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>
56 #include <net/route/nhop.h>
58 #include <fs/nfsclient/nfs_kdtrace.h>
61 dtrace_nfsclient_attrcache_flush_probe_func_t
62 dtrace_nfscl_attrcache_flush_done_probe;
63 uint32_t nfscl_attrcache_flush_done_id;
65 dtrace_nfsclient_attrcache_get_hit_probe_func_t
66 dtrace_nfscl_attrcache_get_hit_probe;
67 uint32_t nfscl_attrcache_get_hit_id;
69 dtrace_nfsclient_attrcache_get_miss_probe_func_t
70 dtrace_nfscl_attrcache_get_miss_probe;
71 uint32_t nfscl_attrcache_get_miss_id;
73 dtrace_nfsclient_attrcache_load_probe_func_t
74 dtrace_nfscl_attrcache_load_done_probe;
75 uint32_t nfscl_attrcache_load_done_id;
76 #endif /* !KDTRACE_HOOKS */
78 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
79 extern struct vop_vector newnfs_vnodeops;
80 extern struct vop_vector newnfs_fifoops;
81 extern uma_zone_t newnfsnode_zone;
82 extern struct buf_ops buf_ops_newnfs;
83 extern uma_zone_t ncl_pbuf_zone;
84 extern short nfsv4_cbport;
85 extern int nfscl_enablecallb;
86 extern int nfs_numnfscbd;
87 extern int nfscl_inited;
88 struct mtx ncl_iod_mutex;
90 extern struct mtx nfsrv_dslock_mtx;
92 extern void (*ncl_call_invalcaches)(struct vnode *);
94 SYSCTL_DECL(_vfs_nfs);
95 static int ncl_fileid_maxwarnings = 10;
96 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
97 &ncl_fileid_maxwarnings, 0,
98 "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
99 static volatile int ncl_fileid_nwarnings;
101 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
105 * Comparison function for vfs_hash functions.
108 newnfs_vncmpf(struct vnode *vp, void *arg)
110 struct nfsfh *nfhp = (struct nfsfh *)arg;
111 struct nfsnode *np = VTONFS(vp);
113 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
114 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
120 * Look up a vnode/nfsnode by file handle.
121 * Callers must check for mount points!!
122 * In all cases, a pointer to a
123 * nfsnode structure is returned.
124 * This variant takes a "struct nfsfh *" as second argument and uses
125 * that structure up, either by hanging off the nfsnode or FREEing it.
128 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
129 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
130 void *stuff, int lkflags)
132 struct nfsnode *np, *dnp;
133 struct vnode *vp, *nvp;
134 struct nfsv4node *newd, *oldd;
137 struct nfsmount *nmp;
139 nmp = VFSTONFS(mntp);
143 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
145 error = vfs_hash_get(mntp, hash, lkflags,
146 td, &nvp, newnfs_vncmpf, nfhp);
147 if (error == 0 && nvp != NULL) {
149 * I believe there is a slight chance that vgonel() could
150 * get called on this vnode between when NFSVOPLOCK() drops
151 * the VI_LOCK() and vget() acquires it again, so that it
152 * hasn't yet had v_usecount incremented. If this were to
153 * happen, the VIRF_DOOMED flag would be set, so check for
154 * that here. Since we now have the v_usecount incremented,
155 * we should be ok until we vrele() it, if the VIRF_DOOMED
156 * flag isn't set now.
159 if (VN_IS_DOOMED(nvp)) {
174 * For NFSv4, check to see if it is the same name and
175 * replace the name, if it is different.
178 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
179 nvp->v_type == VREG &&
180 (np->n_v4->n4_namelen != cnp->cn_namelen ||
181 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
183 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
184 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
185 dnp->n_fhp->nfh_len))) {
187 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
188 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
190 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
191 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
192 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
194 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
195 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
196 dnp->n_fhp->nfh_len))) {
200 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
201 np->n_v4->n4_namelen = cnp->cn_namelen;
202 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
203 dnp->n_fhp->nfh_len);
204 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
210 free(newd, M_NFSV4NODE);
212 free(oldd, M_NFSV4NODE);
217 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
219 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
221 uma_zfree(newnfsnode_zone, np);
226 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
227 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
231 * Initialize the mutex even if the vnode is going to be a loser.
