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
44 #include <sys/param.h>
45 #include <sys/kernel.h>
46 #include <sys/systm.h>
47 #include <sys/resourcevar.h>
49 #include <sys/mount.h>
52 #include <sys/malloc.h>
54 #include <sys/namei.h>
55 #include <sys/socket.h>
56 #include <sys/vnode.h>
57 #include <sys/dirent.h>
58 #include <sys/fcntl.h>
59 #include <sys/lockf.h>
61 #include <sys/sysctl.h>
62 #include <sys/signalvar.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
69 #include <fs/fifofs/fifo.h>
71 #include <rpc/rpcclnt.h>
73 #include <nfs/rpcv2.h>
74 #include <nfs/nfsproto.h>
75 #include <nfsclient/nfs.h>
76 #include <nfsclient/nfsnode.h>
77 #include <nfsclient/nfsmount.h>
78 #include <nfsclient/nfs_lock.h>
79 #include <nfs/xdr_subs.h>
80 #include <nfsclient/nfsm_subs.h>
83 #include <netinet/in.h>
84 #include <netinet/in_var.h>
91 * Ifdef for FreeBSD-current merged buffer cache. It is unfortunate that these
92 * calls are not in getblk() and brelse() so that they would not be necessary
96 #define vfs_busy_pages(bp, f)
99 static vop_read_t nfsfifo_read;
100 static vop_write_t nfsfifo_write;
101 static vop_close_t nfsfifo_close;
102 static int nfs_flush(struct vnode *, int, struct thread *,
104 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
106 static vop_lookup_t nfs_lookup;
107 static vop_create_t nfs_create;
108 static vop_mknod_t nfs_mknod;
109 static vop_open_t nfs_open;
110 static vop_close_t nfs_close;
111 static vop_access_t nfs_access;
112 static vop_getattr_t nfs_getattr;
113 static vop_setattr_t nfs_setattr;
114 static vop_read_t nfs_read;
115 static vop_fsync_t nfs_fsync;
116 static vop_remove_t nfs_remove;
117 static vop_link_t nfs_link;
118 static vop_rename_t nfs_rename;
119 static vop_mkdir_t nfs_mkdir;
120 static vop_rmdir_t nfs_rmdir;
121 static vop_symlink_t nfs_symlink;
122 static vop_readdir_t nfs_readdir;
123 static vop_strategy_t nfs_strategy;
124 static int nfs_lookitup(struct vnode *, const char *, int,
125 struct ucred *, struct thread *, struct nfsnode **);
126 static int nfs_sillyrename(struct vnode *, struct vnode *,
127 struct componentname *);
128 static vop_access_t nfsspec_access;
129 static vop_readlink_t nfs_readlink;
130 static vop_print_t nfs_print;
131 static vop_advlock_t nfs_advlock;
134 * Global vfs data structures for nfs
136 struct vop_vector nfs_vnodeops = {
137 .vop_default = &default_vnodeops,
138 .vop_access = nfs_access,
139 .vop_advlock = nfs_advlock,
140 .vop_close = nfs_close,
141 .vop_create = nfs_create,
142 .vop_fsync = nfs_fsync,
143 .vop_getattr = nfs_getattr,
144 .vop_getpages = nfs_getpages,
145 .vop_putpages = nfs_putpages,
146 .vop_inactive = nfs_inactive,
147 .vop_lease = VOP_NULL,
148 .vop_link = nfs_link,
149 .vop_lookup = nfs_lookup,
150 .vop_mkdir = nfs_mkdir,
151 .vop_mknod = nfs_mknod,
152 .vop_open = nfs_open,
153 .vop_print = nfs_print,
154 .vop_read = nfs_read,
155 .vop_readdir = nfs_readdir,
156 .vop_readlink = nfs_readlink,
157 .vop_reclaim = nfs_reclaim,
158 .vop_remove = nfs_remove,
159 .vop_rename = nfs_rename,
160 .vop_rmdir = nfs_rmdir,
161 .vop_setattr = nfs_setattr,
162 .vop_strategy = nfs_strategy,
163 .vop_symlink = nfs_symlink,
164 .vop_write = nfs_write,
167 struct vop_vector nfs_fifoops = {
168 .vop_default = &fifo_specops,
169 .vop_access = nfsspec_access,
170 .vop_close = nfsfifo_close,
171 .vop_fsync = nfs_fsync,
172 .vop_getattr = nfs_getattr,
173 .vop_inactive = nfs_inactive,
174 .vop_print = nfs_print,
175 .vop_read = nfsfifo_read,
176 .vop_reclaim = nfs_reclaim,
177 .vop_setattr = nfs_setattr,
178 .vop_write = nfsfifo_write,
181 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
182 struct componentname *cnp, struct vattr *vap);
183 static int nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
184 struct ucred *cred, struct thread *td);
185 static int nfs_renamerpc(struct vnode *fdvp, const char *fnameptr,
186 int fnamelen, struct vnode *tdvp,
187 const char *tnameptr, int tnamelen,
188 struct ucred *cred, struct thread *td);
189 static int nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
190 struct sillyrename *sp);
195 struct mtx nfs_iod_mtx;
196 struct proc *nfs_iodwant[NFS_MAXASYNCDAEMON];
197 struct nfsmount *nfs_iodmount[NFS_MAXASYNCDAEMON];
198 int nfs_numasync = 0;
199 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
201 SYSCTL_DECL(_vfs_nfs);
203 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
204 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
205 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
207 static int nfsv3_commit_on_close = 0;
208 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfsv3_commit_on_close, CTLFLAG_RW,
209 &nfsv3_commit_on_close, 0, "write+commit on close, else only write");
211 static int nfs_clean_pages_on_close = 1;
212 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
213 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
215 int nfs_directio_enable = 0;
216 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
217 &nfs_directio_enable, 0, "Enable NFS directio");
220 * This sysctl allows other processes to mmap a file that has been opened
221 * O_DIRECT by a process. In general, having processes mmap the file while
222 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
223 * this by default to prevent DoS attacks - to prevent a malicious user from
224 * opening up files O_DIRECT preventing other users from mmap'ing these
225 * files. "Protected" environments where stricter consistency guarantees are
226 * required can disable this knob. The process that opened the file O_DIRECT
227 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
230 int nfs_directio_allow_mmap = 1;
231 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
232 &nfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
235 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_hits, CTLFLAG_RD,
236 &nfsstats.accesscache_hits, 0, "NFS ACCESS cache hit count");
238 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_misses, CTLFLAG_RD,
239 &nfsstats.accesscache_misses, 0, "NFS ACCESS cache miss count");
242 #define NFSV3ACCESS_ALL (NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY \
243 | NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE \
244 | NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP)
248 * The list of locks after the description of the lock is the ordering
249 * of other locks acquired with the lock held.
250 * np->n_mtx : Protects the fields in the nfsnode.
252 VI_MTX (acquired indirectly)
253 * nmp->nm_mtx : Protects the fields in the nfsmount.
255 * nfs_iod_mtx : Global lock, protects shared nfsiod state.
256 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
259 * rep->r_mtx : Protects the fields in an nfsreq.
263 nfs3_access_otw(struct vnode *vp, int wmode, struct thread *td,
268 int error = 0, attrflag;
270 struct mbuf *mreq, *mrep, *md, *mb;
273 struct nfsnode *np = VTONFS(vp);
275 nfsstats.rpccnt[NFSPROC_ACCESS]++;
276 mreq = nfsm_reqhead(vp, NFSPROC_ACCESS, NFSX_FH(v3) + NFSX_UNSIGNED);
278 bpos = mtod(mb, caddr_t);
280 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
281 *tl = txdr_unsigned(wmode);
282 nfsm_request(vp, NFSPROC_ACCESS, td, cred);
283 nfsm_postop_attr(vp, attrflag);
285 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
286 rmode = fxdr_unsigned(u_int32_t, *tl);
287 mtx_lock(&np->n_mtx);
289 np->n_modeuid = cred->cr_uid;
290 np->n_modestamp = time_second;
291 mtx_unlock(&np->n_mtx);
299 * nfs access vnode op.
300 * For nfs version 2, just return ok. File accesses may fail later.
301 * For nfs version 3, use the access rpc to check accessibility. If file modes
302 * are changed on the server, accesses might still fail later.
305 nfs_access(struct vop_access_args *ap)
307 struct vnode *vp = ap->a_vp;
309 u_int32_t mode, wmode;
310 int v3 = NFS_ISV3(vp);
311 struct nfsnode *np = VTONFS(vp);
314 * Disallow write attempts on filesystems mounted read-only;
315 * unless the file is a socket, fifo, or a block or character
316 * device resident on the filesystem.
318 if ((ap->a_mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
319 switch (vp->v_type) {
329 * For nfs v3, check to see if we have done this recently, and if
330 * so return our cached result instead of making an ACCESS call.
331 * If not, do an access rpc, otherwise you are stuck emulating
332 * ufs_access() locally using the vattr. This may not be correct,
333 * since the server may apply other access criteria such as
334 * client uid-->server uid mapping that we do not know about.
337 if (ap->a_mode & VREAD)
338 mode = NFSV3ACCESS_READ;
341 if (vp->v_type != VDIR) {
342 if (ap->a_mode & VWRITE)
343 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
344 if (ap->a_mode & VEXEC)
345 mode |= NFSV3ACCESS_EXECUTE;
347 if (ap->a_mode & VWRITE)
348 mode |= (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
350 if (ap->a_mode & VEXEC)
351 mode |= NFSV3ACCESS_LOOKUP;
353 /* XXX safety belt, only make blanket request if caching */
354 if (nfsaccess_cache_timeout > 0) {
355 wmode = NFSV3ACCESS_READ | NFSV3ACCESS_MODIFY |
356 NFSV3ACCESS_EXTEND | NFSV3ACCESS_EXECUTE |
357 NFSV3ACCESS_DELETE | NFSV3ACCESS_LOOKUP;
363 * Does our cached result allow us to give a definite yes to
366 mtx_lock(&np->n_mtx);
367 if ((time_second < (np->n_modestamp + nfsaccess_cache_timeout)) &&
368 (ap->a_cred->cr_uid == np->n_modeuid) &&
369 ((np->n_mode & mode) == mode)) {
370 nfsstats.accesscache_hits++;
373 * Either a no, or a don't know. Go to the wire.
375 nfsstats.accesscache_misses++;
376 mtx_unlock(&np->n_mtx);
377 error = nfs3_access_otw(vp, wmode, ap->a_td,ap->a_cred);
378 mtx_lock(&np->n_mtx);
380 if ((np->n_mode & mode) != mode) {
385 mtx_unlock(&np->n_mtx);
388 if ((error = nfsspec_access(ap)) != 0) {
392 * Attempt to prevent a mapped root from accessing a file
393 * which it shouldn't. We try to read a byte from the file
394 * if the user is root and the file is not zero length.
