freebsd-nq/sys/kern/vfs_subr.c

4895 lines
122 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
* (c) UNIX System Laboratories, Inc.
* All or some portions of this file are derived from material licensed
* to the University of California by American Telephone and Telegraph
* Co. or Unix System Laboratories, Inc. and are reproduced herein with
* the permission of UNIX System Laboratories, Inc.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 4. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
* @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95
1994-05-24 10:09:53 +00:00
*/
/*
* External virtual filesystem routines
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_compat.h"
1996-01-04 21:13:23 +00:00
#include "opt_ddb.h"
#include "opt_watchdog.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/bio.h>
#include <sys/buf.h>
#include <sys/condvar.h>
#include <sys/conf.h>
#include <sys/dirent.h>
#include <sys/event.h>
#include <sys/eventhandler.h>
#include <sys/extattr.h>
#include <sys/file.h>
#include <sys/fcntl.h>
#include <sys/jail.h>
#include <sys/kdb.h>
#include <sys/kernel.h>
#include <sys/kthread.h>
#include <sys/lockf.h>
#include <sys/malloc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/pctrie.h>
#include <sys/priv.h>
#include <sys/reboot.h>
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
#include <sys/rwlock.h>
#include <sys/sched.h>
Switch the sleep/wakeup and condition variable implementations to use the sleep queue interface: - Sleep queues attempt to merge some of the benefits of both sleep queues and condition variables. Having sleep qeueus in a hash table avoids having to allocate a queue head for each wait channel. Thus, struct cv has shrunk down to just a single char * pointer now. However, the hash table does not hold threads directly, but queue heads. This means that once you have located a queue in the hash bucket, you no longer have to walk the rest of the hash chain looking for threads. Instead, you have a list of all the threads sleeping on that wait channel. - Outside of the sleepq code and the sleep/cv code the kernel no longer differentiates between cv's and sleep/wakeup. For example, calls to abortsleep() and cv_abort() are replaced with a call to sleepq_abort(). Thus, the TDF_CVWAITQ flag is removed. Also, calls to unsleep() and cv_waitq_remove() have been replaced with calls to sleepq_remove(). - The sched_sleep() function no longer accepts a priority argument as sleep's no longer inherently bump the priority. Instead, this is soley a propery of msleep() which explicitly calls sched_prio() before blocking. - The TDF_ONSLEEPQ flag has been dropped as it was never used. The associated TDF_SET_ONSLEEPQ and TDF_CLR_ON_SLEEPQ macros have also been dropped and replaced with a single explicit clearing of td_wchan. TD_SET_ONSLEEPQ() would really have only made sense if it had taken the wait channel and message as arguments anyway. Now that that only happens in one place, a macro would be overkill.
2004-02-27 18:52:44 +00:00
#include <sys/sleepqueue.h>
#include <sys/smp.h>
1994-05-24 10:09:53 +00:00
#include <sys/stat.h>
#include <sys/sysctl.h>
#include <sys/syslog.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <sys/watchdog.h>
1994-05-24 10:09:53 +00:00
#include <machine/stdarg.h>
#include <security/mac/mac_framework.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm.h>
#include <vm/vm_object.h>
#include <vm/vm_extern.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/vm_kern.h>
#include <vm/uma.h>
1994-05-24 10:09:53 +00:00
2006-09-04 22:15:44 +00:00
#ifdef DDB
#include <ddb/ddb.h>
#endif
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
static void delmntque(struct vnode *vp);
static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
int slpflag, int slptimeo);
static void syncer_shutdown(void *arg, int howto);
static int vtryrecycle(struct vnode *vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
static void v_incr_usecount(struct vnode *);
static void v_decr_usecount(struct vnode *);
static void v_decr_useonly(struct vnode *);
static void v_upgrade_usecount(struct vnode *);
static void vnlru_free(int);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
static void vgonel(struct vnode *);
static void vfs_knllock(void *arg);
static void vfs_knlunlock(void *arg);
static void vfs_knl_assert_locked(void *arg);
static void vfs_knl_assert_unlocked(void *arg);
static void destroy_vpollinfo(struct vpollinfo *vi);
/*
* Number of vnodes in existence. Increased whenever getnewvnode()
* allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
*/
1997-11-22 08:35:46 +00:00
static unsigned long numvnodes;
SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
2010-11-14 16:10:15 +00:00
"Number of vnodes in existence");
static u_long vnodes_created;
SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
0, "Number of vnodes created by getnewvnode");
/*
* Conversion tables for conversion from vnode types to inode formats
* and back.
*/
1994-05-24 10:09:53 +00:00
enum vtype iftovt_tab[16] = {
VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
};
int vttoif_tab[10] = {
1994-05-24 10:09:53 +00:00
0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
1994-05-24 10:09:53 +00:00
};
2000-10-05 18:22:46 +00:00
/*
* List of vnodes that are ready for recycling.
*/
static TAILQ_HEAD(freelst, vnode) vnode_free_list;
/*
* Free vnode target. Free vnodes may simply be files which have been stat'd
* but not read. This is somewhat common, and a small cache of such files
* should be kept to avoid recreation costs.
*/
static u_long wantfreevnodes;
SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
/* Number of vnodes in the free list. */
static u_long freevnodes;
SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
2010-11-14 16:10:15 +00:00
"Number of vnodes in the free list");
static int vlru_allow_cache_src;
SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
&vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
static u_long recycles_count;
SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
"Number of vnodes recycled to avoid exceding kern.maxvnodes");
/*
2000-10-05 18:22:46 +00:00
* Various variables used for debugging the new implementation of
* reassignbuf().
* XXX these are probably of (very) limited utility now.
*/
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
static int reassignbufcalls;
SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
2010-11-14 16:10:15 +00:00
"Number of calls to reassignbuf");
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
/*
* Cache for the mount type id assigned to NFS. This is used for
* special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
*/
int nfs_mount_type = -1;
/* To keep more than one thread at a time from running vfs_getnewfsid */
static struct mtx mntid_mtx;
2000-10-05 18:22:46 +00:00
/*
* Lock for any access to the following:
* vnode_free_list
* numvnodes
* freevnodes
*/
static struct mtx vnode_free_list_mtx;
/* Publicly exported FS */
struct nfs_public nfs_pub;
2000-10-05 18:22:46 +00:00
static uma_zone_t buf_trie_zone;
/* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
static uma_zone_t vnode_zone;
static uma_zone_t vnodepoll_zone;
2000-10-05 18:22:46 +00:00
/*
* The workitem queue.
2002-06-06 15:46:38 +00:00
*
* It is useful to delay writes of file data and filesystem metadata
* for tens of seconds so that quickly created and deleted files need
* not waste disk bandwidth being created and removed. To realize this,
* we append vnodes to a "workitem" queue. When running with a soft
* updates implementation, most pending metadata dependencies should
* not wait for more than a few seconds. Thus, mounted on block devices
* are delayed only about a half the time that file data is delayed.
* Similarly, directory updates are more critical, so are only delayed
* about a third the time that file data is delayed. Thus, there are
* SYNCER_MAXDELAY queues that are processed round-robin at a rate of
* one each second (driven off the filesystem syncer process). The
* syncer_delayno variable indicates the next queue that is to be processed.
* Items that need to be processed soon are placed in this queue:
*
* syncer_workitem_pending[syncer_delayno]
*
* A delay of fifteen seconds is done by placing the request fifteen
* entries later in the queue:
*
* syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
*
*/
static int syncer_delayno;
2002-06-06 15:46:38 +00:00
static long syncer_mask;
LIST_HEAD(synclist, bufobj);
static struct synclist *syncer_workitem_pending;
/*
* The sync_mtx protects:
* bo->bo_synclist
* sync_vnode_count
* syncer_delayno
* syncer_state
* syncer_workitem_pending
* syncer_worklist_len
* rushjob
*/
static struct mtx sync_mtx;
static struct cv sync_wakeup;
#define SYNCER_MAXDELAY 32
1998-12-21 23:38:33 +00:00
static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */
static int syncdelay = 30; /* max time to delay syncing data */
static int filedelay = 30; /* time to delay syncing files */
SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
2010-11-14 16:10:15 +00:00
"Time to delay syncing files (in seconds)");
static int dirdelay = 29; /* time to delay syncing directories */
SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
2010-11-14 16:10:15 +00:00
"Time to delay syncing directories (in seconds)");
static int metadelay = 28; /* time to delay syncing metadata */
SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
2010-11-14 16:10:15 +00:00
"Time to delay syncing metadata (in seconds)");
static int rushjob; /* number of slots to run ASAP */
static int stat_rush_requests; /* number of times I/O speeded up */
SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
2010-11-14 16:10:15 +00:00
"Number of times I/O speeded up (rush requests)");
/*
* When shutting down the syncer, run it at four times normal speed.
*/
#define SYNCER_SHUTDOWN_SPEEDUP 4
static int sync_vnode_count;
static int syncer_worklist_len;
static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
syncer_state;
/*
2000-10-05 18:22:46 +00:00
* Number of vnodes we want to exist at any one time. This is mostly used
* to size hash tables in vnode-related code. It is normally not used in
* getnewvnode(), as wantfreevnodes is normally nonzero.)
*
* XXX desiredvnodes is historical cruft and should not exist.
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
int desiredvnodes;
2002-06-06 15:46:38 +00:00
SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
&desiredvnodes, 0, "Maximum number of vnodes");
SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
&wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
static int vnlru_nowhere;
SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
&vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
1994-05-24 10:09:53 +00:00
/* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
static int vnsz2log;
/*
* Support for the bufobj clean & dirty pctrie.
*/
static void *
buf_trie_alloc(struct pctrie *ptree)
{
return uma_zalloc(buf_trie_zone, M_NOWAIT);
}
static void
buf_trie_free(struct pctrie *ptree, void *node)
{
uma_zfree(buf_trie_zone, node);
}
PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
1994-05-24 10:09:53 +00:00
/*
* Initialize the vnode management data structures.
*
* Reevaluate the following cap on the number of vnodes after the physical
* memory size exceeds 512GB. In the limit, as the physical memory size
* grows, the ratio of physical pages to vnodes approaches sixteen to one.
1994-05-24 10:09:53 +00:00
*/
2004-08-16 08:33:37 +00:00
#ifndef MAXVNODES_MAX
#define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
2004-08-16 08:33:37 +00:00
#endif
static void
vntblinit(void *dummy __unused)
1994-05-24 10:09:53 +00:00
{
u_int i;
int physvnodes, virtvnodes;
1994-05-24 10:09:53 +00:00
/*
* Desiredvnodes is a function of the physical memory size and the
* kernel's heap size. Generally speaking, it scales with the
* physical memory size. The ratio of desiredvnodes to physical pages
* is one to four until desiredvnodes exceeds 98,304. Thereafter, the
* marginal ratio of desiredvnodes to physical pages is one to
* sixteen. However, desiredvnodes is limited by the kernel's heap
* size. The memory required by desiredvnodes vnodes and vm objects
* may not exceed one seventh of the kernel's heap size.
*/
physvnodes = maxproc + vm_cnt.v_page_count / 16 + 3 * min(98304 * 4,
vm_cnt.v_page_count) / 16;
virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
sizeof(struct vnode)));
desiredvnodes = min(physvnodes, virtvnodes);
2004-08-16 08:33:37 +00:00
if (desiredvnodes > MAXVNODES_MAX) {
if (bootverbose)
printf("Reducing kern.maxvnodes %d -> %d\n",
2004-08-16 08:33:37 +00:00
desiredvnodes, MAXVNODES_MAX);
desiredvnodes = MAXVNODES_MAX;
}
wantfreevnodes = desiredvnodes / 4;
mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
1994-05-24 10:09:53 +00:00
TAILQ_INIT(&vnode_free_list);
mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
NULL, NULL, UMA_ALIGN_PTR, 0);
vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
/*
* Preallocate enough nodes to support one-per buf so that
* we can not fail an insert. reassignbuf() callers can not
* tolerate the insertion failure.
*/
buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR,
UMA_ZONE_NOFREE | UMA_ZONE_VM);
uma_prealloc(buf_trie_zone, nbuf);
/*
* Initialize the filesystem syncer.
2002-06-06 15:46:38 +00:00
*/
syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
&syncer_mask);
syncer_maxdelay = syncer_mask + 1;
mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
cv_init(&sync_wakeup, "syncer");
for (i = 1; i <= sizeof(struct vnode); i <<= 1)
vnsz2log++;
vnsz2log--;
1994-05-24 10:09:53 +00:00
}
SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
1994-05-24 10:09:53 +00:00
/*
* Mark a mount point as busy. Used to synchronize access and to delay
2008-11-02 10:15:42 +00:00
* unmounting. Eventually, mountlist_mtx is not released on failure.
*
* vfs_busy() is a custom lock, it can block the caller.
* vfs_busy() only sleeps if the unmount is active on the mount point.
* For a mountpoint mp, vfs_busy-enforced lock is before lock of any
* vnode belonging to mp.
*
* Lookup uses vfs_busy() to traverse mount points.
* root fs var fs
* / vnode lock A / vnode lock (/var) D
* /var vnode lock B /log vnode lock(/var/log) E
* vfs_busy lock C vfs_busy lock F
*
* Within each file system, the lock order is C->A->B and F->D->E.
*
* When traversing across mounts, the system follows that lock order:
*
* C->A->B
* |
* +->F->D->E
*
* The lookup() process for namei("/var") illustrates the process:
* VOP_LOOKUP() obtains B while A is held
* vfs_busy() obtains a shared lock on F while A and B are held
* vput() releases lock on B
* vput() releases lock on A
* VFS_ROOT() obtains lock on D while shared lock on F is held
* vfs_unbusy() releases shared lock on F
* vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
* Attempt to lock A (instead of vp_crossmp) while D is held would
* violate the global order, causing deadlocks.
*
* dounmount() locks B while F is drained.
1994-05-24 10:09:53 +00:00
*/
int
2008-11-02 10:15:42 +00:00
vfs_busy(struct mount *mp, int flags)
1994-05-24 10:09:53 +00:00
{
2008-11-02 10:15:42 +00:00
MPASS((flags & ~MBF_MASK) == 0);
CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
1994-05-24 10:09:53 +00:00
MNT_ILOCK(mp);
MNT_REF(mp);
/*
* If mount point is currenly being unmounted, sleep until the
* mount point fate is decided. If thread doing the unmounting fails,
* it will clear MNTK_UNMOUNT flag before waking us up, indicating
* that this mount point has survived the unmount attempt and vfs_busy
* should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE
* flag in addition to MNTK_UNMOUNT, indicating that mount point is
* about to be really destroyed. vfs_busy needs to release its
* reference on the mount point in this case and return with ENOENT,
* telling the caller that mount mount it tried to busy is no longer
* valid.
*/
while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
MNT_REL(mp);
MNT_IUNLOCK(mp);
CTR1(KTR_VFS, "%s: failed busying before sleeping",
__func__);
return (ENOENT);
}
2008-11-02 10:15:42 +00:00
if (flags & MBF_MNTLSTLOCK)
mtx_unlock(&mountlist_mtx);
mp->mnt_kern_flag |= MNTK_MWAIT;
msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
2008-11-02 10:15:42 +00:00
if (flags & MBF_MNTLSTLOCK)
mtx_lock(&mountlist_mtx);
MNT_ILOCK(mp);
1994-05-24 10:09:53 +00:00
}
2008-11-02 10:15:42 +00:00
if (flags & MBF_MNTLSTLOCK)
mtx_unlock(&mountlist_mtx);
mp->mnt_lockref++;
MNT_IUNLOCK(mp);
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* Free a busy filesystem.
1994-05-24 10:09:53 +00:00
*/
void
vfs_unbusy(struct mount *mp)
1994-05-24 10:09:53 +00:00
{
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
2008-11-02 10:15:42 +00:00
MNT_ILOCK(mp);
MNT_REL(mp);
KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
2008-11-02 10:15:42 +00:00
mp->mnt_lockref--;
if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
CTR1(KTR_VFS, "%s: waking up waiters", __func__);
2008-11-02 10:15:42 +00:00
mp->mnt_kern_flag &= ~MNTK_DRAINING;
wakeup(&mp->mnt_lockref);
}
MNT_IUNLOCK(mp);
1994-05-24 10:09:53 +00:00
}
/*
* Lookup a mount point by filesystem identifier.
*/
struct mount *
vfs_getvfs(fsid_t *fsid)
1994-05-24 10:09:53 +00:00
{
2005-01-28 12:39:10 +00:00
struct mount *mp;
1994-05-24 10:09:53 +00:00
CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1994-05-24 10:09:53 +00:00
if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
vfs_ref(mp);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&mountlist_mtx);
1994-05-24 10:09:53 +00:00
return (mp);
}
1994-05-24 10:09:53 +00:00
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&mountlist_mtx);
CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
return ((struct mount *) 0);
1994-05-24 10:09:53 +00:00
}
/*
* Lookup a mount point by filesystem identifier, busying it before
* returning.
*
* To avoid congestion on mountlist_mtx, implement simple direct-mapped
* cache for popular filesystem identifiers. The cache is lockess, using
* the fact that struct mount's are never freed. In worst case we may
* get pointer to unmounted or even different filesystem, so we have to
* check what we got, and go slow way if so.
*/
struct mount *
vfs_busyfs(fsid_t *fsid)
{
#define FSID_CACHE_SIZE 256
typedef struct mount * volatile vmp_t;
static vmp_t cache[FSID_CACHE_SIZE];
struct mount *mp;
int error;
uint32_t hash;
CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
hash = fsid->val[0] ^ fsid->val[1];
hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
mp = cache[hash];
if (mp == NULL ||
mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
goto slow;
if (vfs_busy(mp, 0) != 0) {
cache[hash] = NULL;
goto slow;
}
if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
return (mp);
else
vfs_unbusy(mp);
slow:
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
error = vfs_busy(mp, MBF_MNTLSTLOCK);
if (error) {
cache[hash] = NULL;
mtx_unlock(&mountlist_mtx);
return (NULL);
}
cache[hash] = mp;
return (mp);
}
}
CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
mtx_unlock(&mountlist_mtx);
return ((struct mount *) 0);
}
/*
2007-04-10 15:22:40 +00:00
* Check if a user can access privileged mount options.
*/
int
vfs_suser(struct mount *mp, struct thread *td)
{
int error;
/*
* If the thread is jailed, but this is not a jail-friendly file
* system, deny immediately.
*/
Update ZFS from version 6 to 13 and bring some FreeBSD-specific changes. This bring huge amount of changes, I'll enumerate only user-visible changes: - Delegated Administration Allows regular users to perform ZFS operations, like file system creation, snapshot creation, etc. - L2ARC Level 2 cache for ZFS - allows to use additional disks for cache. Huge performance improvements mostly for random read of mostly static content. - slog Allow to use additional disks for ZFS Intent Log to speed up operations like fsync(2). - vfs.zfs.super_owner Allows regular users to perform privileged operations on files stored on ZFS file systems owned by him. Very careful with this one. - chflags(2) Not all the flags are supported. This still needs work. - ZFSBoot Support to boot off of ZFS pool. Not finished, AFAIK. Submitted by: dfr - Snapshot properties - New failure modes Before if write requested failed, system paniced. Now one can select from one of three failure modes: - panic - panic on write error - wait - wait for disk to reappear - continue - serve read requests if possible, block write requests - Refquota, refreservation properties Just quota and reservation properties, but don't count space consumed by children file systems, clones and snapshots. - Sparse volumes ZVOLs that don't reserve space in the pool. - External attributes Compatible with extattr(2). - NFSv4-ACLs Not sure about the status, might not be complete yet. Submitted by: trasz - Creation-time properties - Regression tests for zpool(8) command. Obtained from: OpenSolaris
2008-11-17 20:49:29 +00:00
if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
return (EPERM);
/*
* If the file system was mounted outside the jail of the calling
* thread, deny immediately.
