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_kdtrace.h"
39 #include <sys/capability.h>
42 * generally, I don't like #includes inside .h files, but it seems to
43 * be the easiest way to handle the port.
46 #include <fs/nfs/nfsport.h>
47 #include <netinet/if_ether.h>
48 #include <net/if_types.h>
50 #include <fs/nfsclient/nfs_kdtrace.h>
53 dtrace_nfsclient_attrcache_flush_probe_func_t
54 dtrace_nfscl_attrcache_flush_done_probe;
55 uint32_t nfscl_attrcache_flush_done_id;
57 dtrace_nfsclient_attrcache_get_hit_probe_func_t
58 dtrace_nfscl_attrcache_get_hit_probe;
59 uint32_t nfscl_attrcache_get_hit_id;
61 dtrace_nfsclient_attrcache_get_miss_probe_func_t
62 dtrace_nfscl_attrcache_get_miss_probe;
63 uint32_t nfscl_attrcache_get_miss_id;
65 dtrace_nfsclient_attrcache_load_probe_func_t
66 dtrace_nfscl_attrcache_load_done_probe;
67 uint32_t nfscl_attrcache_load_done_id;
68 #endif /* !KDTRACE_HOOKS */
70 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
71 extern struct vop_vector newnfs_vnodeops;
72 extern struct vop_vector newnfs_fifoops;
73 extern uma_zone_t newnfsnode_zone;
74 extern struct buf_ops buf_ops_newnfs;
75 extern int ncl_pbuf_freecnt;
76 extern short nfsv4_cbport;
77 extern int nfscl_enablecallb;
78 extern int nfs_numnfscbd;
79 extern int nfscl_inited;
80 struct mtx nfs_clstate_mutex;
81 struct mtx ncl_iod_mutex;
84 extern void (*ncl_call_invalcaches)(struct vnode *);
87 * Comparison function for vfs_hash functions.
90 newnfs_vncmpf(struct vnode *vp, void *arg)
92 struct nfsfh *nfhp = (struct nfsfh *)arg;
93 struct nfsnode *np = VTONFS(vp);
95 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
96 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
102 * Look up a vnode/nfsnode by file handle.
103 * Callers must check for mount points!!
104 * In all cases, a pointer to a
105 * nfsnode structure is returned.
106 * This variant takes a "struct nfsfh *" as second argument and uses
107 * that structure up, either by hanging off the nfsnode or FREEing it.
110 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
111 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
112 void *stuff, int lkflags)
114 struct nfsnode *np, *dnp;
115 struct vnode *vp, *nvp;
116 struct nfsv4node *newd, *oldd;
119 struct nfsmount *nmp;
121 nmp = VFSTONFS(mntp);
125 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
127 error = vfs_hash_get(mntp, hash, lkflags,
128 td, &nvp, newnfs_vncmpf, nfhp);
129 if (error == 0 && nvp != NULL) {
131 * I believe there is a slight chance that vgonel() could
132 * get called on this vnode between when NFSVOPLOCK() drops
133 * the VI_LOCK() and vget() acquires it again, so that it
134 * hasn't yet had v_usecount incremented. If this were to
135 * happen, the VI_DOOMED flag would be set, so check for
136 * that here. Since we now have the v_usecount incremented,
137 * we should be ok until we vrele() it, if the VI_DOOMED
138 * flag isn't set now.
141 if ((nvp->v_iflag & VI_DOOMED)) {
150 FREE((caddr_t)nfhp, M_NFSFH);
156 * For NFSv4, check to see if it is the same name and
157 * replace the name, if it is different.
160 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
161 nvp->v_type == VREG &&
162 (np->n_v4->n4_namelen != cnp->cn_namelen ||
163 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
165 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
166 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
167 dnp->n_fhp->nfh_len))) {
168 MALLOC(newd, struct nfsv4node *,
169 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
170 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
172 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
173 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
174 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
176 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
177 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
178 dnp->n_fhp->nfh_len))) {
182 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
183 np->n_v4->n4_namelen = cnp->cn_namelen;
184 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
185 dnp->n_fhp->nfh_len);
186 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
192 FREE((caddr_t)newd, M_NFSV4NODE);
194 FREE((caddr_t)oldd, M_NFSV4NODE);
196 FREE((caddr_t)nfhp, M_NFSFH);
201 * Allocate before getnewvnode since doing so afterward
202 * might cause a bogus v_data pointer to get dereferenced
203 * elsewhere if zalloc should block.