395 * After calling nfsspec_access, we should have the correct
398 mtx_lock(&np->n_mtx);
399 if (ap->a_cred->cr_uid == 0 && (ap->a_mode & VREAD)
400 && VTONFS(vp)->n_size > 0) {
405 mtx_unlock(&np->n_mtx);
408 auio.uio_iov = &aiov;
412 auio.uio_segflg = UIO_SYSSPACE;
413 auio.uio_rw = UIO_READ;
414 auio.uio_td = ap->a_td;
416 if (vp->v_type == VREG)
417 error = nfs_readrpc(vp, &auio, ap->a_cred);
418 else if (vp->v_type == VDIR) {
420 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
422 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
423 error = nfs_readdirrpc(vp, &auio, ap->a_cred);
425 } else if (vp->v_type == VLNK)
426 error = nfs_readlinkrpc(vp, &auio, ap->a_cred);
430 mtx_unlock(&np->n_mtx);
435 int nfs_otw_getattr_avoid = 0;
439 * Check to see if the type is ok
440 * and that deletion is not in progress.
441 * For paged in text files, you will need to flush the page cache
442 * if consistency is lost.
446 nfs_open(struct vop_open_args *ap)
448 struct vnode *vp = ap->a_vp;
449 struct nfsnode *np = VTONFS(vp);
452 int fmode = ap->a_mode;
454 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
458 * Get a valid lease. If cached data is stale, flush it.
460 mtx_lock(&np->n_mtx);
461 if (np->n_flag & NMODIFIED) {
462 mtx_unlock(&np->n_mtx);
463 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
464 if (error == EINTR || error == EIO)
467 if (vp->v_type == VDIR)
468 np->n_direofoffset = 0;
469 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
472 mtx_lock(&np->n_mtx);
473 np->n_mtime = vattr.va_mtime;
474 mtx_unlock(&np->n_mtx);
476 struct thread *td = curthread;
478 if (np->n_ac_ts_syscalls != td->td_syscalls ||
479 np->n_ac_ts_tid != td->td_tid ||
480 td->td_proc == NULL ||
481 np->n_ac_ts_pid != td->td_proc->p_pid) {
484 mtx_unlock(&np->n_mtx);
485 error = VOP_GETATTR(vp, &vattr, ap->a_cred, ap->a_td);
488 mtx_lock(&np->n_mtx);
489 if (NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
490 if (vp->v_type == VDIR)
491 np->n_direofoffset = 0;
492 mtx_unlock(&np->n_mtx);
493 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
494 if (error == EINTR || error == EIO) {
497 mtx_lock(&np->n_mtx);
498 np->n_mtime = vattr.va_mtime;
500 mtx_unlock(&np->n_mtx);
503 * If the object has >= 1 O_DIRECT active opens, we disable caching.
505 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
506 if (np->n_directio_opens == 0) {
507 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
510 mtx_lock(&np->n_mtx);
511 np->n_flag |= NNONCACHE;
512 mtx_unlock(&np->n_mtx);
514 np->n_directio_opens++;
516 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
522 * What an NFS client should do upon close after writing is a debatable issue.
523 * Most NFS clients push delayed writes to the server upon close, basically for
525 * 1 - So that any write errors may be reported back to the client process
526 * doing the close system call. By far the two most likely errors are
527 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
528 * 2 - To put a worst case upper bound on cache inconsistency between
529 * multiple clients for the file.
530 * There is also a consistency problem for Version 2 of the protocol w.r.t.
531 * not being able to tell if other clients are writing a file concurrently,
532 * since there is no way of knowing if the changed modify time in the reply
533 * is only due to the write for this client.
534 * (NFS Version 3 provides weak cache consistency data in the reply that
535 * should be sufficient to detect and handle this case.)
537 * The current code does the following:
538 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
539 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
540 * or commit them (this satisfies 1 and 2 except for the
541 * case where the server crashes after this close but
542 * before the commit RPC, which is felt to be "good
543 * enough". Changing the last argument to nfs_flush() to
544 * a 1 would force a commit operation, if it is felt a
545 * commit is necessary now.
549 nfs_close(struct vop_close_args *ap)
551 struct vnode *vp = ap->a_vp;
552 struct nfsnode *np = VTONFS(vp);
554 int fmode = ap->a_fflag;
556 if (vp->v_type == VREG) {
558 * Examine and clean dirty pages, regardless of NMODIFIED.
559 * This closes a major hole in close-to-open consistency.
560 * We want to push out all dirty pages (and buffers) on
561 * close, regardless of whether they were dirtied by
562 * mmap'ed writes or via write().
564 if (nfs_clean_pages_on_close && vp->v_object) {
565 VM_OBJECT_LOCK(vp->v_object);
566 vm_object_page_clean(vp->v_object, 0, 0, 0);
567 VM_OBJECT_UNLOCK(vp->v_object);
569 mtx_lock(&np->n_mtx);
570 if (np->n_flag & NMODIFIED) {
571 mtx_unlock(&np->n_mtx);
574 * Under NFSv3 we have dirty buffers to dispose of. We
575 * must flush them to the NFS server. We have the option
576 * of waiting all the way through the commit rpc or just
577 * waiting for the initial write. The default is to only
578 * wait through the initial write so the data is in the
579 * server's cache, which is roughly similar to the state
580 * a standard disk subsystem leaves the file in on close().
582 * We cannot clear the NMODIFIED bit in np->n_flag due to
583 * potential races with other processes, and certainly
584 * cannot clear it if we don't commit.
586 int cm = nfsv3_commit_on_close ? 1 : 0;
587 error = nfs_flush(vp, MNT_WAIT, ap->a_td, cm);
588 /* np->n_flag &= ~NMODIFIED; */
590 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
591 mtx_lock(&np->n_mtx);
594 * Invalidate the attribute cache in all cases.
595 * An open is going to fetch fresh attrs any way, other procs
596 * on this node that have file open will be forced to do an
597 * otw attr fetch, but this is safe.
600 if (np->n_flag & NWRITEERR) {
601 np->n_flag &= ~NWRITEERR;
604 mtx_unlock(&np->n_mtx);
606 if (nfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
607 mtx_lock(&np->n_mtx);
608 KASSERT((np->n_directio_opens > 0),
609 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
610 np->n_directio_opens--;
611 if (np->n_directio_opens == 0)
612 np->n_flag &= ~NNONCACHE;
613 mtx_unlock(&np->n_mtx);
619 * nfs getattr call from vfs.
622 nfs_getattr(struct vop_getattr_args *ap)
624 struct vnode *vp = ap->a_vp;
625 struct nfsnode *np = VTONFS(vp);
628 struct mbuf *mreq, *mrep, *md, *mb;
629 int v3 = NFS_ISV3(vp);
632 * Update local times for special files.
634 mtx_lock(&np->n_mtx);
635 if (np->n_flag & (NACC | NUPD))
637 mtx_unlock(&np->n_mtx);
639 * First look in the cache.
641 if (nfs_getattrcache(vp, ap->a_vap) == 0)
643 if (v3 && nfsaccess_cache_timeout > 0) {
644 nfsstats.accesscache_misses++;
645 nfs3_access_otw(vp, NFSV3ACCESS_ALL, ap->a_td, ap->a_cred);
646 if (nfs_getattrcache(vp, ap->a_vap) == 0)
649 nfsstats.rpccnt[NFSPROC_GETATTR]++;
650 mreq = nfsm_reqhead(vp, NFSPROC_GETATTR, NFSX_FH(v3));
652 bpos = mtod(mb, caddr_t);
654 nfsm_request(vp, NFSPROC_GETATTR, ap->a_td, ap->a_cred);
656 nfsm_loadattr(vp, ap->a_vap);
667 nfs_setattr(struct vop_setattr_args *ap)
669 struct vnode *vp = ap->a_vp;
670 struct nfsnode *np = VTONFS(vp);
671 struct vattr *vap = ap->a_vap;
680 * Setting of flags and marking of atimes are not supported.
682 if (vap->va_flags != VNOVAL || (vap->va_vaflags & VA_MARK_ATIME))
686 * Disallow write attempts if the filesystem is mounted read-only.
688 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
689 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
690 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
691 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
695 if (vap->va_size != VNOVAL) {
696 switch (vp->v_type) {
703 if (vap->va_mtime.tv_sec == VNOVAL &&
704 vap->va_atime.tv_sec == VNOVAL &&
705 vap->va_mode == (mode_t)VNOVAL &&
706 vap->va_uid == (uid_t)VNOVAL &&
707 vap->va_gid == (gid_t)VNOVAL)
709 vap->va_size = VNOVAL;
713 * Disallow write attempts if the filesystem is
716 if (vp->v_mount->mnt_flag & MNT_RDONLY)
719 * We run vnode_pager_setsize() early (why?),
720 * we must set np->n_size now to avoid vinvalbuf
721 * V_SAVE races that might setsize a lower
724 mtx_lock(&np->n_mtx);
726 mtx_unlock(&np->n_mtx);
727 error = nfs_meta_setsize(vp, ap->a_cred,
728 ap->a_td, vap->va_size);
729 mtx_lock(&np->n_mtx);
730 if (np->n_flag & NMODIFIED) {
732 mtx_unlock(&np->n_mtx);
733 if (vap->va_size == 0)
734 error = nfs_vinvalbuf(vp, 0, ap->a_td, 1);
736 error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
738 vnode_pager_setsize(vp, tsize);
742 mtx_unlock(&np->n_mtx);
744 * np->n_size has already been set to vap->va_size
745 * in nfs_meta_setsize(). We must set it again since
746 * nfs_loadattrcache() could be called through
747 * nfs_meta_setsize() and could modify np->n_size.
749 mtx_lock(&np->n_mtx);
750 np->n_vattr.va_size = np->n_size = vap->va_size;
751 mtx_unlock(&np->n_mtx);
754 mtx_lock(&np->n_mtx);
755 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
756 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
757 mtx_unlock(&np->n_mtx);
758 if ((error = nfs_vinvalbuf(vp, V_SAVE, ap->a_td, 1)) != 0 &&
759 (error == EINTR || error == EIO))
762 mtx_unlock(&np->n_mtx);
764 error = nfs_setattrrpc(vp, vap, ap->a_cred, ap->a_td);
765 if (error && vap->va_size != VNOVAL) {
766 mtx_lock(&np->n_mtx);
767 np->n_size = np->n_vattr.va_size = tsize;
768 vnode_pager_setsize(vp, tsize);
769 mtx_unlock(&np->n_mtx);
776 * Do an nfs setattr rpc.