*/
if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
return (EPERM);
Update ZFS from version 6 to 13 and bring some FreeBSD-specific changes. This bring huge amount of changes, I'll enumerate only user-visible changes: - Delegated Administration Allows regular users to perform ZFS operations, like file system creation, snapshot creation, etc. - L2ARC Level 2 cache for ZFS - allows to use additional disks for cache. Huge performance improvements mostly for random read of mostly static content. - slog Allow to use additional disks for ZFS Intent Log to speed up operations like fsync(2). - vfs.zfs.super_owner Allows regular users to perform privileged operations on files stored on ZFS file systems owned by him. Very careful with this one. - chflags(2) Not all the flags are supported. This still needs work. - ZFSBoot Support to boot off of ZFS pool. Not finished, AFAIK. Submitted by: dfr - Snapshot properties - New failure modes Before if write requested failed, system paniced. Now one can select from one of three failure modes: - panic - panic on write error - wait - wait for disk to reappear - continue - serve read requests if possible, block write requests - Refquota, refreservation properties Just quota and reservation properties, but don't count space consumed by children file systems, clones and snapshots. - Sparse volumes ZVOLs that don't reserve space in the pool. - External attributes Compatible with extattr(2). - NFSv4-ACLs Not sure about the status, might not be complete yet. Submitted by: trasz - Creation-time properties - Regression tests for zpool(8) command. Obtained from: OpenSolaris
2008-11-17 20:49:29 +00:00
/*
* If file system supports delegated administration, we don't check
* for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
* by the file system itself.
* If this is not the user that did original mount, we check for
* the PRIV_VFS_MOUNT_OWNER privilege.
*/
if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
return (error);
}
return (0);
}
1994-05-24 10:09:53 +00:00
/*
* Get a new unique fsid. Try to make its val[0] unique, since this value
* will be used to create fake device numbers for stat(). Also try (but
* not so hard) make its val[0] unique mod 2^16, since some emulators only
* support 16-bit device numbers. We end up with unique val[0]'s for the
* first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
*
* Keep in mind that several mounts may be running in parallel. Starting
* the search one past where the previous search terminated is both a
* micro-optimization and a defense against returning the same fsid to
* different mounts.
1994-05-24 10:09:53 +00:00
*/
void
vfs_getnewfsid(struct mount *mp)
1994-05-24 10:09:53 +00:00
{
static uint16_t mntid_base;
struct mount *nmp;
1994-05-24 10:09:53 +00:00
fsid_t tfsid;
int mtype;
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(&mntid_mtx);
mtype = mp->mnt_vfc->vfc_typenum;
tfsid.val[1] = mtype;
mtype = (mtype & 0xFF) << 24;
for (;;) {
tfsid.val[0] = makedev(255,
mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
mntid_base++;
if ((nmp = vfs_getvfs(&tfsid)) == NULL)
break;
vfs_rel(nmp);
1994-05-24 10:09:53 +00:00
}
mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&mntid_mtx);
1994-05-24 10:09:53 +00:00
}
/*
* Knob to control the precision of file timestamps:
*
* 0 = seconds only; nanoseconds zeroed.
* 1 = seconds and nanoseconds, accurate within 1/HZ.
* 2 = seconds and nanoseconds, truncated to microseconds.
* >=3 = seconds and nanoseconds, maximum precision.
*/
enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
static int timestamp_precision = TSP_USEC;
SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
&timestamp_precision, 0, "File timestamp precision (0: seconds, "
2010-11-14 16:10:15 +00:00
"1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
"3+: sec + ns (max. precision))");
/*
* Get a current timestamp.
*/
void
vfs_timestamp(struct timespec *tsp)
{
struct timeval tv;
switch (timestamp_precision) {
case TSP_SEC:
tsp->tv_sec = time_second;
tsp->tv_nsec = 0;
break;
case TSP_HZ:
getnanotime(tsp);
break;
case TSP_USEC:
microtime(&tv);
TIMEVAL_TO_TIMESPEC(&tv, tsp);
break;
case TSP_NSEC:
default:
nanotime(tsp);
break;
}
}
1994-05-24 10:09:53 +00:00
/*
* Set vnode attributes to VNOVAL
*/
void
vattr_null(struct vattr *vap)
1994-05-24 10:09:53 +00:00
{
vap->va_type = VNON;
vap->va_size = VNOVAL;
vap->va_bytes = VNOVAL;
vap->va_mode = VNOVAL;
vap->va_nlink = VNOVAL;
vap->va_uid = VNOVAL;
vap->va_gid = VNOVAL;
vap->va_fsid = VNOVAL;
vap->va_fileid = VNOVAL;
vap->va_blocksize = VNOVAL;
vap->va_rdev = VNOVAL;
vap->va_atime.tv_sec = VNOVAL;
vap->va_atime.tv_nsec = VNOVAL;
vap->va_mtime.tv_sec = VNOVAL;
vap->va_mtime.tv_nsec = VNOVAL;
vap->va_ctime.tv_sec = VNOVAL;
vap->va_ctime.tv_nsec = VNOVAL;
vap->va_birthtime.tv_sec = VNOVAL;
vap->va_birthtime.tv_nsec = VNOVAL;
vap->va_flags = VNOVAL;
vap->va_gen = VNOVAL;
1994-05-24 10:09:53 +00:00
vap->va_vaflags = 0;
}
/*
* This routine is called when we have too many vnodes. It attempts
* to free <count> vnodes and will potentially free vnodes that still
* have VM backing store (VM backing store is typically the cause
* of a vnode blowout so we want to do this). Therefore, this operation
* is not considered cheap.
*
* A number of conditions may prevent a vnode from being reclaimed.
* the buffer cache may have references on the vnode, a directory
* vnode may still have references due to the namei cache representing
* underlying files, or the vnode may be in active use. It is not
* desireable to reuse such vnodes. These conditions may cause the
* number of vnodes to reach some minimum value regardless of what
* you set kern.maxvnodes to. Do not set kern.maxvnodes too low.
*/
static int
vlrureclaim(struct mount *mp)
{
struct vnode *vp;
int done;
int trigger;
int usevnodes;
int count;
/*
* Calculate the trigger point, don't allow user
* screwups to blow us up. This prevents us from
* recycling vnodes with lots of resident pages. We
* aren't trying to free memory, we are trying to
* free vnodes.
*/
usevnodes = desiredvnodes;
if (usevnodes <= 0)
usevnodes = 1;
trigger = vm_cnt.v_page_count * 2 / usevnodes;
done = 0;
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vn_start_write(NULL, &mp, V_WAIT);
MNT_ILOCK(mp);
count = mp->mnt_nvnodelistsize / 10 + 1;
while (count != 0) {
vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
while (vp != NULL && vp->v_type == VMARKER)
vp = TAILQ_NEXT(vp, v_nmntvnodes);
if (vp == NULL)
break;
TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
--count;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
if (!VI_TRYLOCK(vp))
goto next_iter;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* If it's been deconstructed already, it's still
* referenced, or it exceeds the trigger, skip it.
*/
if (vp->v_usecount ||
(!vlru_allow_cache_src &&
!LIST_EMPTY(&(vp)->v_cache_src)) ||
(vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vp->v_object->resident_page_count > trigger)) {
VI_UNLOCK(vp);
goto next_iter;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
}
MNT_IUNLOCK(mp);
vholdl(vp);
if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdrop(vp);
goto next_iter_mntunlocked;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
}
VI_LOCK(vp);
/*
* v_usecount may have been bumped after VOP_LOCK() dropped
* the vnode interlock and before it was locked again.
*
* It is not necessary to recheck VI_DOOMED because it can
* only be set by another thread that holds both the vnode
* lock and vnode interlock. If another thread has the
* vnode lock before we get to VOP_LOCK() and obtains the
* vnode interlock after VOP_LOCK() drops the vnode
* interlock, the other thread will be unable to drop the
* vnode lock before our VOP_LOCK() call fails.
*/
if (vp->v_usecount ||
(!vlru_allow_cache_src &&
!LIST_EMPTY(&(vp)->v_cache_src)) ||
(vp->v_object != NULL &&
vp->v_object->resident_page_count > trigger)) {
VOP_UNLOCK(vp, LK_INTERLOCK);
vdrop(vp);
goto next_iter_mntunlocked;
}
KASSERT((vp->v_iflag & VI_DOOMED) == 0,
("VI_DOOMED unexpectedly detected in vlrureclaim()"));
atomic_add_long(&recycles_count, 1);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vgonel(vp);
VOP_UNLOCK(vp, 0);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdropl(vp);
done++;
next_iter_mntunlocked:
if (!should_yield())
goto relock_mnt;
goto yield;
next_iter:
if (!should_yield())
continue;
MNT_IUNLOCK(mp);
yield:
kern_yield(PRI_USER);
relock_mnt:
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
MNT_ILOCK(mp);
}
MNT_IUNLOCK(mp);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vn_finished_write(mp);
return done;
}
/*
* Attempt to keep the free list at wantfreevnodes length.
*/
static void
vnlru_free(int count)
{
struct vnode *vp;
mtx_assert(&vnode_free_list_mtx, MA_OWNED);
for (; count > 0; count--) {
vp = TAILQ_FIRST(&vnode_free_list);
/*
* The list can be modified while the free_list_mtx
* has been dropped and vp could be NULL here.
*/
if (!vp)
break;
VNASSERT(vp->v_op != NULL, vp,
("vnlru_free: vnode already reclaimed."));
KASSERT((vp->v_iflag & VI_FREE) != 0,
("Removing vnode not on freelist"));
KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
("Mangling active vnode"));
TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
/*
* Don't recycle if we can't get the interlock.
*/
if (!VI_TRYLOCK(vp)) {
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
continue;
}
VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
vp, ("vp inconsistent on freelist"));
/*
* The clear of VI_FREE prevents activation of the
* vnode. There is no sense in putting the vnode on
* the mount point active list, only to remove it
* later during recycling. Inline the relevant part
* of vholdl(), to avoid triggering assertions or
* activating.
*/
freevnodes--;
vp->v_iflag &= ~VI_FREE;
vp->v_holdcnt++;
mtx_unlock(&vnode_free_list_mtx);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
VI_UNLOCK(vp);
vtryrecycle(vp);
/*
* If the recycled succeeded this vdrop will actually free
* the vnode. If not it will simply place it back on
* the free list.
*/
vdrop(vp);
mtx_lock(&vnode_free_list_mtx);
}
}
/*
* Attempt to recycle vnodes in a context that is always safe to block.
2002-05-16 21:28:32 +00:00
* Calling vlrurecycle() from the bowels of filesystem code has some
* interesting deadlock problems.
*/
static struct proc *vnlruproc;
static int vnlruproc_sig;
2002-06-06 15:46:38 +00:00
static void
vnlru_proc(void)
{
struct mount *mp, *nmp;
int done;
struct proc *p = vnlruproc;
EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
2002-06-06 15:46:38 +00:00
SHUTDOWN_PRI_FIRST);
for (;;) {
kproc_suspend_check(p);
mtx_lock(&vnode_free_list_mtx);
if (freevnodes > wantfreevnodes)
vnlru_free(freevnodes - wantfreevnodes);
if (numvnodes <= desiredvnodes * 9 / 10) {
vnlruproc_sig = 0;
wakeup(&vnlruproc_sig);
msleep(vnlruproc, &vnode_free_list_mtx,
PVFS|PDROP, "vlruwt", hz);
continue;
}
mtx_unlock(&vnode_free_list_mtx);
done = 0;
mtx_lock(&mountlist_mtx);
for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
2008-11-02 10:15:42 +00:00
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
nmp = TAILQ_NEXT(mp, mnt_list);
continue;
}
done += vlrureclaim(mp);
mtx_lock(&mountlist_mtx);
nmp = TAILQ_NEXT(mp, mnt_list);
vfs_unbusy(mp);
}
mtx_unlock(&mountlist_mtx);
if (done == 0) {
#if 0
/* These messages are temporary debugging aids */
if (vnlru_nowhere < 5)
printf("vnlru process getting nowhere..\n");
else if (vnlru_nowhere == 5)
printf("vnlru process messages stopped.\n");
#endif
vnlru_nowhere++;
tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
} else
kern_yield(PRI_USER);
}
}
static struct kproc_desc vnlru_kp = {
"vnlru",
vnlru_proc,
&vnlruproc
};
SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
&vnlru_kp);
1994-05-24 10:09:53 +00:00
/*
* Routines having to do with the management of the vnode table.
*/
/*
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
* Try to recycle a freed vnode. We abort if anyone picks up a reference
* before we actually vgone(). This function must be called with the vnode
* held to prevent the vnode from being returned to the free list midway
* through vgone().
*/
static int
vtryrecycle(struct vnode *vp)
{
struct mount *vnmp;
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
VNASSERT(vp->v_holdcnt, vp,
("vtryrecycle: Recycling vp %p without a reference.", vp));
/*
* This vnode may found and locked via some other list, if so we
* can't recycle it yet.
*/
if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
CTR2(KTR_VFS,
"%s: impossible to recycle, vp %p lock is already held",
__func__, vp);
return (EWOULDBLOCK);
}
/*
* Don't recycle if its filesystem is being suspended.
*/
if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
VOP_UNLOCK(vp, 0);
CTR2(KTR_VFS,
"%s: impossible to recycle, cannot start the write for %p",
__func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
return (EBUSY);
}
/*
* If we got this far, we need to acquire the interlock and see if
* anyone picked up this vnode from another list. If not, we will
* mark it with DOOMED via vgonel() so that anyone who does find it
* will skip over it.
*/
VI_LOCK(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
if (vp->v_usecount) {
VOP_UNLOCK(vp, LK_INTERLOCK);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vn_finished_write(vnmp);
CTR2(KTR_VFS,
"%s: impossible to recycle, %p is already referenced",
__func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
return (EBUSY);
}
if ((vp->v_iflag & VI_DOOMED) == 0) {
atomic_add_long(&recycles_count, 1);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vgonel(vp);
}
VOP_UNLOCK(vp, LK_INTERLOCK);
vn_finished_write(vnmp);
return (0);
}
1994-05-24 10:09:53 +00:00
/*
* Wait for available vnodes.
1994-05-24 10:09:53 +00:00
*/
static int
getnewvnode_wait(int suspended)
1994-05-24 10:09:53 +00:00
{
mtx_assert(&vnode_free_list_mtx, MA_OWNED);
if (numvnodes > desiredvnodes) {
if (suspended) {
/*
* File system is beeing suspended, we cannot risk a
* deadlock here, so allocate new vnode anyway.
*/
if (freevnodes > wantfreevnodes)
vnlru_free(freevnodes - wantfreevnodes);
return (0);
}
if (vnlruproc_sig == 0) {
vnlruproc_sig = 1; /* avoid unnecessary wakeups */
wakeup(vnlruproc);
}
msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
"vlruwk", hz);
}
return (numvnodes > desiredvnodes ? ENFILE : 0);
}
void
getnewvnode_reserve(u_int count)
{
struct thread *td;
td = curthread;
/* First try to be quick and racy. */
if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
td->td_vp_reserv += count;
return;
} else
atomic_subtract_long(&numvnodes, count);
mtx_lock(&vnode_free_list_mtx);
while (count > 0) {
if (getnewvnode_wait(0) == 0) {
count--;
td->td_vp_reserv++;
atomic_add_long(&numvnodes, 1);
}
}
mtx_unlock(&vnode_free_list_mtx);
}
void
getnewvnode_drop_reserve(void)
{
struct thread *td;
td = curthread;
atomic_subtract_long(&numvnodes, td->td_vp_reserv);
td->td_vp_reserv = 0;
}
/*
* Return the next vnode from the free list.
*/
int
getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
struct vnode **vpp)
{
struct vnode *vp;
struct bufobj *bo;
struct thread *td;
int error;
CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
vp = NULL;
td = curthread;
if (td->td_vp_reserv > 0) {
td->td_vp_reserv -= 1;
goto alloc;
}
mtx_lock(&vnode_free_list_mtx);
/*
* Lend our context to reclaim vnodes if they've exceeded the max.
*/
if (freevnodes > wantfreevnodes)
vnlru_free(1);
error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
MNTK_SUSPEND));
#if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
if (error != 0) {
mtx_unlock(&vnode_free_list_mtx);
return (error);
}
#endif
atomic_add_long(&numvnodes, 1);
mtx_unlock(&vnode_free_list_mtx);
alloc:
atomic_add_long(&vnodes_created, 1);
vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
/*
* Setup locks.
*/
vp->v_vnlock = &vp->v_lock;
mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
/*
* By default, don't allow shared locks unless filesystems
* opt-in.
*/
lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
/*
* Initialize bufobj.
*/
bo = &vp->v_bufobj;
bo->__bo_vnode = vp;
rw_init(BO_LOCKPTR(bo), "bufobj interlock");
bo->bo_ops = &buf_ops_bio;
bo->bo_private = vp;
TAILQ_INIT(&bo->bo_clean.bv_hd);
TAILQ_INIT(&bo->bo_dirty.bv_hd);
/*
* Initialize namecache.
*/
LIST_INIT(&vp->v_cache_src);
TAILQ_INIT(&vp->v_cache_dst);
/*
* Finalize various vnode identity bits.
*/
vp->v_type = VNON;
1994-05-24 10:09:53 +00:00
vp->v_tag = tag;
vp->v_op = vops;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
v_incr_usecount(vp);
vp->v_data = NULL;
#ifdef MAC
mac_vnode_init(vp);
if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
mac_vnode_associate_singlelabel(mp, vp);
else if (mp == NULL && vops != &dead_vnodeops)
printf("NULL mp in getnewvnode()\n");
#endif
if (mp != NULL) {
bo->bo_bsize = mp->mnt_stat.f_iosize;
if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
vp->v_vflag |= VV_NOKNOTE;
}
rangelock_init(&vp->v_rl);
/*
* For the filesystems which do not use vfs_hash_insert(),
* still initialize v_hash to have vfs_hash_index() useful.
* E.g., nullfs uses vfs_hash_index() on the lower vnode for
* its own hashing.
*/
vp->v_hash = (uintptr_t)vp >> vnsz2log;
*vpp = vp;
1994-05-24 10:09:53 +00:00
return (0);
}
/*
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
* Delete from old mount point vnode list, if on one.
