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
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_inet6.h"
40 #include <sys/capsicum.h>
43 * generally, I don't like #includes inside .h files, but it seems to
44 * be the easiest way to handle the port.
48 #include <sys/sysctl.h>
49 #include <fs/nfs/nfsport.h>
50 #include <netinet/in_fib.h>
51 #include <netinet/if_ether.h>
52 #include <netinet6/ip6_var.h>
53 #include <net/if_types.h>
55 #include <fs/nfsclient/nfs_kdtrace.h>
58 dtrace_nfsclient_attrcache_flush_probe_func_t
59 dtrace_nfscl_attrcache_flush_done_probe;
60 uint32_t nfscl_attrcache_flush_done_id;
62 dtrace_nfsclient_attrcache_get_hit_probe_func_t
63 dtrace_nfscl_attrcache_get_hit_probe;
64 uint32_t nfscl_attrcache_get_hit_id;
66 dtrace_nfsclient_attrcache_get_miss_probe_func_t
67 dtrace_nfscl_attrcache_get_miss_probe;
68 uint32_t nfscl_attrcache_get_miss_id;
70 dtrace_nfsclient_attrcache_load_probe_func_t
71 dtrace_nfscl_attrcache_load_done_probe;
72 uint32_t nfscl_attrcache_load_done_id;
73 #endif /* !KDTRACE_HOOKS */
75 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
76 extern struct vop_vector newnfs_vnodeops;
77 extern struct vop_vector newnfs_fifoops;
78 extern uma_zone_t newnfsnode_zone;
79 extern struct buf_ops buf_ops_newnfs;
80 extern int ncl_pbuf_freecnt;
81 extern short nfsv4_cbport;
82 extern int nfscl_enablecallb;
83 extern int nfs_numnfscbd;
84 extern int nfscl_inited;
85 struct mtx ncl_iod_mutex;
88 extern void (*ncl_call_invalcaches)(struct vnode *);
90 SYSCTL_DECL(_vfs_nfs);
91 static int ncl_fileid_maxwarnings = 10;
92 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
93 &ncl_fileid_maxwarnings, 0,
94 "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
95 static volatile int ncl_fileid_nwarnings;
97 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
101 * Comparison function for vfs_hash functions.
104 newnfs_vncmpf(struct vnode *vp, void *arg)
106 struct nfsfh *nfhp = (struct nfsfh *)arg;
107 struct nfsnode *np = VTONFS(vp);
109 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
110 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
116 * Look up a vnode/nfsnode by file handle.
117 * Callers must check for mount points!!
118 * In all cases, a pointer to a
119 * nfsnode structure is returned.
120 * This variant takes a "struct nfsfh *" as second argument and uses
121 * that structure up, either by hanging off the nfsnode or FREEing it.
124 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
125 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
126 void *stuff, int lkflags)
128 struct nfsnode *np, *dnp;
129 struct vnode *vp, *nvp;
130 struct nfsv4node *newd, *oldd;
133 struct nfsmount *nmp;
135 nmp = VFSTONFS(mntp);
139 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
141 error = vfs_hash_get(mntp, hash, lkflags,
142 td, &nvp, newnfs_vncmpf, nfhp);
143 if (error == 0 && nvp != NULL) {
145 * I believe there is a slight chance that vgonel() could
146 * get called on this vnode between when NFSVOPLOCK() drops
147 * the VI_LOCK() and vget() acquires it again, so that it
148 * hasn't yet had v_usecount incremented. If this were to
149 * happen, the VI_DOOMED flag would be set, so check for
150 * that here. Since we now have the v_usecount incremented,
151 * we should be ok until we vrele() it, if the VI_DOOMED
152 * flag isn't set now.
155 if ((nvp->v_iflag & VI_DOOMED)) {
164 FREE((caddr_t)nfhp, M_NFSFH);
170 * For NFSv4, check to see if it is the same name and
171 * replace the name, if it is different.