779 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
782 struct nfsv2_sattr *sp;
783 struct nfsnode *np = VTONFS(vp);
786 int error = 0, wccflag = NFSV3_WCCRATTR;
787 struct mbuf *mreq, *mrep, *md, *mb;
788 int v3 = NFS_ISV3(vp);
790 nfsstats.rpccnt[NFSPROC_SETATTR]++;
791 mreq = nfsm_reqhead(vp, NFSPROC_SETATTR, NFSX_FH(v3) + NFSX_SATTR(v3));
793 bpos = mtod(mb, caddr_t);
796 nfsm_v3attrbuild(vap, TRUE);
797 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
800 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
801 if (vap->va_mode == (mode_t)VNOVAL)
802 sp->sa_mode = nfs_xdrneg1;
804 sp->sa_mode = vtonfsv2_mode(vp->v_type, vap->va_mode);
805 if (vap->va_uid == (uid_t)VNOVAL)
806 sp->sa_uid = nfs_xdrneg1;
808 sp->sa_uid = txdr_unsigned(vap->va_uid);
809 if (vap->va_gid == (gid_t)VNOVAL)
810 sp->sa_gid = nfs_xdrneg1;
812 sp->sa_gid = txdr_unsigned(vap->va_gid);
813 sp->sa_size = txdr_unsigned(vap->va_size);
814 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
815 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
817 nfsm_request(vp, NFSPROC_SETATTR, td, cred);
820 nfsm_wcc_data(vp, wccflag);
822 nfsm_loadattr(vp, NULL);
829 * nfs lookup call, one step at a time...
830 * First look in cache
831 * If not found, unlock the directory nfsnode and do the rpc
834 nfs_lookup(struct vop_lookup_args *ap)
836 struct componentname *cnp = ap->a_cnp;
837 struct vnode *dvp = ap->a_dvp;
838 struct vnode **vpp = ap->a_vpp;
839 int flags = cnp->cn_flags;
841 struct nfsmount *nmp;
843 struct mbuf *mreq, *mrep, *md, *mb;
847 int error = 0, attrflag, fhsize;
848 int v3 = NFS_ISV3(dvp);
849 struct thread *td = cnp->cn_thread;
852 if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
853 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
855 if (dvp->v_type != VDIR)
857 nmp = VFSTONFS(dvp->v_mount);
859 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0) {
863 if ((error = cache_lookup(dvp, vpp, cnp)) && error != ENOENT) {
867 if (!VOP_GETATTR(newvp, &vattr, cnp->cn_cred, td)
868 && vattr.va_ctime.tv_sec == VTONFS(newvp)->n_ctime) {
869 nfsstats.lookupcache_hits++;
870 if (cnp->cn_nameiop != LOOKUP &&
872 cnp->cn_flags |= SAVENAME;
884 nfsstats.lookupcache_misses++;
885 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
886 len = cnp->cn_namelen;
887 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
888 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
890 bpos = mtod(mb, caddr_t);
892 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
893 nfsm_request(dvp, NFSPROC_LOOKUP, cnp->cn_thread, cnp->cn_cred);
896 nfsm_postop_attr(dvp, attrflag);
901 nfsm_getfh(fhp, fhsize, v3);
904 * Handle RENAME case...
906 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
907 if (NFS_CMPFH(np, fhp, fhsize)) {
911 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, LK_EXCLUSIVE);
918 nfsm_postop_attr(newvp, attrflag);
919 nfsm_postop_attr(dvp, attrflag);
921 nfsm_loadattr(newvp, NULL);
924 cnp->cn_flags |= SAVENAME;
928 if (flags & ISDOTDOT) {
929 VOP_UNLOCK(dvp, 0, td);
930 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
931 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, td);
935 } else if (NFS_CMPFH(np, fhp, fhsize)) {
939 error = nfs_nget(dvp->v_mount, fhp, fhsize, &np, cnp->cn_lkflags);
947 nfsm_postop_attr(newvp, attrflag);
948 nfsm_postop_attr(dvp, attrflag);
950 nfsm_loadattr(newvp, NULL);
951 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
952 cnp->cn_flags |= SAVENAME;
953 if ((cnp->cn_flags & MAKEENTRY) &&
954 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN))) {
955 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
956 cache_enter(dvp, newvp, cnp);
962 if (newvp != NULLVP) {
966 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
967 (flags & ISLASTCN) && error == ENOENT) {
968 if (dvp->v_mount->mnt_flag & MNT_RDONLY)
973 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
974 cnp->cn_flags |= SAVENAME;
981 * Just call nfs_bioread() to do the work.
984 nfs_read(struct vop_read_args *ap)
986 struct vnode *vp = ap->a_vp;
988 switch (vp->v_type) {
990 return (nfs_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1002 nfs_readlink(struct vop_readlink_args *ap)
1004 struct vnode *vp = ap->a_vp;
1006 if (vp->v_type != VLNK)
1008 return (nfs_bioread(vp, ap->a_uio, 0, ap->a_cred));
1012 * Do a readlink rpc.
1013 * Called by nfs_doio() from below the buffer cache.
1016 nfs_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1019 int error = 0, len, attrflag;
1020 struct mbuf *mreq, *mrep, *md, *mb;
1021 int v3 = NFS_ISV3(vp);
1023 nfsstats.rpccnt[NFSPROC_READLINK]++;
1024 mreq = nfsm_reqhead(vp, NFSPROC_READLINK, NFSX_FH(v3));
1026 bpos = mtod(mb, caddr_t);
1028 nfsm_request(vp, NFSPROC_READLINK, uiop->uio_td, cred);
1030 nfsm_postop_attr(vp, attrflag);
1032 nfsm_strsiz(len, NFS_MAXPATHLEN);
1033 if (len == NFS_MAXPATHLEN) {
1034 struct nfsnode *np = VTONFS(vp);
1035 mtx_lock(&np->n_mtx);
1036 if (np->n_size && np->n_size < NFS_MAXPATHLEN)
1038 mtx_unlock(&np->n_mtx);
1040 nfsm_mtouio(uiop, len);
1052 nfs_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1056 struct mbuf *mreq, *mrep, *md, *mb;
1057 struct nfsmount *nmp;
1058 int error = 0, len, retlen, tsiz, eof, attrflag;
1059 int v3 = NFS_ISV3(vp);
1065 nmp = VFSTONFS(vp->v_mount);
1066 tsiz = uiop->uio_resid;
1067 mtx_lock(&nmp->nm_mtx);
1068 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1069 mtx_unlock(&nmp->nm_mtx);
1072 rsize = nmp->nm_rsize;
1073 mtx_unlock(&nmp->nm_mtx);
1075 nfsstats.rpccnt[NFSPROC_READ]++;
1076 len = (tsiz > rsize) ? rsize : tsiz;
1077 mreq = nfsm_reqhead(vp, NFSPROC_READ, NFSX_FH(v3) + NFSX_UNSIGNED * 3);
1079 bpos = mtod(mb, caddr_t);
1081 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED * 3);
1083 txdr_hyper(uiop->uio_offset, tl);
1084 *(tl + 2) = txdr_unsigned(len);
1086 *tl++ = txdr_unsigned(uiop->uio_offset);
1087 *tl++ = txdr_unsigned(len);
1090 nfsm_request(vp, NFSPROC_READ, uiop->uio_td, cred);
1092 nfsm_postop_attr(vp, attrflag);
1097 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED);
1098 eof = fxdr_unsigned(int, *(tl + 1));
1100 nfsm_loadattr(vp, NULL);
1102 nfsm_strsiz(retlen, rsize);
1103 nfsm_mtouio(uiop, retlen);
1107 if (eof || retlen == 0) {
1110 } else if (retlen < len) {
1122 nfs_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1123 int *iomode, int *must_commit)
1128 struct mbuf *mreq, *mrep, *md, *mb;
1129 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
1130 int error = 0, len, tsiz, wccflag = NFSV3_WCCRATTR, rlen, commit;
1131 int v3 = NFS_ISV3(vp), committed = NFSV3WRITE_FILESYNC;
1135 if (uiop->uio_iovcnt != 1)
1136 panic("nfs: writerpc iovcnt > 1");
1139 tsiz = uiop->uio_resid;
1140 mtx_lock(&nmp->nm_mtx);
1141 if (uiop->uio_offset + tsiz > nmp->nm_maxfilesize) {
1142 mtx_unlock(&nmp->nm_mtx);
1145 wsize = nmp->nm_wsize;
1146 mtx_unlock(&nmp->nm_mtx);
1148 nfsstats.rpccnt[NFSPROC_WRITE]++;
1149 len = (tsiz > wsize) ? wsize : tsiz;
1150 mreq = nfsm_reqhead(vp, NFSPROC_WRITE,
1151 NFSX_FH(v3) + 5 * NFSX_UNSIGNED + nfsm_rndup(len));
1153 bpos = mtod(mb, caddr_t);
1156 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
1157 txdr_hyper(uiop->uio_offset, tl);
1159 *tl++ = txdr_unsigned(len);
1160 *tl++ = txdr_unsigned(*iomode);
1161 *tl = txdr_unsigned(len);
1165 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1166 /* Set both "begin" and "current" to non-garbage. */
1167 x = txdr_unsigned((u_int32_t)uiop->uio_offset);
1168 *tl++ = x; /* "begin offset" */
1169 *tl++ = x; /* "current offset" */
1170 x = txdr_unsigned(len);
1171 *tl++ = x; /* total to this offset */
1172 *tl = x; /* size of this write */
1174 nfsm_uiotom(uiop, len);
1175 nfsm_request(vp, NFSPROC_WRITE, uiop->uio_td, cred);
1177 wccflag = NFSV3_WCCCHK;
1178 nfsm_wcc_data(vp, wccflag);
1180 tl = nfsm_dissect(u_int32_t *, 2 * NFSX_UNSIGNED
1181 + NFSX_V3WRITEVERF);
1182 rlen = fxdr_unsigned(int, *tl++);
1187 } else if (rlen < len) {
1188 backup = len - rlen;
1189 uiop->uio_iov->iov_base =
1190 (char *)uiop->uio_iov->iov_base -
1192 uiop->uio_iov->iov_len += backup;
1193 uiop->uio_offset -= backup;
1194 uiop->uio_resid += backup;
1197 commit = fxdr_unsigned(int, *tl++);
1200 * Return the lowest committment level
1201 * obtained by any of the RPCs.
1203 if (committed == NFSV3WRITE_FILESYNC)
1205 else if (committed == NFSV3WRITE_DATASYNC &&
1206 commit == NFSV3WRITE_UNSTABLE)
1208 mtx_lock(&nmp->nm_mtx);
1209 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0){
1210 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1212 nmp->nm_state |= NFSSTA_HASWRITEVERF;
1213 } else if (bcmp((caddr_t)tl,
1214 (caddr_t)nmp->nm_verf, NFSX_V3WRITEVERF)) {
1216 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
1219 mtx_unlock(&nmp->nm_mtx);
1222 nfsm_loadattr(vp, NULL);
1225 mtx_lock(&(VTONFS(vp))->n_mtx);
1226 VTONFS(vp)->n_mtime = VTONFS(vp)->n_vattr.va_mtime;
1227 mtx_unlock(&(VTONFS(vp))->n_mtx);
1235 if (vp->v_mount->mnt_kern_flag & MNTK_ASYNC)
1236 committed = NFSV3WRITE_FILESYNC;
1237 *iomode = committed;
1239 uiop->uio_resid = tsiz;
1245 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1246 * mode set to specify the file type and the size field for rdev.