1994-05-24 10:09:53 +00:00
*/
1997-11-22 08:35:46 +00:00
static void
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
delmntque(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
struct mount *mp;
int active;
1994-05-24 10:09:53 +00:00
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
mp = vp->v_mount;
if (mp == NULL)
return;
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
MNT_ILOCK(mp);
VI_LOCK(vp);
KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
("Active vnode list size %d > Vnode list size %d",
mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
active = vp->v_iflag & VI_ACTIVE;
vp->v_iflag &= ~VI_ACTIVE;
if (active) {
mtx_lock(&vnode_free_list_mtx);
TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
mp->mnt_activevnodelistsize--;
mtx_unlock(&vnode_free_list_mtx);
}
vp->v_mount = NULL;
VI_UNLOCK(vp);
2005-02-17 10:28:58 +00:00
VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
("bad mount point vnode list size"));
TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
mp->mnt_nvnodelistsize--;
MNT_REL(mp);
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
MNT_IUNLOCK(mp);
}
static void
insmntque_stddtr(struct vnode *vp, void *dtr_arg)
{
vp->v_data = NULL;
vp->v_op = &dead_vnodeops;
vgone(vp);
vput(vp);
}
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
/*
* Insert into list of vnodes for the new mount point, if available.
*/
int
insmntque1(struct vnode *vp, struct mount *mp,
void (*dtr)(struct vnode *, void *), void *dtr_arg)
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
{
KASSERT(vp->v_mount == NULL,
("insmntque: vnode already on per mount vnode list"));
2005-02-17 10:28:58 +00:00
VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
/*
* We acquire the vnode interlock early to ensure that the
* vnode cannot be recycled by another process releasing a
* holdcnt on it before we get it on both the vnode list
* and the active vnode list. The mount mutex protects only
* manipulation of the vnode list and the vnode freelist
* mutex protects only manipulation of the active vnode list.
* Hence the need to hold the vnode interlock throughout.
*/
MNT_ILOCK(mp);
VI_LOCK(vp);
if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
mp->mnt_nvnodelistsize == 0)) &&
(vp->v_vflag & VV_FORCEINSMQ) == 0) {
VI_UNLOCK(vp);
MNT_IUNLOCK(mp);
if (dtr != NULL)
dtr(vp, dtr_arg);
return (EBUSY);
}
vp->v_mount = mp;
MNT_REF(mp);
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
("neg mount point vnode list size"));
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
mp->mnt_nvnodelistsize++;
KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
("Activating already active vnode"));
vp->v_iflag |= VI_ACTIVE;
mtx_lock(&vnode_free_list_mtx);
TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
mp->mnt_activevnodelistsize++;
mtx_unlock(&vnode_free_list_mtx);
VI_UNLOCK(vp);
MNT_IUNLOCK(mp);
return (0);
}
int
insmntque(struct vnode *vp, struct mount *mp)
{
return (insmntque1(vp, mp, insmntque_stddtr, NULL));
1994-05-24 10:09:53 +00:00
}
/*
* Flush out and invalidate all buffers associated with a bufobj
1994-05-24 10:09:53 +00:00
* Called with the underlying object locked.
*/
int
bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
1994-05-24 10:09:53 +00:00
{
int error;
BO_LOCK(bo);
if (flags & V_SAVE) {
error = bufobj_wwait(bo, slpflag, slptimeo);
if (error) {
2005-01-11 10:16:39 +00:00
BO_UNLOCK(bo);
return (error);
}
if (bo->bo_dirty.bv_cnt > 0) {
2005-01-11 10:16:39 +00:00
BO_UNLOCK(bo);
if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
return (error);
/*
* XXX We could save a lock/unlock if this was only
* enabled under INVARIANTS
*/
2005-01-11 10:16:39 +00:00
BO_LOCK(bo);
if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
panic("vinvalbuf: dirty bufs");
}
2002-06-06 15:46:38 +00:00
}
/*
* If you alter this loop please notice that interlock is dropped and
* reacquired in flushbuflist. Special care is needed to ensure that
* no race conditions occur from this.
*/
do {
error = flushbuflist(&bo->bo_clean,
flags, bo, slpflag, slptimeo);
Add the posix_fadvise(2) system call. It is somewhat similar to madvise(2) except that it operates on a file descriptor instead of a memory region. It is currently only supported on regular files. Just as with madvise(2), the advice given to posix_fadvise(2) can be divided into two types. The first type provide hints about data access patterns and are used in the file read and write routines to modify the I/O flags passed down to VOP_READ() and VOP_WRITE(). These modes are thus filesystem independent. Note that to ease implementation (and since this API is only advisory anyway), only a single non-normal range is allowed per file descriptor. The second type of hints are used to hint to the OS that data will or will not be used. These hints are implemented via a new VOP_ADVISE(). A default implementation is provided which does nothing for the WILLNEED request and attempts to move any clean pages to the cache page queue for the DONTNEED request. This latter case required two other changes. First, a new V_CLEANONLY flag was added to vinvalbuf(). This requests vinvalbuf() to only flush clean buffers for the vnode from the buffer cache and to not remove any backing pages from the vnode. This is used to ensure clean pages are not wired into the buffer cache before attempting to move them to the cache page queue. The second change adds a new vm_object_page_cache() method. This method is somewhat similar to vm_object_page_remove() except that instead of freeing each page in the specified range, it attempts to move clean pages to the cache queue if possible. To preserve the ABI of struct file, the f_cdevpriv pointer is now reused in a union to point to the currently active advice region if one is present for regular files. Reviewed by: jilles, kib, arch@ Approved by: re (kib) MFC after: 1 month
2011-11-04 04:02:50 +00:00
if (error == 0 && !(flags & V_CLEANONLY))
error = flushbuflist(&bo->bo_dirty,
flags, bo, slpflag, slptimeo);
if (error != 0 && error != EAGAIN) {
BO_UNLOCK(bo);
return (error);
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
}
} while (error != 0);
/*
* Wait for I/O to complete. XXX needs cleaning up. The vnode can
* have write I/O in-progress but if there is a VM object then the
* VM object can also have read-I/O in-progress.
*/
do {
bufobj_wwait(bo, 0, 0);
BO_UNLOCK(bo);
if (bo->bo_object != NULL) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(bo->bo_object);
vm_object_pip_wait(bo->bo_object, "bovlbx");
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(bo->bo_object);
}
BO_LOCK(bo);
} while (bo->bo_numoutput > 0);
BO_UNLOCK(bo);
/*
* Destroy the copy in the VM cache, too.
*/
Add the posix_fadvise(2) system call. It is somewhat similar to madvise(2) except that it operates on a file descriptor instead of a memory region. It is currently only supported on regular files. Just as with madvise(2), the advice given to posix_fadvise(2) can be divided into two types. The first type provide hints about data access patterns and are used in the file read and write routines to modify the I/O flags passed down to VOP_READ() and VOP_WRITE(). These modes are thus filesystem independent. Note that to ease implementation (and since this API is only advisory anyway), only a single non-normal range is allowed per file descriptor. The second type of hints are used to hint to the OS that data will or will not be used. These hints are implemented via a new VOP_ADVISE(). A default implementation is provided which does nothing for the WILLNEED request and attempts to move any clean pages to the cache page queue for the DONTNEED request. This latter case required two other changes. First, a new V_CLEANONLY flag was added to vinvalbuf(). This requests vinvalbuf() to only flush clean buffers for the vnode from the buffer cache and to not remove any backing pages from the vnode. This is used to ensure clean pages are not wired into the buffer cache before attempting to move them to the cache page queue. The second change adds a new vm_object_page_cache() method. This method is somewhat similar to vm_object_page_remove() except that instead of freeing each page in the specified range, it attempts to move clean pages to the cache queue if possible. To preserve the ABI of struct file, the f_cdevpriv pointer is now reused in a union to point to the currently active advice region if one is present for regular files. Reviewed by: jilles, kib, arch@ Approved by: re (kib) MFC after: 1 month
2011-11-04 04:02:50 +00:00
if (bo->bo_object != NULL &&
(flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(bo->bo_object);
vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
OBJPR_CLEANONLY : 0);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(bo->bo_object);
}
#ifdef INVARIANTS
2005-01-11 10:16:39 +00:00
BO_LOCK(bo);
Add the posix_fadvise(2) system call. It is somewhat similar to madvise(2) except that it operates on a file descriptor instead of a memory region. It is currently only supported on regular files. Just as with madvise(2), the advice given to posix_fadvise(2) can be divided into two types. The first type provide hints about data access patterns and are used in the file read and write routines to modify the I/O flags passed down to VOP_READ() and VOP_WRITE(). These modes are thus filesystem independent. Note that to ease implementation (and since this API is only advisory anyway), only a single non-normal range is allowed per file descriptor. The second type of hints are used to hint to the OS that data will or will not be used. These hints are implemented via a new VOP_ADVISE(). A default implementation is provided which does nothing for the WILLNEED request and attempts to move any clean pages to the cache page queue for the DONTNEED request. This latter case required two other changes. First, a new V_CLEANONLY flag was added to vinvalbuf(). This requests vinvalbuf() to only flush clean buffers for the vnode from the buffer cache and to not remove any backing pages from the vnode. This is used to ensure clean pages are not wired into the buffer cache before attempting to move them to the cache page queue. The second change adds a new vm_object_page_cache() method. This method is somewhat similar to vm_object_page_remove() except that instead of freeing each page in the specified range, it attempts to move clean pages to the cache queue if possible. To preserve the ABI of struct file, the f_cdevpriv pointer is now reused in a union to point to the currently active advice region if one is present for regular files. Reviewed by: jilles, kib, arch@ Approved by: re (kib) MFC after: 1 month
2011-11-04 04:02:50 +00:00
if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
2005-01-11 10:16:39 +00:00
(bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
1994-05-24 10:09:53 +00:00
panic("vinvalbuf: flush failed");
2005-01-11 10:16:39 +00:00
BO_UNLOCK(bo);
#endif
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* Flush out and invalidate all buffers associated with a vnode.
* Called with the underlying object locked.
*/
int
vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
{
CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
ASSERT_VOP_LOCKED(vp, "vinvalbuf");
if (vp->v_object != NULL && vp->v_object->handle != vp)
return (0);
return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
}
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
/*
* Flush out buffers on the specified list.
*
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
*/
static int
2011-10-27 17:43:36 +00:00
flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
int slptimeo)
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
{
struct buf *bp, *nbp;
int retval, error;
daddr_t lblkno;
b_xflags_t xflags;
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
ASSERT_BO_WLOCKED(bo);
retval = 0;
TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
continue;
}
lblkno = 0;
xflags = 0;
if (nbp != NULL) {
lblkno = nbp->b_lblkno;
xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
}
retval = EAGAIN;
error = BUF_TIMELOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
"flushbuf", slpflag, slptimeo);
if (error) {
2005-01-11 10:16:39 +00:00
BO_LOCK(bo);
return (error != ENOLCK ? error : EAGAIN);
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
}
KASSERT(bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p",
bp, bp->b_bufobj, bo));
if (bp->b_bufobj != bo) { /* XXX: necessary ? */
BUF_UNLOCK(bp);
BO_LOCK(bo);
return (EAGAIN);
}
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
/*
* XXX Since there are no node locks for NFS, I
* believe there is a slight chance that a delayed
* write will occur while sleeping just above, so
* check for it.
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
*/
if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
(flags & V_SAVE)) {
bremfree(bp);
bp->b_flags |= B_ASYNC;
bwrite(bp);
2005-01-11 10:16:39 +00:00
BO_LOCK(bo);
return (EAGAIN); /* XXX: why not loop ? */
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
}
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
bp->b_flags &= ~B_ASYNC;
brelse(bp);
2005-01-11 10:16:39 +00:00
BO_LOCK(bo);
if (nbp != NULL &&
(nbp->b_bufobj != bo ||
nbp->b_lblkno != lblkno ||
(nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
break; /* nbp invalid */
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
}
return (retval);
Add support to UFS2 to provide storage for extended attributes. As this code is not actually used by any of the existing interfaces, it seems unlikely to break anything (famous last words). The internal kernel interface to manipulate these attributes is invoked using two new IO_ flags: IO_NORMAL and IO_EXT. These flags may be specified in the ioflags word of VOP_READ, VOP_WRITE, and VOP_TRUNCATE. Specifying IO_NORMAL means that you want to do I/O to the normal data part of the file and IO_EXT means that you want to do I/O to the extended attributes part of the file. IO_NORMAL and IO_EXT are mutually exclusive for VOP_READ and VOP_WRITE, but may be specified individually or together in the case of VOP_TRUNCATE. For example, when removing a file, VOP_TRUNCATE is called with both IO_NORMAL and IO_EXT set. For backward compatibility, if neither IO_NORMAL nor IO_EXT is set, then IO_NORMAL is assumed. Note that the BA_ and IO_ flags have been `merged' so that they may both be used in the same flags word. This merger is possible by assigning the IO_ flags to the low sixteen bits and the BA_ flags the high sixteen bits. This works because the high sixteen bits of the IO_ word is reserved for read-ahead and help with write clustering so will never be used for flags. This merge lets us get away from code of the form: if (ioflags & IO_SYNC) flags |= BA_SYNC; For the future, I have considered adding a new field to the vattr structure, va_extsize. This addition could then be exported through the stat structure to allow applications to find out the size of the extended attribute storage and also would provide a more standard interface for truncating them (via VOP_SETATTR rather than VOP_TRUNCATE). I am also contemplating adding a pathconf parameter (for concreteness, lets call it _PC_MAX_EXTSIZE) which would let an application determine the maximum size of the extended atribute storage. Sponsored by: DARPA & NAI Labs.
2002-07-19 07:29:39 +00:00
}
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
/*
* Truncate a file's buffer and pages to a specified length. This
* is in lieu of the old vinvalbuf mechanism, which performed unneeded
* sync activity.
*/
int
vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
{
struct buf *bp, *nbp;
int anyfreed;
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
int trunclbn;
struct bufobj *bo;
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
vp, cred, blksize, (uintmax_t)length);
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
/*
* Round up to the *next* lbn.
*/
trunclbn = (length + blksize - 1) / blksize;
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
ASSERT_VOP_LOCKED(vp, "vtruncbuf");
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
restart:
bo = &vp->v_bufobj;
BO_LOCK(bo);
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
anyfreed = 1;
for (;anyfreed;) {
anyfreed = 0;
TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno < trunclbn)
continue;
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_LOCKPTR(bo)) == ENOLCK)
goto restart;
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
bp->b_flags &= ~B_ASYNC;
brelse(bp);
anyfreed = 1;
BO_LOCK(bo);
if (nbp != NULL &&
(((nbp->b_xflags & BX_VNCLEAN) == 0) ||
(nbp->b_vp != vp) ||
(nbp->b_flags & B_DELWRI))) {
BO_UNLOCK(bo);
goto restart;
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
}
}
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno < trunclbn)
continue;
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_LOCKPTR(bo)) == ENOLCK)
goto restart;
bremfree(bp);
bp->b_flags |= (B_INVAL | B_RELBUF);
bp->b_flags &= ~B_ASYNC;
brelse(bp);
anyfreed = 1;
BO_LOCK(bo);
if (nbp != NULL &&
(((nbp->b_xflags & BX_VNDIRTY) == 0) ||
(nbp->b_vp != vp) ||
(nbp->b_flags & B_DELWRI) == 0)) {
BO_UNLOCK(bo);
goto restart;
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
}
}
}
if (length > 0) {
restartsync:
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (bp->b_lblkno > 0)
continue;
/*
* Since we hold the vnode lock this should only
* fail if we're racing with the buf daemon.
*/
if (BUF_LOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_LOCKPTR(bo)) == ENOLCK) {
goto restart;
}
2005-02-17 10:28:58 +00:00
VNASSERT((bp->b_flags & B_DELWRI), vp,
("buf(%p) on dirty queue without DELWRI", bp));
bremfree(bp);
bawrite(bp);
BO_LOCK(bo);
goto restartsync;
}
}
bufobj_wwait(bo, 0, 0);
BO_UNLOCK(bo);
Some VM improvements, including elimination of alot of Sig-11 problems. Tor Egge and others have helped with various VM bugs lately, but don't blame him -- blame me!!! pmap.c: 1) Create an object for kernel page table allocations. This fixes a bogus allocation method previously used for such, by grabbing pages from the kernel object, using bogus pindexes. (This was a code cleanup, and perhaps a minor system stability issue.) pmap.c: 2) Pre-set the modify and accessed bits when prudent. This will decrease bus traffic under certain circumstances. vfs_bio.c, vfs_cluster.c: 3) Rather than calculating the beginning virtual byte offset multiple times, stick the offset into the buffer header, so that the calculated offset can be reused. (Long long multiplies are often expensive, and this is a probably unmeasurable performance improvement, and code cleanup.) vfs_bio.c: 4) Handle write recursion more intelligently (but not perfectly) so that it is less likely to cause a system panic, and is also much more robust. vfs_bio.c: 5) getblk incorrectly wrote out blocks that are incorrectly sized. The problem is fixed, and writes blocks out ONLY when B_DELWRI is true. vfs_bio.c: 6) Check that already constituted buffers have fully valid pages. If not, then make sure that the B_CACHE bit is not set. (This was a major source of Sig-11 type problems.) vfs_bio.c: 7) Fix a potential system deadlock due to an incorrectly specified sleep priority while waiting for a buffer write operation. The change that I made opens the system up to serious problems, and we need to examine the issue of process sleep priorities. vfs_cluster.c, vfs_bio.c: 8) Make clustered reads work more correctly (and more completely) when buffers are already constituted, but not fully valid. (This was another system reliability issue.) vfs_subr.c, ffs_inode.c: 9) Create a vtruncbuf function, which is used by filesystems that can truncate files. The vinvalbuf forced a file sync type operation, while vtruncbuf only invalidates the buffers past the new end of file, and also invalidates the appropriate pages. (This was a system reliabiliy and performance issue.) 10) Modify FFS to use vtruncbuf. vm_object.c: 11) Make the object rundown mechanism for OBJT_VNODE type objects work more correctly. Included in that fix, create pager entries for the OBJT_DEAD pager type, so that paging requests that might slip in during race conditions are properly handled. (This was a system reliability issue.) vm_page.c: 12) Make some of the page validation routines be a little less picky about arguments passed to them. Also, support page invalidation change the object generation count so that we handle generation counts a little more robustly. vm_pageout.c: 13) Further reduce pageout daemon activity when the system doesn't need help from it. There should be no additional performance decrease even when the pageout daemon is running. (This was a significant performance issue.) vnode_pager.c: 14) Teach the vnode pager to handle race conditions during vnode deallocations.
1998-03-16 01:56:03 +00:00
vnode_pager_setsize(vp, length);
return (0);
}
static void
buf_vlist_remove(struct buf *bp)
{
struct bufv *bv;
KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
ASSERT_BO_WLOCKED(bp->b_bufobj);
KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
(BX_VNDIRTY|BX_VNCLEAN),
("buf_vlist_remove: Buf %p is on two lists", bp));
if (bp->b_xflags & BX_VNDIRTY)
bv = &bp->b_bufobj->bo_dirty;
else
bv = &bp->b_bufobj->bo_clean;
BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
bv->bv_cnt--;
bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
}
/*
* Add the buffer to the sorted clean or dirty block list.
*
* NOTE: xflags is passed as a constant, optimizing this inline function!
*/
static void
buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
{
struct bufv *bv;
struct buf *n;
int error;
ASSERT_BO_WLOCKED(bo);
KASSERT((bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo));
KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
bp->b_xflags |= xflags;
if (xflags & BX_VNDIRTY)
bv = &bo->bo_dirty;
else
bv = &bo->bo_clean;
/*
* Keep the list ordered. Optimize empty list insertion. Assume
* we tend to grow at the tail so lookup_le should usually be cheaper
* than _ge.
