]> git.saurik.com Git - apple/xnu.git/blob - bsd/nfs/nfs_node.c
xnu-7195.60.75.tar.gz
[apple/xnu.git] / bsd / nfs / nfs_node.c
1 /*
2 * Copyright (c) 2000-2019 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
29 /*
30 * Copyright (c) 1989, 1993
31 * The Regents of the University of California. All rights reserved.
32 *
33 * This code is derived from software contributed to Berkeley by
34 * Rick Macklem at The University of Guelph.
35 *
36 * Redistribution and use in source and binary forms, with or without
37 * modification, are permitted provided that the following conditions
38 * are met:
39 * 1. Redistributions of source code must retain the above copyright
40 * notice, this list of conditions and the following disclaimer.
41 * 2. Redistributions in binary form must reproduce the above copyright
42 * notice, this list of conditions and the following disclaimer in the
43 * documentation and/or other materials provided with the distribution.
44 * 3. All advertising materials mentioning features or use of this software
45 * must display the following acknowledgement:
46 * This product includes software developed by the University of
47 * California, Berkeley and its contributors.
48 * 4. Neither the name of the University nor the names of its contributors
49 * may be used to endorse or promote products derived from this software
50 * without specific prior written permission.
51 *
52 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
53 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
54 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
55 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
56 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
57 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
58 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
59 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
60 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
61 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * SUCH DAMAGE.
63 *
64 * @(#)nfs_node.c 8.6 (Berkeley) 5/22/95
65 * FreeBSD-Id: nfs_node.c,v 1.22 1997/10/28 14:06:20 bde Exp $
66 */
67
68 #include <nfs/nfs_conf.h>
69 #if CONFIG_NFS_CLIENT
70
71 #include <sys/param.h>
72 #include <sys/kernel.h>
73 #include <sys/systm.h>
74 #include <sys/proc.h>
75 #include <sys/kauth.h>
76 #include <sys/mount_internal.h>
77 #include <sys/vnode_internal.h>
78 #include <sys/vnode.h>
79 #include <sys/ubc.h>
80 #include <sys/malloc.h>
81 #include <sys/fcntl.h>
82 #include <sys/time.h>
83
84 #include <nfs/rpcv2.h>
85 #include <nfs/nfsproto.h>
86 #include <nfs/nfs.h>
87 #include <nfs/nfsnode.h>
88 #include <nfs/nfs_gss.h>
89 #include <nfs/nfsmount.h>
90
91 #define NFSNOHASH(fhsum) \
92 (&nfsnodehashtbl[(fhsum) & nfsnodehash])
93 static LIST_HEAD(nfsnodehashhead, nfsnode) * nfsnodehashtbl;
94 static u_long nfsnodehash;
95
96 static lck_grp_t *nfs_node_hash_lck_grp;
97 static lck_grp_t *nfs_node_lck_grp;
98 static lck_grp_t *nfs_data_lck_grp;
99 lck_mtx_t *nfs_node_hash_mutex;
100
101 ZONE_DECLARE(nfsnode_zone, "NFS node",
102 sizeof(struct nfsnode), ZC_ZFREE_CLEARMEM);
103
104 #define NFS_NODE_DBG(...) NFS_DBG(NFS_FAC_NODE, 7, ## __VA_ARGS__)
105
106 /*
107 * Initialize hash links for nfsnodes
108 * and build nfsnode free list.
109 */
110 void
111 nfs_nhinit(void)
112 {
113 nfs_node_hash_lck_grp = lck_grp_alloc_init("nfs_node_hash", LCK_GRP_ATTR_NULL);
114 nfs_node_hash_mutex = lck_mtx_alloc_init(nfs_node_hash_lck_grp, LCK_ATTR_NULL);
115 nfs_node_lck_grp = lck_grp_alloc_init("nfs_node", LCK_GRP_ATTR_NULL);
116 nfs_data_lck_grp = lck_grp_alloc_init("nfs_data", LCK_GRP_ATTR_NULL);
117 }
118
119 void
120 nfs_nhinit_finish(void)
121 {
122 lck_mtx_lock(nfs_node_hash_mutex);
123 if (!nfsnodehashtbl) {
124 nfsnodehashtbl = hashinit(desiredvnodes, M_NFSNODE, &nfsnodehash);
125 }
126 lck_mtx_unlock(nfs_node_hash_mutex);
127 }
128
129 /*
130 * Compute an entry in the NFS hash table structure
131 */
132 u_long
133 nfs_hash(u_char *fhp, int fhsize)
134 {
135 u_long fhsum;
136 int i;
137
138 fhsum = 0;
139 for (i = 0; i < fhsize; i++) {
140 fhsum += *fhp++;
141 }
142 return fhsum;
143 }
144
145
146 int nfs_case_insensitive(mount_t);
147
148 int
149 nfs_case_insensitive(mount_t mp)
150 {
151 struct nfsmount *nmp = VFSTONFS(mp);
152 int answer = 0;
153 int skip = 0;
154
155 if (nfs_mount_gone(nmp)) {
156 return 0;
157 }
158
159 if (nmp->nm_vers == NFS_VER2) {
160 /* V2 has no way to know */
161 return 0;
162 }
163
164 lck_mtx_lock(&nmp->nm_lock);
165 if (nmp->nm_vers == NFS_VER3) {
166 if (!(nmp->nm_state & NFSSTA_GOTPATHCONF)) {
167 /* We're holding the node lock so we just return
168 * with answer as case sensitive. Is very rare
169 * for file systems not to be homogenous w.r.t. pathconf
170 */
171 skip = 1;
172 }
173 } else if (!(nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_HOMOGENEOUS)) {
174 /* no pathconf info cached */
175 skip = 1;
176 }
177
178 if (!skip && (nmp->nm_fsattr.nfsa_flags & NFS_FSFLAG_CASE_INSENSITIVE)) {
179 answer = 1;
180 }
181
182 lck_mtx_unlock(&nmp->nm_lock);
183
184 return answer;
185 }
186
187
188 /*
189 * Look up a vnode/nfsnode by file handle.
190 * Callers must check for mount points!!
191 * In all cases, a pointer to a
192 * nfsnode structure is returned.
