2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)nfs_vnops.c 8.16 (Berkeley) 5/27/95
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
39 * vnode op calls for Sun NFS version 2 and 3
43 #include "opt_kdtrace.h"
45 #include <sys/param.h>
46 #include <sys/kernel.h>
47 #include <sys/systm.h>
48 #include <sys/resourcevar.h>
50 #include <sys/mount.h>
54 #include <sys/malloc.h>
56 #include <sys/namei.h>
57 #include <sys/socket.h>
58 #include <sys/vnode.h>
59 #include <sys/dirent.h>
60 #include <sys/fcntl.h>
61 #include <sys/lockf.h>
63 #include <sys/sysctl.h>
64 #include <sys/signalvar.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
70 #include <nfs/nfsproto.h>
71 #include <nfsclient/nfs.h>
72 #include <nfsclient/nfsnode.h>
73 #include <nfsclient/nfsmount.h>
74 #include <nfs/nfs_kdtrace.h>
75 #include <nfs/nfs_lock.h>
76 #include <nfs/xdr_subs.h>
77 #include <nfsclient/nfsm_subs.h>
80 #include <netinet/in.h>
81 #include <netinet/in_var.h>
83 #include <machine/stdarg.h>
86 #include <sys/dtrace_bsd.h>
88 dtrace_nfsclient_accesscache_flush_probe_func_t
89 dtrace_nfsclient_accesscache_flush_done_probe;
90 uint32_t nfsclient_accesscache_flush_done_id;
92 dtrace_nfsclient_accesscache_get_probe_func_t
93 dtrace_nfsclient_accesscache_get_hit_probe,
94 dtrace_nfsclient_accesscache_get_miss_probe;
95 uint32_t nfsclient_accesscache_get_hit_id;
96 uint32_t nfsclient_accesscache_get_miss_id;
98 dtrace_nfsclient_accesscache_load_probe_func_t
99 dtrace_nfsclient_accesscache_load_done_probe;
100 uint32_t nfsclient_accesscache_load_done_id;
101 #endif /* !KDTRACE_HOOKS */
108 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
109 * calls are not in getblk() and brelse() so that they would not be necessary
113 #define vfs_busy_pages(bp, f)
116 static vop_read_t nfsfifo_read;
117 static vop_write_t nfsfifo_write;
118 static vop_close_t nfsfifo_close;
119 static int nfs_flush(struct vnode *, int, int);
120 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *);
121 static vop_lookup_t nfs_lookup;
122 static vop_create_t nfs_create;
123 static vop_mknod_t nfs_mknod;
124 static vop_open_t nfs_open;
125 static vop_close_t nfs_close;
126 static vop_access_t nfs_access;
127 static vop_getattr_t nfs_getattr;
128 static vop_setattr_t nfs_setattr;
129 static vop_read_t nfs_read;
130 static vop_fsync_t nfs_fsync;
131 static vop_remove_t nfs_remove;
132 static vop_link_t nfs_link;
133 static vop_rename_t nfs_rename;
134 static vop_mkdir_t nfs_mkdir;
135 static vop_rmdir_t nfs_rmdir;
136 static vop_symlink_t nfs_symlink;
137 static vop_readdir_t nfs_readdir;
138 static vop_strategy_t nfs_strategy;
139 static int nfs_lookitup(struct vnode *, const char *, int,
140 struct ucred *, struct thread *, struct nfsnode **);
141 static int nfs_sillyrename(struct vnode *, struct vnode *,
142 struct componentname *);
143 static vop_access_t nfsspec_access;
144 static vop_readlink_t nfs_readlink;
145 static vop_print_t nfs_print;
146 static vop_advlock_t nfs_advlock;
147 static vop_advlockasync_t nfs_advlockasync;
150 * Global vfs data structures for nfs
152 struct vop_vector nfs_vnodeops = {
153 .vop_default = &default_vnodeops,
154 .vop_access = nfs_access,
155 .vop_advlock = nfs_advlock,
156 .vop_advlockasync = nfs_advlockasync,
157 .vop_close = nfs_close,
158 .vop_create = nfs_create,
159 .vop_fsync = nfs_fsync,
160 .vop_getattr = nfs_getattr,
161 .vop_getpages = nfs_getpages,
162 .vop_putpages = nfs_putpages,
163 .vop_inactive = nfs_inactive,
164 .vop_link = nfs_link,
165 .vop_lookup = nfs_lookup,
166 .vop_mkdir = nfs_mkdir,
167 .vop_mknod = nfs_mknod,
168 .vop_open = nfs_open,
169 .vop_print = nfs_print,
170 .vop_read = nfs_read,
171 .vop_readdir = nfs_readdir,
172 .vop_readlink = nfs_readlink,
173 .vop_reclaim = nfs_reclaim,
174 .vop_remove = nfs_remove,
175 .vop_rename = nfs_rename,
176 .vop_rmdir = nfs_rmdir,
177 .vop_setattr = nfs_setattr,
178 .vop_strategy = nfs_strategy,
179 .vop_symlink = nfs_symlink,
180 .vop_write = nfs_write,
183 struct vop_vector nfs_fifoops = {
184 .vop_default = &fifo_specops,
185 .vop_access = nfsspec_access,
186 .vop_close = nfsfifo_close,
187 .vop_fsync = nfs_fsync,
188 .vop_getattr = nfs_getattr,
189 .vop_inactive = nfs_inactive,
190 .vop_print = nfs_print,
191 .vop_read = nfsfifo_read,
192 .vop_reclaim = nfs_reclaim,
193 .vop_setattr = nfs_setattr,
194 .vop_write = nfsfifo_write,
197 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
198 struct componentname *cnp, struct vattr *vap);
199 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
200 struct ucred *cred, struct thread *td);
201 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
202 int fnamelen, struct vnode *tdvp,
203 const char *tnameptr, int tnamelen,
204 struct ucred *cred, struct thread *td);
205 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
206 struct sillyrename *sp);
211 struct mtx nfs_iod_mtx;
212 enum nfsiod_state nfs_iodwant[NFS_MAXASYNCDAEMON];
213 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
214 int nfs_numasync = 0;
215 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
217 SYSCTL_DECL(_vfs_oldnfs);
219 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
220 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
221 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
223 static int nfs_prime_access_cache = 0;
224 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
225 &nfs_prime_access_cache, 0,
226 "Prime NFS ACCESS cache when fetching attributes");
228 static int nfsv3_commit_on_close = 0;
229 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
230 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
232 static int nfs_clean_pages_on_close = 1;
233 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
234 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
236 int nfs_directio_enable = 0;
237 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
238 &nfs_directio_enable, 0, "Enable NFS directio");
241 * This sysctl allows other processes to mmap a file that has been opened
242 * O_DIRECT by a process. In general, having processes mmap the file while
243 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
244 * this by default to prevent DoS attacks - to prevent a malicious user from
245 * opening up files O_DIRECT preventing other users from mmap'ing these
246 * files. "Protected" environments where stricter consistency guarantees are
247 * required can disable this knob. The process that opened the file O_DIRECT
248 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
251 int nfs_directio_allow_mmap = 1;
252 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
253 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
256 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
257 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
259 SYSCTL_INT(_vfs_oldnfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
260 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
263 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
264 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
265 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
269 * The list of locks after the description of the lock is the ordering
270 * of other locks acquired with the lock held.
271 * np->n_mtx : Protects the fields in the nfsnode.
273 VI_MTX (acquired indirectly)
274 * nmp->nm_mtx : Protects the fields in the nfsmount.
276 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
277 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
280 * rep->r_mtx : Protects the fields in an nfsreq.
284 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
285 struct ucred *cred, uint32_t *retmode)
289 int error = 0, attrflag, i, lrupos;
291 struct mbuf *mreq, *mrep, *md, *mb;
294 struct nfsnode *np = VTONFS(vp);
296 nfsstats.rpccnt[NFSPROC_ACCESS]++;
297 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED, M_WAITOK, MT_DATA, 0);
299 bpos = mtod(mb, caddr_t);
301 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
302 *tl = txdr_unsigned(wmode);
303 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
304 nfsm_postop_attr(vp, attrflag);
307 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
308 rmode = fxdr_unsigned(u_int32_t, *tl);
309 mtx_lock(&np->n_mtx);
310 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
311 if (np->n_accesscache[i].uid == cred->cr_uid) {
312 np->n_accesscache[i].mode = rmode;
313 np->n_accesscache[i].stamp = time_second;
316 if (i > 0 && np->n_accesscache[i].stamp <
317 np->n_accesscache[lrupos].stamp)
320 if (i == NFS_ACCESSCACHESIZE) {
321 np->n_accesscache[lrupos].uid = cred->cr_uid;
322 np->n_accesscache[lrupos].mode = rmode;
323 np->n_accesscache[lrupos].stamp = time_second;
325 mtx_unlock(&np->n_mtx);
328 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
334 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
342 * nfs access vnode op.
343 * For nfs version 2, just return ok. File accesses may fail later.
344 * For nfs version 3, use the access rpc to check accessibility. If file modes
345 * are changed on the server, accesses might still fail later.
348 nfs_access(struct vop_access_args *ap)
350 struct vnode *vp = ap->a_vp;
351 int error = 0, i, gotahit;
352 u_int32_t mode, rmode, wmode;
353 int v3 = NFS_ISV3(vp);
354 struct nfsnode *np = VTONFS(vp);
357 * Disallow write attempts on filesystems mounted read-only;
358 * unless the file is a socket, fifo, or a block or character
359 * device resident on the filesystem.
361 if ((ap->a_accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
362 switch (vp->v_type) {
372 * For nfs v3, check to see if we have done this recently, and if
373 * so return our cached result instead of making an ACCESS call.
374 * If not, do an access rpc, otherwise you are stuck emulating
375 * ufs_access() locally using the vattr. This may not be correct,
376 * since the server may apply other access criteria such as
377 * client uid-->server uid mapping that we do not know about.