*/
if (bv->bv_cnt == 0 ||
bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
else
TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
if (error)
panic("buf_vlist_add: Preallocated nodes insufficient.");
bv->bv_cnt++;
}
/*
* Lookup a buffer using the splay tree. Note that we specifically avoid
* shadow buffers used in background bitmap writes.
*
* This code isn't quite efficient as it could be because we are maintaining
* two sorted lists and do not know which list the block resides in.
*
* During a "make buildworld" the desired buffer is found at one of
* the roots more than 60% of the time. Thus, checking both roots
* before performing either splay eliminates unnecessary splays on the
* first tree splayed.
*/
struct buf *
gbincore(struct bufobj *bo, daddr_t lblkno)
{
struct buf *bp;
ASSERT_BO_LOCKED(bo);
bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
if (bp != NULL)
return (bp);
return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
}
1994-05-24 10:09:53 +00:00
/*
* Associate a buffer with a vnode.
*/
void
bgetvp(struct vnode *vp, struct buf *bp)
1994-05-24 10:09:53 +00:00
{
struct bufobj *bo;
2005-02-17 10:28:58 +00:00
bo = &vp->v_bufobj;
ASSERT_BO_WLOCKED(bo);
2005-02-17 10:28:58 +00:00
VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
1999-01-10 01:58:29 +00:00
CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
2005-02-17 10:28:58 +00:00
VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
("bgetvp: bp already attached! %p", bp));
vhold(vp);
1994-05-24 10:09:53 +00:00
bp->b_vp = vp;
bp->b_bufobj = bo;
1994-05-24 10:09:53 +00:00
/*
* Insert onto list for new vnode.
*/
buf_vlist_add(bp, bo, BX_VNCLEAN);
1994-05-24 10:09:53 +00:00
}
/*
* Disassociate a buffer from a vnode.
*/
void
brelvp(struct buf *bp)
1994-05-24 10:09:53 +00:00
{
struct bufobj *bo;
1994-05-24 10:09:53 +00:00
struct vnode *vp;
CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
1994-05-24 10:09:53 +00:00
/*
* Delete from old vnode list, if on one.
*/
vp = bp->b_vp; /* XXX */
bo = bp->b_bufobj;
BO_LOCK(bo);
if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
buf_vlist_remove(bp);
else
panic("brelvp: Buffer %p not on queue.", bp);
if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
bo->bo_flag &= ~BO_ONWORKLST;
mtx_lock(&sync_mtx);
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
mtx_unlock(&sync_mtx);
}
bp->b_vp = NULL;
bp->b_bufobj = NULL;
BO_UNLOCK(bo);
vdrop(vp);
1994-05-24 10:09:53 +00:00
}
/*
* Add an item to the syncer work queue.
*/
static void
vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
{
int slot;
ASSERT_BO_WLOCKED(bo);
mtx_lock(&sync_mtx);
if (bo->bo_flag & BO_ONWORKLST)
LIST_REMOVE(bo, bo_synclist);
else {
bo->bo_flag |= BO_ONWORKLST;
syncer_worklist_len++;
}
if (delay > syncer_maxdelay - 2)
delay = syncer_maxdelay - 2;
slot = (syncer_delayno + delay) & syncer_mask;
LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
mtx_unlock(&sync_mtx);
}
static int
sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
{
int error, len;
mtx_lock(&sync_mtx);
len = syncer_worklist_len - sync_vnode_count;
mtx_unlock(&sync_mtx);
error = SYSCTL_OUT(req, &len, sizeof(len));
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
Un-staticize runningbufwakeup() and staticize updateproc. Add a new private thread flag to indicate that the thread should not sleep if runningbufspace is too large. Set this flag on the bufdaemon and syncer threads so that they skip the waitrunningbufspace() call in bufwrite() rather than than checking the proc pointer vs. the known proc pointers for these two threads. A way of preventing these threads from being starved for I/O but still placing limits on their outstanding I/O would be desirable. Set this flag in ffs_copyonwrite() to prevent bufwrite() calls from blocking on the runningbufspace check while holding snaplk. This prevents snaplk from being held for an arbitrarily long period of time if runningbufspace is high and greatly reduces the contention for snaplk. The disadvantage is that ffs_copyonwrite() can start a large amount of I/O if there are a large number of snapshots, which could cause a deadlock in other parts of the code. Call runningbufwakeup() in ffs_copyonwrite() to decrement runningbufspace before attempting to grab snaplk so that I/O requests waiting on snaplk are not counted in runningbufspace as being in-progress. Increment runningbufspace again before actually launching the original I/O request. Prior to the above two changes, the system could deadlock if enough I/O requests were blocked by snaplk to prevent runningbufspace from falling below lorunningspace and one of the bawrite() calls in ffs_copyonwrite() blocked in waitrunningbufspace() while holding snaplk. See <http://www.holm.cc/stress/log/cons143.html>
2005-09-30 01:30:01 +00:00
static struct proc *updateproc;
2002-03-19 21:25:46 +00:00
static void sched_sync(void);
static struct kproc_desc up_kp = {
"syncer",
sched_sync,
&updateproc
};
SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
static int
sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
{
struct vnode *vp;
struct mount *mp;
*bo = LIST_FIRST(slp);
if (*bo == NULL)
return (0);
vp = (*bo)->__bo_vnode; /* XXX */
if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
return (1);
/*
* We use vhold in case the vnode does not
* successfully sync. vhold prevents the vnode from
* going away when we unlock the sync_mtx so that
* we can acquire the vnode interlock.
*/
vholdl(vp);
mtx_unlock(&sync_mtx);
VI_UNLOCK(vp);
if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
vdrop(vp);
mtx_lock(&sync_mtx);
return (*bo == LIST_FIRST(slp));
}
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
(void) VOP_FSYNC(vp, MNT_LAZY, td);
VOP_UNLOCK(vp, 0);
vn_finished_write(mp);
BO_LOCK(*bo);
if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
/*
* Put us back on the worklist. The worklist
* routine will remove us from our current
* position and then add us back in at a later
* position.
*/
vn_syncer_add_to_worklist(*bo, syncdelay);
}
BO_UNLOCK(*bo);
vdrop(vp);
mtx_lock(&sync_mtx);
return (0);
}
static int first_printf = 1;
/*
* System filesystem synchronizer daemon.
*/
static void
sched_sync(void)
{
struct synclist *next, *slp;
struct bufobj *bo;
long starttime;
struct thread *td = curthread;
int last_work_seen;
int net_worklist_len;
int syncer_final_iter;
int error;
last_work_seen = 0;
syncer_final_iter = 0;
syncer_state = SYNCER_RUNNING;
starttime = time_uptime;
Un-staticize runningbufwakeup() and staticize updateproc. Add a new private thread flag to indicate that the thread should not sleep if runningbufspace is too large. Set this flag on the bufdaemon and syncer threads so that they skip the waitrunningbufspace() call in bufwrite() rather than than checking the proc pointer vs. the known proc pointers for these two threads. A way of preventing these threads from being starved for I/O but still placing limits on their outstanding I/O would be desirable. Set this flag in ffs_copyonwrite() to prevent bufwrite() calls from blocking on the runningbufspace check while holding snaplk. This prevents snaplk from being held for an arbitrarily long period of time if runningbufspace is high and greatly reduces the contention for snaplk. The disadvantage is that ffs_copyonwrite() can start a large amount of I/O if there are a large number of snapshots, which could cause a deadlock in other parts of the code. Call runningbufwakeup() in ffs_copyonwrite() to decrement runningbufspace before attempting to grab snaplk so that I/O requests waiting on snaplk are not counted in runningbufspace as being in-progress. Increment runningbufspace again before actually launching the original I/O request. Prior to the above two changes, the system could deadlock if enough I/O requests were blocked by snaplk to prevent runningbufspace from falling below lorunningspace and one of the bawrite() calls in ffs_copyonwrite() blocked in waitrunningbufspace() while holding snaplk. See <http://www.holm.cc/stress/log/cons143.html>
2005-09-30 01:30:01 +00:00
td->td_pflags |= TDP_NORUNNINGBUF;
EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
2002-06-06 15:46:38 +00:00
SHUTDOWN_PRI_LAST);
mtx_lock(&sync_mtx);
for (;;) {
if (syncer_state == SYNCER_FINAL_DELAY &&
syncer_final_iter == 0) {
mtx_unlock(&sync_mtx);
kproc_suspend_check(td->td_proc);
mtx_lock(&sync_mtx);
}
net_worklist_len = syncer_worklist_len - sync_vnode_count;
2004-07-15 04:29:48 +00:00
if (syncer_state != SYNCER_RUNNING &&
starttime != time_uptime) {
2004-07-15 04:29:48 +00:00
if (first_printf) {
printf("\nSyncing disks, vnodes remaining...");
2004-07-15 04:29:48 +00:00
first_printf = 0;
}
printf("%d ", net_worklist_len);
2004-07-15 04:29:48 +00:00
}
starttime = time_uptime;
/*
* Push files whose dirty time has expired. Be careful
* of interrupt race on slp queue.
*
* Skip over empty worklist slots when shutting down.
*/
do {
slp = &syncer_workitem_pending[syncer_delayno];
syncer_delayno += 1;
if (syncer_delayno == syncer_maxdelay)
syncer_delayno = 0;
next = &syncer_workitem_pending[syncer_delayno];
/*
* If the worklist has wrapped since the
* it was emptied of all but syncer vnodes,
* switch to the FINAL_DELAY state and run
* for one more second.
*/
if (syncer_state == SYNCER_SHUTTING_DOWN &&
net_worklist_len == 0 &&
last_work_seen == syncer_delayno) {
syncer_state = SYNCER_FINAL_DELAY;
syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
}
} while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
syncer_worklist_len > 0);
/*
* Keep track of the last time there was anything
* on the worklist other than syncer vnodes.
* Return to the SHUTTING_DOWN state if any
* new work appears.
*/
if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
last_work_seen = syncer_delayno;
if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
syncer_state = SYNCER_SHUTTING_DOWN;
while (!LIST_EMPTY(slp)) {
error = sync_vnode(slp, &bo, td);
if (error == 1) {
LIST_REMOVE(bo, bo_synclist);
LIST_INSERT_HEAD(next, bo, bo_synclist);
continue;
}
if (first_printf == 0) {
/*
* Drop the sync mutex, because some watchdog
* drivers need to sleep while patting
*/
mtx_unlock(&sync_mtx);
wdog_kern_pat(WD_LASTVAL);
mtx_lock(&sync_mtx);
}
}
if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
syncer_final_iter--;
/*
* The variable rushjob allows the kernel to speed up the
* processing of the filesystem syncer process. A rushjob
* value of N tells the filesystem syncer to process the next
* N seconds worth of work on its queue ASAP. Currently rushjob
* is used by the soft update code to speed up the filesystem
* syncer process when the incore state is getting so far
* ahead of the disk that the kernel memory pool is being
* threatened with exhaustion.
*/
if (rushjob > 0) {
rushjob -= 1;
continue;
}
/*
* Just sleep for a short period of time between
* iterations when shutting down to allow some I/O
* to happen.
*
* If it has taken us less than a second to process the
* current work, then wait. Otherwise start right over
* again. We can still lose time if any single round
* takes more than two seconds, but it does not really
* matter as we are just trying to generally pace the
* filesystem activity.
*/
if (syncer_state != SYNCER_RUNNING ||
time_uptime == starttime) {
thread_lock(td);
sched_prio(td, PPAUSE);
thread_unlock(td);
}
if (syncer_state != SYNCER_RUNNING)
cv_timedwait(&sync_wakeup, &sync_mtx,
hz / SYNCER_SHUTDOWN_SPEEDUP);
else if (time_uptime == starttime)
cv_timedwait(&sync_wakeup, &sync_mtx, hz);
}
}
/*
* Request the syncer daemon to speed up its work.
* We never push it to speed up more than half of its
* normal turn time, otherwise it could take over the cpu.
*/
int
speedup_syncer(void)
{
int ret = 0;
mtx_lock(&sync_mtx);
if (rushjob < syncdelay / 2) {
rushjob += 1;
stat_rush_requests += 1;
ret = 1;
}
mtx_unlock(&sync_mtx);
cv_broadcast(&sync_wakeup);
return (ret);
}
/*
* Tell the syncer to speed up its work and run though its work
* list several times, then tell it to shut down.
*/
static void
syncer_shutdown(void *arg, int howto)
{
if (howto & RB_NOSYNC)
return;
mtx_lock(&sync_mtx);
syncer_state = SYNCER_SHUTTING_DOWN;
rushjob = 0;
mtx_unlock(&sync_mtx);
cv_broadcast(&sync_wakeup);
kproc_shutdown(arg, howto);
}
void
syncer_suspend(void)
{
syncer_shutdown(updateproc, 0);
}
void
syncer_resume(void)
{
mtx_lock(&sync_mtx);
first_printf = 1;
syncer_state = SYNCER_RUNNING;
mtx_unlock(&sync_mtx);
cv_broadcast(&sync_wakeup);
kproc_resume(updateproc);
}
1994-05-24 10:09:53 +00:00
/*
* Reassign a buffer from one vnode to another.
* Used to assign file specific control information
* (indirect blocks) to the vnode to which they belong.
*/
void
reassignbuf(struct buf *bp)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp;
struct bufobj *bo;
int delay;
#ifdef INVARIANTS
struct bufv *bv;
#endif
1994-05-24 10:09:53 +00:00
vp = bp->b_vp;
bo = bp->b_bufobj;
The buffer queue mechanism has been reformulated. Instead of having QUEUE_AGE, QUEUE_LRU, and QUEUE_EMPTY we instead have QUEUE_CLEAN, QUEUE_DIRTY, QUEUE_EMPTY, and QUEUE_EMPTYKVA. With this patch clean and dirty buffers have been separated. Empty buffers with KVM assignments have been separated from truely empty buffers. getnewbuf() has been rewritten and now operates in a 100% optimal fashion. That is, it is able to find precisely the right kind of buffer it needs to allocate a new buffer, defragment KVM, or to free-up an existing buffer when the buffer cache is full (which is a steady-state situation for the buffer cache). Buffer flushing has been reorganized. Previously buffers were flushed in the context of whatever process hit the conditions forcing buffer flushing to occur. This resulted in processes blocking on conditions unrelated to what they were doing. This also resulted in inappropriate VFS stacking chains due to multiple processes getting stuck trying to flush dirty buffers or due to a single process getting into a situation where it might attempt to flush buffers recursively - a situation that was only partially fixed in prior commits. We have added a new daemon called the buf_daemon which is responsible for flushing dirty buffers when the number of dirty buffers exceeds the vfs.hidirtybuffers limit. This daemon attempts to dynamically adjust the rate at which dirty buffers are flushed such that getnewbuf() calls (almost) never block. The number of nbufs and amount of buffer space is now scaled past the 8MB limit that was previously imposed for systems with over 64MB of memory, and the vfs.{lo,hi}dirtybuffers limits have been relaxed somewhat. The number of physical buffers has been increased with the intention that we will manage physical I/O differently in the future. reassignbuf previously attempted to keep the dirtyblkhd list sorted which could result in non-deterministic operation under certain conditions, such as when a large number of dirty buffers are being managed. This algorithm has been changed. reassignbuf now keeps buffers locally sorted if it can do so cheaply, and otherwise gives up and adds buffers to the head of the dirtyblkhd list. The new algorithm is deterministic but not perfect. The new algorithm greatly reduces problems that previously occured when write_behind was turned off in the system. The P_FLSINPROG proc->p_flag bit has been replaced by the more descriptive P_BUFEXHAUST bit. This bit allows processes working with filesystem buffers to use available emergency reserves. Normal processes do not set this bit and are not allowed to dig into emergency reserves. The purpose of this bit is to avoid low-memory deadlocks. A small race condition was fixed in getpbuf() in vm/vm_pager.c. Submitted by: Matthew Dillon <dillon@apollo.backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-04 00:25:38 +00:00
++reassignbufcalls;
CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
bp, bp->b_vp, bp->b_flags);
/*
* B_PAGING flagged buffers cannot be reassigned because their vp
* is not fully linked in.
*/
if (bp->b_flags & B_PAGING)
panic("cannot reassign paging buffer");
1994-05-24 10:09:53 +00:00
/*
* Delete from old vnode list, if on one.
*/
BO_LOCK(bo);
if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
buf_vlist_remove(bp);
else
panic("reassignbuf: Buffer %p not on queue.", bp);
1994-05-24 10:09:53 +00:00
/*
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
* If dirty, put on list of dirty buffers; otherwise insert onto list
* of clean buffers.
1994-05-24 10:09:53 +00:00
*/
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
if (bp->b_flags & B_DELWRI) {
if ((bo->bo_flag & BO_ONWORKLST) == 0) {
switch (vp->v_type) {
case VDIR:
delay = dirdelay;
break;
case VCHR:
delay = metadelay;
break;
default:
delay = filedelay;
}
vn_syncer_add_to_worklist(bo, delay);
}
buf_vlist_add(bp, bo, BX_VNDIRTY);
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
} else {
buf_vlist_add(bp, bo, BX_VNCLEAN);
if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
mtx_lock(&sync_mtx);
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
mtx_unlock(&sync_mtx);
bo->bo_flag &= ~BO_ONWORKLST;
}
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
}
#ifdef INVARIANTS
bv = &bo->bo_clean;
bp = TAILQ_FIRST(&bv->bv_hd);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bp = TAILQ_LAST(&bv->bv_hd, buflists);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bv = &bo->bo_dirty;
bp = TAILQ_FIRST(&bv->bv_hd);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
bp = TAILQ_LAST(&bv->bv_hd, buflists);
KASSERT(bp == NULL || bp->b_bufobj == bo,
("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
#endif
BO_UNLOCK(bo);
1994-05-24 10:09:53 +00:00
}
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* Increment the use and hold counts on the vnode, taking care to reference
* the driver's usecount if this is a chardev. The vholdl() will remove
* the vnode from the free list if it is presently free. Requires the
* vnode interlock and returns with it held.
*/
static void
v_incr_usecount(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vholdl(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vp->v_usecount++;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount++;
dev_unlock();
}
}
/*
* Turn a holdcnt into a use+holdcnt such that only one call to
* v_decr_usecount is needed.
*/
static void
v_upgrade_usecount(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
vp->v_usecount++;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount++;
dev_unlock();
}
}
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* Decrement the vnode use and hold count along with the driver's usecount
* if this is a chardev. The vdropl() below releases the vnode interlock
* as it may free the vnode.
*/
static void
v_decr_usecount(struct vnode *vp)
{
ASSERT_VI_LOCKED(vp, __FUNCTION__);
VNASSERT(vp->v_usecount > 0, vp,
("v_decr_usecount: negative usecount"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vp->v_usecount--;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
vp->v_rdev->si_usecount--;
dev_unlock();
}
vdropl(vp);
}
/*
* Decrement only the use count and driver use count. This is intended to
* be paired with a follow on vdropl() to release the remaining hold count.
* In this way we may vgone() a vnode with a 0 usecount without risk of
* having it end up on a free list because the hold count is kept above 0.
