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 * 4. 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("$FreeBSD$");
37 #include "opt_inet6.h"
38 #include "opt_kdtrace.h"
40 #include <sys/capability.h>
43 * generally, I don't like #includes inside .h files, but it seems to
44 * be the easiest way to handle the port.
47 #include <fs/nfs/nfsport.h>
48 #include <netinet/if_ether.h>
49 #include <net/if_types.h>
51 #include <fs/nfsclient/nfs_kdtrace.h>
54 dtrace_nfsclient_attrcache_flush_probe_func_t
55 dtrace_nfscl_attrcache_flush_done_probe;
56 uint32_t nfscl_attrcache_flush_done_id;
58 dtrace_nfsclient_attrcache_get_hit_probe_func_t
59 dtrace_nfscl_attrcache_get_hit_probe;
60 uint32_t nfscl_attrcache_get_hit_id;
62 dtrace_nfsclient_attrcache_get_miss_probe_func_t
63 dtrace_nfscl_attrcache_get_miss_probe;
64 uint32_t nfscl_attrcache_get_miss_id;
66 dtrace_nfsclient_attrcache_load_probe_func_t
67 dtrace_nfscl_attrcache_load_done_probe;
68 uint32_t nfscl_attrcache_load_done_id;
69 #endif /* !KDTRACE_HOOKS */
71 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
72 extern struct vop_vector newnfs_vnodeops;
73 extern struct vop_vector newnfs_fifoops;
74 extern uma_zone_t newnfsnode_zone;
75 extern struct buf_ops buf_ops_newnfs;
76 extern int ncl_pbuf_freecnt;
77 extern short nfsv4_cbport;
78 extern int nfscl_enablecallb;
79 extern int nfs_numnfscbd;
80 extern int nfscl_inited;
81 struct mtx nfs_clstate_mutex;
82 struct mtx ncl_iod_mutex;
85 extern void (*ncl_call_invalcaches)(struct vnode *);
88 * Comparison function for vfs_hash functions.
91 newnfs_vncmpf(struct vnode *vp, void *arg)
93 struct nfsfh *nfhp = (struct nfsfh *)arg;
94 struct nfsnode *np = VTONFS(vp);
96 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
97 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
103 * Look up a vnode/nfsnode by file handle.
104 * Callers must check for mount points!!
105 * In all cases, a pointer to a
106 * nfsnode structure is returned.
107 * This variant takes a "struct nfsfh *" as second argument and uses
108 * that structure up, either by hanging off the nfsnode or FREEing it.
111 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
112 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
113 void *stuff, int lkflags)
115 struct nfsnode *np, *dnp;
116 struct vnode *vp, *nvp;
117 struct nfsv4node *newd, *oldd;
120 struct nfsmount *nmp;
122 nmp = VFSTONFS(mntp);
126 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
128 error = vfs_hash_get(mntp, hash, lkflags,
129 td, &nvp, newnfs_vncmpf, nfhp);
130 if (error == 0 && nvp != NULL) {
132 * I believe there is a slight chance that vgonel() could
133 * get called on this vnode between when NFSVOPLOCK() drops
134 * the VI_LOCK() and vget() acquires it again, so that it
135 * hasn't yet had v_usecount incremented. If this were to
136 * happen, the VI_DOOMED flag would be set, so check for
137 * that here. Since we now have the v_usecount incremented,
138 * we should be ok until we vrele() it, if the VI_DOOMED
139 * flag isn't set now.
142 if ((nvp->v_iflag & VI_DOOMED)) {
151 FREE((caddr_t)nfhp, M_NFSFH);
157 * For NFSv4, check to see if it is the same name and
158 * replace the name, if it is different.
161 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
162 nvp->v_type == VREG &&
163 (np->n_v4->n4_namelen != cnp->cn_namelen ||
164 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
166 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
167 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
168 dnp->n_fhp->nfh_len))) {
169 MALLOC(newd, struct nfsv4node *,
170 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
171 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
173 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
174 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
175 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
177 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
178 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
179 dnp->n_fhp->nfh_len))) {
183 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
184 np->n_v4->n4_namelen = cnp->cn_namelen;
185 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
186 dnp->n_fhp->nfh_len);
187 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
193 FREE((caddr_t)newd, M_NFSV4NODE);
195 FREE((caddr_t)oldd, M_NFSV4NODE);
197 FREE((caddr_t)nfhp, M_NFSFH);
202 * Allocate before getnewvnode since doing so afterward
203 * might cause a bogus v_data pointer to get dereferenced
204 * elsewhere if zalloc should block.