1249 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1252 struct nfsv2_sattr *sp;
1254 struct vnode *newvp = NULL;
1255 struct nfsnode *np = NULL;
1258 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0;
1259 struct mbuf *mreq, *mrep, *md, *mb;
1261 int v3 = NFS_ISV3(dvp);
1263 if (vap->va_type == VCHR || vap->va_type == VBLK)
1264 rdev = txdr_unsigned(vap->va_rdev);
1265 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1268 return (EOPNOTSUPP);
1270 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1273 nfsstats.rpccnt[NFSPROC_MKNOD]++;
1274 mreq = nfsm_reqhead(dvp, NFSPROC_MKNOD, NFSX_FH(v3) + 4 * NFSX_UNSIGNED +
1275 + nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1277 bpos = mtod(mb, caddr_t);
1278 nfsm_fhtom(dvp, v3);
1279 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1281 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1282 *tl++ = vtonfsv3_type(vap->va_type);
1283 nfsm_v3attrbuild(vap, FALSE);
1284 if (vap->va_type == VCHR || vap->va_type == VBLK) {
1285 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
1286 *tl++ = txdr_unsigned(umajor(vap->va_rdev));
1287 *tl = txdr_unsigned(uminor(vap->va_rdev));
1290 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1291 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1292 sp->sa_uid = nfs_xdrneg1;
1293 sp->sa_gid = nfs_xdrneg1;
1295 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1296 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1298 nfsm_request(dvp, NFSPROC_MKNOD, cnp->cn_thread, cnp->cn_cred);
1300 nfsm_mtofh(dvp, newvp, v3, gotvp);
1306 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1307 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1313 nfsm_wcc_data(dvp, wccflag);
1320 if (cnp->cn_flags & MAKEENTRY)
1321 cache_enter(dvp, newvp, cnp);
1324 mtx_lock(&(VTONFS(dvp))->n_mtx);
1325 VTONFS(dvp)->n_flag |= NMODIFIED;
1327 VTONFS(dvp)->n_attrstamp = 0;
1328 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1334 * just call nfs_mknodrpc() to do the work.
1338 nfs_mknod(struct vop_mknod_args *ap)
1340 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1343 static u_long create_verf;
1345 * nfs file create call
1348 nfs_create(struct vop_create_args *ap)
1350 struct vnode *dvp = ap->a_dvp;
1351 struct vattr *vap = ap->a_vap;
1352 struct componentname *cnp = ap->a_cnp;
1353 struct nfsv2_sattr *sp;
1355 struct nfsnode *np = NULL;
1356 struct vnode *newvp = NULL;
1358 int error = 0, wccflag = NFSV3_WCCRATTR, gotvp = 0, fmode = 0;
1359 struct mbuf *mreq, *mrep, *md, *mb;
1361 int v3 = NFS_ISV3(dvp);
1364 * Oops, not for me..
1366 if (vap->va_type == VSOCK)
1367 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1369 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1372 if (vap->va_vaflags & VA_EXCLUSIVE)
1375 nfsstats.rpccnt[NFSPROC_CREATE]++;
1376 mreq = nfsm_reqhead(dvp, NFSPROC_CREATE, NFSX_FH(v3) + 2 * NFSX_UNSIGNED +
1377 nfsm_rndup(cnp->cn_namelen) + NFSX_SATTR(v3));
1379 bpos = mtod(mb, caddr_t);
1380 nfsm_fhtom(dvp, v3);
1381 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1383 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
1384 if (fmode & O_EXCL) {
1385 *tl = txdr_unsigned(NFSV3CREATE_EXCLUSIVE);
1386 tl = nfsm_build(u_int32_t *, NFSX_V3CREATEVERF);
1388 if (!TAILQ_EMPTY(&in_ifaddrhead))
1389 *tl++ = IA_SIN(TAILQ_FIRST(&in_ifaddrhead))->sin_addr.s_addr;
1392 *tl++ = create_verf;
1393 *tl = ++create_verf;
1395 *tl = txdr_unsigned(NFSV3CREATE_UNCHECKED);
1396 nfsm_v3attrbuild(vap, FALSE);
1399 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1400 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
1401 sp->sa_uid = nfs_xdrneg1;
1402 sp->sa_gid = nfs_xdrneg1;
1404 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1405 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1407 nfsm_request(dvp, NFSPROC_CREATE, cnp->cn_thread, cnp->cn_cred);
1409 nfsm_mtofh(dvp, newvp, v3, gotvp);
1415 error = nfs_lookitup(dvp, cnp->cn_nameptr,
1416 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread, &np);
1422 nfsm_wcc_data(dvp, wccflag);
1426 if (v3 && (fmode & O_EXCL) && error == NFSERR_NOTSUPP) {
1432 } else if (v3 && (fmode & O_EXCL)) {
1434 * We are normally called with only a partially initialized
1435 * VAP. Since the NFSv3 spec says that server may use the
1436 * file attributes to store the verifier, the spec requires
1437 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1438 * in atime, but we can't really assume that all servers will
1439 * so we ensure that our SETATTR sets both atime and mtime.
1441 if (vap->va_mtime.tv_sec == VNOVAL)
1442 vfs_timestamp(&vap->va_mtime);
1443 if (vap->va_atime.tv_sec == VNOVAL)
1444 vap->va_atime = vap->va_mtime;
1445 error = nfs_setattrrpc(newvp, vap, cnp->cn_cred, cnp->cn_thread);
1450 if (cnp->cn_flags & MAKEENTRY)
1451 cache_enter(dvp, newvp, cnp);
1454 mtx_lock(&(VTONFS(dvp))->n_mtx);
1455 VTONFS(dvp)->n_flag |= NMODIFIED;
1457 VTONFS(dvp)->n_attrstamp = 0;
1458 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1463 * nfs file remove call
1464 * To try and make nfs semantics closer to ufs semantics, a file that has
1465 * other processes using the vnode is renamed instead of removed and then
1466 * removed later on the last close.
1467 * - If v_usecount > 1
1468 * If a rename is not already in the works
1469 * call nfs_sillyrename() to set it up
1474 nfs_remove(struct vop_remove_args *ap)
1476 struct vnode *vp = ap->a_vp;
1477 struct vnode *dvp = ap->a_dvp;
1478 struct componentname *cnp = ap->a_cnp;
1479 struct nfsnode *np = VTONFS(vp);
1484 if ((cnp->cn_flags & HASBUF) == 0)
1485 panic("nfs_remove: no name");
1486 if (vrefcnt(vp) < 1)
1487 panic("nfs_remove: bad v_usecount");
1489 if (vp->v_type == VDIR)
1491 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1492 VOP_GETATTR(vp, &vattr, cnp->cn_cred, cnp->cn_thread) == 0 &&
1493 vattr.va_nlink > 1)) {
1495 * Purge the name cache so that the chance of a lookup for
1496 * the name succeeding while the remove is in progress is
1497 * minimized. Without node locking it can still happen, such
1498 * that an I/O op returns ESTALE, but since you get this if
1499 * another host removes the file..
1503 * throw away biocache buffers, mainly to avoid
1504 * unnecessary delayed writes later.
1506 error = nfs_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1508 if (error != EINTR && error != EIO)
1509 error = nfs_removerpc(dvp, cnp->cn_nameptr,
1510 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1512 * Kludge City: If the first reply to the remove rpc is lost..
1513 * the reply to the retransmitted request will be ENOENT
1514 * since the file was in fact removed
1515 * Therefore, we cheat and return success.
1517 if (error == ENOENT)
1519 } else if (!np->n_sillyrename)
1520 error = nfs_sillyrename(dvp, vp, cnp);
1521 np->n_attrstamp = 0;
1526 * nfs file remove rpc called from nfs_inactive
1529 nfs_removeit(struct sillyrename *sp)
1532 * Make sure that the directory vnode is still valid.
1533 * XXX we should lock sp->s_dvp here.
1535 if (sp->s_dvp->v_type == VBAD)
1537 return (nfs_removerpc(sp->s_dvp, sp->s_name, sp->s_namlen, sp->s_cred,
1542 * Nfs remove rpc, called from nfs_remove() and nfs_removeit().
1545 nfs_removerpc(struct vnode *dvp, const char *name, int namelen,
1546 struct ucred *cred, struct thread *td)
1549 int error = 0, wccflag = NFSV3_WCCRATTR;
1550 struct mbuf *mreq, *mrep, *md, *mb;
1551 int v3 = NFS_ISV3(dvp);
1553 nfsstats.rpccnt[NFSPROC_REMOVE]++;
1554 mreq = nfsm_reqhead(dvp, NFSPROC_REMOVE,
1555 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(namelen));
1557 bpos = mtod(mb, caddr_t);
1558 nfsm_fhtom(dvp, v3);
1559 nfsm_strtom(name, namelen, NFS_MAXNAMLEN);
1560 nfsm_request(dvp, NFSPROC_REMOVE, td, cred);
1562 nfsm_wcc_data(dvp, wccflag);
1565 mtx_lock(&(VTONFS(dvp))->n_mtx);
1566 VTONFS(dvp)->n_flag |= NMODIFIED;
1568 VTONFS(dvp)->n_attrstamp = 0;
1569 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1574 * nfs file rename call
1577 nfs_rename(struct vop_rename_args *ap)
1579 struct vnode *fvp = ap->a_fvp;
1580 struct vnode *tvp = ap->a_tvp;
1581 struct vnode *fdvp = ap->a_fdvp;
1582 struct vnode *tdvp = ap->a_tdvp;
1583 struct componentname *tcnp = ap->a_tcnp;
1584 struct componentname *fcnp = ap->a_fcnp;
1588 if ((tcnp->cn_flags & HASBUF) == 0 ||
1589 (fcnp->cn_flags & HASBUF) == 0)
1590 panic("nfs_rename: no name");
1592 /* Check for cross-device rename */
1593 if ((fvp->v_mount != tdvp->v_mount) ||
1594 (tvp && (fvp->v_mount != tvp->v_mount))) {
1600 nfs_printf("nfs_rename: fvp == tvp (can't happen)\n");
1604 if ((error = vn_lock(fvp, LK_EXCLUSIVE, fcnp->cn_thread)) != 0)
1608 * We have to flush B_DELWRI data prior to renaming
1609 * the file. If we don't, the delayed-write buffers
1610 * can be flushed out later after the file has gone stale
1611 * under NFSV3. NFSV2 does not have this problem because
1612 * ( as far as I can tell ) it flushes dirty buffers more
1615 * Skip the rename operation if the fsync fails, this can happen
1616 * due to the server's volume being full, when we pushed out data
1617 * that was written back to our cache earlier. Not checking for
1618 * this condition can result in potential (silent) data loss.