*/
static void
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
v_decr_useonly(struct vnode *vp)
{
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
ASSERT_VI_LOCKED(vp, __FUNCTION__);
VNASSERT(vp->v_usecount > 0, vp,
("v_decr_useonly: negative usecount"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vp->v_usecount--;
if (vp->v_type == VCHR && vp->v_rdev != NULL) {
dev_lock();
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vp->v_rdev->si_usecount--;
dev_unlock();
}
}
1994-05-24 10:09:53 +00:00
/*
* Grab a particular vnode from the free list, increment its
* reference count and lock it. VI_DOOMED is set if the vnode
* is being destroyed. Only callers who specify LK_RETRY will
* see doomed vnodes. If inactive processing was delayed in
* vput try to do it here.
1994-05-24 10:09:53 +00:00
*/
int
vget(struct vnode *vp, int flags, struct thread *td)
1994-05-24 10:09:53 +00:00
{
int error;
1994-05-24 10:09:53 +00:00
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
error = 0;
VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
("vget: invalid lock operation"));
CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
if ((flags & LK_INTERLOCK) == 0)
VI_LOCK(vp);
vholdl(vp);
if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
vdrop(vp);
CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
vp);
return (error);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
}
if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
panic("vget: vn_lock failed to return ENOENT\n");
VI_LOCK(vp);
/* Upgrade our holdcnt to a usecount. */
v_upgrade_usecount(vp);
/*
2010-04-03 11:19:20 +00:00
* We don't guarantee that any particular close will
* trigger inactive processing so just make a best effort
* here at preventing a reference to a removed file. If
* we don't succeed no harm is done.
*/
if (vp->v_iflag & VI_OWEINACT) {
2010-04-03 11:19:20 +00:00
if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
(flags & LK_NOWAIT) == 0)
vinactive(vp, td);
vp->v_iflag &= ~VI_OWEINACT;
}
VI_UNLOCK(vp);
return (0);
}
2002-06-06 15:46:38 +00:00
/*
2000-10-05 18:22:46 +00:00
* Increase the reference count of a vnode.
*/
void
vref(struct vnode *vp)
{
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
VI_LOCK(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
v_incr_usecount(vp);
VI_UNLOCK(vp);
}
/*
* Return reference count of a vnode.
*
* The results of this call are only guaranteed when some mechanism other
* than the VI lock is used to stop other processes from gaining references
* to the vnode. This may be the case if the caller holds the only reference.
* This is also useful when stale data is acceptable as race conditions may
* be accounted for by some other means.
*/
int
vrefcnt(struct vnode *vp)
{
int usecnt;
VI_LOCK(vp);
usecnt = vp->v_usecount;
VI_UNLOCK(vp);
return (usecnt);
}
#define VPUTX_VRELE 1
#define VPUTX_VPUT 2
#define VPUTX_VUNREF 3
static void
vputx(struct vnode *vp, int func)
1994-05-24 10:09:53 +00:00
{
int error;
KASSERT(vp != NULL, ("vputx: null vp"));
if (func == VPUTX_VUNREF)
ASSERT_VOP_LOCKED(vp, "vunref");
else if (func == VPUTX_VPUT)
ASSERT_VOP_LOCKED(vp, "vput");
else
KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
VI_LOCK(vp);
/* Skip this v_writecount check if we're going to panic below. */
2005-02-17 10:28:58 +00:00
VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
("vputx: missed vn_close"));
error = 0;
if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
vp->v_usecount == 1)) {
if (func == VPUTX_VPUT)
VOP_UNLOCK(vp, 0);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
v_decr_usecount(vp);
return;
}
if (vp->v_usecount != 1) {
vprint("vputx: negative ref count", vp);
panic("vputx: negative ref cnt");
}
CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* We want to hold the vnode until the inactive finishes to
* prevent vgone() races. We drop the use count here and the
* hold count below when we're done.
*/
v_decr_useonly(vp);
/*
* We must call VOP_INACTIVE with the node locked. Mark
* as VI_DOINGINACT to avoid recursion.
*/
vp->v_iflag |= VI_OWEINACT;
switch (func) {
case VPUTX_VRELE:
error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
VI_LOCK(vp);
break;
case VPUTX_VPUT:
if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
LK_NOWAIT);
VI_LOCK(vp);
}
break;
case VPUTX_VUNREF:
if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
VI_LOCK(vp);
}
break;
}
if (vp->v_usecount > 0)
vp->v_iflag &= ~VI_OWEINACT;
if (error == 0) {
if (vp->v_iflag & VI_OWEINACT)
vinactive(vp, curthread);
if (func != VPUTX_VUNREF)
VOP_UNLOCK(vp, 0);
}
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdropl(vp);
1994-05-24 10:09:53 +00:00
}
/*
* Vnode put/release.
* If count drops to zero, call inactive routine and return to freelist.
*/
void
vrele(struct vnode *vp)
{
vputx(vp, VPUTX_VRELE);
}
2002-06-06 15:46:38 +00:00
/*
2000-10-05 18:22:46 +00:00
* Release an already locked vnode. This give the same effects as
* unlock+vrele(), but takes less time and avoids releasing and
* re-aquiring the lock (as vrele() acquires the lock internally.)
*/
void
vput(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
vputx(vp, VPUTX_VPUT);
}
/*
* Release an exclusively locked vnode. Do not unlock the vnode lock.
*/
void
vunref(struct vnode *vp)
{
vputx(vp, VPUTX_VUNREF);
}
1994-05-24 10:09:53 +00:00
/*
* Somebody doesn't want the vnode recycled.
1994-05-24 10:09:53 +00:00
*/
void
vhold(struct vnode *vp)
{
VI_LOCK(vp);
vholdl(vp);
VI_UNLOCK(vp);
}
/*
* Increase the hold count and activate if this is the first reference.
*/
void
vholdl(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
struct mount *mp;
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
#ifdef INVARIANTS
/* getnewvnode() calls v_incr_usecount() without holding interlock. */
if (vp->v_type != VNON || vp->v_data != NULL) {
ASSERT_VI_LOCKED(vp, "vholdl");
VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0,
vp, ("vholdl: free vnode is held"));
}
#endif
1994-05-24 10:09:53 +00:00
vp->v_holdcnt++;
if ((vp->v_iflag & VI_FREE) == 0)
return;
VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
/*
* Remove a vnode from the free list, mark it as in use,
* and put it on the active list.
*/
mtx_lock(&vnode_free_list_mtx);
TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
freevnodes--;
vp->v_iflag &= ~(VI_FREE|VI_AGE);
KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
("Activating already active vnode"));
vp->v_iflag |= VI_ACTIVE;
mp = vp->v_mount;
TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
mp->mnt_activevnodelistsize++;
mtx_unlock(&vnode_free_list_mtx);
1994-05-24 10:09:53 +00:00
}
/*
* Note that there is one less who cares about this vnode.
* vdrop() is the opposite of vhold().
1994-05-24 10:09:53 +00:00
*/
void
vdrop(struct vnode *vp)
{
VI_LOCK(vp);
vdropl(vp);
}
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* Drop the hold count of the vnode. If this is the last reference to
* the vnode we place it on the free list unless it has been vgone'd
* (marked VI_DOOMED) in which case we will free it.
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
*/
void
vdropl(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
struct bufobj *bo;
struct mount *mp;
int active;
ASSERT_VI_LOCKED(vp, "vdropl");
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
1994-05-24 10:09:53 +00:00
if (vp->v_holdcnt <= 0)
panic("vdrop: holdcnt %d", vp->v_holdcnt);
1994-05-24 10:09:53 +00:00
vp->v_holdcnt--;
VNASSERT(vp->v_holdcnt >= vp->v_usecount, vp,
("hold count less than use count"));
if (vp->v_holdcnt > 0) {
VI_UNLOCK(vp);
return;
}
if ((vp->v_iflag & VI_DOOMED) == 0) {
/*
* Mark a vnode as free: remove it from its active list
* and put it up for recycling on the freelist.
*/
VNASSERT(vp->v_op != NULL, vp,
("vdropl: vnode already reclaimed."));
VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
("vnode already free"));
VNASSERT(vp->v_holdcnt == 0, vp,
("vdropl: freeing when we shouldn't"));
active = vp->v_iflag & VI_ACTIVE;
vp->v_iflag &= ~VI_ACTIVE;
mp = vp->v_mount;
mtx_lock(&vnode_free_list_mtx);
if (active) {
TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
v_actfreelist);
mp->mnt_activevnodelistsize--;
}
if (vp->v_iflag & VI_AGE) {
TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
} else {
TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
}
freevnodes++;
vp->v_iflag &= ~VI_AGE;
vp->v_iflag |= VI_FREE;
mtx_unlock(&vnode_free_list_mtx);
VI_UNLOCK(vp);
return;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
}
/*
* The vnode has been marked for destruction, so free it.
*/
CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
atomic_subtract_long(&numvnodes, 1);
bo = &vp->v_bufobj;
VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
("cleaned vnode still on the free list."));
VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
("clean blk trie not empty"));
VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
("dirty blk trie not empty"));
VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
VI_UNLOCK(vp);
#ifdef MAC
mac_vnode_destroy(vp);
#endif
if (vp->v_pollinfo != NULL)
destroy_vpollinfo(vp->v_pollinfo);
#ifdef INVARIANTS
/* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
vp->v_op = NULL;
#endif
rangelock_destroy(&vp->v_rl);
lockdestroy(vp->v_vnlock);
mtx_destroy(&vp->v_interlock);
rw_destroy(BO_LOCKPTR(bo));
uma_zfree(vnode_zone, vp);
1994-05-24 10:09:53 +00:00
}
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
/*
* Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
* flags. DOINGINACT prevents us from recursing in calls to vinactive.
* OWEINACT tracks whether a vnode missed a call to inactive due to a
* failed lock upgrade.
*/
void
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vinactive(struct vnode *vp, struct thread *td)
{
struct vm_object *obj;
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
ASSERT_VOP_ELOCKED(vp, "vinactive");
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
ASSERT_VI_LOCKED(vp, "vinactive");
VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
("vinactive: recursed on VI_DOINGINACT"));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vp->v_iflag |= VI_DOINGINACT;
vp->v_iflag &= ~VI_OWEINACT;
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
VI_UNLOCK(vp);
/*
* Before moving off the active list, we must be sure that any
* modified pages are on the vnode's dirty list since these will
* no longer be checked once the vnode is on the inactive list.
* Because the vnode vm object keeps a hold reference on the vnode
* if there is at least one resident non-cached page, the vnode
* cannot leave the active list without the page cleanup done.
*/
obj = vp->v_object;
if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
}
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
VOP_INACTIVE(vp, td);
VI_LOCK(vp);
VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
("vinactive: lost VI_DOINGINACT"));
vp->v_iflag &= ~VI_DOINGINACT;
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
}
1994-05-24 10:09:53 +00:00
/*
* Remove any vnodes in the vnode table belonging to mount point mp.
*
* If FORCECLOSE is not specified, there should not be any active ones,
1994-05-24 10:09:53 +00:00
* return error if any are found (nb: this is a user error, not a
* system error). If FORCECLOSE is specified, detach any active vnodes
1994-05-24 10:09:53 +00:00
* that are found.
*
* If WRITECLOSE is set, only flush out regular file vnodes open for
* writing.
*
* SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
*
* `rootrefs' specifies the base reference count for the root vnode
* of this filesystem. The root vnode is considered busy if its
* v_usecount exceeds this value. On a successful return, vflush(, td)
* will call vrele() on the root vnode exactly rootrefs times.
* If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
* be zero.
1994-05-24 10:09:53 +00:00
*/
#ifdef DIAGNOSTIC
static int busyprt = 0; /* print out busy vnodes */
SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
1994-05-24 10:09:53 +00:00
#endif
int
2010-04-03 11:19:20 +00:00
vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp, *mvp, *rootvp = NULL;
struct vattr vattr;
int busy = 0, error;
1994-05-24 10:09:53 +00:00
CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
rootrefs, flags);
if (rootrefs > 0) {
KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
("vflush: bad args"));
/*
* Get the filesystem root vnode. We can vput() it
* immediately, since with rootrefs > 0, it won't go away.
*/
if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
__func__, error);
return (error);
}
vput(rootvp);
}
1994-05-24 10:09:53 +00:00
loop:
MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vholdl(vp);
error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
if (error) {
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdrop(vp);
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
goto loop;
}
1994-05-24 10:09:53 +00:00
/*
* Skip over a vnodes marked VV_SYSTEM.
1994-05-24 10:09:53 +00:00
*/
if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
VOP_UNLOCK(vp, 0);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdrop(vp);
1994-05-24 10:09:53 +00:00
continue;
}
1994-05-24 10:09:53 +00:00
/*
* If WRITECLOSE is set, flush out unlinked but still open
* files (even if open only for reading) and regular file
2002-06-06 15:46:38 +00:00
* vnodes open for writing.
1994-05-24 10:09:53 +00:00
*/
if (flags & WRITECLOSE) {
if (vp->v_object != NULL) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(vp->v_object);
vm_object_page_clean(vp->v_object, 0, 0, 0);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(vp->v_object);
}
error = VOP_FSYNC(vp, MNT_WAIT, td);
if (error != 0) {
VOP_UNLOCK(vp, 0);
vdrop(vp);
MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
return (error);
}
error = VOP_GETATTR(vp, &vattr, td->td_ucred);
VI_LOCK(vp);
if ((vp->v_type == VNON ||
(error == 0 && vattr.va_nlink > 0)) &&
(vp->v_writecount == 0 || vp->v_type != VREG)) {
VOP_UNLOCK(vp, 0);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdropl(vp);
continue;
}
} else
VI_LOCK(vp);
1994-05-24 10:09:53 +00:00
/*
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
* With v_usecount == 0, all we need to do is clear out the
* vnode data structures and we are done.
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
*
* If FORCECLOSE is set, forcibly close the vnode.
1994-05-24 10:09:53 +00:00
*/
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
VNASSERT(vp->v_usecount == 0 ||
vp->v_op != &devfs_specops ||
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
(vp->v_type != VCHR && vp->v_type != VBLK), vp,
("device VNODE %p is FORCECLOSED", vp));
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vgonel(vp);
} else {
busy++;
1994-05-24 10:09:53 +00:00
#ifdef DIAGNOSTIC
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
if (busyprt)
vprint("vflush: busy vnode", vp);
1994-05-24 10:09:53 +00:00
#endif
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
}
VOP_UNLOCK(vp, 0);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vdropl(vp);
1994-05-24 10:09:53 +00:00
}
if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
/*
* If just the root vnode is busy, and if its refcount
* is equal to `rootrefs', then go ahead and kill it.
*/
VI_LOCK(rootvp);
KASSERT(busy > 0, ("vflush: not busy"));
2005-02-17 10:28:58 +00:00
VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
("vflush: usecount %d < rootrefs %d",
rootvp->v_usecount, rootrefs));
if (busy == 1 && rootvp->v_usecount == rootrefs) {
VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vgone(rootvp);
VOP_UNLOCK(rootvp, 0);
busy = 0;
} else
VI_UNLOCK(rootvp);
}
if (busy) {
CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
busy);
1994-05-24 10:09:53 +00:00
return (EBUSY);
}
for (; rootrefs > 0; rootrefs--)
vrele(rootvp);
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* Recycle an unused vnode to the front of the free list.
*/
int
vrecycle(struct vnode *vp)
{
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
int recycled;
ASSERT_VOP_ELOCKED(vp, "vrecycle");
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
recycled = 0;
VI_LOCK(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
if (vp->v_usecount == 0) {
recycled = 1;
vgonel(vp);
}
VI_UNLOCK(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
return (recycled);
}
1994-05-24 10:09:53 +00:00
/*
* Eliminate all activity associated with a vnode
* in preparation for reuse.
1994-05-24 10:09:53 +00:00
*/
void
vgone(struct vnode *vp)
{
VI_LOCK(vp);
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vgonel(vp);
VI_UNLOCK(vp);
}
static void
notify_lowervp_vfs_dummy(struct mount *mp __unused,
struct vnode *lowervp __unused)
{
}
/*
* Notify upper mounts about reclaimed or unlinked vnode.
*/
void
vfs_notify_upper(struct vnode *vp, int event)
{
static struct vfsops vgonel_vfsops = {
.vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
.vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
};
struct mount *mp, *ump, *mmp;
mp = vp->v_mount;
if (mp == NULL)
return;
MNT_ILOCK(mp);
if (TAILQ_EMPTY(&mp->mnt_uppers))
goto unlock;
MNT_IUNLOCK(mp);
mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
mmp->mnt_op = &vgonel_vfsops;
mmp->mnt_kern_flag |= MNTK_MARKER;
MNT_ILOCK(mp);
mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
ump = TAILQ_NEXT(ump, mnt_upper_link);
continue;
}
TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
MNT_IUNLOCK(mp);
switch (event) {
case VFS_NOTIFY_UPPER_RECLAIM:
VFS_RECLAIM_LOWERVP(ump, vp);
break;
case VFS_NOTIFY_UPPER_UNLINK:
VFS_UNLINK_LOWERVP(ump, vp);
break;
default:
KASSERT(0, ("invalid event %d", event));
break;
}
MNT_ILOCK(mp);
ump = TAILQ_NEXT(mmp, mnt_upper_link);
TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
}
free(mmp, M_TEMP);
mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
wakeup(&mp->mnt_uppers);
}
unlock:
MNT_IUNLOCK(mp);
}
/*
* vgone, with the vp interlock held.
*/
void
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
vgonel(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
struct thread *td;
int oweinact;
int active;
struct mount *mp;
1994-05-24 10:09:53 +00:00
ASSERT_VOP_ELOCKED(vp, "vgonel");
ASSERT_VI_LOCKED(vp, "vgonel");
- Change holdcnt use around vnode recycling. We now always keep a holdcnt ref while we're calling vgone(). This prevents transient refs from re-adding us to the free list. Previously, a vfree() triggered via vinvalbuf() getting rid of all of a vnode's pages could place a partially destructed vnode on the free list where vtryrecycle() could find it. The first call to vtryrecycle would hang up on the vnode lock, but when it failed it would place a now dead vnode onto the free list, and another call to vtryrecycle() would free an already free vnode. There were many complications of having a zero ref count while freeing which can now go away. - Change vdropl() to release the interlock before returning. All callers now respect this, so vdropl() directly frees VI_DOOMED vnodes once the last ref is dropped. This means that we'll never have VI_DOOMED vnodes on the free list. - Seperate v_incr_usecount() into v_incr_usecount(), v_decr_usecount() and v_decr_useonly(). The incr/decr split is so that incr usecount can return with the interlock still held while decr drops the interlock so it can call vdropl() which will potentially free the vnode. The calling function can't drop the lock of an already free'd node. v_decr_useonly() drops a usecount without droping the hold count. This is done so the usecount reaches zero in vput() before we recycle, however the holdcount is still 1 which prevents any new references from placing the vnode back on the free list. - Fix vnlrureclaim() to vhold the vnode since it doesn't do a vget(). We wouldn't want vnlrureclaim() to bump the usecount since this has different semantics. Also change vnlrureclaim() to do a NOWAIT on the vn_lock. When this function runs we're usually in a desperate situation and we wouldn't want to wait for any specific vnode to be released. - Fix a bunch of misc comments to reflect the new behavior. - Add vhold() and vdrop() to vflush() for the same reasons that we do in vlrureclaim(). Previously we held no reference and a vnode could have been freed while we were waiting on the lock. - Get rid of vlruvp() and vfreehead(). Neither are used. vlruvp() should really be rethought before it's reintroduced. - vgonel() always returns with the vnode locked now and never puts the vnode back on a free list. The vnode will be freed as soon as the last reference is released. Sponsored by: Isilon Systems, Inc. Debugging help from: Kris Kennaway, Peter Holm Approved by: re (blanket vfs)
2005-06-16 04:41:42 +00:00
VNASSERT(vp->v_holdcnt, vp,
("vgonel: vp %p has no reference.", vp));
CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
td = curthread;
/*
* Don't vgonel if we're already doomed.
