freebsd-dev/sys/nfsclient/nfs_vnops.c

3546 lines
92 KiB
C
Raw Normal View History

/*-
1994-05-24 10:09:53 +00:00
* Copyright (c) 1989, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* Rick Macklem at The University of Guelph.
*
* 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.
*
* @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
1994-05-24 10:09:53 +00:00
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
1994-05-24 10:09:53 +00:00
/*
* vnode op calls for Sun NFS version 2 and 3
1994-05-24 10:09:53 +00:00
*/
#include "opt_inet.h"
1994-05-24 10:09:53 +00:00
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/systm.h>
#include <sys/resourcevar.h>
#include <sys/proc.h>
1994-05-24 10:09:53 +00:00
#include <sys/mount.h>
#include <sys/bio.h>
1994-05-24 10:09:53 +00:00
#include <sys/buf.h>
#include <sys/jail.h>
#include <sys/lock.h>
1994-05-24 10:09:53 +00:00
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/namei.h>
#include <sys/socket.h>
1994-05-24 10:09:53 +00:00
#include <sys/vnode.h>
#include <sys/dirent.h>
#include <sys/fcntl.h>
#include <sys/lockf.h>
#include <sys/rwlock.h>
#include <sys/stat.h>
#include <sys/sysctl.h>
#include <sys/signalvar.h>
1994-05-24 10:09:53 +00:00
#include <vm/vm.h>
#include <vm/vm_extern.h>
2004-12-01 07:34:08 +00:00
#include <vm/vm_object.h>
1994-05-24 10:09:53 +00:00
#include <nfs/nfsproto.h>
#include <nfsclient/nfs.h>
#include <nfsclient/nfsnode.h>
#include <nfsclient/nfsmount.h>
#include <nfs/nfs_kdtrace.h>
#include <nfs/nfs_lock.h>
1994-05-24 10:09:53 +00:00
#include <nfs/xdr_subs.h>
#include <nfsclient/nfsm_subs.h>
1994-05-24 10:09:53 +00:00
#include <net/if.h>
#include <net/if_var.h>
#include <net/vnet.h>
#include <netinet/in.h>
#include <netinet/in_var.h>
#include <machine/stdarg.h>
#ifdef KDTRACE_HOOKS
#include <sys/dtrace_bsd.h>
dtrace_nfsclient_accesscache_flush_probe_func_t
dtrace_nfsclient_accesscache_flush_done_probe;
uint32_t nfsclient_accesscache_flush_done_id;
dtrace_nfsclient_accesscache_get_probe_func_t
dtrace_nfsclient_accesscache_get_hit_probe,
dtrace_nfsclient_accesscache_get_miss_probe;
uint32_t nfsclient_accesscache_get_hit_id;
uint32_t nfsclient_accesscache_get_miss_id;
dtrace_nfsclient_accesscache_load_probe_func_t
dtrace_nfsclient_accesscache_load_done_probe;
uint32_t nfsclient_accesscache_load_done_id;
#endif /* !KDTRACE_HOOKS */
1994-05-24 10:09:53 +00:00
/* Defs */
#define TRUE 1
#define FALSE 0
/*
* Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
* calls are not in getblk() and brelse() so that they would not be necessary
* here.
*/
#ifndef B_VMIO
#define vfs_busy_pages(bp, f)
#endif
static vop_read_t nfsfifo_read;
static vop_write_t nfsfifo_write;
static vop_close_t nfsfifo_close;
static int nfs_flush(struct vnode *, int, int);
static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
static vop_lookup_t nfs_lookup;
static vop_create_t nfs_create;
static vop_mknod_t nfs_mknod;
static vop_open_t nfs_open;
static vop_close_t nfs_close;
static vop_access_t nfs_access;
static vop_getattr_t nfs_getattr;
static vop_setattr_t nfs_setattr;
static vop_read_t nfs_read;
static vop_fsync_t nfs_fsync;
static vop_remove_t nfs_remove;
static vop_link_t nfs_link;
static vop_rename_t nfs_rename;
static vop_mkdir_t nfs_mkdir;
static vop_rmdir_t nfs_rmdir;
static vop_symlink_t nfs_symlink;
static vop_readdir_t nfs_readdir;
static vop_strategy_t nfs_strategy;
static int nfs_lookitup(struct vnode *, const char *, int,
struct ucred *, struct thread *, struct nfsnode **);
static int nfs_sillyrename(struct vnode *, struct vnode *,
struct componentname *);
static vop_access_t nfsspec_access;
static vop_readlink_t nfs_readlink;
static vop_print_t nfs_print;
static vop_advlock_t nfs_advlock;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
static vop_advlockasync_t nfs_advlockasync;
1994-05-24 10:09:53 +00:00
/*
* Global vfs data structures for nfs
*/
struct vop_vector nfs_vnodeops = {
.vop_default = &default_vnodeops,
.vop_access = nfs_access,
.vop_advlock = nfs_advlock,
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
.vop_advlockasync = nfs_advlockasync,
.vop_close = nfs_close,
.vop_create = nfs_create,
.vop_fsync = nfs_fsync,
.vop_getattr = nfs_getattr,
.vop_getpages = nfs_getpages,
.vop_putpages = nfs_putpages,
.vop_inactive = nfs_inactive,
.vop_link = nfs_link,
.vop_lookup = nfs_lookup,
.vop_mkdir = nfs_mkdir,
.vop_mknod = nfs_mknod,
.vop_open = nfs_open,
.vop_print = nfs_print,
.vop_read = nfs_read,
.vop_readdir = nfs_readdir,
.vop_readlink = nfs_readlink,
.vop_reclaim = nfs_reclaim,
.vop_remove = nfs_remove,
.vop_rename = nfs_rename,
.vop_rmdir = nfs_rmdir,
.vop_setattr = nfs_setattr,
.vop_strategy = nfs_strategy,
.vop_symlink = nfs_symlink,
.vop_write = nfs_write,
1994-05-24 10:09:53 +00:00
};
struct vop_vector nfs_fifoops = {
.vop_default = &fifo_specops,
.vop_access = nfsspec_access,
.vop_close = nfsfifo_close,
.vop_fsync = nfs_fsync,
.vop_getattr = nfs_getattr,
.vop_inactive = nfs_inactive,
.vop_print = nfs_print,
.vop_read = nfsfifo_read,
.vop_reclaim = nfs_reclaim,
.vop_setattr = nfs_setattr,
.vop_write = nfsfifo_write,
1994-05-24 10:09:53 +00:00
};
static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
struct componentname *cnp, struct vattr *vap);
static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
struct ucred *cred, struct thread *td);
static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
int fnamelen, struct vnode *tdvp,
const char *tnameptr, int tnamelen,
struct ucred *cred, struct thread *td);
static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
struct sillyrename *sp);
1994-05-24 10:09:53 +00:00
/*
* Global variables
*/
struct mtx nfs_iod_mtx;
enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
int nfs_numasync = 0;
1994-05-24 10:09:53 +00:00
#define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
SYSCTL_DECL(_vfs_oldnfs);
static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
&nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
static int nfs_prime_access_cache = 0;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
&nfs_prime_access_cache, 0,
"Prime NFS ACCESS cache when fetching attributes");
static int nfsv3_commit_on_close = 0;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
&nfsv3_commit_on_close, 0, "write+commit on close, else only write");
static int nfs_clean_pages_on_close = 1;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
&nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
int nfs_directio_enable = 0;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
&nfs_directio_enable, 0, "Enable NFS directio");
/*
* This sysctl allows other processes to mmap a file that has been opened
* O_DIRECT by a process. In general, having processes mmap the file while
* Direct IO is in progress can lead to Data Inconsistencies. But, we allow
* this by default to prevent DoS attacks - to prevent a malicious user from
* opening up files O_DIRECT preventing other users from mmap'ing these
* files. "Protected" environments where stricter consistency guarantees are
* required can disable this knob. The process that opened the file O_DIRECT
* cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
* meaningful.
*/
int nfs_directio_allow_mmap = 1;
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
&nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
#if 0
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
&nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
&nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
#endif
#define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
| NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
| NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
/*
* SMP Locking Note :
* The list of locks after the description of the lock is the ordering
* of other locks acquired with the lock held.
* np->n_mtx : Protects the fields in the nfsnode.
VM Object Lock
VI_MTX (acquired indirectly)
* nmp->nm_mtx : Protects the fields in the nfsmount.
rep->r_mtx
* nfs_iod_mtx : Global lock, protects shared nfsiod state.
* nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
nmp->nm_mtx
rep->r_mtx
* rep->r_mtx : Protects the fields in an nfsreq.
*/
static int
nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
struct ucred *cred, uint32_t *retmode)
{
const int v3 = 1;
u_int32_t *tl;
int error = 0, attrflag, i, lrupos;
struct mbuf *mreq, *mrep, *md, *mb;
caddr_t bpos, dpos;
u_int32_t rmode;
struct nfsnode *np = VTONFS(vp);
nfsstats.rpccnt[NFSPROC_ACCESS]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED, M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl = txdr_unsigned(wmode);
nfsm_request(vp, NFSPROC_ACCESS, td, cred);
nfsm_postop_attr(vp, attrflag);
if (!error) {
lrupos = 0;
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
rmode = fxdr_unsigned(u_int32_t, *tl);
mtx_lock(&np->n_mtx);
for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
if (np->n_accesscache[i].uid == cred->cr_uid) {
np->n_accesscache[i].mode = rmode;
np->n_accesscache[i].stamp = time_second;
break;
}
if (i > 0 && np->n_accesscache[i].stamp <
np->n_accesscache[lrupos].stamp)
lrupos = i;
}
if (i == NFS_ACCESSCACHESIZE) {
np->n_accesscache[lrupos].uid = cred->cr_uid;
np->n_accesscache[lrupos].mode = rmode;
np->n_accesscache[lrupos].stamp = time_second;
}
mtx_unlock(&np->n_mtx);
if (retmode != NULL)
*retmode = rmode;
KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
}
m_freem(mrep);
nfsmout:
#ifdef KDTRACE_HOOKS
if (error) {
KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
error);
}
#endif
return (error);
}
1994-05-24 10:09:53 +00:00
/*
* nfs access vnode op.
* For nfs version 2, just return ok. File accesses may fail later.
* For nfs version 3, use the access rpc to check accessibility. If file modes
* are changed on the server, accesses might still fail later.
1994-05-24 10:09:53 +00:00
*/
static int
nfs_access(struct vop_access_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
int error = 0, i, gotahit;
u_int32_t mode, rmode, wmode;
int v3 = NFS_ISV3(vp);
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
/*
* Disallow write attempts on filesystems mounted read-only;
* unless the file is a socket, fifo, or a block or character
* device resident on the filesystem.
*/
if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
return (EROFS);
default:
break;
}
}
1994-05-24 10:09:53 +00:00
/*
* For nfs v3, check to see if we have done this recently, and if
* so return our cached result instead of making an ACCESS call.
* If not, do an access rpc, otherwise you are stuck emulating
1994-05-24 10:09:53 +00:00
* ufs_access() locally using the vattr. This may not be correct,
* since the server may apply other access criteria such as
* client uid-->server uid mapping that we do not know about.
1994-05-24 10:09:53 +00:00
*/
if (v3) {
if (ap->a_accmode & VREAD)
mode = NFSV3ACCESS_READ;
1994-05-24 10:09:53 +00:00
else
mode = 0;
if (vp->v_type != VDIR) {
if (ap->a_accmode & VWRITE)
mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
if (ap->a_accmode & VEXEC)
mode |= NFSV3ACCESS_EXECUTE;
} else {
if (ap->a_accmode & VWRITE)
mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
NFSV3ACCESS_DELETE);
if (ap->a_accmode & VEXEC)
mode |= NFSV3ACCESS_LOOKUP;
}
/* XXX safety belt, only make blanket request if caching */
if (nfsaccess_cache_timeout > 0) {
wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
} else {
wmode = mode;
}
/*
* Does our cached result allow us to give a definite yes to
* this request?
*/
gotahit = 0;
mtx_lock(&np->n_mtx);
for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
if (time_second < (np->n_accesscache[i].stamp +
nfsaccess_cache_timeout) &&
(np->n_accesscache[i].mode & mode) == mode) {
nfsstats.accesscache_hits++;
gotahit = 1;
}
break;
}
}
mtx_unlock(&np->n_mtx);
#ifdef KDTRACE_HOOKS
if (gotahit)
KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
ap->a_cred->cr_uid, mode);
else
KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
ap->a_cred->cr_uid, mode);
#endif
if (gotahit == 0) {
/*
* Either a no, or a don't know. Go to the wire.
*/
nfsstats.accesscache_misses++;
error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
&rmode);
if (!error) {
if ((rmode & mode) != mode)
error = EACCES;
}
}
1994-05-24 10:09:53 +00:00
return (error);
} else {
if ((error = nfsspec_access(ap)) != 0) {
return (error);
}
/*
* Attempt to prevent a mapped root from accessing a file
* which it shouldn't. We try to read a byte from the file
* if the user is root and the file is not zero length.
* After calling nfsspec_access, we should have the correct
* file size cached.
