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/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.
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);
200 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
202 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
204 uma_zfree(newnfsnode_zone, np);
205 FREE((caddr_t)nfhp, M_NFSFH);
209 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
210 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
214 * Initialize the mutex even if the vnode is going to be a loser.
215 * This simplifies the logic in reclaim, which can then unconditionally
216 * destroy the mutex (in the case of the loser, or if hash_insert
217 * happened to return an error no special casing is needed).
219 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
222 * Are we getting the root? If so, make sure the vnode flags
225 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
226 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
227 if (vp->v_type == VNON)
229 vp->v_vflag |= VV_ROOT;
234 * For NFSv4, we have to attach the directory file handle and
235 * file name, so that Open Ops can be done later.
237 if (nmp->nm_flag & NFSMNT_NFSV4) {
238 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
239 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
241 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
242 np->n_v4->n4_namelen = cnp->cn_namelen;
243 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
244 dnp->n_fhp->nfh_len);
245 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
252 * NFS supports recursive and shared locking.
254 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
257 error = insmntque(vp, mntp);
260 mtx_destroy(&np->n_mtx);
261 FREE((caddr_t)nfhp, M_NFSFH);
262 if (np->n_v4 != NULL)
263 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
264 uma_zfree(newnfsnode_zone, np);
267 error = vfs_hash_insert(vp, hash, lkflags,
268 td, &nvp, newnfs_vncmpf, nfhp);
273 /* vfs_hash_insert() vput()'s the losing vnode */
282 * Anothe variant of nfs_nget(). This one is only used by reopen. It
283 * takes almost the same args as nfs_nget(), but only succeeds if an entry
284 * exists in the cache. (Since files should already be "open" with a
285 * vnode ref cnt on the node when reopen calls this, it should always
287 * Also, don't get a vnode lock, since it may already be locked by some
288 * other process that is handling it. This is ok, since all other threads
289 * on the client are blocked by the nfsc_lock being exclusively held by the
290 * caller of this function.
293 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
294 struct thread *td, struct nfsnode **npp)
302 /* For forced dismounts, just return error. */
303 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
305 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
307 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
308 nfhp->nfh_len = fhsize;
310 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
313 * First, try to get the vnode locked, but don't block for the lock.
315 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
316 newnfs_vncmpf, nfhp);
317 if (error == 0 && nvp != NULL) {
318 NFSVOPUNLOCK(nvp, 0);
319 } else if (error == EBUSY) {
321 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
322 * and lock the vnode, but just get a v_usecount on it.
323 * LK_NOWAIT is set so that when vget() returns ENOENT,
324 * vfs_hash_get() fails instead of looping.
325 * If this succeeds, it is safe so long as a vflush() with
326 * FORCECLOSE has not been done. Since the Renew thread is
327 * stopped and the MNTK_UNMOUNTF flag is set before doing
328 * a vflush() with FORCECLOSE, we should be ok here.
330 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
333 error = vfs_hash_get(mntp, hash,
334 (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
335 newnfs_vncmpf, nfhp);
348 * Load the attribute cache (that lives in the nfsnode entry) with
349 * the attributes of the second argument and
351 * copy the attributes to *vaper
352 * Similar to nfs_loadattrcache(), except the attributes are passed in
353 * instead of being parsed out of the mbuf list.
356 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
357 void *stuff, int writeattr, int dontshrink)
359 struct vnode *vp = *vpp;
360 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
362 struct nfsmount *nmp;
363 struct timespec mtime_save;
368 * If v_type == VNON it is a new node, so fill in the v_type,
369 * n_mtime fields. Check to see if it represents a special
370 * device, and if so, check for a possible alias. Once the
371 * correct vnode has been obtained, fill in the rest of the
376 if (vp->v_type != nvap->va_type) {
377 vp->v_type = nvap->va_type;
378 if (vp->v_type == VFIFO)
379 vp->v_op = &newnfs_fifoops;
380 np->n_mtime = nvap->va_mtime;
382 nmp = VFSTONFS(vp->v_mount);
383 vap = &np->n_vattr.na_vattr;
384 mtime_save = vap->va_mtime;
386 np->n_vattr.na_filerev = nap->na_filerev;
387 np->n_vattr.na_size = nap->na_size;
388 np->n_vattr.na_mtime = nap->na_mtime;
389 np->n_vattr.na_ctime = nap->na_ctime;
390 np->n_vattr.na_fsid = nap->na_fsid;
391 np->n_vattr.na_mode = nap->na_mode;
393 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
394 sizeof (struct nfsvattr));
398 * For NFSv4, if the node's fsid is not equal to the mount point's
399 * fsid, return the low order 32bits of the node's fsid. This
400 * allows getcwd(3) to work. There is a chance that the fsid might
401 * be the same as a local fs, but since this is in an NFS mount
402 * point, I don't think that will cause any problems?
