freebsd-dev/sys/kern/vfs_cluster.c

1059 lines
28 KiB
C
Raw Normal View History

1994-05-24 10:09:53 +00:00
/*-
* Copyright (c) 1993
* The Regents of the University of California. All rights reserved.
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
* Modifications/enhancements:
* Copyright (c) 1995 John S. Dyson. All rights reserved.
1994-05-24 10:09:53 +00:00
*
* 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_cluster.c 8.7 (Berkeley) 2/13/94
*/
2003-06-11 00:56:59 +00:00
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_debug_cluster.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
1994-05-24 10:09:53 +00:00
#include <sys/proc.h>
#include <sys/bio.h>
1994-05-24 10:09:53 +00:00
#include <sys/buf.h>
#include <sys/vnode.h>
#include <sys/malloc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mount.h>
#include <sys/resourcevar.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>
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
#include <sys/vmmeter.h>
#include <vm/vm.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <sys/sysctl.h>
1994-05-24 10:09:53 +00:00
#if defined(CLUSTERDEBUG)
static int rcluster= 0;
SYSCTL_INT(_debug, OID_AUTO, rcluster, CTLFLAG_RW, &rcluster, 0,
"Debug VFS clustering code");
#endif
static MALLOC_DEFINE(M_SEGMENT, "cl_savebuf", "cluster_save buffer");
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
static struct cluster_save *cluster_collectbufs(struct vnode *vp,
struct buf *last_bp, int gbflags);
static struct buf *cluster_rbuild(struct vnode *vp, u_quad_t filesize,
daddr_t lbn, daddr_t blkno, long size, int run, int gbflags,
struct buf *fbp);
2005-02-10 12:17:48 +00:00
static void cluster_callback(struct buf *);
1994-05-24 10:09:53 +00:00
static int write_behind = 1;
SYSCTL_INT(_vfs, OID_AUTO, write_behind, CTLFLAG_RW, &write_behind, 0,
"Cluster write-behind; 0: disable, 1: enable, 2: backed off");
static int read_max = 64;
SYSCTL_INT(_vfs, OID_AUTO, read_max, CTLFLAG_RW, &read_max, 0,
"Cluster read-ahead max block count");
static int read_min = 1;
SYSCTL_INT(_vfs, OID_AUTO, read_min, CTLFLAG_RW, &read_min, 0,
"Cluster read min block count");
/* Page expended to mark partially backed buffers */
extern vm_page_t bogus_page;
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
1994-05-24 10:09:53 +00:00
/*
* Read data to a buf, including read-ahead if we find this to be beneficial.
* cluster_read replaces bread.
1994-05-24 10:09:53 +00:00
*/
int
cluster_read(struct vnode *vp, u_quad_t filesize, daddr_t lblkno, long size,
struct ucred *cred, long totread, int seqcount, int gbflags,
struct buf **bpp)
1994-05-24 10:09:53 +00:00
{
struct buf *bp, *rbp, *reqbp;
struct bufobj *bo;
daddr_t blkno, origblkno;
int maxra, racluster;
int error, ncontig;
int i;
1994-05-24 10:09:53 +00:00
error = 0;
bo = &vp->v_bufobj;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (!unmapped_buf_allowed)
gbflags &= ~GB_UNMAPPED;
/*
* Try to limit the amount of read-ahead by a few
* ad-hoc parameters. This needs work!!!
*/
racluster = vp->v_mount->mnt_iosize_max / size;
maxra = seqcount;
maxra = min(read_max, maxra);
maxra = min(nbuf/8, maxra);
if (((u_quad_t)(lblkno + maxra + 1) * size) > filesize)
maxra = (filesize / size) - lblkno;
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
/*
* get the requested block
*/
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
*bpp = reqbp = bp = getblk(vp, lblkno, size, 0, 0, gbflags);
origblkno = lblkno;
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 it is in the cache, then check to see if the reads have been
* sequential. If they have, then try some read-ahead, otherwise
* back-off on prospective read-aheads.
*/
1994-05-24 10:09:53 +00:00
if (bp->b_flags & B_CACHE) {
if (!seqcount) {
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 0;
} else if ((bp->b_flags & B_RAM) == 0) {
return 0;
} else {
bp->b_flags &= ~B_RAM;
BO_RLOCK(bo);
for (i = 1; i < maxra; i++) {
/*
* Stop if the buffer does not exist or it
* is invalid (about to go away?)
*/
rbp = gbincore(&vp->v_bufobj, lblkno+i);
if (rbp == NULL || (rbp->b_flags & B_INVAL))
break;
/*
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
* Set another read-ahead mark so we know
* to check again. (If we can lock the
* buffer without waiting)
*/
if ((((i % racluster) == (racluster - 1)) ||
(i == (maxra - 1)))
&& (0 == BUF_LOCK(rbp,
LK_EXCLUSIVE | LK_NOWAIT, NULL))) {
rbp->b_flags |= B_RAM;
BUF_UNLOCK(rbp);
}
}
BO_RUNLOCK(bo);
if (i >= maxra) {
return 0;
}
lblkno += i;
}
reqbp = bp = NULL;
/*
* If it isn't in the cache, then get a chunk from
* disk if sequential, otherwise just get the block.
*/
} else {
1999-01-10 01:58:29 +00:00
off_t firstread = bp->b_offset;
int nblks;
long minread;
1999-01-10 01:58:29 +00:00
KASSERT(bp->b_offset != NOOFFSET,
1999-01-10 01:58:29 +00:00
("cluster_read: no buffer offset"));
ncontig = 0;
/*
* Adjust totread if needed
*/
minread = read_min * size;
if (minread > totread)
totread = minread;
/*
* Compute the total number of blocks that we should read
* synchronously.
