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;
143 static vop_set_text_t nfs_set_text;
146 * Global vfs data structures for nfs
148 struct vop_vector newnfs_vnodeops = {
149 .vop_default = &default_vnodeops,
150 .vop_access = nfs_access,
151 .vop_advlock = nfs_advlock,
152 .vop_advlockasync = nfs_advlockasync,
153 .vop_close = nfs_close,
154 .vop_create = nfs_create,
155 .vop_fsync = nfs_fsync,
156 .vop_getattr = nfs_getattr,
157 .vop_getpages = ncl_getpages,
158 .vop_putpages = ncl_putpages,
159 .vop_inactive = ncl_inactive,
160 .vop_link = nfs_link,
161 .vop_lookup = nfs_lookup,
162 .vop_mkdir = nfs_mkdir,
163 .vop_mknod = nfs_mknod,
164 .vop_open = nfs_open,
165 .vop_pathconf = nfs_pathconf,
166 .vop_print = nfs_print,
167 .vop_read = nfs_read,
168 .vop_readdir = nfs_readdir,
169 .vop_readlink = nfs_readlink,
170 .vop_reclaim = ncl_reclaim,
171 .vop_remove = nfs_remove,
172 .vop_rename = nfs_rename,
173 .vop_rmdir = nfs_rmdir,
174 .vop_setattr = nfs_setattr,
175 .vop_strategy = nfs_strategy,
176 .vop_symlink = nfs_symlink,
177 .vop_write = ncl_write,
178 .vop_getacl = nfs_getacl,
179 .vop_setacl = nfs_setacl,
180 .vop_set_text = nfs_set_text,
183 struct vop_vector newnfs_fifoops = {
184 .vop_default = &fifo_specops,
185 .vop_access = nfsspec_access,
186 .vop_close = nfsfifo_close,
187 .vop_fsync = nfs_fsync,
188 .vop_getattr = nfs_getattr,
189 .vop_inactive = ncl_inactive,
190 .vop_print = nfs_print,
191 .vop_read = nfsfifo_read,
192 .vop_reclaim = ncl_reclaim,
193 .vop_setattr = nfs_setattr,
194 .vop_write = nfsfifo_write,
197 static int nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp,
198 struct componentname *cnp, struct vattr *vap);
199 static int nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
200 int namelen, struct ucred *cred, struct thread *td);
201 static int nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp,
202 char *fnameptr, int fnamelen, struct vnode *tdvp, struct vnode *tvp,
203 char *tnameptr, int tnamelen, struct ucred *cred, struct thread *td);
204 static int nfs_renameit(struct vnode *sdvp, struct vnode *svp,
205 struct componentname *scnp, struct sillyrename *sp);
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 (NFSCL_FORCEDISM(vp->v_mount))
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, 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, 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);
720 mtx_lock(&np->n_mtx);
723 * Invalidate the attribute cache in all cases.
724 * An open is going to fetch fresh attrs any way, other procs
725 * on this node that have file open will be forced to do an
726 * otw attr fetch, but this is safe.
727 * --> A user found that their RPC count dropped by 20% when
728 * this was commented out and I can't see any requirement
729 * for it, so I've disabled it when negative lookups are
730 * enabled. (What does this have to do with negative lookup
731 * caching? Well nothing, except it was reported by the
732 * same user that needed negative lookup caching and I wanted
733 * there to be a way to disable it to see if it
734 * is the cause of some caching/coherency issue that might
737 if (VFSTONFS(vp->v_mount)->nm_negnametimeo == 0) {
739 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
741 if (np->n_flag & NWRITEERR) {
742 np->n_flag &= ~NWRITEERR;
745 mtx_unlock(&np->n_mtx);
750 * Get attributes so "change" is up to date.
752 if (error == 0 && nfscl_mustflush(vp) != 0 &&
753 vp->v_type == VREG &&
754 (VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOCTO) == 0) {
755 ret = nfsrpc_getattr(vp, cred, ap->a_td, &nfsva,
758 np->n_change = nfsva.na_filerev;
759 (void) nfscl_loadattrcache(&vp, &nfsva, NULL,
767 ret = nfsrpc_close(vp, 0, ap->a_td);
771 error = nfscl_maperr(ap->a_td, error, (uid_t)0,
774 if (newnfs_directio_enable)
775 KASSERT((np->n_directio_asyncwr == 0),
776 ("nfs_close: dirty unflushed (%d) directio buffers\n",
777 np->n_directio_asyncwr));
778 if (newnfs_directio_enable && (fmode & O_DIRECT) && (vp->v_type == VREG)) {
779 mtx_lock(&np->n_mtx);
780 KASSERT((np->n_directio_opens > 0),
781 ("nfs_close: unexpectedly value (0) of n_directio_opens\n"));
782 np->n_directio_opens--;
783 if (np->n_directio_opens == 0)
784 np->n_flag &= ~NNONCACHE;
785 mtx_unlock(&np->n_mtx);
793 * nfs getattr call from vfs.
796 nfs_getattr(struct vop_getattr_args *ap)
798 struct vnode *vp = ap->a_vp;
799 struct thread *td = curthread; /* XXX */
800 struct nfsnode *np = VTONFS(vp);
802 struct nfsvattr nfsva;
803 struct vattr *vap = ap->a_vap;
807 * Update local times for special files.
809 mtx_lock(&np->n_mtx);
810 if (np->n_flag & (NACC | NUPD))
812 mtx_unlock(&np->n_mtx);
814 * First look in the cache.
816 if (ncl_getattrcache(vp, &vattr) == 0) {
817 vap->va_type = vattr.va_type;
818 vap->va_mode = vattr.va_mode;
819 vap->va_nlink = vattr.va_nlink;
820 vap->va_uid = vattr.va_uid;
821 vap->va_gid = vattr.va_gid;
822 vap->va_fsid = vattr.va_fsid;
823 vap->va_fileid = vattr.va_fileid;
824 vap->va_size = vattr.va_size;
825 vap->va_blocksize = vattr.va_blocksize;
826 vap->va_atime = vattr.va_atime;
827 vap->va_mtime = vattr.va_mtime;
828 vap->va_ctime = vattr.va_ctime;
829 vap->va_gen = vattr.va_gen;
830 vap->va_flags = vattr.va_flags;
831 vap->va_rdev = vattr.va_rdev;
832 vap->va_bytes = vattr.va_bytes;
833 vap->va_filerev = vattr.va_filerev;
835 * Get the local modify time for the case of a write
838 nfscl_deleggetmodtime(vp, &vap->va_mtime);
842 if (NFS_ISV34(vp) && nfs_prime_access_cache &&
843 nfsaccess_cache_timeout > 0) {
844 NFSINCRGLOBAL(nfsstatsv1.accesscache_misses);
845 nfs34_access_otw(vp, NFSACCESS_ALL, td, ap->a_cred, NULL);
846 if (ncl_getattrcache(vp, ap->a_vap) == 0) {
847 nfscl_deleggetmodtime(vp, &ap->a_vap->va_mtime);
851 error = nfsrpc_getattr(vp, ap->a_cred, td, &nfsva, NULL);
853 error = nfscl_loadattrcache(&vp, &nfsva, vap, NULL, 0, 0);
856 * Get the local modify time for the case of a write
859 nfscl_deleggetmodtime(vp, &vap->va_mtime);
860 } else if (NFS_ISV4(vp)) {
861 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
870 nfs_setattr(struct vop_setattr_args *ap)
872 struct vnode *vp = ap->a_vp;
873 struct nfsnode *np = VTONFS(vp);
874 struct thread *td = curthread; /* XXX */
875 struct vattr *vap = ap->a_vap;
884 * Setting of flags and marking of atimes are not supported.
886 if (vap->va_flags != VNOVAL)
890 * Disallow write attempts if the filesystem is mounted read-only.
892 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
893 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
894 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
895 (vp->v_mount->mnt_flag & MNT_RDONLY))
897 if (vap->va_size != VNOVAL) {
898 switch (vp->v_type) {
905 if (vap->va_mtime.tv_sec == VNOVAL &&
906 vap->va_atime.tv_sec == VNOVAL &&
907 vap->va_mode == (mode_t)VNOVAL &&
908 vap->va_uid == (uid_t)VNOVAL &&
909 vap->va_gid == (gid_t)VNOVAL)
911 vap->va_size = VNOVAL;
915 * Disallow write attempts if the filesystem is
918 if (vp->v_mount->mnt_flag & MNT_RDONLY)
921 * We run vnode_pager_setsize() early (why?),
922 * we must set np->n_size now to avoid vinvalbuf
923 * V_SAVE races that might setsize a lower
926 mtx_lock(&np->n_mtx);
928 mtx_unlock(&np->n_mtx);
929 error = ncl_meta_setsize(vp, ap->a_cred, td,
931 mtx_lock(&np->n_mtx);
932 if (np->n_flag & NMODIFIED) {
934 mtx_unlock(&np->n_mtx);
935 error = ncl_vinvalbuf(vp, vap->va_size == 0 ?
938 vnode_pager_setsize(vp, tsize);
942 * Call nfscl_delegmodtime() to set the modify time
943 * locally, as required.
945 nfscl_delegmodtime(vp);
947 mtx_unlock(&np->n_mtx);
949 * np->n_size has already been set to vap->va_size
950 * in ncl_meta_setsize(). We must set it again since
951 * nfs_loadattrcache() could be called through
952 * ncl_meta_setsize() and could modify np->n_size.
954 mtx_lock(&np->n_mtx);
955 np->n_vattr.na_size = np->n_size = vap->va_size;
956 mtx_unlock(&np->n_mtx);
959 mtx_lock(&np->n_mtx);
960 if ((vap->va_mtime.tv_sec != VNOVAL || vap->va_atime.tv_sec != VNOVAL) &&
961 (np->n_flag & NMODIFIED) && vp->v_type == VREG) {
962 mtx_unlock(&np->n_mtx);
963 error = ncl_vinvalbuf(vp, V_SAVE, td, 1);
964 if (error == EINTR || error == EIO)
967 mtx_unlock(&np->n_mtx);
969 error = nfs_setattrrpc(vp, vap, ap->a_cred, td);
970 if (error && vap->va_size != VNOVAL) {
971 mtx_lock(&np->n_mtx);
972 np->n_size = np->n_vattr.na_size = tsize;
973 vnode_pager_setsize(vp, tsize);
974 mtx_unlock(&np->n_mtx);
980 * Do an nfs setattr rpc.
