1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
30 * SUCH DAMAGE.
31 *
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39
40 #include <sys/capsicum.h>
41
42 /*
43 * generally, I don't like #includes inside .h files, but it seems to
44 * be the easiest way to handle the port.
45 */
46 #include <sys/fail.h>
47 #include <sys/hash.h>
48 #include <sys/sysctl.h>
49 #include <fs/nfs/nfsport.h>
50 #include <netinet/in_fib.h>
51 #include <netinet/if_ether.h>
52 #include <netinet6/ip6_var.h>
53 #include <net/if_types.h>
54
55 #include <fs/nfsclient/nfs_kdtrace.h>
56
57 #ifdef KDTRACE_HOOKS
58 dtrace_nfsclient_attrcache_flush_probe_func_t
59 dtrace_nfscl_attrcache_flush_done_probe;
60 uint32_t nfscl_attrcache_flush_done_id;
61
62 dtrace_nfsclient_attrcache_get_hit_probe_func_t
63 dtrace_nfscl_attrcache_get_hit_probe;
64 uint32_t nfscl_attrcache_get_hit_id;
65
66 dtrace_nfsclient_attrcache_get_miss_probe_func_t
67 dtrace_nfscl_attrcache_get_miss_probe;
68 uint32_t nfscl_attrcache_get_miss_id;
69
70 dtrace_nfsclient_attrcache_load_probe_func_t
71 dtrace_nfscl_attrcache_load_done_probe;
72 uint32_t nfscl_attrcache_load_done_id;
73 #endif /* !KDTRACE_HOOKS */
74
75 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
76 extern struct vop_vector newnfs_vnodeops;
77 extern struct vop_vector newnfs_fifoops;
78 extern uma_zone_t newnfsnode_zone;
79 extern struct buf_ops buf_ops_newnfs;
80 extern int ncl_pbuf_freecnt;
81 extern short nfsv4_cbport;
82 extern int nfscl_enablecallb;
83 extern int nfs_numnfscbd;
84 extern int nfscl_inited;
85 struct mtx ncl_iod_mutex;
86 NFSDLOCKMUTEX;
87
88 extern void (*ncl_call_invalcaches)(struct vnode *);
89
90 SYSCTL_DECL(_vfs_nfs);
91 static int ncl_fileid_maxwarnings = 10;
92 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
93 &ncl_fileid_maxwarnings, 0,
94 "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
95 static volatile int ncl_fileid_nwarnings;
96
97 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
98 struct nfsvattr *);
99
100 /*
101 * Comparison function for vfs_hash functions.
102 */
103 int
104 newnfs_vncmpf(struct vnode *vp, void *arg)
105 {
106 struct nfsfh *nfhp = (struct nfsfh *)arg;
107 struct nfsnode *np = VTONFS(vp);
108
109 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
110 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
111 return (1);
112 return (0);
113 }
114
115 /*
116 * Look up a vnode/nfsnode by file handle.
117 * Callers must check for mount points!!
118 * In all cases, a pointer to a
119 * nfsnode structure is returned.
120 * This variant takes a "struct nfsfh *" as second argument and uses
121 * that structure up, either by hanging off the nfsnode or FREEing it.
122 */
123 int
124 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
125 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
126 void *stuff, int lkflags)
127 {
128 struct nfsnode *np, *dnp;
129 struct vnode *vp, *nvp;
130 struct nfsv4node *newd, *oldd;
131 int error;
132 u_int hash;
133 struct nfsmount *nmp;
134
135 nmp = VFSTONFS(mntp);
136 dnp = VTONFS(dvp);
137 *npp = NULL;
138
139 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
140
141 error = vfs_hash_get(mntp, hash, lkflags,
142 td, &nvp, newnfs_vncmpf, nfhp);
143 if (error == 0 && nvp != NULL) {
144 /*
145 * I believe there is a slight chance that vgonel() could
146 * get called on this vnode between when NFSVOPLOCK() drops
147 * the VI_LOCK() and vget() acquires it again, so that it
148 * hasn't yet had v_usecount incremented. If this were to
149 * happen, the VI_DOOMED flag would be set, so check for
150 * that here. Since we now have the v_usecount incremented,
151 * we should be ok until we vrele() it, if the VI_DOOMED
152 * flag isn't set now.