193 */
194 int
195 nfs_nget(
196 mount_t mp,
197 nfsnode_t dnp,
198 struct componentname *cnp,
199 u_char *fhp,
200 uint32_t fhsize,
201 struct nfs_vattr *nvap,
202 u_int64_t *xidp,
203 uint32_t auth,
204 int flags,
205 nfsnode_t *npp)
206 {
207 nfsnode_t np;
208 struct nfsnodehashhead *nhpp;
209 vnode_t vp;
210 int error, nfsvers;
211 mount_t mp2;
212 struct vnode_fsparam vfsp;
213 uint32_t vid, cn_namelen;
214 u_long nfshash;
215
216 FSDBG_TOP(263, mp, dnp, flags, npp);
217
218 /* Check for unmount in progress */
219 if (!mp || vfs_isforce(mp)) {
220 *npp = NULL;
221 error = ENXIO;
222 FSDBG_BOT(263, mp, dnp, 0xd1e, error);
223 return error;
224 }
225 nfsvers = VFSTONFS(mp)->nm_vers;
226 cn_namelen = cnp ? cnp->cn_namelen : 0;
227 nfshash = nfs_hash(fhp, fhsize);
228 loop:
229 lck_mtx_lock(nfs_node_hash_mutex);
230 nhpp = NFSNOHASH(nfshash);
231 for (np = nhpp->lh_first; np != 0; np = np->n_hash.le_next) {
232 mp2 = (np->n_hflag & NHINIT) ? np->n_mount : NFSTOMP(np);
233 if (mp != mp2 || np->n_fhsize != fhsize ||
234 bcmp(fhp, np->n_fhp, fhsize)) {
235 continue;
236 }
237 if (nvap && (nvap->nva_flags & NFS_FFLAG_TRIGGER_REFERRAL) &&
238 cnp && (cn_namelen > (fhsize - sizeof(dnp)))) {
239 /* The name was too long to fit in the file handle. Check it against the node's name. */
240 int namecmp = 0;
241 const char *vname = vnode_getname(NFSTOV(np));
242 if (vname) {
243 if (cn_namelen != strlen(vname)) {
244 namecmp = 1;
245 } else {
246 namecmp = strncmp(vname, cnp->cn_nameptr, cn_namelen);
247 }
248 vnode_putname(vname);
249 }
250 if (namecmp) { /* full name didn't match */
251 continue;
252 }
253 }
254 FSDBG(263, dnp, np, np->n_flag, 0xcace0000);
255 /* if the node is being initialized or locked, sleep on it */
256 if ((np->n_hflag & NHINIT) || ((np->n_hflag & NHLOCKED) && !(flags & NG_NOCREATE))) {
257 np->n_hflag |= NHLOCKWANT;
258 FSDBG(263, dnp, np, np->n_flag, 0xcace2222);
259 msleep(np, nfs_node_hash_mutex, PDROP | PINOD, "nfs_nget", NULL);
260 FSDBG(263, dnp, np, np->n_flag, 0xcace3333);
261 goto loop;
262 }
263 vp = NFSTOV(np);
264 vid = vnode_vid(vp);
265 lck_mtx_unlock(nfs_node_hash_mutex);
266 if ((error = vnode_getwithvid(vp, vid))) {
267 /*
268 * If vnode is being reclaimed or has already
269 * changed identity, no need to wait.
270 */
271 FSDBG_BOT(263, dnp, *npp, 0xcace0d1e, error);
272 return error;
273 }
274 if ((error = nfs_node_lock(np))) {
275 /* this only fails if the node is now unhashed */
276 /* so let's see if we can find/create it again */
277 FSDBG(263, dnp, *npp, 0xcaced1e2, error);
278 vnode_put(vp);
279 if (flags & NG_NOCREATE) {
280 *npp = 0;
281 FSDBG_BOT(263, dnp, *npp, 0xcaced1e0, ENOENT);
282 return ENOENT;
283 }
284 goto loop;
285 }
286 /* update attributes */
287 if (nvap) {
288 error = nfs_loadattrcache(np, nvap, xidp, 0);
289 }
290 if (error) {
291 nfs_node_unlock(np);
292 vnode_put(vp);
293 } else {
294 if (dnp && cnp && (flags & NG_MAKEENTRY)) {
295 cache_enter(NFSTOV(dnp), vp, cnp);
296 }
297 /*
298 * Update the vnode if the name/and or the parent has
299 * changed. We need to do this so that if getattrlist is
300 * called asking for ATTR_CMN_NAME, that the "most"
301 * correct name is being returned. In addition for
302 * monitored vnodes we need to kick the vnode out of the
303 * name cache. We do this so that if there are hard
304 * links in the same directory the link will not be
305 * found and a lookup will get us here to return the
306 * name of the current link. In addition by removing the
307 * name from the name cache the old name will not be
308 * found after a rename done on another client or the
309 * server. The principle reason to do this is because
310 * Finder is asking for notifications on a directory.
311 * The directory changes, Finder gets notified, reads
312 * the directory (which we have purged) and for each
313 * entry returned calls getattrlist with the name
314 * returned from readdir. gettattrlist has to call
315 * namei/lookup to resolve the name, because its not in
316 * the cache we end up here. We need to update the name
317 * so Finder will get the name it called us with.
318 *
319 * We had an imperfect solution with respect to case
320 * sensitivity. There is a test that is run in
321 * FileBuster that does renames from some name to
322 * another name differing only in case. It then reads
323 * the directory looking for the new name, after it
324 * finds that new name, it ask gettattrlist to verify
325 * that the name is the new name. Usually that works,
326 * but renames generate fsevents and fseventsd will do a
327 * lookup on the name via lstat. Since that test renames
328 * old name to new name back and forth there is a race
329 * that an fsevent will be behind and will access the
330 * file by the old name, on a case insensitive file
331 * system that will work. Problem is if we do a case
332 * sensitive compare, we're going to change the name,
333 * which the test's getattrlist verification step is
334 * going to fail. So we will check the case sensitivity
335 * of the file system and do the appropriate compare. In
336 * a rare instance for non homogeneous file systems
337 * w.r.t. pathconf we will use case sensitive compares.
338 * That could break if the file system is actually case
339 * insensitive.
340 *
341 * Note that V2 does not know the case, so we just
342 * assume case sensitivity.
343 *
344 * This is clearly not perfect due to races, but this is
345 * as good as its going to get. You can defeat the
346 * handling of hard links simply by doing:
347 *
348 * while :; do ls -l > /dev/null; done
349 *
350 * in a terminal window. Even a single ls -l can cause a
351 * race.
352 *
353 * <rant>What we really need is for the caller, that
354 * knows the name being used is valid since it got it
355 * from a readdir to use that name and not ask for the
356 * ATTR_CMN_NAME</rant>
357 */
358 if (dnp && cnp && (vp != NFSTOV(dnp))) {
359 int update_flags = (vnode_ismonitored((NFSTOV(dnp)))) ? VNODE_UPDATE_CACHE : 0;
360 int (*cmp)(const char *s1, const char *s2, size_t n);
361
362 cmp = nfs_case_insensitive(mp) ? strncasecmp : strncmp;
363
364 if (vp->v_name && cn_namelen != strnlen(vp->v_name, MAXPATHLEN)) {
365 update_flags |= VNODE_UPDATE_NAME;
366 }
367 if (vp->v_name && cn_namelen && (*cmp)(cnp->cn_nameptr, vp->v_name, cn_namelen)) {
368 update_flags |= VNODE_UPDATE_NAME;
369 }
370 if ((vp->v_name == NULL && cn_namelen != 0) || (vp->v_name != NULL && cn_namelen == 0)) {
371 update_flags |= VNODE_UPDATE_NAME;
372 }
373 if (vnode_parent(vp) != NFSTOV(dnp)) {
374 update_flags |= VNODE_UPDATE_PARENT;
375 }
376 if (update_flags) {
377 NFS_NODE_DBG("vnode_update_identity old name %s new name %.*s update flags = %x\n",
378 vp->v_name, cn_namelen, cnp->cn_nameptr ? cnp->cn_nameptr : "", update_flags);
379 vnode_update_identity(vp, NFSTOV(dnp), cnp->cn_nameptr, cn_namelen, 0, update_flags);
380 }
381 }
382
383 *npp = np;
384 }
385 FSDBG_BOT(263, dnp, *npp, 0xcace0000, error);
386 return error;
387 }
388
389 FSDBG(263, mp, dnp, npp, 0xaaaaaaaa);
390
391 if (flags & NG_NOCREATE) {
392 lck_mtx_unlock(nfs_node_hash_mutex);
393 *npp = 0;
394 FSDBG_BOT(263, dnp, *npp, 0x80000001, ENOENT);
395 return ENOENT;
396 }
397
398 /*
399 * allocate and initialize nfsnode and stick it in the hash
400 * before calling getnewvnode(). Anyone finding it in the
401 * hash before initialization is complete will wait for it.