205 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
207 error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
209 uma_zfree(newnfsnode_zone, np);
210 FREE((caddr_t)nfhp, M_NFSFH);
214 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
218 * Initialize the mutex even if the vnode is going to be a loser.
219 * This simplifies the logic in reclaim, which can then unconditionally
220 * destroy the mutex (in the case of the loser, or if hash_insert
221 * happened to return an error no special casing is needed).
223 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
226 * Are we getting the root? If so, make sure the vnode flags
229 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
230 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
231 if (vp->v_type == VNON)
233 vp->v_vflag |= VV_ROOT;
238 * For NFSv4, we have to attach the directory file handle and
239 * file name, so that Open Ops can be done later.
241 if (nmp->nm_flag & NFSMNT_NFSV4) {
242 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
243 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
245 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
246 np->n_v4->n4_namelen = cnp->cn_namelen;
247 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
248 dnp->n_fhp->nfh_len);
249 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
256 * NFS supports recursive and shared locking.
258 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
261 error = insmntque(vp, mntp);
264 mtx_destroy(&np->n_mtx);
265 FREE((caddr_t)nfhp, M_NFSFH);
266 if (np->n_v4 != NULL)
267 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
268 uma_zfree(newnfsnode_zone, np);
271 error = vfs_hash_insert(vp, hash, lkflags,
272 td, &nvp, newnfs_vncmpf, nfhp);
277 /* vfs_hash_insert() vput()'s the losing vnode */
286 * Anothe variant of nfs_nget(). This one is only used by reopen. It
287 * takes almost the same args as nfs_nget(), but only succeeds if an entry
288 * exists in the cache. (Since files should already be "open" with a
289 * vnode ref cnt on the node when reopen calls this, it should always
291 * Also, don't get a vnode lock, since it may already be locked by some
292 * other process that is handling it. This is ok, since all other threads
293 * on the client are blocked by the nfsc_lock being exclusively held by the
294 * caller of this function.
297 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
298 struct thread *td, struct nfsnode **npp)
306 /* For forced dismounts, just return error. */
307 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
309 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
311 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
312 nfhp->nfh_len = fhsize;
314 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
317 * First, try to get the vnode locked, but don't block for the lock.
319 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
320 newnfs_vncmpf, nfhp);
321 if (error == 0 && nvp != NULL) {
322 NFSVOPUNLOCK(nvp, 0);
323 } else if (error == EBUSY) {
325 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
326 * and lock the vnode, but just get a v_usecount on it.
327 * LK_NOWAIT is set so that when vget() returns ENOENT,
328 * vfs_hash_get() fails instead of looping.
329 * If this succeeds, it is safe so long as a vflush() with
330 * FORCECLOSE has not been done. Since the Renew thread is
331 * stopped and the MNTK_UNMOUNTF flag is set before doing
332 * a vflush() with FORCECLOSE, we should be ok here.
334 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
337 error = vfs_hash_get(mntp, hash,
338 (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
339 newnfs_vncmpf, nfhp);
352 * Load the attribute cache (that lives in the nfsnode entry) with
353 * the attributes of the second argument and
355 * copy the attributes to *vaper
356 * Similar to nfs_loadattrcache(), except the attributes are passed in
357 * instead of being parsed out of the mbuf list.
360 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
361 void *stuff, int writeattr, int dontshrink)
363 struct vnode *vp = *vpp;
364 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
366 struct nfsmount *nmp;
367 struct timespec mtime_save;
370 * If v_type == VNON it is a new node, so fill in the v_type,
371 * n_mtime fields. Check to see if it represents a special
372 * device, and if so, check for a possible alias. Once the
373 * correct vnode has been obtained, fill in the rest of the
378 if (vp->v_type != nvap->va_type) {
379 vp->v_type = nvap->va_type;
380 if (vp->v_type == VFIFO)
381 vp->v_op = &newnfs_fifoops;
382 np->n_mtime = nvap->va_mtime;
384 nmp = VFSTONFS(vp->v_mount);
385 vap = &np->n_vattr.na_vattr;
386 mtime_save = vap->va_mtime;
388 np->n_vattr.na_filerev = nap->na_filerev;
389 np->n_vattr.na_size = nap->na_size;
390 np->n_vattr.na_mtime = nap->na_mtime;
391 np->n_vattr.na_ctime = nap->na_ctime;
392 np->n_vattr.na_fsid = nap->na_fsid;
393 np->n_vattr.na_mode = nap->na_mode;
395 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
396 sizeof (struct nfsvattr));
400 * For NFSv4, if the node's fsid is not equal to the mount point's
401 * fsid, return the low order 32bits of the node's fsid. This
402 * allows getcwd(3) to work. There is a chance that the fsid might
403 * be the same as a local fs, but since this is in an NFS mount
404 * point, I don't think that will cause any problems?