404 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
405 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
406 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
408 * va_fsid needs to be set to some value derived from
409 * np->n_vattr.na_filesid that is not equal
410 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
411 * from the value used for the top level server volume
412 * in the mounted subtree.
414 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
415 (uint32_t)np->n_vattr.na_filesid[0])
416 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
418 vap->va_fsid = (uint32_t)hash32_buf(
419 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
421 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
422 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 vnode_pager_setsize(vp, np->n_size);
437 } else if (np->n_flag & NMODIFIED) {
439 * We've modified the file: Use the larger
440 * of our size, and the server's size.
442 if (vap->va_size < np->n_size) {
443 vap->va_size = np->n_size;
445 np->n_size = vap->va_size;
446 np->n_flag |= NSIZECHANGED;
448 vnode_pager_setsize(vp, np->n_size);
449 } else if (vap->va_size < np->n_size) {
451 * When shrinking the size, the call to
452 * vnode_pager_setsize() cannot be done
453 * with the mutex held, so delay it until
454 * after the mtx_unlock call.
456 nsize = np->n_size = vap->va_size;
457 np->n_flag |= NSIZECHANGED;
460 np->n_size = vap->va_size;
461 np->n_flag |= NSIZECHANGED;
462 vnode_pager_setsize(vp, np->n_size);
465 np->n_size = vap->va_size;
469 * The following checks are added to prevent a race between (say)
470 * a READDIR+ and a WRITE.
471 * READDIR+, WRITE requests sent out.
472 * READDIR+ resp, WRITE resp received on client.
473 * However, the WRITE resp was handled before the READDIR+ resp
474 * causing the post op attrs from the write to be loaded first
475 * and the attrs from the READDIR+ to be loaded later. If this
476 * happens, we have stale attrs loaded into the attrcache.
477 * We detect this by for the mtime moving back. We invalidate the
478 * attrcache when this happens.
480 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
481 /* Size changed or mtime went backwards */
483 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
486 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
487 if (np->n_flag & NCHG) {
488 if (np->n_flag & NACC)
489 vaper->va_atime = np->n_atim;
490 if (np->n_flag & NUPD)
491 vaper->va_mtime = np->n_mtim;
495 if (np->n_attrstamp != 0)
496 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
500 vnode_pager_setsize(vp, nsize);
505 * Fill in the client id name. For these bytes:
506 * 1 - they must be unique
507 * 2 - they should be persistent across client reboots
508 * 1 is more critical than 2
509 * Use the mount point's unique id plus either the uuid or, if that
510 * isn't set, random junk.
513 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
518 * First, put in the 64bit mount point identifier.
520 if (idlen >= sizeof (u_int64_t)) {
521 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
522 cp += sizeof (u_int64_t);
523 idlen -= sizeof (u_int64_t);
527 * If uuid is non-zero length, use it.
529 uuidlen = strlen(uuid);
530 if (uuidlen > 0 && idlen >= uuidlen) {
531 NFSBCOPY(uuid, cp, uuidlen);
537 * This only normally happens if the uuid isn't set.
540 *cp++ = (u_int8_t)(arc4random() % 256);
546 * Fill in a lock owner name. For now, pid + the process's creation time.
549 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
559 bzero(cp, NFSV4CL_LOCKNAMELEN);
562 if ((flags & F_POSIX) != 0) {
563 p = (struct proc *)id;
569 tl.lval = p->p_stats->p_start.tv_sec;
574 tl.lval = p->p_stats->p_start.tv_usec;
579 } else if ((flags & F_FLOCK) != 0) {
580 bcopy(&id, cp, sizeof(id));
581 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
583 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
584 bzero(cp, NFSV4CL_LOCKNAMELEN);
589 * Find the parent process for the thread passed in as an argument.