402 */
403 np = zalloc_flags(nfsnode_zone, Z_WAITOK | Z_ZERO);
404 np->n_hflag |= (NHINIT | NHLOCKED);
405 np->n_mount = mp;
406 np->n_auth = auth;
407 TAILQ_INIT(&np->n_opens);
408 TAILQ_INIT(&np->n_lock_owners);
409 TAILQ_INIT(&np->n_locks);
410 np->n_dlink.tqe_next = NFSNOLIST;
411 np->n_dreturn.tqe_next = NFSNOLIST;
412 np->n_monlink.le_next = NFSNOLIST;
413
414 /* ugh... need to keep track of ".zfs" directories to workaround server bugs */
415 if ((nvap->nva_type == VDIR) && cnp && (cn_namelen == 4) &&
416 (cnp->cn_nameptr[0] == '.') && (cnp->cn_nameptr[1] == 'z') &&
417 (cnp->cn_nameptr[2] == 'f') && (cnp->cn_nameptr[3] == 's')) {
418 np->n_flag |= NISDOTZFS;
419 }
420 if (dnp && (dnp->n_flag & NISDOTZFS)) {
421 np->n_flag |= NISDOTZFSCHILD;
422 }
423
424 if (dnp && cnp && ((cn_namelen != 2) ||
425 (cnp->cn_nameptr[0] != '.') || (cnp->cn_nameptr[1] != '.'))) {
426 vnode_t dvp = NFSTOV(dnp);
427 if (!vnode_get(dvp)) {
428 if (!vnode_ref(dvp)) {
429 np->n_parent = dvp;
430 }
431 vnode_put(dvp);
432 }
433 }
434
435 /* setup node's file handle */
436 if (fhsize > NFS_SMALLFH) {
437 MALLOC(np->n_fhp, u_char *, fhsize, M_NFSBIGFH, M_WAITOK);
438 if (!np->n_fhp) {
439 lck_mtx_unlock(nfs_node_hash_mutex);
440 NFS_ZFREE(nfsnode_zone, np);
441 *npp = 0;
442 FSDBG_BOT(263, dnp, *npp, 0x80000002, ENOMEM);
443 return ENOMEM;
444 }
445 } else {
446 np->n_fhp = &np->n_fh[0];
447 }
448 bcopy(fhp, np->n_fhp, fhsize);
449 np->n_fhsize = fhsize;
450
451 /* Insert the nfsnode in the hash queue for its new file handle */
452 LIST_INSERT_HEAD(nhpp, np, n_hash);
453 np->n_hflag |= NHHASHED;
454 FSDBG(266, 0, np, np->n_flag, np->n_hflag);
455
456 /* lock the new nfsnode */
457 lck_mtx_init(&np->n_lock, nfs_node_lck_grp, LCK_ATTR_NULL);
458 lck_rw_init(&np->n_datalock, nfs_data_lck_grp, LCK_ATTR_NULL);
459 lck_mtx_init(&np->n_openlock, nfs_open_grp, LCK_ATTR_NULL);
460 lck_mtx_lock(&np->n_lock);
461
462 /* release lock on hash table */
463 lck_mtx_unlock(nfs_node_hash_mutex);
464
465 /* do initial loading of attributes */
466 NACLINVALIDATE(np);
467 NACCESSINVALIDATE(np);
468 error = nfs_loadattrcache(np, nvap, xidp, 1);
469 if (error) {
470 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
471 nfs_node_unlock(np);
472 lck_mtx_lock(nfs_node_hash_mutex);
473 LIST_REMOVE(np, n_hash);
474 np->n_hflag &= ~(NHHASHED | NHINIT | NHLOCKED);
475 if (np->n_hflag & NHLOCKWANT) {
476 np->n_hflag &= ~NHLOCKWANT;
477 wakeup(np);
478 }
479 lck_mtx_unlock(nfs_node_hash_mutex);
480 if (np->n_parent) {
481 if (!vnode_get(np->n_parent)) {
482 vnode_rele(np->n_parent);
483 vnode_put(np->n_parent);
484 }
485 np->n_parent = NULL;
486 }
487 lck_mtx_destroy(&np->n_lock, nfs_node_lck_grp);
488 lck_rw_destroy(&np->n_datalock, nfs_data_lck_grp);
489 lck_mtx_destroy(&np->n_openlock, nfs_open_grp);
490 if (np->n_fhsize > NFS_SMALLFH) {
491 FREE(np->n_fhp, M_NFSBIGFH);
492 }
493 NFS_ZFREE(nfsnode_zone, np);
494 *npp = 0;
495 FSDBG_BOT(263, dnp, *npp, 0x80000003, error);
496 return error;
497 }
498 NFS_CHANGED_UPDATE(nfsvers, np, nvap);
499 if (nvap->nva_type == VDIR) {
500 NFS_CHANGED_UPDATE_NC(nfsvers, np, nvap);
501 }
502
503 /* now, attempt to get a new vnode */
504 vfsp.vnfs_mp = mp;
505 vfsp.vnfs_vtype = nvap->nva_type;
506 vfsp.vnfs_str = "nfs";
507 vfsp.vnfs_dvp = dnp ? NFSTOV(dnp) : NULL;
508 vfsp.vnfs_fsnode = np;
509 #if CONFIG_NFS4
510 if (nfsvers == NFS_VER4) {
511 #if FIFO
512 if (nvap->nva_type == VFIFO) {
513 vfsp.vnfs_vops = fifo_nfsv4nodeop_p;
514 } else
515 #endif /* FIFO */
516 if (nvap->nva_type == VBLK || nvap->nva_type == VCHR) {
517 vfsp.vnfs_vops = spec_nfsv4nodeop_p;
518 } else {
519 vfsp.vnfs_vops = nfsv4_vnodeop_p;
520 }
521 } else
522 #endif /* CONFIG_NFS4 */
523 {
524 #if FIFO
525 if (nvap->nva_type == VFIFO) {
526 vfsp.vnfs_vops = fifo_nfsv2nodeop_p;
527 } else
528 #endif /* FIFO */
529 if (nvap->nva_type == VBLK || nvap->nva_type == VCHR) {
530 vfsp.vnfs_vops = spec_nfsv2nodeop_p;
531 } else {
532 vfsp.vnfs_vops = nfsv2_vnodeop_p;
533 }
534 }
535 vfsp.vnfs_markroot = (flags & NG_MARKROOT) ? 1 : 0;
536 vfsp.vnfs_marksystem = 0;
537 vfsp.vnfs_rdev = 0;
538 vfsp.vnfs_filesize = nvap->nva_size;
539 vfsp.vnfs_cnp = cnp;
540 vfsp.vnfs_flags = VNFS_ADDFSREF;
541 if (!dnp || !cnp || !(flags & NG_MAKEENTRY)) {
542 vfsp.vnfs_flags |= VNFS_NOCACHE;
543 }
544
545 #if CONFIG_TRIGGERS
546 if (((nfsvers >= NFS_VER4)
547 )
548 && (nvap->nva_type == VDIR) && (np->n_vattr.nva_flags & NFS_FFLAG_TRIGGER)
549 && !(flags & NG_MARKROOT)) {
550 struct vnode_trigger_param vtp;
551 bzero(&vtp, sizeof(vtp));
552 bcopy(&vfsp, &vtp.vnt_params, sizeof(vfsp));
553 vtp.vnt_resolve_func = nfs_mirror_mount_trigger_resolve;
554 vtp.vnt_unresolve_func = nfs_mirror_mount_trigger_unresolve;
555 vtp.vnt_rearm_func = nfs_mirror_mount_trigger_rearm;
556 vtp.vnt_flags = VNT_AUTO_REARM | VNT_KERN_RESOLVE;