1620 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1621 VOP_UNLOCK(fvp, 0, fcnp->cn_thread);
1623 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1628 * If the tvp exists and is in use, sillyrename it before doing the
1629 * rename of the new file over it.
1630 * XXX Can't sillyrename a directory.
1632 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1633 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1638 error = nfs_renamerpc(fdvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1639 tdvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1642 if (fvp->v_type == VDIR) {
1643 if (tvp != NULL && tvp->v_type == VDIR)
1658 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1660 if (error == ENOENT)
1666 * nfs file rename rpc called from nfs_remove() above
1669 nfs_renameit(struct vnode *sdvp, struct componentname *scnp,
1670 struct sillyrename *sp)
1673 return (nfs_renamerpc(sdvp, scnp->cn_nameptr, scnp->cn_namelen, sdvp,
1674 sp->s_name, sp->s_namlen, scnp->cn_cred, scnp->cn_thread));
1678 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1681 nfs_renamerpc(struct vnode *fdvp, const char *fnameptr, int fnamelen,
1682 struct vnode *tdvp, const char *tnameptr, int tnamelen, struct ucred *cred,
1686 int error = 0, fwccflag = NFSV3_WCCRATTR, twccflag = NFSV3_WCCRATTR;
1687 struct mbuf *mreq, *mrep, *md, *mb;
1688 int v3 = NFS_ISV3(fdvp);
1690 nfsstats.rpccnt[NFSPROC_RENAME]++;
1691 mreq = nfsm_reqhead(fdvp, NFSPROC_RENAME,
1692 (NFSX_FH(v3) + NFSX_UNSIGNED)*2 + nfsm_rndup(fnamelen) +
1693 nfsm_rndup(tnamelen));
1695 bpos = mtod(mb, caddr_t);
1696 nfsm_fhtom(fdvp, v3);
1697 nfsm_strtom(fnameptr, fnamelen, NFS_MAXNAMLEN);
1698 nfsm_fhtom(tdvp, v3);
1699 nfsm_strtom(tnameptr, tnamelen, NFS_MAXNAMLEN);
1700 nfsm_request(fdvp, NFSPROC_RENAME, td, cred);
1702 nfsm_wcc_data(fdvp, fwccflag);
1703 nfsm_wcc_data(tdvp, twccflag);
1707 mtx_lock(&(VTONFS(fdvp))->n_mtx);
1708 VTONFS(fdvp)->n_flag |= NMODIFIED;
1709 mtx_unlock(&(VTONFS(fdvp))->n_mtx);
1710 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1711 VTONFS(tdvp)->n_flag |= NMODIFIED;
1712 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1714 VTONFS(fdvp)->n_attrstamp = 0;
1716 VTONFS(tdvp)->n_attrstamp = 0;
1721 * nfs hard link create call
1724 nfs_link(struct vop_link_args *ap)
1726 struct vnode *vp = ap->a_vp;
1727 struct vnode *tdvp = ap->a_tdvp;
1728 struct componentname *cnp = ap->a_cnp;
1730 int error = 0, wccflag = NFSV3_WCCRATTR, attrflag = 0;
1731 struct mbuf *mreq, *mrep, *md, *mb;
1734 if (vp->v_mount != tdvp->v_mount) {
1739 * Push all writes to the server, so that the attribute cache
1740 * doesn't get "out of sync" with the server.
1741 * XXX There should be a better way!
1743 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1746 nfsstats.rpccnt[NFSPROC_LINK]++;
1747 mreq = nfsm_reqhead(vp, NFSPROC_LINK,
1748 NFSX_FH(v3)*2 + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1750 bpos = mtod(mb, caddr_t);
1752 nfsm_fhtom(tdvp, v3);
1753 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1754 nfsm_request(vp, NFSPROC_LINK, cnp->cn_thread, cnp->cn_cred);
1756 nfsm_postop_attr(vp, attrflag);
1757 nfsm_wcc_data(tdvp, wccflag);
1761 mtx_lock(&(VTONFS(tdvp))->n_mtx);
1762 VTONFS(tdvp)->n_flag |= NMODIFIED;
1763 mtx_unlock(&(VTONFS(tdvp))->n_mtx);
1765 VTONFS(vp)->n_attrstamp = 0;
1767 VTONFS(tdvp)->n_attrstamp = 0;
1769 * Kludge: Map EEXIST => 0 assuming that it is a reply to a retry.
1771 if (error == EEXIST)
1777 * nfs symbolic link create call
1780 nfs_symlink(struct vop_symlink_args *ap)
1782 struct vnode *dvp = ap->a_dvp;
1783 struct vattr *vap = ap->a_vap;
1784 struct componentname *cnp = ap->a_cnp;
1785 struct nfsv2_sattr *sp;
1787 int slen, error = 0, wccflag = NFSV3_WCCRATTR, gotvp;
1788 struct mbuf *mreq, *mrep, *md, *mb;
1789 struct vnode *newvp = NULL;
1790 int v3 = NFS_ISV3(dvp);
1792 nfsstats.rpccnt[NFSPROC_SYMLINK]++;
1793 slen = strlen(ap->a_target);
1794 mreq = nfsm_reqhead(dvp, NFSPROC_SYMLINK, NFSX_FH(v3) + 2*NFSX_UNSIGNED +
1795 nfsm_rndup(cnp->cn_namelen) + nfsm_rndup(slen) + NFSX_SATTR(v3));
1797 bpos = mtod(mb, caddr_t);
1798 nfsm_fhtom(dvp, v3);
1799 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1801 nfsm_v3attrbuild(vap, FALSE);
1803 nfsm_strtom(ap->a_target, slen, NFS_MAXPATHLEN);
1805 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1806 sp->sa_mode = vtonfsv2_mode(VLNK, vap->va_mode);
1807 sp->sa_uid = nfs_xdrneg1;
1808 sp->sa_gid = nfs_xdrneg1;
1809 sp->sa_size = nfs_xdrneg1;
1810 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1811 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1815 * Issue the NFS request and get the rpc response.
1817 * Only NFSv3 responses returning an error of 0 actually return
1818 * a file handle that can be converted into newvp without having
1819 * to do an extra lookup rpc.
1821 nfsm_request(dvp, NFSPROC_SYMLINK, cnp->cn_thread, cnp->cn_cred);
1824 nfsm_mtofh(dvp, newvp, v3, gotvp);
1825 nfsm_wcc_data(dvp, wccflag);
1829 * out code jumps -> here, mrep is also freed.
1836 * If we get an EEXIST error, silently convert it to no-error
1837 * in case of an NFS retry.
1839 if (error == EEXIST)
1843 * If we do not have (or no longer have) an error, and we could
1844 * not extract the newvp from the response due to the request being
1845 * NFSv2 or the error being EEXIST. We have to do a lookup in order
1846 * to obtain a newvp to return.
1848 if (error == 0 && newvp == NULL) {
1849 struct nfsnode *np = NULL;
1851 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1852 cnp->cn_cred, cnp->cn_thread, &np);
1862 mtx_lock(&(VTONFS(dvp))->n_mtx);
1863 VTONFS(dvp)->n_flag |= NMODIFIED;
1864 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1866 VTONFS(dvp)->n_attrstamp = 0;
1874 nfs_mkdir(struct vop_mkdir_args *ap)
1876 struct vnode *dvp = ap->a_dvp;
1877 struct vattr *vap = ap->a_vap;
1878 struct componentname *cnp = ap->a_cnp;
1879 struct nfsv2_sattr *sp;
1881 struct nfsnode *np = NULL;
1882 struct vnode *newvp = NULL;
1884 int error = 0, wccflag = NFSV3_WCCRATTR;
1886 struct mbuf *mreq, *mrep, *md, *mb;
1888 int v3 = NFS_ISV3(dvp);
1890 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred, cnp->cn_thread)) != 0) {
1893 len = cnp->cn_namelen;
1894 nfsstats.rpccnt[NFSPROC_MKDIR]++;
1895 mreq = nfsm_reqhead(dvp, NFSPROC_MKDIR,
1896 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len) + NFSX_SATTR(v3));
1898 bpos = mtod(mb, caddr_t);
1899 nfsm_fhtom(dvp, v3);
1900 nfsm_strtom(cnp->cn_nameptr, len, NFS_MAXNAMLEN);
1902 nfsm_v3attrbuild(vap, FALSE);
1904 sp = nfsm_build(struct nfsv2_sattr *, NFSX_V2SATTR);
1905 sp->sa_mode = vtonfsv2_mode(VDIR, vap->va_mode);
1906 sp->sa_uid = nfs_xdrneg1;
1907 sp->sa_gid = nfs_xdrneg1;
1908 sp->sa_size = nfs_xdrneg1;
1909 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
1910 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
1912 nfsm_request(dvp, NFSPROC_MKDIR, cnp->cn_thread, cnp->cn_cred);
1914 nfsm_mtofh(dvp, newvp, v3, gotvp);
1916 nfsm_wcc_data(dvp, wccflag);
1919 mtx_lock(&(VTONFS(dvp))->n_mtx);
1920 VTONFS(dvp)->n_flag |= NMODIFIED;
1921 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1923 VTONFS(dvp)->n_attrstamp = 0;
1925 * Kludge: Map EEXIST => 0 assuming that you have a reply to a retry
1926 * if we can succeed in looking up the directory.
1928 if (error == EEXIST || (!error && !gotvp)) {
1933 error = nfs_lookitup(dvp, cnp->cn_nameptr, len, cnp->cn_cred,
1934 cnp->cn_thread, &np);
1937 if (newvp->v_type != VDIR)
1950 * nfs remove directory call
1953 nfs_rmdir(struct vop_rmdir_args *ap)
1955 struct vnode *vp = ap->a_vp;
1956 struct vnode *dvp = ap->a_dvp;
1957 struct componentname *cnp = ap->a_cnp;
1959 int error = 0, wccflag = NFSV3_WCCRATTR;
1960 struct mbuf *mreq, *mrep, *md, *mb;
1961 int v3 = NFS_ISV3(dvp);
1965 nfsstats.rpccnt[NFSPROC_RMDIR]++;
1966 mreq = nfsm_reqhead(dvp, NFSPROC_RMDIR,
1967 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(cnp->cn_namelen));
1969 bpos = mtod(mb, caddr_t);
1970 nfsm_fhtom(dvp, v3);
1971 nfsm_strtom(cnp->cn_nameptr, cnp->cn_namelen, NFS_MAXNAMLEN);
1972 nfsm_request(dvp, NFSPROC_RMDIR, cnp->cn_thread, cnp->cn_cred);
1974 nfsm_wcc_data(dvp, wccflag);
1977 mtx_lock(&(VTONFS(dvp))->n_mtx);
1978 VTONFS(dvp)->n_flag |= NMODIFIED;
1979 mtx_unlock(&(VTONFS(dvp))->n_mtx);
1981 VTONFS(dvp)->n_attrstamp = 0;
1985 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
1987 if (error == ENOENT)
1996 nfs_readdir(struct vop_readdir_args *ap)
1998 struct vnode *vp = ap->a_vp;
1999 struct nfsnode *np = VTONFS(vp);
2000 struct uio *uio = ap->a_uio;
2001 int tresid, error = 0;
2004 if (vp->v_type != VDIR)
2008 * First, check for hit on the EOF offset cache
2010 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2011 (np->n_flag & NMODIFIED) == 0) {
2012 if (VOP_GETATTR(vp, &vattr, ap->a_cred, uio->uio_td) == 0) {
2013 mtx_lock(&np->n_mtx);
2014 if (!NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2015 mtx_unlock(&np->n_mtx);
2016 nfsstats.direofcache_hits++;
2019 mtx_unlock(&np->n_mtx);
2024 * Call nfs_bioread() to do the real work.