380 if (ap->a_accmode & VREAD)
381 mode = NFSV3ACCESS_READ;
384 if (vp->v_type != VDIR) {
385 if (ap->a_accmode & VWRITE)
386 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
387 if (ap->a_accmode & VEXEC)
388 mode |= NFSV3ACCESS_EXECUTE;
390 if (ap->a_accmode & VWRITE)
391 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
393 if (ap->a_accmode & VEXEC)
394 mode |= NFSV3ACCESS_LOOKUP;
396 /* XXX safety belt, only make blanket request if caching */
397 if (nfsaccess_cache_timeout > 0) {
398 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
399 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
400 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
406 * Does our cached result allow us to give a definite yes to
410 mtx_lock(&np->n_mtx);
411 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
412 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
413 if (time_second < (np->n_accesscache[i].stamp +
414 nfsaccess_cache_timeout) &&
415 (np->n_accesscache[i].mode & mode) == mode) {
416 nfsstats.accesscache_hits++;
422 mtx_unlock(&np->n_mtx);
425 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
426 ap->a_cred->cr_uid, mode);
428 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
429 ap->a_cred->cr_uid, mode);
433 * Either a no, or a don't know. Go to the wire.
435 nfsstats.accesscache_misses++;
436 error = nfs3_access_otw(vp, wmode, ap->a_td, ap->a_cred,
439 if ((rmode & mode) != mode)
445 if ((error = nfsspec_access(ap)) != 0) {
449 * Attempt to prevent a mapped root from accessing a file
450 * which it shouldn't. We try to read a byte from the file
451 * if the user is root and the file is not zero length.
452 * After calling nfsspec_access, we should have the correct
455 mtx_lock(&np->n_mtx);
456 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
457 && VTONFS(vp)->n_size > 0) {
462 mtx_unlock(&np->n_mtx);
465 auio.uio_iov = &aiov;
469 auio.uio_segflg = UIO_SYSSPACE;
470 auio.uio_rw = UIO_READ;
471 auio.uio_td = ap->a_td;
473 if (vp->v_type == VREG)
474 error = nfs_readrpc(vp, &auio, ap->a_cred);
475 else if (vp->v_type == VDIR) {
477 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
479 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
480 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
482 } else if (vp->v_type == VLNK)
483 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
487 mtx_unlock(&np->n_mtx);
492 int nfs_otw_getattr_avoid = 0;
496 * Check to see if the type is ok
497 * and that deletion is not in progress.
498 * For paged in text files, you will need to flush the page cache
499 * if consistency is lost.
503 nfs_open(struct vop_open_args *ap)
505 struct vnode *vp = ap->a_vp;
506 struct nfsnode *np = VTONFS(vp);
509 int fmode = ap->a_mode;
512 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
516 * Get a valid lease. If cached data is stale, flush it.
518 mtx_lock(&np->n_mtx);
519 if (np->n_flag & NMODIFIED) {
520 mtx_unlock(&np->n_mtx);
521 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
522 if (error == EINTR || error == EIO)
524 mtx_lock(&np->n_mtx);
526 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
527 if (vp->v_type == VDIR)
528 np->n_direofoffset = 0;
529 mtx_unlock(&np->n_mtx);
530 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
533 mtx_lock(&np->n_mtx);
534 np->n_mtime = vattr.va_mtime;
536 mtx_unlock(&np->n_mtx);
537 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
540 mtx_lock(&np->n_mtx);
541 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
542 if (vp->v_type == VDIR)
543 np->n_direofoffset = 0;
544 mtx_unlock(&np->n_mtx);
545 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
546 if (error == EINTR || error == EIO) {
549 mtx_lock(&np->n_mtx);
550 np->n_mtime = vattr.va_mtime;
554 * If the object has >= 1 O_DIRECT active opens, we disable caching.
556 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
557 if (np->n_directio_opens == 0) {
558 mtx_unlock(&np->n_mtx);
559 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
562 mtx_lock(&np->n_mtx);
563 np->n_flag |= NNONCACHE;
565 np->n_directio_opens++;
569 * If this is an open for writing, capture a reference to the
570 * credentials, so they can be used by nfs_putpages(). Using
571 * these write credentials is preferable to the credentials of
572 * whatever thread happens to be doing the VOP_PUTPAGES() since
573 * the write RPCs are less likely to fail with EACCES.
575 if ((fmode & FWRITE) != 0) {
576 cred = np->n_writecred;
577 np->n_writecred = crhold(ap->a_cred);
580 mtx_unlock(&np->n_mtx);
583 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
589 * What an NFS client should do upon close after writing is a debatable issue.
590 * Most NFS clients push delayed writes to the server upon close, basically for
592 * 1 - So that any write errors may be reported back to the client process
593 * doing the close system call. By far the two most likely errors are
594 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
595 * 2 - To put a worst case upper bound on cache inconsistency between
596 * multiple clients for the file.
597 * There is also a consistency problem for Version 2 of the protocol w.r.t.
598 * not being able to tell if other clients are writing a file concurrently,
599 * since there is no way of knowing if the changed modify time in the reply
600 * is only due to the write for this client.
601 * (NFS Version 3 provides weak cache consistency data in the reply that
602 * should be sufficient to detect and handle this case.)
604 * The current code does the following:
605 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
606 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
607 * or commit them (this satisfies 1 and 2 except for the
608 * case where the server crashes after this close but
609 * before the commit RPC, which is felt to be "good
610 * enough". Changing the last argument to nfs_flush() to
611 * a 1 would force a commit operation, if it is felt a
612 * commit is necessary now.
616 nfs_close(struct vop_close_args *ap)
618 struct vnode *vp = ap->a_vp;
619 struct nfsnode *np = VTONFS(vp);
621 int fmode = ap->a_fflag;
623 if (vp->v_type == VREG) {
625 * Examine and clean dirty pages, regardless of NMODIFIED.
626 * This closes a major hole in close-to-open consistency.
627 * We want to push out all dirty pages (and buffers) on
628 * close, regardless of whether they were dirtied by
629 * mmap'ed writes or via write().
631 if (nfs_clean_pages_on_close && vp->v_object) {
632 VM_OBJECT_WLOCK(vp->v_object);
633 vm_object_page_clean(vp->v_object, 0, 0, 0);
634 VM_OBJECT_WUNLOCK(vp->v_object);
636 mtx_lock(&np->n_mtx);
637 if (np->n_flag & NMODIFIED) {
638 mtx_unlock(&np->n_mtx);
641 * Under NFSv3 we have dirty buffers to dispose of. We
642 * must flush them to the NFS server. We have the option
643 * of waiting all the way through the commit rpc or just
644 * waiting for the initial write. The default is to only
645 * wait through the initial write so the data is in the
646 * server's cache, which is roughly similar to the state
647 * a standard disk subsystem leaves the file in on close().
649 * We cannot clear the NMODIFIED bit in np->n_flag due to
650 * potential races with other processes, and certainly
651 * cannot clear it if we don't commit.
653 int cm = nfsv3_commit_on_close ? 1 : 0;
654 error = nfs_flush(vp, MNT_WAIT, cm);
655 /* np->n_flag &= ~NMODIFIED; */
657 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
658 mtx_lock(&np->n_mtx);
660 if (np->n_flag & NWRITEERR) {
661 np->n_flag &= ~NWRITEERR;
664 mtx_unlock(&np->n_mtx);
666 if (nfs_directio_enable)
667 KASSERT((np->n_directio_asyncwr == 0),
668 ("nfs_close: dirty unflushed (%d) directio buffers\n",
669 np->n_directio_asyncwr));
670 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
671 mtx_lock(&np->n_mtx);
672 KASSERT((np->n_directio_opens > 0),
673 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
674 np->n_directio_opens--;
675 if (np->n_directio_opens == 0)
676 np->n_flag &= ~NNONCACHE;
677 mtx_unlock(&np->n_mtx);
683 * nfs getattr call from vfs.
686 nfs_getattr(struct vop_getattr_args *ap)
688 struct vnode *vp = ap->a_vp;
689 struct nfsnode *np = VTONFS(vp);
690 struct thread *td = curthread;
691 struct vattr *vap = ap->a_vap;
695 struct mbuf *mreq, *mrep, *md, *mb;
696 int v3 = NFS_ISV3(vp);
699 * Update local times for special files.
701 mtx_lock(&np->n_mtx);
702 if (np->n_flag & (NACC | NUPD))
704 mtx_unlock(&np->n_mtx);
706 * First look in the cache.
708 if (nfs_getattrcache(vp, &vattr) == 0)
710 if (v3 && nfs_prime_access_cache && nfsaccess_cache_timeout > 0) {
711 nfsstats.accesscache_misses++;
712 nfs3_access_otw(vp, NFSV3ACCESS_ALL, td, ap->a_cred, NULL);
713 if (nfs_getattrcache(vp, &vattr) == 0)
716 nfsstats.rpccnt[NFSPROC_GETATTR]++;
717 mreq = m_get2(NFSX_FH(v3), M_WAITOK, MT_DATA, 0);
719 bpos = mtod(mb, caddr_t);
721 nfsm_request(vp, NFSPROC_GETATTR, td, ap->a_cred);
723 nfsm_loadattr(vp, &vattr);
727 vap->va_type = vattr.va_type;
728 vap->va_mode = vattr.va_mode;
729 vap->va_nlink = vattr.va_nlink;
730 vap->va_uid = vattr.va_uid;
731 vap->va_gid = vattr.va_gid;
732 vap->va_fsid = vattr.va_fsid;
733 vap->va_fileid = vattr.va_fileid;
734 vap->va_size = vattr.va_size;
735 vap->va_blocksize = vattr.va_blocksize;
736 vap->va_atime = vattr.va_atime;
737 vap->va_mtime = vattr.va_mtime;
738 vap->va_ctime = vattr.va_ctime;
739 vap->va_gen = vattr.va_gen;
740 vap->va_flags = vattr.va_flags;
741 vap->va_rdev = vattr.va_rdev;
742 vap->va_bytes = vattr.va_bytes;
743 vap->va_filerev = vattr.va_filerev;
752 nfs_setattr(struct vop_setattr_args *ap)
754 struct vnode *vp = ap->a_vp;
755 struct nfsnode *np = VTONFS(vp);
756 struct vattr *vap = ap->a_vap;
757 struct thread *td = curthread;
766 * Setting of flags is not supported.
768 if (vap->va_flags != VNOVAL)
772 * Disallow write attempts if the filesystem is mounted read-only.