153 */
154 VI_LOCK(nvp);
155 if ((nvp->v_iflag & VI_DOOMED)) {
156 VI_UNLOCK(nvp);
157 vrele(nvp);
158 error = ENOENT;
159 } else {
160 VI_UNLOCK(nvp);
161 }
162 }
163 if (error) {
164 FREE((caddr_t)nfhp, M_NFSFH);
165 return (error);
166 }
167 if (nvp != NULL) {
168 np = VTONFS(nvp);
169 /*
170 * For NFSv4, check to see if it is the same name and
171 * replace the name, if it is different.
172 */
173 oldd = newd = NULL;
174 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
175 nvp->v_type == VREG &&
176 (np->n_v4->n4_namelen != cnp->cn_namelen ||
177 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
178 cnp->cn_namelen) ||
179 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
180 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
181 dnp->n_fhp->nfh_len))) {
182 MALLOC(newd, struct nfsv4node *,
183 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
184 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
185 NFSLOCKNODE(np);
186 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
187 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
188 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
189 cnp->cn_namelen) ||
190 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
191 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
192 dnp->n_fhp->nfh_len))) {
193 oldd = np->n_v4;
194 np->n_v4 = newd;
195 newd = NULL;
196 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
197 np->n_v4->n4_namelen = cnp->cn_namelen;
198 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
199 dnp->n_fhp->nfh_len);
200 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
201 cnp->cn_namelen);
202 }
203 NFSUNLOCKNODE(np);
204 }
205 if (newd != NULL)
206 FREE((caddr_t)newd, M_NFSV4NODE);
207 if (oldd != NULL)
208 FREE((caddr_t)oldd, M_NFSV4NODE);
209 *npp = np;
210 FREE((caddr_t)nfhp, M_NFSFH);
211 return (0);
212 }
213 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
214
215 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
216 if (error) {
217 uma_zfree(newnfsnode_zone, np);
218 FREE((caddr_t)nfhp, M_NFSFH);
219 return (error);
220 }
221 vp = nvp;
222 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
223 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
224 vp->v_data = np;
225 np->n_vnode = vp;
226 /*
227 * Initialize the mutex even if the vnode is going to be a loser.
228 * This simplifies the logic in reclaim, which can then unconditionally
229 * destroy the mutex (in the case of the loser, or if hash_insert
230 * happened to return an error no special casing is needed).
231 */
232 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
233 lockinit(&np->n_excl, PVFS, "nfsupg", VLKTIMEOUT, LK_NOSHARE |
234 LK_CANRECURSE);
235
236 /*
237 * Are we getting the root? If so, make sure the vnode flags
238 * are correct
239 */
240 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
241 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
242 if (vp->v_type == VNON)
243 vp->v_type = VDIR;
244 vp->v_vflag |= VV_ROOT;
245 }
246
247 np->n_fhp = nfhp;
248 /*
249 * For NFSv4, we have to attach the directory file handle and
250 * file name, so that Open Ops can be done later.
251 */
252 if (nmp->nm_flag & NFSMNT_NFSV4) {
253 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
254 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
255 M_WAITOK);
256 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
257 np->n_v4->n4_namelen = cnp->cn_namelen;
258 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
259 dnp->n_fhp->nfh_len);
260 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
261 cnp->cn_namelen);
262 } else {
263 np->n_v4 = NULL;
264 }
265
266 /*
267 * NFS supports recursive and shared locking.