983 nfs_setattrrpc(struct vnode *vp, struct vattr *vap, struct ucred *cred,
986 struct nfsnode *np = VTONFS(vp);
987 int error, ret, attrflag, i;
988 struct nfsvattr nfsva;
991 mtx_lock(&np->n_mtx);
992 for (i = 0; i < NFS_ACCESSCACHESIZE; i++)
993 np->n_accesscache[i].stamp = 0;
994 np->n_flag |= NDELEGMOD;
995 mtx_unlock(&np->n_mtx);
996 KDTRACE_NFS_ACCESSCACHE_FLUSH_DONE(vp);
998 error = nfsrpc_setattr(vp, vap, NULL, cred, td, &nfsva, &attrflag,
1001 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1005 if (error && NFS_ISV4(vp))
1006 error = nfscl_maperr(td, error, vap->va_uid, vap->va_gid);
1011 * nfs lookup call, one step at a time...
1012 * First look in cache
1013 * If not found, unlock the directory nfsnode and do the rpc
1016 nfs_lookup(struct vop_lookup_args *ap)
1018 struct componentname *cnp = ap->a_cnp;
1019 struct vnode *dvp = ap->a_dvp;
1020 struct vnode **vpp = ap->a_vpp;
1021 struct mount *mp = dvp->v_mount;
1022 int flags = cnp->cn_flags;
1023 struct vnode *newvp;
1024 struct nfsmount *nmp;
1025 struct nfsnode *np, *newnp;
1026 int error = 0, attrflag, dattrflag, ltype, ncticks;
1027 struct thread *td = cnp->cn_thread;
1029 struct nfsvattr dnfsva, nfsva;
1031 struct timespec nctime;
1034 if ((flags & ISLASTCN) && (mp->mnt_flag & MNT_RDONLY) &&
1035 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
1037 if (dvp->v_type != VDIR)
1042 /* For NFSv4, wait until any remove is done. */
1043 mtx_lock(&np->n_mtx);
1044 while (NFSHASNFSV4(nmp) && (np->n_flag & NREMOVEINPROG)) {
1045 np->n_flag |= NREMOVEWANT;
1046 (void) msleep((caddr_t)np, &np->n_mtx, PZERO, "nfslkup", 0);
1048 mtx_unlock(&np->n_mtx);
1050 if ((error = VOP_ACCESS(dvp, VEXEC, cnp->cn_cred, td)) != 0)
1052 error = cache_lookup(dvp, vpp, cnp, &nctime, &ncticks);
1053 if (error > 0 && error != ENOENT)
1057 * Lookups of "." are special and always return the
1058 * current directory. cache_lookup() already handles
1059 * associated locking bookkeeping, etc.
1061 if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
1062 /* XXX: Is this really correct? */
1063 if (cnp->cn_nameiop != LOOKUP &&
1065 cnp->cn_flags |= SAVENAME;
1070 * We only accept a positive hit in the cache if the
1071 * change time of the file matches our cached copy.
1072 * Otherwise, we discard the cache entry and fallback
1073 * to doing a lookup RPC. We also only trust cache
1074 * entries for less than nm_nametimeo seconds.
1076 * To better handle stale file handles and attributes,
1077 * clear the attribute cache of this node if it is a
1078 * leaf component, part of an open() call, and not
1079 * locally modified before fetching the attributes.
1080 * This should allow stale file handles to be detected
1081 * here where we can fall back to a LOOKUP RPC to
1082 * recover rather than having nfs_open() detect the
1083 * stale file handle and failing open(2) with ESTALE.
1086 newnp = VTONFS(newvp);
1087 if (!(nmp->nm_flag & NFSMNT_NOCTO) &&
1088 (flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1089 !(newnp->n_flag & NMODIFIED)) {
1090 mtx_lock(&newnp->n_mtx);
1091 newnp->n_attrstamp = 0;
1092 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1093 mtx_unlock(&newnp->n_mtx);
1095 if (nfscl_nodeleg(newvp, 0) == 0 ||
1096 ((u_int)(ticks - ncticks) < (nmp->nm_nametimeo * hz) &&
1097 VOP_GETATTR(newvp, &vattr, cnp->cn_cred) == 0 &&
1098 timespeccmp(&vattr.va_ctime, &nctime, ==))) {
1099 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1100 if (cnp->cn_nameiop != LOOKUP &&
1102 cnp->cn_flags |= SAVENAME;
1111 } else if (error == ENOENT) {
1112 if (dvp->v_iflag & VI_DOOMED)
1115 * We only accept a negative hit in the cache if the
1116 * modification time of the parent directory matches
1117 * the cached copy in the name cache entry.
1118 * Otherwise, we discard all of the negative cache
1119 * entries for this directory. We also only trust
1120 * negative cache entries for up to nm_negnametimeo
1123 if ((u_int)(ticks - ncticks) < (nmp->nm_negnametimeo * hz) &&
1124 VOP_GETATTR(dvp, &vattr, cnp->cn_cred) == 0 &&
1125 timespeccmp(&vattr.va_mtime, &nctime, ==)) {
1126 NFSINCRGLOBAL(nfsstatsv1.lookupcache_hits);
1129 cache_purge_negative(dvp);
1134 NFSINCRGLOBAL(nfsstatsv1.lookupcache_misses);
1135 error = nfsrpc_lookup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1136 cnp->cn_cred, td, &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1139 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1141 if (newvp != NULLVP) {
1146 if (error != ENOENT) {
1148 error = nfscl_maperr(td, error, (uid_t)0,
1153 /* The requested file was not found. */
1154 if ((cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME) &&
1155 (flags & ISLASTCN)) {
1157 * XXX: UFS does a full VOP_ACCESS(dvp,
1158 * VWRITE) here instead of just checking
1161 if (mp->mnt_flag & MNT_RDONLY)
1163 cnp->cn_flags |= SAVENAME;
1164 return (EJUSTRETURN);
1167 if ((cnp->cn_flags & MAKEENTRY) != 0 && dattrflag) {
1169 * Cache the modification time of the parent
1170 * directory from the post-op attributes in
1171 * the name cache entry. The negative cache
1172 * entry will be ignored once the directory
1173 * has changed. Don't bother adding the entry
1174 * if the directory has already changed.
1176 mtx_lock(&np->n_mtx);
1177 if (timespeccmp(&np->n_vattr.na_mtime,
1178 &dnfsva.na_mtime, ==)) {
1179 mtx_unlock(&np->n_mtx);
1180 cache_enter_time(dvp, NULL, cnp,
1181 &dnfsva.na_mtime, NULL);
1183 mtx_unlock(&np->n_mtx);
1189 * Handle RENAME case...
1191 if (cnp->cn_nameiop == RENAME && (flags & ISLASTCN)) {
1192 if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1193 FREE((caddr_t)nfhp, M_NFSFH);
1196 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1202 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1205 cnp->cn_flags |= SAVENAME;
1209 if (flags & ISDOTDOT) {
1210 ltype = NFSVOPISLOCKED(dvp);
1211 error = vfs_busy(mp, MBF_NOWAIT);
1214 NFSVOPUNLOCK(dvp, 0);
1215 error = vfs_busy(mp, 0);
1216 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1218 if (error == 0 && (dvp->v_iflag & VI_DOOMED)) {
1225 NFSVOPUNLOCK(dvp, 0);
1226 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1232 NFSVOPLOCK(dvp, ltype | LK_RETRY);
1233 if (dvp->v_iflag & VI_DOOMED) {
1245 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1247 } else if (NFS_CMPFH(np, nfhp->nfh_fh, nfhp->nfh_len)) {
1248 FREE((caddr_t)nfhp, M_NFSFH);
1252 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1255 error = nfscl_nget(mp, dvp, nfhp, cnp, td, &np, NULL,
1261 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1263 else if ((flags & (ISLASTCN | ISOPEN)) == (ISLASTCN | ISOPEN) &&
1264 !(np->n_flag & NMODIFIED)) {
1266 * Flush the attribute cache when opening a
1267 * leaf node to ensure that fresh attributes
1268 * are fetched in nfs_open() since we did not
1269 * fetch attributes from the LOOKUP reply.
1271 mtx_lock(&np->n_mtx);
1272 np->n_attrstamp = 0;
1273 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(newvp);
1274 mtx_unlock(&np->n_mtx);
1277 if (cnp->cn_nameiop != LOOKUP && (flags & ISLASTCN))
1278 cnp->cn_flags |= SAVENAME;
1279 if ((cnp->cn_flags & MAKEENTRY) &&
1280 (cnp->cn_nameiop != DELETE || !(flags & ISLASTCN)) &&
1281 attrflag != 0 && (newvp->v_type != VDIR || dattrflag != 0))
1282 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1283 newvp->v_type != VDIR ? NULL : &dnfsva.na_ctime);
1290 * Just call ncl_bioread() to do the work.
1293 nfs_read(struct vop_read_args *ap)
1295 struct vnode *vp = ap->a_vp;
1297 switch (vp->v_type) {
1299 return (ncl_bioread(vp, ap->a_uio, ap->a_ioflag, ap->a_cred));
1303 return (EOPNOTSUPP);
1311 nfs_readlink(struct vop_readlink_args *ap)
1313 struct vnode *vp = ap->a_vp;
1315 if (vp->v_type != VLNK)
1317 return (ncl_bioread(vp, ap->a_uio, 0, ap->a_cred));
1321 * Do a readlink rpc.
1322 * Called by ncl_doio() from below the buffer cache.