174 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
175 nvp->v_type == VREG &&
176 (np->n_v4->n4_namelen != cnp->cn_namelen ||
177 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
179 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
180 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
181 dnp->n_fhp->nfh_len))) {
182 MALLOC(newd, struct nfsv4node *,
183 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
184 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
186 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
187 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
188 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
190 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
191 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
192 dnp->n_fhp->nfh_len))) {
196 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
197 np->n_v4->n4_namelen = cnp->cn_namelen;
198 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
199 dnp->n_fhp->nfh_len);
200 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
206 FREE((caddr_t)newd, M_NFSV4NODE);
208 FREE((caddr_t)oldd, M_NFSV4NODE);
210 FREE((caddr_t)nfhp, M_NFSFH);
213 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
215 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
217 uma_zfree(newnfsnode_zone, np);
218 FREE((caddr_t)nfhp, M_NFSFH);
222 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
223 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
227 * Initialize the mutex even if the vnode is going to be a loser.
228 * This simplifies the logic in reclaim, which can then unconditionally
229 * destroy the mutex (in the case of the loser, or if hash_insert
230 * happened to return an error no special casing is needed).
232 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
235 * Are we getting the root? If so, make sure the vnode flags
238 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
239 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
240 if (vp->v_type == VNON)
242 vp->v_vflag |= VV_ROOT;
247 * For NFSv4, we have to attach the directory file handle and
248 * file name, so that Open Ops can be done later.
250 if (nmp->nm_flag & NFSMNT_NFSV4) {
251 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
252 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
254 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
255 np->n_v4->n4_namelen = cnp->cn_namelen;
256 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
257 dnp->n_fhp->nfh_len);
258 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
265 * NFS supports recursive and shared locking.
267 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
270 error = insmntque(vp, mntp);
273 mtx_destroy(&np->n_mtx);
274 FREE((caddr_t)nfhp, M_NFSFH);
275 if (np->n_v4 != NULL)
276 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
277 uma_zfree(newnfsnode_zone, np);
280 error = vfs_hash_insert(vp, hash, lkflags,
281 td, &nvp, newnfs_vncmpf, nfhp);
286 /* vfs_hash_insert() vput()'s the losing vnode */
295 * Another variant of nfs_nget(). This one is only used by reopen. It
296 * takes almost the same args as nfs_nget(), but only succeeds if an entry
297 * exists in the cache. (Since files should already be "open" with a
298 * vnode ref cnt on the node when reopen calls this, it should always
300 * Also, don't get a vnode lock, since it may already be locked by some
301 * other process that is handling it. This is ok, since all other threads
302 * on the client are blocked by the nfsc_lock being exclusively held by the
303 * caller of this function.
306 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
307 struct thread *td, struct nfsnode **npp)
315 /* For forced dismounts, just return error. */
316 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
318 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
320 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
321 nfhp->nfh_len = fhsize;
323 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
326 * First, try to get the vnode locked, but don't block for the lock.
328 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
329 newnfs_vncmpf, nfhp);
330 if (error == 0 && nvp != NULL) {
331 NFSVOPUNLOCK(nvp, 0);
332 } else if (error == EBUSY) {
334 * It is safe so long as a vflush() with
335 * FORCECLOSE has not been done. Since the Renew thread is
336 * stopped and the MNTK_UNMOUNTF flag is set before doing
337 * a vflush() with FORCECLOSE, we should be ok here.
339 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
342 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
345 } else if ((nvp->v_iflag & VI_DOOMED) != 0) {
364 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
365 struct nfsvattr *newnap)
369 if (ncl_fileid_maxwarnings >= 0 &&
370 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
373 if (ncl_fileid_maxwarnings >= 0) {
374 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
378 printf("newnfs: server '%s' error: fileid changed. "
379 "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
380 "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
381 nmp->nm_com.nmcom_hostname,
382 (uintmax_t)nmp->nm_fsid[0],
383 (uintmax_t)nmp->nm_fsid[1],
384 (uintmax_t)oldnap->na_fileid,
385 (uintmax_t)newnap->na_fileid);
388 printf("newnfs: Logged %d times about fileid corruption; "
389 "going quiet to avoid spamming logs excessively. (Limit "
390 "is: %d).\n", ncl_fileid_nwarnings,
391 ncl_fileid_maxwarnings);
395 * Load the attribute cache (that lives in the nfsnode entry) with
396 * the attributes of the second argument and
398 * copy the attributes to *vaper
399 * Similar to nfs_loadattrcache(), except the attributes are passed in
400 * instead of being parsed out of the mbuf list.