*/
mtx_lock(&np->n_mtx);
if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
&& VTONFS(vp)->n_size > 0) {
struct iovec aiov;
struct uio auio;
char buf[1];
mtx_unlock(&np->n_mtx);
aiov.iov_base = buf;
aiov.iov_len = 1;
auio.uio_iov = &aiov;
auio.uio_iovcnt = 1;
auio.uio_offset = 0;
auio.uio_resid = 1;
auio.uio_segflg = UIO_SYSSPACE;
auio.uio_rw = UIO_READ;
auio.uio_td = ap->a_td;
if (vp->v_type == VREG)
error = nfs_readrpc(vp, &auio, ap->a_cred);
else if (vp->v_type == VDIR) {
char* bp;
bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
aiov.iov_base = bp;
aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
error = nfs_readdirrpc(vp, &auio, ap->a_cred);
free(bp, M_TEMP);
} else if (vp->v_type == VLNK)
error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
else
error = EACCES;
} else
mtx_unlock(&np->n_mtx);
return (error);
}
1994-05-24 10:09:53 +00:00
}
int nfs_otw_getattr_avoid = 0;
1994-05-24 10:09:53 +00:00
/*
* nfs open vnode op
* Check to see if the type is ok
* and that deletion is not in progress.
* For paged in text files, you will need to flush the page cache
* if consistency is lost.
*/
/* ARGSUSED */
static int
nfs_open(struct vop_open_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
struct nfsnode *np = VTONFS(vp);
struct vattr vattr;
int error;
int fmode = ap->a_mode;
struct ucred *cred;
1994-05-24 10:09:53 +00:00
if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
return (EOPNOTSUPP);
/*
* Get a valid lease. If cached data is stale, flush it.
*/
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error == EINTR || error == EIO)
1994-05-24 10:09:53 +00:00
return (error);
mtx_lock(&np->n_mtx);
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
if (vp->v_type == VDIR)
np->n_direofoffset = 0;
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, ap->a_cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
} else {
mtx_unlock(&np->n_mtx);
error = VOP_GETATTR(vp, &vattr, ap->a_cred);
if (error)
return (error);
mtx_lock(&np->n_mtx);
if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
if (vp->v_type == VDIR)
np->n_direofoffset = 0;
mtx_unlock(&np->n_mtx);
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error == EINTR || error == EIO) {
1994-05-24 10:09:53 +00:00
return (error);
}
mtx_lock(&np->n_mtx);
np->n_mtime = vattr.va_mtime;
1994-05-24 10:09:53 +00:00
}
}
/*
* If the object has >= 1 O_DIRECT active opens, we disable caching.
*/
if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
if (np->n_directio_opens == 0) {
mtx_unlock(&np->n_mtx);
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
if (error)
return (error);
mtx_lock(&np->n_mtx);
np->n_flag |= NNONCACHE;
}
np->n_directio_opens++;
}
/*
* If this is an open for writing, capture a reference to the
* credentials, so they can be used by nfs_putpages(). Using
* these write credentials is preferable to the credentials of
* whatever thread happens to be doing the VOP_PUTPAGES() since
* the write RPCs are less likely to fail with EACCES.
*/
if ((fmode & FWRITE) != 0) {
cred = np->n_writecred;
np->n_writecred = crhold(ap->a_cred);
} else
cred = NULL;
mtx_unlock(&np->n_mtx);
if (cred != NULL)
crfree(cred);
vnode_create_vobject(vp, vattr.va_size, ap->a_td);
1994-05-24 10:09:53 +00:00
return (0);
}
/*
* nfs close vnode op
* What an NFS client should do upon close after writing is a debatable issue.
* Most NFS clients push delayed writes to the server upon close, basically for
* two reasons:
* 1 - So that any write errors may be reported back to the client process
* doing the close system call. By far the two most likely errors are
* NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
* 2 - To put a worst case upper bound on cache inconsistency between
* multiple clients for the file.
* There is also a consistency problem for Version 2 of the protocol w.r.t.
* not being able to tell if other clients are writing a file concurrently,
* since there is no way of knowing if the changed modify time in the reply
* is only due to the write for this client.
* (NFS Version 3 provides weak cache consistency data in the reply that
* should be sufficient to detect and handle this case.)
*
* The current code does the following:
* for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
* for NFS Version 3 - flush dirty buffers to the server but don't invalidate
* or commit them (this satisfies 1 and 2 except for the
* case where the server crashes after this close but
* before the commit RPC, which is felt to be "good
* enough". Changing the last argument to nfs_flush() to
* a 1 would force a commit operation, if it is felt a
* commit is necessary now.
1994-05-24 10:09:53 +00:00
*/
/* ARGSUSED */
static int
nfs_close(struct vop_close_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
int error = 0;
int fmode = ap->a_fflag;
1994-05-24 10:09:53 +00:00
if (vp->v_type == VREG) {
/*
* Examine and clean dirty pages, regardless of NMODIFIED.
* This closes a major hole in close-to-open consistency.
* We want to push out all dirty pages (and buffers) on
* close, regardless of whether they were dirtied by
* mmap'ed writes or via write().
*/
if (nfs_clean_pages_on_close && vp->v_object) {
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WLOCK(vp->v_object);
vm_object_page_clean(vp->v_object, 0, 0, 0);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_WUNLOCK(vp->v_object);
}
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
mtx_unlock(&np->n_mtx);
if (NFS_ISV3(vp)) {
/*
* Under NFSv3 we have dirty buffers to dispose of. We
* must flush them to the NFS server. We have the option
* of waiting all the way through the commit rpc or just
* waiting for the initial write. The default is to only
* wait through the initial write so the data is in the
* server's cache, which is roughly similar to the state
* a standard disk subsystem leaves the file in on close().
*
* We cannot clear the NMODIFIED bit in np->n_flag due to
* potential races with other processes, and certainly
* cannot clear it if we don't commit.
*/
int cm = nfsv3_commit_on_close ? 1 : 0;
error = nfs_flush(vp, MNT_WAIT, cm);
/* np->n_flag &= ~NMODIFIED; */
} else
error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
}
if (np->n_flag & NWRITEERR) {
np->n_flag &= ~NWRITEERR;
error = np->n_error;
}
mtx_unlock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
}
if (nfs_directio_enable)
KASSERT((np->n_directio_asyncwr == 0),
("nfs_close: dirty unflushed (%d) directio buffers\n",
np->n_directio_asyncwr));
if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
mtx_lock(&np->n_mtx);
KASSERT((np->n_directio_opens > 0),
("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
np->n_directio_opens--;
if (np->n_directio_opens == 0)
np->n_flag &= ~NNONCACHE;
mtx_unlock(&np->n_mtx);
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs getattr call from vfs.
*/
static int
nfs_getattr(struct vop_getattr_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct thread *td = curthread;
struct vattr *vap = ap->a_vap;
struct vattr vattr;
1994-05-24 10:09:53 +00:00
caddr_t bpos, dpos;
int error = 0;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
/*
* Update local times for special files.
*/
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
if (np->n_flag & (NACC | NUPD))
np->n_flag |= NCHG;
mtx_unlock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
/*
* First look in the cache.
*/
if (nfs_getattrcache(vp, &vattr) == 0)
goto nfsmout;
if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
nfsstats.accesscache_misses++;
nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
if (nfs_getattrcache(vp, &vattr) == 0)
goto nfsmout;
}
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_GETATTR]++;
mreq = m_get2(NFSX_FH(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
if (!error) {
nfsm_loadattr(vp, &vattr);
}
m_freem(mrep);
nfsmout:
vap->va_type = vattr.va_type;
vap->va_mode = vattr.va_mode;
vap->va_nlink = vattr.va_nlink;
vap->va_uid = vattr.va_uid;
vap->va_gid = vattr.va_gid;
vap->va_fsid = vattr.va_fsid;
vap->va_fileid = vattr.va_fileid;
vap->va_size = vattr.va_size;
vap->va_blocksize = vattr.va_blocksize;
vap->va_atime = vattr.va_atime;
vap->va_mtime = vattr.va_mtime;
vap->va_ctime = vattr.va_ctime;
vap->va_gen = vattr.va_gen;
vap->va_flags = vattr.va_flags;
vap->va_rdev = vattr.va_rdev;
vap->va_bytes = vattr.va_bytes;
vap->va_filerev = vattr.va_filerev;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs setattr call.
*/
static int
nfs_setattr(struct vop_setattr_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct vattr *vap = ap->a_vap;
struct thread *td = curthread;
int error = 0;
u_quad_t tsize;
1994-05-24 10:09:53 +00:00
#ifndef nolint
tsize = (u_quad_t)0;
#endif
/*
* Setting of flags is not supported.
*/
if (vap->va_flags != VNOVAL)
return (EOPNOTSUPP);
/*
* Disallow write attempts if the filesystem is mounted read-only.
*/
if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
(vp->v_mount->mnt_flag & MNT_RDONLY)) {
error = EROFS;
goto out;
}
if (vap->va_size != VNOVAL) {
switch (vp->v_type) {
case VDIR:
return (EISDIR);
case VCHR:
case VBLK:
case VSOCK:
case VFIFO:
if (vap->va_mtime.tv_sec == VNOVAL &&
vap->va_atime.tv_sec == VNOVAL &&
vap->va_mode == (mode_t)VNOVAL &&
vap->va_uid == (uid_t)VNOVAL &&
vap->va_gid == (gid_t)VNOVAL)
return (0);
vap->va_size = VNOVAL;
break;
default:
/*
* Disallow write attempts if the filesystem is
* mounted read-only.
*/
if (vp->v_mount->mnt_flag & MNT_RDONLY)
return (EROFS);
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
/*
* We run vnode_pager_setsize() early (why?),
* we must set np->n_size now to avoid vinvalbuf
* V_SAVE races that might setsize a lower
* value.
*/
mtx_lock(&np->n_mtx);
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
tsize = np->n_size;
mtx_unlock(&np->n_mtx);
error = nfs_meta_setsize(vp, ap->a_cred, td,
vap->va_size);
mtx_lock(&np->n_mtx);
if (np->n_flag & NMODIFIED) {
tsize = np->n_size;
mtx_unlock(&np->n_mtx);
if (vap->va_size == 0)
error = nfs_vinvalbuf(vp, 0, td, 1);
else
error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
if (error) {
vnode_pager_setsize(vp, tsize);
goto out;
}
} else
mtx_unlock(&np->n_mtx);
/*
* np->n_size has already been set to vap->va_size
* in nfs_meta_setsize(). We must set it again since
* nfs_loadattrcache() could be called through
* nfs_meta_setsize() and could modify np->n_size.
*/
mtx_lock(&np->n_mtx);
np->n_vattr.va_size = np->n_size = vap->va_size;
mtx_unlock(&np->n_mtx);
};
} else {
mtx_lock(&np->n_mtx);
if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
(np->n_flag & NMODIFIED) && vp->v_type == VREG) {
mtx_unlock(&np->n_mtx);
if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
(error == EINTR || error == EIO))
return error;
} else
mtx_unlock(&np->n_mtx);
}
error = nfs_setattrrpc(vp, vap, ap->a_cred);
if (error && vap->va_size != VNOVAL) {
mtx_lock(&np->n_mtx);
np->n_size = np->n_vattr.va_size = tsize;
vnode_pager_setsize(vp, tsize);
mtx_unlock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
}
out:
return (error);
}
/*
* Do an nfs setattr rpc.
*/
static int
nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
{
struct nfsv2_sattr *sp;
struct nfsnode *np = VTONFS(vp);
caddr_t bpos, dpos;
u_int32_t *tl;
int error = 0, i, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_SETATTR]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
nfsm_v3attrbuild(vap, TRUE);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl = nfs_false;
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
if (vap->va_mode == (mode_t)VNOVAL)
sp->sa_mode = nfs_xdrneg1;
else
sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
if (vap->va_uid == (uid_t)VNOVAL)
sp->sa_uid = nfs_xdrneg1;
else
sp->sa_uid = txdr_unsigned(vap->va_uid);
if (vap->va_gid == (gid_t)VNOVAL)
sp->sa_gid = nfs_xdrneg1;
else
sp->sa_gid = txdr_unsigned(vap->va_gid);
sp->sa_size = txdr_unsigned(vap->va_size);
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
if (v3) {
mtx_lock(&np->n_mtx);
for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
np->n_accesscache[i].stamp = 0;
mtx_unlock(&np->n_mtx);
KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
nfsm_wcc_data(vp, wccflag);
} else
nfsm_loadattr(vp, NULL);
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs lookup call, one step at a time...