206 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
208 error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
210 uma_zfree(newnfsnode_zone, np);
211 FREE((caddr_t)nfhp, M_NFSFH);
215 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
216 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
220 * Initialize the mutex even if the vnode is going to be a loser.
221 * This simplifies the logic in reclaim, which can then unconditionally
222 * destroy the mutex (in the case of the loser, or if hash_insert
223 * happened to return an error no special casing is needed).
225 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
228 * Are we getting the root? If so, make sure the vnode flags
231 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
232 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
233 if (vp->v_type == VNON)
235 vp->v_vflag |= VV_ROOT;
240 * For NFSv4, we have to attach the directory file handle and
241 * file name, so that Open Ops can be done later.
243 if (nmp->nm_flag & NFSMNT_NFSV4) {
244 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
245 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
247 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
248 np->n_v4->n4_namelen = cnp->cn_namelen;
249 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
250 dnp->n_fhp->nfh_len);
251 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
258 * NFS supports recursive and shared locking.
260 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
263 error = insmntque(vp, mntp);
266 mtx_destroy(&np->n_mtx);
267 FREE((caddr_t)nfhp, M_NFSFH);
268 if (np->n_v4 != NULL)
269 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
270 uma_zfree(newnfsnode_zone, np);
273 error = vfs_hash_insert(vp, hash, lkflags,
274 td, &nvp, newnfs_vncmpf, nfhp);
279 /* vfs_hash_insert() vput()'s the losing vnode */
288 * Anothe variant of nfs_nget(). This one is only used by reopen. It
289 * takes almost the same args as nfs_nget(), but only succeeds if an entry
290 * exists in the cache. (Since files should already be "open" with a
291 * vnode ref cnt on the node when reopen calls this, it should always
293 * Also, don't get a vnode lock, since it may already be locked by some
294 * other process that is handling it. This is ok, since all other threads
295 * on the client are blocked by the nfsc_lock being exclusively held by the
296 * caller of this function.
299 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
300 struct thread *td, struct nfsnode **npp)
308 /* For forced dismounts, just return error. */
309 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
311 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
313 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
314 nfhp->nfh_len = fhsize;
316 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
319 * First, try to get the vnode locked, but don't block for the lock.
321 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
322 newnfs_vncmpf, nfhp);
323 if (error == 0 && nvp != NULL) {
324 NFSVOPUNLOCK(nvp, 0);
325 } else if (error == EBUSY) {
327 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
328 * and lock the vnode, but just get a v_usecount on it.
329 * LK_NOWAIT is set so that when vget() returns ENOENT,
330 * vfs_hash_get() fails instead of looping.
331 * If this succeeds, it is safe so long as a vflush() with
332 * FORCECLOSE has not been done. Since the Renew thread is
333 * stopped and the MNTK_UNMOUNTF flag is set before doing
334 * a vflush() with FORCECLOSE, we should be ok here.
336 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
339 error = vfs_hash_get(mntp, hash,
340 (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
341 newnfs_vncmpf, nfhp);
354 * Load the attribute cache (that lives in the nfsnode entry) with
355 * the attributes of the second argument and
357 * copy the attributes to *vaper
358 * Similar to nfs_loadattrcache(), except the attributes are passed in
359 * instead of being parsed out of the mbuf list.