2026 tresid = uio->uio_resid;
2027 error = nfs_bioread(vp, uio, 0, ap->a_cred);
2029 if (!error && uio->uio_resid == tresid) {
2030 nfsstats.direofcache_misses++;
2038 * Called from below the buffer cache by nfs_doio().
2041 nfs_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2044 struct dirent *dp = NULL;
2049 struct mbuf *mreq, *mrep, *md, *mb;
2051 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2052 struct nfsnode *dnp = VTONFS(vp);
2054 int error = 0, tlen, more_dirs = 1, blksiz = 0, bigenough = 1;
2056 int v3 = NFS_ISV3(vp);
2059 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2060 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2061 panic("nfs readdirrpc bad uio");
2065 * If there is no cookie, assume directory was stale.
2067 nfs_dircookie_lock(dnp);
2068 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2071 nfs_dircookie_unlock(dnp);
2073 nfs_dircookie_unlock(dnp);
2074 return (NFSERR_BAD_COOKIE);
2078 * Loop around doing readdir rpc's of size nm_readdirsize
2079 * truncated to a multiple of DIRBLKSIZ.
2080 * The stopping criteria is EOF or buffer full.
2082 while (more_dirs && bigenough) {
2083 nfsstats.rpccnt[NFSPROC_READDIR]++;
2084 mreq = nfsm_reqhead(vp, NFSPROC_READDIR, NFSX_FH(v3) +
2087 bpos = mtod(mb, caddr_t);
2090 tl = nfsm_build(u_int32_t *, 5 * NFSX_UNSIGNED);
2091 *tl++ = cookie.nfsuquad[0];
2092 *tl++ = cookie.nfsuquad[1];
2093 mtx_lock(&dnp->n_mtx);
2094 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2095 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2096 mtx_unlock(&dnp->n_mtx);
2098 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2099 *tl++ = cookie.nfsuquad[0];
2101 *tl = txdr_unsigned(nmp->nm_readdirsize);
2102 nfsm_request(vp, NFSPROC_READDIR, uiop->uio_td, cred);
2104 nfsm_postop_attr(vp, attrflag);
2106 tl = nfsm_dissect(u_int32_t *,
2108 mtx_lock(&dnp->n_mtx);
2109 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2110 dnp->n_cookieverf.nfsuquad[1] = *tl;
2111 mtx_unlock(&dnp->n_mtx);
2117 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2118 more_dirs = fxdr_unsigned(int, *tl);
2120 /* loop thru the dir entries, doctoring them to 4bsd form */
2121 while (more_dirs && bigenough) {
2123 tl = nfsm_dissect(u_int32_t *,
2125 fileno = fxdr_hyper(tl);
2126 len = fxdr_unsigned(int, *(tl + 2));
2128 tl = nfsm_dissect(u_int32_t *,
2130 fileno = fxdr_unsigned(u_quad_t, *tl++);
2131 len = fxdr_unsigned(int, *tl);
2133 if (len <= 0 || len > NFS_MAXNAMLEN) {
2138 tlen = nfsm_rndup(len);
2140 tlen += 4; /* To ensure null termination */
2141 left = DIRBLKSIZ - blksiz;
2142 if ((tlen + DIRHDSIZ) > left) {
2143 dp->d_reclen += left;
2144 uiop->uio_iov->iov_base =
2145 (char *)uiop->uio_iov->iov_base + left;
2146 uiop->uio_iov->iov_len -= left;
2147 uiop->uio_offset += left;
2148 uiop->uio_resid -= left;
2151 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2154 dp = (struct dirent *)uiop->uio_iov->iov_base;
2155 dp->d_fileno = (int)fileno;
2157 dp->d_reclen = tlen + DIRHDSIZ;
2158 dp->d_type = DT_UNKNOWN;
2159 blksiz += dp->d_reclen;
2160 if (blksiz == DIRBLKSIZ)
2162 uiop->uio_offset += DIRHDSIZ;
2163 uiop->uio_resid -= DIRHDSIZ;
2164 uiop->uio_iov->iov_base =
2165 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2166 uiop->uio_iov->iov_len -= DIRHDSIZ;
2167 nfsm_mtouio(uiop, len);
2168 cp = uiop->uio_iov->iov_base;
2170 *cp = '\0'; /* null terminate */
2171 uiop->uio_iov->iov_base =
2172 (char *)uiop->uio_iov->iov_base + tlen;
2173 uiop->uio_iov->iov_len -= tlen;
2174 uiop->uio_offset += tlen;
2175 uiop->uio_resid -= tlen;
2177 nfsm_adv(nfsm_rndup(len));
2179 tl = nfsm_dissect(u_int32_t *,
2182 tl = nfsm_dissect(u_int32_t *,
2186 cookie.nfsuquad[0] = *tl++;
2188 cookie.nfsuquad[1] = *tl++;
2193 more_dirs = fxdr_unsigned(int, *tl);
2196 * If at end of rpc data, get the eof boolean
2199 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2200 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2205 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2206 * by increasing d_reclen for the last record.
2209 left = DIRBLKSIZ - blksiz;
2210 dp->d_reclen += left;
2211 uiop->uio_iov->iov_base =
2212 (char *)uiop->uio_iov->iov_base + left;
2213 uiop->uio_iov->iov_len -= left;
2214 uiop->uio_offset += left;
2215 uiop->uio_resid -= left;
2219 * We are now either at the end of the directory or have filled the
2223 dnp->n_direofoffset = uiop->uio_offset;
2225 if (uiop->uio_resid > 0)
2226 nfs_printf("EEK! readdirrpc resid > 0\n");
2227 nfs_dircookie_lock(dnp);
2228 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2230 nfs_dircookie_unlock(dnp);
2237 * NFS V3 readdir plus RPC. Used in place of nfs_readdirrpc().
2240 nfs_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
2246 struct vnode *newvp;
2248 caddr_t bpos, dpos, dpossav1, dpossav2;
2249 struct mbuf *mreq, *mrep, *md, *mb, *mdsav1, *mdsav2;
2250 struct nameidata nami, *ndp = &nami;
2251 struct componentname *cnp = &ndp->ni_cnd;
2253 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2254 struct nfsnode *dnp = VTONFS(vp), *np;
2257 int error = 0, tlen, more_dirs = 1, blksiz = 0, doit, bigenough = 1, i;
2258 int attrflag, fhsize;
2264 if (uiop->uio_iovcnt != 1 || (uiop->uio_offset & (DIRBLKSIZ - 1)) ||
2265 (uiop->uio_resid & (DIRBLKSIZ - 1)))
2266 panic("nfs readdirplusrpc bad uio");
2272 * If there is no cookie, assume directory was stale.
2274 nfs_dircookie_lock(dnp);
2275 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 0);
2278 nfs_dircookie_unlock(dnp);
2280 nfs_dircookie_unlock(dnp);
2281 return (NFSERR_BAD_COOKIE);
2284 * Loop around doing readdir rpc's of size nm_readdirsize
2285 * truncated to a multiple of DIRBLKSIZ.
2286 * The stopping criteria is EOF or buffer full.
2288 while (more_dirs && bigenough) {
2289 nfsstats.rpccnt[NFSPROC_READDIRPLUS]++;
2290 mreq = nfsm_reqhead(vp, NFSPROC_READDIRPLUS,
2291 NFSX_FH(1) + 6 * NFSX_UNSIGNED);
2293 bpos = mtod(mb, caddr_t);
2295 tl = nfsm_build(u_int32_t *, 6 * NFSX_UNSIGNED);
2296 *tl++ = cookie.nfsuquad[0];
2297 *tl++ = cookie.nfsuquad[1];
2298 mtx_lock(&dnp->n_mtx);
2299 *tl++ = dnp->n_cookieverf.nfsuquad[0];
2300 *tl++ = dnp->n_cookieverf.nfsuquad[1];
2301 mtx_unlock(&dnp->n_mtx);
2302 *tl++ = txdr_unsigned(nmp->nm_readdirsize);
2303 *tl = txdr_unsigned(nmp->nm_rsize);
2304 nfsm_request(vp, NFSPROC_READDIRPLUS, uiop->uio_td, cred);
2305 nfsm_postop_attr(vp, attrflag);
2310 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2311 mtx_lock(&dnp->n_mtx);
2312 dnp->n_cookieverf.nfsuquad[0] = *tl++;
2313 dnp->n_cookieverf.nfsuquad[1] = *tl++;
2314 mtx_unlock(&dnp->n_mtx);
2315 more_dirs = fxdr_unsigned(int, *tl);
2317 /* loop thru the dir entries, doctoring them to 4bsd form */
2318 while (more_dirs && bigenough) {
2319 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2320 fileno = fxdr_hyper(tl);
2321 len = fxdr_unsigned(int, *(tl + 2));
2322 if (len <= 0 || len > NFS_MAXNAMLEN) {
2327 tlen = nfsm_rndup(len);
2329 tlen += 4; /* To ensure null termination*/
2330 left = DIRBLKSIZ - blksiz;
2331 if ((tlen + DIRHDSIZ) > left) {
2332 dp->d_reclen += left;
2333 uiop->uio_iov->iov_base =
2334 (char *)uiop->uio_iov->iov_base + left;
2335 uiop->uio_iov->iov_len -= left;
2336 uiop->uio_offset += left;
2337 uiop->uio_resid -= left;
2340 if ((tlen + DIRHDSIZ) > uiop->uio_resid)
2343 dp = (struct dirent *)uiop->uio_iov->iov_base;
2344 dp->d_fileno = (int)fileno;
2346 dp->d_reclen = tlen + DIRHDSIZ;
2347 dp->d_type = DT_UNKNOWN;
2348 blksiz += dp->d_reclen;
2349 if (blksiz == DIRBLKSIZ)
2351 uiop->uio_offset += DIRHDSIZ;
2352 uiop->uio_resid -= DIRHDSIZ;
2353 uiop->uio_iov->iov_base =
2354 (char *)uiop->uio_iov->iov_base + DIRHDSIZ;
2355 uiop->uio_iov->iov_len -= DIRHDSIZ;
2356 cnp->cn_nameptr = uiop->uio_iov->iov_base;
2357 cnp->cn_namelen = len;
2358 nfsm_mtouio(uiop, len);
2359 cp = uiop->uio_iov->iov_base;
2362 uiop->uio_iov->iov_base =
2363 (char *)uiop->uio_iov->iov_base + tlen;
2364 uiop->uio_iov->iov_len -= tlen;
2365 uiop->uio_offset += tlen;
2366 uiop->uio_resid -= tlen;
2368 nfsm_adv(nfsm_rndup(len));
2369 tl = nfsm_dissect(u_int32_t *, 3 * NFSX_UNSIGNED);
2371 cookie.nfsuquad[0] = *tl++;
2372 cookie.nfsuquad[1] = *tl++;
2377 * Since the attributes are before the file handle
2378 * (sigh), we must skip over the attributes and then
2379 * come back and get them.