590 * If none exists, return NULL, otherwise return a thread for the parent.
591 * (Can be any of the threads, since it is only used for td->td_proc.)
594 nfscl_getparent(struct thread *td)
607 ptd = TAILQ_FIRST(&p->p_threads);
612 * Start up the renew kernel thread.
615 start_nfscl(void *arg)
617 struct nfsclclient *clp;
620 clp = (struct nfsclclient *)arg;
621 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
622 nfscl_renewthread(clp, td);
627 nfscl_start_renewthread(struct nfsclclient *clp)
630 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
636 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
637 * as the first Op after PutFH.
638 * (For NFSv4, the postop attributes are after the Op, so they can't be
639 * parsed here. A separate call to nfscl_postop_attr() is required.)
642 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
643 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
646 struct nfsnode *np = VTONFS(vp);
647 struct nfsvattr nfsva;
650 if (wccflagp != NULL)
652 if (nd->nd_flag & ND_NFSV3) {
654 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
655 if (*tl == newnfs_true) {
656 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
657 if (wccflagp != NULL) {
658 mtx_lock(&np->n_mtx);
659 *wccflagp = (np->n_mtime.tv_sec ==
660 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
661 np->n_mtime.tv_nsec ==
662 fxdr_unsigned(u_int32_t, *(tl + 3)));
663 mtx_unlock(&np->n_mtx);
666 error = nfscl_postop_attr(nd, nap, flagp, stuff);
667 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
668 == (ND_NFSV4 | ND_V4WCCATTR)) {
669 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
670 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
671 NULL, NULL, NULL, NULL, NULL);
675 * Get rid of Op# and status for next op.
677 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
679 nd->nd_flag |= ND_NOMOREDATA;
680 if (wccflagp != NULL &&
681 nfsva.na_vattr.va_mtime.tv_sec != 0) {
682 mtx_lock(&np->n_mtx);
683 *wccflagp = (np->n_mtime.tv_sec ==
684 nfsva.na_vattr.va_mtime.tv_sec &&
685 np->n_mtime.tv_nsec ==
686 nfsva.na_vattr.va_mtime.tv_sec);
687 mtx_unlock(&np->n_mtx);
695 * Get postop attributes.
698 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
705 if (nd->nd_flag & ND_NOMOREDATA)
707 if (nd->nd_flag & ND_NFSV3) {
708 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
709 *retp = fxdr_unsigned(int, *tl);
710 } else if (nd->nd_flag & ND_NFSV4) {
712 * For NFSv4, the postop attr are at the end, so no point
713 * in looking if nd_repstat != 0.
715 if (!nd->nd_repstat) {
716 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
718 /* should never happen since nd_repstat != 0 */
719 nd->nd_flag |= ND_NOMOREDATA;
723 } else if (!nd->nd_repstat) {
724 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
728 error = nfsm_loadattr(nd, nap);
737 * Fill in the setable attributes. The full argument indicates whether
738 * to fill in them all or just mode and time.
741 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
742 struct vnode *vp, int flags, u_int32_t rdev)
745 struct nfsv2_sattr *sp;
746 nfsattrbit_t attrbits;
748 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
750 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
751 if (vap->va_mode == (mode_t)VNOVAL)
752 sp->sa_mode = newnfs_xdrneg1;
754 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
755 if (vap->va_uid == (uid_t)VNOVAL)
756 sp->sa_uid = newnfs_xdrneg1;
758 sp->sa_uid = txdr_unsigned(vap->va_uid);
759 if (vap->va_gid == (gid_t)VNOVAL)
760 sp->sa_gid = newnfs_xdrneg1;
762 sp->sa_gid = txdr_unsigned(vap->va_gid);
763 if (flags & NFSSATTR_SIZE0)
765 else if (flags & NFSSATTR_SIZENEG1)
766 sp->sa_size = newnfs_xdrneg1;
767 else if (flags & NFSSATTR_SIZERDEV)
768 sp->sa_size = txdr_unsigned(rdev);
770 sp->sa_size = txdr_unsigned(vap->va_size);
771 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
772 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
775 if (vap->va_mode != (mode_t)VNOVAL) {
776 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
778 *tl = txdr_unsigned(vap->va_mode);
780 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
783 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
784 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
786 *tl = txdr_unsigned(vap->va_uid);
788 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
791 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
792 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
794 *tl = txdr_unsigned(vap->va_gid);
796 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
799 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
800 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
802 txdr_hyper(vap->va_size, tl);
804 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
807 if (vap->va_atime.tv_sec != VNOVAL) {
808 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
809 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
810 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
811 txdr_nfsv3time(&vap->va_atime, tl);
813 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
814 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
817 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
818 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
820 if (vap->va_mtime.tv_sec != VNOVAL) {
821 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
822 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
823 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
824 txdr_nfsv3time(&vap->va_mtime, tl);
826 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
827 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
830 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
831 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
835 NFSZERO_ATTRBIT(&attrbits);
836 if (vap->va_mode != (mode_t)VNOVAL)
837 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
838 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
839 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
840 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
841 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
842 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
843 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
844 if (vap->va_atime.tv_sec != VNOVAL)
845 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
846 if (vap->va_mtime.tv_sec != VNOVAL)
847 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
848 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
849 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
855 * nfscl_request() - mostly a wrapper for newnfs_request().
