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 * 3. 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 * from 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, 3 and 4
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>
53 #include <sys/malloc.h>
55 #include <sys/namei.h>
56 #include <sys/socket.h>
57 #include <sys/vnode.h>
58 #include <sys/dirent.h>
59 #include <sys/fcntl.h>
60 #include <sys/lockf.h>
62 #include <sys/sysctl.h>
63 #include <sys/signalvar.h>
66 #include <vm/vm_extern.h>
67 #include <vm/vm_object.h>
69 #include <fs/nfs/nfsport.h>
70 #include <fs/nfsclient/nfsnode.h>
71 #include <fs/nfsclient/nfsmount.h>
72 #include <fs/nfsclient/nfs.h>
73 #include <fs/nfsclient/nfs_kdtrace.h>
76 #include <netinet/in.h>
77 #include <netinet/in_var.h>
79 #include <nfs/nfs_lock.h>
82 #include <sys/dtrace_bsd.h>
84 dtrace_nfsclient_accesscache_flush_probe_func_t
85 dtrace_nfscl_accesscache_flush_done_probe;
86 uint32_t nfscl_accesscache_flush_done_id;
88 dtrace_nfsclient_accesscache_get_probe_func_t
89 dtrace_nfscl_accesscache_get_hit_probe,
90 dtrace_nfscl_accesscache_get_miss_probe;
91 uint32_t nfscl_accesscache_get_hit_id;
92 uint32_t nfscl_accesscache_get_miss_id;
94 dtrace_nfsclient_accesscache_load_probe_func_t
95 dtrace_nfscl_accesscache_load_done_probe;
96 uint32_t nfscl_accesscache_load_done_id;
97 #endif /* !KDTRACE_HOOKS */
103 extern struct nfsstatsv1 nfsstatsv1;
104 extern int nfsrv_useacl;
105 extern int nfscl_debuglevel;
106 MALLOC_DECLARE(M_NEWNFSREQ);
108 static vop_read_t nfsfifo_read;
109 static vop_write_t nfsfifo_write;
110 static vop_close_t nfsfifo_close;
111 static int nfs_setattrrpc(struct vnode *, struct vattr *, struct ucred *,
113 static vop_lookup_t nfs_lookup;
114 static vop_create_t nfs_create;
115 static vop_mknod_t nfs_mknod;
116 static vop_open_t nfs_open;
117 static vop_pathconf_t nfs_pathconf;
118 static vop_close_t nfs_close;
119 static vop_access_t nfs_access;
120 static vop_getattr_t nfs_getattr;
121 static vop_setattr_t nfs_setattr;
122 static vop_read_t nfs_read;
123 static vop_fsync_t nfs_fsync;
124 static vop_remove_t nfs_remove;
125 static vop_link_t nfs_link;
126 static vop_rename_t nfs_rename;
127 static vop_mkdir_t nfs_mkdir;
128 static vop_rmdir_t nfs_rmdir;
129 static vop_symlink_t nfs_symlink;
130 static vop_readdir_t nfs_readdir;
131 static vop_strategy_t nfs_strategy;
132 static int nfs_lookitup(struct vnode *, char *, int,
133 struct ucred *, struct thread *, struct nfsnode **);
134 static int nfs_sillyrename(struct vnode *, struct vnode *,
135 struct componentname *);
136 static vop_access_t nfsspec_access;
137 static vop_readlink_t nfs_readlink;
138 static vop_print_t nfs_print;
139 static vop_advlock_t nfs_advlock;
140 static vop_advlockasync_t nfs_advlockasync;
141 static vop_getacl_t nfs_getacl;
142 static vop_setacl_t nfs_setacl;
145 * Global vfs data structures for nfs
147 struct vop_vector newnfs_vnodeops = {
148 .vop_default = &default_vnodeops,
149 .vop_access = nfs_access,
150 .vop_advlock = nfs_advlock,
151 .vop_advlockasync = nfs_advlockasync,
152 .vop_close = nfs_close,
153 .vop_create = nfs_create,
154 .vop_fsync = nfs_fsync,
155 .vop_getattr = nfs_getattr,
156 .vop_getpages = ncl_getpages,
157 .vop_putpages = ncl_putpages,
158 .vop_inactive = ncl_inactive,
159 .vop_link = nfs_link,
160 .vop_lookup = nfs_lookup,
161 .vop_mkdir = nfs_mkdir,
162 .vop_mknod = nfs_mknod,
163 .vop_open = nfs_open,
164 .vop_pathconf = nfs_pathconf,
165 .vop_print = nfs_print,
166 .vop_read = nfs_read,
167 .vop_readdir = nfs_readdir,
168 .vop_readlink = nfs_readlink,
169 .vop_reclaim = ncl_reclaim,
170 .vop_remove = nfs_remove,
171 .vop_rename = nfs_rename,
172 .vop_rmdir = nfs_rmdir,
173 .vop_setattr = nfs_setattr,
174 .vop_strategy = nfs_strategy,
175 .vop_symlink = nfs_symlink,
176 .vop_write = ncl_write,
177 .vop_getacl = nfs_getacl,
178 .vop_setacl = nfs_setacl,
181 struct vop_vector newnfs_fifoops = {
182 .vop_default = &fifo_specops,
183 .vop_access = nfsspec_access,
184 .vop_close = nfsfifo_close,
185 .vop_fsync = nfs_fsync,
186 .vop_getattr = nfs_getattr,
187 .vop_inactive = ncl_inactive,
188 .vop_print = nfs_print,
189 .vop_read = nfsfifo_read,
190 .vop_reclaim = ncl_reclaim,
191 .vop_setattr = nfs_setattr,
192 .vop_write = nfsfifo_write,
195 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
196 struct componentname *cnp, struct vattr *vap);
197 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
198 int namelen, struct ucred *cred, struct thread *td);
199 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
200 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
201 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
202 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
203 struct componentname *scnp, struct sillyrename *sp);
208 #define DIRHDSIZ (sizeof (struct dirent) - (MAXNAMLEN + 1))
210 SYSCTL_DECL(_vfs_nfs);
212 static int nfsaccess_cache_timeout = NFS_MAXATTRTIMO;
213 SYSCTL_INT(_vfs_nfs, OID_AUTO, access_cache_timeout, CTLFLAG_RW,
214 &nfsaccess_cache_timeout, 0, "NFS ACCESS cache timeout");
216 static int nfs_prime_access_cache = 0;
217 SYSCTL_INT(_vfs_nfs, OID_AUTO, prime_access_cache, CTLFLAG_RW,
218 &nfs_prime_access_cache, 0,
219 "Prime NFS ACCESS cache when fetching attributes");
221 static int newnfs_commit_on_close = 0;
222 SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_on_close, CTLFLAG_RW,
223 &newnfs_commit_on_close, 0, "write+commit on close, else only write");
225 static int nfs_clean_pages_on_close = 1;
226 SYSCTL_INT(_vfs_nfs, OID_AUTO, clean_pages_on_close, CTLFLAG_RW,
227 &nfs_clean_pages_on_close, 0, "NFS clean dirty pages on close");
229 int newnfs_directio_enable = 0;
230 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_enable, CTLFLAG_RW,
231 &newnfs_directio_enable, 0, "Enable NFS directio");
233 int nfs_keep_dirty_on_error;
234 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_keep_dirty_on_error, CTLFLAG_RW,
235 &nfs_keep_dirty_on_error, 0, "Retry pageout if error returned");
238 * This sysctl allows other processes to mmap a file that has been opened
239 * O_DIRECT by a process. In general, having processes mmap the file while
240 * Direct IO is in progress can lead to Data Inconsistencies. But, we allow
241 * this by default to prevent DoS attacks - to prevent a malicious user from
242 * opening up files O_DIRECT preventing other users from mmap'ing these
243 * files. "Protected" environments where stricter consistency guarantees are
244 * required can disable this knob. The process that opened the file O_DIRECT
245 * cannot mmap() the file, because mmap'ed IO on an O_DIRECT open() is not
248 int newnfs_directio_allow_mmap = 1;
249 SYSCTL_INT(_vfs_nfs, OID_AUTO, nfs_directio_allow_mmap, CTLFLAG_RW,
250 &newnfs_directio_allow_mmap, 0, "Enable mmaped IO on file with O_DIRECT opens");
252 #define NFSACCESS_ALL (NFSACCESS_READ | NFSACCESS_MODIFY \
253 | NFSACCESS_EXTEND | NFSACCESS_EXECUTE \
254 | NFSACCESS_DELETE | NFSACCESS_LOOKUP)
258 * The list of locks after the description of the lock is the ordering
259 * of other locks acquired with the lock held.
260 * np->n_mtx : Protects the fields in the nfsnode.
262 VI_MTX (acquired indirectly)
263 * nmp->nm_mtx : Protects the fields in the nfsmount.
265 * ncl_iod_mutex : Global lock, protects shared nfsiod state.
266 * nfs_reqq_mtx : Global lock, protects the nfs_reqq list.
269 * rep->r_mtx : Protects the fields in an nfsreq.
273 nfs34_access_otw(struct vnode *vp, int wmode, struct thread *td,
274 struct ucred *cred, u_int32_t *retmode)
276 int error = 0, attrflag, i, lrupos;
278 struct nfsnode *np = VTONFS(vp);
279 struct nfsvattr nfsva;
281 error = nfsrpc_accessrpc(vp, wmode, cred, td, &nfsva, &attrflag,
284 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
287 mtx_lock(&np->n_mtx);
288 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
289 if (np->n_accesscache[i].uid == cred->cr_uid) {
290 np->n_accesscache[i].mode = rmode;
291 np->n_accesscache[i].stamp = time_second;
294 if (i > 0 && np->n_accesscache[i].stamp <
295 np->n_accesscache[lrupos].stamp)
298 if (i == NFS_ACCESSCACHESIZE) {
299 np->n_accesscache[lrupos].uid = cred->cr_uid;
300 np->n_accesscache[lrupos].mode = rmode;
301 np->n_accesscache[lrupos].stamp = time_second;
303 mtx_unlock(&np->n_mtx);
306 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, rmode, 0);
307 } else if (NFS_ISV4(vp)) {
308 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
312 KDTRACE_NFS_ACCESSCACHE_LOAD_DONE(vp, cred->cr_uid, 0,
319 * nfs access vnode op.