* First look in cache
* If not found, unlock the directory nfsnode and do the rpc
1994-05-24 10:09:53 +00:00
*/
static int
nfs_lookup(struct vop_lookup_args *ap)
1994-05-24 10:09:53 +00:00
{
struct componentname *cnp = ap->a_cnp;
struct vnode *dvp = ap->a_dvp;
struct vnode **vpp = ap->a_vpp;
struct mount *mp = dvp->v_mount;
struct vattr dvattr, vattr;
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
struct timespec nctime;
int flags = cnp->cn_flags;
struct vnode *newvp;
1994-05-24 10:09:53 +00:00
struct nfsmount *nmp;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
long len;
nfsfh_t *fhp;
struct nfsnode *np, *newnp;
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
int v3 = NFS_ISV3(dvp);
struct thread *td = cnp->cn_thread;
Close a race with caching of -ve name lookups in the NFS client. Specifically, clients only trust -ve cache entries while the directory remains unchanged and discard any -ve cache entries for a directory when they notice that the modification time of a directory entry changes. The race involves two concurrent lookups as follows: - Thread A does a lookup for file 'foo' which sends a lookup RPC to the server. The lookup fails and the server replies. - The 'foo' file is created (either by the same client or a different client) updating the modification time on the parent directory of 'foo'. - Thread B does a lookup for a different file 'bar' which updates the cached attributes of the parent directory of 'foo' to reflect the new modification time after 'foo' was created. - Thread A finally resumes execution to parse the reply from the NFS server. It adds a -ve cache entry and sets the cached value of the directory's modification time that is used for invalidating -ve cached lookups to the new modification time set by thread B. At this point, future lookups of 'foo' will honor the -ve cached entry until the cached entry is pushed out of the name cache's LRU or the modification time of the parent directory is changed again by some other change. The fix is to read the directory's modification time before sending the lookup RPC and use that cached modification time when setting the directory's cached modification time. Also, we do not add a -ve cache entry if another thread has added -ve cache entry that set the directory's cached modification time to a newer value than the value we read before sending the lookup RPC. Reviewed by: rmacklem MFC after: 1 week
2009-10-16 19:30:48 +00:00
*vpp = NULLVP;
if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
return (EROFS);
if (dvp->v_type != VDIR)
return (ENOTDIR);
nmp = VFSTONFS(mp);
1994-05-24 10:09:53 +00:00
np = VTONFS(dvp);
if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
*vpp = NULLVP;
return (error);
}
error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
if (error > 0 && error != ENOENT)
return (error);
if (error == -1) {
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
/*
* Lookups of "." are special and always return the
* current directory. cache_lookup() already handles
* associated locking bookkeeping, etc.
*/
if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
/* XXX: Is this really correct? */
if (cnp->cn_nameiop != LOOKUP &&
(flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
return (0);
}
/*
* We only accept a positive hit in the cache if the
* change time of the file matches our cached copy.
* Otherwise, we discard the cache entry and fallback
* to doing a lookup RPC. We also only trust cache
* entries for less than nm_nametimeo seconds.
*
* To better handle stale file handles and attributes,
* clear the attribute cache of this node if it is a
* leaf component, part of an open() call, and not
* locally modified before fetching the attributes.
* This should allow stale file handles to be detected
* here where we can fall back to a LOOKUP RPC to
* recover rather than having nfs_open() detect the
* stale file handle and failing open(2) with ESTALE.
*/
newvp = *vpp;
newnp = VTONFS(newvp);
if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
(flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
!(newnp->n_flag & NMODIFIED)) {
mtx_lock(&newnp->n_mtx);
newnp->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
mtx_unlock(&newnp->n_mtx);
}
if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
timespeccmp(&vattr.va_ctime, &nctime, ==)) {
nfsstats.lookupcache_hits++;
if (cnp->cn_nameiop != LOOKUP &&
(flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
return (0);
}
cache_purge(newvp);
if (dvp != newvp)
vput(newvp);
else
vrele(newvp);
*vpp = NULLVP;
} else if (error == ENOENT) {
if (dvp->v_iflag & VI_DOOMED)
return (ENOENT);
/*
* We only accept a negative hit in the cache if the
* modification time of the parent directory matches
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
* the cached copy in the name cache entry.
* Otherwise, we discard all of the negative cache
* entries for this directory. We also only trust
* negative cache entries for up to nm_negnametimeo
* seconds.
*/
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
timespeccmp(&vattr.va_mtime, &nctime, ==)) {
nfsstats.lookupcache_hits++;
return (ENOENT);
}
cache_purge_negative(dvp);
}
Close a race with caching of -ve name lookups in the NFS client. Specifically, clients only trust -ve cache entries while the directory remains unchanged and discard any -ve cache entries for a directory when they notice that the modification time of a directory entry changes. The race involves two concurrent lookups as follows: - Thread A does a lookup for file 'foo' which sends a lookup RPC to the server. The lookup fails and the server replies. - The 'foo' file is created (either by the same client or a different client) updating the modification time on the parent directory of 'foo'. - Thread B does a lookup for a different file 'bar' which updates the cached attributes of the parent directory of 'foo' to reflect the new modification time after 'foo' was created. - Thread A finally resumes execution to parse the reply from the NFS server. It adds a -ve cache entry and sets the cached value of the directory's modification time that is used for invalidating -ve cached lookups to the new modification time set by thread B. At this point, future lookups of 'foo' will honor the -ve cached entry until the cached entry is pushed out of the name cache's LRU or the modification time of the parent directory is changed again by some other change. The fix is to read the directory's modification time before sending the lookup RPC and use that cached modification time when setting the directory's cached modification time. Also, we do not add a -ve cache entry if another thread has added -ve cache entry that set the directory's cached modification time to a newer value than the value we read before sending the lookup RPC. Reviewed by: rmacklem MFC after: 1 week
2009-10-16 19:30:48 +00:00
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
attrflag = dattrflag = 0;
1994-05-24 10:09:53 +00:00
error = 0;
newvp = NULLVP;
1994-05-24 10:09:53 +00:00
nfsstats.lookupcache_misses++;
nfsstats.rpccnt[NFSPROC_LOOKUP]++;
len = cnp->cn_namelen;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len), M_WAITOK,
MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1994-05-24 10:09:53 +00:00
if (error) {
if (v3) {
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
nfsm_postop_attr_va(dvp, dattrflag, &vattr);
m_freem(mrep);
}
goto nfsmout;
1994-05-24 10:09:53 +00:00
}
nfsm_getfh(fhp, fhsize, v3);
1994-05-24 10:09:53 +00:00
/*
* Handle RENAME case...
*/
if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
if (NFS_CMPFH(np, fhp, fhsize)) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (EISDIR);
}
error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
if (error) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (error);
}
newvp = NFSTOV(np);
if (v3) {
nfsm_postop_attr(newvp, attrflag);
nfsm_postop_attr(dvp, attrflag);
} else
nfsm_loadattr(newvp, NULL);
1994-05-24 10:09:53 +00:00
*vpp = newvp;
m_freem(mrep);
cnp->cn_flags |= SAVENAME;
return (0);
}
if (flags & ISDOTDOT) {
ltype = VOP_ISLOCKED(dvp);
error = vfs_busy(mp, MBF_NOWAIT);
if (error != 0) {
vfs_ref(mp);
VOP_UNLOCK(dvp, 0);
error = vfs_busy(mp, 0);
vn_lock(dvp, ltype | LK_RETRY);
vfs_rel(mp);
if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
vfs_unbusy(mp);
error = ENOENT;
}
if (error != 0) {
m_freem(mrep);
return (error);
}
}
VOP_UNLOCK(dvp, 0);
error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
if (error == 0)
newvp = NFSTOV(np);
vfs_unbusy(mp);
if (newvp != dvp)
vn_lock(dvp, ltype | LK_RETRY);
if (dvp->v_iflag & VI_DOOMED) {
if (error == 0) {
if (newvp == dvp)
vrele(newvp);
else
vput(newvp);
}
error = ENOENT;
}
if (error) {
m_freem(mrep);
return (error);
}
} else if (NFS_CMPFH(np, fhp, fhsize)) {
1994-05-24 10:09:53 +00:00
VREF(dvp);
newvp = dvp;
} else {
error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
if (error) {
1994-05-24 10:09:53 +00:00
m_freem(mrep);
return (error);
}
newvp = NFSTOV(np);
/*
* Flush the attribute cache when opening a leaf node
* to ensure that fresh attributes are fetched in
* nfs_open() if we are unable to fetch attributes
* from the LOOKUP reply.
*/
if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
!(np->n_flag & NMODIFIED)) {
mtx_lock(&np->n_mtx);
np->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
mtx_unlock(&np->n_mtx);
}
1994-05-24 10:09:53 +00:00
}
if (v3) {
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
nfsm_postop_attr_va(newvp, attrflag, &vattr);
nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
} else {
nfsm_loadattr(newvp, &vattr);
attrflag = 1;
}
1994-05-24 10:09:53 +00:00
if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
cnp->cn_flags |= SAVENAME;
if ((cnp->cn_flags & MAKEENTRY) &&
(cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
*vpp = newvp;
m_freem(mrep);
nfsmout:
if (error) {
if (newvp != NULLVP) {
vput(newvp);
*vpp = NULLVP;
}
if (error != ENOENT)
goto done;
/* The requested file was not found. */
if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
(flags & ISLASTCN)) {
/*
* XXX: UFS does a full VOP_ACCESS(dvp,
* VWRITE) here instead of just checking
* MNT_RDONLY.
*/
if (mp->mnt_flag & MNT_RDONLY)
return (EROFS);
cnp->cn_flags |= SAVENAME;
return (EJUSTRETURN);
}
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
if ((cnp->cn_flags & MAKEENTRY) && cnp->cn_nameiop != CREATE &&
dattrflag) {
/*
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
* Cache the modification time of the parent
* directory from the post-op attributes in
* the name cache entry. The negative cache
* entry will be ignored once the directory
* has changed. Don't bother adding the entry
* if the directory has already changed.
*/
mtx_lock(&np->n_mtx);
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
if (timespeccmp(&np->n_vattr.va_mtime,
&vattr.va_mtime, ==)) {
Close a race with caching of -ve name lookups in the NFS client. Specifically, clients only trust -ve cache entries while the directory remains unchanged and discard any -ve cache entries for a directory when they notice that the modification time of a directory entry changes. The race involves two concurrent lookups as follows: - Thread A does a lookup for file 'foo' which sends a lookup RPC to the server. The lookup fails and the server replies. - The 'foo' file is created (either by the same client or a different client) updating the modification time on the parent directory of 'foo'. - Thread B does a lookup for a different file 'bar' which updates the cached attributes of the parent directory of 'foo' to reflect the new modification time after 'foo' was created. - Thread A finally resumes execution to parse the reply from the NFS server. It adds a -ve cache entry and sets the cached value of the directory's modification time that is used for invalidating -ve cached lookups to the new modification time set by thread B. At this point, future lookups of 'foo' will honor the -ve cached entry until the cached entry is pushed out of the name cache's LRU or the modification time of the parent directory is changed again by some other change. The fix is to read the directory's modification time before sending the lookup RPC and use that cached modification time when setting the directory's cached modification time. Also, we do not add a -ve cache entry if another thread has added -ve cache entry that set the directory's cached modification time to a newer value than the value we read before sending the lookup RPC. Reviewed by: rmacklem MFC after: 1 week
2009-10-16 19:30:48 +00:00
mtx_unlock(&np->n_mtx);
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
cache_enter_time(dvp, NULL, cnp,
&vattr.va_mtime, NULL);
Close a race with caching of -ve name lookups in the NFS client. Specifically, clients only trust -ve cache entries while the directory remains unchanged and discard any -ve cache entries for a directory when they notice that the modification time of a directory entry changes. The race involves two concurrent lookups as follows: - Thread A does a lookup for file 'foo' which sends a lookup RPC to the server. The lookup fails and the server replies. - The 'foo' file is created (either by the same client or a different client) updating the modification time on the parent directory of 'foo'. - Thread B does a lookup for a different file 'bar' which updates the cached attributes of the parent directory of 'foo' to reflect the new modification time after 'foo' was created. - Thread A finally resumes execution to parse the reply from the NFS server. It adds a -ve cache entry and sets the cached value of the directory's modification time that is used for invalidating -ve cached lookups to the new modification time set by thread B. At this point, future lookups of 'foo' will honor the -ve cached entry until the cached entry is pushed out of the name cache's LRU or the modification time of the parent directory is changed again by some other change. The fix is to read the directory's modification time before sending the lookup RPC and use that cached modification time when setting the directory's cached modification time. Also, we do not add a -ve cache entry if another thread has added -ve cache entry that set the directory's cached modification time to a newer value than the value we read before sending the lookup RPC. Reviewed by: rmacklem MFC after: 1 week
2009-10-16 19:30:48 +00:00
} else
mtx_unlock(&np->n_mtx);
}
return (ENOENT);
}
done:
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* nfs read call.
* Just call nfs_bioread() to do the work.
*/
static int
nfs_read(struct vop_read_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
switch (vp->v_type) {
case VREG:
return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
case VDIR:
return (EISDIR);
default:
return (EOPNOTSUPP);
}
1994-05-24 10:09:53 +00:00
}
/*
* nfs readlink call
*/
static int
nfs_readlink(struct vop_readlink_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
1994-05-24 10:09:53 +00:00
if (vp->v_type != VLNK)
return (EINVAL);
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
return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1994-05-24 10:09:53 +00:00
}
/*
* Do a readlink rpc.
* Called by nfs_doio() from below the buffer cache.