406 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
407 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
408 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
410 * va_fsid needs to be set to some value derived from
411 * np->n_vattr.na_filesid that is not equal
412 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
413 * from the value used for the top level server volume
414 * in the mounted subtree.
416 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
417 (uint32_t)np->n_vattr.na_filesid[0])
418 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
420 vap->va_fsid = (uint32_t)hash32_buf(
421 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
423 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
424 np->n_attrstamp = time_second;
425 if (vap->va_size != np->n_size) {
426 if (vap->va_type == VREG) {
427 if (dontshrink && vap->va_size < np->n_size) {
429 * We've been told not to shrink the file;
430 * zero np->n_attrstamp to indicate that
431 * the attributes are stale.
433 vap->va_size = np->n_size;
435 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
436 } else if (np->n_flag & NMODIFIED) {
438 * We've modified the file: Use the larger
439 * of our size, and the server's size.
441 if (vap->va_size < np->n_size) {
442 vap->va_size = np->n_size;
444 np->n_size = vap->va_size;
445 np->n_flag |= NSIZECHANGED;
448 np->n_size = vap->va_size;
449 np->n_flag |= NSIZECHANGED;
451 vnode_pager_setsize(vp, np->n_size);
453 np->n_size = vap->va_size;
457 * The following checks are added to prevent a race between (say)
458 * a READDIR+ and a WRITE.
459 * READDIR+, WRITE requests sent out.
460 * READDIR+ resp, WRITE resp received on client.
461 * However, the WRITE resp was handled before the READDIR+ resp
462 * causing the post op attrs from the write to be loaded first
463 * and the attrs from the READDIR+ to be loaded later. If this
464 * happens, we have stale attrs loaded into the attrcache.
465 * We detect this by for the mtime moving back. We invalidate the
466 * attrcache when this happens.
468 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
469 /* Size changed or mtime went backwards */
471 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
474 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
475 if (np->n_flag & NCHG) {
476 if (np->n_flag & NACC)
477 vaper->va_atime = np->n_atim;
478 if (np->n_flag & NUPD)
479 vaper->va_mtime = np->n_mtim;
483 if (np->n_attrstamp != 0)
484 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
491 * Fill in the client id name. For these bytes:
492 * 1 - they must be unique
493 * 2 - they should be persistent across client reboots
494 * 1 is more critical than 2
495 * Use the mount point's unique id plus either the uuid or, if that
496 * isn't set, random junk.
499 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
504 * First, put in the 64bit mount point identifier.
506 if (idlen >= sizeof (u_int64_t)) {
507 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
508 cp += sizeof (u_int64_t);
509 idlen -= sizeof (u_int64_t);
513 * If uuid is non-zero length, use it.
515 uuidlen = strlen(uuid);
516 if (uuidlen > 0 && idlen >= uuidlen) {
517 NFSBCOPY(uuid, cp, uuidlen);
523 * This only normally happens if the uuid isn't set.
526 *cp++ = (u_int8_t)(arc4random() % 256);
532 * Fill in a lock owner name. For now, pid + the process's creation time.
535 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
545 bzero(cp, NFSV4CL_LOCKNAMELEN);
548 if ((flags & F_POSIX) != 0) {
549 p = (struct proc *)id;
555 tl.lval = p->p_stats->p_start.tv_sec;
560 tl.lval = p->p_stats->p_start.tv_usec;
565 } else if ((flags & F_FLOCK) != 0) {
566 bcopy(&id, cp, sizeof(id));
567 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
569 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
570 bzero(cp, NFSV4CL_LOCKNAMELEN);
575 * Find the parent process for the thread passed in as an argument.