774 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
775 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
776 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
777 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
781 if (vap->va_size != VNOVAL) {
782 switch (vp->v_type) {
789 if (vap->va_mtime.tv_sec == VNOVAL &&
790 vap->va_atime.tv_sec == VNOVAL &&
791 vap->va_mode == (mode_t)VNOVAL &&
792 vap->va_uid == (uid_t)VNOVAL &&
793 vap->va_gid == (gid_t)VNOVAL)
795 vap->va_size = VNOVAL;
799 * Disallow write attempts if the filesystem is
802 if (vp->v_mount->mnt_flag & MNT_RDONLY)
805 * We run vnode_pager_setsize() early (why?),
806 * we must set np->n_size now to avoid vinvalbuf
807 * V_SAVE races that might setsize a lower
810 mtx_lock(&np->n_mtx);
812 mtx_unlock(&np->n_mtx);
813 error = nfs_meta_setsize(vp, ap->a_cred, td,
815 mtx_lock(&np->n_mtx);
816 if (np->n_flag & NMODIFIED) {
818 mtx_unlock(&np->n_mtx);
819 if (vap->va_size == 0)
820 error = nfs_vinvalbuf(vp, 0, td, 1);
822 error = nfs_vinvalbuf(vp, V_SAVE, td, 1);
824 vnode_pager_setsize(vp, tsize);
828 mtx_unlock(&np->n_mtx);
830 * np->n_size has already been set to vap->va_size
831 * in nfs_meta_setsize(). We must set it again since
832 * nfs_loadattrcache() could be called through
833 * nfs_meta_setsize() and could modify np->n_size.
835 mtx_lock(&np->n_mtx);
836 np->n_vattr.va_size = np->n_size = vap->va_size;
837 mtx_unlock(&np->n_mtx);
840 mtx_lock(&np->n_mtx);
841 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
842 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
843 mtx_unlock(&np->n_mtx);
844 if ((error = nfs_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
845 (error == EINTR || error == EIO))
848 mtx_unlock(&np->n_mtx);
850 error = nfs_setattrrpc(vp, vap, ap->a_cred);
851 if (error && vap->va_size != VNOVAL) {
852 mtx_lock(&np->n_mtx);
853 np->n_size = np->n_vattr.va_size = tsize;
854 vnode_pager_setsize(vp, tsize);
855 mtx_unlock(&np->n_mtx);
862 * Do an nfs setattr rpc.
865 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred)
867 struct nfsv2_sattr *sp;
868 struct nfsnode *np = VTONFS(vp);
871 int error = 0, i, wccflag = NFSV3_WCCRATTR;
872 struct mbuf *mreq, *mrep, *md, *mb;
873 int v3 = NFS_ISV3(vp);
875 nfsstats.rpccnt[NFSPROC_SETATTR]++;
876 mreq = m_get2(NFSX_FH(v3) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
878 bpos = mtod(mb, caddr_t);
881 nfsm_v3attrbuild(vap, TRUE);
882 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
885 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
886 if (vap->va_mode == (mode_t)VNOVAL)
887 sp->sa_mode = nfs_xdrneg1;
889 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
890 if (vap->va_uid == (uid_t)VNOVAL)
891 sp->sa_uid = nfs_xdrneg1;
893 sp->sa_uid = txdr_unsigned(vap->va_uid);
894 if (vap->va_gid == (gid_t)VNOVAL)
895 sp->sa_gid = nfs_xdrneg1;
897 sp->sa_gid = txdr_unsigned(vap->va_gid);
898 sp->sa_size = txdr_unsigned(vap->va_size);
899 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
900 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
902 nfsm_request(vp, NFSPROC_SETATTR, curthread, cred);
904 mtx_lock(&np->n_mtx);
905 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
906 np->n_accesscache[i].stamp = 0;
907 mtx_unlock(&np->n_mtx);
908 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
909 nfsm_wcc_data(vp, wccflag);
911 nfsm_loadattr(vp, NULL);
918 * nfs lookup call, one step at a time...
919 * First look in cache
920 * If not found, unlock the directory nfsnode and do the rpc
923 nfs_lookup(struct vop_lookup_args *ap)
925 struct componentname *cnp = ap->a_cnp;
926 struct vnode *dvp = ap->a_dvp;
927 struct vnode **vpp = ap->a_vpp;
928 struct mount *mp = dvp->v_mount;
929 struct vattr dvattr, vattr;
930 struct timespec nctime;
931 int flags = cnp->cn_flags;
933 struct nfsmount *nmp;
935 struct mbuf *mreq, *mrep, *md, *mb;
938 struct nfsnode *np, *newnp;
939 int error = 0, attrflag, dattrflag, fhsize, ltype, ncticks;
940 int v3 = NFS_ISV3(dvp);
941 struct thread *td = cnp->cn_thread;
944 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
945 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
947 if (dvp->v_type != VDIR)
951 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
955 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
956 if (error > 0 && error != ENOENT)
960 * Lookups of "." are special and always return the
961 * current directory. cache_lookup() already handles
962 * associated locking bookkeeping, etc.
964 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
965 /* XXX: Is this really correct? */
966 if (cnp->cn_nameiop != LOOKUP &&
968 cnp->cn_flags |= SAVENAME;
973 * We only accept a positive hit in the cache if the
974 * change time of the file matches our cached copy.
975 * Otherwise, we discard the cache entry and fallback
976 * to doing a lookup RPC. We also only trust cache
977 * entries for less than nm_nametimeo seconds.
979 * To better handle stale file handles and attributes,
980 * clear the attribute cache of this node if it is a
981 * leaf component, part of an open() call, and not
982 * locally modified before fetching the attributes.
983 * This should allow stale file handles to be detected
984 * here where we can fall back to a LOOKUP RPC to
985 * recover rather than having nfs_open() detect the
986 * stale file handle and failing open(2) with ESTALE.
989 newnp = VTONFS(newvp);
990 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
991 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
992 !(newnp->n_flag & NMODIFIED)) {
993 mtx_lock(&newnp->n_mtx);
994 newnp->n_attrstamp = 0;
995 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
996 mtx_unlock(&newnp->n_mtx);
998 if ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
999 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1000 timespeccmp(&vattr.va_ctime, &nctime, ==)) {
1001 nfsstats.lookupcache_hits++;
1002 if (cnp->cn_nameiop != LOOKUP &&
1004 cnp->cn_flags |= SAVENAME;
1013 } else if (error == ENOENT) {
1014 if (dvp->v_iflag & VI_DOOMED)
1017 * We only accept a negative hit in the cache if the
1018 * modification time of the parent directory matches
1019 * the cached copy in the name cache entry.
1020 * Otherwise, we discard all of the negative cache
1021 * entries for this directory. We also only trust
1022 * negative cache entries for up to nm_negnametimeo
1025 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1026 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1027 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1028 nfsstats.lookupcache_hits++;
1031 cache_purge_negative(dvp);
1034 attrflag = dattrflag = 0;
1037 nfsstats.lookupcache_misses++;
1038 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
1039 len = cnp->cn_namelen;
1040 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len), M_WAITOK,
1043 bpos = mtod(mb, caddr_t);
1044 nfsm_fhtom(dvp, v3);
1045 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1046 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
1049 nfsm_postop_attr_va(dvp, dattrflag, &vattr);
1054 nfsm_getfh(fhp, fhsize, v3);
1057 * Handle RENAME case...
1059 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1060 if (NFS_CMPFH(np, fhp, fhsize)) {
1064 error = nfs_nget(mp, fhp, fhsize, &np, LK_EXCLUSIVE);
1071 nfsm_postop_attr(newvp, attrflag);
1072 nfsm_postop_attr(dvp, attrflag);
1074 nfsm_loadattr(newvp, NULL);
1077 cnp->cn_flags |= SAVENAME;
1081 if (flags & ISDOTDOT) {
1082 ltype = VOP_ISLOCKED(dvp);
1083 error = vfs_busy(mp, MBF_NOWAIT);
1087 error = vfs_busy(mp, 0);
1088 vn_lock(dvp, ltype | LK_RETRY);
1090 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1100 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1105 vn_lock(dvp, ltype | LK_RETRY);
1106 if (dvp->v_iflag & VI_DOOMED) {
1119 } else if (NFS_CMPFH(np, fhp, fhsize)) {
1123 error = nfs_nget(mp, fhp, fhsize, &np, cnp->cn_lkflags);
1131 * Flush the attribute cache when opening a leaf node
1132 * to ensure that fresh attributes are fetched in
1133 * nfs_open() if we are unable to fetch attributes
1134 * from the LOOKUP reply.
1136 if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1137 !(np->n_flag & NMODIFIED)) {
1138 mtx_lock(&np->n_mtx);
1139 np->n_attrstamp = 0;
1140 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1141 mtx_unlock(&np->n_mtx);
1145 nfsm_postop_attr_va(newvp, attrflag, &vattr);
1146 nfsm_postop_attr_va(dvp, dattrflag, &dvattr);
1148 nfsm_loadattr(newvp, &vattr);
1151 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1152 cnp->cn_flags |= SAVENAME;
1153 if ((cnp->cn_flags & MAKEENTRY) &&
1154 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1155 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1156 cache_enter_time(dvp, newvp, cnp, &vattr.va_ctime,
1157 newvp->v_type != VDIR ? NULL : &dvattr.va_ctime);
1162 if (newvp != NULLVP) {
1167 if (error != ENOENT)
1170 /* The requested file was not found. */
1171 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1172 (flags & ISLASTCN)) {
1174 * XXX: UFS does a full VOP_ACCESS(dvp,
1175 * VWRITE) here instead of just checking
1178 if (mp->mnt_flag & MNT_RDONLY)
1180 cnp->cn_flags |= SAVENAME;
1181 return (EJUSTRETURN);
1184 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1186 * Cache the modification time of the parent
1187 * directory from the post-op attributes in
1188 * the name cache entry. The negative cache
1189 * entry will be ignored once the directory
1190 * has changed. Don't bother adding the entry
1191 * if the directory has already changed.