1325 ncl_readlinkrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1327 int error, ret, attrflag;
1328 struct nfsvattr nfsva;
1330 error = nfsrpc_readlink(vp, uiop, cred, uiop->uio_td, &nfsva,
1333 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1337 if (error && NFS_ISV4(vp))
1338 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1347 ncl_readrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred)
1349 int error, ret, attrflag;
1350 struct nfsvattr nfsva;
1351 struct nfsmount *nmp;
1353 nmp = VFSTONFS(vnode_mount(vp));
1356 if (NFSHASPNFS(nmp))
1357 error = nfscl_doiods(vp, uiop, NULL, NULL,
1358 NFSV4OPEN_ACCESSREAD, 0, cred, uiop->uio_td);
1359 NFSCL_DEBUG(4, "readrpc: aft doiods=%d\n", error);
1361 error = nfsrpc_read(vp, uiop, cred, uiop->uio_td, &nfsva,
1364 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1368 if (error && NFS_ISV4(vp))
1369 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1377 ncl_writerpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
1378 int *iomode, int *must_commit, int called_from_strategy)
1380 struct nfsvattr nfsva;
1381 int error, attrflag, ret;
1382 struct nfsmount *nmp;
1384 nmp = VFSTONFS(vnode_mount(vp));
1387 if (NFSHASPNFS(nmp))
1388 error = nfscl_doiods(vp, uiop, iomode, must_commit,
1389 NFSV4OPEN_ACCESSWRITE, 0, cred, uiop->uio_td);
1390 NFSCL_DEBUG(4, "writerpc: aft doiods=%d\n", error);
1392 error = nfsrpc_write(vp, uiop, iomode, must_commit, cred,
1393 uiop->uio_td, &nfsva, &attrflag, NULL,
1394 called_from_strategy);
1396 if (VTONFS(vp)->n_flag & ND_NFSV4)
1397 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 1,
1400 ret = nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
1406 *iomode = NFSWRITE_FILESYNC;
1407 if (error && NFS_ISV4(vp))
1408 error = nfscl_maperr(uiop->uio_td, error, (uid_t)0, (gid_t)0);
1414 * For NFS v2 this is a kludge. Use a create rpc but with the IFMT bits of the
1415 * mode set to specify the file type and the size field for rdev.
1418 nfs_mknodrpc(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
1421 struct nfsvattr nfsva, dnfsva;
1422 struct vnode *newvp = NULL;
1423 struct nfsnode *np = NULL, *dnp;
1426 int error = 0, attrflag, dattrflag;
1429 if (vap->va_type == VCHR || vap->va_type == VBLK)
1430 rdev = vap->va_rdev;
1431 else if (vap->va_type == VFIFO || vap->va_type == VSOCK)
1434 return (EOPNOTSUPP);
1435 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1437 error = nfsrpc_mknod(dvp, cnp->cn_nameptr, cnp->cn_namelen, vap,
1438 rdev, vap->va_type, cnp->cn_cred, cnp->cn_thread, &dnfsva,
1439 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
1442 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1443 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1444 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1447 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1448 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1451 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1454 if (attrflag != 0) {
1455 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1463 } else if (NFS_ISV4(dvp)) {
1464 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1468 mtx_lock(&dnp->n_mtx);
1469 dnp->n_flag |= NMODIFIED;
1471 dnp->n_attrstamp = 0;
1472 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1474 mtx_unlock(&dnp->n_mtx);
1480 * just call nfs_mknodrpc() to do the work.
1484 nfs_mknod(struct vop_mknod_args *ap)
1486 return (nfs_mknodrpc(ap->a_dvp, ap->a_vpp, ap->a_cnp, ap->a_vap));
1489 static struct mtx nfs_cverf_mtx;
1490 MTX_SYSINIT(nfs_cverf_mtx, &nfs_cverf_mtx, "NFS create verifier mutex",
1496 static nfsquad_t cverf;
1498 static int cverf_initialized = 0;
1500 mtx_lock(&nfs_cverf_mtx);
1501 if (cverf_initialized == 0) {
1502 cverf.lval[0] = arc4random();
1503 cverf.lval[1] = arc4random();
1504 cverf_initialized = 1;
1508 mtx_unlock(&nfs_cverf_mtx);
1514 * nfs file create call
1517 nfs_create(struct vop_create_args *ap)
1519 struct vnode *dvp = ap->a_dvp;
1520 struct vattr *vap = ap->a_vap;
1521 struct componentname *cnp = ap->a_cnp;
1522 struct nfsnode *np = NULL, *dnp;
1523 struct vnode *newvp = NULL;
1524 struct nfsmount *nmp;
1525 struct nfsvattr dnfsva, nfsva;
1528 int error = 0, attrflag, dattrflag, fmode = 0;
1532 * Oops, not for me..
1534 if (vap->va_type == VSOCK)
1535 return (nfs_mknodrpc(dvp, ap->a_vpp, cnp, vap));
1537 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)))
1539 if (vap->va_vaflags & VA_EXCLUSIVE)
1542 nmp = VFSTONFS(vnode_mount(dvp));
1544 /* For NFSv4, wait until any remove is done. */
1545 mtx_lock(&dnp->n_mtx);
1546 while (NFSHASNFSV4(nmp) && (dnp->n_flag & NREMOVEINPROG)) {
1547 dnp->n_flag |= NREMOVEWANT;
1548 (void) msleep((caddr_t)dnp, &dnp->n_mtx, PZERO, "nfscrt", 0);
1550 mtx_unlock(&dnp->n_mtx);
1552 cverf = nfs_get_cverf();
1553 error = nfsrpc_create(dvp, cnp->cn_nameptr, cnp->cn_namelen,
1554 vap, cverf, fmode, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva,
1555 &nfhp, &attrflag, &dattrflag, NULL);
1558 (void) nfsrpc_lookup(dvp, cnp->cn_nameptr,
1559 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread,
1560 &dnfsva, &nfsva, &nfhp, &attrflag, &dattrflag,
1563 error = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp,
1564 cnp->cn_thread, &np, NULL, LK_EXCLUSIVE);
1567 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1571 error = nfsrpc_getattr(newvp, cnp->cn_cred,
1572 cnp->cn_thread, &nfsva, NULL);
1574 error = nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
1578 if (newvp != NULL) {
1582 if (NFS_ISV34(dvp) && (fmode & O_EXCL) &&
1583 error == NFSERR_NOTSUPP) {
1587 } else if (NFS_ISV34(dvp) && (fmode & O_EXCL)) {
1588 if (nfscl_checksattr(vap, &nfsva)) {
1589 error = nfsrpc_setattr(newvp, vap, NULL, cnp->cn_cred,
1590 cnp->cn_thread, &nfsva, &attrflag, NULL);
1591 if (error && (vap->va_uid != (uid_t)VNOVAL ||
1592 vap->va_gid != (gid_t)VNOVAL)) {
1593 /* try again without setting uid/gid */
1594 vap->va_uid = (uid_t)VNOVAL;
1595 vap->va_gid = (uid_t)VNOVAL;
1596 error = nfsrpc_setattr(newvp, vap, NULL,
1597 cnp->cn_cred, cnp->cn_thread, &nfsva,
1601 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
1608 if ((cnp->cn_flags & MAKEENTRY) && attrflag)
1609 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
1612 } else if (NFS_ISV4(dvp)) {
1613 error = nfscl_maperr(cnp->cn_thread, error, vap->va_uid,
1616 mtx_lock(&dnp->n_mtx);
1617 dnp->n_flag |= NMODIFIED;
1619 dnp->n_attrstamp = 0;
1620 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1622 mtx_unlock(&dnp->n_mtx);
1627 * nfs file remove call
1628 * To try and make nfs semantics closer to ufs semantics, a file that has
1629 * other processes using the vnode is renamed instead of removed and then
1630 * removed later on the last close.
1631 * - If v_usecount > 1
1632 * If a rename is not already in the works
1633 * call nfs_sillyrename() to set it up
1638 nfs_remove(struct vop_remove_args *ap)
1640 struct vnode *vp = ap->a_vp;
1641 struct vnode *dvp = ap->a_dvp;
1642 struct componentname *cnp = ap->a_cnp;
1643 struct nfsnode *np = VTONFS(vp);
1647 KASSERT((cnp->cn_flags & HASBUF) != 0, ("nfs_remove: no name"));
1648 KASSERT(vrefcnt(vp) > 0, ("nfs_remove: bad v_usecount"));
1649 if (vp->v_type == VDIR)
1651 else if (vrefcnt(vp) == 1 || (np->n_sillyrename &&
1652 VOP_GETATTR(vp, &vattr, cnp->cn_cred) == 0 &&
1653 vattr.va_nlink > 1)) {
1655 * Purge the name cache so that the chance of a lookup for
1656 * the name succeeding while the remove is in progress is
1657 * minimized. Without node locking it can still happen, such
1658 * that an I/O op returns ESTALE, but since you get this if
1659 * another host removes the file..
1663 * throw away biocache buffers, mainly to avoid
1664 * unnecessary delayed writes later.
1666 error = ncl_vinvalbuf(vp, 0, cnp->cn_thread, 1);
1667 if (error != EINTR && error != EIO)
1669 error = nfs_removerpc(dvp, vp, cnp->cn_nameptr,
1670 cnp->cn_namelen, cnp->cn_cred, cnp->cn_thread);
1672 * Kludge City: If the first reply to the remove rpc is lost..
1673 * the reply to the retransmitted request will be ENOENT
1674 * since the file was in fact removed
1675 * Therefore, we cheat and return success.
1677 if (error == ENOENT)
1679 } else if (!np->n_sillyrename)
1680 error = nfs_sillyrename(dvp, vp, cnp);
1681 mtx_lock(&np->n_mtx);
1682 np->n_attrstamp = 0;
1683 mtx_unlock(&np->n_mtx);
1684 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1689 * nfs file remove rpc called from nfs_inactive
1692 ncl_removeit(struct sillyrename *sp, struct vnode *vp)
1695 * Make sure that the directory vnode is still valid.
1696 * XXX we should lock sp->s_dvp here.
1698 if (sp->s_dvp->v_type == VBAD)
1700 return (nfs_removerpc(sp->s_dvp, vp, sp->s_name, sp->s_namlen,
1705 * Nfs remove rpc, called from nfs_remove() and ncl_removeit().