576 * If none exists, return NULL, otherwise return a thread for the parent.
577 * (Can be any of the threads, since it is only used for td->td_proc.)
580 nfscl_getparent(struct thread *td)
593 ptd = TAILQ_FIRST(&p->p_threads);
598 * Start up the renew kernel thread.
601 start_nfscl(void *arg)
603 struct nfsclclient *clp;
606 clp = (struct nfsclclient *)arg;
607 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
608 nfscl_renewthread(clp, td);
613 nfscl_start_renewthread(struct nfsclclient *clp)
616 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
622 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
623 * as the first Op after PutFH.
624 * (For NFSv4, the postop attributes are after the Op, so they can't be
625 * parsed here. A separate call to nfscl_postop_attr() is required.)
628 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
629 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
632 struct nfsnode *np = VTONFS(vp);
633 struct nfsvattr nfsva;
636 if (wccflagp != NULL)
638 if (nd->nd_flag & ND_NFSV3) {
640 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
641 if (*tl == newnfs_true) {
642 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
643 if (wccflagp != NULL) {
644 mtx_lock(&np->n_mtx);
645 *wccflagp = (np->n_mtime.tv_sec ==
646 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
647 np->n_mtime.tv_nsec ==
648 fxdr_unsigned(u_int32_t, *(tl + 3)));
649 mtx_unlock(&np->n_mtx);
652 error = nfscl_postop_attr(nd, nap, flagp, stuff);
653 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
654 == (ND_NFSV4 | ND_V4WCCATTR)) {
655 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
656 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
657 NULL, NULL, NULL, NULL, NULL);
661 * Get rid of Op# and status for next op.
663 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
665 nd->nd_flag |= ND_NOMOREDATA;
666 if (wccflagp != NULL &&
667 nfsva.na_vattr.va_mtime.tv_sec != 0) {
668 mtx_lock(&np->n_mtx);
669 *wccflagp = (np->n_mtime.tv_sec ==
670 nfsva.na_vattr.va_mtime.tv_sec &&
671 np->n_mtime.tv_nsec ==
672 nfsva.na_vattr.va_mtime.tv_sec);
673 mtx_unlock(&np->n_mtx);
681 * Get postop attributes.
684 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
691 if (nd->nd_flag & ND_NOMOREDATA)
693 if (nd->nd_flag & ND_NFSV3) {
694 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
695 *retp = fxdr_unsigned(int, *tl);
696 } else if (nd->nd_flag & ND_NFSV4) {
698 * For NFSv4, the postop attr are at the end, so no point
699 * in looking if nd_repstat != 0.
701 if (!nd->nd_repstat) {
702 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
704 /* should never happen since nd_repstat != 0 */
705 nd->nd_flag |= ND_NOMOREDATA;
709 } else if (!nd->nd_repstat) {
710 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
714 error = nfsm_loadattr(nd, nap);
723 * Fill in the setable attributes. The full argument indicates whether
724 * to fill in them all or just mode and time.