362 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
363 void *stuff, int writeattr, int dontshrink)
365 struct vnode *vp = *vpp;
366 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
368 struct nfsmount *nmp;
369 struct timespec mtime_save;
372 * If v_type == VNON it is a new node, so fill in the v_type,
373 * n_mtime fields. Check to see if it represents a special
374 * device, and if so, check for a possible alias. Once the
375 * correct vnode has been obtained, fill in the rest of the
380 if (vp->v_type != nvap->va_type) {
381 vp->v_type = nvap->va_type;
382 if (vp->v_type == VFIFO)
383 vp->v_op = &newnfs_fifoops;
384 np->n_mtime = nvap->va_mtime;
386 nmp = VFSTONFS(vp->v_mount);
387 vap = &np->n_vattr.na_vattr;
388 mtime_save = vap->va_mtime;
390 np->n_vattr.na_filerev = nap->na_filerev;
391 np->n_vattr.na_size = nap->na_size;
392 np->n_vattr.na_mtime = nap->na_mtime;
393 np->n_vattr.na_ctime = nap->na_ctime;
394 np->n_vattr.na_fsid = nap->na_fsid;
395 np->n_vattr.na_mode = nap->na_mode;
397 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
398 sizeof (struct nfsvattr));
402 * For NFSv4, if the node's fsid is not equal to the mount point's
403 * fsid, return the low order 32bits of the node's fsid. This
404 * allows getcwd(3) to work. There is a chance that the fsid might
405 * be the same as a local fs, but since this is in an NFS mount
406 * point, I don't think that will cause any problems?
408 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
409 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
410 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
412 * va_fsid needs to be set to some value derived from
413 * np->n_vattr.na_filesid that is not equal
414 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
415 * from the value used for the top level server volume
416 * in the mounted subtree.
418 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
419 (uint32_t)np->n_vattr.na_filesid[0])
420 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
422 vap->va_fsid = (uint32_t)hash32_buf(
423 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
425 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
426 np->n_attrstamp = time_second;
427 if (vap->va_size != np->n_size) {
428 if (vap->va_type == VREG) {
429 if (dontshrink && vap->va_size < np->n_size) {
431 * We've been told not to shrink the file;
432 * zero np->n_attrstamp to indicate that
433 * the attributes are stale.
435 vap->va_size = np->n_size;
437 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
438 } else if (np->n_flag & NMODIFIED) {
440 * We've modified the file: Use the larger
441 * of our size, and the server's size.
443 if (vap->va_size < np->n_size) {
444 vap->va_size = np->n_size;
446 np->n_size = vap->va_size;
447 np->n_flag |= NSIZECHANGED;
450 np->n_size = vap->va_size;
451 np->n_flag |= NSIZECHANGED;
453 vnode_pager_setsize(vp, np->n_size);
455 np->n_size = vap->va_size;
459 * The following checks are added to prevent a race between (say)
460 * a READDIR+ and a WRITE.
461 * READDIR+, WRITE requests sent out.
462 * READDIR+ resp, WRITE resp received on client.
463 * However, the WRITE resp was handled before the READDIR+ resp
464 * causing the post op attrs from the write to be loaded first
465 * and the attrs from the READDIR+ to be loaded later. If this
466 * happens, we have stale attrs loaded into the attrcache.
467 * We detect this by for the mtime moving back. We invalidate the
468 * attrcache when this happens.
470 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
471 /* Size changed or mtime went backwards */
473 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
476 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
477 if (np->n_flag & NCHG) {
478 if (np->n_flag & NACC)
479 vaper->va_atime = np->n_atim;
480 if (np->n_flag & NUPD)
481 vaper->va_mtime = np->n_mtim;
485 if (np->n_attrstamp != 0)
486 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
493 * Fill in the client id name. For these bytes:
494 * 1 - they must be unique
495 * 2 - they should be persistent across client reboots
496 * 1 is more critical than 2
497 * Use the mount point's unique id plus either the uuid or, if that
498 * isn't set, random junk.
501 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
506 * First, put in the 64bit mount point identifier.
508 if (idlen >= sizeof (u_int64_t)) {
509 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
510 cp += sizeof (u_int64_t);
511 idlen -= sizeof (u_int64_t);
515 * If uuid is non-zero length, use it.
517 uuidlen = strlen(uuid);
518 if (uuidlen > 0 && idlen >= uuidlen) {
519 NFSBCOPY(uuid, cp, uuidlen);
525 * This only normally happens if the uuid isn't set.
528 *cp++ = (u_int8_t)(arc4random() % 256);
534 * Fill in a lock owner name. For now, pid + the process's creation time.
537 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
547 bzero(cp, NFSV4CL_LOCKNAMELEN);
550 if ((flags & F_POSIX) != 0) {
551 p = (struct proc *)id;
557 tl.lval = p->p_stats->p_start.tv_sec;
562 tl.lval = p->p_stats->p_start.tv_usec;
567 } else if ((flags & F_FLOCK) != 0) {
568 bcopy(&id, cp, sizeof(id));
569 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
571 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
572 bzero(cp, NFSV4CL_LOCKNAMELEN);
577 * Find the parent process for the thread passed in as an argument.
