2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
40 #include "opt_inet6.h"
42 #include <sys/capsicum.h>
45 * generally, I don't like #includes inside .h files, but it seems to
46 * be the easiest way to handle the port.
50 #include <sys/sysctl.h>
51 #include <fs/nfs/nfsport.h>
52 #include <netinet/in_fib.h>
53 #include <netinet/if_ether.h>
54 #include <netinet6/ip6_var.h>
55 #include <net/if_types.h>
57 #include <fs/nfsclient/nfs_kdtrace.h>
60 dtrace_nfsclient_attrcache_flush_probe_func_t
61 dtrace_nfscl_attrcache_flush_done_probe;
62 uint32_t nfscl_attrcache_flush_done_id;
64 dtrace_nfsclient_attrcache_get_hit_probe_func_t
65 dtrace_nfscl_attrcache_get_hit_probe;
66 uint32_t nfscl_attrcache_get_hit_id;
68 dtrace_nfsclient_attrcache_get_miss_probe_func_t
69 dtrace_nfscl_attrcache_get_miss_probe;
70 uint32_t nfscl_attrcache_get_miss_id;
72 dtrace_nfsclient_attrcache_load_probe_func_t
73 dtrace_nfscl_attrcache_load_done_probe;
74 uint32_t nfscl_attrcache_load_done_id;
75 #endif /* !KDTRACE_HOOKS */
77 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
78 extern struct vop_vector newnfs_vnodeops;
79 extern struct vop_vector newnfs_fifoops;
80 extern uma_zone_t newnfsnode_zone;
81 extern struct buf_ops buf_ops_newnfs;
82 extern uma_zone_t ncl_pbuf_zone;
83 extern short nfsv4_cbport;
84 extern int nfscl_enablecallb;
85 extern int nfs_numnfscbd;
86 extern int nfscl_inited;
87 struct mtx ncl_iod_mutex;
89 extern struct mtx nfsrv_dslock_mtx;
91 extern void (*ncl_call_invalcaches)(struct vnode *);
93 SYSCTL_DECL(_vfs_nfs);
94 static int ncl_fileid_maxwarnings = 10;
95 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
96 &ncl_fileid_maxwarnings, 0,
97 "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
98 static volatile int ncl_fileid_nwarnings;
100 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
104 * Comparison function for vfs_hash functions.
107 newnfs_vncmpf(struct vnode *vp, void *arg)
109 struct nfsfh *nfhp = (struct nfsfh *)arg;
110 struct nfsnode *np = VTONFS(vp);
112 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
113 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
119 * Look up a vnode/nfsnode by file handle.
120 * Callers must check for mount points!!
121 * In all cases, a pointer to a
122 * nfsnode structure is returned.
123 * This variant takes a "struct nfsfh *" as second argument and uses
124 * that structure up, either by hanging off the nfsnode or FREEing it.
127 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
128 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
129 void *stuff, int lkflags)
131 struct nfsnode *np, *dnp;
132 struct vnode *vp, *nvp;
133 struct nfsv4node *newd, *oldd;
136 struct nfsmount *nmp;
138 nmp = VFSTONFS(mntp);
142 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
144 error = vfs_hash_get(mntp, hash, lkflags,
145 td, &nvp, newnfs_vncmpf, nfhp);
146 if (error == 0 && nvp != NULL) {
148 * I believe there is a slight chance that vgonel() could
149 * get called on this vnode between when NFSVOPLOCK() drops
150 * the VI_LOCK() and vget() acquires it again, so that it
151 * hasn't yet had v_usecount incremented. If this were to
152 * happen, the VI_DOOMED flag would be set, so check for
153 * that here. Since we now have the v_usecount incremented,
154 * we should be ok until we vrele() it, if the VI_DOOMED
155 * flag isn't set now.
158 if ((nvp->v_iflag & VI_DOOMED)) {
173 * For NFSv4, check to see if it is the same name and
174 * replace the name, if it is different.