1708 nfs_removerpc(struct vnode *dvp, struct vnode *vp, char *name,
1709 int namelen, struct ucred *cred, struct thread *td)
1711 struct nfsvattr dnfsva;
1712 struct nfsnode *dnp = VTONFS(dvp);
1713 int error = 0, dattrflag;
1715 mtx_lock(&dnp->n_mtx);
1716 dnp->n_flag |= NREMOVEINPROG;
1717 mtx_unlock(&dnp->n_mtx);
1718 error = nfsrpc_remove(dvp, name, namelen, vp, cred, td, &dnfsva,
1720 mtx_lock(&dnp->n_mtx);
1721 if ((dnp->n_flag & NREMOVEWANT)) {
1722 dnp->n_flag &= ~(NREMOVEWANT | NREMOVEINPROG);
1723 mtx_unlock(&dnp->n_mtx);
1724 wakeup((caddr_t)dnp);
1726 dnp->n_flag &= ~NREMOVEINPROG;
1727 mtx_unlock(&dnp->n_mtx);
1730 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
1731 mtx_lock(&dnp->n_mtx);
1732 dnp->n_flag |= NMODIFIED;
1734 dnp->n_attrstamp = 0;
1735 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
1737 mtx_unlock(&dnp->n_mtx);
1738 if (error && NFS_ISV4(dvp))
1739 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1744 * nfs file rename call
1747 nfs_rename(struct vop_rename_args *ap)
1749 struct vnode *fvp = ap->a_fvp;
1750 struct vnode *tvp = ap->a_tvp;
1751 struct vnode *fdvp = ap->a_fdvp;
1752 struct vnode *tdvp = ap->a_tdvp;
1753 struct componentname *tcnp = ap->a_tcnp;
1754 struct componentname *fcnp = ap->a_fcnp;
1755 struct nfsnode *fnp = VTONFS(ap->a_fvp);
1756 struct nfsnode *tdnp = VTONFS(ap->a_tdvp);
1757 struct nfsv4node *newv4 = NULL;
1760 KASSERT((tcnp->cn_flags & HASBUF) != 0 &&
1761 (fcnp->cn_flags & HASBUF) != 0, ("nfs_rename: no name"));
1762 /* Check for cross-device rename */
1763 if ((fvp->v_mount != tdvp->v_mount) ||
1764 (tvp && (fvp->v_mount != tvp->v_mount))) {
1770 printf("nfs_rename: fvp == tvp (can't happen)\n");
1774 if ((error = NFSVOPLOCK(fvp, LK_EXCLUSIVE)) != 0)
1778 * We have to flush B_DELWRI data prior to renaming
1779 * the file. If we don't, the delayed-write buffers
1780 * can be flushed out later after the file has gone stale
1781 * under NFSV3. NFSV2 does not have this problem because
1782 * ( as far as I can tell ) it flushes dirty buffers more
1785 * Skip the rename operation if the fsync fails, this can happen
1786 * due to the server's volume being full, when we pushed out data
1787 * that was written back to our cache earlier. Not checking for
1788 * this condition can result in potential (silent) data loss.
1790 error = VOP_FSYNC(fvp, MNT_WAIT, fcnp->cn_thread);
1791 NFSVOPUNLOCK(fvp, 0);
1793 error = VOP_FSYNC(tvp, MNT_WAIT, tcnp->cn_thread);
1798 * If the tvp exists and is in use, sillyrename it before doing the
1799 * rename of the new file over it.
1800 * XXX Can't sillyrename a directory.
1802 if (tvp && vrefcnt(tvp) > 1 && !VTONFS(tvp)->n_sillyrename &&
1803 tvp->v_type != VDIR && !nfs_sillyrename(tdvp, tvp, tcnp)) {
1808 error = nfs_renamerpc(fdvp, fvp, fcnp->cn_nameptr, fcnp->cn_namelen,
1809 tdvp, tvp, tcnp->cn_nameptr, tcnp->cn_namelen, tcnp->cn_cred,
1812 if (error == 0 && NFS_ISV4(tdvp)) {
1814 * For NFSv4, check to see if it is the same name and
1815 * replace the name, if it is different.
1817 MALLOC(newv4, struct nfsv4node *,
1818 sizeof (struct nfsv4node) +
1819 tdnp->n_fhp->nfh_len + tcnp->cn_namelen - 1,
1820 M_NFSV4NODE, M_WAITOK);
1821 mtx_lock(&tdnp->n_mtx);
1822 mtx_lock(&fnp->n_mtx);
1823 if (fnp->n_v4 != NULL && fvp->v_type == VREG &&
1824 (fnp->n_v4->n4_namelen != tcnp->cn_namelen ||
1825 NFSBCMP(tcnp->cn_nameptr, NFS4NODENAME(fnp->n_v4),
1826 tcnp->cn_namelen) ||
1827 tdnp->n_fhp->nfh_len != fnp->n_v4->n4_fhlen ||
1828 NFSBCMP(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1829 tdnp->n_fhp->nfh_len))) {
1831 { char nnn[100]; int nnnl;
1832 nnnl = (tcnp->cn_namelen < 100) ? tcnp->cn_namelen : 99;
1833 bcopy(tcnp->cn_nameptr, nnn, nnnl);
1835 printf("ren replace=%s\n",nnn);
1838 FREE((caddr_t)fnp->n_v4, M_NFSV4NODE);
1841 fnp->n_v4->n4_fhlen = tdnp->n_fhp->nfh_len;
1842 fnp->n_v4->n4_namelen = tcnp->cn_namelen;
1843 NFSBCOPY(tdnp->n_fhp->nfh_fh, fnp->n_v4->n4_data,
1844 tdnp->n_fhp->nfh_len);
1845 NFSBCOPY(tcnp->cn_nameptr,
1846 NFS4NODENAME(fnp->n_v4), tcnp->cn_namelen);
1848 mtx_unlock(&tdnp->n_mtx);
1849 mtx_unlock(&fnp->n_mtx);
1851 FREE((caddr_t)newv4, M_NFSV4NODE);
1854 if (fvp->v_type == VDIR) {
1855 if (tvp != NULL && tvp->v_type == VDIR)
1870 * Kludge: Map ENOENT => 0 assuming that it is a reply to a retry.
1872 if (error == ENOENT)
1878 * nfs file rename rpc called from nfs_remove() above
1881 nfs_renameit(struct vnode *sdvp, struct vnode *svp, struct componentname *scnp,
1882 struct sillyrename *sp)
1885 return (nfs_renamerpc(sdvp, svp, scnp->cn_nameptr, scnp->cn_namelen,
1886 sdvp, NULL, sp->s_name, sp->s_namlen, scnp->cn_cred,
1891 * Do an nfs rename rpc. Called from nfs_rename() and nfs_renameit().
1894 nfs_renamerpc(struct vnode *fdvp, struct vnode *fvp, char *fnameptr,
1895 int fnamelen, struct vnode *tdvp, struct vnode *tvp, char *tnameptr,
1896 int tnamelen, struct ucred *cred, struct thread *td)
1898 struct nfsvattr fnfsva, tnfsva;
1899 struct nfsnode *fdnp = VTONFS(fdvp);
1900 struct nfsnode *tdnp = VTONFS(tdvp);
1901 int error = 0, fattrflag, tattrflag;
1903 error = nfsrpc_rename(fdvp, fvp, fnameptr, fnamelen, tdvp, tvp,
1904 tnameptr, tnamelen, cred, td, &fnfsva, &tnfsva, &fattrflag,
1905 &tattrflag, NULL, NULL);
1906 mtx_lock(&fdnp->n_mtx);
1907 fdnp->n_flag |= NMODIFIED;
1908 if (fattrflag != 0) {
1909 mtx_unlock(&fdnp->n_mtx);
1910 (void) nfscl_loadattrcache(&fdvp, &fnfsva, NULL, NULL, 0, 1);
1912 fdnp->n_attrstamp = 0;
1913 mtx_unlock(&fdnp->n_mtx);
1914 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(fdvp);
1916 mtx_lock(&tdnp->n_mtx);
1917 tdnp->n_flag |= NMODIFIED;
1918 if (tattrflag != 0) {
1919 mtx_unlock(&tdnp->n_mtx);
1920 (void) nfscl_loadattrcache(&tdvp, &tnfsva, NULL, NULL, 0, 1);
1922 tdnp->n_attrstamp = 0;
1923 mtx_unlock(&tdnp->n_mtx);
1924 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1926 if (error && NFS_ISV4(fdvp))
1927 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
1932 * nfs hard link create call
1935 nfs_link(struct vop_link_args *ap)
1937 struct vnode *vp = ap->a_vp;
1938 struct vnode *tdvp = ap->a_tdvp;
1939 struct componentname *cnp = ap->a_cnp;
1940 struct nfsnode *np, *tdnp;
1941 struct nfsvattr nfsva, dnfsva;
1942 int error = 0, attrflag, dattrflag;
1945 * Push all writes to the server, so that the attribute cache
1946 * doesn't get "out of sync" with the server.
1947 * XXX There should be a better way!
1949 VOP_FSYNC(vp, MNT_WAIT, cnp->cn_thread);
1951 error = nfsrpc_link(tdvp, vp, cnp->cn_nameptr, cnp->cn_namelen,
1952 cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &attrflag,
1954 tdnp = VTONFS(tdvp);
1955 mtx_lock(&tdnp->n_mtx);
1956 tdnp->n_flag |= NMODIFIED;
1957 if (dattrflag != 0) {
1958 mtx_unlock(&tdnp->n_mtx);
1959 (void) nfscl_loadattrcache(&tdvp, &dnfsva, NULL, NULL, 0, 1);
1961 tdnp->n_attrstamp = 0;
1962 mtx_unlock(&tdnp->n_mtx);
1963 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(tdvp);
1966 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
1969 mtx_lock(&np->n_mtx);
1970 np->n_attrstamp = 0;
1971 mtx_unlock(&np->n_mtx);
1972 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
1975 * If negative lookup caching is enabled, I might as well
1976 * add an entry for this node. Not necessary for correctness,
1977 * but if negative caching is enabled, then the system
1978 * must care about lookup caching hit rate, so...