320 * For nfs version 2, just return ok. File accesses may fail later.
321 * For nfs version 3, use the access rpc to check accessibility. If file modes
322 * are changed on the server, accesses might still fail later.
325 nfs_access(struct vop_access_args *ap)
327 struct vnode *vp = ap->a_vp;
328 int error = 0, i, gotahit;
329 u_int32_t mode, wmode, rmode;
330 int v34 = NFS_ISV34(vp);
331 struct nfsnode *np = VTONFS(vp);
334 * Disallow write attempts on filesystems mounted read-only;
335 * unless the file is a socket, fifo, or a block or character
336 * device resident on the filesystem.
338 if ((ap->a_accmode & (VWRITE | VAPPEND | VWRITE_NAMED_ATTRS |
339 VDELETE_CHILD | VWRITE_ATTRIBUTES | VDELETE | VWRITE_ACL |
340 VWRITE_OWNER)) != 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) != 0) {
341 switch (vp->v_type) {
351 * For nfs v3 or v4, check to see if we have done this recently, and if
352 * so return our cached result instead of making an ACCESS call.
353 * If not, do an access rpc, otherwise you are stuck emulating
354 * ufs_access() locally using the vattr. This may not be correct,
355 * since the server may apply other access criteria such as
356 * client uid-->server uid mapping that we do not know about.
359 if (ap->a_accmode & VREAD)
360 mode = NFSACCESS_READ;
363 if (vp->v_type != VDIR) {
364 if (ap->a_accmode & VWRITE)
365 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
366 if (ap->a_accmode & VAPPEND)
367 mode |= NFSACCESS_EXTEND;
368 if (ap->a_accmode & VEXEC)
369 mode |= NFSACCESS_EXECUTE;
370 if (ap->a_accmode & VDELETE)
371 mode |= NFSACCESS_DELETE;
373 if (ap->a_accmode & VWRITE)
374 mode |= (NFSACCESS_MODIFY | NFSACCESS_EXTEND);
375 if (ap->a_accmode & VAPPEND)
376 mode |= NFSACCESS_EXTEND;
377 if (ap->a_accmode & VEXEC)
378 mode |= NFSACCESS_LOOKUP;
379 if (ap->a_accmode & VDELETE)
380 mode |= NFSACCESS_DELETE;
381 if (ap->a_accmode & VDELETE_CHILD)
382 mode |= NFSACCESS_MODIFY;
384 /* XXX safety belt, only make blanket request if caching */
385 if (nfsaccess_cache_timeout > 0) {
386 wmode = NFSACCESS_READ | NFSACCESS_MODIFY |
387 NFSACCESS_EXTEND | NFSACCESS_EXECUTE |
388 NFSACCESS_DELETE | NFSACCESS_LOOKUP;
394 * Does our cached result allow us to give a definite yes to
398 mtx_lock(&np->n_mtx);
399 for (i = 0; i < NFS_ACCESSCACHESIZE; i++) {
400 if (ap->a_cred->cr_uid == np->n_accesscache[i].uid) {
401 if (time_second < (np->n_accesscache[i].stamp
402 + nfsaccess_cache_timeout) &&
403 (np->n_accesscache[i].mode & mode) == mode) {
404 NFSINCRGLOBAL(nfsstatsv1.accesscache_hits);
410 mtx_unlock(&np->n_mtx);
413 KDTRACE_NFS_ACCESSCACHE_GET_HIT(vp,
414 ap->a_cred->cr_uid, mode);
416 KDTRACE_NFS_ACCESSCACHE_GET_MISS(vp,
417 ap->a_cred->cr_uid, mode);
421 * Either a no, or a don't know. Go to the wire.
423 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
424 error = nfs34_access_otw(vp, wmode, ap->a_td,
427 (rmode & mode) != mode)
432 if ((error = nfsspec_access(ap)) != 0) {
436 * Attempt to prevent a mapped root from accessing a file
437 * which it shouldn't. We try to read a byte from the file
438 * if the user is root and the file is not zero length.
439 * After calling nfsspec_access, we should have the correct
442 mtx_lock(&np->n_mtx);
443 if (ap->a_cred->cr_uid == 0 && (ap->a_accmode & VREAD)
444 && VTONFS(vp)->n_size > 0) {
449 mtx_unlock(&np->n_mtx);
452 auio.uio_iov = &aiov;
456 auio.uio_segflg = UIO_SYSSPACE;
457 auio.uio_rw = UIO_READ;
458 auio.uio_td = ap->a_td;
460 if (vp->v_type == VREG)
461 error = ncl_readrpc(vp, &auio, ap->a_cred);
462 else if (vp->v_type == VDIR) {
464 bp = malloc(NFS_DIRBLKSIZ, M_TEMP, M_WAITOK);
466 aiov.iov_len = auio.uio_resid = NFS_DIRBLKSIZ;
467 error = ncl_readdirrpc(vp, &auio, ap->a_cred,
470 } else if (vp->v_type == VLNK)
471 error = ncl_readlinkrpc(vp, &auio, ap->a_cred);
475 mtx_unlock(&np->n_mtx);
483 * Check to see if the type is ok
484 * and that deletion is not in progress.
485 * For paged in text files, you will need to flush the page cache
486 * if consistency is lost.
490 nfs_open(struct vop_open_args *ap)
492 struct vnode *vp = ap->a_vp;
493 struct nfsnode *np = VTONFS(vp);
496 int fmode = ap->a_mode;
499 if (vp->v_type != VREG && vp->v_type != VDIR && vp->v_type != VLNK)
503 * For NFSv4, we need to do the Open Op before cache validation,
504 * so that we conform to RFC3530 Sec. 9.3.1.
507 error = nfsrpc_open(vp, fmode, ap->a_cred, ap->a_td);
509 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
516 * Now, if this Open will be doing reading, re-validate/flush the
517 * cache, so that Close/Open coherency is maintained.
519 mtx_lock(&np->n_mtx);
520 if (np->n_flag & NMODIFIED) {
521 mtx_unlock(&np->n_mtx);
522 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
523 if (error == EINTR || error == EIO) {
525 (void) nfsrpc_close(vp, 0, ap->a_td);
528 mtx_lock(&np->n_mtx);
530 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
531 if (vp->v_type == VDIR)
532 np->n_direofoffset = 0;
533 mtx_unlock(&np->n_mtx);
534 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
537 (void) nfsrpc_close(vp, 0, ap->a_td);
540 mtx_lock(&np->n_mtx);
541 np->n_mtime = vattr.va_mtime;
543 np->n_change = vattr.va_filerev;
545 mtx_unlock(&np->n_mtx);
546 error = VOP_GETATTR(vp, &vattr, ap->a_cred);
549 (void) nfsrpc_close(vp, 0, ap->a_td);
552 mtx_lock(&np->n_mtx);
553 if ((NFS_ISV4(vp) && np->n_change != vattr.va_filerev) ||
554 NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
555 if (vp->v_type == VDIR)
556 np->n_direofoffset = 0;
557 mtx_unlock(&np->n_mtx);
558 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
559 if (error == EINTR || error == EIO) {
561 (void) nfsrpc_close(vp, 0, ap->a_td);
564 mtx_lock(&np->n_mtx);
565 np->n_mtime = vattr.va_mtime;
567 np->n_change = vattr.va_filerev;
572 * If the object has >= 1 O_DIRECT active opens, we disable caching.
574 if (newnfs_directio_enable && (fmode & O_DIRECT) &&
575 (vp->v_type == VREG)) {
576 if (np->n_directio_opens == 0) {
577 mtx_unlock(&np->n_mtx);
578 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
581 (void) nfsrpc_close(vp, 0, ap->a_td);
584 mtx_lock(&np->n_mtx);
585 np->n_flag |= NNONCACHE;
587 np->n_directio_opens++;
590 /* If opened for writing via NFSv4.1 or later, mark that for pNFS. */
591 if (NFSHASPNFS(VFSTONFS(vp->v_mount)) && (fmode & FWRITE) != 0)
592 np->n_flag |= NWRITEOPENED;
595 * If this is an open for writing, capture a reference to the
596 * credentials, so they can be used by ncl_putpages(). Using
597 * these write credentials is preferable to the credentials of
598 * whatever thread happens to be doing the VOP_PUTPAGES() since
599 * the write RPCs are less likely to fail with EACCES.
601 if ((fmode & FWRITE) != 0) {
602 cred = np->n_writecred;
603 np->n_writecred = crhold(ap->a_cred);
606 mtx_unlock(&np->n_mtx);
610 vnode_create_vobject(vp, vattr.va_size, ap->a_td);
616 * What an NFS client should do upon close after writing is a debatable issue.
617 * Most NFS clients push delayed writes to the server upon close, basically for
619 * 1 - So that any write errors may be reported back to the client process
620 * doing the close system call. By far the two most likely errors are
621 * NFSERR_NOSPC and NFSERR_DQUOT to indicate space allocation failure.
622 * 2 - To put a worst case upper bound on cache inconsistency between
623 * multiple clients for the file.
624 * There is also a consistency problem for Version 2 of the protocol w.r.t.
625 * not being able to tell if other clients are writing a file concurrently,
626 * since there is no way of knowing if the changed modify time in the reply
627 * is only due to the write for this client.
628 * (NFS Version 3 provides weak cache consistency data in the reply that
629 * should be sufficient to detect and handle this case.)
631 * The current code does the following:
632 * for NFS Version 2 - play it safe and flush/invalidate all dirty buffers
633 * for NFS Version 3 - flush dirty buffers to the server but don't invalidate
634 * or commit them (this satisfies 1 and 2 except for the
635 * case where the server crashes after this close but
636 * before the commit RPC, which is felt to be "good
637 * enough". Changing the last argument to ncl_flush() to
638 * a 1 would force a commit operation, if it is felt a
639 * commit is necessary now.
640 * for NFS Version 4 - flush the dirty buffers and commit them, if
641 * nfscl_mustflush() says this is necessary.