177 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
178 nvp->v_type == VREG &&
179 (np->n_v4->n4_namelen != cnp->cn_namelen ||
180 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
182 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
183 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
184 dnp->n_fhp->nfh_len))) {
186 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
187 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
189 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
190 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
191 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
193 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
194 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
195 dnp->n_fhp->nfh_len))) {
199 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
200 np->n_v4->n4_namelen = cnp->cn_namelen;
201 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
202 dnp->n_fhp->nfh_len);
203 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
209 free(newd, M_NFSV4NODE);
211 free(oldd, M_NFSV4NODE);
216 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
218 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
220 uma_zfree(newnfsnode_zone, np);
225 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
226 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
230 * Initialize the mutex even if the vnode is going to be a loser.
231 * This simplifies the logic in reclaim, which can then unconditionally
232 * destroy the mutex (in the case of the loser, or if hash_insert
233 * happened to return an error no special casing is needed).
235 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
236 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
240 * Are we getting the root? If so, make sure the vnode flags
243 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
244 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
245 if (vp->v_type == VNON)
247 vp->v_vflag |= VV_ROOT;
252 * For NFSv4, we have to attach the directory file handle and
253 * file name, so that Open Ops can be done later.
255 if (nmp->nm_flag & NFSMNT_NFSV4) {
256 np->n_v4 = malloc(sizeof (struct nfsv4node)
257 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
259 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
260 np->n_v4->n4_namelen = cnp->cn_namelen;
261 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
262 dnp->n_fhp->nfh_len);
263 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
270 * NFS supports recursive and shared locking.
272 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
275 error = insmntque(vp, mntp);
278 mtx_destroy(&np->n_mtx);
279 lockdestroy(&np->n_excl);
281 if (np->n_v4 != NULL)
282 free(np->n_v4, M_NFSV4NODE);
283 uma_zfree(newnfsnode_zone, np);
286 error = vfs_hash_insert(vp, hash, lkflags,
287 td, &nvp, newnfs_vncmpf, nfhp);
292 /* vfs_hash_insert() vput()'s the losing vnode */
301 * Another variant of nfs_nget(). This one is only used by reopen. It
302 * takes almost the same args as nfs_nget(), but only succeeds if an entry
303 * exists in the cache. (Since files should already be "open" with a
304 * vnode ref cnt on the node when reopen calls this, it should always
306 * Also, don't get a vnode lock, since it may already be locked by some
307 * other process that is handling it. This is ok, since all other threads
308 * on the client are blocked by the nfsc_lock being exclusively held by the
309 * caller of this function.
312 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
313 struct thread *td, struct nfsnode **npp)
321 /* For forced dismounts, just return error. */
322 if (NFSCL_FORCEDISM(mntp))
324 nfhp = malloc(sizeof (struct nfsfh) + fhsize,
326 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
327 nfhp->nfh_len = fhsize;
329 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
332 * First, try to get the vnode locked, but don't block for the lock.
334 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
335 newnfs_vncmpf, nfhp);
336 if (error == 0 && nvp != NULL) {
337 NFSVOPUNLOCK(nvp, 0);
338 } else if (error == EBUSY) {
340 * It is safe so long as a vflush() with
341 * FORCECLOSE has not been done. Since the Renew thread is
342 * stopped and the MNTK_UNMOUNTF flag is set before doing
343 * a vflush() with FORCECLOSE, we should be ok here.