*/
if (firstread + totread > filesize)
totread = filesize - firstread;
nblks = howmany(totread, size);
if (nblks > racluster)
nblks = racluster;
/*
* Now compute the number of contiguous blocks.
*/
if (nblks > 1) {
error = VOP_BMAP(vp, lblkno, NULL,
&blkno, &ncontig, NULL);
/*
* If this failed to map just do the original block.
*/
if (error || blkno == -1)
ncontig = 0;
}
/*
* If we have contiguous data available do a cluster
* otherwise just read the requested block.
*/
if (ncontig) {
/* Account for our first block. */
ncontig = min(ncontig + 1, nblks);
if (ncontig < nblks)
nblks = ncontig;
bp = cluster_rbuild(vp, filesize, lblkno,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
blkno, size, nblks, gbflags, bp);
lblkno += (bp->b_bufsize / size);
} else {
bp->b_flags |= B_RAM;
bp->b_iocmd = BIO_READ;
lblkno += 1;
1995-05-30 08:16:23 +00:00
}
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
/*
* handle the synchronous read so that it is available ASAP.
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) {
if ((bp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(bp, 0);
}
bp->b_flags &= ~B_INVAL;
bp->b_ioflags &= ~BIO_ERROR;
if ((bp->b_flags & B_ASYNC) || bp->b_iodone != NULL)
BUF_KERNPROC(bp);
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
curthread->td_ru.ru_inblock++;
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
}
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
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 we have been doing sequential I/O, then do some read-ahead.
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
*/
while (lblkno < (origblkno + maxra)) {
error = VOP_BMAP(vp, lblkno, NULL, &blkno, &ncontig, NULL);
if (error)
break;
if (blkno == -1)
break;
/*
* We could throttle ncontig here by maxra but we might as
* well read the data if it is contiguous. We're throttled
* by racluster anyway.
*/
if (ncontig) {
ncontig = min(ncontig + 1, racluster);
rbp = cluster_rbuild(vp, filesize, lblkno, blkno,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
size, ncontig, gbflags, NULL);
lblkno += (rbp->b_bufsize / size);
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
} else {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
rbp = getblk(vp, lblkno, size, 0, 0, gbflags);
lblkno += 1;
if (rbp->b_flags & B_DELWRI) {
bqrelse(rbp);
continue;
}
rbp->b_flags |= B_ASYNC | B_RAM;
rbp->b_iocmd = BIO_READ;
rbp->b_blkno = blkno;
}
if (rbp->b_flags & B_CACHE) {
rbp->b_flags &= ~B_ASYNC;
bqrelse(rbp);
continue;
}
if ((rbp->b_flags & B_CLUSTER) == 0) {
vfs_busy_pages(rbp, 0);
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
}
rbp->b_flags &= ~B_INVAL;
rbp->b_ioflags &= ~BIO_ERROR;
if ((rbp->b_flags & B_ASYNC) || rbp->b_iodone != NULL)
BUF_KERNPROC(rbp);
rbp->b_iooffset = dbtob(rbp->b_blkno);
bstrategy(rbp);
curthread->td_ru.ru_inblock++;
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 (reqbp)
return (bufwait(reqbp));
else
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* If blocks are contiguous on disk, use this to provide clustered
* read ahead. We will read as many blocks as possible sequentially
* and then parcel them up into logical blocks in the buffer hash table.
*/
static struct buf *
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
cluster_rbuild(struct vnode *vp, u_quad_t filesize, daddr_t lbn,
daddr_t blkno, long size, int run, int gbflags, struct buf *fbp)
1994-05-24 10:09:53 +00:00
{
struct bufobj *bo;
struct buf *bp, *tbp;
1994-05-24 10:09:53 +00:00
daddr_t bn;
off_t off;
long tinc, tsize;
int i, inc, j, k, toff;
1994-05-24 10:09:53 +00:00
KASSERT(size == vp->v_mount->mnt_stat.f_iosize,
("cluster_rbuild: size %ld != f_iosize %jd\n",
size, (intmax_t)vp->v_mount->mnt_stat.f_iosize));
1999-01-10 01:58:29 +00:00
/*
* avoid a division
*/
while ((u_quad_t) size * (lbn + run) > filesize) {
1994-05-24 10:09:53 +00:00
--run;
}
if (fbp) {
tbp = fbp;
tbp->b_iocmd = BIO_READ;
} else {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
tbp = getblk(vp, lbn, size, 0, 0, gbflags);
if (tbp->b_flags & B_CACHE)
return tbp;
tbp->b_flags |= B_ASYNC | B_RAM;
tbp->b_iocmd = BIO_READ;
}
tbp->b_blkno = blkno;
if( (tbp->b_flags & B_MALLOC) ||
((tbp->b_flags & B_VMIO) == 0) || (run <= 1) )
return tbp;
bp = trypbuf(&cluster_pbuf_freecnt);
if (bp == 0)
return tbp;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
bp->b_flags = B_ASYNC | B_CLUSTER | B_VMIO;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((gbflags & GB_UNMAPPED) != 0) {
bp->b_flags |= B_UNMAPPED;
bp->b_data = unmapped_buf;
} else {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
}
bp->b_iocmd = BIO_READ;
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
bp->b_iodone = cluster_callback;
bp->b_blkno = blkno;
bp->b_lblkno = lbn;
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
bp->b_offset = tbp->b_offset;
1999-01-10 01:58:29 +00:00
KASSERT(bp->b_offset != NOOFFSET, ("cluster_rbuild: no buffer offset"));
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
pbgetvp(vp, bp);
1994-05-24 10:09:53 +00:00
TAILQ_INIT(&bp->b_cluster.cluster_head);
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
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
1994-05-24 10:09:53 +00:00
inc = btodb(size);
bo = &vp->v_bufobj;
for (bn = blkno, i = 0; i < run; ++i, bn += inc) {
if (i == 0) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
vfs_drain_busy_pages(tbp);
vm_object_pip_add(tbp->b_bufobj->bo_object,
tbp->b_npages);
for (k = 0; k < tbp->b_npages; k++)
vm_page_sbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
} else {
if ((bp->b_npages * PAGE_SIZE) +
round_page(size) > vp->v_mount->mnt_iosize_max) {
break;
}
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
tbp = getblk(vp, lbn + i, size, 0, 0, GB_LOCK_NOWAIT |
(gbflags & GB_UNMAPPED));
/* Don't wait around for locked bufs. */
if (tbp == NULL)
break;
/*
* Stop scanning if the buffer is fully valid
* (marked B_CACHE), or locked (may be doing a
* background write), or if the buffer is not
* VMIO backed. The clustering code can only deal
* with VMIO-backed buffers. The bo lock is not
* required for the BKGRDINPROG check since it
* can not be set without the buf lock.