*/
int
nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
caddr_t bpos, dpos;
int error = 0, len, attrflag;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_READLINK]++;
mreq = m_get2(NFSX_FH(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
if (v3)
nfsm_postop_attr(vp, attrflag);
if (!error) {
nfsm_strsiz(len, NFS_MAXPATHLEN);
if (len == NFS_MAXPATHLEN) {
struct nfsnode *np = VTONFS(vp);
mtx_lock(&np->n_mtx);
if (np->n_size && np->n_size < NFS_MAXPATHLEN)
len = np->n_size;
mtx_unlock(&np->n_mtx);
}
nfsm_mtouio(uiop, len);
}
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs read rpc call
* Ditto above
*/
int
nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
u_int32_t *tl;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct nfsmount *nmp;
off_t end;
int error = 0, len, retlen, tsiz, eof, attrflag;
int v3 = NFS_ISV3(vp);
int rsize;
1994-05-24 10:09:53 +00:00
#ifndef nolint
eof = 0;
#endif
1994-05-24 10:09:53 +00:00
nmp = VFSTONFS(vp->v_mount);
tsiz = uiop->uio_resid;
mtx_lock(&nmp->nm_mtx);
end = uiop->uio_offset + tsiz;
if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
mtx_unlock(&nmp->nm_mtx);
1994-05-24 10:09:53 +00:00
return (EFBIG);
}
rsize = nmp->nm_rsize;
mtx_unlock(&nmp->nm_mtx);
1994-05-24 10:09:53 +00:00
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_READ]++;
len = (tsiz > rsize) ? rsize : tsiz;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED * 3, M_WAITOK,
MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
if (v3) {
txdr_hyper(uiop->uio_offset, tl);
1994-05-24 10:09:53 +00:00
*(tl + 2) = txdr_unsigned(len);
} else {
*tl++ = txdr_unsigned(uiop->uio_offset);
*tl++ = txdr_unsigned(len);
*tl = 0;
}
nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
if (error) {
m_freem(mrep);
goto nfsmout;
}
tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
eof = fxdr_unsigned(int, *(tl + 1));
} else {
nfsm_loadattr(vp, NULL);
}
nfsm_strsiz(retlen, rsize);
1994-05-24 10:09:53 +00:00
nfsm_mtouio(uiop, retlen);
m_freem(mrep);
tsiz -= retlen;
if (v3) {
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
if (eof || retlen == 0) {
tsiz = 0;
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
}
} else if (retlen < len) {
1994-05-24 10:09:53 +00:00
tsiz = 0;
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
}
1994-05-24 10:09:53 +00:00
}
nfsmout:
return (error);
}
/*
* nfs write call
*/
int
nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
int *iomode, int *must_commit)
1994-05-24 10:09:53 +00:00
{
u_int32_t *tl;
int32_t backup;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
off_t end;
int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
int wsize;
KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
*must_commit = 0;
1994-05-24 10:09:53 +00:00
tsiz = uiop->uio_resid;
mtx_lock(&nmp->nm_mtx);
end = uiop->uio_offset + tsiz;
if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
mtx_unlock(&nmp->nm_mtx);
1994-05-24 10:09:53 +00:00
return (EFBIG);
}
wsize = nmp->nm_wsize;
mtx_unlock(&nmp->nm_mtx);
1994-05-24 10:09:53 +00:00
while (tsiz > 0) {
nfsstats.rpccnt[NFSPROC_WRITE]++;
len = (tsiz > wsize) ? wsize : tsiz;
mreq = m_get2(NFSX_FH(v3) + 5 * NFSX_UNSIGNED, M_WAITOK,
MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
txdr_hyper(uiop->uio_offset, tl);
1994-05-24 10:09:53 +00:00
tl += 2;
*tl++ = txdr_unsigned(len);
*tl++ = txdr_unsigned(*iomode);
*tl = txdr_unsigned(len);
1994-05-24 10:09:53 +00:00
} else {
u_int32_t x;
tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
/* Set both "begin" and "current" to non-garbage. */
x = txdr_unsigned((u_int32_t)uiop->uio_offset);
*tl++ = x; /* "begin offset" */
*tl++ = x; /* "current offset" */
x = txdr_unsigned(len);
*tl++ = x; /* total to this offset */
*tl = x; /* size of this write */
1994-05-24 10:09:53 +00:00
}
nfsm_uiotom(uiop, len);
nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
if (v3) {
wccflag = NFSV3_WCCCHK;
nfsm_wcc_data(vp, wccflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
+ NFSX_V3WRITEVERF);
rlen = fxdr_unsigned(int, *tl++);
if (rlen == 0) {
error = NFSERR_IO;
m_freem(mrep);
break;
} else if (rlen < len) {
backup = len - rlen;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base -
backup;
uiop->uio_iov->iov_len += backup;
uiop->uio_offset -= backup;
uiop->uio_resid += backup;
len = rlen;
}
commit = fxdr_unsigned(int, *tl++);
/*
* Return the lowest committment level
* obtained by any of the RPCs.
*/
if (committed == NFSV3WRITE_FILESYNC)
committed = commit;
else if (committed == NFSV3WRITE_DATASYNC &&
commit == NFSV3WRITE_UNSTABLE)
committed = commit;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
nmp->nm_state |= NFSSTA_HASWRITEVERF;
} else if (bcmp((caddr_t)tl,
(caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
*must_commit = 1;
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
}
mtx_unlock(&nmp->nm_mtx);
}
} else {
nfsm_loadattr(vp, NULL);
}
if (wccflag) {
mtx_lock(&(VTONFS(vp))->n_mtx);
VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
mtx_unlock(&(VTONFS(vp))->n_mtx);
}
1994-05-24 10:09:53 +00:00
m_freem(mrep);
if (error)
break;
1994-05-24 10:09:53 +00:00
tsiz -= len;
}
nfsmout:
if (DOINGASYNC(vp))
committed = NFSV3WRITE_FILESYNC;
*iomode = committed;
1994-05-24 10:09:53 +00:00
if (error)
uiop->uio_resid = tsiz;
return (error);
}
/*
* nfs mknod rpc
* For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
* mode set to specify the file type and the size field for rdev.
1994-05-24 10:09:53 +00:00
*/
static int
nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
struct vattr *vap)
1994-05-24 10:09:53 +00:00
{
struct nfsv2_sattr *sp;
u_int32_t *tl;
struct vnode *newvp = NULL;
struct nfsnode *np = NULL;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
struct mbuf *mreq, *mrep, *md, *mb;
u_int32_t rdev;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if (vap->va_type == VCHR || vap->va_type == VBLK)
rdev = txdr_unsigned(vap->va_rdev);
else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
rdev = nfs_xdrneg1;
1994-05-24 10:09:53 +00:00
else {
return (EOPNOTSUPP);
}
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1994-05-24 10:09:53 +00:00
return (error);
nfsstats.rpccnt[NFSPROC_MKNOD]++;
mreq = m_get2(NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
*tl++ = vtonfsv3_type(vap->va_type);
nfsm_v3attrbuild(vap, FALSE);
if (vap->va_type == VCHR || vap->va_type == VBLK) {
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = txdr_unsigned(major(vap->va_rdev));
*tl = txdr_unsigned(minor(vap->va_rdev));
}
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = rdev;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
if (!error) {
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (!gotvp) {
if (newvp) {
vput(newvp);
newvp = NULL;
}
error = nfs_lookitup(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
}
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
if (error) {
if (newvp)
vput(newvp);
} else {
*vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
mtx_unlock(&(VTONFS(dvp))->n_mtx);
return (error);
}
/*
* nfs mknod vop
* just call nfs_mknodrpc() to do the work.
*/
/* ARGSUSED */
static int
nfs_mknod(struct vop_mknod_args *ap)
{
return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1994-05-24 10:09:53 +00:00
}
static u_long create_verf;
1994-05-24 10:09:53 +00:00
/*
* nfs file create call
*/
static int
nfs_create(struct vop_create_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
u_int32_t *tl;
struct nfsnode *np = NULL;
struct vnode *newvp = NULL;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
/*
* Oops, not for me..
*/
if (vap->va_type == VSOCK) {
error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
return (error);
}
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1994-05-24 10:09:53 +00:00
return (error);
}
if (vap->va_vaflags & VA_EXCLUSIVE)
fmode |= O_EXCL;
again:
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_CREATE]++;
mreq = m_get2(NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
if (fmode & O_EXCL) {
*tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
#ifdef INET
CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
IN_IFADDR_RLOCK();
if (!TAILQ_EMPTY(&V_in_ifaddrhead))
*tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
else
#endif
*tl++ = create_verf;
#ifdef INET
IN_IFADDR_RUNLOCK();
CURVNET_RESTORE();
#endif
*tl = ++create_verf;
} else {
*tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
nfsm_v3attrbuild(vap, FALSE);
}
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = 0;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
if (!error) {
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (!gotvp) {
if (newvp) {
vput(newvp);
newvp = NULL;
}
error = nfs_lookitup(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
}
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
if (error) {
if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
fmode &= ~O_EXCL;
goto again;
}
if (newvp)
vput(newvp);
} else if (v3 && (fmode & O_EXCL)) {
/*
* We are normally called with only a partially initialized
* VAP. Since the NFSv3 spec says that server may use the
* file attributes to store the verifier, the spec requires
* us to do a SETATTR RPC. FreeBSD servers store the verifier
* in atime, but we can't really assume that all servers will
* so we ensure that our SETATTR sets both atime and mtime.
*/
if (vap->va_mtime.tv_sec == VNOVAL)
vfs_timestamp(&vap->va_mtime);
if (vap->va_atime.tv_sec == VNOVAL)
vap->va_atime = vap->va_mtime;
error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
if (error)
vput(newvp);
}
if (!error) {
*ap->a_vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
mtx_unlock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs file remove call
* To try and make nfs semantics closer to ufs semantics, a file that has
* other processes using the vnode is renamed instead of removed and then
* removed later on the last close.
* - If v_usecount > 1
* If a rename is not already in the works
* call nfs_sillyrename() to set it up
* else
* do the remove rpc
*/
static int
nfs_remove(struct vop_remove_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
int error = 0;
struct vattr vattr;
1994-05-24 10:09:53 +00:00
KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
if (vp->v_type == VDIR)
error = EPERM;
else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
!VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1994-05-24 10:09:53 +00:00
/*
* Purge the name cache so that the chance of a lookup for
* the name succeeding while the remove is in progress is
* minimized. Without node locking it can still happen, such
* that an I/O op returns ESTALE, but since you get this if
* another host removes the file..
*/
cache_purge(vp);
/*
* throw away biocache buffers, mainly to avoid
* unnecessary delayed writes later.
1994-05-24 10:09:53 +00:00
*/
error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1994-05-24 10:09:53 +00:00
/* Do the rpc */
if (error != EINTR && error != EIO)
error = nfs_removerpc(dvp, cnp->cn_nameptr,
cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1994-05-24 10:09:53 +00:00
/*
* Kludge City: If the first reply to the remove rpc is lost..
* the reply to the retransmitted request will be ENOENT
* since the file was in fact removed
* Therefore, we cheat and return success.
*/
if (error == ENOENT)
error = 0;
} else if (!np->n_sillyrename)
error = nfs_sillyrename(dvp, vp, cnp);
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
np->n_attrstamp = 0;
mtx_unlock(&np->n_mtx);
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs file remove rpc called from nfs_inactive
*/
int
nfs_removeit(struct sillyrename *sp)
1994-05-24 10:09:53 +00:00
{
/*
* Make sure that the directory vnode is still valid.
* XXX we should lock sp->s_dvp here.
*/
if (sp->s_dvp->v_type == VBAD)
return (0);
return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
NULL));
}
/*
* Nfs remove rpc, called from nfs_remove() and nfs_removeit().
*/
static int
nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
struct ucred *cred, struct thread *td)
{
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_REMOVE]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen),
M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
mtx_unlock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs file rename call
*/
static int
nfs_rename(struct vop_rename_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *fvp = ap->a_fvp;
struct vnode *tvp = ap->a_tvp;
struct vnode *fdvp = ap->a_fdvp;
struct vnode *tdvp = ap->a_tdvp;
struct componentname *tcnp = ap->a_tcnp;
struct componentname *fcnp = ap->a_fcnp;
int error;
1994-05-24 10:09:53 +00:00
KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
(fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1994-05-24 10:09:53 +00:00
/* Check for cross-device rename */
if ((fvp->v_mount != tdvp->v_mount) ||
(tvp && (fvp->v_mount != tvp->v_mount))) {
error = EXDEV;
goto out;
}
if (fvp == tvp) {
nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
error = 0;
goto out;
}
if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
goto out;
/*
* We have to flush B_DELWRI data prior to renaming
* the file. If we don't, the delayed-write buffers
* can be flushed out later after the file has gone stale
* under NFSV3. NFSV2 does not have this problem because
* ( as far as I can tell ) it flushes dirty buffers more
* often.
*
* Skip the rename operation if the fsync fails, this can happen
* due to the server's volume being full, when we pushed out data
* that was written back to our cache earlier. Not checking for
* this condition can result in potential (silent) data loss.
*/
error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
VOP_UNLOCK(fvp, 0);
if (!error && tvp)
error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
if (error)
goto out;
/*
* If the tvp exists and is in use, sillyrename it before doing the
* rename of the new file over it.
* XXX Can't sillyrename a directory.