232 * This simplifies the logic in reclaim, which can then unconditionally
233 * destroy the mutex (in the case of the loser, or if hash_insert
234 * happened to return an error no special casing is needed).
236 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
237 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
241 * Are we getting the root? If so, make sure the vnode flags
244 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
245 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
246 if (vp->v_type == VNON)
248 vp->v_vflag |= VV_ROOT;
251 vp->v_vflag |= VV_VMSIZEVNLOCK;
255 * For NFSv4, we have to attach the directory file handle and
256 * file name, so that Open Ops can be done later.
258 if (nmp->nm_flag & NFSMNT_NFSV4) {
259 np->n_v4 = malloc(sizeof (struct nfsv4node)
260 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
262 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
263 np->n_v4->n4_namelen = cnp->cn_namelen;
264 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
265 dnp->n_fhp->nfh_len);
266 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
273 * NFS supports recursive and shared locking.
275 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
278 error = insmntque(vp, mntp);
281 mtx_destroy(&np->n_mtx);
282 lockdestroy(&np->n_excl);
284 if (np->n_v4 != NULL)
285 free(np->n_v4, M_NFSV4NODE);
286 uma_zfree(newnfsnode_zone, np);
289 error = vfs_hash_insert(vp, hash, lkflags,
290 td, &nvp, newnfs_vncmpf, nfhp);
295 /* vfs_hash_insert() vput()'s the losing vnode */
304 * Another variant of nfs_nget(). This one is only used by reopen. It
305 * takes almost the same args as nfs_nget(), but only succeeds if an entry
306 * exists in the cache. (Since files should already be "open" with a
307 * vnode ref cnt on the node when reopen calls this, it should always
309 * Also, don't get a vnode lock, since it may already be locked by some
310 * other process that is handling it. This is ok, since all other threads
311 * on the client are blocked by the nfsc_lock being exclusively held by the
312 * caller of this function.
315 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
316 struct thread *td, struct nfsnode **npp)
324 /* For forced dismounts, just return error. */
325 if (NFSCL_FORCEDISM(mntp))
327 nfhp = malloc(sizeof (struct nfsfh) + fhsize,
329 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
330 nfhp->nfh_len = fhsize;
332 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
335 * First, try to get the vnode locked, but don't block for the lock.
337 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
338 newnfs_vncmpf, nfhp);
339 if (error == 0 && nvp != NULL) {
341 } else if (error == EBUSY) {
343 * It is safe so long as a vflush() with
344 * FORCECLOSE has not been done. Since the Renew thread is
345 * stopped and the MNTK_UNMOUNTF flag is set before doing
346 * a vflush() with FORCECLOSE, we should be ok here.
348 if (NFSCL_FORCEDISM(mntp))
351 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
354 } else if (VN_IS_DOOMED(nvp)) {
373 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
374 struct nfsvattr *newnap)
378 if (ncl_fileid_maxwarnings >= 0 &&
379 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
382 if (ncl_fileid_maxwarnings >= 0) {
383 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
387 printf("newnfs: server '%s' error: fileid changed. "
388 "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
389 "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
390 nmp->nm_com.nmcom_hostname,
391 (uintmax_t)nmp->nm_fsid[0],
392 (uintmax_t)nmp->nm_fsid[1],
393 (uintmax_t)oldnap->na_fileid,
394 (uintmax_t)newnap->na_fileid);
397 printf("newnfs: Logged %d times about fileid corruption; "
398 "going quiet to avoid spamming logs excessively. (Limit "
399 "is: %d).\n", ncl_fileid_nwarnings,
400 ncl_fileid_maxwarnings);
404 * Load the attribute cache (that lives in the nfsnode entry) with
405 * the attributes of the second argument and
407 * copy the attributes to *vaper
408 * Similar to nfs_loadattrcache(), except the attributes are passed in
409 * instead of being parsed out of the mbuf list.