557 error = vnode_create(VNCREATE_TRIGGER, VNCREATE_TRIGGER_SIZE, &vtp, &np->n_vnode);
558 } else
559 #endif
560 {
561 error = vnode_create(VNCREATE_FLAVOR, VCREATESIZE, &vfsp, &np->n_vnode);
562 }
563 if (error) {
564 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
565 nfs_node_unlock(np);
566 lck_mtx_lock(nfs_node_hash_mutex);
567 LIST_REMOVE(np, n_hash);
568 np->n_hflag &= ~(NHHASHED | NHINIT | NHLOCKED);
569 if (np->n_hflag & NHLOCKWANT) {
570 np->n_hflag &= ~NHLOCKWANT;
571 wakeup(np);
572 }
573 lck_mtx_unlock(nfs_node_hash_mutex);
574 if (np->n_parent) {
575 if (!vnode_get(np->n_parent)) {
576 vnode_rele(np->n_parent);
577 vnode_put(np->n_parent);
578 }
579 np->n_parent = NULL;
580 }
581 lck_mtx_destroy(&np->n_lock, nfs_node_lck_grp);
582 lck_rw_destroy(&np->n_datalock, nfs_data_lck_grp);
583 lck_mtx_destroy(&np->n_openlock, nfs_open_grp);
584 if (np->n_fhsize > NFS_SMALLFH) {
585 FREE(np->n_fhp, M_NFSBIGFH);
586 }
587 NFS_ZFREE(nfsnode_zone, np);
588 *npp = 0;
589 FSDBG_BOT(263, dnp, *npp, 0x80000004, error);
590 return error;
591 }
592 vp = np->n_vnode;
593 vnode_settag(vp, VT_NFS);
594 /* node is now initialized */
595
596 /* check if anyone's waiting on this node */
597 lck_mtx_lock(nfs_node_hash_mutex);
598 np->n_hflag &= ~(NHINIT | NHLOCKED);
599 if (np->n_hflag & NHLOCKWANT) {
600 np->n_hflag &= ~NHLOCKWANT;
601 wakeup(np);
602 }
603 lck_mtx_unlock(nfs_node_hash_mutex);
604
605 *npp = np;
606
607 FSDBG_BOT(263, dnp, vp, *npp, error);
608 return error;
609 }
610
611
612 int
613 nfs_vnop_inactive(
614 struct vnop_inactive_args /* {
615 * struct vnodeop_desc *a_desc;
616 * vnode_t a_vp;
617 * vfs_context_t a_context;
618 * } */*ap)
619 {
620 vnode_t vp = ap->a_vp;
621 vfs_context_t ctx = ap->a_context;
622 nfsnode_t np;
623 struct nfs_sillyrename *nsp;
624 struct nfs_vattr *nvattr;
625 int unhash, attrerr, busyerror, error, inuse, busied, force;
626 struct nfs_open_file *nofp;
627 struct componentname cn;
628 struct nfsmount *nmp;
629 mount_t mp;
630
631 if (vp == NULL) {
632 panic("nfs_vnop_inactive: vp == NULL");
633 }
634 np = VTONFS(vp);
635 if (np == NULL) {
636 panic("nfs_vnop_inactive: np == NULL");
637 }
638
639 nmp = NFSTONMP(np);
640 mp = vnode_mount(vp);
641 MALLOC(nvattr, struct nfs_vattr *, sizeof(*nvattr), M_TEMP, M_WAITOK);
642
643 restart:
644 force = (!mp || vfs_isforce(mp));
645 error = 0;
646 inuse = (nfs_mount_state_in_use_start(nmp, NULL) == 0);
647
648 /* There shouldn't be any open or lock state at this point */
649 lck_mtx_lock(&np->n_openlock);
650 if (np->n_openrefcnt && !force) {
651 /*
652 * vnode_rele and vnode_put drop the vnode lock before
653 * calling VNOP_INACTIVE, so there is a race were the
654 * vnode could become active again. Perhaps there are
655 * other places where this can happen, so if we've got
656 * here we need to get out.
657 */
658 #ifdef NFS_NODE_DEBUG
659 NP(np, "nfs_vnop_inactive: still open: %d", np->n_openrefcnt);
660 #endif
661 lck_mtx_unlock(&np->n_openlock);
662 if (inuse) {
663 nfs_mount_state_in_use_end(nmp, 0);
664 }
665 goto out_free;
666 }
667
668 TAILQ_FOREACH(nofp, &np->n_opens, nof_link) {
669 lck_mtx_lock(&nofp->nof_lock);
670 if (nofp->nof_flags & NFS_OPEN_FILE_BUSY) {
671 if (!force) {
672 NP(np, "nfs_vnop_inactive: open file busy");
673 }
674 busied = 0;
675 } else {
676 nofp->nof_flags |= NFS_OPEN_FILE_BUSY;
677 busied = 1;
678 }
679 lck_mtx_unlock(&nofp->nof_lock);
680 if ((np->n_flag & NREVOKE) || (nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
681 if (busied) {
682 nfs_open_file_clear_busy(nofp);
683 }
684 continue;
685 }
686 /*
687 * If we just created the file, we already had it open in
688 * anticipation of getting a subsequent open call. If the
689 * node has gone inactive without being open, we need to
690 * clean up (close) the open done in the create.
691 */
692 #if CONFIG_NFS4
693 if ((nofp->nof_flags & NFS_OPEN_FILE_CREATE) && nofp->nof_creator && !force) {
694 if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN) {
695 lck_mtx_unlock(&np->n_openlock);
696 if (busied) {
697 nfs_open_file_clear_busy(nofp);
698 }
699 if (!nfs4_reopen(nofp, NULL)) {
700 if (inuse) {
701 nfs_mount_state_in_use_end(nmp, 0);
702 }
703 goto restart;
704 }
705 }
706 nofp->nof_flags &= ~NFS_OPEN_FILE_CREATE;
707 lck_mtx_unlock(&np->n_openlock);
708 error = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_BOTH, NFS_OPEN_SHARE_DENY_NONE, ctx);
709 if (error) {
710 NP(np, "nfs_vnop_inactive: create close error: %d", error);
711 nofp->nof_flags |= NFS_OPEN_FILE_CREATE;
712 }
713 if (busied) {
714 nfs_open_file_clear_busy(nofp);
715 }
716 if (inuse) {
717 nfs_mount_state_in_use_end(nmp, error);
718 }
719 goto restart;
720 }
721 #endif
722 if (nofp->nof_flags & NFS_OPEN_FILE_NEEDCLOSE) {
723 /*
724 * If the file is marked as needing reopen, but this was the only
725 * open on the file, just drop the open.