*/
if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
vput(tvp);
tvp = NULL;
}
error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
tcnp->cn_thread);
1994-05-24 10:09:53 +00:00
if (fvp->v_type == VDIR) {
if (tvp != NULL && tvp->v_type == VDIR)
cache_purge(tdvp);
cache_purge(fdvp);
}
1994-05-24 10:09:53 +00:00
out:
if (tdvp == tvp)
vrele(tdvp);
else
vput(tdvp);
if (tvp)
vput(tvp);
vrele(fdvp);
vrele(fvp);
/*
* Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
*/
if (error == ENOENT)
error = 0;
return (error);
}
/*
* nfs file rename rpc called from nfs_remove() above
*/
static int
nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
struct sillyrename *sp)
{
return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
}
/*
* Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
*/
static int
nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
struct thread *td)
1994-05-24 10:09:53 +00:00
{
caddr_t bpos, dpos;
int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(fdvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_RENAME]++;
mreq = m_get2((NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
nfsm_rndup(tnamelen), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(fdvp, v3);
nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
nfsm_fhtom(tdvp, v3);
nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
if (v3) {
nfsm_wcc_data(fdvp, fwccflag);
nfsm_wcc_data(tdvp, twccflag);
}
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(fdvp))->n_mtx);
VTONFS(fdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(fdvp))->n_mtx);
mtx_lock(&(VTONFS(tdvp))->n_mtx);
VTONFS(tdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(tdvp))->n_mtx);
if (!fwccflag) {
VTONFS(fdvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
}
if (!twccflag) {
VTONFS(tdvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs hard link create call
*/
static int
nfs_link(struct vop_link_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *tdvp = ap->a_tdvp;
struct componentname *cnp = ap->a_cnp;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
struct mbuf *mreq, *mrep, *md, *mb;
int v3;
1994-05-24 10:09:53 +00:00
if (vp->v_mount != tdvp->v_mount) {
return (EXDEV);
}
/*
* Push all writes to the server, so that the attribute cache
* doesn't get "out of sync" with the server.
* XXX There should be a better way!
*/
VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
v3 = NFS_ISV3(vp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_LINK]++;
mreq = m_get2(NFSX_FH(v3)*2 + NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
nfsm_fhtom(tdvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
nfsm_wcc_data(tdvp, wccflag);
}
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(tdvp))->n_mtx);
VTONFS(tdvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(tdvp))->n_mtx);
if (!attrflag) {
VTONFS(vp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
}
if (!wccflag) {
VTONFS(tdvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs symbolic link create call
*/
static int
nfs_symlink(struct vop_symlink_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
caddr_t bpos, dpos;
int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
struct mbuf *mreq, *mrep, *md, *mb;
struct vnode *newvp = NULL;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_SYMLINK]++;
slen = strlen(ap->a_target);
mreq = m_get2(NFSX_FH(v3) + 2*NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3),
M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
if (v3) {
nfsm_v3attrbuild(vap, FALSE);
}
1994-05-24 10:09:53 +00:00
nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
if (!v3) {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = nfs_xdrneg1;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
/*
* Issue the NFS request and get the rpc response.
*
* Only NFSv3 responses returning an error of 0 actually return
* a file handle that can be converted into newvp without having
* to do an extra lookup rpc.
*/
nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
if (v3) {
if (error == 0)
nfsm_mtofh(dvp, newvp, v3, gotvp);
nfsm_wcc_data(dvp, wccflag);
}
/*
* out code jumps -> here, mrep is also freed.
*/
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
/*
Remove hacks from the NFSv2/3 client intended to handle a lack of a server-side RPC retranmission cache for non-idempotent operations: these hacks substituted 0 (success) for the expected EEXIST in the event that a target name already existed for LINK, SYMLINK, and MKDIR operations, under the assumption that EEXIST represented a second application of the original RPC rather than a true failure. Background: certain NFS operations (in this case, LINK, SYMLINK, and MKDIR) are not idempotent, as they leave behind persisting state on the server that prevents them from being replayed without an error;if an UDP RPC reply is lost leading to a retransmission by theclient, the second reply will return EEXIST rather than success, asthe new object has already been created. The NFS client previouslysilently mapped the EEXIST return into success to paper over thisproblem. However, in all modern NFS server implementations, a reply cache is kept in order to retransmit the original reply to a retransmitted request, rather than performing the operation a second time, allowing this hack to be avoided. This allows link()-based filelocking over NFS to operate correctly, as an application requestingthe creation of a new link for a file to tell if it succeededatomically or not. Other NFS clients, including Solaris and Linux, generally follow this behavior for the same reasons. Most clients also now default to TCP, which also helps avoid the issue of retransmitted but non-idempotent requests in most cases. Reported by: Adam McDougall <mcdouga9 at egr dot msu dot edu>, Timo Sirainen <tss at iki dot fi> Reviewed by: mohans MFC after: 1 week
2007-11-19 16:03:21 +00:00
* If we do not have an error and we could not extract the newvp from
* the response due to the request being NFSv2, we have to do a
* lookup in order to obtain a newvp to return.
*/
if (error == 0 && newvp == NULL) {
struct nfsnode *np = NULL;
error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
cnp->cn_cred, cnp->cn_thread, &np);
if (!error)
newvp = NFSTOV(np);
}
if (error) {
if (newvp)
vput(newvp);
} else {
*ap->a_vpp = newvp;
}
mtx_lock(&(VTONFS(dvp))->n_mtx);
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs make dir call
*/
static int
nfs_mkdir(struct vop_mkdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *dvp = ap->a_dvp;
struct vattr *vap = ap->a_vap;
struct componentname *cnp = ap->a_cnp;
struct nfsv2_sattr *sp;
int len;
struct nfsnode *np = NULL;
struct vnode *newvp = NULL;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
int gotvp = 0;
struct mbuf *mreq, *mrep, *md, *mb;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1994-05-24 10:09:53 +00:00
return (error);
len = cnp->cn_namelen;
nfsstats.rpccnt[NFSPROC_MKDIR]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) +
NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
if (v3) {
nfsm_v3attrbuild(vap, FALSE);
1994-05-24 10:09:53 +00:00
} else {
sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
sp->sa_uid = nfs_xdrneg1;
sp->sa_gid = nfs_xdrneg1;
sp->sa_size = nfs_xdrneg1;
txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1994-05-24 10:09:53 +00:00
}
nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
if (!error)
nfsm_mtofh(dvp, newvp, v3, gotvp);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
Remove hacks from the NFSv2/3 client intended to handle a lack of a server-side RPC retranmission cache for non-idempotent operations: these hacks substituted 0 (success) for the expected EEXIST in the event that a target name already existed for LINK, SYMLINK, and MKDIR operations, under the assumption that EEXIST represented a second application of the original RPC rather than a true failure. Background: certain NFS operations (in this case, LINK, SYMLINK, and MKDIR) are not idempotent, as they leave behind persisting state on the server that prevents them from being replayed without an error;if an UDP RPC reply is lost leading to a retransmission by theclient, the second reply will return EEXIST rather than success, asthe new object has already been created. The NFS client previouslysilently mapped the EEXIST return into success to paper over thisproblem. However, in all modern NFS server implementations, a reply cache is kept in order to retransmit the original reply to a retransmitted request, rather than performing the operation a second time, allowing this hack to be avoided. This allows link()-based filelocking over NFS to operate correctly, as an application requestingthe creation of a new link for a file to tell if it succeededatomically or not. Other NFS clients, including Solaris and Linux, generally follow this behavior for the same reasons. Most clients also now default to TCP, which also helps avoid the issue of retransmitted but non-idempotent requests in most cases. Reported by: Adam McDougall <mcdouga9 at egr dot msu dot edu>, Timo Sirainen <tss at iki dot fi> Reviewed by: mohans MFC after: 1 week
2007-11-19 16:03:21 +00:00
if (error == 0 && newvp == NULL) {
error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
cnp->cn_thread, &np);
if (!error) {
newvp = NFSTOV(np);
if (newvp->v_type != VDIR)
error = EEXIST;
1994-05-24 10:09:53 +00:00
}
}
if (error) {
if (newvp)
vput(newvp);
} else
*ap->a_vpp = newvp;
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* nfs remove directory call
*/
static int
nfs_rmdir(struct vop_rmdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct vnode *dvp = ap->a_dvp;
struct componentname *cnp = ap->a_cnp;
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
if (dvp == vp)
return (EINVAL);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_RMDIR]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED +
nfsm_rndup(cnp->cn_namelen), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
1994-05-24 10:09:53 +00:00
nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
if (v3)
nfsm_wcc_data(dvp, wccflag);
m_freem(mrep);
nfsmout:
mtx_lock(&(VTONFS(dvp))->n_mtx);
1994-05-24 10:09:53 +00:00
VTONFS(dvp)->n_flag |= NMODIFIED;
mtx_unlock(&(VTONFS(dvp))->n_mtx);
if (!wccflag) {
VTONFS(dvp)->n_attrstamp = 0;
KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
}
1994-05-24 10:09:53 +00:00
cache_purge(dvp);
cache_purge(vp);
/*
* Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
*/
if (error == ENOENT)
error = 0;
return (error);
}
/*
* nfs readdir call
*/
static int
nfs_readdir(struct vop_readdir_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
struct uio *uio = ap->a_uio;
int tresid, error = 0;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
if (vp->v_type != VDIR)
return(EPERM);
1994-05-24 10:09:53 +00:00
/*
* First, check for hit on the EOF offset cache
*/
if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
1994-05-24 10:09:53 +00:00
(np->n_flag & NMODIFIED) == 0) {
if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
mtx_lock(&np->n_mtx);
if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
mtx_unlock(&np->n_mtx);
nfsstats.direofcache_hits++;
goto out;
} else
mtx_unlock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
}
}
/*
* Call nfs_bioread() to do the real work.
*/
tresid = uio->uio_resid;
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
error = nfs_bioread(vp, uio, 0, ap->a_cred);
1994-05-24 10:09:53 +00:00
if (!error && uio->uio_resid == tresid) {
1994-05-24 10:09:53 +00:00
nfsstats.direofcache_misses++;
}
out:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Readdir rpc call.
* Called from below the buffer cache by nfs_doio().
*/
int
nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
int len, left;
struct dirent *dp = NULL;
u_int32_t *tl;
caddr_t cp;
nfsuint64 *cookiep;
caddr_t bpos, dpos;
struct mbuf *mreq, *mrep, *md, *mb;
nfsuint64 cookie;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct nfsnode *dnp = VTONFS(vp);
u_quad_t fileno;
int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
int attrflag;
int v3 = NFS_ISV3(vp);
KASSERT(uiop->uio_iovcnt == 1 &&
(uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
(uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
("nfs readdirrpc bad uio"));
1994-05-24 10:09:53 +00:00
/*
* If there is no cookie, assume directory was stale.
*/
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep) {
cookie = *cookiep;
nfs_dircookie_unlock(dnp);
} else {
nfs_dircookie_unlock(dnp);
return (NFSERR_BAD_COOKIE);
}
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
1994-05-24 10:09:53 +00:00
* The stopping criteria is EOF or buffer full.
*/
while (more_dirs && bigenough) {
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_READDIR]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_READDIR(v3), M_WAITOK,
MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, v3);
if (v3) {
tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
mtx_lock(&dnp->n_mtx);
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
mtx_unlock(&dnp->n_mtx);
} else {
tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
}
*tl = txdr_unsigned(nmp->nm_readdirsize);
nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
if (v3) {
nfsm_postop_attr(vp, attrflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
mtx_lock(&dnp->n_mtx);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl;
mtx_unlock(&dnp->n_mtx);
} else {
m_freem(mrep);
goto nfsmout;
}
}
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
1994-05-24 10:09:53 +00:00
/* loop thru the dir entries, doctoring them to 4bsd form */
while (more_dirs && bigenough) {
if (v3) {
tl = nfsm_dissect(u_int32_t *,
3 * NFSX_UNSIGNED);
fileno = fxdr_hyper(tl);
len = fxdr_unsigned(int, *(tl + 2));
} else {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
fileno = fxdr_unsigned(u_quad_t, *tl++);
len = fxdr_unsigned(int, *tl);
}
1994-05-24 10:09:53 +00:00
if (len <= 0 || len > NFS_MAXNAMLEN) {
error = EBADRPC;
m_freem(mrep);
goto nfsmout;
}
tlen = nfsm_rndup(len);
if (tlen == len)
tlen += 4; /* To ensure null termination */
left = DIRBLKSIZ - blksiz;
if ((tlen + DIRHDSIZ) > left) {
dp->d_reclen += left;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
blksiz = 0;
1994-05-24 10:09:53 +00:00
}
if ((tlen + DIRHDSIZ) > uiop->uio_resid)
bigenough = 0;
if (bigenough) {
dp = (struct dirent *)uiop->uio_iov->iov_base;
dp->d_fileno = (int)fileno;
dp->d_namlen = len;
dp->d_reclen = tlen + DIRHDSIZ;
dp->d_type = DT_UNKNOWN;
blksiz += dp->d_reclen;
if (blksiz == DIRBLKSIZ)
blksiz = 0;
uiop->uio_offset += DIRHDSIZ;
uiop->uio_resid -= DIRHDSIZ;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + DIRHDSIZ;
uiop->uio_iov->iov_len -= DIRHDSIZ;
nfsm_mtouio(uiop, len);
cp = uiop->uio_iov->iov_base;
tlen -= len;
*cp = '\0'; /* null terminate */
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + tlen;
uiop->uio_iov->iov_len -= tlen;
uiop->uio_offset += tlen;
uiop->uio_resid -= tlen;
} else
nfsm_adv(nfsm_rndup(len));
if (v3) {
tl = nfsm_dissect(u_int32_t *,
3 * NFSX_UNSIGNED);
} else {
tl = nfsm_dissect(u_int32_t *,
2 * NFSX_UNSIGNED);
}
if (bigenough) {
cookie.nfsuquad[0] = *tl++;
if (v3)
cookie.nfsuquad[1] = *tl++;
} else if (v3)
tl += 2;
else
tl++;
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = (fxdr_unsigned(int, *tl) == 0);
}
m_freem(mrep);
}
/*
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
1994-05-24 10:09:53 +00:00
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
}
/*
* We are now either at the end of the directory or have filled the
* block.