412 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
413 void *stuff, int writeattr, int dontshrink)
415 struct vnode *vp = *vpp;
416 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
418 struct nfsmount *nmp;
419 struct timespec mtime_save;
420 int 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 } else if (np->n_flag & NMODIFIED) {
519 * We've modified the file: Use the larger
520 * of our size, and the server's size.
522 if (vap->va_size < np->n_size) {
523 vap->va_size = np->n_size;
525 np->n_size = vap->va_size;
526 np->n_flag |= NSIZECHANGED;
529 np->n_size = vap->va_size;
530 np->n_flag |= NSIZECHANGED;
533 np->n_size = vap->va_size;
537 * The following checks are added to prevent a race between (say)
538 * a READDIR+ and a WRITE.
539 * READDIR+, WRITE requests sent out.
540 * READDIR+ resp, WRITE resp received on client.
541 * However, the WRITE resp was handled before the READDIR+ resp
542 * causing the post op attrs from the write to be loaded first
543 * and the attrs from the READDIR+ to be loaded later. If this
544 * happens, we have stale attrs loaded into the attrcache.
545 * We detect this by for the mtime moving back. We invalidate the
546 * attrcache when this happens.
548 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
549 /* Size changed or mtime went backwards */
551 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
554 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
555 if (np->n_flag & NCHG) {
556 if (np->n_flag & NACC)
557 vaper->va_atime = np->n_atim;
558 if (np->n_flag & NUPD)
559 vaper->va_mtime = np->n_mtim;
565 if (np->n_attrstamp != 0)
566 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
568 (void)ncl_pager_setsize(vp, NULL);
573 * Call vnode_pager_setsize() if the size of the node changed, as
574 * recorded in nfsnode vs. v_object, or delay the call if notifying
575 * the pager is not possible at the moment.
577 * If nsizep is non-NULL, the call is delayed and the new node size is
578 * provided. Caller should itself call vnode_pager_setsize() if
579 * function returned true. If nsizep is NULL, function tries to call
580 * vnode_pager_setsize() itself if needed and possible, and the nfs
581 * node is unlocked unconditionally, the return value is not useful.
584 ncl_pager_setsize(struct vnode *vp, u_quad_t *nsizep)
595 vap = &np->n_vattr.na_vattr;
596 nsize = vap->va_size;
597 object = vp->v_object;
600 if (object != NULL && nsize != object->un_pager.vnp.vnp_size) {
601 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
602 (curthread->td_pflags2 & TDP2_SBPAGES) == 0)
605 np->n_flag |= NVNSETSZSKIP;
607 if (nsizep == NULL) {
610 vnode_pager_setsize(vp, nsize);
619 * Fill in the client id name. For these bytes:
620 * 1 - they must be unique
621 * 2 - they should be persistent across client reboots
622 * 1 is more critical than 2
623 * Use the mount point's unique id plus either the uuid or, if that
624 * isn't set, random junk.
627 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
632 * First, put in the 64bit mount point identifier.
634 if (idlen >= sizeof (u_int64_t)) {
635 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
636 cp += sizeof (u_int64_t);
637 idlen -= sizeof (u_int64_t);
641 * If uuid is non-zero length, use it.
643 uuidlen = strlen(uuid);
644 if (uuidlen > 0 && idlen >= uuidlen) {
645 NFSBCOPY(uuid, cp, uuidlen);
651 * This only normally happens if the uuid isn't set.
654 *cp++ = (u_int8_t)(arc4random() % 256);
660 * Fill in a lock owner name. For now, pid + the process's creation time.
663 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
672 /* Return the single open_owner of all 0 bytes. */
673 bzero(cp, NFSV4CL_LOCKNAMELEN);
676 if ((flags & F_POSIX) != 0) {
677 p = (struct proc *)id;
683 tl.lval = p->p_stats->p_start.tv_sec;
688 tl.lval = p->p_stats->p_start.tv_usec;
693 } else if ((flags & F_FLOCK) != 0) {
694 bcopy(&id, cp, sizeof(id));
695 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
697 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
698 bzero(cp, NFSV4CL_LOCKNAMELEN);
703 * Find the parent process for the thread passed in as an argument.