1193 mtx_lock(&np->n_mtx);
1194 if (timespeccmp(&np->n_vattr.va_mtime,
1195 &vattr.va_mtime, ==)) {
1196 mtx_unlock(&np->n_mtx);
1197 cache_enter_time(dvp, NULL, cnp,
1198 &vattr.va_mtime, NULL);
1200 mtx_unlock(&np->n_mtx);
1210 * Just call nfs_bioread() to do the work.
1213 nfs_read(struct vop_read_args *ap)
1215 struct vnode *vp = ap->a_vp;
1217 switch (vp->v_type) {
1219 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1223 return (EOPNOTSUPP);
1231 nfs_readlink(struct vop_readlink_args *ap)
1233 struct vnode *vp = ap->a_vp;
1235 if (vp->v_type != VLNK)
1237 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1241 * Do a readlink rpc.
1242 * Called by nfs_doio() from below the buffer cache.
1245 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1248 int error = 0, len, attrflag;
1249 struct mbuf *mreq, *mrep, *md, *mb;
1250 int v3 = NFS_ISV3(vp);
1252 nfsstats.rpccnt[NFSPROC_READLINK]++;
1253 mreq = m_get2(NFSX_FH(v3), M_WAITOK, MT_DATA, 0);
1255 bpos = mtod(mb, caddr_t);
1257 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1259 nfsm_postop_attr(vp, attrflag);
1261 nfsm_strsiz(len, NFS_MAXPATHLEN);
1262 if (len == NFS_MAXPATHLEN) {
1263 struct nfsnode *np = VTONFS(vp);
1264 mtx_lock(&np->n_mtx);
1265 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1267 mtx_unlock(&np->n_mtx);
1269 nfsm_mtouio(uiop, len);
1281 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1285 struct mbuf *mreq, *mrep, *md, *mb;
1286 struct nfsmount *nmp;
1288 int error = 0, len, retlen, tsiz, eof, attrflag;
1289 int v3 = NFS_ISV3(vp);
1295 nmp = VFSTONFS(vp->v_mount);
1296 tsiz = uiop->uio_resid;
1297 mtx_lock(&nmp->nm_mtx);
1298 end = uiop->uio_offset + tsiz;
1299 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1300 mtx_unlock(&nmp->nm_mtx);
1303 rsize = nmp->nm_rsize;
1304 mtx_unlock(&nmp->nm_mtx);
1306 nfsstats.rpccnt[NFSPROC_READ]++;
1307 len = (tsiz > rsize) ? rsize : tsiz;
1308 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED * 3, M_WAITOK,
1311 bpos = mtod(mb, caddr_t);
1313 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1315 txdr_hyper(uiop->uio_offset, tl);
1316 *(tl + 2) = txdr_unsigned(len);
1318 *tl++ = txdr_unsigned(uiop->uio_offset);
1319 *tl++ = txdr_unsigned(len);
1322 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1324 nfsm_postop_attr(vp, attrflag);
1329 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1330 eof = fxdr_unsigned(int, *(tl + 1));
1332 nfsm_loadattr(vp, NULL);
1334 nfsm_strsiz(retlen, rsize);
1335 nfsm_mtouio(uiop, retlen);
1339 if (eof || retlen == 0) {
1342 } else if (retlen < len) {
1354 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1355 int *iomode, int *must_commit)
1360 struct mbuf *mreq, *mrep, *md, *mb;
1361 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1363 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1364 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1367 KASSERT(uiop->uio_iovcnt == 1, ("nfs: writerpc iovcnt > 1"));
1369 tsiz = uiop->uio_resid;
1370 mtx_lock(&nmp->nm_mtx);
1371 end = uiop->uio_offset + tsiz;
1372 if (end > nmp->nm_maxfilesize || end < uiop->uio_offset) {
1373 mtx_unlock(&nmp->nm_mtx);
1376 wsize = nmp->nm_wsize;
1377 mtx_unlock(&nmp->nm_mtx);
1379 nfsstats.rpccnt[NFSPROC_WRITE]++;
1380 len = (tsiz > wsize) ? wsize : tsiz;
1381 mreq = m_get2(NFSX_FH(v3) + 5 * NFSX_UNSIGNED, M_WAITOK,
1384 bpos = mtod(mb, caddr_t);
1387 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1388 txdr_hyper(uiop->uio_offset, tl);
1390 *tl++ = txdr_unsigned(len);
1391 *tl++ = txdr_unsigned(*iomode);
1392 *tl = txdr_unsigned(len);
1396 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1397 /* Set both "begin" and "current" to non-garbage. */
1398 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1399 *tl++ = x; /* "begin offset" */
1400 *tl++ = x; /* "current offset" */
1401 x = txdr_unsigned(len);
1402 *tl++ = x; /* total to this offset */
1403 *tl = x; /* size of this write */
1405 nfsm_uiotom(uiop, len);
1406 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1408 wccflag = NFSV3_WCCCHK;
1409 nfsm_wcc_data(vp, wccflag);
1411 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1412 + NFSX_V3WRITEVERF);
1413 rlen = fxdr_unsigned(int, *tl++);
1418 } else if (rlen < len) {
1419 backup = len - rlen;
1420 uiop->uio_iov->iov_base =
1421 (char *)uiop->uio_iov->iov_base -
1423 uiop->uio_iov->iov_len += backup;
1424 uiop->uio_offset -= backup;
1425 uiop->uio_resid += backup;
1428 commit = fxdr_unsigned(int, *tl++);
1431 * Return the lowest committment level
1432 * obtained by any of the RPCs.
1434 if (committed == NFSV3WRITE_FILESYNC)
1436 else if (committed == NFSV3WRITE_DATASYNC &&
1437 commit == NFSV3WRITE_UNSTABLE)
1439 mtx_lock(&nmp->nm_mtx);
1440 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1441 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1443 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1444 } else if (bcmp((caddr_t)tl,
1445 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1447 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1450 mtx_unlock(&nmp->nm_mtx);
1453 nfsm_loadattr(vp, NULL);
1456 mtx_lock(&(VTONFS(vp))->n_mtx);
1457 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1458 mtx_unlock(&(VTONFS(vp))->n_mtx);
1467 committed = NFSV3WRITE_FILESYNC;
1468 *iomode = committed;
1470 uiop->uio_resid = tsiz;
1476 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1477 * mode set to specify the file type and the size field for rdev.
1480 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1483 struct nfsv2_sattr *sp;
1485 struct vnode *newvp = NULL;
1486 struct nfsnode *np = NULL;
1489 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1490 struct mbuf *mreq, *mrep, *md, *mb;
1492 int v3 = NFS_ISV3(dvp);
1494 if (vap->va_type == VCHR || vap->va_type == VBLK)
1495 rdev = txdr_unsigned(vap->va_rdev);
1496 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1499 return (EOPNOTSUPP);
1501 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
1503 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1504 mreq = m_get2(NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1505 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
1507 bpos = mtod(mb, caddr_t);
1508 nfsm_fhtom(dvp, v3);
1509 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1511 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1512 *tl++ = vtonfsv3_type(vap->va_type);
1513 nfsm_v3attrbuild(vap, FALSE);
1514 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1515 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1516 *tl++ = txdr_unsigned(major(vap->va_rdev));
1517 *tl = txdr_unsigned(minor(vap->va_rdev));
1520 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1521 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1522 sp->sa_uid = nfs_xdrneg1;
1523 sp->sa_gid = nfs_xdrneg1;
1525 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1526 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1528 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1530 nfsm_mtofh(dvp, newvp, v3, gotvp);
1536 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1537 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1543 nfsm_wcc_data(dvp, wccflag);
1552 mtx_lock(&(VTONFS(dvp))->n_mtx);
1553 VTONFS(dvp)->n_flag |= NMODIFIED;
1555 VTONFS(dvp)->n_attrstamp = 0;
1556 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1558 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1564 * just call nfs_mknodrpc() to do the work.
1568 nfs_mknod(struct vop_mknod_args *ap)
1570 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1573 static u_long create_verf;
1575 * nfs file create call
1578 nfs_create(struct vop_create_args *ap)
1580 struct vnode *dvp = ap->a_dvp;
1581 struct vattr *vap = ap->a_vap;
1582 struct componentname *cnp = ap->a_cnp;
1583 struct nfsv2_sattr *sp;
1585 struct nfsnode *np = NULL;
1586 struct vnode *newvp = NULL;
1588 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1589 struct mbuf *mreq, *mrep, *md, *mb;
1591 int v3 = NFS_ISV3(dvp);
1594 * Oops, not for me..
1596 if (vap->va_type == VSOCK) {
1597 error = nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap);
1601 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0) {
1604 if (vap->va_vaflags & VA_EXCLUSIVE)
1607 nfsstats.rpccnt[NFSPROC_CREATE]++;
1608 mreq = m_get2(NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1609 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
1611 bpos = mtod(mb, caddr_t);
1612 nfsm_fhtom(dvp, v3);
1613 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1615 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1616 if (fmode & O_EXCL) {
1617 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1618 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1620 CURVNET_SET(CRED_TO_VNET(cnp->cn_cred));
1622 if (!TAILQ_EMPTY(&V_in_ifaddrhead))
1623 *tl++ = IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr.s_addr;
1626 *tl++ = create_verf;
1628 IN_IFADDR_RUNLOCK();
1631 *tl = ++create_verf;
1633 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1634 nfsm_v3attrbuild(vap, FALSE);
1637 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1638 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1639 sp->sa_uid = nfs_xdrneg1;
1640 sp->sa_gid = nfs_xdrneg1;
1642 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1643 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1645 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1647 nfsm_mtofh(dvp, newvp, v3, gotvp);
1653 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1654 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1660 nfsm_wcc_data(dvp, wccflag);
1664 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1670 } else if (v3 && (fmode & O_EXCL)) {
1672 * We are normally called with only a partially initialized
1673 * VAP. Since the NFSv3 spec says that server may use the
1674 * file attributes to store the verifier, the spec requires
1675 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1676 * in atime, but we can't really assume that all servers will
1677 * so we ensure that our SETATTR sets both atime and mtime.