727 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
728 struct vnode *vp, int flags, u_int32_t rdev)
731 struct nfsv2_sattr *sp;
732 nfsattrbit_t attrbits;
733 struct timeval curtime;
735 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
737 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
738 if (vap->va_mode == (mode_t)VNOVAL)
739 sp->sa_mode = newnfs_xdrneg1;
741 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
742 if (vap->va_uid == (uid_t)VNOVAL)
743 sp->sa_uid = newnfs_xdrneg1;
745 sp->sa_uid = txdr_unsigned(vap->va_uid);
746 if (vap->va_gid == (gid_t)VNOVAL)
747 sp->sa_gid = newnfs_xdrneg1;
749 sp->sa_gid = txdr_unsigned(vap->va_gid);
750 if (flags & NFSSATTR_SIZE0)
752 else if (flags & NFSSATTR_SIZENEG1)
753 sp->sa_size = newnfs_xdrneg1;
754 else if (flags & NFSSATTR_SIZERDEV)
755 sp->sa_size = txdr_unsigned(rdev);
757 sp->sa_size = txdr_unsigned(vap->va_size);
758 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
759 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
762 getmicrotime(&curtime);
763 if (vap->va_mode != (mode_t)VNOVAL) {
764 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
766 *tl = txdr_unsigned(vap->va_mode);
768 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
771 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
772 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
774 *tl = txdr_unsigned(vap->va_uid);
776 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
779 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
780 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
782 *tl = txdr_unsigned(vap->va_gid);
784 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
787 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
788 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
790 txdr_hyper(vap->va_size, tl);
792 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
795 if (vap->va_atime.tv_sec != VNOVAL) {
796 if (vap->va_atime.tv_sec != curtime.tv_sec) {
797 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
798 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
799 txdr_nfsv3time(&vap->va_atime, tl);
801 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
802 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
805 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
806 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
808 if (vap->va_mtime.tv_sec != VNOVAL) {
809 if (vap->va_mtime.tv_sec != curtime.tv_sec) {
810 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
811 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
812 txdr_nfsv3time(&vap->va_mtime, tl);
814 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
815 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
818 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
819 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
823 NFSZERO_ATTRBIT(&attrbits);
824 if (vap->va_mode != (mode_t)VNOVAL)
825 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
826 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
827 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
828 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
829 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
830 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
831 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
832 if (vap->va_atime.tv_sec != VNOVAL)
833 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
834 if (vap->va_mtime.tv_sec != VNOVAL)
835 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
836 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
837 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
843 * nfscl_request() - mostly a wrapper for newnfs_request().
846 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
847 struct ucred *cred, void *stuff)
850 struct nfsmount *nmp;
852 nmp = VFSTONFS(vp->v_mount);
853 if (nd->nd_flag & ND_NFSV4)
855 else if (nd->nd_flag & ND_NFSV3)
859 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
860 NFS_PROG, vers, NULL, 1, NULL);
865 * fill in this bsden's variant of statfs using nfsstatfs.
868 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
870 struct statfs *sbp = (struct statfs *)statfs;
872 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
873 sbp->f_bsize = NFS_FABLKSIZE;
874 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
875 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
877 * Although sf_abytes is uint64_t and f_bavail is int64_t,
878 * the value after dividing by NFS_FABLKSIZE is small
879 * enough that it will fit in 63bits, so it is ok to
880 * assign it to f_bavail without fear that it will become
883 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
884 sbp->f_files = sfp->sf_tfiles;
885 /* Since f_ffree is int64_t, clip it to 63bits. */
886 if (sfp->sf_ffiles > INT64_MAX)
887 sbp->f_ffree = INT64_MAX;
889 sbp->f_ffree = sfp->sf_ffiles;
890 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
892 * The type casts to (int32_t) ensure that this code is
893 * compatible with the old NFS client, in that it will
894 * propagate bit31 to the high order bits. This may or may
895 * not be correct for NFSv2, but since it is a legacy
896 * environment, I'd rather retain backwards compatibility.
898 sbp->f_bsize = (int32_t)sfp->sf_bsize;
899 sbp->f_blocks = (int32_t)sfp->sf_blocks;
900 sbp->f_bfree = (int32_t)sfp->sf_bfree;
901 sbp->f_bavail = (int32_t)sfp->sf_bavail;
908 * Use the fsinfo stuff to update the mount point.
911 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
914 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
915 fsp->fs_wtpref >= NFS_FABLKSIZE)
916 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
917 ~(NFS_FABLKSIZE - 1);
918 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
919 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
920 if (nmp->nm_wsize == 0)
921 nmp->nm_wsize = fsp->fs_wtmax;
923 if (nmp->nm_wsize < NFS_FABLKSIZE)
924 nmp->nm_wsize = NFS_FABLKSIZE;
925 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
926 fsp->fs_rtpref >= NFS_FABLKSIZE)
927 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
928 ~(NFS_FABLKSIZE - 1);
929 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
930 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
931 if (nmp->nm_rsize == 0)
932 nmp->nm_rsize = fsp->fs_rtmax;
934 if (nmp->nm_rsize < NFS_FABLKSIZE)
935 nmp->nm_rsize = NFS_FABLKSIZE;
936 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
937 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
938 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
939 ~(NFS_DIRBLKSIZ - 1);
940 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
941 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
942 if (nmp->nm_readdirsize == 0)
943 nmp->nm_readdirsize = fsp->fs_rtmax;
945 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
946 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
947 if (fsp->fs_maxfilesize > 0 &&
948 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
949 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
950 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
951 nmp->nm_state |= NFSSTA_GOTFSINFO;
955 * Get a pointer to my IP addrress and return it.