345 if (NFSCL_FORCEDISM(mntp))
348 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
351 } else if ((nvp->v_iflag & VI_DOOMED) != 0) {
370 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
371 struct nfsvattr *newnap)
375 if (ncl_fileid_maxwarnings >= 0 &&
376 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
379 if (ncl_fileid_maxwarnings >= 0) {
380 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
384 printf("newnfs: server '%s' error: fileid changed. "
385 "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
386 "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
387 nmp->nm_com.nmcom_hostname,
388 (uintmax_t)nmp->nm_fsid[0],
389 (uintmax_t)nmp->nm_fsid[1],
390 (uintmax_t)oldnap->na_fileid,
391 (uintmax_t)newnap->na_fileid);
394 printf("newnfs: Logged %d times about fileid corruption; "
395 "going quiet to avoid spamming logs excessively. (Limit "
396 "is: %d).\n", ncl_fileid_nwarnings,
397 ncl_fileid_maxwarnings);
401 * Load the attribute cache (that lives in the nfsnode entry) with
402 * the attributes of the second argument and
404 * copy the attributes to *vaper
405 * Similar to nfs_loadattrcache(), except the attributes are passed in
406 * instead of being parsed out of the mbuf list.
409 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
410 void *stuff, int writeattr, int dontshrink)
412 struct vnode *vp = *vpp;
413 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
415 struct nfsmount *nmp;
416 struct timespec mtime_save;
418 int setnsize, error, force_fid_err;
425 * If v_type == VNON it is a new node, so fill in the v_type,
426 * n_mtime fields. Check to see if it represents a special
427 * device, and if so, check for a possible alias. Once the
428 * correct vnode has been obtained, fill in the rest of the
433 if (vp->v_type != nvap->va_type) {
434 vp->v_type = nvap->va_type;
435 if (vp->v_type == VFIFO)
436 vp->v_op = &newnfs_fifoops;
437 np->n_mtime = nvap->va_mtime;
439 nmp = VFSTONFS(vp->v_mount);
440 vap = &np->n_vattr.na_vattr;
441 mtime_save = vap->va_mtime;
443 np->n_vattr.na_filerev = nap->na_filerev;
444 np->n_vattr.na_size = nap->na_size;
445 np->n_vattr.na_mtime = nap->na_mtime;
446 np->n_vattr.na_ctime = nap->na_ctime;
447 np->n_vattr.na_fsid = nap->na_fsid;
448 np->n_vattr.na_mode = nap->na_mode;
451 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
454 * BROKEN NFS SERVER OR MIDDLEWARE
456 * Certain NFS servers (certain old proprietary filers ca.
457 * 2006) or broken middleboxes (e.g. WAN accelerator products)
458 * will respond to GETATTR requests with results for a
461 * The WAN accelerator we've observed not only serves stale
462 * cache results for a given file, it also occasionally serves
463 * results for wholly different files. This causes surprising
464 * problems; for example the cached size attribute of a file
465 * may truncate down and then back up, resulting in zero
466 * regions in file contents read by applications. We observed
467 * this reliably with Clang and .c files during parallel build.
468 * A pcap revealed packet fragmentation and GETATTR RPC
469 * responses with wholly wrong fileids.
471 if ((np->n_vattr.na_fileid != 0 &&
472 np->n_vattr.na_fileid != nap->na_fileid) ||
474 nfscl_warn_fileid(nmp, &np->n_vattr, nap);
478 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
479 sizeof (struct nfsvattr));
483 * For NFSv4, if the node's fsid is not equal to the mount point's
484 * fsid, return the low order 32bits of the node's fsid. This
485 * allows getcwd(3) to work. There is a chance that the fsid might
486 * be the same as a local fs, but since this is in an NFS mount
487 * point, I don't think that will cause any problems?
489 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
490 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
491 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
493 * va_fsid needs to be set to some value derived from
494 * np->n_vattr.na_filesid that is not equal
495 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
496 * from the value used for the top level server volume
497 * in the mounted subtree.
500 if ((uint32_t)vap->va_fsid == np->n_vattr.na_filesid[0])
501 vap->va_fsid = hash32_buf(
502 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
505 np->n_attrstamp = time_second;
506 if (vap->va_size != np->n_size) {
507 if (vap->va_type == VREG) {
508 if (dontshrink && vap->va_size < np->n_size) {
510 * We've been told not to shrink the file;
511 * zero np->n_attrstamp to indicate that
512 * the attributes are stale.