*/
if (vp->v_iflag & VI_DOOMED)
return;
vp->v_iflag |= VI_DOOMED;
1994-05-24 10:09:53 +00:00
/*
* Check to see if the vnode is in use. If so, we have to call
* VOP_CLOSE() and VOP_INACTIVE().
1994-05-24 10:09:53 +00:00
*/
active = vp->v_usecount;
oweinact = (vp->v_iflag & VI_OWEINACT);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
VI_UNLOCK(vp);
vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
1994-05-24 10:09:53 +00:00
/*
* If purging an active vnode, it must be closed and
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
* deactivated before being reclaimed.
1994-05-24 10:09:53 +00:00
*/
if (active)
VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
if (oweinact || active) {
VI_LOCK(vp);
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
if ((vp->v_iflag & VI_DOINGINACT) == 0)
vinactive(vp, td);
VI_UNLOCK(vp);
1994-05-24 10:09:53 +00:00
}
if (vp->v_type == VSOCK)
vfs_unp_reclaim(vp);
/*
* Clean out any buffers associated with the vnode.
* If the flush fails, just toss the buffers.
*/
mp = NULL;
if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
(void) vn_start_secondary_write(vp, &mp, V_WAIT);
if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
while (vinvalbuf(vp, 0, 0, 0) != 0)
;
}
#ifdef INVARIANTS
BO_LOCK(&vp->v_bufobj);
KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
vp->v_bufobj.bo_clean.bv_cnt == 0,
("vp %p bufobj not invalidated", vp));
vp->v_bufobj.bo_flag |= BO_DEAD;
BO_UNLOCK(&vp->v_bufobj);
#endif
1994-05-24 10:09:53 +00:00
/*
* Reclaim the vnode.
*/
if (VOP_RECLAIM(vp, td))
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
panic("vgone: cannot reclaim");
if (mp != NULL)
vn_finished_secondary_write(mp);
2005-02-17 10:28:58 +00:00
VNASSERT(vp->v_object == NULL, vp,
("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
/*
* Clear the advisory locks and wake up waiting threads.
*/
(void)VOP_ADVLOCKPURGE(vp);
/*
* Delete from old mount point vnode list.
*/
When we traverse the vnodes on a mountpoint we need to look out for our cached 'next vnode' being removed from this mountpoint. If we find that it was recycled, we restart our traversal from the start of the list. Code to do that is in all local disk filesystems (and a few other places) and looks roughly like this: MNT_ILOCK(mp); loop: for (vp = TAILQ_FIRST(&mp...); (vp = nvp) != NULL; nvp = TAILQ_NEXT(vp,...)) { if (vp->v_mount != mp) goto loop; MNT_IUNLOCK(mp); ... MNT_ILOCK(mp); } MNT_IUNLOCK(mp); The code which takes vnodes off a mountpoint looks like this: MNT_ILOCK(vp->v_mount); ... TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); ... MNT_IUNLOCK(vp->v_mount); ... vp->v_mount = something; (Take a moment and try to spot the locking error before you read on.) On a SMP system, one CPU could have removed nvp from our mountlist but not yet gotten to assign a new value to vp->v_mount while another CPU simultaneously get to the top of the traversal loop where it finds that (vp->v_mount != mp) is not true despite the fact that the vnode has indeed been removed from our mountpoint. Fix: Introduce the macro MNT_VNODE_FOREACH() to traverse the list of vnodes on a mountpoint while taking into account that vnodes may be removed from the list as we go. This saves approx 65 lines of duplicated code. Split the insmntque() which potentially moves a vnode from one mount point to another into delmntque() and insmntque() which does just what the names say. Fix delmntque() to set vp->v_mount to NULL while holding the mountpoint lock.
2004-07-04 08:52:35 +00:00
delmntque(vp);
cache_purge(vp);
1994-05-24 10:09:53 +00:00
/*
* Done with purge, reset to the standard lock and invalidate
* the vnode.
1994-05-24 10:09:53 +00:00
*/
VI_LOCK(vp);
vp->v_vnlock = &vp->v_lock;
vp->v_op = &dead_vnodeops;
vp->v_tag = "none";
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vp->v_type = VBAD;
1994-05-24 10:09:53 +00:00
}
/*
* Calculate the total number of references to a special device.
*/
int
vcount(struct vnode *vp)
1994-05-24 10:09:53 +00:00
{
int count;
dev_lock();
count = vp->v_rdev->si_usecount;
dev_unlock();
1994-05-24 10:09:53 +00:00
return (count);
}
/*
* Same as above, but using the struct cdev *as argument
*/
int
count_dev(struct cdev *dev)
{
2003-10-17 11:56:48 +00:00
int count;
dev_lock();
2003-10-17 11:56:48 +00:00
count = dev->si_usecount;
dev_unlock();
2003-10-17 11:56:48 +00:00
return(count);
}
1994-05-24 10:09:53 +00:00
/*
* Print out a description of a vnode.
*/
static char *typename[] =
{"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
"VMARKER"};
1994-05-24 10:09:53 +00:00
void
vn_printf(struct vnode *vp, const char *fmt, ...)
1994-05-24 10:09:53 +00:00
{
va_list ap;
char buf[256], buf2[16];
u_long flags;
1994-05-24 10:09:53 +00:00
va_start(ap, fmt);
vprintf(fmt, ap);
va_end(ap);
printf("%p: ", (void *)vp);
printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n",
vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
1994-05-24 10:09:53 +00:00
buf[0] = '\0';
buf[1] = '\0';
if (vp->v_vflag & VV_ROOT)
strlcat(buf, "|VV_ROOT", sizeof(buf));
if (vp->v_vflag & VV_ISTTY)
strlcat(buf, "|VV_ISTTY", sizeof(buf));
if (vp->v_vflag & VV_NOSYNC)
strlcat(buf, "|VV_NOSYNC", sizeof(buf));
if (vp->v_vflag & VV_ETERNALDEV)
strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
if (vp->v_vflag & VV_CACHEDLABEL)
strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
if (vp->v_vflag & VV_TEXT)
strlcat(buf, "|VV_TEXT", sizeof(buf));
if (vp->v_vflag & VV_COPYONWRITE)
strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
if (vp->v_vflag & VV_SYSTEM)
strlcat(buf, "|VV_SYSTEM", sizeof(buf));
if (vp->v_vflag & VV_PROCDEP)
strlcat(buf, "|VV_PROCDEP", sizeof(buf));
if (vp->v_vflag & VV_NOKNOTE)
strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
if (vp->v_vflag & VV_DELETED)
strlcat(buf, "|VV_DELETED", sizeof(buf));
if (vp->v_vflag & VV_MD)
strlcat(buf, "|VV_MD", sizeof(buf));
if (vp->v_vflag & VV_FORCEINSMQ)
strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
if (flags != 0) {
snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
strlcat(buf, buf2, sizeof(buf));
}
if (vp->v_iflag & VI_MOUNT)
strlcat(buf, "|VI_MOUNT", sizeof(buf));
if (vp->v_iflag & VI_AGE)
strlcat(buf, "|VI_AGE", sizeof(buf));
if (vp->v_iflag & VI_DOOMED)
strlcat(buf, "|VI_DOOMED", sizeof(buf));
if (vp->v_iflag & VI_FREE)
strlcat(buf, "|VI_FREE", sizeof(buf));
if (vp->v_iflag & VI_ACTIVE)
strlcat(buf, "|VI_ACTIVE", sizeof(buf));
if (vp->v_iflag & VI_DOINGINACT)
strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
if (vp->v_iflag & VI_OWEINACT)
strlcat(buf, "|VI_OWEINACT", sizeof(buf));
flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
if (flags != 0) {
snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
strlcat(buf, buf2, sizeof(buf));
}
printf(" flags (%s)\n", buf + 1);
if (mtx_owned(VI_MTX(vp)))
printf(" VI_LOCKed");
if (vp->v_object != NULL)
printf(" v_object %p ref %d pages %d "
"cleanbuf %d dirtybuf %d\n",
- Remove vx_lock, vx_unlock, vx_wait, etc. - Add a vn_start_write/vn_finished_write around vlrureclaim so we don't do writing ops without suspending. This could suspend the vlruproc which should not be a problem under normal circumstances. - Manually implement VMIGHTFREE in vlrureclaim as this was the only instance where it was used. - Acquire a lock before calling vgone() as it now requires it. - Move the acquisition of the vnode interlock from vtryrecycle() to getnewvnode() so that if it fails we don't drop and reacquire the vnode_free_list_mtx. - Check for a usecount or holdcount at the end of vtryrecycle() in case someone grabbed a ref while we were recycling. Abort the recycle, and on the final ref drop this vnode will be placed on the head of the free list. - Move the redundant VOP_INACTIVE protection code into the local vinactive() routine to avoid code bloat. - Keep the vnode lock held across calls to vgone() in several places. - vgonel() no longer uses XLOCK, instead callers must hold an exclusive vnode lock. The VI_DOOMED flag is set to allow other threads to detect a vnode which is no longer valid. This flag is set until the last reference is gone, and there are no chances for a new ref. vgonel() holds this lock across the entire function, which greatly simplifies logic. _ Only vfree() in one place in vgone() not three. - Adjust vget() to check the VI_DOOMED flag prior to waiting on the lock in the LK_NOWAIT case. In other cases, check after we have slept and acquired an exlusive lock. This will simulate the old vx_wait() behavior. Sponsored by: Isilon Systems, Inc.
2005-03-13 11:54:28 +00:00
vp->v_object, vp->v_object->ref_count,
vp->v_object->resident_page_count,
vp->v_bufobj.bo_dirty.bv_cnt,
vp->v_bufobj.bo_clean.bv_cnt);
printf(" ");
lockmgr_printinfo(vp->v_vnlock);
if (vp->v_data != NULL)
1994-05-24 10:09:53 +00:00
VOP_PRINT(vp);
}
#ifdef DDB
1994-05-24 10:09:53 +00:00
/*
* List all of the locked vnodes in the system.
* Called when debugging the kernel.
*/
DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
1994-05-24 10:09:53 +00:00
{
struct mount *mp;
struct vnode *vp;
1994-05-24 10:09:53 +00:00
/*
* Note: because this is DDB, we can't obey the locking semantics
* for these structures, which means we could catch an inconsistent
* state and dereference a nasty pointer. Not much to be done
* about that.
*/
db_printf("Locked vnodes\n");
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
2005-02-10 08:55:08 +00:00
vprint("", vp);
}
1994-05-24 10:09:53 +00:00
}
}
2006-09-04 22:15:44 +00:00
/*
* Show details about the given vnode.
*/
DB_SHOW_COMMAND(vnode, db_show_vnode)
{
struct vnode *vp;
if (!have_addr)
return;
vp = (struct vnode *)addr;
vn_printf(vp, "vnode ");
}
/*
* Show details about the given mount point.
*/
DB_SHOW_COMMAND(mount, db_show_mount)
{
struct mount *mp;
struct vfsopt *opt;
struct statfs *sp;
struct vnode *vp;
char buf[512];
uint64_t mflags;
u_int flags;
if (!have_addr) {
/* No address given, print short info about all mount points. */
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
db_printf("%p %s on %s (%s)\n", mp,
mp->mnt_stat.f_mntfromname,
mp->mnt_stat.f_mntonname,
mp->mnt_stat.f_fstypename);
if (db_pager_quit)
break;
}
db_printf("\nMore info: show mount <addr>\n");
return;
}
mp = (struct mount *)addr;
db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
buf[0] = '\0';
mflags = mp->mnt_flag;
#define MNT_FLAG(flag) do { \
if (mflags & (flag)) { \
if (buf[0] != '\0') \
strlcat(buf, ", ", sizeof(buf)); \
strlcat(buf, (#flag) + 4, sizeof(buf)); \
mflags &= ~(flag); \
} \
} while (0)
MNT_FLAG(MNT_RDONLY);
MNT_FLAG(MNT_SYNCHRONOUS);
MNT_FLAG(MNT_NOEXEC);
MNT_FLAG(MNT_NOSUID);
MNT_FLAG(MNT_NFS4ACLS);
MNT_FLAG(MNT_UNION);
MNT_FLAG(MNT_ASYNC);
MNT_FLAG(MNT_SUIDDIR);
MNT_FLAG(MNT_SOFTDEP);
MNT_FLAG(MNT_NOSYMFOLLOW);
MNT_FLAG(MNT_GJOURNAL);
MNT_FLAG(MNT_MULTILABEL);
MNT_FLAG(MNT_ACLS);
MNT_FLAG(MNT_NOATIME);
MNT_FLAG(MNT_NOCLUSTERR);
MNT_FLAG(MNT_NOCLUSTERW);
MNT_FLAG(MNT_SUJ);
MNT_FLAG(MNT_EXRDONLY);
MNT_FLAG(MNT_EXPORTED);
MNT_FLAG(MNT_DEFEXPORTED);
MNT_FLAG(MNT_EXPORTANON);
MNT_FLAG(MNT_EXKERB);
MNT_FLAG(MNT_EXPUBLIC);
MNT_FLAG(MNT_LOCAL);
MNT_FLAG(MNT_QUOTA);
MNT_FLAG(MNT_ROOTFS);
MNT_FLAG(MNT_USER);
MNT_FLAG(MNT_IGNORE);
MNT_FLAG(MNT_UPDATE);
MNT_FLAG(MNT_DELEXPORT);
MNT_FLAG(MNT_RELOAD);
MNT_FLAG(MNT_FORCE);
MNT_FLAG(MNT_SNAPSHOT);
MNT_FLAG(MNT_BYFSID);
#undef MNT_FLAG
if (mflags != 0) {
if (buf[0] != '\0')
strlcat(buf, ", ", sizeof(buf));
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
"0x%016jx", mflags);
}
db_printf(" mnt_flag = %s\n", buf);
buf[0] = '\0';
flags = mp->mnt_kern_flag;
#define MNT_KERN_FLAG(flag) do { \
if (flags & (flag)) { \
if (buf[0] != '\0') \
strlcat(buf, ", ", sizeof(buf)); \
strlcat(buf, (#flag) + 5, sizeof(buf)); \
flags &= ~(flag); \
} \
} while (0)
MNT_KERN_FLAG(MNTK_UNMOUNTF);
MNT_KERN_FLAG(MNTK_ASYNC);
MNT_KERN_FLAG(MNTK_SOFTDEP);
MNT_KERN_FLAG(MNTK_NOINSMNTQ);
MNT_KERN_FLAG(MNTK_DRAINING);
MNT_KERN_FLAG(MNTK_REFEXPIRE);
MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
MNT_KERN_FLAG(MNTK_SHARED_WRITES);
MNT_KERN_FLAG(MNTK_NO_IOPF);
MNT_KERN_FLAG(MNTK_VGONE_UPPER);
MNT_KERN_FLAG(MNTK_VGONE_WAITER);
MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
MNT_KERN_FLAG(MNTK_MARKER);
MNT_KERN_FLAG(MNTK_USES_BCACHE);
MNT_KERN_FLAG(MNTK_NOASYNC);
MNT_KERN_FLAG(MNTK_UNMOUNT);
MNT_KERN_FLAG(MNTK_MWAIT);
MNT_KERN_FLAG(MNTK_SUSPEND);
MNT_KERN_FLAG(MNTK_SUSPEND2);
MNT_KERN_FLAG(MNTK_SUSPENDED);
MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
MNT_KERN_FLAG(MNTK_NOKNOTE);
#undef MNT_KERN_FLAG
if (flags != 0) {
if (buf[0] != '\0')
strlcat(buf, ", ", sizeof(buf));
snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
"0x%08x", flags);
}
db_printf(" mnt_kern_flag = %s\n", buf);
db_printf(" mnt_opt = ");
opt = TAILQ_FIRST(mp->mnt_opt);
if (opt != NULL) {
db_printf("%s", opt->name);
opt = TAILQ_NEXT(opt, link);
while (opt != NULL) {
db_printf(", %s", opt->name);
opt = TAILQ_NEXT(opt, link);
}
}
db_printf("\n");
sp = &mp->mnt_stat;
db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx "
"bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
"ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
"asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
(u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
(uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
(uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
(intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
(intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
(uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
(uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
(u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
db_printf(" mnt_cred = { uid=%u ruid=%u",
(u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
if (jailed(mp->mnt_cred))
db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
db_printf(" }\n");
db_printf(" mnt_ref = %d\n", mp->mnt_ref);
db_printf(" mnt_gen = %d\n", mp->mnt_gen);
db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
db_printf(" mnt_activevnodelistsize = %d\n",
mp->mnt_activevnodelistsize);
db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount);
db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max);
db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed);
db_printf(" mnt_lockref = %d\n", mp->mnt_lockref);
db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
db_printf(" mnt_secondary_accwrites = %d\n",
mp->mnt_secondary_accwrites);
db_printf(" mnt_gjprovider = %s\n",
mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
db_printf("\n\nList of active vnodes\n");
TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
if (vp->v_type != VMARKER) {
vn_printf(vp, "vnode ");
if (db_pager_quit)
break;
}
}
db_printf("\n\nList of inactive vnodes\n");
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
vn_printf(vp, "vnode ");
if (db_pager_quit)
break;
}
}
}
2006-09-04 22:15:44 +00:00
#endif /* DDB */
1994-05-24 10:09:53 +00:00
/*
* Fill in a struct xvfsconf based on a struct vfsconf.
*/
static int
vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
{
struct xvfsconf xvfsp;
bzero(&xvfsp, sizeof(xvfsp));
strcpy(xvfsp.vfc_name, vfsp->vfc_name);
xvfsp.vfc_typenum = vfsp->vfc_typenum;
xvfsp.vfc_refcount = vfsp->vfc_refcount;
xvfsp.vfc_flags = vfsp->vfc_flags;
/*
* These are unused in userland, we keep them
* to not break binary compatibility.
*/
xvfsp.vfc_vfsops = NULL;
xvfsp.vfc_next = NULL;
return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
}
#ifdef COMPAT_FREEBSD32
struct xvfsconf32 {
uint32_t vfc_vfsops;
char vfc_name[MFSNAMELEN];
int32_t vfc_typenum;
int32_t vfc_refcount;
int32_t vfc_flags;
uint32_t vfc_next;
};
static int
vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
{
struct xvfsconf32 xvfsp;
strcpy(xvfsp.vfc_name, vfsp->vfc_name);
xvfsp.vfc_typenum = vfsp->vfc_typenum;
xvfsp.vfc_refcount = vfsp->vfc_refcount;
xvfsp.vfc_flags = vfsp->vfc_flags;
xvfsp.vfc_vfsops = 0;
xvfsp.vfc_next = 0;
return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
}
#endif
/*
* Top level filesystem related information gathering.
