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("$FreeBSD$");
38 * generally, I don't like #includes inside .h files, but it seems to
39 * be the easiest way to handle the port.
42 #include <fs/nfs/nfsport.h>
43 #include <netinet/if_ether.h>
44 #include <net/if_types.h>
46 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
47 extern struct vop_vector newnfs_vnodeops;
48 extern struct vop_vector newnfs_fifoops;
49 extern uma_zone_t newnfsnode_zone;
50 extern struct buf_ops buf_ops_newnfs;
51 extern int ncl_pbuf_freecnt;
52 extern short nfsv4_cbport;
53 extern int nfscl_enablecallb;
54 extern int nfs_numnfscbd;
55 extern int nfscl_inited;
56 struct mtx nfs_clstate_mutex;
57 struct mtx ncl_iod_mutex;
60 extern void (*ncl_call_invalcaches)(struct vnode *);
63 * Comparison function for vfs_hash functions.
66 newnfs_vncmpf(struct vnode *vp, void *arg)
68 struct nfsfh *nfhp = (struct nfsfh *)arg;
69 struct nfsnode *np = VTONFS(vp);
71 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
72 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
78 * Look up a vnode/nfsnode by file handle.
79 * Callers must check for mount points!!
80 * In all cases, a pointer to a
81 * nfsnode structure is returned.
82 * This variant takes a "struct nfsfh *" as second argument and uses
83 * that structure up, either by hanging off the nfsnode or FREEing it.
86 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
87 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
88 void *stuff, int lkflags)
90 struct nfsnode *np, *dnp;
91 struct vnode *vp, *nvp;
92 struct nfsv4node *newd, *oldd;
101 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
103 error = vfs_hash_get(mntp, hash, lkflags,
104 td, &nvp, newnfs_vncmpf, nfhp);
105 if (error == 0 && nvp != NULL) {
107 * I believe there is a slight chance that vgonel() could
108 * get called on this vnode between when vn_lock() drops
109 * the VI_LOCK() and vget() acquires it again, so that it
110 * hasn't yet had v_usecount incremented. If this were to
111 * happen, the VI_DOOMED flag would be set, so check for
112 * that here. Since we now have the v_usecount incremented,
113 * we should be ok until we vrele() it, if the VI_DOOMED
114 * flag isn't set now.
117 if ((nvp->v_iflag & VI_DOOMED)) {
126 FREE((caddr_t)nfhp, M_NFSFH);
132 * For NFSv4, check to see if it is the same name and
133 * replace the name, if it is different.
136 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
137 nvp->v_type == VREG &&
138 (np->n_v4->n4_namelen != cnp->cn_namelen ||
139 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
141 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
142 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
143 dnp->n_fhp->nfh_len))) {
144 MALLOC(newd, struct nfsv4node *,
145 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
146 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
148 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
149 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
150 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
152 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
153 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
154 dnp->n_fhp->nfh_len))) {
158 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
159 np->n_v4->n4_namelen = cnp->cn_namelen;
160 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
161 dnp->n_fhp->nfh_len);
162 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
168 FREE((caddr_t)newd, M_NFSV4NODE);
170 FREE((caddr_t)oldd, M_NFSV4NODE);
172 FREE((caddr_t)nfhp, M_NFSFH);
177 * Allocate before getnewvnode since doing so afterward
178 * might cause a bogus v_data pointer to get dereferenced
179 * elsewhere if zalloc should block.
181 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
183 error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
185 uma_zfree(newnfsnode_zone, np);
186 FREE((caddr_t)nfhp, M_NFSFH);
190 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
194 * Initialize the mutex even if the vnode is going to be a loser.
195 * This simplifies the logic in reclaim, which can then unconditionally
196 * destroy the mutex (in the case of the loser, or if hash_insert
197 * happened to return an error no special casing is needed).
199 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
202 * Are we getting the root? If so, make sure the vnode flags
205 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
206 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
207 if (vp->v_type == VNON)
209 vp->v_vflag |= VV_ROOT;
214 * For NFSv4, we have to attach the directory file handle and
215 * file name, so that Open Ops can be done later.