*/
if ((tbp->b_vflags & BV_BKGRDINPROG) ||
(tbp->b_flags & B_CACHE) ||
(tbp->b_flags & B_VMIO) == 0) {
bqrelse(tbp);
break;
}
/*
* The buffer must be completely invalid in order to
* take part in the cluster. If it is partially valid
* then we stop.
*/
off = tbp->b_offset;
tsize = size;
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(tbp->b_bufobj->bo_object);
for (j = 0; tsize > 0; j++) {
toff = off & PAGE_MASK;
tinc = tsize;
if (toff + tinc > PAGE_SIZE)
tinc = PAGE_SIZE - toff;
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_ASSERT_WLOCKED(tbp->b_pages[j]->object);
if ((tbp->b_pages[j]->valid &
vm_page_bits(toff, tinc)) != 0)
break;
if (vm_page_xbusied(tbp->b_pages[j]))
break;
vm_object_pip_add(tbp->b_bufobj->bo_object, 1);
vm_page_sbusy(tbp->b_pages[j]);
off += tinc;
tsize -= tinc;
}
if (tsize > 0) {
clean_sbusy:
vm_object_pip_add(tbp->b_bufobj->bo_object, -j);
for (k = 0; k < j; k++)
vm_page_sunbusy(tbp->b_pages[k]);
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
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
bqrelse(tbp);
break;
}
VM_OBJECT_WUNLOCK(tbp->b_bufobj->bo_object);
/*
* Set a read-ahead mark as appropriate
*/
if ((fbp && (i == 1)) || (i == (run - 1)))
tbp->b_flags |= B_RAM;
/*
* Set the buffer up for an async read (XXX should
* we do this only if we do not wind up brelse()ing?).
* Set the block number if it isn't set, otherwise
* if it is make sure it matches the block number we
* expect.
*/
tbp->b_flags |= B_ASYNC;
tbp->b_iocmd = BIO_READ;
if (tbp->b_blkno == tbp->b_lblkno) {
tbp->b_blkno = bn;
} else if (tbp->b_blkno != bn) {
VM_OBJECT_WLOCK(tbp->b_bufobj->bo_object);
goto clean_sbusy;
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
}
1994-05-24 10:09:53 +00:00
}
/*
* XXX fbp from caller may not be B_ASYNC, but we are going
* to biodone() it in cluster_callback() anyway
*/
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
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(tbp->b_bufobj->bo_object);
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
for (j = 0; j < tbp->b_npages; j += 1) {
vm_page_t m;
m = tbp->b_pages[j];
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages-1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
if (m->valid == VM_PAGE_BITS_ALL)
tbp->b_pages[j] = bogus_page;
1994-05-24 10:09:53 +00:00
}
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(tbp->b_bufobj->bo_object);
/*
* Don't inherit tbp->b_bufsize as it may be larger due to
* a non-page-aligned size. Instead just aggregate using
* 'size'.
*/
if (tbp->b_bcount != size)
printf("warning: tbp->b_bcount wrong %ld vs %ld\n", tbp->b_bcount, size);
if (tbp->b_bufsize != size)
printf("warning: tbp->b_bufsize wrong %ld vs %ld\n", tbp->b_bufsize, size);
bp->b_bcount += size;
bp->b_bufsize += size;
1994-05-24 10:09:53 +00:00
}
/*
* Fully valid pages in the cluster are already good and do not need
* to be re-read from disk. Replace the page with bogus_page
*/
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(bp->b_bufobj->bo_object);
for (j = 0; j < bp->b_npages; j++) {
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_ASSERT_WLOCKED(bp->b_pages[j]->object);
if (bp->b_pages[j]->valid == VM_PAGE_BITS_ALL)
bp->b_pages[j] = bogus_page;
}
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(bp->b_bufobj->bo_object);
if (bp->b_bufsize > bp->b_kvasize)
1998-07-11 10:45:45 +00:00
panic("cluster_rbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
bp->b_kvasize = bp->b_bufsize;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((bp->b_flags & B_UNMAPPED) == 0) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
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 (bp);
1994-05-24 10:09:53 +00:00
}
/*
* Cleanup after a clustered read or write.
* This is complicated by the fact that any of the buffers might have
* extra memory (if there were no empty buffer headers at allocbuf time)
* that we will need to shift around.