1679 if (vap->va_mtime.tv_sec == VNOVAL)
1680 vfs_timestamp(&vap->va_mtime);
1681 if (vap->va_atime.tv_sec == VNOVAL)
1682 vap->va_atime = vap->va_mtime;
1683 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred);
1690 mtx_lock(&(VTONFS(dvp))->n_mtx);
1691 VTONFS(dvp)->n_flag |= NMODIFIED;
1693 VTONFS(dvp)->n_attrstamp = 0;
1694 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1696 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1701 * nfs file remove call
1702 * To try and make nfs semantics closer to ufs semantics, a file that has
1703 * other processes using the vnode is renamed instead of removed and then
1704 * removed later on the last close.
1705 * - If v_usecount > 1
1706 * If a rename is not already in the works
1707 * call nfs_sillyrename() to set it up
1712 nfs_remove(struct vop_remove_args *ap)
1714 struct vnode *vp = ap->a_vp;
1715 struct vnode *dvp = ap->a_dvp;
1716 struct componentname *cnp = ap->a_cnp;
1717 struct nfsnode *np = VTONFS(vp);
1721 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1722 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1723 if (vp->v_type == VDIR)
1725 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1726 !VOP_GETATTR(vp, &vattr, cnp->cn_cred) && vattr.va_nlink > 1)) {
1728 * Purge the name cache so that the chance of a lookup for
1729 * the name succeeding while the remove is in progress is
1730 * minimized. Without node locking it can still happen, such
1731 * that an I/O op returns ESTALE, but since you get this if
1732 * another host removes the file..
1736 * throw away biocache buffers, mainly to avoid
1737 * unnecessary delayed writes later.
1739 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1741 if (error != EINTR && error != EIO)
1742 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1743 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1745 * Kludge City: If the first reply to the remove rpc is lost..
1746 * the reply to the retransmitted request will be ENOENT
1747 * since the file was in fact removed
1748 * Therefore, we cheat and return success.
1750 if (error == ENOENT)
1752 } else if (!np->n_sillyrename)
1753 error = nfs_sillyrename(dvp, vp, cnp);
1754 mtx_lock(&np->n_mtx);
1755 np->n_attrstamp = 0;
1756 mtx_unlock(&np->n_mtx);
1757 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1762 * nfs file remove rpc called from nfs_inactive
1765 nfs_removeit(struct sillyrename *sp)
1768 * Make sure that the directory vnode is still valid.
1769 * XXX we should lock sp->s_dvp here.
1771 if (sp->s_dvp->v_type == VBAD)
1773 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1778 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1781 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1782 struct ucred *cred, struct thread *td)
1785 int error = 0, wccflag = NFSV3_WCCRATTR;
1786 struct mbuf *mreq, *mrep, *md, *mb;
1787 int v3 = NFS_ISV3(dvp);
1789 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1790 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen),
1791 M_WAITOK, MT_DATA, 0);
1793 bpos = mtod(mb, caddr_t);
1794 nfsm_fhtom(dvp, v3);
1795 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1796 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1798 nfsm_wcc_data(dvp, wccflag);
1801 mtx_lock(&(VTONFS(dvp))->n_mtx);
1802 VTONFS(dvp)->n_flag |= NMODIFIED;
1804 VTONFS(dvp)->n_attrstamp = 0;
1805 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1807 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1812 * nfs file rename call
1815 nfs_rename(struct vop_rename_args *ap)
1817 struct vnode *fvp = ap->a_fvp;
1818 struct vnode *tvp = ap->a_tvp;
1819 struct vnode *fdvp = ap->a_fdvp;
1820 struct vnode *tdvp = ap->a_tdvp;
1821 struct componentname *tcnp = ap->a_tcnp;
1822 struct componentname *fcnp = ap->a_fcnp;
1825 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1826 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1827 /* Check for cross-device rename */
1828 if ((fvp->v_mount != tdvp->v_mount) ||
1829 (tvp && (fvp->v_mount != tvp->v_mount))) {
1835 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1839 if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1843 * We have to flush B_DELWRI data prior to renaming
1844 * the file. If we don't, the delayed-write buffers
1845 * can be flushed out later after the file has gone stale
1846 * under NFSV3. NFSV2 does not have this problem because
1847 * ( as far as I can tell ) it flushes dirty buffers more
1850 * Skip the rename operation if the fsync fails, this can happen
1851 * due to the server's volume being full, when we pushed out data
1852 * that was written back to our cache earlier. Not checking for
1853 * this condition can result in potential (silent) data loss.
1855 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1858 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1863 * If the tvp exists and is in use, sillyrename it before doing the
1864 * rename of the new file over it.
1865 * XXX Can't sillyrename a directory.
1867 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1868 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1873 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1874 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1877 if (fvp->v_type == VDIR) {
1878 if (tvp != NULL && tvp->v_type == VDIR)
1893 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1895 if (error == ENOENT)
1901 * nfs file rename rpc called from nfs_remove() above
1904 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1905 struct sillyrename *sp)
1908 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1909 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1913 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1916 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1917 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1921 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1922 struct mbuf *mreq, *mrep, *md, *mb;
1923 int v3 = NFS_ISV3(fdvp);
1925 nfsstats.rpccnt[NFSPROC_RENAME]++;
1926 mreq = m_get2((NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1927 nfsm_rndup(tnamelen), M_WAITOK, MT_DATA, 0);
1929 bpos = mtod(mb, caddr_t);
1930 nfsm_fhtom(fdvp, v3);
1931 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1932 nfsm_fhtom(tdvp, v3);
1933 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1934 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1936 nfsm_wcc_data(fdvp, fwccflag);
1937 nfsm_wcc_data(tdvp, twccflag);
1941 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1942 VTONFS(fdvp)->n_flag |= NMODIFIED;
1943 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1944 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1945 VTONFS(tdvp)->n_flag |= NMODIFIED;
1946 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1948 VTONFS(fdvp)->n_attrstamp = 0;
1949 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1952 VTONFS(tdvp)->n_attrstamp = 0;
1953 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1959 * nfs hard link create call
1962 nfs_link(struct vop_link_args *ap)
1964 struct vnode *vp = ap->a_vp;
1965 struct vnode *tdvp = ap->a_tdvp;
1966 struct componentname *cnp = ap->a_cnp;
1968 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1969 struct mbuf *mreq, *mrep, *md, *mb;
1972 if (vp->v_mount != tdvp->v_mount) {
1977 * Push all writes to the server, so that the attribute cache
1978 * doesn't get "out of sync" with the server.
1979 * XXX There should be a better way!
1981 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1984 nfsstats.rpccnt[NFSPROC_LINK]++;
1985 mreq = m_get2(NFSX_FH(v3)*2 + NFSX_UNSIGNED +
1986 nfsm_rndup(cnp->cn_namelen), M_WAITOK, MT_DATA, 0);
1988 bpos = mtod(mb, caddr_t);
1990 nfsm_fhtom(tdvp, v3);
1991 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1992 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1994 nfsm_postop_attr(vp, attrflag);
1995 nfsm_wcc_data(tdvp, wccflag);
1999 mtx_lock(&(VTONFS(tdvp))->n_mtx);
2000 VTONFS(tdvp)->n_flag |= NMODIFIED;
2001 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
2003 VTONFS(vp)->n_attrstamp = 0;
2004 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
2007 VTONFS(tdvp)->n_attrstamp = 0;
2008 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
2014 * nfs symbolic link create call
2017 nfs_symlink(struct vop_symlink_args *ap)
2019 struct vnode *dvp = ap->a_dvp;
2020 struct vattr *vap = ap->a_vap;
2021 struct componentname *cnp = ap->a_cnp;
2022 struct nfsv2_sattr *sp;
2024 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
2025 struct mbuf *mreq, *mrep, *md, *mb;
2026 struct vnode *newvp = NULL;
2027 int v3 = NFS_ISV3(dvp);
2029 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
2030 slen = strlen(ap->a_target);
2031 mreq = m_get2(NFSX_FH(v3) + 2*NFSX_UNSIGNED +
2032 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3),
2033 M_WAITOK, MT_DATA, 0);
2035 bpos = mtod(mb, caddr_t);
2036 nfsm_fhtom(dvp, v3);
2037 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2039 nfsm_v3attrbuild(vap, FALSE);
2041 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
2043 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2044 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
2045 sp->sa_uid = nfs_xdrneg1;
2046 sp->sa_gid = nfs_xdrneg1;
2047 sp->sa_size = nfs_xdrneg1;
2048 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2049 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2053 * Issue the NFS request and get the rpc response.
2055 * Only NFSv3 responses returning an error of 0 actually return
2056 * a file handle that can be converted into newvp without having
2057 * to do an extra lookup rpc.
2059 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
2062 nfsm_mtofh(dvp, newvp, v3, gotvp);
2063 nfsm_wcc_data(dvp, wccflag);
2067 * out code jumps -> here, mrep is also freed.
2074 * If we do not have an error and we could not extract the newvp from
2075 * the response due to the request being NFSv2, we have to do a
2076 * lookup in order to obtain a newvp to return.