726 */
727 nofp->nof_flags &= ~NFS_OPEN_FILE_NEEDCLOSE;
728 if ((nofp->nof_flags & NFS_OPEN_FILE_REOPEN) && (nofp->nof_opencnt == 1)) {
729 nofp->nof_flags &= ~NFS_OPEN_FILE_REOPEN;
730 nofp->nof_r--;
731 nofp->nof_opencnt--;
732 nofp->nof_access = 0;
733 } else if (!force) {
734 lck_mtx_unlock(&np->n_openlock);
735 if (nofp->nof_flags & NFS_OPEN_FILE_REOPEN) {
736 int should_restart = 0;
737 if (busied) {
738 nfs_open_file_clear_busy(nofp);
739 }
740 #if CONFIG_NFS4
741 if (!nfs4_reopen(nofp, NULL)) {
742 should_restart = 1;
743 }
744 #endif
745 if (should_restart) {
746 if (inuse) {
747 nfs_mount_state_in_use_end(nmp, 0);
748 }
749 goto restart;
750 }
751 }
752 error = nfs_close(np, nofp, NFS_OPEN_SHARE_ACCESS_READ, NFS_OPEN_SHARE_DENY_NONE, ctx);
753 if (error) {
754 NP(np, "nfs_vnop_inactive: need close error: %d", error);
755 nofp->nof_flags |= NFS_OPEN_FILE_NEEDCLOSE;
756 }
757 if (busied) {
758 nfs_open_file_clear_busy(nofp);
759 }
760 if (inuse) {
761 nfs_mount_state_in_use_end(nmp, error);
762 }
763 goto restart;
764 }
765 }
766 if (nofp->nof_opencnt && !force) {
767 NP(np, "nfs_vnop_inactive: file still open: %d", nofp->nof_opencnt);
768 }
769 if (!force && (nofp->nof_access || nofp->nof_deny ||
770 nofp->nof_mmap_access || nofp->nof_mmap_deny ||
771 nofp->nof_r || nofp->nof_w || nofp->nof_rw ||
772 nofp->nof_r_dw || nofp->nof_w_dw || nofp->nof_rw_dw ||
773 nofp->nof_r_drw || nofp->nof_w_drw || nofp->nof_rw_drw ||
774 nofp->nof_d_r || nofp->nof_d_w || nofp->nof_d_rw ||
775 nofp->nof_d_r_dw || nofp->nof_d_w_dw || nofp->nof_d_rw_dw ||
776 nofp->nof_d_r_drw || nofp->nof_d_w_drw || nofp->nof_d_rw_drw)) {
777 NP(np, "nfs_vnop_inactive: non-zero access: %d %d %d %d # %u.%u %u.%u %u.%u dw %u.%u %u.%u %u.%u drw %u.%u %u.%u %u.%u",
778 nofp->nof_access, nofp->nof_deny,
779 nofp->nof_mmap_access, nofp->nof_mmap_deny,
780 nofp->nof_r, nofp->nof_d_r,
781 nofp->nof_w, nofp->nof_d_w,
782 nofp->nof_rw, nofp->nof_d_rw,
783 nofp->nof_r_dw, nofp->nof_d_r_dw,
784 nofp->nof_w_dw, nofp->nof_d_w_dw,
785 nofp->nof_rw_dw, nofp->nof_d_rw_dw,
786 nofp->nof_r_drw, nofp->nof_d_r_drw,
787 nofp->nof_w_drw, nofp->nof_d_w_drw,
788 nofp->nof_rw_drw, nofp->nof_d_rw_drw);
789 }
790 if (busied) {
791 nfs_open_file_clear_busy(nofp);
792 }
793 }
794 lck_mtx_unlock(&np->n_openlock);
795
796 if (inuse && nfs_mount_state_in_use_end(nmp, error)) {
797 goto restart;
798 }
799
800 nfs_node_lock_force(np);
801
802 if (vnode_vtype(vp) != VDIR) {
803 nsp = np->n_sillyrename;
804 np->n_sillyrename = NULL;
805 } else {
806 nsp = NULL;
807 }
808
809 FSDBG_TOP(264, vp, np, np->n_flag, nsp);
810
811 if (!nsp) {
812 /* no silly file to clean up... */
813 /* clear all flags other than these */
814 np->n_flag &= (NMODIFIED);
815 nfs_node_unlock(np);
816 FSDBG_BOT(264, vp, np, np->n_flag, 0);
817 goto out_free;
818 }
819 nfs_node_unlock(np);
820
821 /* Remove the silly file that was rename'd earlier */
822
823 /* flush all the buffers */
824 nfs_vinvalbuf2(vp, V_SAVE, vfs_context_thread(ctx), nsp->nsr_cred, 1);
825
826 /* try to get the latest attributes */
827 attrerr = nfs_getattr(np, nvattr, ctx, NGA_UNCACHED);
828
829 /* Check if we should remove it from the node hash. */
830 /* Leave it if inuse or it has multiple hard links. */
831 if (vnode_isinuse(vp, 0) || (!attrerr && (nvattr->nva_nlink > 1))) {
832 unhash = 0;
833 } else {
834 unhash = 1;
835 ubc_setsize(vp, 0);
836 }
837
838 /* mark this node and the directory busy while we do the remove */
839 busyerror = nfs_node_set_busy2(nsp->nsr_dnp, np, vfs_context_thread(ctx));
840
841 /* lock the node while we remove the silly file */
842 lck_mtx_lock(nfs_node_hash_mutex);
843 while (np->n_hflag & NHLOCKED) {
844 np->n_hflag |= NHLOCKWANT;
845 msleep(np, nfs_node_hash_mutex, PINOD, "nfs_inactive", NULL);
846 }
847 np->n_hflag |= NHLOCKED;
848 lck_mtx_unlock(nfs_node_hash_mutex);
849
850 /* purge the name cache to deter others from finding it */
851 bzero(&cn, sizeof(cn));
852 cn.cn_nameptr = nsp->nsr_name;
853 cn.cn_namelen = nsp->nsr_namlen;
854 nfs_name_cache_purge(nsp->nsr_dnp, np, &cn, ctx);
855
856 FSDBG(264, np, np->n_size, np->n_vattr.nva_size, 0xf00d00f1);
857
858 /* now remove the silly file */
859 nfs_removeit(nsp);
860
861 /* clear all flags other than these */
862 nfs_node_lock_force(np);
863 np->n_flag &= (NMODIFIED);
864 nfs_node_unlock(np);
865
866 if (!busyerror) {
867 nfs_node_clear_busy2(nsp->nsr_dnp, np);
868 }
869
870 if (unhash && vnode_isinuse(vp, 0)) {
871 /* vnode now inuse after silly remove? */
872 unhash = 0;
873 ubc_setsize(vp, np->n_size);
874 }
875
876 lck_mtx_lock(nfs_node_hash_mutex);
877 if (unhash) {
878 /*
879 * remove nfsnode from hash now so we can't accidentally find it
880 * again if another object gets created with the same filehandle
881 * before this vnode gets reclaimed
882 */
883 if (np->n_hflag & NHHASHED) {
884 LIST_REMOVE(np, n_hash);
885 np->n_hflag &= ~NHHASHED;
886 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
887 }
888 vnode_recycle(vp);
889 }
890 /* unlock the node */
891 np->n_hflag &= ~NHLOCKED;
892 if (np->n_hflag & NHLOCKWANT) {
893 np->n_hflag &= ~NHLOCKWANT;
894 wakeup(np);
895 }
896 lck_mtx_unlock(nfs_node_hash_mutex);
897
898 /* cleanup sillyrename info */
899 if (nsp->nsr_cred != NOCRED) {
900 kauth_cred_unref(&nsp->nsr_cred);
901 }
902 vnode_rele(NFSTOV(nsp->nsr_dnp));
903 FREE(nsp, M_TEMP);
904 FSDBG_BOT(264, vp, np, np->n_flag, 0);
905 out_free:
906 FREE(nvattr, M_TEMP);
907 return 0;
908 }
909
910 /*
911 * Reclaim an nfsnode so that it can be used for other purposes.