403 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
404 void *stuff, int writeattr, int dontshrink)
406 struct vnode *vp = *vpp;
407 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
409 struct nfsmount *nmp;
410 struct timespec mtime_save;
412 int setnsize, error, force_fid_err;
419 * If v_type == VNON it is a new node, so fill in the v_type,
420 * n_mtime fields. Check to see if it represents a special
421 * device, and if so, check for a possible alias. Once the
422 * correct vnode has been obtained, fill in the rest of the
427 if (vp->v_type != nvap->va_type) {
428 vp->v_type = nvap->va_type;
429 if (vp->v_type == VFIFO)
430 vp->v_op = &newnfs_fifoops;
431 np->n_mtime = nvap->va_mtime;
433 nmp = VFSTONFS(vp->v_mount);
434 vap = &np->n_vattr.na_vattr;
435 mtime_save = vap->va_mtime;
437 np->n_vattr.na_filerev = nap->na_filerev;
438 np->n_vattr.na_size = nap->na_size;
439 np->n_vattr.na_mtime = nap->na_mtime;
440 np->n_vattr.na_ctime = nap->na_ctime;
441 np->n_vattr.na_fsid = nap->na_fsid;
442 np->n_vattr.na_mode = nap->na_mode;
445 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
448 * BROKEN NFS SERVER OR MIDDLEWARE
450 * Certain NFS servers (certain old proprietary filers ca.
451 * 2006) or broken middleboxes (e.g. WAN accelerator products)
452 * will respond to GETATTR requests with results for a
455 * The WAN accelerator we've observed not only serves stale
456 * cache results for a given file, it also occasionally serves
457 * results for wholly different files. This causes surprising
458 * problems; for example the cached size attribute of a file
459 * may truncate down and then back up, resulting in zero
460 * regions in file contents read by applications. We observed
461 * this reliably with Clang and .c files during parallel build.
462 * A pcap revealed packet fragmentation and GETATTR RPC
463 * responses with wholly wrong fileids.
465 if ((np->n_vattr.na_fileid != 0 &&
466 np->n_vattr.na_fileid != nap->na_fileid) ||
468 nfscl_warn_fileid(nmp, &np->n_vattr, nap);
472 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
473 sizeof (struct nfsvattr));
477 * For NFSv4, if the node's fsid is not equal to the mount point's
478 * fsid, return the low order 32bits of the node's fsid. This
479 * allows getcwd(3) to work. There is a chance that the fsid might
480 * be the same as a local fs, but since this is in an NFS mount
481 * point, I don't think that will cause any problems?
483 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
484 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
485 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
487 * va_fsid needs to be set to some value derived from
488 * np->n_vattr.na_filesid that is not equal
489 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
490 * from the value used for the top level server volume
491 * in the mounted subtree.
493 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
494 (uint32_t)np->n_vattr.na_filesid[0])
495 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
497 vap->va_fsid = (uint32_t)hash32_buf(
498 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
500 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
501 np->n_attrstamp = time_second;
502 if (vap->va_size != np->n_size) {
503 if (vap->va_type == VREG) {
504 if (dontshrink && vap->va_size < np->n_size) {
506 * We've been told not to shrink the file;
507 * zero np->n_attrstamp to indicate that
508 * the attributes are stale.
510 vap->va_size = np->n_size;
512 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
513 vnode_pager_setsize(vp, np->n_size);
514 } else if (np->n_flag & NMODIFIED) {
516 * We've modified the file: Use the larger
517 * of our size, and the server's size.