704 * If none exists, return NULL, otherwise return a thread for the parent.
705 * (Can be any of the threads, since it is only used for td->td_proc.)
708 nfscl_getparent(struct thread *td)
721 ptd = TAILQ_FIRST(&p->p_threads);
726 * Start up the renew kernel thread.
729 start_nfscl(void *arg)
731 struct nfsclclient *clp;
734 clp = (struct nfsclclient *)arg;
735 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
736 nfscl_renewthread(clp, td);
741 nfscl_start_renewthread(struct nfsclclient *clp)
744 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
750 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
751 * as the first Op after PutFH.
752 * (For NFSv4, the postop attributes are after the Op, so they can't be
753 * parsed here. A separate call to nfscl_postop_attr() is required.)
756 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
757 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
760 struct nfsnode *np = VTONFS(vp);
761 struct nfsvattr nfsva;
764 if (wccflagp != NULL)
766 if (nd->nd_flag & ND_NFSV3) {
768 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
769 if (*tl == newnfs_true) {
770 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
771 if (wccflagp != NULL) {
773 *wccflagp = (np->n_mtime.tv_sec ==
774 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
775 np->n_mtime.tv_nsec ==
776 fxdr_unsigned(u_int32_t, *(tl + 3)));
780 error = nfscl_postop_attr(nd, nap, flagp, stuff);
781 if (wccflagp != NULL && *flagp == 0)
783 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
784 == (ND_NFSV4 | ND_V4WCCATTR)) {
785 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
786 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
787 NULL, NULL, NULL, NULL, NULL);
791 * Get rid of Op# and status for next op.
793 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
795 nd->nd_flag |= ND_NOMOREDATA;
796 if (wccflagp != NULL &&
797 nfsva.na_vattr.va_mtime.tv_sec != 0) {
799 *wccflagp = (np->n_mtime.tv_sec ==
800 nfsva.na_vattr.va_mtime.tv_sec &&
801 np->n_mtime.tv_nsec ==
802 nfsva.na_vattr.va_mtime.tv_sec);
811 * Get postop attributes.
814 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
821 if (nd->nd_flag & ND_NOMOREDATA)
823 if (nd->nd_flag & ND_NFSV3) {
824 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
825 *retp = fxdr_unsigned(int, *tl);
826 } else if (nd->nd_flag & ND_NFSV4) {
828 * For NFSv4, the postop attr are at the end, so no point
829 * in looking if nd_repstat != 0.
831 if (!nd->nd_repstat) {
832 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
834 /* should never happen since nd_repstat != 0 */
835 nd->nd_flag |= ND_NOMOREDATA;
839 } else if (!nd->nd_repstat) {
840 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
844 error = nfsm_loadattr(nd, nap);
853 * nfscl_request() - mostly a wrapper for newnfs_request().
856 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
857 struct ucred *cred, void *stuff)
860 struct nfsmount *nmp;
862 nmp = VFSTONFS(vp->v_mount);
863 if (nd->nd_flag & ND_NFSV4)
865 else if (nd->nd_flag & ND_NFSV3)
869 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
870 NFS_PROG, vers, NULL, 1, NULL, NULL);
875 * fill in this bsden's variant of statfs using nfsstatfs.
878 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
880 struct statfs *sbp = (struct statfs *)statfs;
882 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
883 sbp->f_bsize = NFS_FABLKSIZE;
884 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
885 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
887 * Although sf_abytes is uint64_t and f_bavail is int64_t,
888 * the value after dividing by NFS_FABLKSIZE is small
889 * enough that it will fit in 63bits, so it is ok to
890 * assign it to f_bavail without fear that it will become
893 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
894 sbp->f_files = sfp->sf_tfiles;
895 /* Since f_ffree is int64_t, clip it to 63bits. */
896 if (sfp->sf_ffiles > INT64_MAX)
897 sbp->f_ffree = INT64_MAX;
899 sbp->f_ffree = sfp->sf_ffiles;
900 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
902 * The type casts to (int32_t) ensure that this code is
903 * compatible with the old NFS client, in that it will
904 * propagate bit31 to the high order bits. This may or may
905 * not be correct for NFSv2, but since it is a legacy
906 * environment, I'd rather retain backwards compatibility.