912 */
913 int
914 nfs_vnop_reclaim(
915 struct vnop_reclaim_args /* {
916 * struct vnodeop_desc *a_desc;
917 * vnode_t a_vp;
918 * vfs_context_t a_context;
919 * } */*ap)
920 {
921 vnode_t vp = ap->a_vp;
922 nfsnode_t np = VTONFS(vp);
923 struct nfs_open_file *nofp, *nextnofp;
924 struct nfs_file_lock *nflp, *nextnflp;
925 struct nfs_lock_owner *nlop, *nextnlop;
926 struct nfsmount *nmp = np->n_mount ? VFSTONFS(np->n_mount) : NFSTONMP(np);
927 mount_t mp = vnode_mount(vp);
928 int force;
929
930 FSDBG_TOP(265, vp, np, np->n_flag, 0);
931 force = (!mp || vfs_isforce(mp) || nfs_mount_gone(nmp));
932
933
934 /* There shouldn't be any open or lock state at this point */
935 lck_mtx_lock(&np->n_openlock);
936
937 #if CONFIG_NFS4
938 if (nmp && (nmp->nm_vers >= NFS_VER4)) {
939 /* need to drop a delegation */
940 if (np->n_dreturn.tqe_next != NFSNOLIST) {
941 /* remove this node from the delegation return list */
942 lck_mtx_lock(&nmp->nm_lock);
943 if (np->n_dreturn.tqe_next != NFSNOLIST) {
944 TAILQ_REMOVE(&nmp->nm_dreturnq, np, n_dreturn);
945 np->n_dreturn.tqe_next = NFSNOLIST;
946 }
947 lck_mtx_unlock(&nmp->nm_lock);
948 }
949 if (np->n_dlink.tqe_next != NFSNOLIST) {
950 /* remove this node from the delegation list */
951 lck_mtx_lock(&nmp->nm_lock);
952 if (np->n_dlink.tqe_next != NFSNOLIST) {
953 TAILQ_REMOVE(&nmp->nm_delegations, np, n_dlink);
954 np->n_dlink.tqe_next = NFSNOLIST;
955 }
956 lck_mtx_unlock(&nmp->nm_lock);
957 }
958 if ((np->n_openflags & N_DELEG_MASK) && !force) {
959 /* try to return the delegation */
960 np->n_openflags &= ~N_DELEG_MASK;
961 }
962 if (np->n_attrdirfh) {
963 FREE(np->n_attrdirfh, M_TEMP);
964 np->n_attrdirfh = NULL;
965 }
966 }
967 #endif
968
969 /* clean up file locks */
970 TAILQ_FOREACH_SAFE(nflp, &np->n_locks, nfl_link, nextnflp) {
971 if (!(nflp->nfl_flags & NFS_FILE_LOCK_DEAD) && !force) {
972 NP(np, "nfs_vnop_reclaim: lock 0x%llx 0x%llx 0x%x (bc %d)",
973 nflp->nfl_start, nflp->nfl_end, nflp->nfl_flags, nflp->nfl_blockcnt);
974 }
975 if (!(nflp->nfl_flags & (NFS_FILE_LOCK_BLOCKED | NFS_FILE_LOCK_DEAD))) {
976 /* try sending an unlock RPC if it wasn't delegated */
977 if (!(nflp->nfl_flags & NFS_FILE_LOCK_DELEGATED) && !force) {
978 nmp->nm_funcs->nf_unlock_rpc(np, nflp->nfl_owner, F_WRLCK, nflp->nfl_start, nflp->nfl_end, R_RECOVER,
979 NULL, nflp->nfl_owner->nlo_open_owner->noo_cred);
980 }
981 lck_mtx_lock(&nflp->nfl_owner->nlo_lock);
982 TAILQ_REMOVE(&nflp->nfl_owner->nlo_locks, nflp, nfl_lolink);
983 lck_mtx_unlock(&nflp->nfl_owner->nlo_lock);
984 }
985 TAILQ_REMOVE(&np->n_locks, nflp, nfl_link);
986 nfs_file_lock_destroy(nflp);
987 }
988 /* clean up lock owners */
989 TAILQ_FOREACH_SAFE(nlop, &np->n_lock_owners, nlo_link, nextnlop) {
990 if (!TAILQ_EMPTY(&nlop->nlo_locks) && !force) {
991 NP(np, "nfs_vnop_reclaim: lock owner with locks");
992 }
993 TAILQ_REMOVE(&np->n_lock_owners, nlop, nlo_link);
994 nfs_lock_owner_destroy(nlop);
995 }
996 /* clean up open state */
997 if (np->n_openrefcnt && !force) {
998 NP(np, "nfs_vnop_reclaim: still open: %d", np->n_openrefcnt);
999 }
1000 TAILQ_FOREACH_SAFE(nofp, &np->n_opens, nof_link, nextnofp) {
1001 if (nofp->nof_flags & NFS_OPEN_FILE_BUSY) {
1002 NP(np, "nfs_vnop_reclaim: open file busy");
1003 }
1004 if (!(np->n_flag & NREVOKE) && !(nofp->nof_flags & NFS_OPEN_FILE_LOST)) {
1005 if (nofp->nof_opencnt && !force) {
1006 NP(np, "nfs_vnop_reclaim: file still open: %d", nofp->nof_opencnt);
1007 }
1008 if (!force && (nofp->nof_access || nofp->nof_deny ||
1009 nofp->nof_mmap_access || nofp->nof_mmap_deny ||
1010 nofp->nof_r || nofp->nof_w || nofp->nof_rw ||
1011 nofp->nof_r_dw || nofp->nof_w_dw || nofp->nof_rw_dw ||
1012 nofp->nof_r_drw || nofp->nof_w_drw || nofp->nof_rw_drw ||
1013 nofp->nof_d_r || nofp->nof_d_w || nofp->nof_d_rw ||
1014 nofp->nof_d_r_dw || nofp->nof_d_w_dw || nofp->nof_d_rw_dw ||
1015 nofp->nof_d_r_drw || nofp->nof_d_w_drw || nofp->nof_d_rw_drw)) {
1016 NP(np, "nfs_vnop_reclaim: non-zero access: %d %d %d %d # %u.%u %u.%u %u.%u dw %u.%u %u.%u %u.%u drw %u.%u %u.%u %u.%u",
1017 nofp->nof_access, nofp->nof_deny,
1018 nofp->nof_mmap_access, nofp->nof_mmap_deny,
1019 nofp->nof_r, nofp->nof_d_r,
1020 nofp->nof_w, nofp->nof_d_w,
1021 nofp->nof_rw, nofp->nof_d_rw,
1022 nofp->nof_r_dw, nofp->nof_d_r_dw,
1023 nofp->nof_w_dw, nofp->nof_d_w_dw,
1024 nofp->nof_rw_dw, nofp->nof_d_rw_dw,
1025 nofp->nof_r_drw, nofp->nof_d_r_drw,
1026 nofp->nof_w_drw, nofp->nof_d_w_drw,
1027 nofp->nof_rw_drw, nofp->nof_d_rw_drw);
1028 #if CONFIG_NFS4
1029 /* try sending a close RPC if it wasn't delegated */
1030 if (nofp->nof_r || nofp->nof_w || nofp->nof_rw ||
1031 nofp->nof_r_dw || nofp->nof_w_dw || nofp->nof_rw_dw ||
1032 nofp->nof_r_drw || nofp->nof_w_drw || nofp->nof_rw_drw) {
1033 nfs4_close_rpc(np, nofp, NULL, nofp->nof_owner->noo_cred, R_RECOVER);
1034 }
1035 #endif
1036 }
1037 }
1038 TAILQ_REMOVE(&np->n_opens, nofp, nof_link);
1039 nfs_open_file_destroy(nofp);
1040 }
1041 lck_mtx_unlock(&np->n_openlock);
1042
1043 if (np->n_monlink.le_next != NFSNOLIST) {
1044 /* Wait for any in-progress getattr to complete, */
1045 /* then remove this node from the monitored node list. */
1046 lck_mtx_lock(&nmp->nm_lock);
1047 while (np->n_mflag & NMMONSCANINPROG) {
1048 struct timespec ts = { .tv_sec = 1, .tv_nsec = 0 };
1049 np->n_mflag |= NMMONSCANWANT;
1050 msleep(&np->n_mflag, &nmp->nm_lock, PZERO - 1, "nfswaitmonscan", &ts);
1051 }
1052 if (np->n_monlink.le_next != NFSNOLIST) {
1053 LIST_REMOVE(np, n_monlink);
1054 np->n_monlink.le_next = NFSNOLIST;
1055 }
1056 lck_mtx_unlock(&nmp->nm_lock);
1057 }
1058
1059 lck_mtx_lock(nfs_buf_mutex);
1060 if (!force && (!LIST_EMPTY(&np->n_dirtyblkhd) || !LIST_EMPTY(&np->n_cleanblkhd))) {
1061 NP(np, "nfs_reclaim: dropping %s buffers", (!LIST_EMPTY(&np->n_dirtyblkhd) ? "dirty" : "clean"));
1062 }
1063 lck_mtx_unlock(nfs_buf_mutex);
1064 nfs_vinvalbuf(vp, V_IGNORE_WRITEERR, ap->a_context, 0);
1065
1066 lck_mtx_lock(nfs_node_hash_mutex);
1067
1068 if ((vnode_vtype(vp) != VDIR) && np->n_sillyrename) {
1069 if (!force) {
1070 NP(np, "nfs_reclaim: leaving unlinked file %s", np->n_sillyrename->nsr_name);
1071 }
1072 if (np->n_sillyrename->nsr_cred != NOCRED) {
1073 kauth_cred_unref(&np->n_sillyrename->nsr_cred);
1074 }
1075 vnode_rele(NFSTOV(np->n_sillyrename->nsr_dnp));
1076 FREE(np->n_sillyrename, M_TEMP);
1077 }
1078
1079 vnode_removefsref(vp);
1080
1081 if (np->n_hflag & NHHASHED) {
1082 LIST_REMOVE(np, n_hash);
1083 np->n_hflag &= ~NHHASHED;
1084 FSDBG(266, 0, np, np->n_flag, 0xb1eb1e);
1085 }
1086 lck_mtx_unlock(nfs_node_hash_mutex);
1087
1088 /*
1089 * Free up any directory cookie structures and large file handle
1090 * structures that might be associated with this nfs node.