*/
2005-02-10 12:17:48 +00:00
static void
1994-05-24 10:09:53 +00:00
cluster_callback(bp)
struct buf *bp;
{
struct buf *nbp, *tbp;
1994-05-24 10:09:53 +00:00
int error = 0;
/*
* Must propogate errors to all the components.
*/
if (bp->b_ioflags & BIO_ERROR)
1994-05-24 10:09:53 +00:00
error = bp->b_error;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((bp->b_flags & B_UNMAPPED) == 0) {
pmap_qremove(trunc_page((vm_offset_t) bp->b_data),
bp->b_npages);
}
1994-05-24 10:09:53 +00:00
/*
* Move memory from the large cluster buffer into the component
* buffers and mark IO as done on these.
*/
for (tbp = TAILQ_FIRST(&bp->b_cluster.cluster_head);
tbp; tbp = nbp) {
nbp = TAILQ_NEXT(&tbp->b_cluster, cluster_entry);
1994-05-24 10:09:53 +00:00
if (error) {
tbp->b_ioflags |= BIO_ERROR;
1994-05-24 10:09:53 +00:00
tbp->b_error = error;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
} else {
tbp->b_dirtyoff = tbp->b_dirtyend = 0;
tbp->b_flags &= ~B_INVAL;
tbp->b_ioflags &= ~BIO_ERROR;
/*
* XXX the bdwrite()/bqrelse() issued during
* cluster building clears B_RELBUF (see bqrelse()
* comment). If direct I/O was specified, we have
* to restore it here to allow the buffer and VM
* to be freed.
*/
if (tbp->b_flags & B_DIRECT)
tbp->b_flags |= B_RELBUF;
The VFS/BIO subsystem contained a number of hacks in order to optimize piecemeal, middle-of-file writes for NFS. These hacks have caused no end of trouble, especially when combined with mmap(). I've removed them. Instead, NFS will issue a read-before-write to fully instantiate the struct buf containing the write. NFS does, however, optimize piecemeal appends to files. For most common file operations, you will not notice the difference. The sole remaining fragment in the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache coherency issues with read-merge-write style operations. NFS also optimizes the write-covers-entire-buffer case by avoiding the read-before-write. There is quite a bit of room for further optimization in these areas. The VM system marks pages fully-valid (AKA vm_page_t->valid = VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This is not correct operation. The vm_pager_get_pages() code is now responsible for marking VM pages all-valid. A number of VM helper routines have been added to aid in zeroing-out the invalid portions of a VM page prior to the page being marked all-valid. This operation is necessary to properly support mmap(). The zeroing occurs most often when dealing with file-EOF situations. Several bugs have been fixed in the NFS subsystem, including bits handling file and directory EOF situations and buf->b_flags consistancy issues relating to clearing B_ERROR & B_INVAL, and handling B_DONE. getblk() and allocbuf() have been rewritten. B_CACHE operation is now formally defined in comments and more straightforward in implementation. B_CACHE for VMIO buffers is based on the validity of the backing store. B_CACHE for non-VMIO buffers is based simply on whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear, and vise-versa). biodone() is now responsible for setting B_CACHE when a successful read completes. B_CACHE is also set when a bdwrite() is initiated and when a bwrite() is initiated. VFS VOP_BWRITE routines (there are only two - nfs_bwrite() and bwrite()) are now expected to set B_CACHE. This means that bowrite() and bawrite() also set B_CACHE indirectly. There are a number of places in the code which were previously using buf->b_bufsize (which is DEV_BSIZE aligned) when they should have been using buf->b_bcount. These have been fixed. getblk() now clears B_DONE on return because the rest of the system is so bad about dealing with B_DONE. Major fixes to NFS/TCP have been made. A server-side bug could cause requests to be lost by the server due to nfs_realign() overwriting other rpc's in the same TCP mbuf chain. The server's kernel must be recompiled to get the benefit of the fixes. Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
}
bufdone(tbp);
1994-05-24 10:09:53 +00:00
}
2004-11-15 08:06:05 +00:00
pbrelvp(bp);
relpbuf(bp, &cluster_pbuf_freecnt);
1994-05-24 10:09:53 +00:00
}
/*
* cluster_wbuild_wb:
*
* Implement modified write build for cluster.
*
* write_behind = 0 write behind disabled
* write_behind = 1 write behind normal (default)
* write_behind = 2 write behind backed-off
*/
static __inline int
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
cluster_wbuild_wb(struct vnode *vp, long size, daddr_t start_lbn, int len,
int gbflags)
{
int r = 0;
switch (write_behind) {
case 2:
if (start_lbn < len)
break;
start_lbn -= len;
/* FALLTHROUGH */
case 1:
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
r = cluster_wbuild(vp, size, start_lbn, len, gbflags);
/* FALLTHROUGH */
default:
/* FALLTHROUGH */
break;
}
return(r);
}
1994-05-24 10:09:53 +00:00
/*
* Do clustered write for FFS.