2078 if (error == 0 && newvp == NULL) {
2079 struct nfsnode *np = NULL;
2081 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2082 cnp->cn_cred, cnp->cn_thread, &np);
2092 mtx_lock(&(VTONFS(dvp))->n_mtx);
2093 VTONFS(dvp)->n_flag |= NMODIFIED;
2094 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2096 VTONFS(dvp)->n_attrstamp = 0;
2097 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2106 nfs_mkdir(struct vop_mkdir_args *ap)
2108 struct vnode *dvp = ap->a_dvp;
2109 struct vattr *vap = ap->a_vap;
2110 struct componentname *cnp = ap->a_cnp;
2111 struct nfsv2_sattr *sp;
2113 struct nfsnode *np = NULL;
2114 struct vnode *newvp = NULL;
2116 int error = 0, wccflag = NFSV3_WCCRATTR;
2118 struct mbuf *mreq, *mrep, *md, *mb;
2120 int v3 = NFS_ISV3(dvp);
2122 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2124 len = cnp->cn_namelen;
2125 nfsstats.rpccnt[NFSPROC_MKDIR]++;
2126 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) +
2127 NFSX_SATTR(v3), M_WAITOK, MT_DATA, 0);
2129 bpos = mtod(mb, caddr_t);
2130 nfsm_fhtom(dvp, v3);
2131 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
2133 nfsm_v3attrbuild(vap, FALSE);
2135 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
2136 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
2137 sp->sa_uid = nfs_xdrneg1;
2138 sp->sa_gid = nfs_xdrneg1;
2139 sp->sa_size = nfs_xdrneg1;
2140 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
2141 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
2143 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
2145 nfsm_mtofh(dvp, newvp, v3, gotvp);
2147 nfsm_wcc_data(dvp, wccflag);
2150 mtx_lock(&(VTONFS(dvp))->n_mtx);
2151 VTONFS(dvp)->n_flag |= NMODIFIED;
2152 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2154 VTONFS(dvp)->n_attrstamp = 0;
2155 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2157 if (error == 0 && newvp == NULL) {
2158 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
2159 cnp->cn_thread, &np);
2162 if (newvp->v_type != VDIR)
2175 * nfs remove directory call
2178 nfs_rmdir(struct vop_rmdir_args *ap)
2180 struct vnode *vp = ap->a_vp;
2181 struct vnode *dvp = ap->a_dvp;
2182 struct componentname *cnp = ap->a_cnp;
2184 int error = 0, wccflag = NFSV3_WCCRATTR;
2185 struct mbuf *mreq, *mrep, *md, *mb;
2186 int v3 = NFS_ISV3(dvp);
2190 nfsstats.rpccnt[NFSPROC_RMDIR]++;
2191 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED +
2192 nfsm_rndup(cnp->cn_namelen), M_WAITOK, MT_DATA, 0);
2194 bpos = mtod(mb, caddr_t);
2195 nfsm_fhtom(dvp, v3);
2196 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
2197 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
2199 nfsm_wcc_data(dvp, wccflag);
2202 mtx_lock(&(VTONFS(dvp))->n_mtx);
2203 VTONFS(dvp)->n_flag |= NMODIFIED;
2204 mtx_unlock(&(VTONFS(dvp))->n_mtx);
2206 VTONFS(dvp)->n_attrstamp = 0;
2207 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2212 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2214 if (error == ENOENT)
2223 nfs_readdir(struct vop_readdir_args *ap)
2225 struct vnode *vp = ap->a_vp;
2226 struct nfsnode *np = VTONFS(vp);
2227 struct uio *uio = ap->a_uio;
2228 int tresid, error = 0;
2231 if (vp->v_type != VDIR)
2235 * First, check for hit on the EOF offset cache
2237 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2238 (np->n_flag & NMODIFIED) == 0) {
2239 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2240 mtx_lock(&np->n_mtx);
2241 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2242 mtx_unlock(&np->n_mtx);
2243 nfsstats.direofcache_hits++;
2246 mtx_unlock(&np->n_mtx);
2251 * Call nfs_bioread() to do the real work.
2253 tresid = uio->uio_resid;
2254 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2256 if (!error && uio->uio_resid == tresid) {
2257 nfsstats.direofcache_misses++;
2265 * Called from below the buffer cache by nfs_doio().
2268 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2271 struct dirent *dp = NULL;
2276 struct mbuf *mreq, *mrep, *md, *mb;
2278 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2279 struct nfsnode *dnp = VTONFS(vp);
2281 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2283 int v3 = NFS_ISV3(vp);
2285 KASSERT(uiop->uio_iovcnt == 1 &&
2286 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2287 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2288 ("nfs readdirrpc bad uio"));
2291 * If there is no cookie, assume directory was stale.
2293 nfs_dircookie_lock(dnp);
2294 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2297 nfs_dircookie_unlock(dnp);
2299 nfs_dircookie_unlock(dnp);
2300 return (NFSERR_BAD_COOKIE);
2304 * Loop around doing readdir rpc's of size nm_readdirsize
2305 * truncated to a multiple of DIRBLKSIZ.
2306 * The stopping criteria is EOF or buffer full.
2308 while (more_dirs && bigenough) {
2309 nfsstats.rpccnt[NFSPROC_READDIR]++;
2310 mreq = m_get2(NFSX_FH(v3) + NFSX_READDIR(v3), M_WAITOK,
2313 bpos = mtod(mb, caddr_t);
2316 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2317 *tl++ = cookie.nfsuquad[0];
2318 *tl++ = cookie.nfsuquad[1];
2319 mtx_lock(&dnp->n_mtx);
2320 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2321 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2322 mtx_unlock(&dnp->n_mtx);
2324 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2325 *tl++ = cookie.nfsuquad[0];
2327 *tl = txdr_unsigned(nmp->nm_readdirsize);
2328 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2330 nfsm_postop_attr(vp, attrflag);
2332 tl = nfsm_dissect(u_int32_t *,
2334 mtx_lock(&dnp->n_mtx);
2335 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2336 dnp->n_cookieverf.nfsuquad[1] = *tl;
2337 mtx_unlock(&dnp->n_mtx);
2343 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2344 more_dirs = fxdr_unsigned(int, *tl);
2346 /* loop thru the dir entries, doctoring them to 4bsd form */
2347 while (more_dirs && bigenough) {
2349 tl = nfsm_dissect(u_int32_t *,
2351 fileno = fxdr_hyper(tl);
2352 len = fxdr_unsigned(int, *(tl + 2));
2354 tl = nfsm_dissect(u_int32_t *,
2356 fileno = fxdr_unsigned(u_quad_t, *tl++);
2357 len = fxdr_unsigned(int, *tl);
2359 if (len <= 0 || len > NFS_MAXNAMLEN) {
2364 tlen = nfsm_rndup(len);
2366 tlen += 4; /* To ensure null termination */
2367 left = DIRBLKSIZ - blksiz;
2368 if ((tlen + DIRHDSIZ) > left) {
2369 dp->d_reclen += left;
2370 uiop->uio_iov->iov_base =
2371 (char *)uiop->uio_iov->iov_base + left;
2372 uiop->uio_iov->iov_len -= left;
2373 uiop->uio_offset += left;
2374 uiop->uio_resid -= left;
2377 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2380 dp = (struct dirent *)uiop->uio_iov->iov_base;
2381 dp->d_fileno = (int)fileno;
2383 dp->d_reclen = tlen + DIRHDSIZ;
2384 dp->d_type = DT_UNKNOWN;
2385 blksiz += dp->d_reclen;
2386 if (blksiz == DIRBLKSIZ)
2388 uiop->uio_offset += DIRHDSIZ;
2389 uiop->uio_resid -= DIRHDSIZ;
2390 uiop->uio_iov->iov_base =
2391 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2392 uiop->uio_iov->iov_len -= DIRHDSIZ;
2393 nfsm_mtouio(uiop, len);
2394 cp = uiop->uio_iov->iov_base;
2396 *cp = '\0'; /* null terminate */
2397 uiop->uio_iov->iov_base =
2398 (char *)uiop->uio_iov->iov_base + tlen;
2399 uiop->uio_iov->iov_len -= tlen;
2400 uiop->uio_offset += tlen;
2401 uiop->uio_resid -= tlen;
2403 nfsm_adv(nfsm_rndup(len));
2405 tl = nfsm_dissect(u_int32_t *,
2408 tl = nfsm_dissect(u_int32_t *,
2412 cookie.nfsuquad[0] = *tl++;
2414 cookie.nfsuquad[1] = *tl++;
2419 more_dirs = fxdr_unsigned(int, *tl);
2422 * If at end of rpc data, get the eof boolean
2425 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2426 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2431 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2432 * by increasing d_reclen for the last record.
2435 left = DIRBLKSIZ - blksiz;
2436 dp->d_reclen += left;
2437 uiop->uio_iov->iov_base =
2438 (char *)uiop->uio_iov->iov_base + left;
2439 uiop->uio_iov->iov_len -= left;
2440 uiop->uio_offset += left;
2441 uiop->uio_resid -= left;
2445 * We are now either at the end of the directory or have filled the
2449 dnp->n_direofoffset = uiop->uio_offset;
2451 if (uiop->uio_resid > 0)
2452 nfs_printf("EEK! readdirrpc resid > 0\n");
2453 nfs_dircookie_lock(dnp);
2454 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2456 nfs_dircookie_unlock(dnp);
2463 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2466 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2472 struct vnode *newvp;
2474 caddr_t bpos, dpos, dpossav1, dpossav2;
2475 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2476 struct nameidata nami, *ndp = &nami;
2477 struct componentname *cnp = &ndp->ni_cnd;
2479 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2480 struct nfsnode *dnp = VTONFS(vp), *np;
2481 struct vattr vattr, dvattr;
2484 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2485 int attrflag, dattrflag, fhsize;
2490 KASSERT(uiop->uio_iovcnt == 1 &&
2491 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2492 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2493 ("nfs readdirplusrpc bad uio"));
2498 * If there is no cookie, assume directory was stale.
2500 nfs_dircookie_lock(dnp);
2501 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2504 nfs_dircookie_unlock(dnp);
2506 nfs_dircookie_unlock(dnp);
2507 return (NFSERR_BAD_COOKIE);
2510 * Loop around doing readdir rpc's of size nm_readdirsize
2511 * truncated to a multiple of DIRBLKSIZ.
2512 * The stopping criteria is EOF or buffer full.