578 * If none exists, return NULL, otherwise return a thread for the parent.
579 * (Can be any of the threads, since it is only used for td->td_proc.)
582 nfscl_getparent(struct thread *td)
595 ptd = TAILQ_FIRST(&p->p_threads);
600 * Start up the renew kernel thread.
603 start_nfscl(void *arg)
605 struct nfsclclient *clp;
608 clp = (struct nfsclclient *)arg;
609 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
610 nfscl_renewthread(clp, td);
615 nfscl_start_renewthread(struct nfsclclient *clp)
618 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
624 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
625 * as the first Op after PutFH.
626 * (For NFSv4, the postop attributes are after the Op, so they can't be
627 * parsed here. A separate call to nfscl_postop_attr() is required.)
630 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
631 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
634 struct nfsnode *np = VTONFS(vp);
635 struct nfsvattr nfsva;
638 if (wccflagp != NULL)
640 if (nd->nd_flag & ND_NFSV3) {
642 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
643 if (*tl == newnfs_true) {
644 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
645 if (wccflagp != NULL) {
646 mtx_lock(&np->n_mtx);
647 *wccflagp = (np->n_mtime.tv_sec ==
648 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
649 np->n_mtime.tv_nsec ==
650 fxdr_unsigned(u_int32_t, *(tl + 3)));
651 mtx_unlock(&np->n_mtx);
654 error = nfscl_postop_attr(nd, nap, flagp, stuff);
655 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
656 == (ND_NFSV4 | ND_V4WCCATTR)) {
657 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
658 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
659 NULL, NULL, NULL, NULL, NULL);
663 * Get rid of Op# and status for next op.
665 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
667 nd->nd_flag |= ND_NOMOREDATA;
668 if (wccflagp != NULL &&
669 nfsva.na_vattr.va_mtime.tv_sec != 0) {
670 mtx_lock(&np->n_mtx);
671 *wccflagp = (np->n_mtime.tv_sec ==
672 nfsva.na_vattr.va_mtime.tv_sec &&
673 np->n_mtime.tv_nsec ==
674 nfsva.na_vattr.va_mtime.tv_sec);
675 mtx_unlock(&np->n_mtx);
683 * Get postop attributes.
686 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
693 if (nd->nd_flag & ND_NOMOREDATA)
695 if (nd->nd_flag & ND_NFSV3) {
696 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
697 *retp = fxdr_unsigned(int, *tl);
698 } else if (nd->nd_flag & ND_NFSV4) {
700 * For NFSv4, the postop attr are at the end, so no point
701 * in looking if nd_repstat != 0.
703 if (!nd->nd_repstat) {
704 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
706 /* should never happen since nd_repstat != 0 */
707 nd->nd_flag |= ND_NOMOREDATA;
711 } else if (!nd->nd_repstat) {
712 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
716 error = nfsm_loadattr(nd, nap);
725 * Fill in the setable attributes. The full argument indicates whether
726 * to fill in them all or just mode and time.