*
* Three cases:
* 1. Write is not sequential (write asynchronously)
* Write is sequential:
* 2. beginning of cluster - begin cluster
* 3. middle of a cluster - add to cluster
* 4. end of a cluster - asynchronously write cluster
*/
void
cluster_write(struct vnode *vp, struct buf *bp, u_quad_t filesize, int seqcount,
int gbflags)
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
daddr_t lbn;
int maxclen, cursize;
int lblocksize;
int async;
1994-05-24 10:09:53 +00:00
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (!unmapped_buf_allowed)
gbflags &= ~GB_UNMAPPED;
if (vp->v_type == VREG) {
async = DOINGASYNC(vp);
lblocksize = vp->v_mount->mnt_stat.f_iosize;
} else {
async = 0;
lblocksize = bp->b_bufsize;
}
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
lbn = bp->b_lblkno;
KASSERT(bp->b_offset != NOOFFSET, ("cluster_write: no buffer offset"));
1994-05-24 10:09:53 +00:00
/* Initialize vnode to beginning of file. */
if (lbn == 0)
vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
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 (vp->v_clen == 0 || lbn != vp->v_lastw + 1 ||
(bp->b_blkno != vp->v_lasta + btodb(lblocksize))) {
maxclen = vp->v_mount->mnt_iosize_max / lblocksize - 1;
1994-05-24 10:09:53 +00:00
if (vp->v_clen != 0) {
/*
* Next block is not sequential.
1995-05-30 08:16:23 +00:00
*
1994-05-24 10:09:53 +00:00
* If we are not writing at end of file, the process
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
* seeked to another point in the file since its last
* write, or we have reached our maximum cluster size,
* then push the previous cluster. Otherwise try
* reallocating to make it sequential.
*
* Change to algorithm: only push previous cluster if
* it was sequential from the point of view of the
* seqcount heuristic, otherwise leave the buffer
* intact so we can potentially optimize the I/O
* later on in the buf_daemon or update daemon
* flush.
1994-05-24 10:09:53 +00:00
*/
cursize = vp->v_lastw - vp->v_cstart + 1;
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 (((u_quad_t) bp->b_offset + lblocksize) != filesize ||
lbn != vp->v_lastw + 1 || vp->v_clen <= cursize) {
if (!async && seqcount > 0) {
These changes appear to give us benefits with both small (32MB) and large (1G) memory machine configurations. I was able to run 'dbench 32' on a 32MB system without bring the machine to a grinding halt. * buffer cache hash table now dynamically allocated. This will have no effect on memory consumption for smaller systems and will help scale the buffer cache for larger systems. * minor enhancement to pmap_clearbit(). I noticed that all the calls to it used constant arguments. Making it an inline allows the constants to propogate to deeper inlines and should produce better code. * removal of inherent vfs_ioopt support through the emplacement of appropriate #ifdef's, with John's permission. If we do not find a use for it by the end of the year we will remove it entirely. * removal of getnewbufloops* counters & sysctl's - no longer necessary for debugging, getnewbuf() is now optimal. * buffer hash table functions removed from sys/buf.h and localized to vfs_bio.c * VFS_BIO_NEED_DIRTYFLUSH flag and support code added ( bwillwrite() ), allowing processes to block when too many dirty buffers are present in the system. * removal of a softdep test in bdwrite() that is no longer necessary now that bdwrite() no longer attempts to flush dirty buffers. * slight optimization added to bqrelse() - there is no reason to test for available buffer space on B_DELWRI buffers. * addition of reverse-scanning code to vfs_bio_awrite(). vfs_bio_awrite() will attempt to locate clusterable areas in both the forward and reverse direction relative to the offset of the buffer passed to it. This will probably not make much of a difference now, but I believe we will start to rely on it heavily in the future if we decide to shift some of the burden of the clustering closer to the actual I/O initiation. * Removal of the newbufcnt and lastnewbuf counters that Kirk added. They do not fix any race conditions that haven't already been fixed by the gbincore() test done after the only call to getnewbuf(). getnewbuf() is a static, so there is no chance of it being misused by other modules. ( Unless Kirk can think of a specific thing that this code fixes. I went through it very carefully and didn't see anything ). * removal of VOP_ISLOCKED() check in flushbufqueues(). I do not think this check is necessary, the buffer should flush properly whether the vnode is locked or not. ( yes? ). * removal of extra arguments passed to getnewbuf() that are not necessary. * missed cluster_wbuild() that had to be a cluster_wbuild_wb() in vfs_cluster.c * vn_write() now calls bwillwrite() *PRIOR* to locking the vnode, which should greatly aid flushing operations in heavy load situations - both the pageout and update daemons will be able to operate more efficiently. * removal of b_usecount. We may add it back in later but for now it is useless. Prior implementations of the buffer cache never had enough buffers for it to be useful, and current implementations which make more buffers available might not benefit relative to the amount of sophistication required to implement a b_usecount. Straight LRU should work just as well, especially when most things are VMIO backed. I expect that (even though John will not like this assumption) directories will become VMIO backed some point soon. Submitted by: Matthew Dillon <dillon@backplane.com> Reviewed by: Kirk McKusick <mckusick@mckusick.com>
1999-07-08 06:06:00 +00:00
cluster_wbuild_wb(vp, lblocksize,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
vp->v_cstart, cursize, gbflags);
}
} else {
struct buf **bpp, **endbp;
struct cluster_save *buflist;
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
buflist = cluster_collectbufs(vp, bp, gbflags);
endbp = &buflist->bs_children
[buflist->bs_nchildren - 1];
if (VOP_REALLOCBLKS(vp, buflist)) {
/*
* Failed, push the previous cluster
* if *really* writing sequentially
* in the logical file (seqcount > 1),
* otherwise delay it in the hopes that
* the low level disk driver can
* optimize the write ordering.
*/
for (bpp = buflist->bs_children;
bpp < endbp; bpp++)
brelse(*bpp);
free(buflist, M_SEGMENT);
if (seqcount > 1) {
cluster_wbuild_wb(vp,
lblocksize, vp->v_cstart,
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
cursize, gbflags);
}
} else {
/*
* Succeeded, keep building cluster.