268 */
269 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
270 VN_LOCK_AREC(vp);
271 VN_LOCK_ASHARE(vp);
272 error = insmntque(vp, mntp);
273 if (error != 0) {
274 *npp = NULL;
275 mtx_destroy(&np->n_mtx);
276 lockdestroy(&np->n_excl);
277 FREE((caddr_t)nfhp, M_NFSFH);
278 if (np->n_v4 != NULL)
279 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
280 uma_zfree(newnfsnode_zone, np);
281 return (error);
282 }
283 error = vfs_hash_insert(vp, hash, lkflags,
284 td, &nvp, newnfs_vncmpf, nfhp);
285 if (error)
286 return (error);
287 if (nvp != NULL) {
288 *npp = VTONFS(nvp);
289 /* vfs_hash_insert() vput()'s the losing vnode */
290 return (0);
291 }
292 *npp = np;
293
294 return (0);
295 }
296
297 /*
298 * Another variant of nfs_nget(). This one is only used by reopen. It
299 * takes almost the same args as nfs_nget(), but only succeeds if an entry
300 * exists in the cache. (Since files should already be "open" with a
301 * vnode ref cnt on the node when reopen calls this, it should always
302 * succeed.)
303 * Also, don't get a vnode lock, since it may already be locked by some
304 * other process that is handling it. This is ok, since all other threads
305 * on the client are blocked by the nfsc_lock being exclusively held by the
306 * caller of this function.
307 */
308 int
309 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
310 struct thread *td, struct nfsnode **npp)
311 {
312 struct vnode *nvp;
313 u_int hash;
314 struct nfsfh *nfhp;
315 int error;
316
317 *npp = NULL;
318 /* For forced dismounts, just return error. */
319 if (NFSCL_FORCEDISM(mntp))
320 return (EINTR);
321 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
322 M_NFSFH, M_WAITOK);
323 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
324 nfhp->nfh_len = fhsize;
325
326 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
327
328 /*
329 * First, try to get the vnode locked, but don't block for the lock.
330 */
331 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
332 newnfs_vncmpf, nfhp);
333 if (error == 0 && nvp != NULL) {
334 NFSVOPUNLOCK(nvp, 0);
335 } else if (error == EBUSY) {
336 /*
337 * It is safe so long as a vflush() with
338 * FORCECLOSE has not been done. Since the Renew thread is
339 * stopped and the MNTK_UNMOUNTF flag is set before doing
340 * a vflush() with FORCECLOSE, we should be ok here.
341 */
342 if (NFSCL_FORCEDISM(mntp))
343 error = EINTR;
344 else {
345 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
346 if (nvp == NULL) {
347 error = ENOENT;
348 } else if ((nvp->v_iflag & VI_DOOMED) != 0) {
349 error = ENOENT;
350 vrele(nvp);
351 } else {
352 error = 0;
353 }
354 }
355 }
356 FREE(nfhp, M_NFSFH);
357 if (error)
358 return (error);
359 if (nvp != NULL) {
360 *npp = VTONFS(nvp);
361 return (0);
362 }
363 return (EINVAL);
364 }
365
366 static void
367 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
368 struct nfsvattr *newnap)
369 {
370 int off;
371
372 if (ncl_fileid_maxwarnings >= 0 &&
373 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
374 return;
375 off = 0;
376 if (ncl_fileid_maxwarnings >= 0) {
377 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
378 off = 1;
379 }
380
381 printf("newnfs: server '%s' error: fileid changed. "
382 "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
383 "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
384 nmp->nm_com.nmcom_hostname,
385 (uintmax_t)nmp->nm_fsid[0],
386 (uintmax_t)nmp->nm_fsid[1],
387 (uintmax_t)oldnap->na_fileid,
388 (uintmax_t)newnap->na_fileid);
389
390 if (off)
391 printf("newnfs: Logged %d times about fileid corruption; "
392 "going quiet to avoid spamming logs excessively. (Limit "
393 "is: %d).\n", ncl_fileid_nwarnings,
394 ncl_fileid_maxwarnings);
395 }
396
397 /*
398 * Load the attribute cache (that lives in the nfsnode entry) with
399 * the attributes of the second argument and
400 * Iff vaper not NULL
401 * copy the attributes to *vaper
402 * Similar to nfs_loadattrcache(), except the attributes are passed in
403 * instead of being parsed out of the mbuf list.