*/
if (bigenough)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
nfs_printf("EEK! readdirrpc resid > 0\n");
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
nfs_dircookie_unlock(dnp);
1994-05-24 10:09:53 +00:00
}
nfsmout:
return (error);
}
/*
* NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
1994-05-24 10:09:53 +00:00
*/
int
nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1994-05-24 10:09:53 +00:00
{
int len, left;
struct dirent *dp;
u_int32_t *tl;
caddr_t cp;
struct vnode *newvp;
nfsuint64 *cookiep;
caddr_t bpos, dpos, dpossav1, dpossav2;
struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
1994-05-24 10:09:53 +00:00
struct nameidata nami, *ndp = &nami;
struct componentname *cnp = &ndp->ni_cnd;
nfsuint64 cookie;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
struct nfsnode *dnp = VTONFS(vp), *np;
struct vattr vattr, dvattr;
nfsfh_t *fhp;
u_quad_t fileno;
int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
int attrflag, dattrflag, fhsize;
#ifndef nolint
dp = NULL;
#endif
KASSERT(uiop->uio_iovcnt == 1 &&
(uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
(uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
("nfs readdirplusrpc bad uio"));
1994-05-24 10:09:53 +00:00
ndp->ni_dvp = vp;
newvp = NULLVP;
/*
* If there is no cookie, assume directory was stale.
*/
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
if (cookiep) {
cookie = *cookiep;
nfs_dircookie_unlock(dnp);
} else {
nfs_dircookie_unlock(dnp);
return (NFSERR_BAD_COOKIE);
}
1994-05-24 10:09:53 +00:00
/*
* Loop around doing readdir rpc's of size nm_readdirsize
* truncated to a multiple of DIRBLKSIZ.
1994-05-24 10:09:53 +00:00
* The stopping criteria is EOF or buffer full.
*/
while (more_dirs && bigenough) {
nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
mreq = m_get2(NFSX_FH(1) + 6 * NFSX_UNSIGNED, M_WAITOK,
MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, 1);
tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
*tl++ = cookie.nfsuquad[0];
*tl++ = cookie.nfsuquad[1];
mtx_lock(&dnp->n_mtx);
*tl++ = dnp->n_cookieverf.nfsuquad[0];
*tl++ = dnp->n_cookieverf.nfsuquad[1];
mtx_unlock(&dnp->n_mtx);
*tl++ = txdr_unsigned(nmp->nm_readdirsize);
*tl = txdr_unsigned(nmp->nm_rsize);
nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
nfsm_postop_attr_va(vp, dattrflag, &dvattr);
if (error) {
m_freem(mrep);
goto nfsmout;
}
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
mtx_lock(&dnp->n_mtx);
dnp->n_cookieverf.nfsuquad[0] = *tl++;
dnp->n_cookieverf.nfsuquad[1] = *tl++;
mtx_unlock(&dnp->n_mtx);
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
1995-05-30 08:16:23 +00:00
1994-05-24 10:09:53 +00:00
/* loop thru the dir entries, doctoring them to 4bsd form */
while (more_dirs && bigenough) {
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
fileno = fxdr_hyper(tl);
len = fxdr_unsigned(int, *(tl + 2));
1994-05-24 10:09:53 +00:00
if (len <= 0 || len > NFS_MAXNAMLEN) {
error = EBADRPC;
m_freem(mrep);
goto nfsmout;
}
tlen = nfsm_rndup(len);
if (tlen == len)
tlen += 4; /* To ensure null termination*/
left = DIRBLKSIZ - blksiz;
if ((tlen + DIRHDSIZ) > left) {
dp->d_reclen += left;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
blksiz = 0;
}
1994-05-24 10:09:53 +00:00
if ((tlen + DIRHDSIZ) > uiop->uio_resid)
bigenough = 0;
if (bigenough) {
1994-05-24 10:09:53 +00:00
dp = (struct dirent *)uiop->uio_iov->iov_base;
dp->d_fileno = (int)fileno;
1994-05-24 10:09:53 +00:00
dp->d_namlen = len;
dp->d_reclen = tlen + DIRHDSIZ;
dp->d_type = DT_UNKNOWN;
blksiz += dp->d_reclen;
if (blksiz == DIRBLKSIZ)
blksiz = 0;
uiop->uio_offset += DIRHDSIZ;
1994-05-24 10:09:53 +00:00
uiop->uio_resid -= DIRHDSIZ;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + DIRHDSIZ;
1994-05-24 10:09:53 +00:00
uiop->uio_iov->iov_len -= DIRHDSIZ;
cnp->cn_nameptr = uiop->uio_iov->iov_base;
cnp->cn_namelen = len;
nfsm_mtouio(uiop, len);
cp = uiop->uio_iov->iov_base;
tlen -= len;
*cp = '\0';
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + tlen;
1994-05-24 10:09:53 +00:00
uiop->uio_iov->iov_len -= tlen;
uiop->uio_offset += tlen;
1994-05-24 10:09:53 +00:00
uiop->uio_resid -= tlen;
} else
nfsm_adv(nfsm_rndup(len));
tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
if (bigenough) {
cookie.nfsuquad[0] = *tl++;
cookie.nfsuquad[1] = *tl++;
} else
tl += 2;
/*
* Since the attributes are before the file handle
* (sigh), we must skip over the attributes and then
* come back and get them.
*/
attrflag = fxdr_unsigned(int, *tl);
if (attrflag) {
dpossav1 = dpos;
mdsav1 = md;
nfsm_adv(NFSX_V3FATTR);
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
doit = fxdr_unsigned(int, *tl);
/*
* Skip loading the attrs for "..". There's a
* race between loading the attrs here and
* lookups that look for the directory currently
* being read (in the parent). We try to acquire
* the exclusive lock on ".." here, owning the
* lock on the directory being read. Lookup will
* hold the lock on ".." and try to acquire the
* lock on the directory being read.
*
* There are other ways of fixing this, one would
* be to do a trylock on the ".." vnode and skip
* loading the attrs on ".." if it happens to be
* locked by another process. But skipping the
* attrload on ".." seems the easiest option.
*/
if (strcmp(dp->d_name, "..") == 0) {
doit = 0;
/*
* We've already skipped over the attrs,
* skip over the filehandle. And store d_type
* as VDIR.
*/
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
i = fxdr_unsigned(int, *tl);
nfsm_adv(nfsm_rndup(i));
dp->d_type = IFTODT(VTTOIF(VDIR));
}
if (doit) {
nfsm_getfh(fhp, fhsize, 1);
if (NFS_CMPFH(dnp, fhp, fhsize)) {
VREF(vp);
newvp = vp;
np = dnp;
} else {
error = nfs_nget(vp->v_mount, fhp,
fhsize, &np, LK_EXCLUSIVE);
if (error)
doit = 0;
else
newvp = NFSTOV(np);
}
}
if (doit && bigenough) {
dpossav2 = dpos;
dpos = dpossav1;
mdsav2 = md;
md = mdsav1;
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
nfsm_loadattr(newvp, &vattr);
dpos = dpossav2;
md = mdsav2;
Close a race in NFS lookup processing that could result in stale name cache entries on one client when a directory was renamed on another client. The root cause for the stale entry being trusted is that each per-vnode nfsnode structure has a single 'n_ctime' timestamp used to validate positive name cache entries. However, if there are multiple entries for a single vnode, they all share a single timestamp. To fix this, extend the name cache to allow filesystems to optionally store a timestamp value in each name cache entry. The NFS clients now fetch the timestamp associated with each name cache entry and use that to validate cache hits instead of the timestamps previously stored in the nfsnode. Another part of the fix is that the NFS clients now use timestamps from the post-op attributes of RPCs when adding name cache entries rather than pulling the timestamps out of the file's attribute cache. The latter is subject to races with other lookups updating the attribute cache concurrently. Some more details: - Add a variant of nfsm_postop_attr() to the old NFS client that can return a vattr structure with a copy of the post-op attributes. - Handle lookups of "." as a special case in the NFS clients since the name cache does not store name cache entries for ".", so we cannot get a useful timestamp. It didn't really make much sense to recheck the attributes on the the directory to validate the namecache hit for "." anyway. - ABI compat shims for the name cache routines are present in this commit so that it is safe to MFC. MFC after: 2 weeks
2012-01-20 20:02:01 +00:00
dp->d_type = IFTODT(VTTOIF(vattr.va_type));
ndp->ni_vp = newvp;
if (newvp->v_type != VDIR || dattrflag != 0)
cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
cnp, &vattr.va_ctime,
newvp->v_type != VDIR ? NULL :
&dvattr.va_ctime);
}
1994-05-24 10:09:53 +00:00
} else {
/* Just skip over the file handle */
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
i = fxdr_unsigned(int, *tl);
if (i) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
fhsize = fxdr_unsigned(int, *tl);
nfsm_adv(nfsm_rndup(fhsize));
}
1994-05-24 10:09:53 +00:00
}
if (newvp != NULLVP) {
if (newvp == vp)
vrele(newvp);
else
vput(newvp);
newvp = NULLVP;
1994-05-24 10:09:53 +00:00
}
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = fxdr_unsigned(int, *tl);
}
/*
* If at end of rpc data, get the eof boolean
*/
if (!more_dirs) {
tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
1994-05-24 10:09:53 +00:00
more_dirs = (fxdr_unsigned(int, *tl) == 0);
}
m_freem(mrep);
}
/*
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
* Fill last record, iff any, out to a multiple of DIRBLKSIZ
1994-05-24 10:09:53 +00:00
* by increasing d_reclen for the last record.
*/
if (blksiz > 0) {
left = DIRBLKSIZ - blksiz;
dp->d_reclen += left;
uiop->uio_iov->iov_base =
(char *)uiop->uio_iov->iov_base + left;
uiop->uio_iov->iov_len -= left;
uiop->uio_offset += left;
uiop->uio_resid -= left;
}
/*
* We are now either at the end of the directory or have filled the
* block.
*/
if (bigenough)
dnp->n_direofoffset = uiop->uio_offset;
else {
if (uiop->uio_resid > 0)
nfs_printf("EEK! readdirplusrpc resid > 0\n");
nfs_dircookie_lock(dnp);
cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
*cookiep = cookie;
nfs_dircookie_unlock(dnp);
1994-05-24 10:09:53 +00:00
}
nfsmout:
if (newvp != NULLVP) {
if (newvp == vp)
vrele(newvp);
else
vput(newvp);
newvp = NULLVP;
}
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Silly rename. To make the NFS filesystem that is stateless look a little
* more like the "ufs" a remove of an active vnode is translated to a rename
* to a funny looking filename that is removed by nfs_inactive on the
* nfsnode. There is the potential for another process on a different client
* to create the same funny name between the nfs_lookitup() fails and the
* nfs_rename() completes, but...
*/
static int
nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
1994-05-24 10:09:53 +00:00
{
struct sillyrename *sp;
struct nfsnode *np;
1994-05-24 10:09:53 +00:00
int error;
short pid;
unsigned int lticks;
1994-05-24 10:09:53 +00:00
cache_purge(dvp);
np = VTONFS(vp);
KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
sp = malloc(sizeof (struct sillyrename),
M_NFSREQ, M_WAITOK);
sp->s_cred = crhold(cnp->cn_cred);
1994-05-24 10:09:53 +00:00
sp->s_dvp = dvp;
sp->s_removeit = nfs_removeit;
1994-05-24 10:09:53 +00:00
VREF(dvp);
/*
* Fudge together a funny name.
* Changing the format of the funny name to accomodate more
* sillynames per directory.
* The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
* CPU ticks since boot.
*/
pid = cnp->cn_thread->td_proc->p_pid;
lticks = (unsigned int)ticks;
for ( ; ; ) {
sp->s_namlen = sprintf(sp->s_name,
".nfs.%08x.%04x4.4", lticks,
pid);
if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
cnp->cn_thread, NULL))
break;
lticks++;
1994-05-24 10:09:53 +00:00
}
error = nfs_renameit(dvp, cnp, sp);
if (error)
1994-05-24 10:09:53 +00:00
goto bad;
error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
cnp->cn_thread, &np);
1994-05-24 10:09:53 +00:00
np->n_sillyrename = sp;
return (0);
bad:
vrele(sp->s_dvp);
crfree(sp->s_cred);
free((caddr_t)sp, M_NFSREQ);
return (error);
}
/*
* Look up a file name and optionally either update the file handle or
* allocate an nfsnode, depending on the value of npp.