514 vap->va_size = np->n_size;
516 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
517 vnode_pager_setsize(vp, np->n_size);
518 } else if (np->n_flag & NMODIFIED) {
520 * We've modified the file: Use the larger
521 * of our size, and the server's size.
523 if (vap->va_size < np->n_size) {
524 vap->va_size = np->n_size;
526 np->n_size = vap->va_size;
527 np->n_flag |= NSIZECHANGED;
529 vnode_pager_setsize(vp, np->n_size);
530 } else if (vap->va_size < np->n_size) {
532 * When shrinking the size, the call to
533 * vnode_pager_setsize() cannot be done
534 * with the mutex held, so delay it until
535 * after the mtx_unlock call.
537 nsize = np->n_size = vap->va_size;
538 np->n_flag |= NSIZECHANGED;
541 np->n_size = vap->va_size;
542 np->n_flag |= NSIZECHANGED;
543 vnode_pager_setsize(vp, np->n_size);
546 np->n_size = vap->va_size;
550 * The following checks are added to prevent a race between (say)
551 * a READDIR+ and a WRITE.
552 * READDIR+, WRITE requests sent out.
553 * READDIR+ resp, WRITE resp received on client.
554 * However, the WRITE resp was handled before the READDIR+ resp
555 * causing the post op attrs from the write to be loaded first
556 * and the attrs from the READDIR+ to be loaded later. If this
557 * happens, we have stale attrs loaded into the attrcache.
558 * We detect this by for the mtime moving back. We invalidate the
559 * attrcache when this happens.
561 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
562 /* Size changed or mtime went backwards */
564 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
567 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
568 if (np->n_flag & NCHG) {
569 if (np->n_flag & NACC)
570 vaper->va_atime = np->n_atim;
571 if (np->n_flag & NUPD)
572 vaper->va_mtime = np->n_mtim;
578 if (np->n_attrstamp != 0)
579 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
583 vnode_pager_setsize(vp, nsize);
588 * Fill in the client id name. For these bytes:
589 * 1 - they must be unique
590 * 2 - they should be persistent across client reboots
591 * 1 is more critical than 2
592 * Use the mount point's unique id plus either the uuid or, if that
593 * isn't set, random junk.
596 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
601 * First, put in the 64bit mount point identifier.
603 if (idlen >= sizeof (u_int64_t)) {
604 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
605 cp += sizeof (u_int64_t);
606 idlen -= sizeof (u_int64_t);
610 * If uuid is non-zero length, use it.
612 uuidlen = strlen(uuid);
613 if (uuidlen > 0 && idlen >= uuidlen) {
614 NFSBCOPY(uuid, cp, uuidlen);
620 * This only normally happens if the uuid isn't set.
623 *cp++ = (u_int8_t)(arc4random() % 256);
629 * Fill in a lock owner name. For now, pid + the process's creation time.
632 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
641 /* Return the single open_owner of all 0 bytes. */
642 bzero(cp, NFSV4CL_LOCKNAMELEN);
645 if ((flags & F_POSIX) != 0) {
646 p = (struct proc *)id;
652 tl.lval = p->p_stats->p_start.tv_sec;
657 tl.lval = p->p_stats->p_start.tv_usec;
662 } else if ((flags & F_FLOCK) != 0) {
663 bcopy(&id, cp, sizeof(id));
664 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
666 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
667 bzero(cp, NFSV4CL_LOCKNAMELEN);
672 * Find the parent process for the thread passed in as an argument.
673 * If none exists, return NULL, otherwise return a thread for the parent.
674 * (Can be any of the threads, since it is only used for td->td_proc.)
677 nfscl_getparent(struct thread *td)
690 ptd = TAILQ_FIRST(&p->p_threads);
695 * Start up the renew kernel thread.