*/
static int
sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
{
struct vfsconf *vfsp;
int error;
error = 0;
vfsconf_slock();
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
#ifdef COMPAT_FREEBSD32
if (req->flags & SCTL_MASK32)
error = vfsconf2x32(req, vfsp);
else
#endif
error = vfsconf2x(req, vfsp);
if (error)
break;
}
vfsconf_sunlock();
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
"S,xvfsconf", "List of all configured filesystems");
#ifndef BURN_BRIDGES
2002-03-19 21:25:46 +00:00
static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
static int
vfs_sysctl(SYSCTL_HANDLER_ARGS)
{
int *name = (int *)arg1 - 1; /* XXX */
u_int namelen = arg2 + 1; /* XXX */
struct vfsconf *vfsp;
log(LOG_WARNING, "userland calling deprecated sysctl, "
"please rebuild world\n");
#if 1 || defined(COMPAT_PRELITE2)
/* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
if (namelen == 1)
return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
#endif
switch (name[1]) {
case VFS_MAXTYPENUM:
if (namelen != 2)
return (ENOTDIR);
return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
case VFS_CONF:
if (namelen != 3)
return (ENOTDIR); /* overloaded */
vfsconf_slock();
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
if (vfsp->vfc_typenum == name[2])
break;
}
vfsconf_sunlock();
if (vfsp == NULL)
return (EOPNOTSUPP);
#ifdef COMPAT_FREEBSD32
if (req->flags & SCTL_MASK32)
return (vfsconf2x32(req, vfsp));
else
#endif
return (vfsconf2x(req, vfsp));
}
return (EOPNOTSUPP);
}
static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
CTLFLAG_MPSAFE, vfs_sysctl,
"Generic filesystem");
#if 1 || defined(COMPAT_PRELITE2)
static int
sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
{
int error;
struct vfsconf *vfsp;
struct ovfsconf ovfs;
vfsconf_slock();
TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
bzero(&ovfs, sizeof(ovfs));
ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */
strcpy(ovfs.vfc_name, vfsp->vfc_name);
ovfs.vfc_index = vfsp->vfc_typenum;
ovfs.vfc_refcount = vfsp->vfc_refcount;
ovfs.vfc_flags = vfsp->vfc_flags;
error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
if (error != 0) {
vfsconf_sunlock();
return (error);
}
}
vfsconf_sunlock();
return (0);
}
#endif /* 1 || COMPAT_PRELITE2 */
#endif /* !BURN_BRIDGES */
#define KINFO_VNODESLOP 10
#ifdef notyet
1994-05-24 10:09:53 +00:00
/*
* Dump vnode list (via sysctl).
*/
/* ARGSUSED */
static int
sysctl_vnode(SYSCTL_HANDLER_ARGS)
1994-05-24 10:09:53 +00:00
{
struct xvnode *xvn;
struct mount *mp;
struct vnode *vp;
int error, len, n;
/*
* Stale numvnodes access is not fatal here.
*/
req->lock = 0;
len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
if (!req->oldptr)
/* Make an estimate */
return (SYSCTL_OUT(req, 0, len));
These changes embody the support of the fully coherent merged VM buffer cache, much higher filesystem I/O performance, and much better paging performance. It represents the culmination of over 6 months of R&D. The majority of the merged VM/cache work is by John Dyson. The following highlights the most significant changes. Additionally, there are (mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to support the new VM/buffer scheme. vfs_bio.c: Significant rewrite of most of vfs_bio to support the merged VM buffer cache scheme. The scheme is almost fully compatible with the old filesystem interface. Significant improvement in the number of opportunities for write clustering. vfs_cluster.c, vfs_subr.c Upgrade and performance enhancements in vfs layer code to support merged VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff. vm_object.c: Yet more improvements in the collapse code. Elimination of some windows that can cause list corruption. vm_pageout.c: Fixed it, it really works better now. Somehow in 2.0, some "enhancements" broke the code. This code has been reworked from the ground-up. vm_fault.c, vm_page.c, pmap.c, vm_object.c Support for small-block filesystems with merged VM/buffer cache scheme. pmap.c vm_map.c Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of kernel PTs. vm_glue.c Much simpler and more effective swapping code. No more gratuitous swapping. proc.h Fixed the problem that the p_lock flag was not being cleared on a fork. swap_pager.c, vnode_pager.c Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the code doesn't need it anymore. machdep.c Changes to better support the parameter values for the merged VM/buffer cache scheme. machdep.c, kern_exec.c, vm_glue.c Implemented a seperate submap for temporary exec string space and another one to contain process upages. This eliminates all map fragmentation problems that previously existed. ffs_inode.c, ufs_inode.c, ufs_readwrite.c Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on busy buffers. Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
n = 0;
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(&mountlist_mtx);
TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2008-11-02 10:15:42 +00:00
if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
1994-05-24 10:09:53 +00:00
continue;
MNT_ILOCK(mp);
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
if (n == len)
break;
vref(vp);
xvn[n].xv_size = sizeof *xvn;
xvn[n].xv_vnode = vp;
xvn[n].xv_id = 0; /* XXX compat */
#define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
XV_COPY(usecount);
XV_COPY(writecount);
XV_COPY(holdcnt);
XV_COPY(mount);
XV_COPY(numoutput);
XV_COPY(type);
#undef XV_COPY
xvn[n].xv_flag = vp->v_vflag;
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
break;
case VBLK:
case VCHR:
if (vp->v_rdev == NULL) {
vrele(vp);
continue;
}
xvn[n].xv_dev = dev2udev(vp->v_rdev);
break;
case VSOCK:
xvn[n].xv_socket = vp->v_socket;
break;
case VFIFO:
xvn[n].xv_fifo = vp->v_fifoinfo;
break;
case VNON:
case VBAD:
default:
/* shouldn't happen? */
vrele(vp);
continue;
}
vrele(vp);
++n;
1994-05-24 10:09:53 +00:00
}
MNT_IUNLOCK(mp);
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_lock(&mountlist_mtx);
vfs_unbusy(mp);
if (n == len)
break;
1994-05-24 10:09:53 +00:00
}
Change and clean the mutex lock interface. mtx_enter(lock, type) becomes: mtx_lock(lock) for sleep locks (MTX_DEF-initialized locks) mtx_lock_spin(lock) for spin locks (MTX_SPIN-initialized) similarily, for releasing a lock, we now have: mtx_unlock(lock) for MTX_DEF and mtx_unlock_spin(lock) for MTX_SPIN. We change the caller interface for the two different types of locks because the semantics are entirely different for each case, and this makes it explicitly clear and, at the same time, it rids us of the extra `type' argument. The enter->lock and exit->unlock change has been made with the idea that we're "locking data" and not "entering locked code" in mind. Further, remove all additional "flags" previously passed to the lock acquire/release routines with the exception of two: MTX_QUIET and MTX_NOSWITCH The functionality of these flags is preserved and they can be passed to the lock/unlock routines by calling the corresponding wrappers: mtx_{lock, unlock}_flags(lock, flag(s)) and mtx_{lock, unlock}_spin_flags(lock, flag(s)) for MTX_DEF and MTX_SPIN locks, respectively. Re-inline some lock acq/rel code; in the sleep lock case, we only inline the _obtain_lock()s in order to ensure that the inlined code fits into a cache line. In the spin lock case, we inline recursion and actually only perform a function call if we need to spin. This change has been made with the idea that we generally tend to avoid spin locks and that also the spin locks that we do have and are heavily used (i.e. sched_lock) do recurse, and therefore in an effort to reduce function call overhead for some architectures (such as alpha), we inline recursion for this case. Create a new malloc type for the witness code and retire from using the M_DEV type. The new type is called M_WITNESS and is only declared if WITNESS is enabled. Begin cleaning up some machdep/mutex.h code - specifically updated the "optimized" inlined code in alpha/mutex.h and wrote MTX_LOCK_SPIN and MTX_UNLOCK_SPIN asm macros for the i386/mutex.h as we presently need those. Finally, caught up to the interface changes in all sys code. Contributors: jake, jhb, jasone (in no particular order)
2001-02-09 06:11:45 +00:00
mtx_unlock(&mountlist_mtx);
1994-05-24 10:09:53 +00:00
error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
free(xvn, M_TEMP);
return (error);
1994-05-24 10:09:53 +00:00
}
SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
"");
#endif
/*
* Unmount all filesystems. The list is traversed in reverse order
* of mounting to avoid dependencies.
*/
void
vfs_unmountall(void)
{
struct mount *mp;
struct thread *td;
int error;
CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
td = curthread;
/*
* Since this only runs when rebooting, it is not interlocked.
*/
while(!TAILQ_EMPTY(&mountlist)) {
mp = TAILQ_LAST(&mountlist, mntlist);
vfs_ref(mp);
error = dounmount(mp, MNT_FORCE, td);
if (error != 0) {
TAILQ_REMOVE(&mountlist, mp, mnt_list);
/*
* XXX: Due to the way in which we mount the root
* file system off of devfs, devfs will generate a
* "busy" warning when we try to unmount it before
* the root. Don't print a warning as a result in
* order to avoid false positive errors that may
* cause needless upset.
*/
if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
printf("unmount of %s failed (",
mp->mnt_stat.f_mntonname);
if (error == EBUSY)
printf("BUSY)\n");
else
printf("%d)\n", error);
}
} else {
/* The unmount has removed mp from the mountlist */
}
}
1994-05-24 10:09:53 +00:00
}
/*
* perform msync on all vnodes under a mount point
* the mount point must be locked.
*/
void
vfs_msync(struct mount *mp, int flags)
{
struct vnode *vp, *mvp;
struct vm_object *obj;
CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
obj = vp->v_object;
if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
(flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
if (!vget(vp,
LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
curthread)) {
if (vp->v_vflag & VV_NOSYNC) { /* unlinked */
vput(vp);
continue;
}
obj = vp->v_object;
if (obj != NULL) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(obj);
vm_object_page_clean(obj, 0, 0,
flags == MNT_WAIT ?
OBJPC_SYNC : OBJPC_NOSYNC);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(obj);
}
vput(vp);
}
} else
VI_UNLOCK(vp);
}
}
static void
destroy_vpollinfo_free(struct vpollinfo *vi)
{
knlist_destroy(&vi->vpi_selinfo.si_note);
mtx_destroy(&vi->vpi_lock);
uma_zfree(vnodepoll_zone, vi);
}
static void
destroy_vpollinfo(struct vpollinfo *vi)
{
knlist_clear(&vi->vpi_selinfo.si_note, 1);
seldrain(&vi->vpi_selinfo);
destroy_vpollinfo_free(vi);
}
/*
* Initalize per-vnode helper structure to hold poll-related state.
*/
void
v_addpollinfo(struct vnode *vp)
{
struct vpollinfo *vi;
if (vp->v_pollinfo != NULL)
return;
vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
VI_LOCK(vp);
if (vp->v_pollinfo != NULL) {
VI_UNLOCK(vp);
destroy_vpollinfo_free(vi);
return;
}
vp->v_pollinfo = vi;
VI_UNLOCK(vp);
}
/*
* Record a process's interest in events which might happen to
* a vnode. Because poll uses the historic select-style interface
* internally, this routine serves as both the ``check for any
* pending events'' and the ``record my interest in future events''
* functions. (These are done together, while the lock is held,
* to avoid race conditions.)
*/
int
vn_pollrecord(struct vnode *vp, struct thread *td, int events)
{
v_addpollinfo(vp);
mtx_lock(&vp->v_pollinfo->vpi_lock);
if (vp->v_pollinfo->vpi_revents & events) {
/*
* This leaves events we are not interested
* in available for the other process which
* which presumably had requested them
* (otherwise they would never have been
* recorded).
*/
events &= vp->v_pollinfo->vpi_revents;
vp->v_pollinfo->vpi_revents &= ~events;
mtx_unlock(&vp->v_pollinfo->vpi_lock);
return (events);
}
vp->v_pollinfo->vpi_events |= events;
selrecord(td, &vp->v_pollinfo->vpi_selinfo);
mtx_unlock(&vp->v_pollinfo->vpi_lock);
return (0);
}
/*
* Routine to create and manage a filesystem syncer vnode.
*/
2002-03-19 21:25:46 +00:00
#define sync_close ((int (*)(struct vop_close_args *))nullop)
static int sync_fsync(struct vop_fsync_args *);
static int sync_inactive(struct vop_inactive_args *);
static int sync_reclaim(struct vop_reclaim_args *);
static struct vop_vector sync_vnodeops = {
.vop_bypass = VOP_EOPNOTSUPP,
.vop_close = sync_close, /* close */
.vop_fsync = sync_fsync, /* fsync */
.vop_inactive = sync_inactive, /* inactive */
.vop_reclaim = sync_reclaim, /* reclaim */
.vop_lock1 = vop_stdlock, /* lock */
.vop_unlock = vop_stdunlock, /* unlock */
.vop_islocked = vop_stdislocked, /* islocked */
};
/*
* Create a new filesystem syncer vnode for the specified mount point.
*/
void
vfs_allocate_syncvnode(struct mount *mp)
{
struct vnode *vp;
struct bufobj *bo;
static long start, incr, next;
int error;
/* Allocate a new vnode */
error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
if (error != 0)
panic("vfs_allocate_syncvnode: getnewvnode() failed");
vp->v_type = VNON;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vp->v_vflag |= VV_FORCEINSMQ;
error = insmntque(vp, mp);
if (error != 0)
panic("vfs_allocate_syncvnode: insmntque() failed");
vp->v_vflag &= ~VV_FORCEINSMQ;
VOP_UNLOCK(vp, 0);
/*
* Place the vnode onto the syncer worklist. We attempt to
* scatter them about on the list so that they will go off
* at evenly distributed times even if all the filesystems
* are mounted at once.
*/
next += incr;
if (next == 0 || next > syncer_maxdelay) {
start /= 2;
incr /= 2;
if (start == 0) {
start = syncer_maxdelay / 2;
incr = syncer_maxdelay;
}
next = start;
}
bo = &vp->v_bufobj;
BO_LOCK(bo);
vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
/* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
mtx_lock(&sync_mtx);
sync_vnode_count++;
if (mp->mnt_syncer == NULL) {
mp->mnt_syncer = vp;
vp = NULL;
}
mtx_unlock(&sync_mtx);
BO_UNLOCK(bo);
if (vp != NULL) {
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
vgone(vp);
vput(vp);
}
}
void
vfs_deallocate_syncvnode(struct mount *mp)
{
struct vnode *vp;
mtx_lock(&sync_mtx);
vp = mp->mnt_syncer;
if (vp != NULL)
mp->mnt_syncer = NULL;
mtx_unlock(&sync_mtx);
if (vp != NULL)
vrele(vp);
}
/*
* Do a lazy sync of the filesystem.
*/
1998-12-21 23:38:33 +00:00
static int
sync_fsync(struct vop_fsync_args *ap)
{
struct vnode *syncvp = ap->a_vp;
struct mount *mp = syncvp->v_mount;
int error, save;
struct bufobj *bo;
/*
* We only need to do something if this is a lazy evaluation.
*/
if (ap->a_waitfor != MNT_LAZY)
return (0);
/*
* Move ourselves to the back of the sync list.
*/
bo = &syncvp->v_bufobj;
BO_LOCK(bo);
vn_syncer_add_to_worklist(bo, syncdelay);
BO_UNLOCK(bo);
/*
* Walk the list of vnodes pushing all that are dirty and
* not already on the sync list.
*/
if (vfs_busy(mp, MBF_NOWAIT) != 0)
return (0);
if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
vfs_unbusy(mp);
return (0);
}
save = curthread_pflags_set(TDP_SYNCIO);
vfs_msync(mp, MNT_NOWAIT);
error = VFS_SYNC(mp, MNT_LAZY);
curthread_pflags_restore(save);
vn_finished_write(mp);
vfs_unbusy(mp);
return (error);
}
/*
* The syncer vnode is no referenced.
*/
1998-12-21 23:38:33 +00:00
static int
sync_inactive(struct vop_inactive_args *ap)
{
vgone(ap->a_vp);
return (0);
}
/*
* The syncer vnode is no longer needed and is being decommissioned.
*
* Modifications to the worklist must be protected by sync_mtx.
*/
1998-12-21 23:38:33 +00:00
static int
sync_reclaim(struct vop_reclaim_args *ap)
{
struct vnode *vp = ap->a_vp;
struct bufobj *bo;
bo = &vp->v_bufobj;
BO_LOCK(bo);
mtx_lock(&sync_mtx);
if (vp->v_mount->mnt_syncer == vp)
vp->v_mount->mnt_syncer = NULL;
if (bo->bo_flag & BO_ONWORKLST) {
LIST_REMOVE(bo, bo_synclist);
syncer_worklist_len--;
sync_vnode_count--;
bo->bo_flag &= ~BO_ONWORKLST;
}
mtx_unlock(&sync_mtx);
BO_UNLOCK(bo);
return (0);
}
/*
* Check if vnode represents a disk device
*/
int
vn_isdisk(struct vnode *vp, int *errp)
{
2003-10-12 14:04:39 +00:00
int error;
if (vp->v_type != VCHR) {
error = ENOTBLK;
goto out;
}
2003-10-12 14:04:39 +00:00
error = 0;
dev_lock();
if (vp->v_rdev == NULL)
2003-10-12 14:04:39 +00:00
error = ENXIO;
else if (vp->v_rdev->si_devsw == NULL)
error = ENXIO;
else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
2003-10-12 14:04:39 +00:00
error = ENOTBLK;
dev_unlock();
out:
if (errp != NULL)
2003-10-12 14:04:39 +00:00
*errp = error;
return (error == 0);
}
/*
2002-05-16 21:28:32 +00:00
* Common filesystem object access control check routine. Accepts a
* vnode's type, "mode", uid and gid, requested access mode, credentials,
* and optional call-by-reference privused argument allowing vaccess()
* to indicate to the caller whether privilege was used to satisfy the
* request (obsoleted). Returns 0 on success, or an errno on failure.
*/
int
vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
accmode_t accmode, struct ucred *cred, int *privused)
{
accmode_t dac_granted;
accmode_t priv_granted;
KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
("invalid bit in accmode"));
KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
2010-04-03 11:19:20 +00:00
("VAPPEND without VWRITE"));
/*
* Look for a normal, non-privileged way to access the file/directory
* as requested. If it exists, go with that.
*/
if (privused != NULL)
*privused = 0;
dac_granted = 0;
/* Check the owner. */
if (cred->cr_uid == file_uid) {
dac_granted |= VADMIN;
if (file_mode & S_IXUSR)
dac_granted |= VEXEC;
if (file_mode & S_IRUSR)
dac_granted |= VREAD;
if (file_mode & S_IWUSR)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
goto privcheck;
}
/* Otherwise, check the groups (first match) */
if (groupmember(file_gid, cred)) {
if (file_mode & S_IXGRP)
dac_granted |= VEXEC;
if (file_mode & S_IRGRP)
dac_granted |= VREAD;
if (file_mode & S_IWGRP)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
goto privcheck;
}
/* Otherwise, check everyone else. */
if (file_mode & S_IXOTH)
dac_granted |= VEXEC;
if (file_mode & S_IROTH)
dac_granted |= VREAD;
if (file_mode & S_IWOTH)
dac_granted |= (VWRITE | VAPPEND);
if ((accmode & dac_granted) == accmode)
return (0);
privcheck:
/*
* Build a privilege mask to determine if the set of privileges
* satisfies the requirements when combined with the granted mask
* from above. For each privilege, if the privilege is required,
* bitwise or the request type onto the priv_granted mask.