956 * Return NULL if you can't find one.
959 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
961 struct sockaddr_in sad, *sin;
963 u_int8_t *retp = NULL;
964 static struct in_addr laddr;
968 * Loop up a route for the destination address.
970 if (nmp->nm_nam->sa_family == AF_INET) {
971 bzero(&sad, sizeof (sad));
972 sin = (struct sockaddr_in *)nmp->nm_nam;
973 sad.sin_family = AF_INET;
974 sad.sin_len = sizeof (struct sockaddr_in);
975 sad.sin_addr.s_addr = sin->sin_addr.s_addr;
976 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
977 rt = rtalloc1((struct sockaddr *)&sad, 0, 0UL);
979 if (rt->rt_ifp != NULL &&
980 rt->rt_ifa != NULL &&
981 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
982 rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
983 sin = (struct sockaddr_in *)
984 rt->rt_ifa->ifa_addr;
985 laddr.s_addr = sin->sin_addr.s_addr;
986 retp = (u_int8_t *)&laddr;
992 } else if (nmp->nm_nam->sa_family == AF_INET6) {
993 struct sockaddr_in6 sad6, *sin6;
994 static struct in6_addr laddr6;
996 bzero(&sad6, sizeof (sad6));
997 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
998 sad6.sin6_family = AF_INET6;
999 sad6.sin6_len = sizeof (struct sockaddr_in6);
1000 sad6.sin6_addr = sin6->sin6_addr;
1001 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1002 rt = rtalloc1((struct sockaddr *)&sad6, 0, 0UL);
1004 if (rt->rt_ifp != NULL &&
1005 rt->rt_ifa != NULL &&
1006 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1007 rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1008 sin6 = (struct sockaddr_in6 *)
1009 rt->rt_ifa->ifa_addr;
1010 laddr6 = sin6->sin6_addr;
1011 retp = (u_int8_t *)&laddr6;
1023 * Copy NFS uid, gids from the cred structure.
1026 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1030 KASSERT(cr->cr_ngroups >= 0,
1031 ("newnfs_copyincred: negative cr_ngroups"));
1032 nfscr->nfsc_uid = cr->cr_uid;
1033 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1034 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1035 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1040 * Do any client specific initialization.
1045 static int inited = 0;
1051 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1055 * Check each of the attributes to be set, to ensure they aren't already
1056 * the correct value. Disable setting ones already correct.
1059 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1062 if (vap->va_mode != (mode_t)VNOVAL) {
1063 if (vap->va_mode == nvap->na_mode)
1064 vap->va_mode = (mode_t)VNOVAL;
1066 if (vap->va_uid != (uid_t)VNOVAL) {
1067 if (vap->va_uid == nvap->na_uid)
1068 vap->va_uid = (uid_t)VNOVAL;
1070 if (vap->va_gid != (gid_t)VNOVAL) {
1071 if (vap->va_gid == nvap->na_gid)
1072 vap->va_gid = (gid_t)VNOVAL;
1074 if (vap->va_size != VNOVAL) {
1075 if (vap->va_size == nvap->na_size)
1076 vap->va_size = VNOVAL;
1080 * We are normally called with only a partially initialized
1081 * VAP. Since the NFSv3 spec says that server may use the
1082 * file attributes to store the verifier, the spec requires
1083 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1084 * in atime, but we can't really assume that all servers will
1085 * so we ensure that our SETATTR sets both atime and mtime.
1087 if (vap->va_mtime.tv_sec == VNOVAL)
1088 vfs_timestamp(&vap->va_mtime);
1089 if (vap->va_atime.tv_sec == VNOVAL)
1090 vap->va_atime = vap->va_mtime;
1095 * Map nfsv4 errors to errno.h errors.
1096 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1097 * error should only be returned for the Open, Create and Setattr Ops.
1098 * As such, most calls can just pass in 0 for those arguments.