698 start_nfscl(void *arg)
700 struct nfsclclient *clp;
703 clp = (struct nfsclclient *)arg;
704 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
705 nfscl_renewthread(clp, td);
710 nfscl_start_renewthread(struct nfsclclient *clp)
713 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
719 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
720 * as the first Op after PutFH.
721 * (For NFSv4, the postop attributes are after the Op, so they can't be
722 * parsed here. A separate call to nfscl_postop_attr() is required.)
725 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
726 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
729 struct nfsnode *np = VTONFS(vp);
730 struct nfsvattr nfsva;
733 if (wccflagp != NULL)
735 if (nd->nd_flag & ND_NFSV3) {
737 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
738 if (*tl == newnfs_true) {
739 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
740 if (wccflagp != NULL) {
741 mtx_lock(&np->n_mtx);
742 *wccflagp = (np->n_mtime.tv_sec ==
743 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
744 np->n_mtime.tv_nsec ==
745 fxdr_unsigned(u_int32_t, *(tl + 3)));
746 mtx_unlock(&np->n_mtx);
749 error = nfscl_postop_attr(nd, nap, flagp, stuff);
750 if (wccflagp != NULL && *flagp == 0)
752 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
753 == (ND_NFSV4 | ND_V4WCCATTR)) {
754 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
755 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
756 NULL, NULL, NULL, NULL, NULL);
760 * Get rid of Op# and status for next op.
762 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
764 nd->nd_flag |= ND_NOMOREDATA;
765 if (wccflagp != NULL &&
766 nfsva.na_vattr.va_mtime.tv_sec != 0) {
767 mtx_lock(&np->n_mtx);
768 *wccflagp = (np->n_mtime.tv_sec ==
769 nfsva.na_vattr.va_mtime.tv_sec &&
770 np->n_mtime.tv_nsec ==
771 nfsva.na_vattr.va_mtime.tv_sec);
772 mtx_unlock(&np->n_mtx);
780 * Get postop attributes.
783 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
790 if (nd->nd_flag & ND_NOMOREDATA)
792 if (nd->nd_flag & ND_NFSV3) {
793 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
794 *retp = fxdr_unsigned(int, *tl);
795 } else if (nd->nd_flag & ND_NFSV4) {
797 * For NFSv4, the postop attr are at the end, so no point
798 * in looking if nd_repstat != 0.
800 if (!nd->nd_repstat) {
801 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
803 /* should never happen since nd_repstat != 0 */
804 nd->nd_flag |= ND_NOMOREDATA;
808 } else if (!nd->nd_repstat) {
809 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
813 error = nfsm_loadattr(nd, nap);
822 * nfscl_request() - mostly a wrapper for newnfs_request().
825 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
826 struct ucred *cred, void *stuff)
829 struct nfsmount *nmp;
831 nmp = VFSTONFS(vp->v_mount);
832 if (nd->nd_flag & ND_NFSV4)
834 else if (nd->nd_flag & ND_NFSV3)
838 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
839 NFS_PROG, vers, NULL, 1, NULL, NULL);
844 * fill in this bsden's variant of statfs using nfsstatfs.
847 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
849 struct statfs *sbp = (struct statfs *)statfs;
851 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
852 sbp->f_bsize = NFS_FABLKSIZE;
853 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
854 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
856 * Although sf_abytes is uint64_t and f_bavail is int64_t,
857 * the value after dividing by NFS_FABLKSIZE is small
858 * enough that it will fit in 63bits, so it is ok to
859 * assign it to f_bavail without fear that it will become
862 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
863 sbp->f_files = sfp->sf_tfiles;
864 /* Since f_ffree is int64_t, clip it to 63bits. */
865 if (sfp->sf_ffiles > INT64_MAX)
866 sbp->f_ffree = INT64_MAX;
868 sbp->f_ffree = sfp->sf_ffiles;
869 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
871 * The type casts to (int32_t) ensure that this code is
872 * compatible with the old NFS client, in that it will
873 * propagate bit31 to the high order bits. This may or may
874 * not be correct for NFSv2, but since it is a legacy
875 * environment, I'd rather retain backwards compatibility.