642 * It is necessary if there is no write delegation held,
643 * in order to satisfy open/close coherency.
644 * If the file isn't cached on local stable storage,
645 * it may be necessary in order to detect "out of space"
646 * errors from the server, if the write delegation
647 * issued by the server doesn't allow the file to grow.
651 nfs_close(struct vop_close_args *ap)
653 struct vnode *vp = ap->a_vp;
654 struct nfsnode *np = VTONFS(vp);
655 struct nfsvattr nfsva;
657 int error = 0, ret, localcred = 0;
658 int fmode = ap->a_fflag;
660 if ((vp->v_mount->mnt_kern_flag & MNTK_UNMOUNTF))
663 * During shutdown, a_cred isn't valid, so just use root.
665 if (ap->a_cred == NOCRED) {
666 cred = newnfs_getcred();
671 if (vp->v_type == VREG) {
673 * Examine and clean dirty pages, regardless of NMODIFIED.
674 * This closes a major hole in close-to-open consistency.
675 * We want to push out all dirty pages (and buffers) on
676 * close, regardless of whether they were dirtied by
677 * mmap'ed writes or via write().
679 if (nfs_clean_pages_on_close && vp->v_object) {
680 VM_OBJECT_WLOCK(vp->v_object);
681 vm_object_page_clean(vp->v_object, 0, 0, 0);
682 VM_OBJECT_WUNLOCK(vp->v_object);
684 mtx_lock(&np->n_mtx);
685 if (np->n_flag & NMODIFIED) {
686 mtx_unlock(&np->n_mtx);
689 * Under NFSv3 we have dirty buffers to dispose of. We
690 * must flush them to the NFS server. We have the option
691 * of waiting all the way through the commit rpc or just
692 * waiting for the initial write. The default is to only
693 * wait through the initial write so the data is in the
694 * server's cache, which is roughly similar to the state
695 * a standard disk subsystem leaves the file in on close().
697 * We cannot clear the NMODIFIED bit in np->n_flag due to
698 * potential races with other processes, and certainly
699 * cannot clear it if we don't commit.
700 * These races occur when there is no longer the old
701 * traditional vnode locking implemented for Vnode Ops.
703 int cm = newnfs_commit_on_close ? 1 : 0;
704 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td, cm, 0);
705 /* np->n_flag &= ~NMODIFIED; */
706 } else if (NFS_ISV4(vp)) {
707 if (nfscl_mustflush(vp) != 0) {
708 int cm = newnfs_commit_on_close ? 1 : 0;
709 error = ncl_flush(vp, MNT_WAIT, cred, ap->a_td,
712 * as above w.r.t races when clearing
714 * np->n_flag &= ~NMODIFIED;
718 error = ncl_vinvalbuf(vp, V_SAVE, ap->a_td, 1);
719 mtx_lock(&np->n_mtx);
722 * Invalidate the attribute cache in all cases.
723 * An open is going to fetch fresh attrs any way, other procs
724 * on this node that have file open will be forced to do an
725 * otw attr fetch, but this is safe.
726 * --> A user found that their RPC count dropped by 20% when
727 * this was commented out and I can't see any requirement
728 * for it, so I've disabled it when negative lookups are
729 * enabled. (What does this have to do with negative lookup
730 * caching? Well nothing, except it was reported by the
731 * same user that needed negative lookup caching and I wanted
732 * there to be a way to disable it to see if it
733 * is the cause of some caching/coherency issue that might
736 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
738 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
740 if (np->n_flag & NWRITEERR) {
741 np->n_flag &= ~NWRITEERR;
744 mtx_unlock(&np->n_mtx);
749 * Get attributes so "change" is up to date.
751 if (error == 0 && nfscl_mustflush(vp) != 0 &&
752 vp->v_type == VREG &&
753 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
754 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
757 np->n_change = nfsva.na_filerev;
758 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
766 ret = nfsrpc_close(vp, 0, ap->a_td);
770 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
773 if (newnfs_directio_enable)
774 KASSERT((np->n_directio_asyncwr == 0),
775 ("nfs_close: dirty unflushed (%d) directio buffers\n",
776 np->n_directio_asyncwr));
777 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
778 mtx_lock(&np->n_mtx);
779 KASSERT((np->n_directio_opens > 0),
780 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
781 np->n_directio_opens--;
782 if (np->n_directio_opens == 0)
783 np->n_flag &= ~NNONCACHE;
784 mtx_unlock(&np->n_mtx);
792 * nfs getattr call from vfs.
795 nfs_getattr(struct vop_getattr_args *ap)
797 struct vnode *vp = ap->a_vp;
798 struct thread *td = curthread; /* XXX */
799 struct nfsnode *np = VTONFS(vp);
801 struct nfsvattr nfsva;
802 struct vattr *vap = ap->a_vap;
806 * Update local times for special files.
808 mtx_lock(&np->n_mtx);
809 if (np->n_flag & (NACC | NUPD))
811 mtx_unlock(&np->n_mtx);
813 * First look in the cache.
815 if (ncl_getattrcache(vp, &vattr) == 0) {
816 vap->va_type = vattr.va_type;
817 vap->va_mode = vattr.va_mode;
818 vap->va_nlink = vattr.va_nlink;
819 vap->va_uid = vattr.va_uid;
820 vap->va_gid = vattr.va_gid;
821 vap->va_fsid = vattr.va_fsid;
822 vap->va_fileid = vattr.va_fileid;
823 vap->va_size = vattr.va_size;
824 vap->va_blocksize = vattr.va_blocksize;
825 vap->va_atime = vattr.va_atime;
826 vap->va_mtime = vattr.va_mtime;
827 vap->va_ctime = vattr.va_ctime;
828 vap->va_gen = vattr.va_gen;
829 vap->va_flags = vattr.va_flags;
830 vap->va_rdev = vattr.va_rdev;
831 vap->va_bytes = vattr.va_bytes;
832 vap->va_filerev = vattr.va_filerev;
834 * Get the local modify time for the case of a write
837 nfscl_deleggetmodtime(vp, &vap->va_mtime);
841 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
842 nfsaccess_cache_timeout > 0) {
843 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
844 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
845 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
846 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
850 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
852 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
855 * Get the local modify time for the case of a write
858 nfscl_deleggetmodtime(vp, &vap->va_mtime);
859 } else if (NFS_ISV4(vp)) {
860 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
869 nfs_setattr(struct vop_setattr_args *ap)
871 struct vnode *vp = ap->a_vp;
872 struct nfsnode *np = VTONFS(vp);
873 struct thread *td = curthread; /* XXX */
874 struct vattr *vap = ap->a_vap;
883 * Setting of flags and marking of atimes are not supported.
885 if (vap->va_flags != VNOVAL)
889 * Disallow write attempts if the filesystem is mounted read-only.
891 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
892 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
893 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
894 (vp->v_mount->mnt_flag & MNT_RDONLY))
896 if (vap->va_size != VNOVAL) {
897 switch (vp->v_type) {
904 if (vap->va_mtime.tv_sec == VNOVAL &&
905 vap->va_atime.tv_sec == VNOVAL &&
906 vap->va_mode == (mode_t)VNOVAL &&
907 vap->va_uid == (uid_t)VNOVAL &&
908 vap->va_gid == (gid_t)VNOVAL)
910 vap->va_size = VNOVAL;
914 * Disallow write attempts if the filesystem is
917 if (vp->v_mount->mnt_flag & MNT_RDONLY)
920 * We run vnode_pager_setsize() early (why?),
921 * we must set np->n_size now to avoid vinvalbuf
922 * V_SAVE races that might setsize a lower
925 mtx_lock(&np->n_mtx);
927 mtx_unlock(&np->n_mtx);
928 error = ncl_meta_setsize(vp, ap->a_cred, td,
930 mtx_lock(&np->n_mtx);
931 if (np->n_flag & NMODIFIED) {
933 mtx_unlock(&np->n_mtx);
934 if (vap->va_size == 0)
935 error = ncl_vinvalbuf(vp, 0, td, 1);
937 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
939 vnode_pager_setsize(vp, tsize);
943 * Call nfscl_delegmodtime() to set the modify time
944 * locally, as required.
946 nfscl_delegmodtime(vp);
948 mtx_unlock(&np->n_mtx);
950 * np->n_size has already been set to vap->va_size
951 * in ncl_meta_setsize(). We must set it again since
952 * nfs_loadattrcache() could be called through
953 * ncl_meta_setsize() and could modify np->n_size.
955 mtx_lock(&np->n_mtx);
956 np->n_vattr.na_size = np->n_size = vap->va_size;
957 mtx_unlock(&np->n_mtx);
960 mtx_lock(&np->n_mtx);
961 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
962 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
963 mtx_unlock(&np->n_mtx);
964 if ((error = ncl_vinvalbuf(vp, V_SAVE, td, 1)) != 0 &&
965 (error == EINTR || error == EIO))
968 mtx_unlock(&np->n_mtx);
970 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
971 if (error && vap->va_size != VNOVAL) {
972 mtx_lock(&np->n_mtx);
973 np->n_size = np->n_vattr.na_size = tsize;
974 vnode_pager_setsize(vp, tsize);
975 mtx_unlock(&np->n_mtx);
981 * Do an nfs setattr rpc.
984 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
987 struct nfsnode *np = VTONFS(vp);
988 int error, ret, attrflag, i;
989 struct nfsvattr nfsva;
992 mtx_lock(&np->n_mtx);
993 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
994 np->n_accesscache[i].stamp = 0;
995 np->n_flag |= NDELEGMOD;
996 mtx_unlock(&np->n_mtx);
997 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
999 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1002 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1006 if (error && NFS_ISV4(vp))
1007 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1012 * nfs lookup call, one step at a time...