217 if (nmp->nm_flag & NFSMNT_NFSV4) {
218 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
219 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
221 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
222 np->n_v4->n4_namelen = cnp->cn_namelen;
223 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
224 dnp->n_fhp->nfh_len);
225 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
232 * NFS supports recursive and shared locking.
236 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
237 error = insmntque(vp, mntp);
240 mtx_destroy(&np->n_mtx);
241 FREE((caddr_t)nfhp, M_NFSFH);
242 if (np->n_v4 != NULL)
243 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
244 uma_zfree(newnfsnode_zone, np);
247 error = vfs_hash_insert(vp, hash, lkflags,
248 td, &nvp, newnfs_vncmpf, nfhp);
253 /* vfs_hash_insert() vput()'s the losing vnode */
262 * Anothe variant of nfs_nget(). This one is only used by reopen. It
263 * takes almost the same args as nfs_nget(), but only succeeds if an entry
264 * exists in the cache. (Since files should already be "open" with a
265 * vnode ref cnt on the node when reopen calls this, it should always
267 * Also, don't get a vnode lock, since it may already be locked by some
268 * other process that is handling it. This is ok, since all other threads
269 * on the client are blocked by the nfsc_lock being exclusively held by the
270 * caller of this function.
273 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
274 struct thread *td, struct nfsnode **npp)
282 /* For forced dismounts, just return error. */
283 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
285 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
287 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
288 nfhp->nfh_len = fhsize;
290 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
293 * First, try to get the vnode locked, but don't block for the lock.
295 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
296 newnfs_vncmpf, nfhp);
297 if (error == 0 && nvp != NULL) {
299 } else if (error == EBUSY) {
301 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
302 * and lock the vnode, but just get a v_usecount on it.
303 * LK_NOWAIT is set so that when vget() returns ENOENT,
304 * vfs_hash_get() fails instead of looping.
305 * If this succeeds, it is safe so long as a vflush() with
306 * FORCECLOSE has not been done. Since the Renew thread is
307 * stopped and the MNTK_UNMOUNTF flag is set before doing
308 * a vflush() with FORCECLOSE, we should be ok here.
310 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
313 error = vfs_hash_get(mntp, hash,
314 (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
315 newnfs_vncmpf, nfhp);
328 * Load the attribute cache (that lives in the nfsnode entry) with
329 * the attributes of the second argument and
331 * copy the attributes to *vaper
332 * Similar to nfs_loadattrcache(), except the attributes are passed in
333 * instead of being parsed out of the mbuf list.
336 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
337 void *stuff, int writeattr, int dontshrink)
339 struct vnode *vp = *vpp;
340 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
342 struct nfsmount *nmp;
343 struct timespec mtime_save;
346 * If v_type == VNON it is a new node, so fill in the v_type,
347 * n_mtime fields. Check to see if it represents a special
348 * device, and if so, check for a possible alias. Once the
349 * correct vnode has been obtained, fill in the rest of the
354 if (vp->v_type != nvap->va_type) {
355 vp->v_type = nvap->va_type;
356 if (vp->v_type == VFIFO)
357 vp->v_op = &newnfs_fifoops;
358 np->n_mtime = nvap->va_mtime;
360 nmp = VFSTONFS(vp->v_mount);
361 vap = &np->n_vattr.na_vattr;
362 mtime_save = vap->va_mtime;
364 np->n_vattr.na_filerev = nap->na_filerev;
365 np->n_vattr.na_size = nap->na_size;
366 np->n_vattr.na_mtime = nap->na_mtime;
367 np->n_vattr.na_ctime = nap->na_ctime;
368 np->n_vattr.na_fsid = nap->na_fsid;
370 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
371 sizeof (struct nfsvattr));
375 * For NFSv4, if the node's fsid is not equal to the mount point's
376 * fsid, return the low order 32bits of the node's fsid. This
377 * allows getcwd(3) to work. There is a chance that the fsid might
378 * be the same as a local fs, but since this is in an NFS mount
379 * point, I don't think that will cause any problems?