877 sbp->f_bsize = (int32_t)sfp->sf_bsize;
878 sbp->f_blocks = (int32_t)sfp->sf_blocks;
879 sbp->f_bfree = (int32_t)sfp->sf_bfree;
880 sbp->f_bavail = (int32_t)sfp->sf_bavail;
887 * Use the fsinfo stuff to update the mount point.
890 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
893 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
894 fsp->fs_wtpref >= NFS_FABLKSIZE)
895 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
896 ~(NFS_FABLKSIZE - 1);
897 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
898 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
899 if (nmp->nm_wsize == 0)
900 nmp->nm_wsize = fsp->fs_wtmax;
902 if (nmp->nm_wsize < NFS_FABLKSIZE)
903 nmp->nm_wsize = NFS_FABLKSIZE;
904 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
905 fsp->fs_rtpref >= NFS_FABLKSIZE)
906 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
907 ~(NFS_FABLKSIZE - 1);
908 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
909 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
910 if (nmp->nm_rsize == 0)
911 nmp->nm_rsize = fsp->fs_rtmax;
913 if (nmp->nm_rsize < NFS_FABLKSIZE)
914 nmp->nm_rsize = NFS_FABLKSIZE;
915 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
916 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
917 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
918 ~(NFS_DIRBLKSIZ - 1);
919 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
920 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
921 if (nmp->nm_readdirsize == 0)
922 nmp->nm_readdirsize = fsp->fs_rtmax;
924 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
925 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
926 if (fsp->fs_maxfilesize > 0 &&
927 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
928 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
929 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
930 nmp->nm_state |= NFSSTA_GOTFSINFO;
934 * Lookups source address which should be used to communicate with
935 * @nmp and stores it inside @pdst.
937 * Returns 0 on success.
940 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
942 #if defined(INET6) || defined(INET)
945 fibnum = curthread->td_proc->p_fibnum;
948 if (nmp->nm_nam->sa_family == AF_INET) {
949 struct sockaddr_in *sin;
950 struct nhop4_extended nh_ext;
952 sin = (struct sockaddr_in *)nmp->nm_nam;
953 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
954 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
960 if (IN_LOOPBACK(ntohl(nh_ext.nh_src.s_addr))) {
961 /* Ignore loopback addresses */
966 *((struct in_addr *)paddr) = nh_ext.nh_src;
968 return (u_int8_t *)paddr;
972 if (nmp->nm_nam->sa_family == AF_INET6) {
973 struct sockaddr_in6 *sin6;
975 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
977 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
978 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
979 sin6->sin6_scope_id, NULL, paddr, NULL);
984 if (IN6_IS_ADDR_LOOPBACK(paddr))
987 /* Scope is embedded in */
990 return (u_int8_t *)paddr;
997 * Copy NFS uid, gids from the cred structure.
1000 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1004 KASSERT(cr->cr_ngroups >= 0,
1005 ("newnfs_copyincred: negative cr_ngroups"));
1006 nfscr->nfsc_uid = cr->cr_uid;
1007 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1008 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1009 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1014 * Do any client specific initialization.
1019 static int inited = 0;
1025 ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1029 * Check each of the attributes to be set, to ensure they aren't already
1030 * the correct value. Disable setting ones already correct.