908 sbp->f_bsize = (int32_t)sfp->sf_bsize;
909 sbp->f_blocks = (int32_t)sfp->sf_blocks;
910 sbp->f_bfree = (int32_t)sfp->sf_bfree;
911 sbp->f_bavail = (int32_t)sfp->sf_bavail;
918 * Use the fsinfo stuff to update the mount point.
921 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
924 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
925 fsp->fs_wtpref >= NFS_FABLKSIZE)
926 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
927 ~(NFS_FABLKSIZE - 1);
928 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
929 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
930 if (nmp->nm_wsize == 0)
931 nmp->nm_wsize = fsp->fs_wtmax;
933 if (nmp->nm_wsize < NFS_FABLKSIZE)
934 nmp->nm_wsize = NFS_FABLKSIZE;
935 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
936 fsp->fs_rtpref >= NFS_FABLKSIZE)
937 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
938 ~(NFS_FABLKSIZE - 1);
939 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
940 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
941 if (nmp->nm_rsize == 0)
942 nmp->nm_rsize = fsp->fs_rtmax;
944 if (nmp->nm_rsize < NFS_FABLKSIZE)
945 nmp->nm_rsize = NFS_FABLKSIZE;
946 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
947 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
948 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
949 ~(NFS_DIRBLKSIZ - 1);
950 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
951 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
952 if (nmp->nm_readdirsize == 0)
953 nmp->nm_readdirsize = fsp->fs_rtmax;
955 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
956 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
957 if (fsp->fs_maxfilesize > 0 &&
958 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
959 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
960 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
961 nmp->nm_state |= NFSSTA_GOTFSINFO;
965 * Lookups source address which should be used to communicate with
966 * @nmp and stores it inside @pdst.
968 * Returns 0 on success.
971 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
973 #if defined(INET6) || defined(INET)
976 fibnum = curthread->td_proc->p_fibnum;
979 if (nmp->nm_nam->sa_family == AF_INET) {
980 struct epoch_tracker et;
981 struct nhop_object *nh;
982 struct sockaddr_in *sin;
983 struct in_addr addr = {};
985 sin = (struct sockaddr_in *)nmp->nm_nam;
987 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
988 nh = fib4_lookup(fibnum, sin->sin_addr, 0, NHR_NONE, 0);
991 addr = IA_SIN(ifatoia(nh->nh_ifa))->sin_addr;
996 if (IN_LOOPBACK(ntohl(addr.s_addr))) {
997 /* Ignore loopback addresses */
1002 *((struct in_addr *)paddr) = addr;
1004 return (u_int8_t *)paddr;
1008 if (nmp->nm_nam->sa_family == AF_INET6) {
1009 struct sockaddr_in6 *sin6;
1012 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1014 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1015 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1016 sin6->sin6_scope_id, NULL, paddr, NULL);
1021 if (IN6_IS_ADDR_LOOPBACK(paddr))
1024 /* Scope is embedded in */
1027 return (u_int8_t *)paddr;
1034 * Copy NFS uid, gids from the cred structure.
1037 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1041 KASSERT(cr->cr_ngroups >= 0,
1042 ("newnfs_copyincred: negative cr_ngroups"));
1043 nfscr->nfsc_uid = cr->cr_uid;
1044 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1045 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1046 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1051 * Do any client specific initialization.
1056 static int inited = 0;
1062 ncl_pbuf_zone = pbuf_zsecond_create("nfspbuf", nswbuf / 2);
1066 * Check each of the attributes to be set, to ensure they aren't already
1067 * the correct value. Disable setting ones already correct.