519 if (vap->va_size < np->n_size) {
520 vap->va_size = np->n_size;
522 np->n_size = vap->va_size;
523 np->n_flag |= NSIZECHANGED;
525 vnode_pager_setsize(vp, np->n_size);
526 } else if (vap->va_size < np->n_size) {
528 * When shrinking the size, the call to
529 * vnode_pager_setsize() cannot be done
530 * with the mutex held, so delay it until
531 * after the mtx_unlock call.
533 nsize = np->n_size = vap->va_size;
534 np->n_flag |= NSIZECHANGED;
537 np->n_size = vap->va_size;
538 np->n_flag |= NSIZECHANGED;
539 vnode_pager_setsize(vp, np->n_size);
542 np->n_size = vap->va_size;
546 * The following checks are added to prevent a race between (say)
547 * a READDIR+ and a WRITE.
548 * READDIR+, WRITE requests sent out.
549 * READDIR+ resp, WRITE resp received on client.
550 * However, the WRITE resp was handled before the READDIR+ resp
551 * causing the post op attrs from the write to be loaded first
552 * and the attrs from the READDIR+ to be loaded later. If this
553 * happens, we have stale attrs loaded into the attrcache.
554 * We detect this by for the mtime moving back. We invalidate the
555 * attrcache when this happens.
557 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
558 /* Size changed or mtime went backwards */
560 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
563 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
564 if (np->n_flag & NCHG) {
565 if (np->n_flag & NACC)
566 vaper->va_atime = np->n_atim;
567 if (np->n_flag & NUPD)
568 vaper->va_mtime = np->n_mtim;
574 if (np->n_attrstamp != 0)
575 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
579 vnode_pager_setsize(vp, nsize);
584 * Fill in the client id name. For these bytes:
585 * 1 - they must be unique
586 * 2 - they should be persistent across client reboots
587 * 1 is more critical than 2
588 * Use the mount point's unique id plus either the uuid or, if that
589 * isn't set, random junk.
592 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
597 * First, put in the 64bit mount point identifier.
599 if (idlen >= sizeof (u_int64_t)) {
600 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
601 cp += sizeof (u_int64_t);
602 idlen -= sizeof (u_int64_t);
606 * If uuid is non-zero length, use it.
608 uuidlen = strlen(uuid);
609 if (uuidlen > 0 && idlen >= uuidlen) {
610 NFSBCOPY(uuid, cp, uuidlen);
616 * This only normally happens if the uuid isn't set.
619 *cp++ = (u_int8_t)(arc4random() % 256);
625 * Fill in a lock owner name. For now, pid + the process's creation time.
628 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
637 /* Return the single open_owner of all 0 bytes. */
638 bzero(cp, NFSV4CL_LOCKNAMELEN);
641 if ((flags & F_POSIX) != 0) {
642 p = (struct proc *)id;
648 tl.lval = p->p_stats->p_start.tv_sec;
653 tl.lval = p->p_stats->p_start.tv_usec;
658 } else if ((flags & F_FLOCK) != 0) {
659 bcopy(&id, cp, sizeof(id));
660 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
662 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
663 bzero(cp, NFSV4CL_LOCKNAMELEN);
668 * Find the parent process for the thread passed in as an argument.
669 * If none exists, return NULL, otherwise return a thread for the parent.
670 * (Can be any of the threads, since it is only used for td->td_proc.)
673 nfscl_getparent(struct thread *td)
686 ptd = TAILQ_FIRST(&p->p_threads);
691 * Start up the renew kernel thread.
694 start_nfscl(void *arg)
696 struct nfsclclient *clp;
699 clp = (struct nfsclclient *)arg;
700 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
701 nfscl_renewthread(clp, td);
706 nfscl_start_renewthread(struct nfsclclient *clp)
709 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
715 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
716 * as the first Op after PutFH.