729 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
730 struct vnode *vp, int flags, u_int32_t rdev)
733 struct nfsv2_sattr *sp;
734 nfsattrbit_t attrbits;
735 struct timeval curtime;
737 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
739 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
740 if (vap->va_mode == (mode_t)VNOVAL)
741 sp->sa_mode = newnfs_xdrneg1;
743 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
744 if (vap->va_uid == (uid_t)VNOVAL)
745 sp->sa_uid = newnfs_xdrneg1;
747 sp->sa_uid = txdr_unsigned(vap->va_uid);
748 if (vap->va_gid == (gid_t)VNOVAL)
749 sp->sa_gid = newnfs_xdrneg1;
751 sp->sa_gid = txdr_unsigned(vap->va_gid);
752 if (flags & NFSSATTR_SIZE0)
754 else if (flags & NFSSATTR_SIZENEG1)
755 sp->sa_size = newnfs_xdrneg1;
756 else if (flags & NFSSATTR_SIZERDEV)
757 sp->sa_size = txdr_unsigned(rdev);
759 sp->sa_size = txdr_unsigned(vap->va_size);
760 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
761 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
764 getmicrotime(&curtime);
765 if (vap->va_mode != (mode_t)VNOVAL) {
766 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
768 *tl = txdr_unsigned(vap->va_mode);
770 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
773 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
774 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
776 *tl = txdr_unsigned(vap->va_uid);
778 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
781 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
782 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
784 *tl = txdr_unsigned(vap->va_gid);
786 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
789 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
790 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
792 txdr_hyper(vap->va_size, tl);
794 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
797 if (vap->va_atime.tv_sec != VNOVAL) {
798 if (vap->va_atime.tv_sec != curtime.tv_sec) {
799 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
800 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
801 txdr_nfsv3time(&vap->va_atime, tl);
803 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
804 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
807 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
808 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
810 if (vap->va_mtime.tv_sec != VNOVAL) {
811 if (vap->va_mtime.tv_sec != curtime.tv_sec) {
812 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
813 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
814 txdr_nfsv3time(&vap->va_mtime, tl);
816 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
817 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
820 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
821 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
825 NFSZERO_ATTRBIT(&attrbits);
826 if (vap->va_mode != (mode_t)VNOVAL)
827 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
828 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
829 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
830 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
831 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
832 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
833 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
834 if (vap->va_atime.tv_sec != VNOVAL)
835 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
836 if (vap->va_mtime.tv_sec != VNOVAL)
837 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
838 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
839 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
845 * nfscl_request() - mostly a wrapper for newnfs_request().
848 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
849 struct ucred *cred, void *stuff)
852 struct nfsmount *nmp;
854 nmp = VFSTONFS(vp->v_mount);
855 if (nd->nd_flag & ND_NFSV4)
857 else if (nd->nd_flag & ND_NFSV3)
861 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
862 NFS_PROG, vers, NULL, 1, NULL);
867 * fill in this bsden's variant of statfs using nfsstatfs.
870 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
872 struct statfs *sbp = (struct statfs *)statfs;
874 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
875 sbp->f_bsize = NFS_FABLKSIZE;
876 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
877 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
879 * Although sf_abytes is uint64_t and f_bavail is int64_t,
880 * the value after dividing by NFS_FABLKSIZE is small
881 * enough that it will fit in 63bits, so it is ok to
882 * assign it to f_bavail without fear that it will become
885 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
886 sbp->f_files = sfp->sf_tfiles;
887 /* Since f_ffree is int64_t, clip it to 63bits. */
888 if (sfp->sf_ffiles > INT64_MAX)
889 sbp->f_ffree = INT64_MAX;
891 sbp->f_ffree = sfp->sf_ffiles;
892 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
894 * The type casts to (int32_t) ensure that this code is
895 * compatible with the old NFS client, in that it will
896 * propagate bit31 to the high order bits. This may or may
897 * not be correct for NFSv2, but since it is a legacy
898 * environment, I'd rather retain backwards compatibility.
900 sbp->f_bsize = (int32_t)sfp->sf_bsize;
901 sbp->f_blocks = (int32_t)sfp->sf_blocks;
902 sbp->f_bfree = (int32_t)sfp->sf_bfree;
903 sbp->f_bavail = (int32_t)sfp->sf_bavail;
910 * Use the fsinfo stuff to update the mount point.
913 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
916 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
917 fsp->fs_wtpref >= NFS_FABLKSIZE)
918 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
919 ~(NFS_FABLKSIZE - 1);
920 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
921 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
922 if (nmp->nm_wsize == 0)
923 nmp->nm_wsize = fsp->fs_wtmax;
925 if (nmp->nm_wsize < NFS_FABLKSIZE)
926 nmp->nm_wsize = NFS_FABLKSIZE;
927 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
928 fsp->fs_rtpref >= NFS_FABLKSIZE)
929 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
930 ~(NFS_FABLKSIZE - 1);
931 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
932 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
933 if (nmp->nm_rsize == 0)
934 nmp->nm_rsize = fsp->fs_rtmax;
936 if (nmp->nm_rsize < NFS_FABLKSIZE)
937 nmp->nm_rsize = NFS_FABLKSIZE;
938 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
939 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
940 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
941 ~(NFS_DIRBLKSIZ - 1);
942 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
943 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
944 if (nmp->nm_readdirsize == 0)
945 nmp->nm_readdirsize = fsp->fs_rtmax;
947 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
948 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
949 if (fsp->fs_maxfilesize > 0 &&
950 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
951 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
952 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
953 nmp->nm_state |= NFSSTA_GOTFSINFO;
957 * Get a pointer to my IP addrress and return it.