1980 if (VFSTONFS(vp->v_mount)->nm_negnametimeo != 0 &&
1981 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
1982 cache_enter_time(tdvp, vp, cnp, &nfsva.na_ctime, NULL);
1984 if (error && NFS_ISV4(vp))
1985 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
1991 * nfs symbolic link create call
1994 nfs_symlink(struct vop_symlink_args *ap)
1996 struct vnode *dvp = ap->a_dvp;
1997 struct vattr *vap = ap->a_vap;
1998 struct componentname *cnp = ap->a_cnp;
1999 struct nfsvattr nfsva, dnfsva;
2001 struct nfsnode *np = NULL, *dnp;
2002 struct vnode *newvp = NULL;
2003 int error = 0, attrflag, dattrflag, ret;
2005 vap->va_type = VLNK;
2006 error = nfsrpc_symlink(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2007 ap->a_target, vap, cnp->cn_cred, cnp->cn_thread, &dnfsva,
2008 &nfsva, &nfhp, &attrflag, &dattrflag, NULL);
2010 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2011 &np, NULL, LK_EXCLUSIVE);
2017 if (newvp != NULL) {
2019 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2021 } else if (!error) {
2023 * If we do not have an error and we could not extract the
2024 * newvp from the response due to the request being NFSv2, we
2025 * have to do a lookup in order to obtain a newvp to return.
2027 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2028 cnp->cn_cred, cnp->cn_thread, &np);
2036 error = nfscl_maperr(cnp->cn_thread, error,
2037 vap->va_uid, vap->va_gid);
2043 mtx_lock(&dnp->n_mtx);
2044 dnp->n_flag |= NMODIFIED;
2045 if (dattrflag != 0) {
2046 mtx_unlock(&dnp->n_mtx);
2047 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2049 dnp->n_attrstamp = 0;
2050 mtx_unlock(&dnp->n_mtx);
2051 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2054 * If negative lookup caching is enabled, I might as well
2055 * add an entry for this node. Not necessary for correctness,
2056 * but if negative caching is enabled, then the system
2057 * must care about lookup caching hit rate, so...
2059 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2060 (cnp->cn_flags & MAKEENTRY) && attrflag != 0 && error == 0) {
2061 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime, NULL);
2070 nfs_mkdir(struct vop_mkdir_args *ap)
2072 struct vnode *dvp = ap->a_dvp;
2073 struct vattr *vap = ap->a_vap;
2074 struct componentname *cnp = ap->a_cnp;
2075 struct nfsnode *np = NULL, *dnp;
2076 struct vnode *newvp = NULL;
2079 struct nfsvattr nfsva, dnfsva;
2080 int error = 0, attrflag, dattrflag, ret;
2082 if ((error = VOP_GETATTR(dvp, &vattr, cnp->cn_cred)) != 0)
2084 vap->va_type = VDIR;
2085 error = nfsrpc_mkdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2086 vap, cnp->cn_cred, cnp->cn_thread, &dnfsva, &nfsva, &nfhp,
2087 &attrflag, &dattrflag, NULL);
2089 mtx_lock(&dnp->n_mtx);
2090 dnp->n_flag |= NMODIFIED;
2091 if (dattrflag != 0) {
2092 mtx_unlock(&dnp->n_mtx);
2093 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2095 dnp->n_attrstamp = 0;
2096 mtx_unlock(&dnp->n_mtx);
2097 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2100 ret = nfscl_nget(dvp->v_mount, dvp, nfhp, cnp, cnp->cn_thread,
2101 &np, NULL, LK_EXCLUSIVE);
2105 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL,
2110 if (!error && newvp == NULL) {
2111 error = nfs_lookitup(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2112 cnp->cn_cred, cnp->cn_thread, &np);
2115 if (newvp->v_type != VDIR)
2123 error = nfscl_maperr(cnp->cn_thread, error,
2124 vap->va_uid, vap->va_gid);
2127 * If negative lookup caching is enabled, I might as well
2128 * add an entry for this node. Not necessary for correctness,
2129 * but if negative caching is enabled, then the system
2130 * must care about lookup caching hit rate, so...
2132 if (VFSTONFS(dvp->v_mount)->nm_negnametimeo != 0 &&
2133 (cnp->cn_flags & MAKEENTRY) &&
2134 attrflag != 0 && dattrflag != 0)
2135 cache_enter_time(dvp, newvp, cnp, &nfsva.na_ctime,
2143 * nfs remove directory call
2146 nfs_rmdir(struct vop_rmdir_args *ap)
2148 struct vnode *vp = ap->a_vp;
2149 struct vnode *dvp = ap->a_dvp;
2150 struct componentname *cnp = ap->a_cnp;
2151 struct nfsnode *dnp;
2152 struct nfsvattr dnfsva;
2153 int error, dattrflag;
2157 error = nfsrpc_rmdir(dvp, cnp->cn_nameptr, cnp->cn_namelen,
2158 cnp->cn_cred, cnp->cn_thread, &dnfsva, &dattrflag, NULL);
2160 mtx_lock(&dnp->n_mtx);
2161 dnp->n_flag |= NMODIFIED;
2162 if (dattrflag != 0) {
2163 mtx_unlock(&dnp->n_mtx);
2164 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2166 dnp->n_attrstamp = 0;
2167 mtx_unlock(&dnp->n_mtx);
2168 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(dvp);
2173 if (error && NFS_ISV4(dvp))
2174 error = nfscl_maperr(cnp->cn_thread, error, (uid_t)0,
2177 * Kludge: Map ENOENT => 0 assuming that you have a reply to a retry.
2179 if (error == ENOENT)
2188 nfs_readdir(struct vop_readdir_args *ap)
2190 struct vnode *vp = ap->a_vp;
2191 struct nfsnode *np = VTONFS(vp);
2192 struct uio *uio = ap->a_uio;
2193 ssize_t tresid, left;
2197 if (ap->a_eofflag != NULL)
2199 if (vp->v_type != VDIR)
2203 * First, check for hit on the EOF offset cache
2205 if (np->n_direofoffset > 0 && uio->uio_offset >= np->n_direofoffset &&
2206 (np->n_flag & NMODIFIED) == 0) {
2207 if (VOP_GETATTR(vp, &vattr, ap->a_cred) == 0) {
2208 mtx_lock(&np->n_mtx);
2209 if ((NFS_ISV4(vp) && np->n_change == vattr.va_filerev) ||
2210 !NFS_TIMESPEC_COMPARE(&np->n_mtime, &vattr.va_mtime)) {
2211 mtx_unlock(&np->n_mtx);
2212 NFSINCRGLOBAL(nfsstatsv1.direofcache_hits);
2213 if (ap->a_eofflag != NULL)
2217 mtx_unlock(&np->n_mtx);
2222 * NFS always guarantees that directory entries don't straddle
2223 * DIRBLKSIZ boundaries. As such, we need to limit the size
2224 * to an exact multiple of DIRBLKSIZ, to avoid copying a partial
2227 left = uio->uio_resid % DIRBLKSIZ;
2228 if (left == uio->uio_resid)
2230 uio->uio_resid -= left;
2233 * Call ncl_bioread() to do the real work.
2235 tresid = uio->uio_resid;
2236 error = ncl_bioread(vp, uio, 0, ap->a_cred);
2238 if (!error && uio->uio_resid == tresid) {
2239 NFSINCRGLOBAL(nfsstatsv1.direofcache_misses);
2240 if (ap->a_eofflag != NULL)
2244 /* Add the partial DIRBLKSIZ (left) back in. */
2245 uio->uio_resid += left;
2251 * Called from below the buffer cache by ncl_doio().
2254 ncl_readdirrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2257 struct nfsvattr nfsva;
2258 nfsuint64 *cookiep, cookie;
2259 struct nfsnode *dnp = VTONFS(vp);
2260 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2261 int error = 0, eof, attrflag;
2263 KASSERT(uiop->uio_iovcnt == 1 &&
2264 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2265 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2266 ("nfs readdirrpc bad uio"));
2269 * If there is no cookie, assume directory was stale.
2271 ncl_dircookie_lock(dnp);
2272 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2275 ncl_dircookie_unlock(dnp);
2277 ncl_dircookie_unlock(dnp);
2278 return (NFSERR_BAD_COOKIE);
2281 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2282 (void)ncl_fsinfo(nmp, vp, cred, td);
2284 error = nfsrpc_readdir(vp, uiop, &cookie, cred, td, &nfsva,
2285 &attrflag, &eof, NULL);
2287 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2291 * We are now either at the end of the directory or have filled
2295 dnp->n_direofoffset = uiop->uio_offset;
2297 if (uiop->uio_resid > 0)
2298 printf("EEK! readdirrpc resid > 0\n");
2299 ncl_dircookie_lock(dnp);
2300 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2302 ncl_dircookie_unlock(dnp);
2304 } else if (NFS_ISV4(vp)) {
2305 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2311 * NFS V3 readdir plus RPC. Used in place of ncl_readdirrpc().
2314 ncl_readdirplusrpc(struct vnode *vp, struct uio *uiop, struct ucred *cred,
2317 struct nfsvattr nfsva;
2318 nfsuint64 *cookiep, cookie;
2319 struct nfsnode *dnp = VTONFS(vp);
2320 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2321 int error = 0, attrflag, eof;
2323 KASSERT(uiop->uio_iovcnt == 1 &&
2324 (uiop->uio_offset & (DIRBLKSIZ - 1)) == 0 &&
2325 (uiop->uio_resid & (DIRBLKSIZ - 1)) == 0,
2326 ("nfs readdirplusrpc bad uio"));
2329 * If there is no cookie, assume directory was stale.
2331 ncl_dircookie_lock(dnp);
2332 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 0);
2335 ncl_dircookie_unlock(dnp);
2337 ncl_dircookie_unlock(dnp);
2338 return (NFSERR_BAD_COOKIE);
2341 if (NFSHASNFSV3(nmp) && !NFSHASGOTFSINFO(nmp))
2342 (void)ncl_fsinfo(nmp, vp, cred, td);
2343 error = nfsrpc_readdirplus(vp, uiop, &cookie, cred, td, &nfsva,
2344 &attrflag, &eof, NULL);
2346 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0, 1);
2350 * We are now either at end of the directory or have filled the
2354 dnp->n_direofoffset = uiop->uio_offset;
2356 if (uiop->uio_resid > 0)
2357 printf("EEK! readdirplusrpc resid > 0\n");
2358 ncl_dircookie_lock(dnp);
2359 cookiep = ncl_getcookie(dnp, uiop->uio_offset, 1);
2361 ncl_dircookie_unlock(dnp);
2363 } else if (NFS_ISV4(vp)) {
2364 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2370 * Silly rename. To make the NFS filesystem that is stateless look a little
2371 * more like the "ufs" a remove of an active vnode is translated to a rename
2372 * to a funny looking filename that is removed by nfs_inactive on the
2373 * nfsnode. There is the potential for another process on a different client
2374 * to create the same funny name between the nfs_lookitup() fails and the
2375 * nfs_rename() completes, but...