*/
for (bpp = buflist->bs_children;
bpp <= endbp; bpp++)
bdwrite(*bpp);
free(buflist, M_SEGMENT);
vp->v_lastw = lbn;
vp->v_lasta = bp->b_blkno;
return;
}
}
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
* Consider beginning a cluster. If at end of file, make
* cluster as large as possible, otherwise find size of
* existing cluster.
1994-05-24 10:09:53 +00:00
*/
if ((vp->v_type == VREG) &&
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
((u_quad_t) bp->b_offset + lblocksize) != filesize &&
(bp->b_blkno == bp->b_lblkno) &&
(VOP_BMAP(vp, lbn, NULL, &bp->b_blkno, &maxclen, NULL) ||
bp->b_blkno == -1)) {
1994-05-24 10:09:53 +00:00
bawrite(bp);
vp->v_clen = 0;
vp->v_lasta = bp->b_blkno;
vp->v_cstart = lbn + 1;
vp->v_lastw = lbn;
return;
}
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
vp->v_clen = maxclen;
if (!async && maxclen == 0) { /* I/O not contiguous */
1994-05-24 10:09:53 +00:00
vp->v_cstart = lbn + 1;
bawrite(bp);
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 { /* Wait for rest of cluster */
1994-05-24 10:09:53 +00:00
vp->v_cstart = lbn;
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
bdwrite(bp);
1994-05-24 10:09:53 +00:00
}
} else if (lbn == vp->v_cstart + vp->v_clen) {
/*
* At end of cluster, write it out if seqcount tells us we
* are operating sequentially, otherwise let the buf or
* update daemon handle it.
1994-05-24 10:09:53 +00:00
*/
bdwrite(bp);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (seqcount > 1) {
cluster_wbuild_wb(vp, lblocksize, vp->v_cstart,
vp->v_clen + 1, gbflags);
}
1994-05-24 10:09:53 +00:00
vp->v_clen = 0;
vp->v_cstart = lbn + 1;
Implement a low-memory deadlock solution. Removed most of the hacks that were trying to deal with low-memory situations prior to now. The new code is based on the concept that I/O must be able to function in a low memory situation. All major modules related to I/O (except networking) have been adjusted to allow allocation out of the system reserve memory pool. These modules now detect a low memory situation but rather then block they instead continue to operate, then return resources to the memory pool instead of cache them or leave them wired. Code has been added to stall in a low-memory situation prior to a vnode being locked. Thus situations where a process blocks in a low-memory condition while holding a locked vnode have been reduced to near nothing. Not only will I/O continue to operate, but many prior deadlock conditions simply no longer exist. Implement a number of VFS/BIO fixes (found by Ian): in biodone(), bogus-page replacement code, the loop was not properly incrementing loop variables prior to a continue statement. We do not believe this code can be hit anyway but we aren't taking any chances. We'll turn the whole section into a panic (as it already is in brelse()) after the release is rolled. In biodone(), the foff calculation was incorrectly clamped to the iosize, causing the wrong foff to be calculated for pages in the case of an I/O error or biodone() called without initiating I/O. The problem always caused a panic before. Now it doesn't. The problem is mainly an issue with NFS. Fixed casts for ~PAGE_MASK. This code worked properly before only because the calculations use signed arithmatic. Better to properly extend PAGE_MASK first before inverting it for the 64 bit masking op. In brelse(), the bogus_page fixup code was improperly throwing away the original contents of 'm' when it did the j-loop to fix the bogus pages. The result was that it would potentially invalidate parts of the *WRONG* page(!), leading to corruption. There may still be cases where a background bitmap write is being duplicated, causing potential corruption. We have identified a potentially serious bug related to this but the fix is still TBD. So instead this patch contains a KASSERT to detect the problem and panic the machine rather then continue to corrupt the filesystem. The problem does not occur very often.. it is very hard to reproduce, and it may or may not be the cause of the corruption people have reported. Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>) Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
} else if (vm_page_count_severe()) {
/*
* We are low on memory, get it going NOW
*/
bawrite(bp);
} else {
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
* In the middle of a cluster, so just delay the I/O for now.
1994-05-24 10:09:53 +00:00
*/
bdwrite(bp);
}
1994-05-24 10:09:53 +00:00
vp->v_lastw = lbn;
vp->v_lasta = bp->b_blkno;
}
/*
* This is an awful lot like cluster_rbuild...wish they could be combined.
* The last lbn argument is the current block on which I/O is being
* performed. Check to see that it doesn't fall in the middle of
* the current block (if last_bp == NULL).
*/
int
cluster_wbuild(struct vnode *vp, long size, daddr_t start_lbn, int len,
int gbflags)
1994-05-24 10:09:53 +00:00
{
struct buf *bp, *tbp;
struct bufobj *bo;
int i, j;
int totalwritten = 0;
int dbsize = btodb(size);
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if (!unmapped_buf_allowed)
gbflags &= ~GB_UNMAPPED;
bo = &vp->v_bufobj;
while (len > 0) {
/*
* If the buffer is not delayed-write (i.e. dirty), or it
* is delayed-write but either locked or inval, it cannot
2001-02-06 09:18:39 +00:00
* partake in the clustered write.
*/
BO_LOCK(bo);
if ((tbp = gbincore(&vp->v_bufobj, start_lbn)) == NULL ||
(tbp->b_vflags & BV_BKGRDINPROG)) {
BO_UNLOCK(bo);
++start_lbn;
--len;
continue;
}
if (BUF_LOCK(tbp,
LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, BO_LOCKPTR(bo))) {
++start_lbn;
--len;
continue;
}
if ((tbp->b_flags & (B_INVAL | B_DELWRI)) != B_DELWRI) {
BUF_UNLOCK(tbp);
++start_lbn;
--len;
continue;
}
if (tbp->b_pin_count > 0) {
BUF_UNLOCK(tbp);
++start_lbn;
--len;
continue;
}
bremfree(tbp);
tbp->b_flags &= ~B_DONE;
/*
* Extra memory in the buffer, punt on this buffer.