404 */
405 int
406 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
407 void *stuff, int writeattr, int dontshrink)
408 {
409 struct vnode *vp = *vpp;
410 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
411 struct nfsnode *np;
412 struct nfsmount *nmp;
413 struct timespec mtime_save;
414 u_quad_t nsize;
415 int setnsize, error, force_fid_err;
416
417 error = 0;
418 setnsize = 0;
419 nsize = 0;
420
421 /*
422 * If v_type == VNON it is a new node, so fill in the v_type,
423 * n_mtime fields. Check to see if it represents a special
424 * device, and if so, check for a possible alias. Once the
425 * correct vnode has been obtained, fill in the rest of the
426 * information.
427 */
428 np = VTONFS(vp);
429 NFSLOCKNODE(np);
430 if (vp->v_type != nvap->va_type) {
431 vp->v_type = nvap->va_type;
432 if (vp->v_type == VFIFO)
433 vp->v_op = &newnfs_fifoops;
434 np->n_mtime = nvap->va_mtime;
435 }
436 nmp = VFSTONFS(vp->v_mount);
437 vap = &np->n_vattr.na_vattr;
438 mtime_save = vap->va_mtime;
439 if (writeattr) {
440 np->n_vattr.na_filerev = nap->na_filerev;
441 np->n_vattr.na_size = nap->na_size;
442 np->n_vattr.na_mtime = nap->na_mtime;
443 np->n_vattr.na_ctime = nap->na_ctime;
444 np->n_vattr.na_fsid = nap->na_fsid;
445 np->n_vattr.na_mode = nap->na_mode;
446 } else {
447 force_fid_err = 0;
448 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
449 force_fid_err);
450 /*
451 * BROKEN NFS SERVER OR MIDDLEWARE
452 *
453 * Certain NFS servers (certain old proprietary filers ca.
454 * 2006) or broken middleboxes (e.g. WAN accelerator products)
455 * will respond to GETATTR requests with results for a
456 * different fileid.
457 *
458 * The WAN accelerator we've observed not only serves stale
459 * cache results for a given file, it also occasionally serves
460 * results for wholly different files. This causes surprising
461 * problems; for example the cached size attribute of a file
462 * may truncate down and then back up, resulting in zero
463 * regions in file contents read by applications. We observed
464 * this reliably with Clang and .c files during parallel build.
465 * A pcap revealed packet fragmentation and GETATTR RPC
466 * responses with wholly wrong fileids.
467 */
468 if ((np->n_vattr.na_fileid != 0 &&
469 np->n_vattr.na_fileid != nap->na_fileid) ||
470 force_fid_err) {
471 nfscl_warn_fileid(nmp, &np->n_vattr, nap);
472 error = EIDRM;
473 goto out;
474 }
475 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
476 sizeof (struct nfsvattr));
477 }
478
479 /*
480 * For NFSv4, if the node's fsid is not equal to the mount point's
481 * fsid, return the low order 32bits of the node's fsid. This
482 * allows getcwd(3) to work. There is a chance that the fsid might
483 * be the same as a local fs, but since this is in an NFS mount
484 * point, I don't think that will cause any problems?
485 */
486 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
487 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
488 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
489 /*
490 * va_fsid needs to be set to some value derived from
491 * np->n_vattr.na_filesid that is not equal
492 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
493 * from the value used for the top level server volume
494 * in the mounted subtree.
495 */
496 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
497 (uint32_t)np->n_vattr.na_filesid[0])
498 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
499 else
500 vap->va_fsid = (uint32_t)hash32_buf(
501 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
502 } else
503 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
504 np->n_attrstamp = time_second;
505 if (vap->va_size != np->n_size) {
506 if (vap->va_type == VREG) {
507 if (dontshrink && vap->va_size < np->n_size) {
508 /*
509 * We've been told not to shrink the file;
510 * zero np->n_attrstamp to indicate that
511 * the attributes are stale.
512 */
513 vap->va_size = np->n_size;
514 np->n_attrstamp = 0;
515 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
516 vnode_pager_setsize(vp, np->n_size);
517 } else if (np->n_flag & NMODIFIED) {
518 /*
519 * We've modified the file: Use the larger
520 * of our size, and the server's size.