1070 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1073 if (vap->va_mode != (mode_t)VNOVAL) {
1074 if (vap->va_mode == nvap->na_mode)
1075 vap->va_mode = (mode_t)VNOVAL;
1077 if (vap->va_uid != (uid_t)VNOVAL) {
1078 if (vap->va_uid == nvap->na_uid)
1079 vap->va_uid = (uid_t)VNOVAL;
1081 if (vap->va_gid != (gid_t)VNOVAL) {
1082 if (vap->va_gid == nvap->na_gid)
1083 vap->va_gid = (gid_t)VNOVAL;
1085 if (vap->va_size != VNOVAL) {
1086 if (vap->va_size == nvap->na_size)
1087 vap->va_size = VNOVAL;
1091 * We are normally called with only a partially initialized
1092 * VAP. Since the NFSv3 spec says that server may use the
1093 * file attributes to store the verifier, the spec requires
1094 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1095 * in atime, but we can't really assume that all servers will
1096 * so we ensure that our SETATTR sets both atime and mtime.
1097 * Set the VA_UTIMES_NULL flag for this case, so that
1098 * the server's time will be used. This is needed to
1099 * work around a bug in some Solaris servers, where
1100 * setting the time TOCLIENT causes the Setattr RPC
1101 * to return NFS_OK, but not set va_mode.
1103 if (vap->va_mtime.tv_sec == VNOVAL) {
1104 vfs_timestamp(&vap->va_mtime);
1105 vap->va_vaflags |= VA_UTIMES_NULL;
1107 if (vap->va_atime.tv_sec == VNOVAL)
1108 vap->va_atime = vap->va_mtime;
1113 * Map nfsv4 errors to errno.h errors.
1114 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1115 * error should only be returned for the Open, Create and Setattr Ops.
1116 * As such, most calls can just pass in 0 for those arguments.
1119 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1123 if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1130 case NFSERR_BADOWNER:
1131 tprintf(p, LOG_INFO,
1132 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1135 case NFSERR_BADNAME:
1136 case NFSERR_BADCHAR:
1137 printf("nfsv4 char/name not handled by server\n");
1139 case NFSERR_STALECLIENTID:
1140 case NFSERR_STALESTATEID:
1141 case NFSERR_EXPIRED:
1142 case NFSERR_BADSTATEID:
1143 case NFSERR_BADSESSION:
1144 printf("nfsv4 recover err returned %d\n", error);
1146 case NFSERR_BADHANDLE:
1147 case NFSERR_SERVERFAULT:
1148 case NFSERR_BADTYPE:
1149 case NFSERR_FHEXPIRED:
1150 case NFSERR_RESOURCE:
1152 case NFSERR_NOFILEHANDLE:
1153 case NFSERR_MINORVERMISMATCH:
1154 case NFSERR_OLDSTATEID:
1155 case NFSERR_BADSEQID:
1156 case NFSERR_LEASEMOVED:
1157 case NFSERR_RECLAIMBAD:
1159 case NFSERR_OPILLEGAL:
1160 printf("nfsv4 client/server protocol prob err=%d\n",
1164 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1170 * Check to see if the process for this owner exists. Return 1 if it doesn't
1174 nfscl_procdoesntexist(u_int8_t *own)
1184 /* For the single open_owner of all 0 bytes, just return 0. */
1185 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1188 if (i == NFSV4CL_LOCKNAMELEN)
1191 tl.cval[0] = *own++;
1192 tl.cval[1] = *own++;
1193 tl.cval[2] = *own++;
1194 tl.cval[3] = *own++;
1196 p = pfind_any_locked(pid);
1199 if (p->p_stats == NULL) {
1203 tl.cval[0] = *own++;
1204 tl.cval[1] = *own++;
1205 tl.cval[2] = *own++;
1206 tl.cval[3] = *own++;
1207 if (tl.lval != p->p_stats->p_start.tv_sec) {
1210 tl.cval[0] = *own++;
1211 tl.cval[1] = *own++;
1212 tl.cval[2] = *own++;
1214 if (tl.lval != p->p_stats->p_start.tv_usec)
1222 * - nfs pseudo system call for the client
1228 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1231 struct nfscbd_args nfscbdarg;
1232 struct nfsd_nfscbd_args nfscbdarg2;
1233 struct nameidata nd;
1234 struct nfscl_dumpmntopts dumpmntopts;
1235 cap_rights_t rights;
1239 struct nfsmount *nmp;
1241 if (uap->flag & NFSSVC_CBADDSOCK) {
1242 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1246 * Since we don't know what rights might be required,
1247 * pretend that we need them all. It is better to be too
1248 * careful than too reckless.