1013 * First look in cache
1014 * If not found, unlock the directory nfsnode and do the rpc
1017 nfs_lookup(struct vop_lookup_args *ap)
1019 struct componentname *cnp = ap->a_cnp;
1020 struct vnode *dvp = ap->a_dvp;
1021 struct vnode **vpp = ap->a_vpp;
1022 struct mount *mp = dvp->v_mount;
1023 int flags = cnp->cn_flags;
1024 struct vnode *newvp;
1025 struct nfsmount *nmp;
1026 struct nfsnode *np, *newnp;
1027 int error = 0, attrflag, dattrflag, ltype, ncticks;
1028 struct thread *td = cnp->cn_thread;
1030 struct nfsvattr dnfsva, nfsva;
1032 struct timespec nctime;
1035 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1036 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1038 if (dvp->v_type != VDIR)
1043 /* For NFSv4, wait until any remove is done. */
1044 mtx_lock(&np->n_mtx);
1045 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1046 np->n_flag |= NREMOVEWANT;
1047 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1049 mtx_unlock(&np->n_mtx);
1051 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1053 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1054 if (error > 0 && error != ENOENT)
1058 * Lookups of "." are special and always return the
1059 * current directory. cache_lookup() already handles
1060 * associated locking bookkeeping, etc.
1062 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1063 /* XXX: Is this really correct? */
1064 if (cnp->cn_nameiop != LOOKUP &&
1066 cnp->cn_flags |= SAVENAME;
1071 * We only accept a positive hit in the cache if the
1072 * change time of the file matches our cached copy.
1073 * Otherwise, we discard the cache entry and fallback
1074 * to doing a lookup RPC. We also only trust cache
1075 * entries for less than nm_nametimeo seconds.
1077 * To better handle stale file handles and attributes,
1078 * clear the attribute cache of this node if it is a
1079 * leaf component, part of an open() call, and not
1080 * locally modified before fetching the attributes.
1081 * This should allow stale file handles to be detected
1082 * here where we can fall back to a LOOKUP RPC to
1083 * recover rather than having nfs_open() detect the
1084 * stale file handle and failing open(2) with ESTALE.
1087 newnp = VTONFS(newvp);
1088 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1089 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1090 !(newnp->n_flag & NMODIFIED)) {
1091 mtx_lock(&newnp->n_mtx);
1092 newnp->n_attrstamp = 0;
1093 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1094 mtx_unlock(&newnp->n_mtx);
1096 if (nfscl_nodeleg(newvp, 0) == 0 ||
1097 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1098 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1099 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1100 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1101 if (cnp->cn_nameiop != LOOKUP &&
1103 cnp->cn_flags |= SAVENAME;
1112 } else if (error == ENOENT) {
1113 if (dvp->v_iflag & VI_DOOMED)
1116 * We only accept a negative hit in the cache if the
1117 * modification time of the parent directory matches
1118 * the cached copy in the name cache entry.
1119 * Otherwise, we discard all of the negative cache
1120 * entries for this directory. We also only trust
1121 * negative cache entries for up to nm_negnametimeo
1124 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1125 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1126 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1127 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1130 cache_purge_negative(dvp);
1135 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1136 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1137 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1140 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1142 if (newvp != NULLVP) {
1147 if (error != ENOENT) {
1149 error = nfscl_maperr(td, error, (uid_t)0,
1154 /* The requested file was not found. */
1155 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1156 (flags & ISLASTCN)) {
1158 * XXX: UFS does a full VOP_ACCESS(dvp,
1159 * VWRITE) here instead of just checking
1162 if (mp->mnt_flag & MNT_RDONLY)
1164 cnp->cn_flags |= SAVENAME;
1165 return (EJUSTRETURN);
1168 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1170 * Cache the modification time of the parent
1171 * directory from the post-op attributes in
1172 * the name cache entry. The negative cache
1173 * entry will be ignored once the directory
1174 * has changed. Don't bother adding the entry
1175 * if the directory has already changed.
1177 mtx_lock(&np->n_mtx);
1178 if (timespeccmp(&np->n_vattr.na_mtime,
1179 &dnfsva.na_mtime, ==)) {
1180 mtx_unlock(&np->n_mtx);
1181 cache_enter_time(dvp, NULL, cnp,
1182 &dnfsva.na_mtime, NULL);
1184 mtx_unlock(&np->n_mtx);
1190 * Handle RENAME case...
1192 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1193 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1194 FREE((caddr_t)nfhp, M_NFSFH);
1197 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1203 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1206 cnp->cn_flags |= SAVENAME;
1210 if (flags & ISDOTDOT) {
1211 ltype = NFSVOPISLOCKED(dvp);
1212 error = vfs_busy(mp, MBF_NOWAIT);
1215 NFSVOPUNLOCK(dvp, 0);
1216 error = vfs_busy(mp, 0);
1217 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1219 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1226 NFSVOPUNLOCK(dvp, 0);
1227 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1233 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1234 if (dvp->v_iflag & VI_DOOMED) {
1246 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1248 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1249 FREE((caddr_t)nfhp, M_NFSFH);
1253 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1256 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1262 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1264 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1265 !(np->n_flag & NMODIFIED)) {
1267 * Flush the attribute cache when opening a
1268 * leaf node to ensure that fresh attributes
1269 * are fetched in nfs_open() since we did not
1270 * fetch attributes from the LOOKUP reply.
1272 mtx_lock(&np->n_mtx);
1273 np->n_attrstamp = 0;
1274 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1275 mtx_unlock(&np->n_mtx);
1278 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1279 cnp->cn_flags |= SAVENAME;
1280 if ((cnp->cn_flags & MAKEENTRY) &&
1281 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1282 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1283 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1284 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1291 * Just call ncl_bioread() to do the work.
1294 nfs_read(struct vop_read_args *ap)
1296 struct vnode *vp = ap->a_vp;
1298 switch (vp->v_type) {
1300 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1304 return (EOPNOTSUPP);
1312 nfs_readlink(struct vop_readlink_args *ap)
1314 struct vnode *vp = ap->a_vp;
1316 if (vp->v_type != VLNK)
1318 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1322 * Do a readlink rpc.
1323 * Called by ncl_doio() from below the buffer cache.
1326 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1328 int error, ret, attrflag;
1329 struct nfsvattr nfsva;
1331 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1334 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1338 if (error && NFS_ISV4(vp))
1339 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1348 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1350 int error, ret, attrflag;
1351 struct nfsvattr nfsva;
1352 struct nfsmount *nmp;
1354 nmp = VFSTONFS(vnode_mount(vp));
1357 if (NFSHASPNFS(nmp))
1358 error = nfscl_doiods(vp, uiop, NULL, NULL,
1359 NFSV4OPEN_ACCESSREAD, cred, uiop->uio_td);
1360 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1362 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1365 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1369 if (error && NFS_ISV4(vp))
1370 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1378 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1379 int *iomode, int *must_commit, int called_from_strategy)
1381 struct nfsvattr nfsva;
1382 int error, attrflag, ret;
1383 struct nfsmount *nmp;
1385 nmp = VFSTONFS(vnode_mount(vp));
1388 if (NFSHASPNFS(nmp))
1389 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1390 NFSV4OPEN_ACCESSWRITE, cred, uiop->uio_td);
1391 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1393 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1394 uiop->uio_td, &nfsva, &attrflag, NULL,
1395 called_from_strategy);
1397 if (VTONFS(vp)->n_flag & ND_NFSV4)
1398 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1401 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1407 *iomode = NFSWRITE_FILESYNC;
1408 if (error && NFS_ISV4(vp))
1409 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1415 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1416 * mode set to specify the file type and the size field for rdev.
1419 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1422 struct nfsvattr nfsva, dnfsva;
1423 struct vnode *newvp = NULL;
1424 struct nfsnode *np = NULL, *dnp;
1427 int error = 0, attrflag, dattrflag;
1430 if (vap->va_type == VCHR || vap->va_type == VBLK)
1431 rdev = vap->va_rdev;
1432 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1435 return (EOPNOTSUPP);
1436 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1438 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1439 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1440 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1443 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1444 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1445 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1448 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1449 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1452 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1455 if (attrflag != 0) {
1456 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1464 } else if (NFS_ISV4(dvp)) {
1465 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1469 mtx_lock(&dnp->n_mtx);
1470 dnp->n_flag |= NMODIFIED;
1472 dnp->n_attrstamp = 0;
1473 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1475 mtx_unlock(&dnp->n_mtx);
1481 * just call nfs_mknodrpc() to do the work.
1485 nfs_mknod(struct vop_mknod_args *ap)
1487 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1490 static struct mtx nfs_cverf_mtx;
1491 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1497 static nfsquad_t cverf;
1499 static int cverf_initialized = 0;
1501 mtx_lock(&nfs_cverf_mtx);
1502 if (cverf_initialized == 0) {
1503 cverf.lval[0] = arc4random();
1504 cverf.lval[1] = arc4random();
1505 cverf_initialized = 1;
1509 mtx_unlock(&nfs_cverf_mtx);
1515 * nfs file create call
1518 nfs_create(struct vop_create_args *ap)
1520 struct vnode *dvp = ap->a_dvp;
1521 struct vattr *vap = ap->a_vap;
1522 struct componentname *cnp = ap->a_cnp;
1523 struct nfsnode *np = NULL, *dnp;
1524 struct vnode *newvp = NULL;
1525 struct nfsmount *nmp;
1526 struct nfsvattr dnfsva, nfsva;
1529 int error = 0, attrflag, dattrflag, fmode = 0;
1533 * Oops, not for me..