521 */
522 if (vap->va_size < np->n_size) {
523 vap->va_size = np->n_size;
524 } else {
525 np->n_size = vap->va_size;
526 np->n_flag |= NSIZECHANGED;
527 }
528 vnode_pager_setsize(vp, np->n_size);
529 } else if (vap->va_size < np->n_size) {
530 /*
531 * When shrinking the size, the call to
532 * vnode_pager_setsize() cannot be done
533 * with the mutex held, so delay it until
534 * after the mtx_unlock call.
535 */
536 nsize = np->n_size = vap->va_size;
537 np->n_flag |= NSIZECHANGED;
538 setnsize = 1;
539 } else {
540 np->n_size = vap->va_size;
541 np->n_flag |= NSIZECHANGED;
542 vnode_pager_setsize(vp, np->n_size);
543 }
544 } else {
545 np->n_size = vap->va_size;
546 }
547 }
548 /*
549 * The following checks are added to prevent a race between (say)
550 * a READDIR+ and a WRITE.
551 * READDIR+, WRITE requests sent out.
552 * READDIR+ resp, WRITE resp received on client.
553 * However, the WRITE resp was handled before the READDIR+ resp
554 * causing the post op attrs from the write to be loaded first
555 * and the attrs from the READDIR+ to be loaded later. If this
556 * happens, we have stale attrs loaded into the attrcache.
557 * We detect this by for the mtime moving back. We invalidate the
558 * attrcache when this happens.
559 */
560 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
561 /* Size changed or mtime went backwards */
562 np->n_attrstamp = 0;
563 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
564 }
565 if (vaper != NULL) {
566 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
567 if (np->n_flag & NCHG) {
568 if (np->n_flag & NACC)
569 vaper->va_atime = np->n_atim;
570 if (np->n_flag & NUPD)
571 vaper->va_mtime = np->n_mtim;
572 }
573 }
574
575 out:
576 #ifdef KDTRACE_HOOKS
577 if (np->n_attrstamp != 0)
578 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
579 #endif
580 NFSUNLOCKNODE(np);
581 if (setnsize)
582 vnode_pager_setsize(vp, nsize);
583 return (error);
584 }
585
586 /*
587 * Fill in the client id name. For these bytes:
588 * 1 - they must be unique
589 * 2 - they should be persistent across client reboots
590 * 1 is more critical than 2
591 * Use the mount point's unique id plus either the uuid or, if that
592 * isn't set, random junk.
593 */
594 void
595 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
596 {
597 int uuidlen;
598
599 /*
600 * First, put in the 64bit mount point identifier.
601 */
602 if (idlen >= sizeof (u_int64_t)) {
603 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
604 cp += sizeof (u_int64_t);
605 idlen -= sizeof (u_int64_t);
606 }
607
608 /*
609 * If uuid is non-zero length, use it.
610 */
611 uuidlen = strlen(uuid);
612 if (uuidlen > 0 && idlen >= uuidlen) {
613 NFSBCOPY(uuid, cp, uuidlen);
614 cp += uuidlen;
615 idlen -= uuidlen;
616 }
617
618 /*
619 * This only normally happens if the uuid isn't set.
620 */
621 while (idlen > 0) {
622 *cp++ = (u_int8_t)(arc4random() % 256);
623 idlen--;
624 }
625 }
626
627 /*
628 * Fill in a lock owner name. For now, pid + the process's creation time.
629 */
630 void
631 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
632 {
633 union {
634 u_int32_t lval;
635 u_int8_t cval[4];
636 } tl;
637 struct proc *p;
638
639 if (id == NULL) {
640 /* Return the single open_owner of all 0 bytes. */
641 bzero(cp, NFSV4CL_LOCKNAMELEN);
642 return;
643 }
644 if ((flags & F_POSIX) != 0) {
645 p = (struct proc *)id;
646 tl.lval = p->p_pid;
647 *cp++ = tl.cval[0];
648 *cp++ = tl.cval[1];
649 *cp++ = tl.cval[2];
650 *cp++ = tl.cval[3];
651 tl.lval = p->p_stats->p_start.tv_sec;
652 *cp++ = tl.cval[0];
653 *cp++ = tl.cval[1];
654 *cp++ = tl.cval[2];
655 *cp++ = tl.cval[3];
656 tl.lval = p->p_stats->p_start.tv_usec;
657 *cp++ = tl.cval[0];
658 *cp++ = tl.cval[1];
659 *cp++ = tl.cval[2];
660 *cp = tl.cval[3];
661 } else if ((flags & F_FLOCK) != 0) {
662 bcopy(&id, cp, sizeof(id));
663 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
664 } else {
665 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
666 bzero(cp, NFSV4CL_LOCKNAMELEN);
667 }
668 }
669
670 /*
671 * Find the parent process for the thread passed in as an argument.