958 * Return NULL if you can't find one.
961 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
963 struct sockaddr_in sad, *sin;
965 u_int8_t *retp = NULL;
966 static struct in_addr laddr;
970 * Loop up a route for the destination address.
972 if (nmp->nm_nam->sa_family == AF_INET) {
973 bzero(&sad, sizeof (sad));
974 sin = (struct sockaddr_in *)nmp->nm_nam;
975 sad.sin_family = AF_INET;
976 sad.sin_len = sizeof (struct sockaddr_in);
977 sad.sin_addr.s_addr = sin->sin_addr.s_addr;
978 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
979 rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
980 curthread->td_proc->p_fibnum);
982 if (rt->rt_ifp != NULL &&
983 rt->rt_ifa != NULL &&
984 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
985 rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
986 sin = (struct sockaddr_in *)
987 rt->rt_ifa->ifa_addr;
988 laddr.s_addr = sin->sin_addr.s_addr;
989 retp = (u_int8_t *)&laddr;
995 } else if (nmp->nm_nam->sa_family == AF_INET6) {
996 struct sockaddr_in6 sad6, *sin6;
997 static struct in6_addr laddr6;
999 bzero(&sad6, sizeof (sad6));
1000 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1001 sad6.sin6_family = AF_INET6;
1002 sad6.sin6_len = sizeof (struct sockaddr_in6);
1003 sad6.sin6_addr = sin6->sin6_addr;
1004 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1005 rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1006 curthread->td_proc->p_fibnum);
1008 if (rt->rt_ifp != NULL &&
1009 rt->rt_ifa != NULL &&
1010 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1011 rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1012 sin6 = (struct sockaddr_in6 *)
1013 rt->rt_ifa->ifa_addr;
1014 laddr6 = sin6->sin6_addr;
1015 retp = (u_int8_t *)&laddr6;
1027 * Copy NFS uid, gids from the cred structure.
1030 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1034 KASSERT(cr->cr_ngroups >= 0,
1035 ("newnfs_copyincred: negative cr_ngroups"));
1036 nfscr->nfsc_uid = cr->cr_uid;
1037 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1038 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1039 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1044 * Do any client specific initialization.
1049 static int inited = 0;
1055 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1059 * Check each of the attributes to be set, to ensure they aren't already
1060 * the correct value. Disable setting ones already correct.
1063 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1066 if (vap->va_mode != (mode_t)VNOVAL) {
1067 if (vap->va_mode == nvap->na_mode)
1068 vap->va_mode = (mode_t)VNOVAL;
1070 if (vap->va_uid != (uid_t)VNOVAL) {
1071 if (vap->va_uid == nvap->na_uid)
1072 vap->va_uid = (uid_t)VNOVAL;
1074 if (vap->va_gid != (gid_t)VNOVAL) {
1075 if (vap->va_gid == nvap->na_gid)
1076 vap->va_gid = (gid_t)VNOVAL;
1078 if (vap->va_size != VNOVAL) {
1079 if (vap->va_size == nvap->na_size)
1080 vap->va_size = VNOVAL;
1084 * We are normally called with only a partially initialized
1085 * VAP. Since the NFSv3 spec says that server may use the
1086 * file attributes to store the verifier, the spec requires
1087 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1088 * in atime, but we can't really assume that all servers will
1089 * so we ensure that our SETATTR sets both atime and mtime.
1091 if (vap->va_mtime.tv_sec == VNOVAL)
1092 vfs_timestamp(&vap->va_mtime);
1093 if (vap->va_atime.tv_sec == VNOVAL)
1094 vap->va_atime = vap->va_mtime;
1099 * Map nfsv4 errors to errno.h errors.
1100 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1101 * error should only be returned for the Open, Create and Setattr Ops.
1102 * As such, most calls can just pass in 0 for those arguments.