717 * (For NFSv4, the postop attributes are after the Op, so they can't be
718 * parsed here. A separate call to nfscl_postop_attr() is required.)
721 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
722 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
725 struct nfsnode *np = VTONFS(vp);
726 struct nfsvattr nfsva;
729 if (wccflagp != NULL)
731 if (nd->nd_flag & ND_NFSV3) {
733 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
734 if (*tl == newnfs_true) {
735 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
736 if (wccflagp != NULL) {
737 mtx_lock(&np->n_mtx);
738 *wccflagp = (np->n_mtime.tv_sec ==
739 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
740 np->n_mtime.tv_nsec ==
741 fxdr_unsigned(u_int32_t, *(tl + 3)));
742 mtx_unlock(&np->n_mtx);
745 error = nfscl_postop_attr(nd, nap, flagp, stuff);
746 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
747 == (ND_NFSV4 | ND_V4WCCATTR)) {
748 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
749 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
750 NULL, NULL, NULL, NULL, NULL);
754 * Get rid of Op# and status for next op.
756 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
758 nd->nd_flag |= ND_NOMOREDATA;
759 if (wccflagp != NULL &&
760 nfsva.na_vattr.va_mtime.tv_sec != 0) {
761 mtx_lock(&np->n_mtx);
762 *wccflagp = (np->n_mtime.tv_sec ==
763 nfsva.na_vattr.va_mtime.tv_sec &&
764 np->n_mtime.tv_nsec ==
765 nfsva.na_vattr.va_mtime.tv_sec);
766 mtx_unlock(&np->n_mtx);
774 * Get postop attributes.
777 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
784 if (nd->nd_flag & ND_NOMOREDATA)
786 if (nd->nd_flag & ND_NFSV3) {
787 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
788 *retp = fxdr_unsigned(int, *tl);
789 } else if (nd->nd_flag & ND_NFSV4) {
791 * For NFSv4, the postop attr are at the end, so no point
792 * in looking if nd_repstat != 0.
794 if (!nd->nd_repstat) {
795 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
797 /* should never happen since nd_repstat != 0 */
798 nd->nd_flag |= ND_NOMOREDATA;
802 } else if (!nd->nd_repstat) {
803 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
807 error = nfsm_loadattr(nd, nap);
816 * Fill in the setable attributes. The full argument indicates whether
817 * to fill in them all or just mode and time.
820 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
821 struct vnode *vp, int flags, u_int32_t rdev)
824 struct nfsv2_sattr *sp;
825 nfsattrbit_t attrbits;
827 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
829 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
830 if (vap->va_mode == (mode_t)VNOVAL)
831 sp->sa_mode = newnfs_xdrneg1;
833 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
834 if (vap->va_uid == (uid_t)VNOVAL)
835 sp->sa_uid = newnfs_xdrneg1;
837 sp->sa_uid = txdr_unsigned(vap->va_uid);
838 if (vap->va_gid == (gid_t)VNOVAL)
839 sp->sa_gid = newnfs_xdrneg1;
841 sp->sa_gid = txdr_unsigned(vap->va_gid);
842 if (flags & NFSSATTR_SIZE0)
844 else if (flags & NFSSATTR_SIZENEG1)
845 sp->sa_size = newnfs_xdrneg1;
846 else if (flags & NFSSATTR_SIZERDEV)
847 sp->sa_size = txdr_unsigned(rdev);
849 sp->sa_size = txdr_unsigned(vap->va_size);
850 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
851 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
854 if (vap->va_mode != (mode_t)VNOVAL) {
855 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
857 *tl = txdr_unsigned(vap->va_mode);
859 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
862 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
863 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
865 *tl = txdr_unsigned(vap->va_uid);
867 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
870 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
871 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
873 *tl = txdr_unsigned(vap->va_gid);
875 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
878 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
879 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
881 txdr_hyper(vap->va_size, tl);
883 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
886 if (vap->va_atime.tv_sec != VNOVAL) {
887 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
888 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
889 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
890 txdr_nfsv3time(&vap->va_atime, tl);
892 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
893 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
896 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
897 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
899 if (vap->va_mtime.tv_sec != VNOVAL) {
900 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
901 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
902 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
903 txdr_nfsv3time(&vap->va_mtime, tl);
905 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
906 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
909 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
910 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
914 NFSZERO_ATTRBIT(&attrbits);
915 if (vap->va_mode != (mode_t)VNOVAL)
916 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
917 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
918 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
919 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
920 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
921 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
922 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
923 if (vap->va_atime.tv_sec != VNOVAL)
924 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
925 if (vap->va_mtime.tv_sec != VNOVAL)
926 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
927 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
928 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
934 * nfscl_request() - mostly a wrapper for newnfs_request().