381 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
382 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
383 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
385 * va_fsid needs to be set to some value derived from
386 * np->n_vattr.na_filesid that is not equal
387 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
388 * from the value used for the top level server volume
389 * in the mounted subtree.
391 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
392 (uint32_t)np->n_vattr.na_filesid[0])
393 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
395 vap->va_fsid = (uint32_t)hash32_buf(
396 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
398 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
399 np->n_attrstamp = time_second;
400 if (vap->va_size != np->n_size) {
401 if (vap->va_type == VREG) {
402 if (dontshrink && vap->va_size < np->n_size) {
404 * We've been told not to shrink the file;
405 * zero np->n_attrstamp to indicate that
406 * the attributes are stale.
408 vap->va_size = np->n_size;
410 } else if (np->n_flag & NMODIFIED) {
412 * We've modified the file: Use the larger
413 * of our size, and the server's size.
415 if (vap->va_size < np->n_size) {
416 vap->va_size = np->n_size;
418 np->n_size = vap->va_size;
419 np->n_flag |= NSIZECHANGED;
422 np->n_size = vap->va_size;
423 np->n_flag |= NSIZECHANGED;
425 vnode_pager_setsize(vp, np->n_size);
427 np->n_size = vap->va_size;
431 * The following checks are added to prevent a race between (say)
432 * a READDIR+ and a WRITE.
433 * READDIR+, WRITE requests sent out.
434 * READDIR+ resp, WRITE resp received on client.
435 * However, the WRITE resp was handled before the READDIR+ resp
436 * causing the post op attrs from the write to be loaded first
437 * and the attrs from the READDIR+ to be loaded later. If this
438 * happens, we have stale attrs loaded into the attrcache.
439 * We detect this by for the mtime moving back. We invalidate the
440 * attrcache when this happens.
442 if (timespeccmp(&mtime_save, &vap->va_mtime, >))
443 /* Size changed or mtime went backwards */
446 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
447 if (np->n_flag & NCHG) {
448 if (np->n_flag & NACC)
449 vaper->va_atime = np->n_atim;
450 if (np->n_flag & NUPD)
451 vaper->va_mtime = np->n_mtim;
459 * Fill in the client id name. For these bytes:
460 * 1 - they must be unique
461 * 2 - they should be persistent across client reboots
462 * 1 is more critical than 2
463 * Use the mount point's unique id plus either the uuid or, if that
464 * isn't set, random junk.
467 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
472 * First, put in the 64bit mount point identifier.
474 if (idlen >= sizeof (u_int64_t)) {
475 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
476 cp += sizeof (u_int64_t);
477 idlen -= sizeof (u_int64_t);
481 * If uuid is non-zero length, use it.
483 uuidlen = strlen(uuid);
484 if (uuidlen > 0 && idlen >= uuidlen) {
485 NFSBCOPY(uuid, cp, uuidlen);
491 * This only normally happens if the uuid isn't set.
494 *cp++ = (u_int8_t)(arc4random() % 256);
500 * Fill in a lock owner name. For now, pid + the process's creation time.
503 nfscl_filllockowner(struct thread *td, u_int8_t *cp)
518 printf("NULL pid\n");
527 if (p->p_stats == NULL) {
528 printf("pstats null\n");
532 tl.lval = p->p_stats->p_start.tv_sec;
537 tl.lval = p->p_stats->p_start.tv_usec;
545 * Find the parent process for the thread passed in as an argument.
546 * If none exists, return NULL, otherwise return a thread for the parent.
547 * (Can be any of the threads, since it is only used for td->td_proc.)
550 nfscl_getparent(struct thread *td)
563 ptd = TAILQ_FIRST(&p->p_threads);
568 * Start up the renew kernel thread.
571 start_nfscl(void *arg)
573 struct nfsclclient *clp;
576 clp = (struct nfsclclient *)arg;
577 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
578 nfscl_renewthread(clp, td);
583 nfscl_start_renewthread(struct nfsclclient *clp)
586 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
592 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
593 * as the first Op after PutFH.