*/
priv_granted = 0;
if (type == VDIR) {
/*
* For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
* requests, instead of PRIV_VFS_EXEC.
*/
if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
!priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
priv_granted |= VEXEC;
} else {
/*
* Ensure that at least one execute bit is on. Otherwise,
* a privileged user will always succeed, and we don't want
* this to happen unless the file really is executable.
*/
if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
(file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
!priv_check_cred(cred, PRIV_VFS_EXEC, 0))
priv_granted |= VEXEC;
}
if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
!priv_check_cred(cred, PRIV_VFS_READ, 0))
priv_granted |= VREAD;
if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
!priv_check_cred(cred, PRIV_VFS_WRITE, 0))
priv_granted |= (VWRITE | VAPPEND);
if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
!priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
priv_granted |= VADMIN;
if ((accmode & (priv_granted | dac_granted)) == accmode) {
/* XXX audit: privilege used */
if (privused != NULL)
*privused = 1;
return (0);
}
return ((accmode & VADMIN) ? EPERM : EACCES);
}
/*
* Credential check based on process requesting service, and per-attribute
* permissions.
*/
int
extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
struct thread *td, accmode_t accmode)
{
/*
* Kernel-invoked always succeeds.
*/
if (cred == NOCRED)
return (0);
/*
* Do not allow privileged processes in jail to directly manipulate
* system attributes.
*/
switch (attrnamespace) {
case EXTATTR_NAMESPACE_SYSTEM:
/* Potentially should be: return (EPERM); */
return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
case EXTATTR_NAMESPACE_USER:
return (VOP_ACCESS(vp, accmode, cred, td));
default:
return (EPERM);
}
}
#ifdef DEBUG_VFS_LOCKS
/*
* This only exists to supress warnings from unlocked specfs accesses. It is
* no longer ok to have an unlocked VFS.
*/
#define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \
(vp)->v_type == VCHR || (vp)->v_type == VBAD)
int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */
2010-11-14 16:10:15 +00:00
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
"Drop into debugger on lock violation");
2004-01-05 23:40:46 +00:00
int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */
2010-11-14 16:10:15 +00:00
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
0, "Check for interlock across VOPs");
int vfs_badlock_print = 1; /* Print lock violations. */
2010-11-14 16:10:15 +00:00
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
0, "Print lock violations");
#ifdef KDB
int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */
2010-11-14 16:10:15 +00:00
SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
&vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
#endif
static void
vfs_badlock(const char *msg, const char *str, struct vnode *vp)
{
#ifdef KDB
if (vfs_badlock_backtrace)
kdb_backtrace();
#endif
if (vfs_badlock_print)
printf("%s: %p %s\n", str, (void *)vp, msg);
if (vfs_badlock_ddb)
kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
}
void
assert_vi_locked(struct vnode *vp, const char *str)
{
if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
vfs_badlock("interlock is not locked but should be", str, vp);
}
void
assert_vi_unlocked(struct vnode *vp, const char *str)
{
if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
vfs_badlock("interlock is locked but should not be", str, vp);
}
void
assert_vop_locked(struct vnode *vp, const char *str)
{
int locked;
if (!IGNORE_LOCK(vp)) {
locked = VOP_ISLOCKED(vp);
if (locked == 0 || locked == LK_EXCLOTHER)
vfs_badlock("is not locked but should be", str, vp);
}
}
void
assert_vop_unlocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
vfs_badlock("is locked but should not be", str, vp);
}
void
assert_vop_elocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
vfs_badlock("is not exclusive locked but should be", str, vp);
}
#if 0
void
assert_vop_elocked_other(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
vfs_badlock("is not exclusive locked by another thread",
str, vp);
}
void
assert_vop_slocked(struct vnode *vp, const char *str)
{
if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
vfs_badlock("is not locked shared but should be", str, vp);
}
#endif /* 0 */
#endif /* DEBUG_VFS_LOCKS */
void
vop_rename_fail(struct vop_rename_args *ap)
{
if (ap->a_tvp != NULL)
vput(ap->a_tvp);
if (ap->a_tdvp == ap->a_tvp)
vrele(ap->a_tdvp);
else
vput(ap->a_tdvp);
vrele(ap->a_fdvp);
vrele(ap->a_fvp);
}
void
vop_rename_pre(void *ap)
{
struct vop_rename_args *a = ap;
#ifdef DEBUG_VFS_LOCKS
if (a->a_tvp)
ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
/* Check the source (from). */
if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
(a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
/* Check the target. */
if (a->a_tvp)
ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
#endif
if (a->a_tdvp != a->a_fdvp)
vhold(a->a_fdvp);
if (a->a_tvp != a->a_fvp)
vhold(a->a_fvp);
vhold(a->a_tdvp);
if (a->a_tvp)
vhold(a->a_tvp);
}
void
vop_strategy_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
2004-01-05 23:40:46 +00:00
struct vop_strategy_args *a;
struct buf *bp;
2004-01-05 23:40:46 +00:00
a = ap;
bp = a->a_bp;
/*
* Cluster ops lock their component buffers but not the IO container.
*/
if ((bp->b_flags & B_CLUSTER) != 0)
return;
if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
if (vfs_badlock_print)
printf(
2004-01-05 23:40:46 +00:00
"VOP_STRATEGY: bp is not locked but should be\n");
if (vfs_badlock_ddb)
kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
}
#endif
}
void
vop_lock_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lock1_args *a = ap;
if ((a->a_flags & LK_INTERLOCK) == 0)
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
else
ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
#endif
}
void
vop_lock_post(void *ap, int rc)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_lock1_args *a = ap;
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
#endif
}
void
vop_unlock_pre(void *ap)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_unlock_args *a = ap;
if (a->a_flags & LK_INTERLOCK)
ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
#endif
}
void
vop_unlock_post(void *ap, int rc)
{
#ifdef DEBUG_VFS_LOCKS
struct vop_unlock_args *a = ap;
if (a->a_flags & LK_INTERLOCK)
ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
#endif
}
void
vop_create_post(void *ap, int rc)
{
struct vop_create_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
void
vop_deleteextattr_post(void *ap, int rc)
{
struct vop_deleteextattr_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
}
void
vop_link_post(void *ap, int rc)
{
struct vop_link_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
}
}
void
vop_mkdir_post(void *ap, int rc)
{
struct vop_mkdir_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
}
void
vop_mknod_post(void *ap, int rc)
{
struct vop_mknod_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
void
vop_remove_post(void *ap, int rc)
{
struct vop_remove_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
}
}
void
vop_rename_post(void *ap, int rc)
{
struct vop_rename_args *a = ap;
if (!rc) {
VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
if (a->a_tvp)
VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
}
if (a->a_tdvp != a->a_fdvp)
vdrop(a->a_fdvp);
if (a->a_tvp != a->a_fvp)
vdrop(a->a_fvp);
vdrop(a->a_tdvp);
if (a->a_tvp)
vdrop(a->a_tvp);
}
void
vop_rmdir_post(void *ap, int rc)
{
struct vop_rmdir_args *a = ap;
if (!rc) {
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
}
}
void
vop_setattr_post(void *ap, int rc)
{
struct vop_setattr_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
}
void
vop_setextattr_post(void *ap, int rc)
{
struct vop_setextattr_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
}
void
vop_symlink_post(void *ap, int rc)
{
struct vop_symlink_args *a = ap;
if (!rc)
VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
}
static struct knlist fs_knlist;
static void
vfs_event_init(void *arg)
{
knlist_init_mtx(&fs_knlist, NULL);
}
/* XXX - correct order? */
SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
void
vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
{
KNOTE_UNLOCKED(&fs_knlist, event);
}
static int filt_fsattach(struct knote *kn);
static void filt_fsdetach(struct knote *kn);
static int filt_fsevent(struct knote *kn, long hint);
struct filterops fs_filtops = {
.f_isfd = 0,
.f_attach = filt_fsattach,
.f_detach = filt_fsdetach,
.f_event = filt_fsevent
};
static int
filt_fsattach(struct knote *kn)
{
kn->kn_flags |= EV_CLEAR;
knlist_add(&fs_knlist, kn, 0);
return (0);
}
static void
filt_fsdetach(struct knote *kn)
{
knlist_remove(&fs_knlist, kn, 0);
}
static int
filt_fsevent(struct knote *kn, long hint)
{
kn->kn_fflags |= hint;
return (kn->kn_fflags != 0);
}
static int
sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
{
struct vfsidctl vc;
int error;
struct mount *mp;
error = SYSCTL_IN(req, &vc, sizeof(vc));
if (error)
return (error);
if (vc.vc_vers != VFS_CTL_VERS1)
return (EINVAL);
mp = vfs_getvfs(&vc.vc_fsid);
if (mp == NULL)
return (ENOENT);
/* ensure that a specific sysctl goes to the right filesystem. */
if (strcmp(vc.vc_fstypename, "*") != 0 &&
strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
vfs_rel(mp);
return (EINVAL);
}
VCTLTOREQ(&vc, req);
error = VFS_SYSCTL(mp, vc.vc_op, req);
vfs_rel(mp);
return (error);
}
SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
NULL, 0, sysctl_vfs_ctl, "",
"Sysctl by fsid");
/*
* Function to initialize a va_filerev field sensibly.
* XXX: Wouldn't a random number make a lot more sense ??
*/
u_quad_t
init_va_filerev(void)
{
struct bintime bt;
getbinuptime(&bt);
return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
}
static int filt_vfsread(struct knote *kn, long hint);
static int filt_vfswrite(struct knote *kn, long hint);
static int filt_vfsvnode(struct knote *kn, long hint);
static void filt_vfsdetach(struct knote *kn);
static struct filterops vfsread_filtops = {
.f_isfd = 1,
.f_detach = filt_vfsdetach,
.f_event = filt_vfsread
};
static struct filterops vfswrite_filtops = {
.f_isfd = 1,
.f_detach = filt_vfsdetach,
.f_event = filt_vfswrite
};
static struct filterops vfsvnode_filtops = {
.f_isfd = 1,
.f_detach = filt_vfsdetach,
.f_event = filt_vfsvnode
};
static void
vfs_knllock(void *arg)
{
struct vnode *vp = arg;
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
}
static void
vfs_knlunlock(void *arg)
{
struct vnode *vp = arg;
VOP_UNLOCK(vp, 0);
}
static void
vfs_knl_assert_locked(void *arg)
{
#ifdef DEBUG_VFS_LOCKS
struct vnode *vp = arg;
ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
#endif
}
static void
vfs_knl_assert_unlocked(void *arg)
{
#ifdef DEBUG_VFS_LOCKS
struct vnode *vp = arg;
ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
#endif
}
int
vfs_kqfilter(struct vop_kqfilter_args *ap)
{
struct vnode *vp = ap->a_vp;
struct knote *kn = ap->a_kn;
struct knlist *knl;
switch (kn->kn_filter) {
case EVFILT_READ:
kn->kn_fop = &vfsread_filtops;
break;
case EVFILT_WRITE:
kn->kn_fop = &vfswrite_filtops;
break;
case EVFILT_VNODE:
kn->kn_fop = &vfsvnode_filtops;
break;
default:
return (EINVAL);
}
kn->kn_hook = (caddr_t)vp;
v_addpollinfo(vp);
if (vp->v_pollinfo == NULL)
return (ENOMEM);
knl = &vp->v_pollinfo->vpi_selinfo.si_note;
vhold(vp);
knlist_add(knl, kn, 0);
return (0);
}
/*
* Detach knote from vnode
*/
static void
filt_vfsdetach(struct knote *kn)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
vdrop(vp);
}
/*ARGSUSED*/
static int
filt_vfsread(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
struct vattr va;
int res;
/*
* filesystem is gone, so set the EOF flag and schedule
* the knote for deletion.
*/
if (hint == NOTE_REVOKE) {
VI_LOCK(vp);
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
VI_UNLOCK(vp);
return (1);
}
if (VOP_GETATTR(vp, &va, curthread->td_ucred))
return (0);
VI_LOCK(vp);
kn->kn_data = va.va_size - kn->kn_fp->f_offset;
res = (kn->kn_data != 0);
VI_UNLOCK(vp);
return (res);
}
/*ARGSUSED*/
static int
filt_vfswrite(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
VI_LOCK(vp);
/*
* filesystem is gone, so set the EOF flag and schedule
* the knote for deletion.
*/
if (hint == NOTE_REVOKE)
kn->kn_flags |= (EV_EOF | EV_ONESHOT);
kn->kn_data = 0;
VI_UNLOCK(vp);
return (1);
}
static int
filt_vfsvnode(struct knote *kn, long hint)
{
struct vnode *vp = (struct vnode *)kn->kn_hook;
int res;
VI_LOCK(vp);
if (kn->kn_sfflags & hint)
kn->kn_fflags |= hint;
if (hint == NOTE_REVOKE) {
kn->kn_flags |= EV_EOF;
VI_UNLOCK(vp);
return (1);
}
res = (kn->kn_fflags != 0);
VI_UNLOCK(vp);
return (res);
}
int
vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
{
int error;
if (dp->d_reclen > ap->a_uio->uio_resid)
return (ENAMETOOLONG);
error = uiomove(dp, dp->d_reclen, ap->a_uio);
if (error) {
if (ap->a_ncookies != NULL) {
if (ap->a_cookies != NULL)
free(ap->a_cookies, M_TEMP);
ap->a_cookies = NULL;
*ap->a_ncookies = 0;
}
return (error);
}
if (ap->a_ncookies == NULL)
return (0);
KASSERT(ap->a_cookies,
("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
*ap->a_cookies = realloc(*ap->a_cookies,
(*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
(*ap->a_cookies)[*ap->a_ncookies] = off;
return (0);
}
/*
* Mark for update the access time of the file if the filesystem
* supports VOP_MARKATIME. This functionality is used by execve and
* mmap, so we want to avoid the I/O implied by directly setting
* va_atime for the sake of efficiency.
*/
void
vfs_mark_atime(struct vnode *vp, struct ucred *cred)
{
struct mount *mp;
mp = vp->v_mount;
ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
(void)VOP_MARKATIME(vp);
}
/*
* The purpose of this routine is to remove granularity from accmode_t,
* reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
* VADMIN and VAPPEND.
*
* If it returns 0, the caller is supposed to continue with the usual
* access checks using 'accmode' as modified by this routine. If it
* returns nonzero value, the caller is supposed to return that value
* as errno.
*
* Note that after this routine runs, accmode may be zero.
*/
int
vfs_unixify_accmode(accmode_t *accmode)
{
/*
* There is no way to specify explicit "deny" rule using
* file mode or POSIX.1e ACLs.
*/
if (*accmode & VEXPLICIT_DENY) {
*accmode = 0;
return (0);
}
/*
* None of these can be translated into usual access bits.
* Also, the common case for NFSv4 ACLs is to not contain
* either of these bits. Caller should check for VWRITE
* on the containing directory instead.
*/
if (*accmode & (VDELETE_CHILD | VDELETE))
return (EPERM);
if (*accmode & VADMIN_PERMS) {
*accmode &= ~VADMIN_PERMS;
*accmode |= VADMIN;
}
/*
* There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
* or VSYNCHRONIZE using file mode or POSIX.1e ACL.
*/
*accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
return (0);
}
/*
* These are helper functions for filesystems to traverse all
* their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
*
* This interface replaces MNT_VNODE_FOREACH.
*/
MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
struct vnode *
__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
{
struct vnode *vp;
if (should_yield())
kern_yield(PRI_USER);
MNT_ILOCK(mp);
KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
while (vp != NULL && (vp->v_type == VMARKER ||
(vp->v_iflag & VI_DOOMED) != 0))
vp = TAILQ_NEXT(vp, v_nmntvnodes);
/* Check if we are done */
if (vp == NULL) {
__mnt_vnode_markerfree_all(mvp, mp);
/* MNT_IUNLOCK(mp); -- done in above function */
mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
return (NULL);
}
TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
VI_LOCK(vp);
MNT_IUNLOCK(mp);
return (vp);
}
struct vnode *
__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
{
struct vnode *vp;
*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
MNT_ILOCK(mp);
MNT_REF(mp);
(*mvp)->v_type = VMARKER;
vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
while (vp != NULL && (vp->v_type == VMARKER ||
(vp->v_iflag & VI_DOOMED) != 0))
vp = TAILQ_NEXT(vp, v_nmntvnodes);
/* Check if we are done */
if (vp == NULL) {
MNT_REL(mp);
MNT_IUNLOCK(mp);
free(*mvp, M_VNODE_MARKER);
*mvp = NULL;
return (NULL);
}
(*mvp)->v_mount = mp;
TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
VI_LOCK(vp);
MNT_IUNLOCK(mp);
return (vp);
}
void
__mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
{
if (*mvp == NULL) {
MNT_IUNLOCK(mp);
return;
}
mtx_assert(MNT_MTX(mp), MA_OWNED);
KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
MNT_REL(mp);
MNT_IUNLOCK(mp);
free(*mvp, M_VNODE_MARKER);
*mvp = NULL;
}
/*
* These are helper functions for filesystems to traverse their
* active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
*/
static void
mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
{
KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
MNT_ILOCK(mp);
MNT_REL(mp);
MNT_IUNLOCK(mp);
free(*mvp, M_VNODE_MARKER);
*mvp = NULL;
}
static struct vnode *
mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
{
struct vnode *vp, *nvp;
mtx_assert(&vnode_free_list_mtx, MA_OWNED);
KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
restart:
vp = TAILQ_NEXT(*mvp, v_actfreelist);
TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
while (vp != NULL) {
if (vp->v_type == VMARKER) {
vp = TAILQ_NEXT(vp, v_actfreelist);
continue;
}
if (!VI_TRYLOCK(vp)) {
if (mp_ncpus == 1 || should_yield()) {
TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
mtx_unlock(&vnode_free_list_mtx);
pause("vnacti", 1);
mtx_lock(&vnode_free_list_mtx);
goto restart;
}
continue;
}
KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
("alien vnode on the active list %p %p", vp, mp));
if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
break;
nvp = TAILQ_NEXT(vp, v_actfreelist);
VI_UNLOCK(vp);
vp = nvp;
}
/* Check if we are done */
if (vp == NULL) {
mtx_unlock(&vnode_free_list_mtx);
mnt_vnode_markerfree_active(mvp, mp);
return (NULL);
}
TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
mtx_unlock(&vnode_free_list_mtx);
ASSERT_VI_LOCKED(vp, "active iter");
KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
return (vp);
}
struct vnode *
__mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
{
if (should_yield())
kern_yield(PRI_USER);
mtx_lock(&vnode_free_list_mtx);
return (mnt_vnode_next_active(mvp, mp));
}
struct vnode *
__mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
{
struct vnode *vp;
*mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
MNT_ILOCK(mp);
MNT_REF(mp);
MNT_IUNLOCK(mp);
(*mvp)->v_type = VMARKER;
(*mvp)->v_mount = mp;
mtx_lock(&vnode_free_list_mtx);
vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
if (vp == NULL) {
mtx_unlock(&vnode_free_list_mtx);
mnt_vnode_markerfree_active(mvp, mp);
return (NULL);
}
TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
return (mnt_vnode_next_active(mvp, mp));
}
void
__mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
{
if (*mvp == NULL)
return;
mtx_lock(&vnode_free_list_mtx);
TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
mtx_unlock(&vnode_free_list_mtx);
mnt_vnode_markerfree_active(mvp, mp);
}