2378 nfs_sillyrename(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
2380 struct sillyrename *sp;
2384 unsigned int lticks;
2388 KASSERT(vp->v_type != VDIR, ("nfs: sillyrename dir"));
2389 MALLOC(sp, struct sillyrename *, sizeof (struct sillyrename),
2390 M_NEWNFSREQ, M_WAITOK);
2391 sp->s_cred = crhold(cnp->cn_cred);
2396 * Fudge together a funny name.
2397 * Changing the format of the funny name to accommodate more
2398 * sillynames per directory.
2399 * The name is now changed to .nfs.<ticks>.<pid>.4, where ticks is
2400 * CPU ticks since boot.
2402 pid = cnp->cn_thread->td_proc->p_pid;
2403 lticks = (unsigned int)ticks;
2405 sp->s_namlen = sprintf(sp->s_name,
2406 ".nfs.%08x.%04x4.4", lticks,
2408 if (nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2409 cnp->cn_thread, NULL))
2413 error = nfs_renameit(dvp, vp, cnp, sp);
2416 error = nfs_lookitup(dvp, sp->s_name, sp->s_namlen, sp->s_cred,
2417 cnp->cn_thread, &np);
2418 np->n_sillyrename = sp;
2423 free((caddr_t)sp, M_NEWNFSREQ);
2428 * Look up a file name and optionally either update the file handle or
2429 * allocate an nfsnode, depending on the value of npp.
2430 * npp == NULL --> just do the lookup
2431 * *npp == NULL --> allocate a new nfsnode and make sure attributes are
2433 * *npp != NULL --> update the file handle in the vnode
2436 nfs_lookitup(struct vnode *dvp, char *name, int len, struct ucred *cred,
2437 struct thread *td, struct nfsnode **npp)
2439 struct vnode *newvp = NULL, *vp;
2440 struct nfsnode *np, *dnp = VTONFS(dvp);
2441 struct nfsfh *nfhp, *onfhp;
2442 struct nfsvattr nfsva, dnfsva;
2443 struct componentname cn;
2444 int error = 0, attrflag, dattrflag;
2447 error = nfsrpc_lookup(dvp, name, len, cred, td, &dnfsva, &nfsva,
2448 &nfhp, &attrflag, &dattrflag, NULL);
2450 (void) nfscl_loadattrcache(&dvp, &dnfsva, NULL, NULL, 0, 1);
2451 if (npp && !error) {
2456 * For NFSv4, check to see if it is the same name and
2457 * replace the name, if it is different.
2459 if (np->n_v4 != NULL && nfsva.na_type == VREG &&
2460 (np->n_v4->n4_namelen != len ||
2461 NFSBCMP(name, NFS4NODENAME(np->n_v4), len) ||
2462 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
2463 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2464 dnp->n_fhp->nfh_len))) {
2466 { char nnn[100]; int nnnl;
2467 nnnl = (len < 100) ? len : 99;
2468 bcopy(name, nnn, nnnl);
2470 printf("replace=%s\n",nnn);
2473 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
2474 MALLOC(np->n_v4, struct nfsv4node *,
2475 sizeof (struct nfsv4node) +
2476 dnp->n_fhp->nfh_len + len - 1,
2477 M_NFSV4NODE, M_WAITOK);
2478 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
2479 np->n_v4->n4_namelen = len;
2480 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
2481 dnp->n_fhp->nfh_len);
2482 NFSBCOPY(name, NFS4NODENAME(np->n_v4), len);
2484 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len,
2488 * Rehash node for new file handle.
2490 vfs_hash_rehash(vp, hash);
2493 FREE((caddr_t)onfhp, M_NFSFH);
2495 } else if (NFS_CMPFH(dnp, nfhp->nfh_fh, nfhp->nfh_len)) {
2496 FREE((caddr_t)nfhp, M_NFSFH);
2500 cn.cn_nameptr = name;
2501 cn.cn_namelen = len;
2502 error = nfscl_nget(dvp->v_mount, dvp, nfhp, &cn, td,
2503 &np, NULL, LK_EXCLUSIVE);
2508 if (!attrflag && *npp == NULL) {
2516 (void) nfscl_loadattrcache(&newvp, &nfsva, NULL, NULL,
2519 if (npp && *npp == NULL) {
2530 if (error && NFS_ISV4(dvp))
2531 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2536 * Nfs Version 3 and 4 commit rpc
2539 ncl_commit(struct vnode *vp, u_quad_t offset, int cnt, struct ucred *cred,
2542 struct nfsvattr nfsva;
2543 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2546 int error, attrflag;
2551 if (NFSHASPNFS(nmp) && (np->n_flag & NDSCOMMIT) != 0) {
2552 uio.uio_offset = offset;
2553 uio.uio_resid = cnt;
2554 error = nfscl_doiods(vp, &uio, NULL, NULL,
2555 NFSV4OPEN_ACCESSWRITE, 1, cred, td);
2557 mtx_lock(&np->n_mtx);
2558 np->n_flag &= ~NDSCOMMIT;
2559 mtx_unlock(&np->n_mtx);
2563 mtx_lock(&nmp->nm_mtx);
2564 if ((nmp->nm_state & NFSSTA_HASWRITEVERF) == 0) {
2565 mtx_unlock(&nmp->nm_mtx);
2568 mtx_unlock(&nmp->nm_mtx);
2569 error = nfsrpc_commit(vp, offset, cnt, cred, td, &nfsva,
2573 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL,
2575 if (error != 0 && NFS_ISV4(vp))
2576 error = nfscl_maperr(td, error, (uid_t)0, (gid_t)0);
2582 * For async requests when nfsiod(s) are running, queue the request by
2583 * calling ncl_asyncio(), otherwise just all ncl_doio() to do the
2587 nfs_strategy(struct vop_strategy_args *ap)
2595 KASSERT(bp->b_vp == vp, ("missing b_getvp"));
2596 KASSERT(!(bp->b_flags & B_DONE),
2597 ("nfs_strategy: buffer %p unexpectedly marked B_DONE", bp));
2598 BUF_ASSERT_HELD(bp);
2600 if (vp->v_type == VREG && bp->b_blkno == bp->b_lblkno)
2601 bp->b_blkno = bp->b_lblkno * (vp->v_bufobj.bo_bsize /
2603 if (bp->b_iocmd == BIO_READ)
2609 * If the op is asynchronous and an i/o daemon is waiting
2610 * queue the request, wake it up and wait for completion
2611 * otherwise just do it ourselves.
2613 if ((bp->b_flags & B_ASYNC) == 0 ||
2614 ncl_asyncio(VFSTONFS(vp->v_mount), bp, NOCRED, curthread))
2615 (void) ncl_doio(vp, bp, cr, curthread, 1);
2620 * fsync vnode op. Just call ncl_flush() with commit == 1.
2624 nfs_fsync(struct vop_fsync_args *ap)
2627 if (ap->a_vp->v_type != VREG) {
2629 * For NFS, metadata is changed synchronously on the server,
2630 * so there is nothing to flush. Also, ncl_flush() clears
2631 * the NMODIFIED flag and that shouldn't be done here for
2636 return (ncl_flush(ap->a_vp, ap->a_waitfor, ap->a_td, 1, 0));
2640 * Flush all the blocks associated with a vnode.
2641 * Walk through the buffer pool and push any dirty pages
2642 * associated with the vnode.
2643 * If the called_from_renewthread argument is TRUE, it has been called
2644 * from the NFSv4 renew thread and, as such, cannot block indefinitely
2645 * waiting for a buffer write to complete.
2648 ncl_flush(struct vnode *vp, int waitfor, struct thread *td,
2649 int commit, int called_from_renewthread)
2651 struct nfsnode *np = VTONFS(vp);
2655 struct nfsmount *nmp = VFSTONFS(vp->v_mount);
2656 int error = 0, slptimeo = 0, slpflag = 0, retv, bvecpos;
2657 int passone = 1, trycnt = 0;
2658 u_quad_t off, endoff, toff;
2659 struct ucred* wcred = NULL;
2660 struct buf **bvec = NULL;
2662 #ifndef NFS_COMMITBVECSIZ
2663 #define NFS_COMMITBVECSIZ 20
2665 struct buf *bvec_on_stack[NFS_COMMITBVECSIZ];
2666 int bvecsize = 0, bveccount;
2668 if (called_from_renewthread != 0)
2670 if (nmp->nm_flag & NFSMNT_INT)
2676 * A b_flags == (B_DELWRI | B_NEEDCOMMIT) block has been written to the
2677 * server, but has not been committed to stable storage on the server
2678 * yet. On the first pass, the byte range is worked out and the commit
2679 * rpc is done. On the second pass, ncl_writebp() is called to do the
2686 if (NFS_ISV34(vp) && commit) {
2687 if (bvec != NULL && bvec != bvec_on_stack)
2690 * Count up how many buffers waiting for a commit.
2694 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2695 if (!BUF_ISLOCKED(bp) &&
2696 (bp->b_flags & (B_DELWRI | B_NEEDCOMMIT))
2697 == (B_DELWRI | B_NEEDCOMMIT))
2701 * Allocate space to remember the list of bufs to commit. It is
2702 * important to use M_NOWAIT here to avoid a race with nfs_write.
2703 * If we can't get memory (for whatever reason), we will end up
2704 * committing the buffers one-by-one in the loop below.