1535 if (vap->va_type == VSOCK)
1536 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1538 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1540 if (vap->va_vaflags & VA_EXCLUSIVE)
1543 nmp = VFSTONFS(vnode_mount(dvp));
1545 /* For NFSv4, wait until any remove is done. */
1546 mtx_lock(&dnp->n_mtx);
1547 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1548 dnp->n_flag |= NREMOVEWANT;
1549 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1551 mtx_unlock(&dnp->n_mtx);
1553 cverf = nfs_get_cverf();
1554 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1555 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1556 &nfhp, &attrflag, &dattrflag, NULL);
1559 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1560 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1561 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1564 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1565 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1568 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1572 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1573 cnp->cn_thread, &nfsva, NULL);
1575 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1579 if (newvp != NULL) {
1583 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1584 error == NFSERR_NOTSUPP) {
1588 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1589 if (nfscl_checksattr(vap, &nfsva)) {
1590 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1591 cnp->cn_thread, &nfsva, &attrflag, NULL);
1592 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1593 vap->va_gid != (gid_t)VNOVAL)) {
1594 /* try again without setting uid/gid */
1595 vap->va_uid = (uid_t)VNOVAL;
1596 vap->va_gid = (uid_t)VNOVAL;
1597 error = nfsrpc_setattr(newvp, vap, NULL,
1598 cnp->cn_cred, cnp->cn_thread, &nfsva,
1602 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1609 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1610 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1613 } else if (NFS_ISV4(dvp)) {
1614 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1617 mtx_lock(&dnp->n_mtx);
1618 dnp->n_flag |= NMODIFIED;
1620 dnp->n_attrstamp = 0;
1621 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1623 mtx_unlock(&dnp->n_mtx);
1628 * nfs file remove call
1629 * To try and make nfs semantics closer to ufs semantics, a file that has
1630 * other processes using the vnode is renamed instead of removed and then
1631 * removed later on the last close.
1632 * - If v_usecount > 1
1633 * If a rename is not already in the works
1634 * call nfs_sillyrename() to set it up
1639 nfs_remove(struct vop_remove_args *ap)
1641 struct vnode *vp = ap->a_vp;
1642 struct vnode *dvp = ap->a_dvp;
1643 struct componentname *cnp = ap->a_cnp;
1644 struct nfsnode *np = VTONFS(vp);
1648 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1649 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1650 if (vp->v_type == VDIR)
1652 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1653 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1654 vattr.va_nlink > 1)) {
1656 * Purge the name cache so that the chance of a lookup for
1657 * the name succeeding while the remove is in progress is
1658 * minimized. Without node locking it can still happen, such
1659 * that an I/O op returns ESTALE, but since you get this if
1660 * another host removes the file..
1664 * throw away biocache buffers, mainly to avoid
1665 * unnecessary delayed writes later.
1667 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1669 if (error != EINTR && error != EIO)
1670 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1671 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1673 * Kludge City: If the first reply to the remove rpc is lost..
1674 * the reply to the retransmitted request will be ENOENT
1675 * since the file was in fact removed
1676 * Therefore, we cheat and return success.
1678 if (error == ENOENT)
1680 } else if (!np->n_sillyrename)
1681 error = nfs_sillyrename(dvp, vp, cnp);
1682 mtx_lock(&np->n_mtx);
1683 np->n_attrstamp = 0;
1684 mtx_unlock(&np->n_mtx);
1685 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1690 * nfs file remove rpc called from nfs_inactive
1693 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1696 * Make sure that the directory vnode is still valid.
1697 * XXX we should lock sp->s_dvp here.
1699 if (sp->s_dvp->v_type == VBAD)
1701 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1706 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1709 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1710 int namelen, struct ucred *cred, struct thread *td)
1712 struct nfsvattr dnfsva;
1713 struct nfsnode *dnp = VTONFS(dvp);
1714 int error = 0, dattrflag;
1716 mtx_lock(&dnp->n_mtx);
1717 dnp->n_flag |= NREMOVEINPROG;
1718 mtx_unlock(&dnp->n_mtx);
1719 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1721 mtx_lock(&dnp->n_mtx);
1722 if ((dnp->n_flag & NREMOVEWANT)) {
1723 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1724 mtx_unlock(&dnp->n_mtx);
1725 wakeup((caddr_t)dnp);
1727 dnp->n_flag &= ~NREMOVEINPROG;
1728 mtx_unlock(&dnp->n_mtx);
1731 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1732 mtx_lock(&dnp->n_mtx);
1733 dnp->n_flag |= NMODIFIED;
1735 dnp->n_attrstamp = 0;
1736 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1738 mtx_unlock(&dnp->n_mtx);
1739 if (error && NFS_ISV4(dvp))
1740 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1745 * nfs file rename call
1748 nfs_rename(struct vop_rename_args *ap)
1750 struct vnode *fvp = ap->a_fvp;
1751 struct vnode *tvp = ap->a_tvp;
1752 struct vnode *fdvp = ap->a_fdvp;
1753 struct vnode *tdvp = ap->a_tdvp;
1754 struct componentname *tcnp = ap->a_tcnp;
1755 struct componentname *fcnp = ap->a_fcnp;
1756 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1757 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1758 struct nfsv4node *newv4 = NULL;
1761 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1762 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1763 /* Check for cross-device rename */
1764 if ((fvp->v_mount != tdvp->v_mount) ||
1765 (tvp && (fvp->v_mount != tvp->v_mount))) {
1771 printf("nfs_rename: fvp == tvp (can't happen)\n");
1775 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1779 * We have to flush B_DELWRI data prior to renaming
1780 * the file. If we don't, the delayed-write buffers
1781 * can be flushed out later after the file has gone stale
1782 * under NFSV3. NFSV2 does not have this problem because
1783 * ( as far as I can tell ) it flushes dirty buffers more
1786 * Skip the rename operation if the fsync fails, this can happen
1787 * due to the server's volume being full, when we pushed out data
1788 * that was written back to our cache earlier. Not checking for
1789 * this condition can result in potential (silent) data loss.
1791 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1792 NFSVOPUNLOCK(fvp, 0);
1794 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1799 * If the tvp exists and is in use, sillyrename it before doing the
1800 * rename of the new file over it.
1801 * XXX Can't sillyrename a directory.
1803 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1804 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1809 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1810 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1813 if (error == 0 && NFS_ISV4(tdvp)) {
1815 * For NFSv4, check to see if it is the same name and
1816 * replace the name, if it is different.
1818 MALLOC(newv4, struct nfsv4node *,
1819 sizeof (struct nfsv4node) +
1820 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1821 M_NFSV4NODE, M_WAITOK);
1822 mtx_lock(&tdnp->n_mtx);
1823 mtx_lock(&fnp->n_mtx);
1824 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1825 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1826 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1827 tcnp->cn_namelen) ||
1828 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1829 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1830 tdnp->n_fhp->nfh_len))) {
1832 { char nnn[100]; int nnnl;
1833 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1834 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1836 printf("ren replace=%s\n",nnn);
1839 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1842 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1843 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1844 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1845 tdnp->n_fhp->nfh_len);
1846 NFSBCOPY(tcnp->cn_nameptr,
1847 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1849 mtx_unlock(&tdnp->n_mtx);
1850 mtx_unlock(&fnp->n_mtx);
1852 FREE((caddr_t)newv4, M_NFSV4NODE);
1855 if (fvp->v_type == VDIR) {
1856 if (tvp != NULL && tvp->v_type == VDIR)
1871 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1873 if (error == ENOENT)
1879 * nfs file rename rpc called from nfs_remove() above
1882 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1883 struct sillyrename *sp)
1886 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1887 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1892 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1895 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1896 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1897 int tnamelen, struct ucred *cred, struct thread *td)
1899 struct nfsvattr fnfsva, tnfsva;
1900 struct nfsnode *fdnp = VTONFS(fdvp);
1901 struct nfsnode *tdnp = VTONFS(tdvp);
1902 int error = 0, fattrflag, tattrflag;
1904 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1905 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1906 &tattrflag, NULL, NULL);
1907 mtx_lock(&fdnp->n_mtx);
1908 fdnp->n_flag |= NMODIFIED;
1909 if (fattrflag != 0) {
1910 mtx_unlock(&fdnp->n_mtx);
1911 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1913 fdnp->n_attrstamp = 0;
1914 mtx_unlock(&fdnp->n_mtx);
1915 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1917 mtx_lock(&tdnp->n_mtx);
1918 tdnp->n_flag |= NMODIFIED;
1919 if (tattrflag != 0) {
1920 mtx_unlock(&tdnp->n_mtx);
1921 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1923 tdnp->n_attrstamp = 0;
1924 mtx_unlock(&tdnp->n_mtx);
1925 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1927 if (error && NFS_ISV4(fdvp))
1928 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1933 * nfs hard link create call
1936 nfs_link(struct vop_link_args *ap)
1938 struct vnode *vp = ap->a_vp;
1939 struct vnode *tdvp = ap->a_tdvp;
1940 struct componentname *cnp = ap->a_cnp;
1941 struct nfsnode *np, *tdnp;
1942 struct nfsvattr nfsva, dnfsva;
1943 int error = 0, attrflag, dattrflag;
1946 * Push all writes to the server, so that the attribute cache
1947 * doesn't get "out of sync" with the server.
1948 * XXX There should be a better way!
1950 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1952 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1953 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1955 tdnp = VTONFS(tdvp);
1956 mtx_lock(&tdnp->n_mtx);
1957 tdnp->n_flag |= NMODIFIED;
1958 if (dattrflag != 0) {
1959 mtx_unlock(&tdnp->n_mtx);
1960 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1962 tdnp->n_attrstamp = 0;
1963 mtx_unlock(&tdnp->n_mtx);
1964 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1967 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1970 mtx_lock(&np->n_mtx);
1971 np->n_attrstamp = 0;
1972 mtx_unlock(&np->n_mtx);
1973 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1976 * If negative lookup caching is enabled, I might as well
1977 * add an entry for this node. Not necessary for correctness,
1978 * but if negative caching is enabled, then the system
1979 * must care about lookup caching hit rate, so...
1981 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1982 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1983 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1985 if (error && NFS_ISV4(vp))
1986 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1992 * nfs symbolic link create call
1995 nfs_symlink(struct vop_symlink_args *ap)
1997 struct vnode *dvp = ap->a_dvp;
1998 struct vattr *vap = ap->a_vap;
1999 struct componentname *cnp = ap->a_cnp;
2000 struct nfsvattr nfsva, dnfsva;
2002 struct nfsnode *np = NULL, *dnp;
2003 struct vnode *newvp = NULL;
2004 int error = 0, attrflag, dattrflag, ret;
2006 vap->va_type = VLNK;
2007 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2008 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2009 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2011 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2012 &np, NULL, LK_EXCLUSIVE);
2018 if (newvp != NULL) {
2020 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2022 } else if (!error) {
2024 * If we do not have an error and we could not extract the
2025 * newvp from the response due to the request being NFSv2, we
2026 * have to do a lookup in order to obtain a newvp to return.