594 * (For NFSv4, the postop attributes are after the Op, so they can't be
595 * parsed here. A separate call to nfscl_postop_attr() is required.)
598 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
599 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
602 struct nfsnode *np = VTONFS(vp);
603 struct nfsvattr nfsva;
606 if (wccflagp != NULL)
608 if (nd->nd_flag & ND_NFSV3) {
610 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
611 if (*tl == newnfs_true) {
612 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
613 if (wccflagp != NULL) {
614 mtx_lock(&np->n_mtx);
615 *wccflagp = (np->n_mtime.tv_sec ==
616 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
617 np->n_mtime.tv_nsec ==
618 fxdr_unsigned(u_int32_t, *(tl + 3)));
619 mtx_unlock(&np->n_mtx);
622 error = nfscl_postop_attr(nd, nap, flagp, stuff);
623 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
624 == (ND_NFSV4 | ND_V4WCCATTR)) {
625 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
626 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
627 NULL, NULL, NULL, NULL, NULL);
631 * Get rid of Op# and status for next op.
633 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
635 nd->nd_flag |= ND_NOMOREDATA;
636 if (wccflagp != NULL &&
637 nfsva.na_vattr.va_mtime.tv_sec != 0) {
638 mtx_lock(&np->n_mtx);
639 *wccflagp = (np->n_mtime.tv_sec ==
640 nfsva.na_vattr.va_mtime.tv_sec &&
641 np->n_mtime.tv_nsec ==
642 nfsva.na_vattr.va_mtime.tv_sec);
643 mtx_unlock(&np->n_mtx);
651 * Get postop attributes.
654 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
661 if (nd->nd_flag & ND_NOMOREDATA)
663 if (nd->nd_flag & ND_NFSV3) {
664 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
665 *retp = fxdr_unsigned(int, *tl);
666 } else if (nd->nd_flag & ND_NFSV4) {
668 * For NFSv4, the postop attr are at the end, so no point
669 * in looking if nd_repstat != 0.
671 if (!nd->nd_repstat) {
672 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
674 /* should never happen since nd_repstat != 0 */
675 nd->nd_flag |= ND_NOMOREDATA;
679 } else if (!nd->nd_repstat) {
680 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
684 error = nfsm_loadattr(nd, nap);
693 * Fill in the setable attributes. The full argument indicates whether
694 * to fill in them all or just mode and time.
697 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
698 struct vnode *vp, int flags, u_int32_t rdev)
701 struct nfsv2_sattr *sp;
702 nfsattrbit_t attrbits;
703 struct timeval curtime;
705 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
707 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
708 if (vap->va_mode == (mode_t)VNOVAL)
709 sp->sa_mode = newnfs_xdrneg1;
711 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
712 if (vap->va_uid == (uid_t)VNOVAL)
713 sp->sa_uid = newnfs_xdrneg1;
715 sp->sa_uid = txdr_unsigned(vap->va_uid);
716 if (vap->va_gid == (gid_t)VNOVAL)
717 sp->sa_gid = newnfs_xdrneg1;
719 sp->sa_gid = txdr_unsigned(vap->va_gid);
720 if (flags & NFSSATTR_SIZE0)
722 else if (flags & NFSSATTR_SIZENEG1)
723 sp->sa_size = newnfs_xdrneg1;
724 else if (flags & NFSSATTR_SIZERDEV)
725 sp->sa_size = txdr_unsigned(rdev);
727 sp->sa_size = txdr_unsigned(vap->va_size);
728 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
729 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
732 getmicrotime(&curtime);
733 if (vap->va_mode != (mode_t)VNOVAL) {
734 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
736 *tl = txdr_unsigned(vap->va_mode);
738 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
741 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
742 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
744 *tl = txdr_unsigned(vap->va_uid);
746 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
749 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
750 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
752 *tl = txdr_unsigned(vap->va_gid);
754 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
757 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
758 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
760 txdr_hyper(vap->va_size, tl);
762 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
765 if (vap->va_atime.tv_sec != VNOVAL) {
766 if (vap->va_atime.tv_sec != curtime.tv_sec) {
767 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
768 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
769 txdr_nfsv3time(&vap->va_atime, tl);
771 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
772 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
775 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
776 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
778 if (vap->va_mtime.tv_sec != VNOVAL) {
779 if (vap->va_mtime.tv_sec != curtime.tv_sec) {
780 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
781 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
782 txdr_nfsv3time(&vap->va_mtime, tl);
784 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
785 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
788 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
789 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
793 NFSZERO_ATTRBIT(&attrbits);
794 if (vap->va_mode != (mode_t)VNOVAL)
795 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
796 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
797 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
798 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
799 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
800 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
801 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
802 if (vap->va_atime.tv_sec != VNOVAL)
803 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
804 if (vap->va_mtime.tv_sec != VNOVAL)
805 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
806 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
807 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
813 * nfscl_request() - mostly a wrapper for newnfs_request().