2706 if (bveccount > NFS_COMMITBVECSIZ) {
2708 * Release the vnode interlock to avoid a lock
2712 bvec = (struct buf **)
2713 malloc(bveccount * sizeof(struct buf *),
2717 bvec = bvec_on_stack;
2718 bvecsize = NFS_COMMITBVECSIZ;
2720 bvecsize = bveccount;
2722 bvec = bvec_on_stack;
2723 bvecsize = NFS_COMMITBVECSIZ;
2725 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2726 if (bvecpos >= bvecsize)
2728 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2729 nbp = TAILQ_NEXT(bp, b_bobufs);
2732 if ((bp->b_flags & (B_DELWRI | B_NEEDCOMMIT)) !=
2733 (B_DELWRI | B_NEEDCOMMIT)) {
2735 nbp = TAILQ_NEXT(bp, b_bobufs);
2741 * Work out if all buffers are using the same cred
2742 * so we can deal with them all with one commit.
2744 * NOTE: we are not clearing B_DONE here, so we have
2745 * to do it later on in this routine if we intend to
2746 * initiate I/O on the bp.
2748 * Note: to avoid loopback deadlocks, we do not
2749 * assign b_runningbufspace.
2752 wcred = bp->b_wcred;
2753 else if (wcred != bp->b_wcred)
2755 vfs_busy_pages(bp, 1);
2759 * bp is protected by being locked, but nbp is not
2760 * and vfs_busy_pages() may sleep. We have to
2763 nbp = TAILQ_NEXT(bp, b_bobufs);
2766 * A list of these buffers is kept so that the
2767 * second loop knows which buffers have actually
2768 * been committed. This is necessary, since there
2769 * may be a race between the commit rpc and new
2770 * uncommitted writes on the file.
2772 bvec[bvecpos++] = bp;
2773 toff = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2777 toff += (u_quad_t)(bp->b_dirtyend - bp->b_dirtyoff);
2785 * Commit data on the server, as required.
2786 * If all bufs are using the same wcred, then use that with
2787 * one call for all of them, otherwise commit each one
2790 if (wcred != NOCRED)
2791 retv = ncl_commit(vp, off, (int)(endoff - off),
2795 for (i = 0; i < bvecpos; i++) {
2798 off = ((u_quad_t)bp->b_blkno) * DEV_BSIZE +
2800 size = (u_quad_t)(bp->b_dirtyend
2802 retv = ncl_commit(vp, off, (int)size,
2808 if (retv == NFSERR_STALEWRITEVERF)
2809 ncl_clearcommit(vp->v_mount);
2812 * Now, either mark the blocks I/O done or mark the
2813 * blocks dirty, depending on whether the commit
2816 for (i = 0; i < bvecpos; i++) {
2818 bp->b_flags &= ~(B_NEEDCOMMIT | B_CLUSTEROK);
2821 * Error, leave B_DELWRI intact
2823 vfs_unbusy_pages(bp);
2827 * Success, remove B_DELWRI ( bundirty() ).
2829 * b_dirtyoff/b_dirtyend seem to be NFS
2830 * specific. We should probably move that
2831 * into bundirty(). XXX
2834 bp->b_flags |= B_ASYNC;
2836 bp->b_flags &= ~B_DONE;
2837 bp->b_ioflags &= ~BIO_ERROR;
2838 bp->b_dirtyoff = bp->b_dirtyend = 0;
2845 * Start/do any write(s) that are required.
2849 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
2850 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT, NULL)) {
2851 if (waitfor != MNT_WAIT || passone)
2854 error = BUF_TIMELOCK(bp,
2855 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
2856 BO_LOCKPTR(bo), "nfsfsync", slpflag, slptimeo);
2861 if (error == ENOLCK) {
2865 if (called_from_renewthread != 0) {
2867 * Return EIO so the flush will be retried
2873 if (newnfs_sigintr(nmp, td)) {
2877 if (slpflag == PCATCH) {
2883 if ((bp->b_flags & B_DELWRI) == 0)
2884 panic("nfs_fsync: not dirty");
2885 if ((passone || !commit) && (bp->b_flags & B_NEEDCOMMIT)) {
2891 if (passone || !commit)
2892 bp->b_flags |= B_ASYNC;
2894 bp->b_flags |= B_ASYNC;
2896 if (newnfs_sigintr(nmp, td)) {
2907 if (waitfor == MNT_WAIT) {
2908 while (bo->bo_numoutput) {
2909 error = bufobj_wwait(bo, slpflag, slptimeo);
2912 if (called_from_renewthread != 0) {
2914 * Return EIO so that the flush will be
2920 error = newnfs_sigintr(nmp, td);
2923 if (slpflag == PCATCH) {
2930 if (bo->bo_dirty.bv_cnt != 0 && commit) {
2935 * Wait for all the async IO requests to drain
2938 mtx_lock(&np->n_mtx);
2939 while (np->n_directio_asyncwr > 0) {
2940 np->n_flag |= NFSYNCWAIT;
2941 error = newnfs_msleep(td, &np->n_directio_asyncwr,
2942 &np->n_mtx, slpflag | (PRIBIO + 1),
2945 if (newnfs_sigintr(nmp, td)) {
2946 mtx_unlock(&np->n_mtx);
2952 mtx_unlock(&np->n_mtx);
2955 if (NFSHASPNFS(nmp)) {
2956 nfscl_layoutcommit(vp, td);
2958 * Invalidate the attribute cache, since writes to a DS
2959 * won't update the size attribute.
2961 mtx_lock(&np->n_mtx);
2962 np->n_attrstamp = 0;
2964 mtx_lock(&np->n_mtx);
2965 if (np->n_flag & NWRITEERR) {
2966 error = np->n_error;
2967 np->n_flag &= ~NWRITEERR;
2969 if (commit && bo->bo_dirty.bv_cnt == 0 &&
2970 bo->bo_numoutput == 0 && np->n_directio_asyncwr == 0)
2971 np->n_flag &= ~NMODIFIED;
2972 mtx_unlock(&np->n_mtx);
2974 if (bvec != NULL && bvec != bvec_on_stack)
2976 if (error == 0 && commit != 0 && waitfor == MNT_WAIT &&
2977 (bo->bo_dirty.bv_cnt != 0 || bo->bo_numoutput != 0 ||
2978 np->n_directio_asyncwr != 0)) {
2980 /* try, try again... */
2987 vn_printf(vp, "ncl_flush failed");
2988 error = called_from_renewthread != 0 ? EIO : EBUSY;
2994 * NFS advisory byte-level locks.
2997 nfs_advlock(struct vop_advlock_args *ap)
2999 struct vnode *vp = ap->a_vp;
3001 struct nfsnode *np = VTONFS(ap->a_vp);
3002 struct proc *p = (struct proc *)ap->a_id;
3003 struct thread *td = curthread; /* XXX */
3005 int ret, error = EOPNOTSUPP;
3008 if (NFS_ISV4(vp) && (ap->a_flags & (F_POSIX | F_FLOCK)) != 0) {
3009 if (vp->v_type != VREG)
3011 if ((ap->a_flags & F_POSIX) != 0)
3014 cred = td->td_ucred;
3015 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3016 if (vp->v_iflag & VI_DOOMED) {
3017 NFSVOPUNLOCK(vp, 0);
3022 * If this is unlocking a write locked region, flush and
3023 * commit them before unlocking. This is required by
3024 * RFC3530 Sec. 9.3.2.
3026 if (ap->a_op == F_UNLCK &&
3027 nfscl_checkwritelocked(vp, ap->a_fl, cred, td, ap->a_id,
3029 (void) ncl_flush(vp, MNT_WAIT, td, 1, 0);
3032 * Loop around doing the lock op, while a blocking lock
3033 * must wait for the lock op to succeed.
3036 ret = nfsrpc_advlock(vp, np->n_size, ap->a_op,
3037 ap->a_fl, 0, cred, td, ap->a_id, ap->a_flags);
3038 if (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3039 ap->a_op == F_SETLK) {
3040 NFSVOPUNLOCK(vp, 0);
3041 error = nfs_catnap(PZERO | PCATCH, ret,
3045 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3046 if (vp->v_iflag & VI_DOOMED) {
3047 NFSVOPUNLOCK(vp, 0);
3051 } while (ret == NFSERR_DENIED && (ap->a_flags & F_WAIT) &&
3052 ap->a_op == F_SETLK);
3053 if (ret == NFSERR_DENIED) {
3054 NFSVOPUNLOCK(vp, 0);
3056 } else if (ret == EINVAL || ret == EBADF || ret == EINTR) {
3057 NFSVOPUNLOCK(vp, 0);
3059 } else if (ret != 0) {
3060 NFSVOPUNLOCK(vp, 0);
3065 * Now, if we just got a lock, invalidate data in the buffer
3066 * cache, as required, so that the coherency conforms with
3067 * RFC3530 Sec. 9.3.2.
3069 if (ap->a_op == F_SETLK) {
3070 if ((np->n_flag & NMODIFIED) == 0) {
3071 np->n_attrstamp = 0;
3072 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3073 ret = VOP_GETATTR(vp, &va, cred);
3075 if ((np->n_flag & NMODIFIED) || ret ||
3076 np->n_change != va.va_filerev) {
3077 (void) ncl_vinvalbuf(vp, V_SAVE, td, 1);
3078 np->n_attrstamp = 0;
3079 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
3080 ret = VOP_GETATTR(vp, &va, cred);
3082 np->n_mtime = va.va_mtime;
3083 np->n_change = va.va_filerev;
3086 /* Mark that a file lock has been acquired. */
3087 mtx_lock(&np->n_mtx);
3088 np->n_flag |= NHASBEENLOCKED;
3089 mtx_unlock(&np->n_mtx);
3091 NFSVOPUNLOCK(vp, 0);
3093 } else if (!NFS_ISV4(vp)) {
3094 error = NFSVOPLOCK(vp, LK_SHARED);
3097 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3098 size = VTONFS(vp)->n_size;
3099 NFSVOPUNLOCK(vp, 0);
3100 error = lf_advlock(ap, &(vp->v_lockf), size);
3102 if (nfs_advlock_p != NULL)
3103 error = nfs_advlock_p(ap);
3105 NFSVOPUNLOCK(vp, 0);
3109 if (error == 0 && ap->a_op == F_SETLK) {
3110 error = NFSVOPLOCK(vp, LK_SHARED);
3112 /* Mark that a file lock has been acquired. */
3113 mtx_lock(&np->n_mtx);
3114 np->n_flag |= NHASBEENLOCKED;
3115 mtx_unlock(&np->n_mtx);
3116 NFSVOPUNLOCK(vp, 0);
3124 * NFS advisory byte-level locks.