* npp == NULL --> just do the lookup
* *npp == NULL --> allocate a new nfsnode and make sure attributes are
* handled too
* *npp != NULL --> update the file handle in the vnode
1994-05-24 10:09:53 +00:00
*/
static int
nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
struct thread *td, struct nfsnode **npp)
1994-05-24 10:09:53 +00:00
{
struct vnode *newvp = NULL;
struct nfsnode *np, *dnp = VTONFS(dvp);
caddr_t bpos, dpos;
int error = 0, fhlen, attrflag;
struct mbuf *mreq, *mrep, *md, *mb;
nfsfh_t *nfhp;
int v3 = NFS_ISV3(dvp);
1994-05-24 10:09:53 +00:00
nfsstats.rpccnt[NFSPROC_LOOKUP]++;
mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len),
M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(dvp, v3);
nfsm_strtom(name, len, NFS_MAXNAMLEN);
nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
if (npp && !error) {
nfsm_getfh(nfhp, fhlen, v3);
if (*npp) {
np = *npp;
if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
free((caddr_t)np->n_fhp, M_NFSBIGFH);
np->n_fhp = &np->n_fh;
} else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
np->n_fhsize = fhlen;
newvp = NFSTOV(np);
} else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
VREF(dvp);
newvp = dvp;
} else {
error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
if (error) {
m_freem(mrep);
return (error);
}
newvp = NFSTOV(np);
}
if (v3) {
nfsm_postop_attr(newvp, attrflag);
if (!attrflag && *npp == NULL) {
m_freem(mrep);
if (newvp == dvp)
vrele(newvp);
else
vput(newvp);
return (ENOENT);
}
} else
nfsm_loadattr(newvp, NULL);
1994-05-24 10:09:53 +00:00
}
m_freem(mrep);
nfsmout:
if (npp && *npp == NULL) {
if (error) {
if (newvp) {
if (newvp == dvp)
vrele(newvp);
else
vput(newvp);
}
} else
*npp = np;
}
return (error);
}
/*
* Nfs Version 3 commit rpc
*/
int
nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
struct thread *td)
{
u_int32_t *tl;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
caddr_t bpos, dpos;
int error = 0, wccflag = NFSV3_WCCRATTR;
struct mbuf *mreq, *mrep, *md, *mb;
mtx_lock(&nmp->nm_mtx);
if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
mtx_unlock(&nmp->nm_mtx);
return (0);
}
mtx_unlock(&nmp->nm_mtx);
nfsstats.rpccnt[NFSPROC_COMMIT]++;
mreq = m_get2(NFSX_FH(1), M_WAITOK, MT_DATA, 0);
mb = mreq;
bpos = mtod(mb, caddr_t);
nfsm_fhtom(vp, 1);
tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
txdr_hyper(offset, tl);
tl += 2;
*tl = txdr_unsigned(cnt);
nfsm_request(vp, NFSPROC_COMMIT, td, cred);
nfsm_wcc_data(vp, wccflag);
if (!error) {
tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
NFSX_V3WRITEVERF)) {
bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
NFSX_V3WRITEVERF);
error = NFSERR_STALEWRITEVERF;
}
1994-05-24 10:09:53 +00:00
}
m_freem(mrep);
nfsmout:
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* Strategy routine.
* For async requests when nfsiod(s) are running, queue the request by
* calling nfs_asyncio(), otherwise just all nfs_doio() to do the
* request.
*/
static int
nfs_strategy(struct vop_strategy_args *ap)
1994-05-24 10:09:53 +00:00
{
struct buf *bp = ap->a_bp;
1994-05-24 10:09:53 +00:00
struct ucred *cr;
KASSERT(!(bp->b_flags & B_DONE),
("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
if (bp->b_iocmd == BIO_READ)
1994-05-24 10:09:53 +00:00
cr = bp->b_rcred;
else
cr = bp->b_wcred;
1994-05-24 10:09:53 +00:00
/*
* If the op is asynchronous and an i/o daemon is waiting
* queue the request, wake it up and wait for completion
* otherwise just do it ourselves.
*/
if ((bp->b_flags & B_ASYNC) == 0 ||
nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
(void)nfs_doio(ap->a_vp, bp, cr, curthread);
return (0);
1994-05-24 10:09:53 +00:00
}
/*
* fsync vnode op. Just call nfs_flush() with commit == 1.
1994-05-24 10:09:53 +00:00
*/
/* ARGSUSED */
static int
nfs_fsync(struct vop_fsync_args *ap)
1994-05-24 10:09:53 +00:00
{
return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
}
/*
* Flush all the blocks associated with a vnode.
* Walk through the buffer pool and push any dirty pages
* associated with the vnode.
*/
static int
nfs_flush(struct vnode *vp, int waitfor, int commit)
{
struct nfsnode *np = VTONFS(vp);
struct buf *bp;
int i;
1994-05-24 10:09:53 +00:00
struct buf *nbp;
struct nfsmount *nmp = VFSTONFS(vp->v_mount);
int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
int passone = 1;
u_quad_t off, endoff, toff;
struct ucred* wcred = NULL;
struct buf **bvec = NULL;
struct bufobj *bo;
struct thread *td = curthread;
#ifndef NFS_COMMITBVECSIZ
#define NFS_COMMITBVECSIZ 20
#endif
struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
int bvecsize = 0, bveccount;
1994-05-24 10:09:53 +00:00
if (nmp->nm_flag & NFSMNT_INT)
slpflag = PCATCH;
if (!commit)
passone = 0;
bo = &vp->v_bufobj;
/*
* A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2005-12-28 10:03:48 +00:00
* server, but has not been committed to stable storage on the server
* yet. On the first pass, the byte range is worked out and the commit
* rpc is done. On the second pass, nfs_writebp() is called to do the
* job.
*/
again:
off = (u_quad_t)-1;
endoff = 0;
bvecpos = 0;
if (NFS_ISV3(vp) && commit) {
if (bvec != NULL && bvec != bvec_on_stack)
free(bvec, M_TEMP);
/*
* Count up how many buffers waiting for a commit.
*/
bveccount = 0;
BO_LOCK(bo);
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (!BUF_ISLOCKED(bp) &&
(bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
== (B_DELWRI | B_NEEDCOMMIT))
bveccount++;
}
/*
* Allocate space to remember the list of bufs to commit. It is
* important to use M_NOWAIT here to avoid a race with nfs_write.
* If we can't get memory (for whatever reason), we will end up
* committing the buffers one-by-one in the loop below.
*/
if (bveccount > NFS_COMMITBVECSIZ) {
/*
* Release the vnode interlock to avoid a lock
* order reversal.
*/
BO_UNLOCK(bo);
bvec = (struct buf **)
malloc(bveccount * sizeof(struct buf *),
M_TEMP, M_NOWAIT);
BO_LOCK(bo);
if (bvec == NULL) {
bvec = bvec_on_stack;
bvecsize = NFS_COMMITBVECSIZ;
} else
bvecsize = bveccount;
} else {
bvec = bvec_on_stack;
bvecsize = NFS_COMMITBVECSIZ;
}
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (bvecpos >= bvecsize)
break;
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
nbp = TAILQ_NEXT(bp, b_bobufs);
continue;
}
if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
(B_DELWRI | B_NEEDCOMMIT)) {
BUF_UNLOCK(bp);
nbp = TAILQ_NEXT(bp, b_bobufs);
continue;
}
BO_UNLOCK(bo);
bremfree(bp);
/*
* Work out if all buffers are using the same cred
* so we can deal with them all with one commit.
*
* NOTE: we are not clearing B_DONE here, so we have
* to do it later on in this routine if we intend to
* initiate I/O on the bp.
*
* Note: to avoid loopback deadlocks, we do not
* assign b_runningbufspace.
*/
if (wcred == NULL)
wcred = bp->b_wcred;
else if (wcred != bp->b_wcred)
wcred = NOCRED;
vfs_busy_pages(bp, 1);
BO_LOCK(bo);
/*
* bp is protected by being locked, but nbp is not
* and vfs_busy_pages() may sleep. We have to
* recalculate nbp.
*/
nbp = TAILQ_NEXT(bp, b_bobufs);
/*
* A list of these buffers is kept so that the
* second loop knows which buffers have actually
* been committed. This is necessary, since there
* may be a race between the commit rpc and new
* uncommitted writes on the file.
*/
bvec[bvecpos++] = bp;
toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
bp->b_dirtyoff;
if (toff < off)
off = toff;
toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
if (toff > endoff)
endoff = toff;
}
BO_UNLOCK(bo);
}
if (bvecpos > 0) {
/*
* Commit data on the server, as required.
* If all bufs are using the same wcred, then use that with
* one call for all of them, otherwise commit each one
* separately.
*/
if (wcred != NOCRED)
retv = nfs_commit(vp, off, (int)(endoff - off),
wcred, td);
else {
retv = 0;
for (i = 0; i < bvecpos; i++) {
off_t off, size;
bp = bvec[i];
off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
bp->b_dirtyoff;
size = (u_quad_t)(bp->b_dirtyend
- bp->b_dirtyoff);
retv = nfs_commit(vp, off, (int)size,
bp->b_wcred, td);
if (retv) break;
}
}
if (retv == NFSERR_STALEWRITEVERF)
nfs_clearcommit(vp->v_mount);
/*
* Now, either mark the blocks I/O done or mark the
* blocks dirty, depending on whether the commit
* succeeded.
*/
for (i = 0; i < bvecpos; i++) {
bp = bvec[i];
2004-09-15 21:49:22 +00:00
bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
if (retv) {
/*
* Error, leave B_DELWRI intact
*/
vfs_unbusy_pages(bp);
brelse(bp);
} else {
/*
* Success, remove B_DELWRI ( bundirty() ).
*
* b_dirtyoff/b_dirtyend seem to be NFS
* specific. We should probably move that
* into bundirty(). XXX
*/
bufobj_wref(bo);
bp->b_flags |= B_ASYNC;
bundirty(bp);
bp->b_flags &= ~B_DONE;
bp->b_ioflags &= ~BIO_ERROR;
bp->b_dirtyoff = bp->b_dirtyend = 0;
bufdone(bp);
}
}
}
/*
* Start/do any write(s) that are required.
*/
1994-05-24 10:09:53 +00:00
loop:
BO_LOCK(bo);
TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
if (waitfor != MNT_WAIT || passone)
1994-05-24 10:09:53 +00:00
continue;
error = BUF_TIMELOCK(bp,
LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
if (error == 0) {
BUF_UNLOCK(bp);
goto loop;
}
if (error == ENOLCK) {
error = 0;
goto loop;
}
if (nfs_sigintr(nmp, td)) {
error = EINTR;
goto done;
}
if (slpflag == PCATCH) {
1994-05-24 10:09:53 +00:00
slpflag = 0;
slptimeo = 2 * hz;
}
goto loop;
}
if ((bp->b_flags & B_DELWRI) == 0)
panic("nfs_fsync: not dirty");
if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
BUF_UNLOCK(bp);
continue;
}
BO_UNLOCK(bo);
1994-05-24 10:09:53 +00:00
bremfree(bp);
if (passone || !commit)
bp->b_flags |= B_ASYNC;
else
2004-09-15 21:49:22 +00:00
bp->b_flags |= B_ASYNC;
bwrite(bp);
if (nfs_sigintr(nmp, td)) {
error = EINTR;
goto done;
}
1994-05-24 10:09:53 +00:00
goto loop;
}
if (passone) {
passone = 0;
BO_UNLOCK(bo);
goto again;
}
if (waitfor == MNT_WAIT) {
while (bo->bo_numoutput) {
error = bufobj_wwait(bo, slpflag, slptimeo);
1994-05-24 10:09:53 +00:00
if (error) {
BO_UNLOCK(bo);
error = nfs_sigintr(nmp, td);
if (error)
goto done;
if (slpflag == PCATCH) {
1994-05-24 10:09:53 +00:00
slpflag = 0;
slptimeo = 2 * hz;
}
BO_LOCK(bo);
1994-05-24 10:09:53 +00:00
}
}
if (bo->bo_dirty.bv_cnt != 0 && commit) {
BO_UNLOCK(bo);
1994-05-24 10:09:53 +00:00
goto loop;
}
/*
* Wait for all the async IO requests to drain
*/
BO_UNLOCK(bo);
mtx_lock(&np->n_mtx);
while (np->n_directio_asyncwr > 0) {
np->n_flag |= NFSYNCWAIT;
error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
&np->n_mtx, slpflag | (PRIBIO + 1),
"nfsfsync", 0);
if (error) {
if (nfs_sigintr(nmp, td)) {
mtx_unlock(&np->n_mtx);
error = EINTR;
goto done;
}
}
}
mtx_unlock(&np->n_mtx);
} else
BO_UNLOCK(bo);
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
if (np->n_flag & NWRITEERR) {
error = np->n_error;
np->n_flag &= ~NWRITEERR;
}
if (commit && bo->bo_dirty.bv_cnt == 0 &&
bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
np->n_flag &= ~NMODIFIED;
mtx_unlock(&np->n_mtx);
done:
if (bvec != NULL && bvec != bvec_on_stack)
free(bvec, M_TEMP);
1994-05-24 10:09:53 +00:00
return (error);
}
/*
* NFS advisory byte-level locks.
*/
static int
nfs_advlock(struct vop_advlock_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
u_quad_t size;
int error;
error = vn_lock(vp, LK_SHARED);
if (error)
return (error);
if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
size = VTONFS(vp)->n_size;
VOP_UNLOCK(vp, 0);
error = lf_advlock(ap, &(vp->v_lockf), size);
} else {
if (nfs_advlock_p)
error = nfs_advlock_p(ap);
else
error = ENOLCK;
}
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
return (error);
}
/*
* NFS advisory byte-level locks.