816 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
817 struct ucred *cred, void *stuff)
820 struct nfsmount *nmp;
822 nmp = VFSTONFS(vp->v_mount);
823 if (nd->nd_flag & ND_NFSV4)
825 else if (nd->nd_flag & ND_NFSV3)
829 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
830 NFS_PROG, vers, NULL, 1, NULL);
835 * fill in this bsden's variant of statfs using nfsstatfs.
838 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
840 struct statfs *sbp = (struct statfs *)statfs;
843 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
844 sbp->f_bsize = NFS_FABLKSIZE;
845 tquad.qval = sfp->sf_tbytes;
846 sbp->f_blocks = (long)(tquad.qval / ((u_quad_t)NFS_FABLKSIZE));
847 tquad.qval = sfp->sf_fbytes;
848 sbp->f_bfree = (long)(tquad.qval / ((u_quad_t)NFS_FABLKSIZE));
849 tquad.qval = sfp->sf_abytes;
850 sbp->f_bavail = (long)(tquad.qval / ((u_quad_t)NFS_FABLKSIZE));
851 tquad.qval = sfp->sf_tfiles;
852 sbp->f_files = (tquad.lval[0] & 0x7fffffff);
853 tquad.qval = sfp->sf_ffiles;
854 sbp->f_ffree = (tquad.lval[0] & 0x7fffffff);
855 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
856 sbp->f_bsize = (int32_t)sfp->sf_bsize;
857 sbp->f_blocks = (int32_t)sfp->sf_blocks;
858 sbp->f_bfree = (int32_t)sfp->sf_bfree;
859 sbp->f_bavail = (int32_t)sfp->sf_bavail;
866 * Use the fsinfo stuff to update the mount point.
869 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
872 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
873 fsp->fs_wtpref >= NFS_FABLKSIZE)
874 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
875 ~(NFS_FABLKSIZE - 1);
876 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
877 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
878 if (nmp->nm_wsize == 0)
879 nmp->nm_wsize = fsp->fs_wtmax;
881 if (nmp->nm_wsize < NFS_FABLKSIZE)
882 nmp->nm_wsize = NFS_FABLKSIZE;
883 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
884 fsp->fs_rtpref >= NFS_FABLKSIZE)
885 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
886 ~(NFS_FABLKSIZE - 1);
887 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
888 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
889 if (nmp->nm_rsize == 0)
890 nmp->nm_rsize = fsp->fs_rtmax;
892 if (nmp->nm_rsize < NFS_FABLKSIZE)
893 nmp->nm_rsize = NFS_FABLKSIZE;
894 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
895 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
896 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
897 ~(NFS_DIRBLKSIZ - 1);
898 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
899 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
900 if (nmp->nm_readdirsize == 0)
901 nmp->nm_readdirsize = fsp->fs_rtmax;
903 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
904 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
905 if (fsp->fs_maxfilesize > 0 &&
906 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
907 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
908 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
909 nmp->nm_state |= NFSSTA_GOTFSINFO;
913 * Get a pointer to my IP addrress and return it.
914 * Return NULL if you can't find one.
917 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
919 struct sockaddr_in sad, *sin;
921 u_int8_t *retp = NULL;
922 static struct in_addr laddr;
926 * Loop up a route for the destination address.