672 * If none exists, return NULL, otherwise return a thread for the parent.
673 * (Can be any of the threads, since it is only used for td->td_proc.)
674 */
675 NFSPROC_T *
676 nfscl_getparent(struct thread *td)
677 {
678 struct proc *p;
679 struct thread *ptd;
680
681 if (td == NULL)
682 return (NULL);
683 p = td->td_proc;
684 if (p->p_pid == 0)
685 return (NULL);
686 p = p->p_pptr;
687 if (p == NULL)
688 return (NULL);
689 ptd = TAILQ_FIRST(&p->p_threads);
690 return (ptd);
691 }
692
693 /*
694 * Start up the renew kernel thread.
695 */
696 static void
697 start_nfscl(void *arg)
698 {
699 struct nfsclclient *clp;
700 struct thread *td;
701
702 clp = (struct nfsclclient *)arg;
703 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
704 nfscl_renewthread(clp, td);
705 kproc_exit(0);
706 }
707
708 void
709 nfscl_start_renewthread(struct nfsclclient *clp)
710 {
711
712 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
713 "nfscl");
714 }
715
716 /*
717 * Handle wcc_data.
718 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
719 * as the first Op after PutFH.
720 * (For NFSv4, the postop attributes are after the Op, so they can't be
721 * parsed here. A separate call to nfscl_postop_attr() is required.)
722 */
723 int
724 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
725 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
726 {
727 u_int32_t *tl;
728 struct nfsnode *np = VTONFS(vp);
729 struct nfsvattr nfsva;
730 int error = 0;
731
732 if (wccflagp != NULL)
733 *wccflagp = 0;
734 if (nd->nd_flag & ND_NFSV3) {
735 *flagp = 0;
736 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
737 if (*tl == newnfs_true) {
738 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
739 if (wccflagp != NULL) {
740 mtx_lock(&np->n_mtx);
741 *wccflagp = (np->n_mtime.tv_sec ==
742 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
743 np->n_mtime.tv_nsec ==
744 fxdr_unsigned(u_int32_t, *(tl + 3)));
745 mtx_unlock(&np->n_mtx);
746 }
747 }
748 error = nfscl_postop_attr(nd, nap, flagp, stuff);
749 if (wccflagp != NULL && *flagp == 0)
750 *wccflagp = 0;
751 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
752 == (ND_NFSV4 | ND_V4WCCATTR)) {
753 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
754 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
755 NULL, NULL, NULL, NULL, NULL);
756 if (error)
757 return (error);
758 /*
759 * Get rid of Op# and status for next op.
760 */
761 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
762 if (*++tl)
763 nd->nd_flag |= ND_NOMOREDATA;
764 if (wccflagp != NULL &&
765 nfsva.na_vattr.va_mtime.tv_sec != 0) {
766 mtx_lock(&np->n_mtx);
767 *wccflagp = (np->n_mtime.tv_sec ==
768 nfsva.na_vattr.va_mtime.tv_sec &&
769 np->n_mtime.tv_nsec ==
770 nfsva.na_vattr.va_mtime.tv_sec);
771 mtx_unlock(&np->n_mtx);
772 }
773 }
774 nfsmout:
775 return (error);
776 }
777
778 /*
779 * Get postop attributes.
780 */
781 int
782 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
783 void *stuff)
784 {
785 u_int32_t *tl;
786 int error = 0;
787
788 *retp = 0;
789 if (nd->nd_flag & ND_NOMOREDATA)
790 return (error);
791 if (nd->nd_flag & ND_NFSV3) {
792 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
793 *retp = fxdr_unsigned(int, *tl);
794 } else if (nd->nd_flag & ND_NFSV4) {
795 /*
796 * For NFSv4, the postop attr are at the end, so no point
797 * in looking if nd_repstat != 0.