937 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
938 struct ucred *cred, void *stuff)
941 struct nfsmount *nmp;
943 nmp = VFSTONFS(vp->v_mount);
944 if (nd->nd_flag & ND_NFSV4)
946 else if (nd->nd_flag & ND_NFSV3)
950 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
951 NFS_PROG, vers, NULL, 1, NULL, NULL);
956 * fill in this bsden's variant of statfs using nfsstatfs.
959 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
961 struct statfs *sbp = (struct statfs *)statfs;
963 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
964 sbp->f_bsize = NFS_FABLKSIZE;
965 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
966 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
968 * Although sf_abytes is uint64_t and f_bavail is int64_t,
969 * the value after dividing by NFS_FABLKSIZE is small
970 * enough that it will fit in 63bits, so it is ok to
971 * assign it to f_bavail without fear that it will become
974 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
975 sbp->f_files = sfp->sf_tfiles;
976 /* Since f_ffree is int64_t, clip it to 63bits. */
977 if (sfp->sf_ffiles > INT64_MAX)
978 sbp->f_ffree = INT64_MAX;
980 sbp->f_ffree = sfp->sf_ffiles;
981 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
983 * The type casts to (int32_t) ensure that this code is
984 * compatible with the old NFS client, in that it will
985 * propagate bit31 to the high order bits. This may or may
986 * not be correct for NFSv2, but since it is a legacy
987 * environment, I'd rather retain backwards compatibility.
989 sbp->f_bsize = (int32_t)sfp->sf_bsize;
990 sbp->f_blocks = (int32_t)sfp->sf_blocks;
991 sbp->f_bfree = (int32_t)sfp->sf_bfree;
992 sbp->f_bavail = (int32_t)sfp->sf_bavail;
999 * Use the fsinfo stuff to update the mount point.
1002 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1005 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1006 fsp->fs_wtpref >= NFS_FABLKSIZE)
1007 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1008 ~(NFS_FABLKSIZE - 1);
1009 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1010 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1011 if (nmp->nm_wsize == 0)
1012 nmp->nm_wsize = fsp->fs_wtmax;
1014 if (nmp->nm_wsize < NFS_FABLKSIZE)
1015 nmp->nm_wsize = NFS_FABLKSIZE;
1016 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1017 fsp->fs_rtpref >= NFS_FABLKSIZE)
1018 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1019 ~(NFS_FABLKSIZE - 1);
1020 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1021 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1022 if (nmp->nm_rsize == 0)
1023 nmp->nm_rsize = fsp->fs_rtmax;
1025 if (nmp->nm_rsize < NFS_FABLKSIZE)
1026 nmp->nm_rsize = NFS_FABLKSIZE;
1027 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1028 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1029 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1030 ~(NFS_DIRBLKSIZ - 1);
1031 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1032 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1033 if (nmp->nm_readdirsize == 0)
1034 nmp->nm_readdirsize = fsp->fs_rtmax;
1036 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1037 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1038 if (fsp->fs_maxfilesize > 0 &&
1039 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1040 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1041 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1042 nmp->nm_state |= NFSSTA_GOTFSINFO;
1046 * Lookups source address which should be used to communicate with
1047 * @nmp and stores it inside @pdst.