1091 */
1092 nfs_node_lock_force(np);
1093 if ((vnode_vtype(vp) == VDIR) && np->n_cookiecache) {
1094 NFS_ZFREE(ZV_NFSDIROFF, np->n_cookiecache);
1095 }
1096 if (np->n_fhsize > NFS_SMALLFH) {
1097 FREE(np->n_fhp, M_NFSBIGFH);
1098 }
1099 if (np->n_vattr.nva_acl) {
1100 kauth_acl_free(np->n_vattr.nva_acl);
1101 }
1102 nfs_node_unlock(np);
1103 vnode_clearfsnode(vp);
1104
1105 if (np->n_parent) {
1106 if (!vnode_get(np->n_parent)) {
1107 vnode_rele(np->n_parent);
1108 vnode_put(np->n_parent);
1109 }
1110 np->n_parent = NULL;
1111 }
1112
1113 lck_mtx_destroy(&np->n_lock, nfs_node_lck_grp);
1114 lck_rw_destroy(&np->n_datalock, nfs_data_lck_grp);
1115 lck_mtx_destroy(&np->n_openlock, nfs_open_grp);
1116
1117 FSDBG_BOT(265, vp, np, np->n_flag, 0xd1ed1e);
1118 NFS_ZFREE(nfsnode_zone, np);
1119 return 0;
1120 }
1121
1122 /*
1123 * Acquire an NFS node lock
1124 */
1125
1126 int
1127 nfs_node_lock_internal(nfsnode_t np, int force)
1128 {
1129 FSDBG_TOP(268, np, force, 0, 0);
1130 lck_mtx_lock(&np->n_lock);
1131 if (!force && !(np->n_hflag && NHHASHED)) {
1132 FSDBG_BOT(268, np, 0xdead, 0, 0);
1133 lck_mtx_unlock(&np->n_lock);
1134 return ENOENT;
1135 }
1136 FSDBG_BOT(268, np, force, 0, 0);
1137 return 0;
1138 }
1139
1140 int
1141 nfs_node_lock(nfsnode_t np)
1142 {
1143 return nfs_node_lock_internal(np, 0);
1144 }
1145
1146 void
1147 nfs_node_lock_force(nfsnode_t np)
1148 {
1149 nfs_node_lock_internal(np, 1);
1150 }
1151
1152 /*
1153 * Release an NFS node lock
1154 */
1155 void
1156 nfs_node_unlock(nfsnode_t np)
1157 {
1158 FSDBG(269, np, current_thread(), 0, 0);
1159 lck_mtx_unlock(&np->n_lock);
1160 }
1161
1162 /*
1163 * Acquire 2 NFS node locks
1164 * - locks taken in reverse address order
1165 * - both or neither of the locks are taken
1166 * - only one lock taken per node (dup nodes are skipped)
1167 */
1168 int
1169 nfs_node_lock2(nfsnode_t np1, nfsnode_t np2)
1170 {
1171 nfsnode_t first, second;
1172 int error;
1173
1174 first = (np1 > np2) ? np1 : np2;
1175 second = (np1 > np2) ? np2 : np1;
1176 if ((error = nfs_node_lock(first))) {
1177 return error;
1178 }
1179 if (np1 == np2) {
1180 return error;
1181 }
1182 if ((error = nfs_node_lock(second))) {
1183 nfs_node_unlock(first);
1184 }
1185 return error;
1186 }
1187
1188 void
1189 nfs_node_unlock2(nfsnode_t np1, nfsnode_t np2)
1190 {
1191 nfs_node_unlock(np1);
1192 if (np1 != np2) {
1193 nfs_node_unlock(np2);
1194 }
1195 }
1196
1197 /*
1198 * Manage NFS node busy state.
1199 * (Similar to NFS node locks above)
1200 */
1201 int
1202 nfs_node_set_busy(nfsnode_t np, thread_t thd)
1203 {
1204 struct timespec ts = { .tv_sec = 2, .tv_nsec = 0 };
1205 int error;
1206
1207 if ((error = nfs_node_lock(np))) {
1208 return error;
1209 }
1210 while (ISSET(np->n_flag, NBUSY)) {
1211 SET(np->n_flag, NBUSYWANT);
1212 msleep(np, &np->n_lock, PZERO - 1, "nfsbusywant", &ts);
1213 if ((error = nfs_sigintr(NFSTONMP(np), NULL, thd, 0))) {
1214 break;
1215 }
1216 }
1217 if (!error) {
1218 SET(np->n_flag, NBUSY);
1219 }
1220 nfs_node_unlock(np);
1221 return error;
1222 }
1223
1224 void
1225 nfs_node_clear_busy(nfsnode_t np)
1226 {
1227 int wanted;
1228
1229 nfs_node_lock_force(np);
1230 wanted = ISSET(np->n_flag, NBUSYWANT);
1231 CLR(np->n_flag, NBUSY | NBUSYWANT);
1232 nfs_node_unlock(np);
1233 if (wanted) {
1234 wakeup(np);
1235 }
1236 }
1237
1238 int
1239 nfs_node_set_busy2(nfsnode_t np1, nfsnode_t np2, thread_t thd)
1240 {
1241 nfsnode_t first, second;
1242 int error;
1243
1244 first = (np1 > np2) ? np1 : np2;
1245 second = (np1 > np2) ? np2 : np1;
1246 if ((error = nfs_node_set_busy(first, thd))) {
1247 return error;
1248 }
1249 if (np1 == np2) {
1250 return error;
1251 }
1252 if ((error = nfs_node_set_busy(second, thd))) {
1253 nfs_node_clear_busy(first);
1254 }
1255 return error;
1256 }
1257
1258 void
1259 nfs_node_clear_busy2(nfsnode_t np1, nfsnode_t np2)
1260 {
1261 nfs_node_clear_busy(np1);
1262 if (np1 != np2) {
1263 nfs_node_clear_busy(np2);
1264 }
1265 }
1266
1267 /* helper function to sort four nodes in reverse address order (no dupes) */
1268 static void
1269 nfs_node_sort4(nfsnode_t np1, nfsnode_t np2, nfsnode_t np3, nfsnode_t np4, nfsnode_t *list, int *lcntp)
1270 {
1271 nfsnode_t na[2], nb[2];
1272 int a, b, i, lcnt;
1273
1274 /* sort pairs then merge */
1275 na[0] = (np1 > np2) ? np1 : np2;
1276 na[1] = (np1 > np2) ? np2 : np1;
1277 nb[0] = (np3 > np4) ? np3 : np4;
1278 nb[1] = (np3 > np4) ? np4 : np3;
1279 for (a = b = i = lcnt = 0; i < 4; i++) {
1280 if (a >= 2) {
1281 list[lcnt] = nb[b++];
1282 } else if ((b >= 2) || (na[a] >= nb[b])) {
1283 list[lcnt] = na[a++];
1284 } else {
1285 list[lcnt] = nb[b++];
1286 }
1287 if ((lcnt <= 0) || (list[lcnt] != list[lcnt - 1])) {