1033 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1036 if (vap->va_mode != (mode_t)VNOVAL) {
1037 if (vap->va_mode == nvap->na_mode)
1038 vap->va_mode = (mode_t)VNOVAL;
1040 if (vap->va_uid != (uid_t)VNOVAL) {
1041 if (vap->va_uid == nvap->na_uid)
1042 vap->va_uid = (uid_t)VNOVAL;
1044 if (vap->va_gid != (gid_t)VNOVAL) {
1045 if (vap->va_gid == nvap->na_gid)
1046 vap->va_gid = (gid_t)VNOVAL;
1048 if (vap->va_size != VNOVAL) {
1049 if (vap->va_size == nvap->na_size)
1050 vap->va_size = VNOVAL;
1054 * We are normally called with only a partially initialized
1055 * VAP. Since the NFSv3 spec says that server may use the
1056 * file attributes to store the verifier, the spec requires
1057 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1058 * in atime, but we can't really assume that all servers will
1059 * so we ensure that our SETATTR sets both atime and mtime.
1060 * Set the VA_UTIMES_NULL flag for this case, so that
1061 * the server's time will be used. This is needed to
1062 * work around a bug in some Solaris servers, where
1063 * setting the time TOCLIENT causes the Setattr RPC
1064 * to return NFS_OK, but not set va_mode.
1066 if (vap->va_mtime.tv_sec == VNOVAL) {
1067 vfs_timestamp(&vap->va_mtime);
1068 vap->va_vaflags |= VA_UTIMES_NULL;
1070 if (vap->va_atime.tv_sec == VNOVAL)
1071 vap->va_atime = vap->va_mtime;
1076 * Map nfsv4 errors to errno.h errors.
1077 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1078 * error should only be returned for the Open, Create and Setattr Ops.
1079 * As such, most calls can just pass in 0 for those arguments.
1082 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1086 if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1093 case NFSERR_BADOWNER:
1094 tprintf(p, LOG_INFO,
1095 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1098 case NFSERR_BADNAME:
1099 case NFSERR_BADCHAR:
1100 printf("nfsv4 char/name not handled by server\n");
1102 case NFSERR_STALECLIENTID:
1103 case NFSERR_STALESTATEID:
1104 case NFSERR_EXPIRED:
1105 case NFSERR_BADSTATEID:
1106 case NFSERR_BADSESSION:
1107 printf("nfsv4 recover err returned %d\n", error);
1109 case NFSERR_BADHANDLE:
1110 case NFSERR_SERVERFAULT:
1111 case NFSERR_BADTYPE:
1112 case NFSERR_FHEXPIRED:
1113 case NFSERR_RESOURCE:
1115 case NFSERR_NOFILEHANDLE:
1116 case NFSERR_MINORVERMISMATCH:
1117 case NFSERR_OLDSTATEID:
1118 case NFSERR_BADSEQID:
1119 case NFSERR_LEASEMOVED:
1120 case NFSERR_RECLAIMBAD:
1122 case NFSERR_OPILLEGAL:
1123 printf("nfsv4 client/server protocol prob err=%d\n",
1127 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1133 * Check to see if the process for this owner exists. Return 1 if it doesn't
1137 nfscl_procdoesntexist(u_int8_t *own)
1147 /* For the single open_owner of all 0 bytes, just return 0. */
1148 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1151 if (i == NFSV4CL_LOCKNAMELEN)
1154 tl.cval[0] = *own++;
1155 tl.cval[1] = *own++;
1156 tl.cval[2] = *own++;
1157 tl.cval[3] = *own++;
1159 p = pfind_any_locked(pid);
1162 if (p->p_stats == NULL) {
1166 tl.cval[0] = *own++;
1167 tl.cval[1] = *own++;
1168 tl.cval[2] = *own++;
1169 tl.cval[3] = *own++;
1170 if (tl.lval != p->p_stats->p_start.tv_sec) {
1173 tl.cval[0] = *own++;
1174 tl.cval[1] = *own++;
1175 tl.cval[2] = *own++;
1177 if (tl.lval != p->p_stats->p_start.tv_usec)
1185 * - nfs pseudo system call for the client
1191 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1194 struct nfscbd_args nfscbdarg;
1195 struct nfsd_nfscbd_args nfscbdarg2;
1196 struct nameidata nd;
1197 struct nfscl_dumpmntopts dumpmntopts;
1198 cap_rights_t rights;
1202 struct nfsmount *nmp;
1204 if (uap->flag & NFSSVC_CBADDSOCK) {
1205 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1209 * Since we don't know what rights might be required,
1210 * pretend that we need them all. It is better to be too
1211 * careful than too reckless.