928 if (nmp->nm_nam->sa_family == AF_INET) {
929 bzero(&sad, sizeof (sad));
930 sin = (struct sockaddr_in *)nmp->nm_nam;
931 sad.sin_family = AF_INET;
932 sad.sin_len = sizeof (struct sockaddr_in);
933 sad.sin_addr.s_addr = sin->sin_addr.s_addr;
934 rt = rtalloc1((struct sockaddr *)&sad, 0, 0UL);
936 if (rt->rt_ifp != NULL &&
937 rt->rt_ifa != NULL &&
938 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
939 rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
940 sin = (struct sockaddr_in *)
941 rt->rt_ifa->ifa_addr;
942 laddr.s_addr = sin->sin_addr.s_addr;
943 retp = (u_int8_t *)&laddr;
948 } else if (nmp->nm_nam->sa_family == AF_INET6) {
949 struct sockaddr_in6 sad6, *sin6;
950 static struct in6_addr laddr6;
952 bzero(&sad6, sizeof (sad6));
953 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
954 sad6.sin6_family = AF_INET6;
955 sad6.sin6_len = sizeof (struct sockaddr_in6);
956 sad6.sin6_addr = sin6->sin6_addr;
957 rt = rtalloc1((struct sockaddr *)&sad6, 0, 0UL);
959 if (rt->rt_ifp != NULL &&
960 rt->rt_ifa != NULL &&
961 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
962 rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
963 sin6 = (struct sockaddr_in6 *)
964 rt->rt_ifa->ifa_addr;
965 laddr6 = sin6->sin6_addr;
966 retp = (u_int8_t *)&laddr6;
977 * Copy NFS uid, gids from the cred structure.
980 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
984 KASSERT(cr->cr_ngroups >= 0,
985 ("newnfs_copyincred: negative cr_ngroups"));
986 nfscr->nfsc_uid = cr->cr_uid;
987 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
988 for (i = 0; i < nfscr->nfsc_ngroups; i++)
989 nfscr->nfsc_groups[i] = cr->cr_groups[i];
994 * Do any client specific initialization.
999 static int inited = 0;
1005 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1009 * Check each of the attributes to be set, to ensure they aren't already
1010 * the correct value. Disable setting ones already correct.
1013 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1016 if (vap->va_mode != (mode_t)VNOVAL) {
1017 if (vap->va_mode == nvap->na_mode)
1018 vap->va_mode = (mode_t)VNOVAL;
1020 if (vap->va_uid != (uid_t)VNOVAL) {
1021 if (vap->va_uid == nvap->na_uid)
1022 vap->va_uid = (uid_t)VNOVAL;
1024 if (vap->va_gid != (gid_t)VNOVAL) {
1025 if (vap->va_gid == nvap->na_gid)
1026 vap->va_gid = (gid_t)VNOVAL;
1028 if (vap->va_size != VNOVAL) {
1029 if (vap->va_size == nvap->na_size)
1030 vap->va_size = VNOVAL;
1034 * We are normally called with only a partially initialized
1035 * VAP. Since the NFSv3 spec says that server may use the
1036 * file attributes to store the verifier, the spec requires
1037 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1038 * in atime, but we can't really assume that all servers will
1039 * so we ensure that our SETATTR sets both atime and mtime.
1041 if (vap->va_mtime.tv_sec == VNOVAL)
1042 vfs_timestamp(&vap->va_mtime);
1043 if (vap->va_atime.tv_sec == VNOVAL)
1044 vap->va_atime = vap->va_mtime;
1049 * Map nfsv4 errors to errno.h errors.
1050 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1051 * error should only be returned for the Open, Create and Setattr Ops.
1052 * As such, most calls can just pass in 0 for those arguments.