1288 lcnt++; /* omit dups */
1289 }
1290 }
1291 if (list[lcnt - 1] == NULL) {
1292 lcnt--;
1293 }
1294 *lcntp = lcnt;
1295 }
1296
1297 int
1298 nfs_node_set_busy4(nfsnode_t np1, nfsnode_t np2, nfsnode_t np3, nfsnode_t np4, thread_t thd)
1299 {
1300 nfsnode_t list[4];
1301 int i, lcnt, error;
1302
1303 nfs_node_sort4(np1, np2, np3, np4, list, &lcnt);
1304
1305 /* Now we can lock using list[0 - lcnt-1] */
1306 for (i = 0; i < lcnt; ++i) {
1307 if ((error = nfs_node_set_busy(list[i], thd))) {
1308 /* Drop any locks we acquired. */
1309 while (--i >= 0) {
1310 nfs_node_clear_busy(list[i]);
1311 }
1312 return error;
1313 }
1314 }
1315 return 0;
1316 }
1317
1318 void
1319 nfs_node_clear_busy4(nfsnode_t np1, nfsnode_t np2, nfsnode_t np3, nfsnode_t np4)
1320 {
1321 nfsnode_t list[4];
1322 int lcnt;
1323
1324 nfs_node_sort4(np1, np2, np3, np4, list, &lcnt);
1325 while (--lcnt >= 0) {
1326 nfs_node_clear_busy(list[lcnt]);
1327 }
1328 }
1329
1330 /*
1331 * Acquire an NFS node data lock
1332 */
1333 void
1334 nfs_data_lock(nfsnode_t np, int locktype)
1335 {
1336 nfs_data_lock_internal(np, locktype, 1);
1337 }
1338 void
1339 nfs_data_lock_noupdate(nfsnode_t np, int locktype)
1340 {
1341 nfs_data_lock_internal(np, locktype, 0);
1342 }
1343 void
1344 nfs_data_lock_internal(nfsnode_t np, int locktype, int updatesize)
1345 {
1346 FSDBG_TOP(270, np, locktype, np->n_datalockowner, 0);
1347 if (locktype == NFS_DATA_LOCK_SHARED) {
1348 if (updatesize && ISSET(np->n_flag, NUPDATESIZE)) {
1349 nfs_data_update_size(np, 0);
1350 }
1351 lck_rw_lock_shared(&np->n_datalock);
1352 } else {
1353 lck_rw_lock_exclusive(&np->n_datalock);
1354 np->n_datalockowner = current_thread();
1355 if (updatesize && ISSET(np->n_flag, NUPDATESIZE)) {
1356 nfs_data_update_size(np, 1);
1357 }
1358 }
1359 FSDBG_BOT(270, np, locktype, np->n_datalockowner, 0);
1360 }
1361
1362 /*
1363 * Release an NFS node data lock
1364 */
1365 void
1366 nfs_data_unlock(nfsnode_t np)
1367 {
1368 nfs_data_unlock_internal(np, 1);
1369 }
1370 void
1371 nfs_data_unlock_noupdate(nfsnode_t np)
1372 {
1373 nfs_data_unlock_internal(np, 0);
1374 }
1375 void
1376 nfs_data_unlock_internal(nfsnode_t np, int updatesize)
1377 {
1378 int mine = (np->n_datalockowner == current_thread());
1379 FSDBG_TOP(271, np, np->n_datalockowner, current_thread(), 0);
1380 if (updatesize && mine && ISSET(np->n_flag, NUPDATESIZE)) {
1381 nfs_data_update_size(np, 1);
1382 }
1383 np->n_datalockowner = NULL;
1384 lck_rw_done(&np->n_datalock);
1385 if (updatesize && !mine && ISSET(np->n_flag, NUPDATESIZE)) {
1386 nfs_data_update_size(np, 0);
1387 }
1388 FSDBG_BOT(271, np, np->n_datalockowner, current_thread(), 0);
1389 }
1390
1391
1392 /*
1393 * update an NFS node's size
1394 */
1395 void
1396 nfs_data_update_size(nfsnode_t np, int datalocked)
1397 {
1398 int error;
1399
1400 FSDBG_TOP(272, np, np->n_flag, np->n_size, np->n_newsize);
1401 if (!datalocked) {
1402 nfs_data_lock(np, NFS_DATA_LOCK_EXCLUSIVE);
1403 /* grabbing data lock will automatically update size */
1404 nfs_data_unlock(np);
1405 FSDBG_BOT(272, np, np->n_flag, np->n_size, np->n_newsize);
1406 return;
1407 }
1408 error = nfs_node_lock(np);
1409 if (error || !ISSET(np->n_flag, NUPDATESIZE)) {
1410 if (!error) {
1411 nfs_node_unlock(np);
1412 }
1413 FSDBG_BOT(272, np, np->n_flag, np->n_size, np->n_newsize);
1414 return;
1415 }
1416 CLR(np->n_flag, NUPDATESIZE);
1417 np->n_size = np->n_newsize;
1418 /* make sure we invalidate buffers the next chance we get */
1419 SET(np->n_flag, NNEEDINVALIDATE);
1420 nfs_node_unlock(np);
1421 ubc_setsize(NFSTOV(np), (off_t)np->n_size); /* XXX error? */
1422 FSDBG_BOT(272, np, np->n_flag, np->n_size, np->n_newsize);
1423 }
1424
1425 #define DODEBUG 1
1426
1427 int
1428 nfs_mount_is_dirty(mount_t mp)
1429 {
1430 u_long i;
1431 nfsnode_t np;
1432 #ifdef DODEBUG
1433 struct timeval now, then, diff;
1434 u_long ncnt = 0;
1435 microuptime(&now);
1436 #endif
1437 lck_mtx_lock(nfs_node_hash_mutex);
1438 for (i = 0; i <= nfsnodehash; i++) {
1439 LIST_FOREACH(np, &nfsnodehashtbl[i], n_hash) {
1440 #ifdef DODEBUG
1441 ncnt++;
1442 #endif
1443 if (np->n_mount == mp && !LIST_EMPTY(&np->n_dirtyblkhd)) {
1444 goto out;
1445 }
1446 }
1447 }
1448 out:
1449 lck_mtx_unlock(nfs_node_hash_mutex);
1450 #ifdef DODEBUG
1451 microuptime(&then);
1452 timersub(&then, &now, &diff);
1453
1454 NFS_DBG(NFS_FAC_SOCK, 7, "mount_is_dirty for %s took %lld mics for %ld slots and %ld nodes return %d\n",
1455 vfs_statfs(mp)->f_mntfromname, (uint64_t)diff.tv_sec * 1000000LL + diff.tv_usec, i, ncnt, (i <= nfsnodehash));
1456 #endif
1457
1458 return i <= nfsnodehash;
1459 }
1460
1461 #endif /* CONFIG_NFS_CLIENT */