858 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
859 struct ucred *cred, void *stuff)
862 struct nfsmount *nmp;
864 nmp = VFSTONFS(vp->v_mount);
865 if (nd->nd_flag & ND_NFSV4)
867 else if (nd->nd_flag & ND_NFSV3)
871 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
872 NFS_PROG, vers, NULL, 1, NULL, NULL);
877 * fill in this bsden's variant of statfs using nfsstatfs.
880 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
882 struct statfs *sbp = (struct statfs *)statfs;
884 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
885 sbp->f_bsize = NFS_FABLKSIZE;
886 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
887 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
889 * Although sf_abytes is uint64_t and f_bavail is int64_t,
890 * the value after dividing by NFS_FABLKSIZE is small
891 * enough that it will fit in 63bits, so it is ok to
892 * assign it to f_bavail without fear that it will become
895 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
896 sbp->f_files = sfp->sf_tfiles;
897 /* Since f_ffree is int64_t, clip it to 63bits. */
898 if (sfp->sf_ffiles > INT64_MAX)
899 sbp->f_ffree = INT64_MAX;
901 sbp->f_ffree = sfp->sf_ffiles;
902 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
904 * The type casts to (int32_t) ensure that this code is
905 * compatible with the old NFS client, in that it will
906 * propagate bit31 to the high order bits. This may or may
907 * not be correct for NFSv2, but since it is a legacy
908 * environment, I'd rather retain backwards compatibility.
910 sbp->f_bsize = (int32_t)sfp->sf_bsize;
911 sbp->f_blocks = (int32_t)sfp->sf_blocks;
912 sbp->f_bfree = (int32_t)sfp->sf_bfree;
913 sbp->f_bavail = (int32_t)sfp->sf_bavail;
920 * Use the fsinfo stuff to update the mount point.
923 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
926 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
927 fsp->fs_wtpref >= NFS_FABLKSIZE)
928 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
929 ~(NFS_FABLKSIZE - 1);
930 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
931 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
932 if (nmp->nm_wsize == 0)
933 nmp->nm_wsize = fsp->fs_wtmax;
935 if (nmp->nm_wsize < NFS_FABLKSIZE)
936 nmp->nm_wsize = NFS_FABLKSIZE;
937 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
938 fsp->fs_rtpref >= NFS_FABLKSIZE)
939 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
940 ~(NFS_FABLKSIZE - 1);
941 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
942 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
943 if (nmp->nm_rsize == 0)
944 nmp->nm_rsize = fsp->fs_rtmax;
946 if (nmp->nm_rsize < NFS_FABLKSIZE)
947 nmp->nm_rsize = NFS_FABLKSIZE;
948 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
949 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
950 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
951 ~(NFS_DIRBLKSIZ - 1);
952 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
953 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
954 if (nmp->nm_readdirsize == 0)
955 nmp->nm_readdirsize = fsp->fs_rtmax;
957 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
958 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
959 if (fsp->fs_maxfilesize > 0 &&
960 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
961 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
962 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
963 nmp->nm_state |= NFSSTA_GOTFSINFO;
967 * Get a pointer to my IP addrress and return it.
968 * Return NULL if you can't find one.