1055 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1066 case NFSERR_BADOWNER:
1067 tprintf(p, LOG_INFO,
1068 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1071 case NFSERR_STALECLIENTID:
1072 case NFSERR_STALESTATEID:
1073 case NFSERR_EXPIRED:
1074 case NFSERR_BADSTATEID:
1075 printf("nfsv4 recover err returned %d\n", error);
1077 case NFSERR_BADHANDLE:
1078 case NFSERR_SERVERFAULT:
1079 case NFSERR_BADTYPE:
1080 case NFSERR_FHEXPIRED:
1081 case NFSERR_RESOURCE:
1083 case NFSERR_NOFILEHANDLE:
1084 case NFSERR_MINORVERMISMATCH:
1085 case NFSERR_OLDSTATEID:
1086 case NFSERR_BADSEQID:
1087 case NFSERR_LEASEMOVED:
1088 case NFSERR_RECLAIMBAD:
1090 case NFSERR_BADCHAR:
1091 case NFSERR_BADNAME:
1092 case NFSERR_OPILLEGAL:
1093 printf("nfsv4 client/server protocol prob err=%d\n",
1097 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1103 * Locate a process by number; return only "live" processes -- i.e., neither
1104 * zombies nor newly born but incompletely initialized processes. By not
1105 * returning processes in the PRS_NEW state, we allow callers to avoid
1106 * testing for that condition to avoid dereferencing p_ucred, et al.
1107 * Identical to pfind() in kern_proc.c, except it assume the list is
1110 static struct proc *
1111 pfind_locked(pid_t pid)
1115 LIST_FOREACH(p, PIDHASH(pid), p_hash)
1116 if (p->p_pid == pid) {
1117 if (p->p_state == PRS_NEW) {
1128 * Check to see if the process for this owner exists. Return 1 if it doesn't
1132 nfscl_procdoesntexist(u_int8_t *own)
1142 tl.cval[0] = *own++;
1143 tl.cval[1] = *own++;
1144 tl.cval[2] = *own++;
1145 tl.cval[3] = *own++;
1147 p = pfind_locked(pid);
1150 if (p->p_stats == NULL) {
1154 tl.cval[0] = *own++;
1155 tl.cval[1] = *own++;
1156 tl.cval[2] = *own++;
1157 tl.cval[3] = *own++;
1158 if (tl.lval != p->p_stats->p_start.tv_sec) {
1161 tl.cval[0] = *own++;
1162 tl.cval[1] = *own++;
1163 tl.cval[2] = *own++;
1165 if (tl.lval != p->p_stats->p_start.tv_usec)
1173 * - nfs pseudo system call for the client
1179 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1182 struct nfscbd_args nfscbdarg;
1183 struct nfsd_nfscbd_args nfscbdarg2;
1186 if (uap->flag & NFSSVC_CBADDSOCK) {
1187 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1190 if ((error = fget(td, nfscbdarg.sock, &fp)) != 0) {
1193 if (fp->f_type != DTYPE_SOCKET) {
1197 error = nfscbd_addsock(fp);
1199 if (!error && nfscl_enablecallb == 0) {
1200 nfsv4_cbport = nfscbdarg.port;
1201 nfscl_enablecallb = 1;
1203 } else if (uap->flag & NFSSVC_NFSCBD) {
1204 if (uap->argp == NULL)
1206 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1207 sizeof(nfscbdarg2));
1210 error = nfscbd_nfsd(td, &nfscbdarg2);
1217 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1220 * Called once to initialize data structures...
1223 nfscl_modevent(module_t mod, int type, void *data)
1226 static int loaded = 0;
1233 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1235 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1240 ncl_call_invalcaches = ncl_invalcaches;
1241 nfsd_call_nfscl = nfssvc_nfscl;
1246 if (nfs_numnfscbd != 0) {
1252 * XXX: Unloading of nfscl module is unsupported.
1255 ncl_call_invalcaches = NULL;
1256 nfsd_call_nfscl = NULL;
1257 /* and get rid of the mutexes */
1258 mtx_destroy(&nfs_clstate_mutex);
1259 mtx_destroy(&ncl_iod_mutex);
1271 static moduledata_t nfscl_mod = {
1276 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1278 /* So that loader and kldload(2) can find us, wherever we are.. */
1279 MODULE_VERSION(nfscl, 1);
1280 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1281 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1282 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1283 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);