1049 * Returns 0 on success.
1052 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1054 #if defined(INET6) || defined(INET)
1057 fibnum = curthread->td_proc->p_fibnum;
1060 if (nmp->nm_nam->sa_family == AF_INET) {
1061 struct sockaddr_in *sin;
1062 struct nhop4_extended nh_ext;
1064 sin = (struct sockaddr_in *)nmp->nm_nam;
1065 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1066 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1072 if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1074 /* Ignore loopback addresses */
1079 *((struct in_addr *)paddr) = nh_ext.nh_src;
1081 return (u_int8_t *)paddr;
1085 if (nmp->nm_nam->sa_family == AF_INET6) {
1086 struct sockaddr_in6 *sin6;
1088 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1090 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1091 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1092 sin6->sin6_scope_id, NULL, paddr, NULL);
1097 if (IN6_IS_ADDR_LOOPBACK(paddr))
1100 /* Scope is embedded in */
1103 return (u_int8_t *)paddr;
1110 * Copy NFS uid, gids from the cred structure.
1113 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1117 KASSERT(cr->cr_ngroups >= 0,
1118 ("newnfs_copyincred: negative cr_ngroups"));
1119 nfscr->nfsc_uid = cr->cr_uid;
1120 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1121 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1122 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1127 * Do any client specific initialization.
1132 static int inited = 0;
1138 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1142 * Check each of the attributes to be set, to ensure they aren't already
1143 * the correct value. Disable setting ones already correct.
1146 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1149 if (vap->va_mode != (mode_t)VNOVAL) {
1150 if (vap->va_mode == nvap->na_mode)
1151 vap->va_mode = (mode_t)VNOVAL;
1153 if (vap->va_uid != (uid_t)VNOVAL) {
1154 if (vap->va_uid == nvap->na_uid)
1155 vap->va_uid = (uid_t)VNOVAL;
1157 if (vap->va_gid != (gid_t)VNOVAL) {
1158 if (vap->va_gid == nvap->na_gid)
1159 vap->va_gid = (gid_t)VNOVAL;
1161 if (vap->va_size != VNOVAL) {
1162 if (vap->va_size == nvap->na_size)
1163 vap->va_size = VNOVAL;
1167 * We are normally called with only a partially initialized
1168 * VAP. Since the NFSv3 spec says that server may use the
1169 * file attributes to store the verifier, the spec requires
1170 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1171 * in atime, but we can't really assume that all servers will
1172 * so we ensure that our SETATTR sets both atime and mtime.
1173 * Set the VA_UTIMES_NULL flag for this case, so that
1174 * the server's time will be used. This is needed to
1175 * work around a bug in some Solaris servers, where
1176 * setting the time TOCLIENT causes the Setattr RPC
1177 * to return NFS_OK, but not set va_mode.
1179 if (vap->va_mtime.tv_sec == VNOVAL) {
1180 vfs_timestamp(&vap->va_mtime);
1181 vap->va_vaflags |= VA_UTIMES_NULL;
1183 if (vap->va_atime.tv_sec == VNOVAL)
1184 vap->va_atime = vap->va_mtime;
1189 * Map nfsv4 errors to errno.h errors.
1190 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1191 * error should only be returned for the Open, Create and Setattr Ops.
1192 * As such, most calls can just pass in 0 for those arguments.