2381 attrflag = fxdr_unsigned(int, *tl);
2385 nfsm_adv(NFSX_V3FATTR);
2386 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2387 doit = fxdr_unsigned(int, *tl);
2389 * Skip loading the attrs for "..". There's a
2390 * race between loading the attrs here and
2391 * lookups that look for the directory currently
2392 * being read (in the parent). We try to acquire
2393 * the exclusive lock on ".." here, owning the
2394 * lock on the directory being read. Lookup will
2395 * hold the lock on ".." and try to acquire the
2396 * lock on the directory being read.
2398 * There are other ways of fixing this, one would
2399 * be to do a trylock on the ".." vnode and skip
2400 * loading the attrs on ".." if it happens to be
2401 * locked by another process. But skipping the
2402 * attrload on ".." seems the easiest option.
2404 if (strcmp(dp->d_name, "..") == 0) {
2407 * We've already skipped over the attrs,
2408 * skip over the filehandle. And store d_type
2411 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2412 i = fxdr_unsigned(int, *tl);
2413 nfsm_adv(nfsm_rndup(i));
2414 dp->d_type = IFTODT(VTTOIF(VDIR));
2417 nfsm_getfh(fhp, fhsize, 1);
2418 if (NFS_CMPFH(dnp, fhp, fhsize)) {
2423 error = nfs_nget(vp->v_mount, fhp,
2424 fhsize, &np, LK_EXCLUSIVE);
2431 if (doit && bigenough) {
2436 nfsm_loadattr(newvp, NULL);
2440 IFTODT(VTTOIF(np->n_vattr.va_type));
2442 /* Update n_ctime, so subsequent lookup doesn't purge entry */
2443 np->n_ctime = np->n_vattr.va_ctime.tv_sec;
2444 cache_enter(ndp->ni_dvp, ndp->ni_vp, cnp);
2447 /* Just skip over the file handle */
2448 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2449 i = fxdr_unsigned(int, *tl);
2451 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2452 fhsize = fxdr_unsigned(int, *tl);
2453 nfsm_adv(nfsm_rndup(fhsize));
2456 if (newvp != NULLVP) {
2463 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2464 more_dirs = fxdr_unsigned(int, *tl);
2467 * If at end of rpc data, get the eof boolean
2470 tl = nfsm_dissect(u_int32_t *, NFSX_UNSIGNED);
2471 more_dirs = (fxdr_unsigned(int, *tl) == 0);
2476 * Fill last record, iff any, out to a multiple of DIRBLKSIZ
2477 * by increasing d_reclen for the last record.
2480 left = DIRBLKSIZ - blksiz;
2481 dp->d_reclen += left;
2482 uiop->uio_iov->iov_base =
2483 (char *)uiop->uio_iov->iov_base + left;
2484 uiop->uio_iov->iov_len -= left;
2485 uiop->uio_offset += left;
2486 uiop->uio_resid -= left;
2490 * We are now either at the end of the directory or have filled the
2494 dnp->n_direofoffset = uiop->uio_offset;
2496 if (uiop->uio_resid > 0)
2497 nfs_printf("EEK! readdirplusrpc resid > 0\n");
2498 nfs_dircookie_lock(dnp);
2499 cookiep = nfs_getcookie(dnp, uiop->uio_offset, 1);
2501 nfs_dircookie_unlock(dnp);
2504 if (newvp != NULLVP) {
2515 * Silly rename. To make the NFS filesystem that is stateless look a little
2516 * more like the "ufs" a remove of an active vnode is translated to a rename
2517 * to a funny looking filename that is removed by nfs_inactive on the
2518 * nfsnode. There is the potential for another process on a different client
2519 * to create the same funny name between the nfs_lookitup() fails and the
2520 * nfs_rename() completes, but...
2523 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2525 struct sillyrename *sp;
2529 unsigned int lticks;
2534 if (vp->v_type == VDIR)
2535 panic("nfs: sillyrename dir");
2537 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2538 M_NFSREQ, M_WAITOK);
2539 sp->s_cred = crhold(cnp->cn_cred);
2541 sp->s_removeit = nfs_removeit;
2545 * Fudge together a funny name.
2546 * Changing the format of the funny name to accomodate more
2547 * sillynames per directory.
2548 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2549 * CPU ticks since boot.
2551 pid = cnp->cn_thread->td_proc->p_pid;
2552 lticks = (unsigned int)ticks;
2554 sp->s_namlen = sprintf(sp->s_name,
2555 ".nfs.%08x.%04x4.4", lticks,
2557 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2558 cnp->cn_thread, NULL))
2562 error = nfs_renameit(dvp, cnp, sp);
2565 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2566 cnp->cn_thread, &np);
2567 np->n_sillyrename = sp;
2572 free((caddr_t)sp, M_NFSREQ);
2577 * Look up a file name and optionally either update the file handle or
2578 * allocate an nfsnode, depending on the value of npp.
2579 * npp == NULL --> just do the lookup
2580 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2582 * *npp != NULL --> update the file handle in the vnode
2585 nfs_lookitup(struct vnode *dvp, const char *name, int len, struct ucred *cred,
2586 struct thread *td, struct nfsnode **npp)
2588 struct vnode *newvp = NULL;
2589 struct nfsnode *np, *dnp = VTONFS(dvp);
2591 int error = 0, fhlen, attrflag;
2592 struct mbuf *mreq, *mrep, *md, *mb;
2594 int v3 = NFS_ISV3(dvp);
2596 nfsstats.rpccnt[NFSPROC_LOOKUP]++;
2597 mreq = nfsm_reqhead(dvp, NFSPROC_LOOKUP,
2598 NFSX_FH(v3) + NFSX_UNSIGNED + nfsm_rndup(len));
2600 bpos = mtod(mb, caddr_t);
2601 nfsm_fhtom(dvp, v3);
2602 nfsm_strtom(name, len, NFS_MAXNAMLEN);
2603 nfsm_request(dvp, NFSPROC_LOOKUP, td, cred);
2604 if (npp && !error) {
2605 nfsm_getfh(nfhp, fhlen, v3);
2608 if (np->n_fhsize > NFS_SMALLFH && fhlen <= NFS_SMALLFH) {
2609 free((caddr_t)np->n_fhp, M_NFSBIGFH);
2610 np->n_fhp = &np->n_fh;
2611 } else if (np->n_fhsize <= NFS_SMALLFH && fhlen>NFS_SMALLFH)
2612 np->n_fhp =(nfsfh_t *)malloc(fhlen, M_NFSBIGFH, M_WAITOK);
2613 bcopy((caddr_t)nfhp, (caddr_t)np->n_fhp, fhlen);
2614 np->n_fhsize = fhlen;
2616 } else if (NFS_CMPFH(dnp, nfhp, fhlen)) {
2620 error = nfs_nget(dvp->v_mount, nfhp, fhlen, &np, LK_EXCLUSIVE);
2628 nfsm_postop_attr(newvp, attrflag);
2629 if (!attrflag && *npp == NULL) {
2638 nfsm_loadattr(newvp, NULL);
2642 if (npp && *npp == NULL) {
2657 * Nfs Version 3 commit rpc
2660 nfs_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2664 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2666 int error = 0, wccflag = NFSV3_WCCRATTR;
2667 struct mbuf *mreq, *mrep, *md, *mb;
2669 mtx_lock(&nmp->nm_mtx);
2670 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2671 mtx_unlock(&nmp->nm_mtx);
2674 mtx_unlock(&nmp->nm_mtx);
2675 nfsstats.rpccnt[NFSPROC_COMMIT]++;
2676 mreq = nfsm_reqhead(vp, NFSPROC_COMMIT, NFSX_FH(1));
2678 bpos = mtod(mb, caddr_t);
2680 tl = nfsm_build(u_int32_t *, 3 * NFSX_UNSIGNED);
2681 txdr_hyper(offset, tl);
2683 *tl = txdr_unsigned(cnt);
2684 nfsm_request(vp, NFSPROC_COMMIT, td, cred);
2685 nfsm_wcc_data(vp, wccflag);
2687 tl = nfsm_dissect(u_int32_t *, NFSX_V3WRITEVERF);
2688 if (bcmp((caddr_t)nmp->nm_verf, (caddr_t)tl,
2689 NFSX_V3WRITEVERF)) {
2690 bcopy((caddr_t)tl, (caddr_t)nmp->nm_verf,
2692 error = NFSERR_STALEWRITEVERF;
2702 * For async requests when nfsiod(s) are running, queue the request by
2703 * calling nfs_asyncio(), otherwise just all nfs_doio() to do the
2707 nfs_strategy(struct vop_strategy_args *ap)
2709 struct buf *bp = ap->a_bp;
2712 KASSERT(!(bp->b_flags & B_DONE), ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2713 KASSERT(BUF_REFCNT(bp) > 0, ("nfs_strategy: buffer %p not locked", bp));
2715 if (bp->b_iocmd == BIO_READ)
2721 * If the op is asynchronous and an i/o daemon is waiting
2722 * queue the request, wake it up and wait for completion
2723 * otherwise just do it ourselves.
2725 if ((bp->b_flags & B_ASYNC) == 0 ||
2726 nfs_asyncio(VFSTONFS(ap->a_vp->v_mount), bp, NOCRED, curthread))
2727 (void)nfs_doio(ap->a_vp, bp, cr, curthread);
2732 * fsync vnode op. Just call nfs_flush() with commit == 1.
2736 nfs_fsync(struct vop_fsync_args *ap)
2738 return (nfs_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1));
2742 * Flush all the blocks associated with a vnode.
2743 * Walk through the buffer pool and push any dirty pages
2744 * associated with the vnode.
2747 nfs_flush(struct vnode *vp, int waitfor, struct thread *td,
2750 struct nfsnode *np = VTONFS(vp);
2754 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2755 int s, error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2757 u_quad_t off, endoff, toff;
2758 struct ucred* wcred = NULL;
2759 struct buf **bvec = NULL;
2760 #ifndef NFS_COMMITBVECSIZ
2761 #define NFS_COMMITBVECSIZ 20
2763 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2764 int bvecsize = 0, bveccount;
2766 if (nmp->nm_flag & NFSMNT_INT)
2771 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2772 * server, but has not been committed to stable storage on the server
2773 * yet. On the first pass, the byte range is worked out and the commit
2774 * rpc is done. On the second pass, nfs_writebp() is called to do the
2781 if (NFS_ISV3(vp) && commit) {
2783 if (bvec != NULL && bvec != bvec_on_stack)
2786 * Count up how many buffers waiting for a commit.