3127 nfs_advlockasync(struct vop_advlockasync_args *ap)
3129 struct vnode *vp = ap->a_vp;
3134 return (EOPNOTSUPP);
3135 error = NFSVOPLOCK(vp, LK_SHARED);
3138 if ((VFSTONFS(vp->v_mount)->nm_flag & NFSMNT_NOLOCKD) != 0) {
3139 size = VTONFS(vp)->n_size;
3140 NFSVOPUNLOCK(vp, 0);
3141 error = lf_advlockasync(ap, &(vp->v_lockf), size);
3143 NFSVOPUNLOCK(vp, 0);
3150 * Print out the contents of an nfsnode.
3153 nfs_print(struct vop_print_args *ap)
3155 struct vnode *vp = ap->a_vp;
3156 struct nfsnode *np = VTONFS(vp);
3158 printf("\tfileid %jd fsid 0x%jx", (uintmax_t)np->n_vattr.na_fileid,
3159 (uintmax_t)np->n_vattr.na_fsid);
3160 if (vp->v_type == VFIFO)
3167 * This is the "real" nfs::bwrite(struct buf*).
3168 * We set B_CACHE if this is a VMIO buffer.
3171 ncl_writebp(struct buf *bp, int force __unused, struct thread *td)
3173 int oldflags, rtval;
3175 BUF_ASSERT_HELD(bp);
3177 if (bp->b_flags & B_INVAL) {
3182 oldflags = bp->b_flags;
3183 bp->b_flags |= B_CACHE;
3186 * Undirty the bp. We will redirty it later if the I/O fails.
3189 bp->b_flags &= ~B_DONE;
3190 bp->b_ioflags &= ~BIO_ERROR;
3191 bp->b_iocmd = BIO_WRITE;
3193 bufobj_wref(bp->b_bufobj);
3194 curthread->td_ru.ru_oublock++;
3197 * Note: to avoid loopback deadlocks, we do not
3198 * assign b_runningbufspace.
3200 vfs_busy_pages(bp, 1);
3203 bp->b_iooffset = dbtob(bp->b_blkno);
3206 if ((oldflags & B_ASYNC) != 0)
3209 rtval = bufwait(bp);
3210 if (oldflags & B_DELWRI)
3217 * nfs special file access vnode op.
3218 * Essentially just get vattr and then imitate iaccess() since the device is
3219 * local to the client.
3222 nfsspec_access(struct vop_access_args *ap)
3225 struct ucred *cred = ap->a_cred;
3226 struct vnode *vp = ap->a_vp;
3227 accmode_t accmode = ap->a_accmode;
3232 * Disallow write attempts on filesystems mounted read-only;
3233 * unless the file is a socket, fifo, or a block or character
3234 * device resident on the filesystem.
3236 if ((accmode & VWRITE) && (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3237 switch (vp->v_type) {
3247 error = VOP_GETATTR(vp, vap, cred);
3250 error = vaccess(vp->v_type, vap->va_mode, vap->va_uid, vap->va_gid,
3251 accmode, cred, NULL);
3257 * Read wrapper for fifos.
3260 nfsfifo_read(struct vop_read_args *ap)
3262 struct nfsnode *np = VTONFS(ap->a_vp);
3268 mtx_lock(&np->n_mtx);
3270 vfs_timestamp(&np->n_atim);
3271 mtx_unlock(&np->n_mtx);
3272 error = fifo_specops.vop_read(ap);
3277 * Write wrapper for fifos.
3280 nfsfifo_write(struct vop_write_args *ap)
3282 struct nfsnode *np = VTONFS(ap->a_vp);
3287 mtx_lock(&np->n_mtx);
3289 vfs_timestamp(&np->n_mtim);
3290 mtx_unlock(&np->n_mtx);
3291 return(fifo_specops.vop_write(ap));
3295 * Close wrapper for fifos.
3297 * Update the times on the nfsnode then do fifo close.
3300 nfsfifo_close(struct vop_close_args *ap)
3302 struct vnode *vp = ap->a_vp;
3303 struct nfsnode *np = VTONFS(vp);
3307 mtx_lock(&np->n_mtx);
3308 if (np->n_flag & (NACC | NUPD)) {
3310 if (np->n_flag & NACC)
3312 if (np->n_flag & NUPD)
3315 if (vrefcnt(vp) == 1 &&
3316 (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
3318 if (np->n_flag & NACC)
3319 vattr.va_atime = np->n_atim;
3320 if (np->n_flag & NUPD)
3321 vattr.va_mtime = np->n_mtim;
3322 mtx_unlock(&np->n_mtx);
3323 (void)VOP_SETATTR(vp, &vattr, ap->a_cred);
3327 mtx_unlock(&np->n_mtx);
3329 return (fifo_specops.vop_close(ap));
3333 * Just call ncl_writebp() with the force argument set to 1.
3335 * NOTE: B_DONE may or may not be set in a_bp on call.
3338 nfs_bwrite(struct buf *bp)
3341 return (ncl_writebp(bp, 1, curthread));
3344 struct buf_ops buf_ops_newnfs = {
3345 .bop_name = "buf_ops_nfs",
3346 .bop_write = nfs_bwrite,
3347 .bop_strategy = bufstrategy,
3348 .bop_sync = bufsync,
3349 .bop_bdflush = bufbdflush,
3353 nfs_getacl(struct vop_getacl_args *ap)
3357 if (ap->a_type != ACL_TYPE_NFS4)
3358 return (EOPNOTSUPP);
3359 error = nfsrpc_getacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3361 if (error > NFSERR_STALE) {
3362 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3369 nfs_setacl(struct vop_setacl_args *ap)
3373 if (ap->a_type != ACL_TYPE_NFS4)
3374 return (EOPNOTSUPP);
3375 error = nfsrpc_setacl(ap->a_vp, ap->a_cred, ap->a_td, ap->a_aclp,
3377 if (error > NFSERR_STALE) {
3378 (void) nfscl_maperr(ap->a_td, error, (uid_t)0, (gid_t)0);
3385 nfs_set_text(struct vop_set_text_args *ap)
3387 struct vnode *vp = ap->a_vp;
3391 * If the text file has been mmap'd, flush any dirty pages to the
3392 * buffer cache and then...
3393 * Make sure all writes are pushed to the NFS server. If this is not
3394 * done, the modify time of the file can change while the text
3395 * file is being executed. This will cause the process that is
3396 * executing the text file to be terminated.
3398 if (vp->v_object != NULL) {
3399 VM_OBJECT_WLOCK(vp->v_object);
3400 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3401 VM_OBJECT_WUNLOCK(vp->v_object);
3404 /* Now, flush the buffer cache. */
3405 ncl_flush(vp, MNT_WAIT, curthread, 0, 0);
3407 /* And, finally, make sure that n_mtime is up to date. */
3409 mtx_lock(&np->n_mtx);
3410 np->n_mtime = np->n_vattr.na_mtime;
3411 mtx_unlock(&np->n_mtx);
3413 vp->v_vflag |= VV_TEXT;
3418 * Return POSIX pathconf information applicable to nfs filesystems.
3421 nfs_pathconf(struct vop_pathconf_args *ap)
3423 struct nfsv3_pathconf pc;
3424 struct nfsvattr nfsva;
3425 struct vnode *vp = ap->a_vp;
3426 struct thread *td = curthread;
3427 int attrflag, error;
3429 if ((NFS_ISV34(vp) && (ap->a_name == _PC_LINK_MAX ||
3430 ap->a_name == _PC_NAME_MAX || ap->a_name == _PC_CHOWN_RESTRICTED ||
3431 ap->a_name == _PC_NO_TRUNC)) ||
3432 (NFS_ISV4(vp) && ap->a_name == _PC_ACL_NFS4)) {
3434 * Since only the above 4 a_names are returned by the NFSv3
3435 * Pathconf RPC, there is no point in doing it for others.
3436 * For NFSv4, the Pathconf RPC (actually a Getattr Op.) can
3437 * be used for _PC_NFS4_ACL as well.
3439 error = nfsrpc_pathconf(vp, &pc, td->td_ucred, td, &nfsva,
3442 (void) nfscl_loadattrcache(&vp, &nfsva, NULL, NULL, 0,
3448 * For NFSv2 (or NFSv3 when not one of the above 4 a_names),
3451 pc.pc_linkmax = LINK_MAX;
3452 pc.pc_namemax = NFS_MAXNAMLEN;
3454 pc.pc_chownrestricted = 1;
3455 pc.pc_caseinsensitive = 0;
3456 pc.pc_casepreserving = 1;
3459 switch (ap->a_name) {
3461 *ap->a_retval = pc.pc_linkmax;
3464 *ap->a_retval = pc.pc_namemax;
3466 case _PC_CHOWN_RESTRICTED:
3467 *ap->a_retval = pc.pc_chownrestricted;
3470 *ap->a_retval = pc.pc_notrunc;
3472 case _PC_ACL_EXTENDED:
3476 if (NFS_ISV4(vp) && nfsrv_useacl != 0 && attrflag != 0 &&
3477 NFSISSET_ATTRBIT(&nfsva.na_suppattr, NFSATTRBIT_ACL))
3482 case _PC_ACL_PATH_MAX:
3484 *ap->a_retval = ACL_MAX_ENTRIES;
3488 case _PC_MAC_PRESENT:
3497 case _PC_ALLOC_SIZE_MIN:
3498 *ap->a_retval = vp->v_mount->mnt_stat.f_bsize;
3500 case _PC_FILESIZEBITS:
3506 case _PC_REC_INCR_XFER_SIZE:
3507 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3509 case _PC_REC_MAX_XFER_SIZE:
3510 *ap->a_retval = -1; /* means ``unlimited'' */
3512 case _PC_REC_MIN_XFER_SIZE:
3513 *ap->a_retval = vp->v_mount->mnt_stat.f_iosize;
3515 case _PC_REC_XFER_ALIGN:
3516 *ap->a_retval = PAGE_SIZE;
3518 case _PC_SYMLINK_MAX:
3519 *ap->a_retval = NFS_MAXPATHLEN;
3523 error = vop_stdpathconf(ap);