2028 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2029 cnp->cn_cred, cnp->cn_thread, &np);
2037 error = nfscl_maperr(cnp->cn_thread, error,
2038 vap->va_uid, vap->va_gid);
2044 mtx_lock(&dnp->n_mtx);
2045 dnp->n_flag |= NMODIFIED;
2046 if (dattrflag != 0) {
2047 mtx_unlock(&dnp->n_mtx);
2048 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2050 dnp->n_attrstamp = 0;
2051 mtx_unlock(&dnp->n_mtx);
2052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2055 * If negative lookup caching is enabled, I might as well
2056 * add an entry for this node. Not necessary for correctness,
2057 * but if negative caching is enabled, then the system
2058 * must care about lookup caching hit rate, so...
2060 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2061 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2062 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2071 nfs_mkdir(struct vop_mkdir_args *ap)
2073 struct vnode *dvp = ap->a_dvp;
2074 struct vattr *vap = ap->a_vap;
2075 struct componentname *cnp = ap->a_cnp;
2076 struct nfsnode *np = NULL, *dnp;
2077 struct vnode *newvp = NULL;
2080 struct nfsvattr nfsva, dnfsva;
2081 int error = 0, attrflag, dattrflag, ret;
2083 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2085 vap->va_type = VDIR;
2086 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2087 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2088 &attrflag, &dattrflag, NULL);
2090 mtx_lock(&dnp->n_mtx);
2091 dnp->n_flag |= NMODIFIED;
2092 if (dattrflag != 0) {
2093 mtx_unlock(&dnp->n_mtx);
2094 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2096 dnp->n_attrstamp = 0;
2097 mtx_unlock(&dnp->n_mtx);
2098 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2101 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2102 &np, NULL, LK_EXCLUSIVE);
2106 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2111 if (!error && newvp == NULL) {
2112 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2113 cnp->cn_cred, cnp->cn_thread, &np);
2116 if (newvp->v_type != VDIR)
2124 error = nfscl_maperr(cnp->cn_thread, error,
2125 vap->va_uid, vap->va_gid);
2128 * If negative lookup caching is enabled, I might as well
2129 * add an entry for this node. Not necessary for correctness,
2130 * but if negative caching is enabled, then the system
2131 * must care about lookup caching hit rate, so...
2133 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2134 (cnp->cn_flags & MAKEENTRY) &&
2135 attrflag != 0 && dattrflag != 0)
2136 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2144 * nfs remove directory call
2147 nfs_rmdir(struct vop_rmdir_args *ap)
2149 struct vnode *vp = ap->a_vp;
2150 struct vnode *dvp = ap->a_dvp;
2151 struct componentname *cnp = ap->a_cnp;
2152 struct nfsnode *dnp;
2153 struct nfsvattr dnfsva;
2154 int error, dattrflag;
2158 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2159 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2161 mtx_lock(&dnp->n_mtx);
2162 dnp->n_flag |= NMODIFIED;
2163 if (dattrflag != 0) {
2164 mtx_unlock(&dnp->n_mtx);
2165 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2167 dnp->n_attrstamp = 0;
2168 mtx_unlock(&dnp->n_mtx);
2169 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2174 if (error && NFS_ISV4(dvp))
2175 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2178 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2180 if (error == ENOENT)
2189 nfs_readdir(struct vop_readdir_args *ap)
2191 struct vnode *vp = ap->a_vp;
2192 struct nfsnode *np = VTONFS(vp);
2193 struct uio *uio = ap->a_uio;
2194 ssize_t tresid, left;
2198 if (ap->a_eofflag != NULL)
2200 if (vp->v_type != VDIR)
2204 * First, check for hit on the EOF offset cache
2206 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2207 (np->n_flag & NMODIFIED) == 0) {
2208 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2209 mtx_lock(&np->n_mtx);
2210 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2211 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2212 mtx_unlock(&np->n_mtx);
2213 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2214 if (ap->a_eofflag != NULL)
2218 mtx_unlock(&np->n_mtx);
2223 * NFS always guarantees that directory entries don't straddle
2224 * DIRBLKSIZ boundaries. As such, we need to limit the size
2225 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2228 left = uio->uio_resid % DIRBLKSIZ;
2229 if (left == uio->uio_resid)
2231 uio->uio_resid -= left;
2234 * Call ncl_bioread() to do the real work.
2236 tresid = uio->uio_resid;
2237 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2239 if (!error && uio->uio_resid == tresid) {
2240 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2241 if (ap->a_eofflag != NULL)
2245 /* Add the partial DIRBLKSIZ (left) back in. */
2246 uio->uio_resid += left;
2252 * Called from below the buffer cache by ncl_doio().
2255 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2258 struct nfsvattr nfsva;
2259 nfsuint64 *cookiep, cookie;
2260 struct nfsnode *dnp = VTONFS(vp);
2261 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2262 int error = 0, eof, attrflag;
2264 KASSERT(uiop->uio_iovcnt == 1 &&
2265 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2266 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2267 ("nfs readdirrpc bad uio"));
2270 * If there is no cookie, assume directory was stale.
2272 ncl_dircookie_lock(dnp);
2273 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2276 ncl_dircookie_unlock(dnp);
2278 ncl_dircookie_unlock(dnp);
2279 return (NFSERR_BAD_COOKIE);
2282 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2283 (void)ncl_fsinfo(nmp, vp, cred, td);
2285 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2286 &attrflag, &eof, NULL);
2288 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2292 * We are now either at the end of the directory or have filled
2296 dnp->n_direofoffset = uiop->uio_offset;
2298 if (uiop->uio_resid > 0)
2299 printf("EEK! readdirrpc resid > 0\n");
2300 ncl_dircookie_lock(dnp);
2301 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2303 ncl_dircookie_unlock(dnp);
2305 } else if (NFS_ISV4(vp)) {
2306 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2312 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2315 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2318 struct nfsvattr nfsva;
2319 nfsuint64 *cookiep, cookie;
2320 struct nfsnode *dnp = VTONFS(vp);
2321 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2322 int error = 0, attrflag, eof;
2324 KASSERT(uiop->uio_iovcnt == 1 &&
2325 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2326 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2327 ("nfs readdirplusrpc bad uio"));
2330 * If there is no cookie, assume directory was stale.
2332 ncl_dircookie_lock(dnp);
2333 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2336 ncl_dircookie_unlock(dnp);
2338 ncl_dircookie_unlock(dnp);
2339 return (NFSERR_BAD_COOKIE);
2342 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2343 (void)ncl_fsinfo(nmp, vp, cred, td);
2344 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2345 &attrflag, &eof, NULL);
2347 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2351 * We are now either at end of the directory or have filled the
2355 dnp->n_direofoffset = uiop->uio_offset;
2357 if (uiop->uio_resid > 0)
2358 printf("EEK! readdirplusrpc resid > 0\n");
2359 ncl_dircookie_lock(dnp);
2360 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2362 ncl_dircookie_unlock(dnp);
2364 } else if (NFS_ISV4(vp)) {
2365 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2371 * Silly rename. To make the NFS filesystem that is stateless look a little
2372 * more like the "ufs" a remove of an active vnode is translated to a rename
2373 * to a funny looking filename that is removed by nfs_inactive on the
2374 * nfsnode. There is the potential for another process on a different client
2375 * to create the same funny name between the nfs_lookitup() fails and the
2376 * nfs_rename() completes, but...
2379 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2381 struct sillyrename *sp;
2385 unsigned int lticks;
2389 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2390 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2391 M_NEWNFSREQ, M_WAITOK);
2392 sp->s_cred = crhold(cnp->cn_cred);
2397 * Fudge together a funny name.
2398 * Changing the format of the funny name to accommodate more
2399 * sillynames per directory.
2400 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2401 * CPU ticks since boot.
2403 pid = cnp->cn_thread->td_proc->p_pid;
2404 lticks = (unsigned int)ticks;
2406 sp->s_namlen = sprintf(sp->s_name,
2407 ".nfs.%08x.%04x4.4", lticks,
2409 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2410 cnp->cn_thread, NULL))
2414 error = nfs_renameit(dvp, vp, cnp, sp);
2417 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2418 cnp->cn_thread, &np);
2419 np->n_sillyrename = sp;
2424 free((caddr_t)sp, M_NEWNFSREQ);
2429 * Look up a file name and optionally either update the file handle or
2430 * allocate an nfsnode, depending on the value of npp.
2431 * npp == NULL --> just do the lookup
2432 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2434 * *npp != NULL --> update the file handle in the vnode
2437 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2438 struct thread *td, struct nfsnode **npp)
2440 struct vnode *newvp = NULL, *vp;
2441 struct nfsnode *np, *dnp = VTONFS(dvp);
2442 struct nfsfh *nfhp, *onfhp;
2443 struct nfsvattr nfsva, dnfsva;
2444 struct componentname cn;
2445 int error = 0, attrflag, dattrflag;
2448 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2449 &nfhp, &attrflag, &dattrflag, NULL);
2451 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2452 if (npp && !error) {
2457 * For NFSv4, check to see if it is the same name and
2458 * replace the name, if it is different.
2460 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2461 (np->n_v4->n4_namelen != len ||
2462 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2463 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2464 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2465 dnp->n_fhp->nfh_len))) {
2467 { char nnn[100]; int nnnl;
2468 nnnl = (len < 100) ? len : 99;
2469 bcopy(name, nnn, nnnl);
2471 printf("replace=%s\n",nnn);
2474 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2475 MALLOC(np->n_v4, struct nfsv4node *,
2476 sizeof (struct nfsv4node) +
2477 dnp->n_fhp->nfh_len + len - 1,
2478 M_NFSV4NODE, M_WAITOK);
2479 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2480 np->n_v4->n4_namelen = len;
2481 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2482 dnp->n_fhp->nfh_len);
2483 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2485 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2489 * Rehash node for new file handle.