1101 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1112 case NFSERR_BADOWNER:
1113 tprintf(p, LOG_INFO,
1114 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1117 case NFSERR_STALECLIENTID:
1118 case NFSERR_STALESTATEID:
1119 case NFSERR_EXPIRED:
1120 case NFSERR_BADSTATEID:
1121 printf("nfsv4 recover err returned %d\n", error);
1123 case NFSERR_BADHANDLE:
1124 case NFSERR_SERVERFAULT:
1125 case NFSERR_BADTYPE:
1126 case NFSERR_FHEXPIRED:
1127 case NFSERR_RESOURCE:
1129 case NFSERR_NOFILEHANDLE:
1130 case NFSERR_MINORVERMISMATCH:
1131 case NFSERR_OLDSTATEID:
1132 case NFSERR_BADSEQID:
1133 case NFSERR_LEASEMOVED:
1134 case NFSERR_RECLAIMBAD:
1136 case NFSERR_BADCHAR:
1137 case NFSERR_BADNAME:
1138 case NFSERR_OPILLEGAL:
1139 printf("nfsv4 client/server protocol prob err=%d\n",
1143 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1149 * Locate a process by number; return only "live" processes -- i.e., neither
1150 * zombies nor newly born but incompletely initialized processes. By not
1151 * returning processes in the PRS_NEW state, we allow callers to avoid
1152 * testing for that condition to avoid dereferencing p_ucred, et al.
1153 * Identical to pfind() in kern_proc.c, except it assume the list is
1156 static struct proc *
1157 pfind_locked(pid_t pid)
1161 LIST_FOREACH(p, PIDHASH(pid), p_hash)
1162 if (p->p_pid == pid) {
1164 if (p->p_state == PRS_NEW) {
1174 * Check to see if the process for this owner exists. Return 1 if it doesn't
1178 nfscl_procdoesntexist(u_int8_t *own)
1188 tl.cval[0] = *own++;
1189 tl.cval[1] = *own++;
1190 tl.cval[2] = *own++;
1191 tl.cval[3] = *own++;
1193 p = pfind_locked(pid);
1196 if (p->p_stats == NULL) {
1200 tl.cval[0] = *own++;
1201 tl.cval[1] = *own++;
1202 tl.cval[2] = *own++;
1203 tl.cval[3] = *own++;
1204 if (tl.lval != p->p_stats->p_start.tv_sec) {
1207 tl.cval[0] = *own++;
1208 tl.cval[1] = *own++;
1209 tl.cval[2] = *own++;
1211 if (tl.lval != p->p_stats->p_start.tv_usec)
1219 * - nfs pseudo system call for the client
1225 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1228 struct nfscbd_args nfscbdarg;
1229 struct nfsd_nfscbd_args nfscbdarg2;
1232 if (uap->flag & NFSSVC_CBADDSOCK) {
1233 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1237 * Since we don't know what rights might be required,
1238 * pretend that we need them all. It is better to be too
1239 * careful than too reckless.
1241 if ((error = fget(td, nfscbdarg.sock, CAP_SOCK_ALL, &fp))
1245 if (fp->f_type != DTYPE_SOCKET) {
1249 error = nfscbd_addsock(fp);
1251 if (!error && nfscl_enablecallb == 0) {
1252 nfsv4_cbport = nfscbdarg.port;
1253 nfscl_enablecallb = 1;
1255 } else if (uap->flag & NFSSVC_NFSCBD) {
1256 if (uap->argp == NULL)
1258 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1259 sizeof(nfscbdarg2));
1262 error = nfscbd_nfsd(td, &nfscbdarg2);
1269 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1272 * Called once to initialize data structures...
1275 nfscl_modevent(module_t mod, int type, void *data)
1278 static int loaded = 0;
1285 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1287 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1292 ncl_call_invalcaches = ncl_invalcaches;
1293 nfsd_call_nfscl = nfssvc_nfscl;
1298 if (nfs_numnfscbd != 0) {
1304 * XXX: Unloading of nfscl module is unsupported.
1307 ncl_call_invalcaches = NULL;
1308 nfsd_call_nfscl = NULL;
1309 /* and get rid of the mutexes */
1310 mtx_destroy(&nfs_clstate_mutex);
1311 mtx_destroy(&ncl_iod_mutex);
1323 static moduledata_t nfscl_mod = {
1328 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1330 /* So that loader and kldload(2) can find us, wherever we are.. */
1331 MODULE_VERSION(nfscl, 1);
1332 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1333 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1334 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1335 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);