971 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
973 struct sockaddr_in sad, *sin;
975 u_int8_t *retp = NULL;
976 static struct in_addr laddr;
980 * Loop up a route for the destination address.
982 if (nmp->nm_nam->sa_family == AF_INET) {
983 bzero(&sad, sizeof (sad));
984 sin = (struct sockaddr_in *)nmp->nm_nam;
985 sad.sin_family = AF_INET;
986 sad.sin_len = sizeof (struct sockaddr_in);
987 sad.sin_addr.s_addr = sin->sin_addr.s_addr;
988 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
989 rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
990 curthread->td_proc->p_fibnum);
992 if (rt->rt_ifp != NULL &&
993 rt->rt_ifa != NULL &&
994 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
995 rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
996 sin = (struct sockaddr_in *)
997 rt->rt_ifa->ifa_addr;
998 laddr.s_addr = sin->sin_addr.s_addr;
999 retp = (u_int8_t *)&laddr;
1005 } else if (nmp->nm_nam->sa_family == AF_INET6) {
1006 struct sockaddr_in6 sad6, *sin6;
1007 static struct in6_addr laddr6;
1009 bzero(&sad6, sizeof (sad6));
1010 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1011 sad6.sin6_family = AF_INET6;
1012 sad6.sin6_len = sizeof (struct sockaddr_in6);
1013 sad6.sin6_addr = sin6->sin6_addr;
1014 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1015 rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1016 curthread->td_proc->p_fibnum);
1018 if (rt->rt_ifp != NULL &&
1019 rt->rt_ifa != NULL &&
1020 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1021 rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1022 sin6 = (struct sockaddr_in6 *)
1023 rt->rt_ifa->ifa_addr;
1024 laddr6 = sin6->sin6_addr;
1025 retp = (u_int8_t *)&laddr6;
1037 * Copy NFS uid, gids from the cred structure.
1040 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1044 KASSERT(cr->cr_ngroups >= 0,
1045 ("newnfs_copyincred: negative cr_ngroups"));
1046 nfscr->nfsc_uid = cr->cr_uid;
1047 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1048 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1049 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1054 * Do any client specific initialization.
1059 static int inited = 0;
1065 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1069 * Check each of the attributes to be set, to ensure they aren't already
1070 * the correct value. Disable setting ones already correct.
1073 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1076 if (vap->va_mode != (mode_t)VNOVAL) {
1077 if (vap->va_mode == nvap->na_mode)
1078 vap->va_mode = (mode_t)VNOVAL;
1080 if (vap->va_uid != (uid_t)VNOVAL) {
1081 if (vap->va_uid == nvap->na_uid)
1082 vap->va_uid = (uid_t)VNOVAL;
1084 if (vap->va_gid != (gid_t)VNOVAL) {
1085 if (vap->va_gid == nvap->na_gid)
1086 vap->va_gid = (gid_t)VNOVAL;
1088 if (vap->va_size != VNOVAL) {
1089 if (vap->va_size == nvap->na_size)
1090 vap->va_size = VNOVAL;
1094 * We are normally called with only a partially initialized
1095 * VAP. Since the NFSv3 spec says that server may use the
1096 * file attributes to store the verifier, the spec requires
1097 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1098 * in atime, but we can't really assume that all servers will
1099 * so we ensure that our SETATTR sets both atime and mtime.
1100 * Set the VA_UTIMES_NULL flag for this case, so that
1101 * the server's time will be used. This is needed to
1102 * work around a bug in some Solaris servers, where
1103 * setting the time TOCLIENT causes the Setattr RPC
1104 * to return NFS_OK, but not set va_mode.
1106 if (vap->va_mtime.tv_sec == VNOVAL) {
1107 vfs_timestamp(&vap->va_mtime);
1108 vap->va_vaflags |= VA_UTIMES_NULL;
1110 if (vap->va_atime.tv_sec == VNOVAL)
1111 vap->va_atime = vap->va_mtime;
1116 * Map nfsv4 errors to errno.h errors.
1117 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1118 * error should only be returned for the Open, Create and Setattr Ops.
1119 * As such, most calls can just pass in 0 for those arguments.