2491 vfs_hash_rehash(vp, hash);
2494 FREE((caddr_t)onfhp, M_NFSFH);
2496 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2497 FREE((caddr_t)nfhp, M_NFSFH);
2501 cn.cn_nameptr = name;
2502 cn.cn_namelen = len;
2503 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2504 &np, NULL, LK_EXCLUSIVE);
2509 if (!attrflag && *npp == NULL) {
2517 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2520 if (npp && *npp == NULL) {
2531 if (error && NFS_ISV4(dvp))
2532 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2537 * Nfs Version 3 and 4 commit rpc
2540 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2543 struct nfsvattr nfsva;
2544 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2545 int error, attrflag;
2547 mtx_lock(&nmp->nm_mtx);
2548 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2549 mtx_unlock(&nmp->nm_mtx);
2552 mtx_unlock(&nmp->nm_mtx);
2553 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2556 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2558 if (error != 0 && NFS_ISV4(vp))
2559 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2565 * For async requests when nfsiod(s) are running, queue the request by
2566 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2570 nfs_strategy(struct vop_strategy_args *ap)
2578 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2579 KASSERT(!(bp->b_flags & B_DONE),
2580 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2581 BUF_ASSERT_HELD(bp);
2583 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2584 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2586 if (bp->b_iocmd == BIO_READ)
2592 * If the op is asynchronous and an i/o daemon is waiting
2593 * queue the request, wake it up and wait for completion
2594 * otherwise just do it ourselves.
2596 if ((bp->b_flags & B_ASYNC) == 0 ||
2597 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2598 (void) ncl_doio(vp, bp, cr, curthread, 1);
2603 * fsync vnode op. Just call ncl_flush() with commit == 1.
2607 nfs_fsync(struct vop_fsync_args *ap)
2610 if (ap->a_vp->v_type != VREG) {
2612 * For NFS, metadata is changed synchronously on the server,
2613 * so there is nothing to flush. Also, ncl_flush() clears
2614 * the NMODIFIED flag and that shouldn't be done here for
2619 return (ncl_flush(ap->a_vp, ap->a_waitfor, NULL, ap->a_td, 1, 0));
2623 * Flush all the blocks associated with a vnode.
2624 * Walk through the buffer pool and push any dirty pages
2625 * associated with the vnode.
2626 * If the called_from_renewthread argument is TRUE, it has been called
2627 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2628 * waiting for a buffer write to complete.
2631 ncl_flush(struct vnode *vp, int waitfor, struct ucred *cred, struct thread *td,
2632 int commit, int called_from_renewthread)
2634 struct nfsnode *np = VTONFS(vp);
2638 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2639 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2640 int passone = 1, trycnt = 0;
2641 u_quad_t off, endoff, toff;
2642 struct ucred* wcred = NULL;
2643 struct buf **bvec = NULL;
2645 #ifndef NFS_COMMITBVECSIZ
2646 #define NFS_COMMITBVECSIZ 20
2648 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2649 int bvecsize = 0, bveccount;
2651 if (called_from_renewthread != 0)
2653 if (nmp->nm_flag & NFSMNT_INT)
2659 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2660 * server, but has not been committed to stable storage on the server
2661 * yet. On the first pass, the byte range is worked out and the commit
2662 * rpc is done. On the second pass, ncl_writebp() is called to do the
2669 if (NFS_ISV34(vp) && commit) {
2670 if (bvec != NULL && bvec != bvec_on_stack)
2673 * Count up how many buffers waiting for a commit.
2677 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2678 if (!BUF_ISLOCKED(bp) &&
2679 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2680 == (B_DELWRI | B_NEEDCOMMIT))
2684 * Allocate space to remember the list of bufs to commit. It is
2685 * important to use M_NOWAIT here to avoid a race with nfs_write.
2686 * If we can't get memory (for whatever reason), we will end up
2687 * committing the buffers one-by-one in the loop below.
2689 if (bveccount > NFS_COMMITBVECSIZ) {
2691 * Release the vnode interlock to avoid a lock
2695 bvec = (struct buf **)
2696 malloc(bveccount * sizeof(struct buf *),
2700 bvec = bvec_on_stack;
2701 bvecsize = NFS_COMMITBVECSIZ;
2703 bvecsize = bveccount;
2705 bvec = bvec_on_stack;
2706 bvecsize = NFS_COMMITBVECSIZ;
2708 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2709 if (bvecpos >= bvecsize)
2711 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2712 nbp = TAILQ_NEXT(bp, b_bobufs);
2715 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2716 (B_DELWRI | B_NEEDCOMMIT)) {
2718 nbp = TAILQ_NEXT(bp, b_bobufs);
2724 * Work out if all buffers are using the same cred
2725 * so we can deal with them all with one commit.
2727 * NOTE: we are not clearing B_DONE here, so we have
2728 * to do it later on in this routine if we intend to
2729 * initiate I/O on the bp.
2731 * Note: to avoid loopback deadlocks, we do not
2732 * assign b_runningbufspace.
2735 wcred = bp->b_wcred;
2736 else if (wcred != bp->b_wcred)
2738 vfs_busy_pages(bp, 1);
2742 * bp is protected by being locked, but nbp is not
2743 * and vfs_busy_pages() may sleep. We have to
2746 nbp = TAILQ_NEXT(bp, b_bobufs);
2749 * A list of these buffers is kept so that the
2750 * second loop knows which buffers have actually
2751 * been committed. This is necessary, since there
2752 * may be a race between the commit rpc and new
2753 * uncommitted writes on the file.
2755 bvec[bvecpos++] = bp;
2756 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2760 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2768 * Commit data on the server, as required.
2769 * If all bufs are using the same wcred, then use that with
2770 * one call for all of them, otherwise commit each one
2773 if (wcred != NOCRED)
2774 retv = ncl_commit(vp, off, (int)(endoff - off),
2778 for (i = 0; i < bvecpos; i++) {
2781 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2783 size = (u_quad_t)(bp->b_dirtyend
2785 retv = ncl_commit(vp, off, (int)size,
2791 if (retv == NFSERR_STALEWRITEVERF)
2792 ncl_clearcommit(vp->v_mount);
2795 * Now, either mark the blocks I/O done or mark the
2796 * blocks dirty, depending on whether the commit
2799 for (i = 0; i < bvecpos; i++) {
2801 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2804 * Error, leave B_DELWRI intact
2806 vfs_unbusy_pages(bp);
2810 * Success, remove B_DELWRI ( bundirty() ).
2812 * b_dirtyoff/b_dirtyend seem to be NFS
2813 * specific. We should probably move that
2814 * into bundirty(). XXX
2817 bp->b_flags |= B_ASYNC;
2819 bp->b_flags &= ~B_DONE;
2820 bp->b_ioflags &= ~BIO_ERROR;
2821 bp->b_dirtyoff = bp->b_dirtyend = 0;
2828 * Start/do any write(s) that are required.
2832 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2833 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2834 if (waitfor != MNT_WAIT || passone)
2837 error = BUF_TIMELOCK(bp,
2838 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2839 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2844 if (error == ENOLCK) {
2848 if (called_from_renewthread != 0) {
2850 * Return EIO so the flush will be retried
2856 if (newnfs_sigintr(nmp, td)) {
2860 if (slpflag == PCATCH) {
2866 if ((bp->b_flags & B_DELWRI) == 0)
2867 panic("nfs_fsync: not dirty");
2868 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2874 if (passone || !commit)
2875 bp->b_flags |= B_ASYNC;
2877 bp->b_flags |= B_ASYNC;
2879 if (newnfs_sigintr(nmp, td)) {
2890 if (waitfor == MNT_WAIT) {
2891 while (bo->bo_numoutput) {
2892 error = bufobj_wwait(bo, slpflag, slptimeo);
2895 if (called_from_renewthread != 0) {
2897 * Return EIO so that the flush will be
2903 error = newnfs_sigintr(nmp, td);
2906 if (slpflag == PCATCH) {
2913 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2918 * Wait for all the async IO requests to drain
2921 mtx_lock(&np->n_mtx);
2922 while (np->n_directio_asyncwr > 0) {
2923 np->n_flag |= NFSYNCWAIT;
2924 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2925 &np->n_mtx, slpflag | (PRIBIO + 1),
2928 if (newnfs_sigintr(nmp, td)) {
2929 mtx_unlock(&np->n_mtx);
2935 mtx_unlock(&np->n_mtx);
2938 if (NFSHASPNFS(nmp)) {
2939 nfscl_layoutcommit(vp, td);
2941 * Invalidate the attribute cache, since writes to a DS
2942 * won't update the size attribute.
2944 mtx_lock(&np->n_mtx);
2945 np->n_attrstamp = 0;
2947 mtx_lock(&np->n_mtx);
2948 if (np->n_flag & NWRITEERR) {
2949 error = np->n_error;
2950 np->n_flag &= ~NWRITEERR;
2952 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2953 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2954 np->n_flag &= ~NMODIFIED;
2955 mtx_unlock(&np->n_mtx);
2957 if (bvec != NULL && bvec != bvec_on_stack)
2959 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2960 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2961 np->n_directio_asyncwr != 0) && trycnt++ < 5) {
2962 /* try, try again... */
2967 printf("try%d\n", trycnt);
2974 * NFS advisory byte-level locks.
2977 nfs_advlock(struct vop_advlock_args *ap)
2979 struct vnode *vp = ap->a_vp;
2981 struct nfsnode *np = VTONFS(ap->a_vp);
2982 struct proc *p = (struct proc *)ap->a_id;
2983 struct thread *td = curthread; /* XXX */
2985 int ret, error = EOPNOTSUPP;
2988 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
2989 if (vp->v_type != VREG)
2991 if ((ap->a_flags & F_POSIX) != 0)
2994 cred = td->td_ucred;
2995 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2996 if (vp->v_iflag & VI_DOOMED) {
2997 NFSVOPUNLOCK(vp, 0);
3002 * If this is unlocking a write locked region, flush and
3003 * commit them before unlocking. This is required by
3004 * RFC3530 Sec. 9.3.2.
3006 if (ap->a_op == F_UNLCK &&
3007 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3009 (void) ncl_flush(vp, MNT_WAIT, cred, td, 1, 0);
3012 * Loop around doing the lock op, while a blocking lock
3013 * must wait for the lock op to succeed.