1250 error = fget(td, nfscbdarg.sock,
1251 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1254 if (fp->f_type != DTYPE_SOCKET) {
1258 error = nfscbd_addsock(fp);
1260 if (!error && nfscl_enablecallb == 0) {
1261 nfsv4_cbport = nfscbdarg.port;
1262 nfscl_enablecallb = 1;
1264 } else if (uap->flag & NFSSVC_NFSCBD) {
1265 if (uap->argp == NULL)
1267 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1268 sizeof(nfscbdarg2));
1271 error = nfscbd_nfsd(td, &nfscbdarg2);
1272 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1273 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1274 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1275 dumpmntopts.ndmnt_blen > 1024))
1278 error = nfsrv_lookupfilename(&nd,
1279 dumpmntopts.ndmnt_fname, td);
1280 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1286 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1287 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1288 dumpmntopts.ndmnt_blen);
1290 error = copyout(buf, dumpmntopts.ndmnt_buf,
1291 dumpmntopts.ndmnt_blen);
1294 } else if (uap->flag & NFSSVC_FORCEDISM) {
1295 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1296 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1299 mtx_lock(&mountlist_mtx);
1300 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1301 if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1302 0 && strcmp(mp->mnt_stat.f_fstypename,
1303 "nfs") == 0 && mp->mnt_data != NULL) {
1306 if (nfsv4_findmirror(nmp) != NULL) {
1312 mtx_lock(&nmp->nm_mtx);
1313 if ((nmp->nm_privflag &
1314 NFSMNTP_FORCEDISM) == 0) {
1316 (NFSMNTP_FORCEDISM |
1317 NFSMNTP_CANCELRPCS);
1318 mtx_unlock(&nmp->nm_mtx);
1320 mtx_unlock(&nmp->nm_mtx);
1327 mtx_unlock(&mountlist_mtx);
1331 * Call newnfs_nmcancelreqs() to cause
1332 * any RPCs in progress on the mount point to
1334 * This will cause any process waiting for an
1335 * RPC to complete while holding a vnode lock
1336 * on the mounted-on vnode (such as "df" or
1337 * a non-forced "umount") to fail.
1338 * This will unlock the mounted-on vnode so
1339 * a forced dismount can succeed.
1340 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1341 * so that nfs_unmount() can complete.
1343 newnfs_nmcancelreqs(nmp);
1344 mtx_lock(&nmp->nm_mtx);
1345 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1347 mtx_unlock(&nmp->nm_mtx);
1348 } else if (error == 0)
1358 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1361 * Called once to initialize data structures...
1364 nfscl_modevent(module_t mod, int type, void *data)
1367 static int loaded = 0;
1374 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1379 ncl_call_invalcaches = ncl_invalcaches;
1380 nfsd_call_nfscl = nfssvc_nfscl;
1385 if (nfs_numnfscbd != 0) {
1391 * XXX: Unloading of nfscl module is unsupported.
1394 ncl_call_invalcaches = NULL;
1395 nfsd_call_nfscl = NULL;
1396 uma_zdestroy(ncl_pbuf_zone);
1397 /* and get rid of the mutexes */
1398 mtx_destroy(&ncl_iod_mutex);
1410 static moduledata_t nfscl_mod = {
1415 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1417 /* So that loader and kldload(2) can find us, wherever we are.. */
1418 MODULE_VERSION(nfscl, 1);
1419 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1420 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1421 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);