*/
static int
nfs_advlockasync(struct vop_advlockasync_args *ap)
{
struct vnode *vp = ap->a_vp;
u_quad_t size;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
int error;
error = vn_lock(vp, LK_SHARED);
if (error)
return (error);
if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
size = VTONFS(vp)->n_size;
VOP_UNLOCK(vp, 0);
error = lf_advlockasync(ap, &(vp->v_lockf), size);
} else {
VOP_UNLOCK(vp, 0);
error = EOPNOTSUPP;
Add the new kernel-mode NFS Lock Manager. To use it instead of the user-mode lock manager, build a kernel with the NFSLOCKD option and add '-k' to 'rpc_lockd_flags' in rc.conf. Highlights include: * Thread-safe kernel RPC client - many threads can use the same RPC client handle safely with replies being de-multiplexed at the socket upcall (typically driven directly by the NIC interrupt) and handed off to whichever thread matches the reply. For UDP sockets, many RPC clients can share the same socket. This allows the use of a single privileged UDP port number to talk to an arbitrary number of remote hosts. * Single-threaded kernel RPC server. Adding support for multi-threaded server would be relatively straightforward and would follow approximately the Solaris KPI. A single thread should be sufficient for the NLM since it should rarely block in normal operation. * Kernel mode NLM server supporting cancel requests and granted callbacks. I've tested the NLM server reasonably extensively - it passes both my own tests and the NFS Connectathon locking tests running on Solaris, Mac OS X and Ubuntu Linux. * Userland NLM client supported. While the NLM server doesn't have support for the local NFS client's locking needs, it does have to field async replies and granted callbacks from remote NLMs that the local client has contacted. We relay these replies to the userland rpc.lockd over a local domain RPC socket. * Robust deadlock detection for the local lock manager. In particular it will detect deadlocks caused by a lock request that covers more than one blocking request. As required by the NLM protocol, all deadlock detection happens synchronously - a user is guaranteed that if a lock request isn't rejected immediately, the lock will eventually be granted. The old system allowed for a 'deferred deadlock' condition where a blocked lock request could wake up and find that some other deadlock-causing lock owner had beaten them to the lock. * Since both local and remote locks are managed by the same kernel locking code, local and remote processes can safely use file locks for mutual exclusion. Local processes have no fairness advantage compared to remote processes when contending to lock a region that has just been unlocked - the local lock manager enforces a strict first-come first-served model for both local and remote lockers. Sponsored by: Isilon Systems PR: 95247 107555 115524 116679 MFC after: 2 weeks
2008-03-26 15:23:12 +00:00
}
return (error);
1994-05-24 10:09:53 +00:00
}
/*
* Print out the contents of an nfsnode.
*/
static int
nfs_print(struct vop_print_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
nfs_printf("\tfileid %ld fsid 0x%x",
np->n_vattr.va_fileid, np->n_vattr.va_fsid);
1994-05-24 10:09:53 +00:00
if (vp->v_type == VFIFO)
fifo_printinfo(vp);
printf("\n");
return (0);
1994-05-24 10:09:53 +00:00
}
/*
* This is the "real" nfs::bwrite(struct buf*).
* We set B_CACHE if this is a VMIO buffer.
*/
int
2004-09-15 21:49:22 +00:00
nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
{
int s;
int oldflags = bp->b_flags;
#if 0
int retv = 1;
off_t off;
#endif
- Add real assertions to lockmgr locking primitives. A couple of notes for this: * WITNESS support, when enabled, is only used for shared locks in order to avoid problems with the "disowned" locks * KA_HELD and KA_UNHELD only exists in the lockmgr namespace in order to assert for a generic thread (not curthread) owning or not the lock. Really, this kind of check is bogus but it seems very widespread in the consumers code. So, for the moment, we cater this untrusted behaviour, until the consumers are not fixed and the options could be removed (hopefully during 8.0-CURRENT lifecycle) * Implementing KA_HELD and KA_UNHELD (not surported natively by WITNESS) made necessary the introduction of LA_MASKASSERT which specifies the range for default lock assertion flags * About other aspects, lockmgr_assert() follows exactly what other locking primitives offer about this operation. - Build real assertions for buffer cache locks on the top of lockmgr_assert(). They can be used with the BUF_ASSERT_*(bp) paradigm. - Add checks at lock destruction time and use a cookie for verifying lock integrity at any operation. - Redefine BUF_LOCKFREE() in order to not use a direct assert but let it rely on the aforementioned destruction time check. KPI results evidently broken, so __FreeBSD_version bumping and manpage update result necessary and will be committed soon. Side note: lockmgr_assert() will be used soon in order to implement real assertions in the vnode namespace replacing the legacy and still bogus "VOP_ISLOCKED()" way. Tested by: kris (earlier version) Reviewed by: jhb
2008-02-13 20:44:19 +00:00
BUF_ASSERT_HELD(bp);
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
if (bp->b_flags & B_INVAL) {
brelse(bp);
return(0);
}
bp->b_flags |= B_CACHE;
This mega-commit is meant to fix numerous interrelated problems. There has been some bitrot and incorrect assumptions in the vfs_bio code. These problems have manifest themselves worse on NFS type filesystems, but can still affect local filesystems under certain circumstances. Most of the problems have involved mmap consistancy, and as a side-effect broke the vfs.ioopt code. This code might have been committed seperately, but almost everything is interrelated. 1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that are fully valid. 2) Rather than deactivating erroneously read initial (header) pages in kern_exec, we now free them. 3) Fix the rundown of non-VMIO buffers that are in an inconsistent (missing vp) state. 4) Fix the disassociation of pages from buffers in brelse. The previous code had rotted and was faulty in a couple of important circumstances. 5) Remove a gratuitious buffer wakeup in vfs_vmio_release. 6) Remove a crufty and currently unused cluster mechanism for VBLK files in vfs_bio_awrite. When the code is functional, I'll add back a cleaner version. 7) The page busy count wakeups assocated with the buffer cache usage were incorrectly cleaned up in a previous commit by me. Revert to the original, correct version, but with a cleaner implementation. 8) The cluster read code now tries to keep data associated with buffers more aggressively (without breaking the heuristics) when it is presumed that the read data (buffers) will be soon needed. 9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The delay loop waiting is not useful for filesystem locks, due to the length of the time intervals. 10) Correct and clean-up spec_getpages. 11) Implement a fully functional nfs_getpages, nfs_putpages. 12) Fix nfs_write so that modifications are coherent with the NFS data on the server disk (at least as well as NFS seems to allow.) 13) Properly support MS_INVALIDATE on NFS. 14) Properly pass down MS_INVALIDATE to lower levels of the VM code from vm_map_clean. 15) Better support the notion of pages being busy but valid, so that fewer in-transit waits occur. (use p->busy more for pageouts instead of PG_BUSY.) Since the page is fully valid, it is still usable for reads. 16) It is possible (in error) for cached pages to be busy. Make the page allocation code handle that case correctly. (It should probably be a printf or panic, but I want the system to handle coding errors robustly. I'll probably add a printf.) 17) Correct the design and usage of vm_page_sleep. It didn't handle consistancy problems very well, so make the design a little less lofty. After vm_page_sleep, if it ever blocked, it is still important to relookup the page (if the object generation count changed), and verify it's status (always.) 18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up. 19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush. 20) Fix vm_pager_put_pages and it's descendents to support an int flag instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
/*
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
* Undirty the bp. We will redirty it later if the I/O fails.
*/
s = splbio();
bundirty(bp);
bp->b_flags &= ~B_DONE;
bp->b_ioflags &= ~BIO_ERROR;
bp->b_iocmd = BIO_WRITE;
bufobj_wref(bp->b_bufobj);
curthread->td_ru.ru_oublock++;
splx(s);
/*
* Note: to avoid loopback deadlocks, we do not
* assign b_runningbufspace.
*/
vfs_busy_pages(bp, 1);
BUF_KERNPROC(bp);
bp->b_iooffset = dbtob(bp->b_blkno);
bstrategy(bp);
if( (oldflags & B_ASYNC) == 0) {
int rtval = bufwait(bp);
if (oldflags & B_DELWRI) {
s = splbio();
reassignbuf(bp);
splx(s);
}
brelse(bp);
return (rtval);
}
return (0);
}
1994-05-24 10:09:53 +00:00
/*
* nfs special file access vnode op.
* Essentially just get vattr and then imitate iaccess() since the device is
* local to the client.
*/
static int
nfsspec_access(struct vop_access_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vattr *vap;
struct ucred *cred = ap->a_cred;
struct vnode *vp = ap->a_vp;
accmode_t accmode = ap->a_accmode;
1994-05-24 10:09:53 +00:00
struct vattr vattr;
int error;
/*
* Disallow write attempts on filesystems mounted read-only;
* unless the file is a socket, fifo, or a block or character
* device resident on the filesystem.
*/
if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
switch (vp->v_type) {
case VREG:
case VDIR:
case VLNK:
return (EROFS);
default:
break;
}
}
1994-05-24 10:09:53 +00:00
vap = &vattr;
error = VOP_GETATTR(vp, vap, cred);
if (error)
goto out;
error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
accmode, cred, NULL);
out:
return error;
1994-05-24 10:09:53 +00:00
}
/*
* Read wrapper for fifos.
*/
static int
nfsfifo_read(struct vop_read_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
int error;
1994-05-24 10:09:53 +00:00
/*
* Set access flag.
*/
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
np->n_flag |= NACC;
vfs_timestamp(&np->n_atim);
mtx_unlock(&np->n_mtx);
error = fifo_specops.vop_read(ap);
return error;
1994-05-24 10:09:53 +00:00
}
/*
* Write wrapper for fifos.
*/
static int
nfsfifo_write(struct vop_write_args *ap)
1994-05-24 10:09:53 +00:00
{
struct nfsnode *np = VTONFS(ap->a_vp);
1994-05-24 10:09:53 +00:00
/*
* Set update flag.
*/
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
np->n_flag |= NUPD;
vfs_timestamp(&np->n_mtim);
mtx_unlock(&np->n_mtx);
return(fifo_specops.vop_write(ap));
1994-05-24 10:09:53 +00:00
}
/*
* Close wrapper for fifos.
*
* Update the times on the nfsnode then do fifo close.
*/
static int
nfsfifo_close(struct vop_close_args *ap)
1994-05-24 10:09:53 +00:00
{
struct vnode *vp = ap->a_vp;
struct nfsnode *np = VTONFS(vp);
1994-05-24 10:09:53 +00:00
struct vattr vattr;
struct timespec ts;
1994-05-24 10:09:53 +00:00
mtx_lock(&np->n_mtx);
1994-05-24 10:09:53 +00:00
if (np->n_flag & (NACC | NUPD)) {
vfs_timestamp(&ts);
if (np->n_flag & NACC)
np->n_atim = ts;
if (np->n_flag & NUPD)
np->n_mtim = ts;
1994-05-24 10:09:53 +00:00
np->n_flag |= NCHG;
if (vrefcnt(vp) == 1 &&
1994-05-24 10:09:53 +00:00
(vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
VATTR_NULL(&vattr);
if (np->n_flag & NACC)
vattr.va_atime = np->n_atim;
if (np->n_flag & NUPD)
vattr.va_mtime = np->n_mtim;
mtx_unlock(&np->n_mtx);
(void)VOP_SETATTR(vp, &vattr, ap->a_cred);
goto out;
1994-05-24 10:09:53 +00:00
}
}
mtx_unlock(&np->n_mtx);
out:
return (fifo_specops.vop_close(ap));
1994-05-24 10:09:53 +00:00
}
This fixes a large number of bugs in our NFS client side code. A recent commit by Kirk also fixed a softupdates bug that could easily be triggered by server side NFS. * An edge case with shared R+W mmap()'s and truncate whereby the system would inappropriately clear the dirty bits on still-dirty data. (applicable to all filesystems) THIS FIX TEMPORARILY DISABLED PENDING FURTHER TESTING. see vm/vm_page.c line 1641 * The straddle case for VM pages and buffer cache buffers when truncating. (applicable to NFS client side) * Possible SMP database corruption due to vm_pager_unmap_page() not clearing the TLB for the other cpu's. (applicable to NFS client side but could effect all filesystems). Note: not considered serious since the corruption occurs beyond the file EOF. * When flusing a dirty buffer due to B_CACHE getting cleared, we were accidently setting B_CACHE again (that is, bwrite() sets B_CACHE), when we really want it to stay clear after the write is complete. This resulted in a corrupt buffer. (applicable to all filesystems but probably only triggered by NFS) * We have to call vtruncbuf() when ftruncate()ing to remove any buffer cache buffers. This is still tentitive, I may be able to remove it due to the second bug fix. (applicable to NFS client side) * vnode_pager_setsize() race against nfs_vinvalbuf()... we have to set n_size before calling nfs_vinvalbuf or the NFS code may recursively vnode_pager_setsize() to the original value before the truncate. This is what was causing the user mmap bus faults in the nfs tester program. (applicable to NFS client side) * Fix to softupdates (see ufs/ffs/ffs_inode.c 1.73, commit made by Kirk). Testing program written by: Avadis Tevanian, Jr. Testing program supplied by: jkh / Apple (see Dec2001 posting to freebsd-hackers with Subject 'NFS: How to make FreeBS fall on its face in one easy step') MFC after: 1 week
2001-12-14 01:16:57 +00:00
/*
* Just call nfs_writebp() with the force argument set to 1.
*
* NOTE: B_DONE may or may not be set in a_bp on call.
*/
static int
nfs_bwrite(struct buf *bp)
{
return (nfs_writebp(bp, 1, curthread));
}
struct buf_ops buf_ops_nfs = {
.bop_name = "buf_ops_nfs",
.bop_write = nfs_bwrite,
.bop_strategy = bufstrategy,
.bop_sync = bufsync,
.bop_bdflush = bufbdflush,
};