3016 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3017 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3018 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3019 ap->a_op == F_SETLK) {
3020 NFSVOPUNLOCK(vp, 0);
3021 error = nfs_catnap(PZERO | PCATCH, ret,
3025 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3026 if (vp->v_iflag & VI_DOOMED) {
3027 NFSVOPUNLOCK(vp, 0);
3031 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3032 ap->a_op == F_SETLK);
3033 if (ret == NFSERR_DENIED) {
3034 NFSVOPUNLOCK(vp, 0);
3036 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3037 NFSVOPUNLOCK(vp, 0);
3039 } else if (ret != 0) {
3040 NFSVOPUNLOCK(vp, 0);
3045 * Now, if we just got a lock, invalidate data in the buffer
3046 * cache, as required, so that the coherency conforms with
3047 * RFC3530 Sec. 9.3.2.
3049 if (ap->a_op == F_SETLK) {
3050 if ((np->n_flag & NMODIFIED) == 0) {
3051 np->n_attrstamp = 0;
3052 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3053 ret = VOP_GETATTR(vp, &va, cred);
3055 if ((np->n_flag & NMODIFIED) || ret ||
3056 np->n_change != va.va_filerev) {
3057 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3058 np->n_attrstamp = 0;
3059 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3060 ret = VOP_GETATTR(vp, &va, cred);
3062 np->n_mtime = va.va_mtime;
3063 np->n_change = va.va_filerev;
3066 /* Mark that a file lock has been acquired. */
3067 mtx_lock(&np->n_mtx);
3068 np->n_flag |= NHASBEENLOCKED;
3069 mtx_unlock(&np->n_mtx);
3071 NFSVOPUNLOCK(vp, 0);
3073 } else if (!NFS_ISV4(vp)) {
3074 error = NFSVOPLOCK(vp, LK_SHARED);
3077 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3078 size = VTONFS(vp)->n_size;
3079 NFSVOPUNLOCK(vp, 0);
3080 error = lf_advlock(ap, &(vp->v_lockf), size);
3082 if (nfs_advlock_p != NULL)
3083 error = nfs_advlock_p(ap);
3085 NFSVOPUNLOCK(vp, 0);
3089 if (error == 0 && ap->a_op == F_SETLK) {
3090 error = NFSVOPLOCK(vp, LK_SHARED);
3092 /* Mark that a file lock has been acquired. */
3093 mtx_lock(&np->n_mtx);
3094 np->n_flag |= NHASBEENLOCKED;
3095 mtx_unlock(&np->n_mtx);
3096 NFSVOPUNLOCK(vp, 0);
3104 * NFS advisory byte-level locks.
3107 nfs_advlockasync(struct vop_advlockasync_args *ap)
3109 struct vnode *vp = ap->a_vp;
3114 return (EOPNOTSUPP);
3115 error = NFSVOPLOCK(vp, LK_SHARED);
3118 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3119 size = VTONFS(vp)->n_size;
3120 NFSVOPUNLOCK(vp, 0);
3121 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3123 NFSVOPUNLOCK(vp, 0);
3130 * Print out the contents of an nfsnode.
3133 nfs_print(struct vop_print_args *ap)
3135 struct vnode *vp = ap->a_vp;
3136 struct nfsnode *np = VTONFS(vp);
3138 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3139 (uintmax_t)np->n_vattr.na_fsid);
3140 if (vp->v_type == VFIFO)
3147 * This is the "real" nfs::bwrite(struct buf*).
3148 * We set B_CACHE if this is a VMIO buffer.
3151 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3154 int oldflags = bp->b_flags;
3160 BUF_ASSERT_HELD(bp);
3162 if (bp->b_flags & B_INVAL) {
3167 bp->b_flags |= B_CACHE;
3170 * Undirty the bp. We will redirty it later if the I/O fails.
3175 bp->b_flags &= ~B_DONE;
3176 bp->b_ioflags &= ~BIO_ERROR;
3177 bp->b_iocmd = BIO_WRITE;
3179 bufobj_wref(bp->b_bufobj);
3180 curthread->td_ru.ru_oublock++;
3184 * Note: to avoid loopback deadlocks, we do not
3185 * assign b_runningbufspace.
3187 vfs_busy_pages(bp, 1);
3190 bp->b_iooffset = dbtob(bp->b_blkno);
3193 if( (oldflags & B_ASYNC) == 0) {
3194 int rtval = bufwait(bp);
3196 if (oldflags & B_DELWRI) {
3209 * nfs special file access vnode op.
3210 * Essentially just get vattr and then imitate iaccess() since the device is
3211 * local to the client.
3214 nfsspec_access(struct vop_access_args *ap)
3217 struct ucred *cred = ap->a_cred;
3218 struct vnode *vp = ap->a_vp;
3219 accmode_t accmode = ap->a_accmode;
3224 * Disallow write attempts on filesystems mounted read-only;
3225 * unless the file is a socket, fifo, or a block or character
3226 * device resident on the filesystem.
3228 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3229 switch (vp->v_type) {
3239 error = VOP_GETATTR(vp, vap, cred);
3242 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3243 accmode, cred, NULL);
3249 * Read wrapper for fifos.
3252 nfsfifo_read(struct vop_read_args *ap)
3254 struct nfsnode *np = VTONFS(ap->a_vp);
3260 mtx_lock(&np->n_mtx);
3262 vfs_timestamp(&np->n_atim);
3263 mtx_unlock(&np->n_mtx);
3264 error = fifo_specops.vop_read(ap);
3269 * Write wrapper for fifos.
3272 nfsfifo_write(struct vop_write_args *ap)
3274 struct nfsnode *np = VTONFS(ap->a_vp);
3279 mtx_lock(&np->n_mtx);
3281 vfs_timestamp(&np->n_mtim);
3282 mtx_unlock(&np->n_mtx);
3283 return(fifo_specops.vop_write(ap));
3287 * Close wrapper for fifos.
3289 * Update the times on the nfsnode then do fifo close.
3292 nfsfifo_close(struct vop_close_args *ap)
3294 struct vnode *vp = ap->a_vp;
3295 struct nfsnode *np = VTONFS(vp);
3299 mtx_lock(&np->n_mtx);
3300 if (np->n_flag & (NACC | NUPD)) {
3302 if (np->n_flag & NACC)
3304 if (np->n_flag & NUPD)
3307 if (vrefcnt(vp) == 1 &&
3308 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3310 if (np->n_flag & NACC)
3311 vattr.va_atime = np->n_atim;
3312 if (np->n_flag & NUPD)
3313 vattr.va_mtime = np->n_mtim;
3314 mtx_unlock(&np->n_mtx);
3315 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3319 mtx_unlock(&np->n_mtx);
3321 return (fifo_specops.vop_close(ap));
3325 * Just call ncl_writebp() with the force argument set to 1.
3327 * NOTE: B_DONE may or may not be set in a_bp on call.
3330 nfs_bwrite(struct buf *bp)
3333 return (ncl_writebp(bp, 1, curthread));
3336 struct buf_ops buf_ops_newnfs = {
3337 .bop_name = "buf_ops_nfs",
3338 .bop_write = nfs_bwrite,
3339 .bop_strategy = bufstrategy,
3340 .bop_sync = bufsync,
3341 .bop_bdflush = bufbdflush,
3345 nfs_getacl(struct vop_getacl_args *ap)
3349 if (ap->a_type != ACL_TYPE_NFS4)
3350 return (EOPNOTSUPP);
3351 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3353 if (error > NFSERR_STALE) {
3354 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3361 nfs_setacl(struct vop_setacl_args *ap)
3365 if (ap->a_type != ACL_TYPE_NFS4)
3366 return (EOPNOTSUPP);
3367 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3369 if (error > NFSERR_STALE) {
3370 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3377 * Return POSIX pathconf information applicable to nfs filesystems.
3380 nfs_pathconf(struct vop_pathconf_args *ap)
3382 struct nfsv3_pathconf pc;
3383 struct nfsvattr nfsva;
3384 struct vnode *vp = ap->a_vp;
3385 struct thread *td = curthread;
3386 int attrflag, error;
3388 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3389 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3390 ap->a_name == _PC_NO_TRUNC)) ||
3391 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3393 * Since only the above 4 a_names are returned by the NFSv3
3394 * Pathconf RPC, there is no point in doing it for others.
3395 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3396 * be used for _PC_NFS4_ACL as well.
3398 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3401 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3407 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3410 pc.pc_linkmax = LINK_MAX;
3411 pc.pc_namemax = NFS_MAXNAMLEN;
3413 pc.pc_chownrestricted = 1;
3414 pc.pc_caseinsensitive = 0;
3415 pc.pc_casepreserving = 1;
3418 switch (ap->a_name) {
3420 *ap->a_retval = pc.pc_linkmax;
3423 *ap->a_retval = pc.pc_namemax;
3426 *ap->a_retval = PATH_MAX;
3429 *ap->a_retval = PIPE_BUF;
3431 case _PC_CHOWN_RESTRICTED:
3432 *ap->a_retval = pc.pc_chownrestricted;
3435 *ap->a_retval = pc.pc_notrunc;
3437 case _PC_ACL_EXTENDED:
3441 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3442 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3447 case _PC_ACL_PATH_MAX:
3449 *ap->a_retval = ACL_MAX_ENTRIES;
3453 case _PC_MAC_PRESENT:
3457 /* _PC_ASYNC_IO should have been handled by upper layers. */
3458 KASSERT(0, ("_PC_ASYNC_IO should not get here"));
3467 case _PC_ALLOC_SIZE_MIN:
3468 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3470 case _PC_FILESIZEBITS:
3476 case _PC_REC_INCR_XFER_SIZE:
3477 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3479 case _PC_REC_MAX_XFER_SIZE:
3480 *ap->a_retval = -1; /* means ``unlimited'' */
3482 case _PC_REC_MIN_XFER_SIZE:
3483 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3485 case _PC_REC_XFER_ALIGN:
3486 *ap->a_retval = PAGE_SIZE;
3488 case _PC_SYMLINK_MAX:
3489 *ap->a_retval = NFS_MAXPATHLEN;