1213 error = fget(td, nfscbdarg.sock,
1214 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1217 if (fp->f_type != DTYPE_SOCKET) {
1221 error = nfscbd_addsock(fp);
1223 if (!error && nfscl_enablecallb == 0) {
1224 nfsv4_cbport = nfscbdarg.port;
1225 nfscl_enablecallb = 1;
1227 } else if (uap->flag & NFSSVC_NFSCBD) {
1228 if (uap->argp == NULL)
1230 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1231 sizeof(nfscbdarg2));
1234 error = nfscbd_nfsd(td, &nfscbdarg2);
1235 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1236 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1237 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1238 dumpmntopts.ndmnt_blen > 1024))
1241 error = nfsrv_lookupfilename(&nd,
1242 dumpmntopts.ndmnt_fname, td);
1243 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1249 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1250 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1251 dumpmntopts.ndmnt_blen);
1253 error = copyout(buf, dumpmntopts.ndmnt_buf,
1254 dumpmntopts.ndmnt_blen);
1257 } else if (uap->flag & NFSSVC_FORCEDISM) {
1258 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1259 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1262 mtx_lock(&mountlist_mtx);
1263 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1264 if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1265 0 && strcmp(mp->mnt_stat.f_fstypename,
1266 "nfs") == 0 && mp->mnt_data != NULL) {
1269 if (nfsv4_findmirror(nmp) != NULL) {
1275 mtx_lock(&nmp->nm_mtx);
1276 if ((nmp->nm_privflag &
1277 NFSMNTP_FORCEDISM) == 0) {
1279 (NFSMNTP_FORCEDISM |
1280 NFSMNTP_CANCELRPCS);
1281 mtx_unlock(&nmp->nm_mtx);
1283 mtx_unlock(&nmp->nm_mtx);
1290 mtx_unlock(&mountlist_mtx);
1294 * Call newnfs_nmcancelreqs() to cause
1295 * any RPCs in progress on the mount point to
1297 * This will cause any process waiting for an
1298 * RPC to complete while holding a vnode lock
1299 * on the mounted-on vnode (such as "df" or
1300 * a non-forced "umount") to fail.
1301 * This will unlock the mounted-on vnode so
1302 * a forced dismount can succeed.
1303 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1304 * so that nfs_unmount() can complete.
1306 newnfs_nmcancelreqs(nmp);
1307 mtx_lock(&nmp->nm_mtx);
1308 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1310 mtx_unlock(&nmp->nm_mtx);
1311 } else if (error == 0)
1321 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1324 * Called once to initialize data structures...
1327 nfscl_modevent(module_t mod, int type, void *data)
1330 static int loaded = 0;
1337 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1342 ncl_call_invalcaches = ncl_invalcaches;
1343 nfsd_call_nfscl = nfssvc_nfscl;
1348 if (nfs_numnfscbd != 0) {
1354 * XXX: Unloading of nfscl module is unsupported.
1357 ncl_call_invalcaches = NULL;
1358 nfsd_call_nfscl = NULL;
1359 uma_zdestroy(ncl_pbuf_zone);
1360 /* and get rid of the mutexes */
1361 mtx_destroy(&ncl_iod_mutex);
1373 static moduledata_t nfscl_mod = {
1378 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1380 /* So that loader and kldload(2) can find us, wherever we are.. */
1381 MODULE_VERSION(nfscl, 1);
1382 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1383 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1384 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1385 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);