1105 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1116 case NFSERR_BADOWNER:
1117 tprintf(p, LOG_INFO,
1118 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1121 case NFSERR_STALECLIENTID:
1122 case NFSERR_STALESTATEID:
1123 case NFSERR_EXPIRED:
1124 case NFSERR_BADSTATEID:
1125 printf("nfsv4 recover err returned %d\n", error);
1127 case NFSERR_BADHANDLE:
1128 case NFSERR_SERVERFAULT:
1129 case NFSERR_BADTYPE:
1130 case NFSERR_FHEXPIRED:
1131 case NFSERR_RESOURCE:
1133 case NFSERR_NOFILEHANDLE:
1134 case NFSERR_MINORVERMISMATCH:
1135 case NFSERR_OLDSTATEID:
1136 case NFSERR_BADSEQID:
1137 case NFSERR_LEASEMOVED:
1138 case NFSERR_RECLAIMBAD:
1140 case NFSERR_BADCHAR:
1141 case NFSERR_BADNAME:
1142 case NFSERR_OPILLEGAL:
1143 printf("nfsv4 client/server protocol prob err=%d\n",
1147 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1153 * Locate a process by number; return only "live" processes -- i.e., neither
1154 * zombies nor newly born but incompletely initialized processes. By not
1155 * returning processes in the PRS_NEW state, we allow callers to avoid
1156 * testing for that condition to avoid dereferencing p_ucred, et al.
1157 * Identical to pfind() in kern_proc.c, except it assume the list is
1160 static struct proc *
1161 pfind_locked(pid_t pid)
1165 LIST_FOREACH(p, PIDHASH(pid), p_hash)
1166 if (p->p_pid == pid) {
1168 if (p->p_state == PRS_NEW) {
1178 * Check to see if the process for this owner exists. Return 1 if it doesn't
1182 nfscl_procdoesntexist(u_int8_t *own)
1192 tl.cval[0] = *own++;
1193 tl.cval[1] = *own++;
1194 tl.cval[2] = *own++;
1195 tl.cval[3] = *own++;
1197 p = pfind_locked(pid);
1200 if (p->p_stats == NULL) {
1204 tl.cval[0] = *own++;
1205 tl.cval[1] = *own++;
1206 tl.cval[2] = *own++;
1207 tl.cval[3] = *own++;
1208 if (tl.lval != p->p_stats->p_start.tv_sec) {
1211 tl.cval[0] = *own++;
1212 tl.cval[1] = *own++;
1213 tl.cval[2] = *own++;
1215 if (tl.lval != p->p_stats->p_start.tv_usec)
1223 * - nfs pseudo system call for the client
1229 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1232 struct nfscbd_args nfscbdarg;
1233 struct nfsd_nfscbd_args nfscbdarg2;
1236 if (uap->flag & NFSSVC_CBADDSOCK) {
1237 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1241 * Since we don't know what rights might be required,
1242 * pretend that we need them all. It is better to be too
1243 * careful than too reckless.
1245 if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1249 if (fp->f_type != DTYPE_SOCKET) {
1253 error = nfscbd_addsock(fp);
1255 if (!error && nfscl_enablecallb == 0) {
1256 nfsv4_cbport = nfscbdarg.port;
1257 nfscl_enablecallb = 1;
1259 } else if (uap->flag & NFSSVC_NFSCBD) {
1260 if (uap->argp == NULL)
1262 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1263 sizeof(nfscbdarg2));
1266 error = nfscbd_nfsd(td, &nfscbdarg2);
1273 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1276 * Called once to initialize data structures...
1279 nfscl_modevent(module_t mod, int type, void *data)
1282 static int loaded = 0;
1289 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1291 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1296 ncl_call_invalcaches = ncl_invalcaches;
1297 nfsd_call_nfscl = nfssvc_nfscl;
1302 if (nfs_numnfscbd != 0) {
1308 * XXX: Unloading of nfscl module is unsupported.
1311 ncl_call_invalcaches = NULL;
1312 nfsd_call_nfscl = NULL;
1313 /* and get rid of the mutexes */
1314 mtx_destroy(&nfs_clstate_mutex);
1315 mtx_destroy(&ncl_iod_mutex);
1327 static moduledata_t nfscl_mod = {
1332 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1334 /* So that loader and kldload(2) can find us, wherever we are.. */
1335 MODULE_VERSION(nfscl, 1);
1336 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1337 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1338 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1339 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);