* XXX we could handle this in most cases, but we would
* have to push the extra memory down to after our max
* possible cluster size and then potentially pull it back
* up if the cluster was terminated prematurely--too much
* hassle.
*/
if (((tbp->b_flags & (B_CLUSTEROK | B_MALLOC | B_VMIO)) !=
(B_CLUSTEROK | B_VMIO)) ||
(tbp->b_bcount != tbp->b_bufsize) ||
(tbp->b_bcount != size) ||
(len == 1) ||
((bp = (vp->v_vflag & VV_MD) != 0 ?
trypbuf(&cluster_pbuf_freecnt) :
getpbuf(&cluster_pbuf_freecnt)) == NULL)) {
totalwritten += tbp->b_bufsize;
bawrite(tbp);
++start_lbn;
--len;
continue;
}
/*
* We got a pbuf to make the cluster in.
* so initialise it.
*/
TAILQ_INIT(&bp->b_cluster.cluster_head);
bp->b_bcount = 0;
bp->b_bufsize = 0;
bp->b_npages = 0;
if (tbp->b_wcred != NOCRED)
bp->b_wcred = crhold(tbp->b_wcred);
bp->b_blkno = tbp->b_blkno;
bp->b_lblkno = tbp->b_lblkno;
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
bp->b_offset = tbp->b_offset;
/*
* We are synthesizing a buffer out of vm_page_t's, but
* if the block size is not page aligned then the starting
* address may not be either. Inherit the b_data offset
* from the original buffer.
*/
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((gbflags & GB_UNMAPPED) == 0 ||
(tbp->b_flags & B_VMIO) == 0) {
bp->b_data = (char *)((vm_offset_t)bp->b_data |
((vm_offset_t)tbp->b_data & PAGE_MASK));
} else {
bp->b_flags |= B_UNMAPPED;
bp->b_data = unmapped_buf;
}
bp->b_flags |= B_CLUSTER | (tbp->b_flags & (B_VMIO |
B_NEEDCOMMIT));
bp->b_iodone = cluster_callback;
pbgetvp(vp, bp);
/*
* From this location in the file, scan forward to see
* if there are buffers with adjacent data that need to
* be written as well.
*/
for (i = 0; i < len; ++i, ++start_lbn) {
if (i != 0) { /* If not the first buffer */
/*
* If the adjacent data is not even in core it
* can't need to be written.
*/
BO_LOCK(bo);
if ((tbp = gbincore(bo, start_lbn)) == NULL ||
(tbp->b_vflags & BV_BKGRDINPROG)) {
BO_UNLOCK(bo);
break;
}
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 it IS in core, but has different
* characteristics, or is locked (which
* means it could be undergoing a background
* I/O or be in a weird state), then don't
* cluster with it.
*/
if (BUF_LOCK(tbp,
LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK,
BO_LOCKPTR(bo)))
break;
if ((tbp->b_flags & (B_VMIO | B_CLUSTEROK |
B_INVAL | B_DELWRI | B_NEEDCOMMIT))
!= (B_DELWRI | B_CLUSTEROK |
(bp->b_flags & (B_VMIO | B_NEEDCOMMIT))) ||
tbp->b_wcred != bp->b_wcred) {
BUF_UNLOCK(tbp);
break;
}
1994-05-24 10:09:53 +00:00
/*
* Check that the combined cluster
* would make sense with regard to pages
* and would not be too large
*/
if ((tbp->b_bcount != size) ||
((bp->b_blkno + (dbsize * i)) !=
tbp->b_blkno) ||
((tbp->b_npages + bp->b_npages) >
(vp->v_mount->mnt_iosize_max / PAGE_SIZE))) {
BUF_UNLOCK(tbp);
break;
}
/*
* Do not pull in pinned buffers.
*/
if (tbp->b_pin_count > 0) {
BUF_UNLOCK(tbp);
break;
}
/*
* Ok, it's passed all the tests,
* so remove it from the free list
* and mark it busy. We will use it.
*/
bremfree(tbp);
tbp->b_flags &= ~B_DONE;
} /* end of code for non-first buffers only */
/*
* If the IO is via the VM then we do some
* special VM hackery (yuck). Since the buffer's
* block size may not be page-aligned it is possible
* for a page to be shared between two buffers. We
* have to get rid of the duplication when building
* the cluster.
*/
if (tbp->b_flags & B_VMIO) {
vm_page_t m;
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(tbp->b_bufobj->bo_object);
if (i == 0) {
vfs_drain_busy_pages(tbp);
2013-08-23 16:54:38 +00:00
} else { /* if not first buffer */
for (j = 0; j < tbp->b_npages; j += 1) {
m = tbp->b_pages[j];
if (vm_page_xbusied(m)) {
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(
tbp->b_object);
bqrelse(tbp);
goto finishcluster;
}
}
}
for (j = 0; j < tbp->b_npages; j += 1) {
m = tbp->b_pages[j];
vm_page_sbusy(m);
vm_object_pip_add(m->object, 1);
if ((bp->b_npages == 0) ||
(bp->b_pages[bp->b_npages - 1] != m)) {
bp->b_pages[bp->b_npages] = m;
bp->b_npages++;
}
}
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(tbp->b_bufobj->bo_object);
}
bp->b_bcount += size;
bp->b_bufsize += size;
/*
* If any of the clustered buffers have their
* B_BARRIER flag set, transfer that request to
* the cluster.