1195 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1199 if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1206 case NFSERR_BADOWNER:
1207 tprintf(p, LOG_INFO,
1208 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1211 case NFSERR_BADNAME:
1212 case NFSERR_BADCHAR:
1213 printf("nfsv4 char/name not handled by server\n");
1215 case NFSERR_STALECLIENTID:
1216 case NFSERR_STALESTATEID:
1217 case NFSERR_EXPIRED:
1218 case NFSERR_BADSTATEID:
1219 case NFSERR_BADSESSION:
1220 printf("nfsv4 recover err returned %d\n", error);
1222 case NFSERR_BADHANDLE:
1223 case NFSERR_SERVERFAULT:
1224 case NFSERR_BADTYPE:
1225 case NFSERR_FHEXPIRED:
1226 case NFSERR_RESOURCE:
1228 case NFSERR_NOFILEHANDLE:
1229 case NFSERR_MINORVERMISMATCH:
1230 case NFSERR_OLDSTATEID:
1231 case NFSERR_BADSEQID:
1232 case NFSERR_LEASEMOVED:
1233 case NFSERR_RECLAIMBAD:
1235 case NFSERR_OPILLEGAL:
1236 printf("nfsv4 client/server protocol prob err=%d\n",
1240 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1246 * Check to see if the process for this owner exists. Return 1 if it doesn't
1250 nfscl_procdoesntexist(u_int8_t *own)
1260 /* For the single open_owner of all 0 bytes, just return 0. */
1261 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1264 if (i == NFSV4CL_LOCKNAMELEN)
1267 tl.cval[0] = *own++;
1268 tl.cval[1] = *own++;
1269 tl.cval[2] = *own++;
1270 tl.cval[3] = *own++;
1272 p = pfind_locked(pid);
1275 if (p->p_stats == NULL) {
1279 tl.cval[0] = *own++;
1280 tl.cval[1] = *own++;
1281 tl.cval[2] = *own++;
1282 tl.cval[3] = *own++;
1283 if (tl.lval != p->p_stats->p_start.tv_sec) {
1286 tl.cval[0] = *own++;
1287 tl.cval[1] = *own++;
1288 tl.cval[2] = *own++;
1290 if (tl.lval != p->p_stats->p_start.tv_usec)
1298 * - nfs pseudo system call for the client
1304 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1307 struct nfscbd_args nfscbdarg;
1308 struct nfsd_nfscbd_args nfscbdarg2;
1309 struct nameidata nd;
1310 struct nfscl_dumpmntopts dumpmntopts;
1311 cap_rights_t rights;
1315 if (uap->flag & NFSSVC_CBADDSOCK) {
1316 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1320 * Since we don't know what rights might be required,
1321 * pretend that we need them all. It is better to be too
1322 * careful than too reckless.
1324 error = fget(td, nfscbdarg.sock,
1325 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1328 if (fp->f_type != DTYPE_SOCKET) {
1332 error = nfscbd_addsock(fp);
1334 if (!error && nfscl_enablecallb == 0) {
1335 nfsv4_cbport = nfscbdarg.port;
1336 nfscl_enablecallb = 1;
1338 } else if (uap->flag & NFSSVC_NFSCBD) {
1339 if (uap->argp == NULL)
1341 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1342 sizeof(nfscbdarg2));
1345 error = nfscbd_nfsd(td, &nfscbdarg2);
1346 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1347 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1348 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1349 dumpmntopts.ndmnt_blen > 1024))
1352 error = nfsrv_lookupfilename(&nd,
1353 dumpmntopts.ndmnt_fname, td);
1354 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1360 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1361 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1362 dumpmntopts.ndmnt_blen);
1364 error = copyout(buf, dumpmntopts.ndmnt_buf,
1365 dumpmntopts.ndmnt_blen);
1374 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1377 * Called once to initialize data structures...
1380 nfscl_modevent(module_t mod, int type, void *data)
1383 static int loaded = 0;
1390 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1395 ncl_call_invalcaches = ncl_invalcaches;
1396 nfsd_call_nfscl = nfssvc_nfscl;
1401 if (nfs_numnfscbd != 0) {
1407 * XXX: Unloading of nfscl module is unsupported.
1410 ncl_call_invalcaches = NULL;
1411 nfsd_call_nfscl = NULL;
1412 /* and get rid of the mutexes */
1413 mtx_destroy(&ncl_iod_mutex);
1425 static moduledata_t nfscl_mod = {
1430 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1432 /* So that loader and kldload(2) can find us, wherever we are.. */
1433 MODULE_VERSION(nfscl, 1);
1434 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1435 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1436 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1437 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);