2514 while (more_dirs && bigenough) {
2515 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2516 mreq = m_get2(NFSX_FH(1) + 6 * NFSX_UNSIGNED, M_WAITOK,
2519 bpos = mtod(mb, caddr_t);
2521 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2522 *tl++ = cookie.nfsuquad[0];
2523 *tl++ = cookie.nfsuquad[1];
2524 mtx_lock(&dnp->n_mtx);
2525 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2526 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2527 mtx_unlock(&dnp->n_mtx);
2528 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2529 *tl = txdr_unsigned(nmp->nm_rsize);
2530 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2531 nfsm_postop_attr_va(vp, dattrflag, &dvattr);
2536 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2537 mtx_lock(&dnp->n_mtx);
2538 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2539 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2540 mtx_unlock(&dnp->n_mtx);
2541 more_dirs = fxdr_unsigned(int, *tl);
2543 /* loop thru the dir entries, doctoring them to 4bsd form */
2544 while (more_dirs && bigenough) {
2545 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2546 fileno = fxdr_hyper(tl);
2547 len = fxdr_unsigned(int, *(tl + 2));
2548 if (len <= 0 || len > NFS_MAXNAMLEN) {
2553 tlen = nfsm_rndup(len);
2555 tlen += 4; /* To ensure null termination*/
2556 left = DIRBLKSIZ - blksiz;
2557 if ((tlen + DIRHDSIZ) > left) {
2558 dp->d_reclen += left;
2559 uiop->uio_iov->iov_base =
2560 (char *)uiop->uio_iov->iov_base + left;
2561 uiop->uio_iov->iov_len -= left;
2562 uiop->uio_offset += left;
2563 uiop->uio_resid -= left;
2566 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2569 dp = (struct dirent *)uiop->uio_iov->iov_base;
2570 dp->d_fileno = (int)fileno;
2572 dp->d_reclen = tlen + DIRHDSIZ;
2573 dp->d_type = DT_UNKNOWN;
2574 blksiz += dp->d_reclen;
2575 if (blksiz == DIRBLKSIZ)
2577 uiop->uio_offset += DIRHDSIZ;
2578 uiop->uio_resid -= DIRHDSIZ;
2579 uiop->uio_iov->iov_base =
2580 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2581 uiop->uio_iov->iov_len -= DIRHDSIZ;
2582 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2583 cnp->cn_namelen = len;
2584 nfsm_mtouio(uiop, len);
2585 cp = uiop->uio_iov->iov_base;
2588 uiop->uio_iov->iov_base =
2589 (char *)uiop->uio_iov->iov_base + tlen;
2590 uiop->uio_iov->iov_len -= tlen;
2591 uiop->uio_offset += tlen;
2592 uiop->uio_resid -= tlen;
2594 nfsm_adv(nfsm_rndup(len));
2595 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2597 cookie.nfsuquad[0] = *tl++;
2598 cookie.nfsuquad[1] = *tl++;
2603 * Since the attributes are before the file handle
2604 * (sigh), we must skip over the attributes and then
2605 * come back and get them.
2607 attrflag = fxdr_unsigned(int, *tl);
2611 nfsm_adv(NFSX_V3FATTR);
2612 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2613 doit = fxdr_unsigned(int, *tl);
2615 * Skip loading the attrs for "..". There's a
2616 * race between loading the attrs here and
2617 * lookups that look for the directory currently
2618 * being read (in the parent). We try to acquire
2619 * the exclusive lock on ".." here, owning the
2620 * lock on the directory being read. Lookup will
2621 * hold the lock on ".." and try to acquire the
2622 * lock on the directory being read.
2624 * There are other ways of fixing this, one would
2625 * be to do a trylock on the ".." vnode and skip
2626 * loading the attrs on ".." if it happens to be
2627 * locked by another process. But skipping the
2628 * attrload on ".." seems the easiest option.
2630 if (strcmp(dp->d_name, "..") == 0) {
2633 * We've already skipped over the attrs,
2634 * skip over the filehandle. And store d_type
2637 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2638 i = fxdr_unsigned(int, *tl);
2639 nfsm_adv(nfsm_rndup(i));
2640 dp->d_type = IFTODT(VTTOIF(VDIR));
2643 nfsm_getfh(fhp, fhsize, 1);
2644 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2649 error = nfs_nget(vp->v_mount, fhp,
2650 fhsize, &np, LK_EXCLUSIVE);
2657 if (doit && bigenough) {
2662 nfsm_loadattr(newvp, &vattr);
2665 dp->d_type = IFTODT(VTTOIF(vattr.va_type));
2667 if (newvp->v_type != VDIR || dattrflag != 0)
2668 cache_enter_time(ndp->ni_dvp, ndp->ni_vp,
2669 cnp, &vattr.va_ctime,
2670 newvp->v_type != VDIR ? NULL :
2674 /* Just skip over the file handle */
2675 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2676 i = fxdr_unsigned(int, *tl);
2678 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2679 fhsize = fxdr_unsigned(int, *tl);
2680 nfsm_adv(nfsm_rndup(fhsize));
2683 if (newvp != NULLVP) {
2690 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2691 more_dirs = fxdr_unsigned(int, *tl);
2694 * If at end of rpc data, get the eof boolean
2697 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2698 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2703 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2704 * by increasing d_reclen for the last record.
2707 left = DIRBLKSIZ - blksiz;
2708 dp->d_reclen += left;
2709 uiop->uio_iov->iov_base =
2710 (char *)uiop->uio_iov->iov_base + left;
2711 uiop->uio_iov->iov_len -= left;
2712 uiop->uio_offset += left;
2713 uiop->uio_resid -= left;
2717 * We are now either at the end of the directory or have filled the
2721 dnp->n_direofoffset = uiop->uio_offset;
2723 if (uiop->uio_resid > 0)
2724 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2725 nfs_dircookie_lock(dnp);
2726 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2728 nfs_dircookie_unlock(dnp);
2731 if (newvp != NULLVP) {
2742 * Silly rename. To make the NFS filesystem that is stateless look a little
2743 * more like the "ufs" a remove of an active vnode is translated to a rename
2744 * to a funny looking filename that is removed by nfs_inactive on the
2745 * nfsnode. There is the potential for another process on a different client
2746 * to create the same funny name between the nfs_lookitup() fails and the
2747 * nfs_rename() completes, but...
2750 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2752 struct sillyrename *sp;
2756 unsigned int lticks;
2760 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2761 sp = malloc(sizeof (struct sillyrename),
2762 M_NFSREQ, M_WAITOK);
2763 sp->s_cred = crhold(cnp->cn_cred);
2765 sp->s_removeit = nfs_removeit;
2769 * Fudge together a funny name.
2770 * Changing the format of the funny name to accomodate more
2771 * sillynames per directory.
2772 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2773 * CPU ticks since boot.
2775 pid = cnp->cn_thread->td_proc->p_pid;
2776 lticks = (unsigned int)ticks;
2778 sp->s_namlen = sprintf(sp->s_name,
2779 ".nfs.%08x.%04x4.4", lticks,
2781 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2782 cnp->cn_thread, NULL))
2786 error = nfs_renameit(dvp, cnp, sp);
2789 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2790 cnp->cn_thread, &np);
2791 np->n_sillyrename = sp;
2796 free((caddr_t)sp, M_NFSREQ);
2801 * Look up a file name and optionally either update the file handle or
2802 * allocate an nfsnode, depending on the value of npp.
2803 * npp == NULL --> just do the lookup
2804 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2806 * *npp != NULL --> update the file handle in the vnode
2809 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2810 struct thread *td, struct nfsnode **npp)
2812 struct vnode *newvp = NULL;
2813 struct nfsnode *np, *dnp = VTONFS(dvp);
2815 int error = 0, fhlen, attrflag;
2816 struct mbuf *mreq, *mrep, *md, *mb;
2818 int v3 = NFS_ISV3(dvp);
2820 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2821 mreq = m_get2(NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len),
2822 M_WAITOK, MT_DATA, 0);
2824 bpos = mtod(mb, caddr_t);
2825 nfsm_fhtom(dvp, v3);
2826 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2827 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2828 if (npp && !error) {
2829 nfsm_getfh(nfhp, fhlen, v3);
2832 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2833 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2834 np->n_fhp = &np->n_fh;
2835 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2836 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2837 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2838 np->n_fhsize = fhlen;
2840 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2844 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2852 nfsm_postop_attr(newvp, attrflag);
2853 if (!attrflag && *npp == NULL) {
2862 nfsm_loadattr(newvp, NULL);
2866 if (npp && *npp == NULL) {
2881 * Nfs Version 3 commit rpc
2884 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2888 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2890 int error = 0, wccflag = NFSV3_WCCRATTR;
2891 struct mbuf *mreq, *mrep, *md, *mb;
2893 mtx_lock(&nmp->nm_mtx);
2894 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2895 mtx_unlock(&nmp->nm_mtx);
2898 mtx_unlock(&nmp->nm_mtx);
2899 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2900 mreq = m_get2(NFSX_FH(1), M_WAITOK, MT_DATA, 0);
2902 bpos = mtod(mb, caddr_t);
2904 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2905 txdr_hyper(offset, tl);
2907 *tl = txdr_unsigned(cnt);
2908 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2909 nfsm_wcc_data(vp, wccflag);
2911 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2912 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2913 NFSX_V3WRITEVERF)) {
2914 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2916 error = NFSERR_STALEWRITEVERF;
2926 * For async requests when nfsiod(s) are running, queue the request by
2927 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2931 nfs_strategy(struct vop_strategy_args *ap)
2933 struct buf *bp = ap->a_bp;
2936 KASSERT(!(bp->b_flags & B_DONE),
2937 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2938 BUF_ASSERT_HELD(bp);
2940 if (bp->b_iocmd == BIO_READ)
2946 * If the op is asynchronous and an i/o daemon is waiting
2947 * queue the request, wake it up and wait for completion
2948 * otherwise just do it ourselves.
2950 if ((bp->b_flags & B_ASYNC) == 0 ||
2951 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2952 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2957 * fsync vnode op. Just call nfs_flush() with commit == 1.
2961 nfs_fsync(struct vop_fsync_args *ap)
2964 return (nfs_flush(ap->a_vp, ap->a_waitfor, 1));
2968 * Flush all the blocks associated with a vnode.
2969 * Walk through the buffer pool and push any dirty pages
2970 * associated with the vnode.
2973 nfs_flush(struct vnode *vp, int waitfor, int commit)
2975 struct nfsnode *np = VTONFS(vp);
2979 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2980 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2982 u_quad_t off, endoff, toff;
2983 struct ucred* wcred = NULL;
2984 struct buf **bvec = NULL;
2986 struct thread *td = curthread;
2987 #ifndef NFS_COMMITBVECSIZ
2988 #define NFS_COMMITBVECSIZ 20
2990 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2991 int bvecsize = 0, bveccount;
2993 if (nmp->nm_flag & NFSMNT_INT)
2999 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
3000 * server, but has not been committed to stable storage on the server
3001 * yet. On the first pass, the byte range is worked out and the commit
3002 * rpc is done. On the second pass, nfs_writebp() is called to do the
3009 if (NFS_ISV3(vp) && commit) {
3010 if (bvec != NULL && bvec != bvec_on_stack)
3013 * Count up how many buffers waiting for a commit.