2790 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2791 if (BUF_REFCNT(bp) == 0 &&
2792 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2793 == (B_DELWRI | B_NEEDCOMMIT))
2797 * Allocate space to remember the list of bufs to commit. It is
2798 * important to use M_NOWAIT here to avoid a race with nfs_write.
2799 * If we can't get memory (for whatever reason), we will end up
2800 * committing the buffers one-by-one in the loop below.
2802 if (bveccount > NFS_COMMITBVECSIZ) {
2804 * Release the vnode interlock to avoid a lock
2808 bvec = (struct buf **)
2809 malloc(bveccount * sizeof(struct buf *),
2813 bvec = bvec_on_stack;
2814 bvecsize = NFS_COMMITBVECSIZ;
2816 bvecsize = bveccount;
2818 bvec = bvec_on_stack;
2819 bvecsize = NFS_COMMITBVECSIZ;
2821 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2822 if (bvecpos >= bvecsize)
2824 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2825 nbp = TAILQ_NEXT(bp, b_bobufs);
2828 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2829 (B_DELWRI | B_NEEDCOMMIT)) {
2831 nbp = TAILQ_NEXT(bp, b_bobufs);
2837 * Work out if all buffers are using the same cred
2838 * so we can deal with them all with one commit.
2840 * NOTE: we are not clearing B_DONE here, so we have
2841 * to do it later on in this routine if we intend to
2842 * initiate I/O on the bp.
2844 * Note: to avoid loopback deadlocks, we do not
2845 * assign b_runningbufspace.
2848 wcred = bp->b_wcred;
2849 else if (wcred != bp->b_wcred)
2851 vfs_busy_pages(bp, 1);
2855 * bp is protected by being locked, but nbp is not
2856 * and vfs_busy_pages() may sleep. We have to
2859 nbp = TAILQ_NEXT(bp, b_bobufs);
2862 * A list of these buffers is kept so that the
2863 * second loop knows which buffers have actually
2864 * been committed. This is necessary, since there
2865 * may be a race between the commit rpc and new
2866 * uncommitted writes on the file.
2868 bvec[bvecpos++] = bp;
2869 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2873 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2882 * Commit data on the server, as required.
2883 * If all bufs are using the same wcred, then use that with
2884 * one call for all of them, otherwise commit each one
2887 if (wcred != NOCRED)
2888 retv = nfs_commit(vp, off, (int)(endoff - off),
2892 for (i = 0; i < bvecpos; i++) {
2895 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2897 size = (u_quad_t)(bp->b_dirtyend
2899 retv = nfs_commit(vp, off, (int)size,
2905 if (retv == NFSERR_STALEWRITEVERF)
2906 nfs_clearcommit(vp->v_mount);
2909 * Now, either mark the blocks I/O done or mark the
2910 * blocks dirty, depending on whether the commit
2913 for (i = 0; i < bvecpos; i++) {
2915 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2918 * Error, leave B_DELWRI intact
2920 vfs_unbusy_pages(bp);
2924 * Success, remove B_DELWRI ( bundirty() ).
2926 * b_dirtyoff/b_dirtyend seem to be NFS
2927 * specific. We should probably move that
2928 * into bundirty(). XXX
2931 bufobj_wref(&vp->v_bufobj);
2932 bp->b_flags |= B_ASYNC;
2934 bp->b_flags &= ~B_DONE;
2935 bp->b_ioflags &= ~BIO_ERROR;
2936 bp->b_dirtyoff = bp->b_dirtyend = 0;
2944 * Start/do any write(s) that are required.
2949 TAILQ_FOREACH_SAFE(bp, &vp->v_bufobj.bo_dirty.bv_hd, b_bobufs, nbp) {
2950 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2951 if (waitfor != MNT_WAIT || passone)
2954 error = BUF_TIMELOCK(bp,
2955 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2956 VI_MTX(vp), "nfsfsync", slpflag, slptimeo);
2962 if (error == ENOLCK)
2964 if (nfs_sigintr(nmp, NULL, td)) {
2968 if (slpflag == PCATCH) {
2974 if ((bp->b_flags & B_DELWRI) == 0)
2975 panic("nfs_fsync: not dirty");
2976 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2982 if (passone || !commit)
2983 bp->b_flags |= B_ASYNC;
2985 bp->b_flags |= B_ASYNC;
2988 if (nfs_sigintr(nmp, NULL, td)) {
3000 if (waitfor == MNT_WAIT) {
3001 while (vp->v_bufobj.bo_numoutput) {
3002 error = bufobj_wwait(&vp->v_bufobj, slpflag, slptimeo);
3005 error = nfs_sigintr(nmp, NULL, td);
3008 if (slpflag == PCATCH) {
3015 if (vp->v_bufobj.bo_dirty.bv_cnt != 0 && commit) {
3020 * Wait for all the async IO requests to drain
3023 mtx_lock(&np->n_mtx);
3024 while (np->n_directio_asyncwr > 0) {
3025 np->n_flag |= NFSYNCWAIT;
3026 error = nfs_msleep(td, (caddr_t)&np->n_directio_asyncwr,
3027 &np->n_mtx, slpflag | (PRIBIO + 1),
3030 if (nfs_sigintr(nmp, (struct nfsreq *)0, td)) {
3031 mtx_unlock(&np->n_mtx);
3037 mtx_unlock(&np->n_mtx);
3040 mtx_lock(&np->n_mtx);
3041 if (np->n_flag & NWRITEERR) {
3042 error = np->n_error;
3043 np->n_flag &= ~NWRITEERR;
3045 if (commit && vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
3046 vp->v_bufobj.bo_numoutput == 0 && np->n_directio_asyncwr == 0)
3047 np->n_flag &= ~NMODIFIED;
3048 mtx_unlock(&np->n_mtx);
3050 if (bvec != NULL && bvec != bvec_on_stack)
3056 * NFS advisory byte-level locks.
3059 nfs_advlock(struct vop_advlock_args *ap)
3064 if ((VFSTONFS(ap->a_vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3065 struct nfsnode *np = VTONFS(ap->a_vp);
3067 error = lf_advlock(ap, &(np->n_lockf), np->n_size);
3070 error = nfs_dolock(ap);
3077 * Print out the contents of an nfsnode.
3080 nfs_print(struct vop_print_args *ap)
3082 struct vnode *vp = ap->a_vp;
3083 struct nfsnode *np = VTONFS(vp);
3085 nfs_printf("\tfileid %ld fsid 0x%x",
3086 np->n_vattr.va_fileid, np->n_vattr.va_fsid);
3087 if (vp->v_type == VFIFO)
3094 * This is the "real" nfs::bwrite(struct buf*).
3095 * We set B_CACHE if this is a VMIO buffer.
3098 nfs_writebp(struct buf *bp, int force __unused, struct thread *td)
3101 int oldflags = bp->b_flags;
3107 if (BUF_REFCNT(bp) == 0)
3108 panic("bwrite: buffer is not locked???");
3110 if (bp->b_flags & B_INVAL) {
3115 bp->b_flags |= B_CACHE;
3118 * Undirty the bp. We will redirty it later if the I/O fails.
3123 bp->b_flags &= ~B_DONE;
3124 bp->b_ioflags &= ~BIO_ERROR;
3125 bp->b_iocmd = BIO_WRITE;
3127 bufobj_wref(bp->b_bufobj);
3128 curthread->td_proc->p_stats->p_ru.ru_oublock++;
3132 * Note: to avoid loopback deadlocks, we do not
3133 * assign b_runningbufspace.
3135 vfs_busy_pages(bp, 1);
3138 bp->b_iooffset = dbtob(bp->b_blkno);
3141 if( (oldflags & B_ASYNC) == 0) {
3142 int rtval = bufwait(bp);
3144 if (oldflags & B_DELWRI) {
3157 * nfs special file access vnode op.
3158 * Essentially just get vattr and then imitate iaccess() since the device is
3159 * local to the client.
3162 nfsspec_access(struct vop_access_args *ap)
3165 struct ucred *cred = ap->a_cred;
3166 struct vnode *vp = ap->a_vp;
3167 mode_t mode = ap->a_mode;
3172 * Disallow write attempts on filesystems mounted read-only;
3173 * unless the file is a socket, fifo, or a block or character
3174 * device resident on the filesystem.
3176 if ((mode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3177 switch (vp->v_type) {
3187 error = VOP_GETATTR(vp, vap, cred, ap->a_td);
3190 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3197 * Read wrapper for fifos.
3200 nfsfifo_read(struct vop_read_args *ap)
3202 struct nfsnode *np = VTONFS(ap->a_vp);
3208 mtx_lock(&np->n_mtx);
3210 getnanotime(&np->n_atim);
3211 mtx_unlock(&np->n_mtx);
3212 error = fifo_specops.vop_read(ap);
3217 * Write wrapper for fifos.
3220 nfsfifo_write(struct vop_write_args *ap)
3222 struct nfsnode *np = VTONFS(ap->a_vp);
3227 mtx_lock(&np->n_mtx);
3229 getnanotime(&np->n_mtim);
3230 mtx_unlock(&np->n_mtx);
3231 return(fifo_specops.vop_write(ap));
3235 * Close wrapper for fifos.
3237 * Update the times on the nfsnode then do fifo close.
3240 nfsfifo_close(struct vop_close_args *ap)
3242 struct vnode *vp = ap->a_vp;
3243 struct nfsnode *np = VTONFS(vp);
3247 mtx_lock(&np->n_mtx);
3248 if (np->n_flag & (NACC | NUPD)) {
3250 if (np->n_flag & NACC)
3252 if (np->n_flag & NUPD)
3255 if (vrefcnt(vp) == 1 &&
3256 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3258 if (np->n_flag & NACC)
3259 vattr.va_atime = np->n_atim;
3260 if (np->n_flag & NUPD)
3261 vattr.va_mtime = np->n_mtim;
3262 mtx_unlock(&np->n_mtx);
3263 (void)VOP_SETATTR(vp, &vattr, ap->a_cred, ap->a_td);
3267 mtx_unlock(&np->n_mtx);
3269 return (fifo_specops.vop_close(ap));
3273 * Just call nfs_writebp() with the force argument set to 1.
3275 * NOTE: B_DONE may or may not be set in a_bp on call.
3278 nfs_bwrite(struct buf *bp)
3281 return (nfs_writebp(bp, 1, curthread));
3284 struct buf_ops buf_ops_nfs = {
3285 .bop_name = "buf_ops_nfs",
3286 .bop_write = nfs_bwrite,
3287 .bop_strategy = bufstrategy,
3288 .bop_sync = bufsync,
3289 .bop_bdflush = bufbdflush,