*/
bp->b_flags |= (tbp->b_flags & B_BARRIER);
tbp->b_flags &= ~(B_DONE | B_BARRIER);
tbp->b_flags |= B_ASYNC;
tbp->b_ioflags &= ~BIO_ERROR;
tbp->b_iocmd = BIO_WRITE;
bundirty(tbp);
reassignbuf(tbp); /* put on clean list */
bufobj_wref(tbp->b_bufobj);
BUF_KERNPROC(tbp);
TAILQ_INSERT_TAIL(&bp->b_cluster.cluster_head,
tbp, b_cluster.cluster_entry);
1994-05-24 10:09:53 +00:00
}
finishcluster:
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
if ((bp->b_flags & B_UNMAPPED) == 0) {
pmap_qenter(trunc_page((vm_offset_t) bp->b_data),
(vm_page_t *)bp->b_pages, bp->b_npages);
}
if (bp->b_bufsize > bp->b_kvasize)
1998-07-11 10:45:45 +00:00
panic(
"cluster_wbuild: b_bufsize(%ld) > b_kvasize(%d)\n",
bp->b_bufsize, bp->b_kvasize);
bp->b_kvasize = bp->b_bufsize;
totalwritten += bp->b_bufsize;
bp->b_dirtyoff = 0;
bp->b_dirtyend = bp->b_bufsize;
bawrite(bp);
1994-05-24 10:09:53 +00:00
len -= i;
1994-05-24 10:09:53 +00:00
}
return totalwritten;
1994-05-24 10:09:53 +00:00
}
/*
* Collect together all the buffers in a cluster.
* Plus add one additional buffer.
*/
static struct cluster_save *
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
cluster_collectbufs(struct vnode *vp, struct buf *last_bp, int gbflags)
1994-05-24 10:09:53 +00:00
{
struct cluster_save *buflist;
struct buf *bp;
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
daddr_t lbn;
1994-05-24 10:09:53 +00:00
int i, len;
len = vp->v_lastw - vp->v_cstart + 1;
buflist = malloc(sizeof(struct buf *) * (len + 1) + sizeof(*buflist),
M_SEGMENT, M_WAITOK);
1994-05-24 10:09:53 +00:00
buflist->bs_nchildren = 0;
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
buflist->bs_children = (struct buf **) (buflist + 1);
for (lbn = vp->v_cstart, i = 0; i < len; lbn++, i++) {
Implement the concept of the unmapped VMIO buffers, i.e. buffers which do not map the b_pages pages into buffer_map KVA. The use of the unmapped buffers eliminate the need to perform TLB shootdown for mapping on the buffer creation and reuse, greatly reducing the amount of IPIs for shootdown on big-SMP machines and eliminating up to 25-30% of the system time on i/o intensive workloads. The unmapped buffer should be explicitely requested by the GB_UNMAPPED flag by the consumer. For unmapped buffer, no KVA reservation is performed at all. The consumer might request unmapped buffer which does have a KVA reserve, to manually map it without recursing into buffer cache and blocking, with the GB_KVAALLOC flag. When the mapped buffer is requested and unmapped buffer already exists, the cache performs an upgrade, possibly reusing the KVA reservation. Unmapped buffer is translated into unmapped bio in g_vfs_strategy(). Unmapped bio carry a pointer to the vm_page_t array, offset and length instead of the data pointer. The provider which processes the bio should explicitely specify a readiness to accept unmapped bio, otherwise g_down geom thread performs the transient upgrade of the bio request by mapping the pages into the new bio_transient_map KVA submap. The bio_transient_map submap claims up to 10% of the buffer map, and the total buffer_map + bio_transient_map KVA usage stays the same. Still, it could be manually tuned by kern.bio_transient_maxcnt tunable, in the units of the transient mappings. Eventually, the bio_transient_map could be removed after all geom classes and drivers can accept unmapped i/o requests. Unmapped support can be turned off by the vfs.unmapped_buf_allowed tunable, disabling which makes the buffer (or cluster) creation requests to ignore GB_UNMAPPED and GB_KVAALLOC flags. Unmapped buffers are only enabled by default on the architectures where pmap_copy_page() was implemented and tested. In the rework, filesystem metadata is not the subject to maxbufspace limit anymore. Since the metadata buffers are always mapped, the buffers still have to fit into the buffer map, which provides a reasonable (but practically unreachable) upper bound on it. The non-metadata buffer allocations, both mapped and unmapped, is accounted against maxbufspace, as before. Effectively, this means that the maxbufspace is forced on mapped and unmapped buffers separately. The pre-patch bufspace limiting code did not worked, because buffer_map fragmentation does not allow the limit to be reached. By Jeff Roberson request, the getnewbuf() function was split into smaller single-purpose functions. Sponsored by: The FreeBSD Foundation Discussed with: jeff (previous version) Tested by: pho, scottl (previous version), jhb, bf MFC after: 2 weeks
2013-03-19 14:13:12 +00:00
(void)bread_gb(vp, lbn, last_bp->b_bcount, NOCRED,
gbflags, &bp);
buflist->bs_children[i] = bp;
if (bp->b_blkno == bp->b_lblkno)
VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno,
NULL, NULL);
}
buflist->bs_children[i] = bp = last_bp;
if (bp->b_blkno == bp->b_lblkno)
VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, NULL, NULL);
1994-05-24 10:09:53 +00:00
buflist->bs_nchildren = i + 1;
return (buflist);
}