3017 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3018 if (!BUF_ISLOCKED(bp) &&
3019 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
3020 == (B_DELWRI | B_NEEDCOMMIT))
3024 * Allocate space to remember the list of bufs to commit. It is
3025 * important to use M_NOWAIT here to avoid a race with nfs_write.
3026 * If we can't get memory (for whatever reason), we will end up
3027 * committing the buffers one-by-one in the loop below.
3029 if (bveccount > NFS_COMMITBVECSIZ) {
3031 * Release the vnode interlock to avoid a lock
3035 bvec = (struct buf **)
3036 malloc(bveccount * sizeof(struct buf *),
3040 bvec = bvec_on_stack;
3041 bvecsize = NFS_COMMITBVECSIZ;
3043 bvecsize = bveccount;
3045 bvec = bvec_on_stack;
3046 bvecsize = NFS_COMMITBVECSIZ;
3048 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3049 if (bvecpos >= bvecsize)
3051 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3052 nbp = TAILQ_NEXT(bp, b_bobufs);
3055 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
3056 (B_DELWRI | B_NEEDCOMMIT)) {
3058 nbp = TAILQ_NEXT(bp, b_bobufs);
3064 * Work out if all buffers are using the same cred
3065 * so we can deal with them all with one commit.
3067 * NOTE: we are not clearing B_DONE here, so we have
3068 * to do it later on in this routine if we intend to
3069 * initiate I/O on the bp.
3071 * Note: to avoid loopback deadlocks, we do not
3072 * assign b_runningbufspace.
3075 wcred = bp->b_wcred;
3076 else if (wcred != bp->b_wcred)
3078 vfs_busy_pages(bp, 1);
3082 * bp is protected by being locked, but nbp is not
3083 * and vfs_busy_pages() may sleep. We have to
3086 nbp = TAILQ_NEXT(bp, b_bobufs);
3089 * A list of these buffers is kept so that the
3090 * second loop knows which buffers have actually
3091 * been committed. This is necessary, since there
3092 * may be a race between the commit rpc and new
3093 * uncommitted writes on the file.
3095 bvec[bvecpos++] = bp;
3096 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3100 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
3108 * Commit data on the server, as required.
3109 * If all bufs are using the same wcred, then use that with
3110 * one call for all of them, otherwise commit each one
3113 if (wcred != NOCRED)
3114 retv = nfs_commit(vp, off, (int)(endoff - off),
3118 for (i = 0; i < bvecpos; i++) {
3121 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
3123 size = (u_quad_t)(bp->b_dirtyend
3125 retv = nfs_commit(vp, off, (int)size,
3131 if (retv == NFSERR_STALEWRITEVERF)
3132 nfs_clearcommit(vp->v_mount);
3135 * Now, either mark the blocks I/O done or mark the
3136 * blocks dirty, depending on whether the commit
3139 for (i = 0; i < bvecpos; i++) {
3141 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
3144 * Error, leave B_DELWRI intact
3146 vfs_unbusy_pages(bp);
3150 * Success, remove B_DELWRI ( bundirty() ).
3152 * b_dirtyoff/b_dirtyend seem to be NFS
3153 * specific. We should probably move that
3154 * into bundirty(). XXX
3157 bp->b_flags |= B_ASYNC;
3159 bp->b_flags &= ~B_DONE;
3160 bp->b_ioflags &= ~BIO_ERROR;
3161 bp->b_dirtyoff = bp->b_dirtyend = 0;
3168 * Start/do any write(s) that are required.
3172 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
3173 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
3174 if (waitfor != MNT_WAIT || passone)
3177 error = BUF_TIMELOCK(bp,
3178 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
3179 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
3184 if (error == ENOLCK) {
3188 if (nfs_sigintr(nmp, td)) {
3192 if (slpflag == PCATCH) {
3198 if ((bp->b_flags & B_DELWRI) == 0)
3199 panic("nfs_fsync: not dirty");
3200 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
3206 if (passone || !commit)
3207 bp->b_flags |= B_ASYNC;
3209 bp->b_flags |= B_ASYNC;
3211 if (nfs_sigintr(nmp, td)) {
3222 if (waitfor == MNT_WAIT) {
3223 while (bo->bo_numoutput) {
3224 error = bufobj_wwait(bo, slpflag, slptimeo);
3227 error = nfs_sigintr(nmp, td);
3230 if (slpflag == PCATCH) {
3237 if (bo->bo_dirty.bv_cnt != 0 && commit) {
3242 * Wait for all the async IO requests to drain
3245 mtx_lock(&np->n_mtx);
3246 while (np->n_directio_asyncwr > 0) {
3247 np->n_flag |= NFSYNCWAIT;
3248 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3249 &np->n_mtx, slpflag | (PRIBIO + 1),
3252 if (nfs_sigintr(nmp, td)) {
3253 mtx_unlock(&np->n_mtx);
3259 mtx_unlock(&np->n_mtx);
3262 mtx_lock(&np->n_mtx);
3263 if (np->n_flag & NWRITEERR) {
3264 error = np->n_error;
3265 np->n_flag &= ~NWRITEERR;
3267 if (commit && bo->bo_dirty.bv_cnt == 0 &&
3268 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3269 np->n_flag &= ~NMODIFIED;
3270 mtx_unlock(&np->n_mtx);
3272 if (bvec != NULL && bvec != bvec_on_stack)
3278 * NFS advisory byte-level locks.
3281 nfs_advlock(struct vop_advlock_args *ap)
3283 struct vnode *vp = ap->a_vp;
3287 error = vn_lock(vp, LK_SHARED);
3290 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3291 size = VTONFS(vp)->n_size;
3293 error = lf_advlock(ap, &(vp->v_lockf), size);
3296 error = nfs_advlock_p(ap);
3305 * NFS advisory byte-level locks.
3308 nfs_advlockasync(struct vop_advlockasync_args *ap)
3310 struct vnode *vp = ap->a_vp;
3314 error = vn_lock(vp, LK_SHARED);
3317 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3318 size = VTONFS(vp)->n_size;
3320 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3329 * Print out the contents of an nfsnode.
3332 nfs_print(struct vop_print_args *ap)
3334 struct vnode *vp = ap->a_vp;
3335 struct nfsnode *np = VTONFS(vp);
3337 nfs_printf("\tfileid %ld fsid 0x%x",
3338 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3339 if (vp->v_type == VFIFO)
3346 * This is the "real" nfs::bwrite(struct buf*).
3347 * We set B_CACHE if this is a VMIO buffer.
3350 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3353 int oldflags = bp->b_flags;
3359 BUF_ASSERT_HELD(bp);
3361 if (bp->b_flags & B_INVAL) {
3366 bp->b_flags |= B_CACHE;
3369 * Undirty the bp. We will redirty it later if the I/O fails.
3374 bp->b_flags &= ~B_DONE;
3375 bp->b_ioflags &= ~BIO_ERROR;
3376 bp->b_iocmd = BIO_WRITE;
3378 bufobj_wref(bp->b_bufobj);
3379 curthread->td_ru.ru_oublock++;
3383 * Note: to avoid loopback deadlocks, we do not
3384 * assign b_runningbufspace.
3386 vfs_busy_pages(bp, 1);
3389 bp->b_iooffset = dbtob(bp->b_blkno);
3392 if( (oldflags & B_ASYNC) == 0) {
3393 int rtval = bufwait(bp);
3395 if (oldflags & B_DELWRI) {
3408 * nfs special file access vnode op.
3409 * Essentially just get vattr and then imitate iaccess() since the device is
3410 * local to the client.
3413 nfsspec_access(struct vop_access_args *ap)
3416 struct ucred *cred = ap->a_cred;
3417 struct vnode *vp = ap->a_vp;
3418 accmode_t accmode = ap->a_accmode;
3423 * Disallow write attempts on filesystems mounted read-only;
3424 * unless the file is a socket, fifo, or a block or character
3425 * device resident on the filesystem.
3427 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3428 switch (vp->v_type) {
3438 error = VOP_GETATTR(vp, vap, cred);
3441 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3442 accmode, cred, NULL);
3448 * Read wrapper for fifos.
3451 nfsfifo_read(struct vop_read_args *ap)
3453 struct nfsnode *np = VTONFS(ap->a_vp);
3459 mtx_lock(&np->n_mtx);
3461 vfs_timestamp(&np->n_atim);
3462 mtx_unlock(&np->n_mtx);
3463 error = fifo_specops.vop_read(ap);
3468 * Write wrapper for fifos.
3471 nfsfifo_write(struct vop_write_args *ap)
3473 struct nfsnode *np = VTONFS(ap->a_vp);
3478 mtx_lock(&np->n_mtx);
3480 vfs_timestamp(&np->n_mtim);
3481 mtx_unlock(&np->n_mtx);
3482 return(fifo_specops.vop_write(ap));
3486 * Close wrapper for fifos.
3488 * Update the times on the nfsnode then do fifo close.
3491 nfsfifo_close(struct vop_close_args *ap)
3493 struct vnode *vp = ap->a_vp;
3494 struct nfsnode *np = VTONFS(vp);
3498 mtx_lock(&np->n_mtx);
3499 if (np->n_flag & (NACC | NUPD)) {
3501 if (np->n_flag & NACC)
3503 if (np->n_flag & NUPD)
3506 if (vrefcnt(vp) == 1 &&
3507 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3509 if (np->n_flag & NACC)
3510 vattr.va_atime = np->n_atim;
3511 if (np->n_flag & NUPD)
3512 vattr.va_mtime = np->n_mtim;
3513 mtx_unlock(&np->n_mtx);
3514 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3518 mtx_unlock(&np->n_mtx);
3520 return (fifo_specops.vop_close(ap));
3524 * Just call nfs_writebp() with the force argument set to 1.
3526 * NOTE: B_DONE may or may not be set in a_bp on call.
3529 nfs_bwrite(struct buf *bp)
3532 return (nfs_writebp(bp, 1, curthread));
3535 struct buf_ops buf_ops_nfs = {
3536 .bop_name = "buf_ops_nfs",
3537 .bop_write = nfs_bwrite,
3538 .bop_strategy = bufstrategy,
3539 .bop_sync = bufsync,
3540 .bop_bdflush = bufbdflush,