1122 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1133 case NFSERR_BADOWNER:
1134 tprintf(p, LOG_INFO,
1135 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1138 case NFSERR_BADNAME:
1139 case NFSERR_BADCHAR:
1140 printf("nfsv4 char/name not handled by server\n");
1142 case NFSERR_STALECLIENTID:
1143 case NFSERR_STALESTATEID:
1144 case NFSERR_EXPIRED:
1145 case NFSERR_BADSTATEID:
1146 case NFSERR_BADSESSION:
1147 printf("nfsv4 recover err returned %d\n", error);
1149 case NFSERR_BADHANDLE:
1150 case NFSERR_SERVERFAULT:
1151 case NFSERR_BADTYPE:
1152 case NFSERR_FHEXPIRED:
1153 case NFSERR_RESOURCE:
1155 case NFSERR_NOFILEHANDLE:
1156 case NFSERR_MINORVERMISMATCH:
1157 case NFSERR_OLDSTATEID:
1158 case NFSERR_BADSEQID:
1159 case NFSERR_LEASEMOVED:
1160 case NFSERR_RECLAIMBAD:
1162 case NFSERR_OPILLEGAL:
1163 printf("nfsv4 client/server protocol prob err=%d\n",
1167 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1173 * Check to see if the process for this owner exists. Return 1 if it doesn't
1177 nfscl_procdoesntexist(u_int8_t *own)
1187 tl.cval[0] = *own++;
1188 tl.cval[1] = *own++;
1189 tl.cval[2] = *own++;
1190 tl.cval[3] = *own++;
1192 p = pfind_locked(pid);
1195 if (p->p_stats == NULL) {
1199 tl.cval[0] = *own++;
1200 tl.cval[1] = *own++;
1201 tl.cval[2] = *own++;
1202 tl.cval[3] = *own++;
1203 if (tl.lval != p->p_stats->p_start.tv_sec) {
1206 tl.cval[0] = *own++;
1207 tl.cval[1] = *own++;
1208 tl.cval[2] = *own++;
1210 if (tl.lval != p->p_stats->p_start.tv_usec)
1218 * - nfs pseudo system call for the client
1224 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1227 struct nfscbd_args nfscbdarg;
1228 struct nfsd_nfscbd_args nfscbdarg2;
1229 struct nameidata nd;
1230 struct nfscl_dumpmntopts dumpmntopts;
1231 cap_rights_t rights;
1235 if (uap->flag & NFSSVC_CBADDSOCK) {
1236 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1240 * Since we don't know what rights might be required,
1241 * pretend that we need them all. It is better to be too
1242 * careful than too reckless.
1244 error = fget(td, nfscbdarg.sock,
1245 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1248 if (fp->f_type != DTYPE_SOCKET) {
1252 error = nfscbd_addsock(fp);
1254 if (!error && nfscl_enablecallb == 0) {
1255 nfsv4_cbport = nfscbdarg.port;
1256 nfscl_enablecallb = 1;
1258 } else if (uap->flag & NFSSVC_NFSCBD) {
1259 if (uap->argp == NULL)
1261 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1262 sizeof(nfscbdarg2));
1265 error = nfscbd_nfsd(td, &nfscbdarg2);
1266 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1267 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1268 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1269 dumpmntopts.ndmnt_blen > 1024))
1272 error = nfsrv_lookupfilename(&nd,
1273 dumpmntopts.ndmnt_fname, td);
1274 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1280 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1281 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1282 dumpmntopts.ndmnt_blen);
1284 error = copyout(buf, dumpmntopts.ndmnt_buf,
1285 dumpmntopts.ndmnt_blen);
1294 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1297 * Called once to initialize data structures...
1300 nfscl_modevent(module_t mod, int type, void *data)
1303 static int loaded = 0;
1310 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1312 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1317 ncl_call_invalcaches = ncl_invalcaches;
1318 nfsd_call_nfscl = nfssvc_nfscl;
1323 if (nfs_numnfscbd != 0) {
1329 * XXX: Unloading of nfscl module is unsupported.
1332 ncl_call_invalcaches = NULL;
1333 nfsd_call_nfscl = NULL;
1334 /* and get rid of the mutexes */
1335 mtx_destroy(&nfs_clstate_mutex);
1336 mtx_destroy(&ncl_iod_mutex);
1348 static moduledata_t nfscl_mod = {
1353 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1355 /* So that loader and kldload(2) can find us, wherever we are.. */
1356 MODULE_VERSION(nfscl, 1);
1357 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1358 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1359 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1360 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);