798 */
799 if (!nd->nd_repstat) {
800 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
801 if (*(tl + 1))
802 /* should never happen since nd_repstat != 0 */
803 nd->nd_flag |= ND_NOMOREDATA;
804 else
805 *retp = 1;
806 }
807 } else if (!nd->nd_repstat) {
808 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
809 *retp = 1;
810 }
811 if (*retp) {
812 error = nfsm_loadattr(nd, nap);
813 if (error)
814 *retp = 0;
815 }
816 nfsmout:
817 return (error);
818 }
819
820 /*
821 * Fill in the setable attributes. The full argument indicates whether
822 * to fill in them all or just mode and time.
823 */
824 void
825 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
826 struct vnode *vp, int flags, u_int32_t rdev)
827 {
828 u_int32_t *tl;
829 struct nfsv2_sattr *sp;
830 nfsattrbit_t attrbits;
831
832 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
833 case ND_NFSV2:
834 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
835 if (vap->va_mode == (mode_t)VNOVAL)
836 sp->sa_mode = newnfs_xdrneg1;
837 else
838 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
839 if (vap->va_uid == (uid_t)VNOVAL)
840 sp->sa_uid = newnfs_xdrneg1;
841 else
842 sp->sa_uid = txdr_unsigned(vap->va_uid);
843 if (vap->va_gid == (gid_t)VNOVAL)
844 sp->sa_gid = newnfs_xdrneg1;
845 else
846 sp->sa_gid = txdr_unsigned(vap->va_gid);
847 if (flags & NFSSATTR_SIZE0)
848 sp->sa_size = 0;
849 else if (flags & NFSSATTR_SIZENEG1)
850 sp->sa_size = newnfs_xdrneg1;
851 else if (flags & NFSSATTR_SIZERDEV)
852 sp->sa_size = txdr_unsigned(rdev);
853 else
854 sp->sa_size = txdr_unsigned(vap->va_size);
855 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
856 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
857 break;
858 case ND_NFSV3:
859 if (vap->va_mode != (mode_t)VNOVAL) {
860 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
861 *tl++ = newnfs_true;
862 *tl = txdr_unsigned(vap->va_mode);
863 } else {
864 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
865 *tl = newnfs_false;
866 }
867 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
868 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
869 *tl++ = newnfs_true;
870 *tl = txdr_unsigned(vap->va_uid);
871 } else {
872 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
873 *tl = newnfs_false;
874 }
875 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
876 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
877 *tl++ = newnfs_true;
878 *tl = txdr_unsigned(vap->va_gid);
879 } else {
880 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
881 *tl = newnfs_false;
882 }
883 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
884 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
885 *tl++ = newnfs_true;
886 txdr_hyper(vap->va_size, tl);
887 } else {
888 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
889 *tl = newnfs_false;
890 }
891 if (vap->va_atime.tv_sec != VNOVAL) {
892 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
893 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
894 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
895 txdr_nfsv3time(&vap->va_atime, tl);
896 } else {
897 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
898 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
899 }
900 } else {
901 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
902 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
903 }
904 if (vap->va_mtime.tv_sec != VNOVAL) {
905 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
906 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
907 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
908 txdr_nfsv3time(&vap->va_mtime, tl);
909 } else {
910 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
911 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
912 }
913 } else {
914 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
915 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
916 }
917 break;
918 case ND_NFSV4:
919 NFSZERO_ATTRBIT(&attrbits);
920 if (vap->va_mode != (mode_t)VNOVAL)
921 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
922 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
923 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
924 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
925 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
926 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
927 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
928 if (vap->va_atime.tv_sec != VNOVAL)
929 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
930 if (vap->va_mtime.tv_sec != VNOVAL)
931 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
932 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
933 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
934 break;
935 }
936 }
937
938 /*
939 * nfscl_request() - mostly a wrapper for newnfs_request().
940 */
941 int
942 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
943 struct ucred *cred, void *stuff)
944 {
945 int ret, vers;
946 struct nfsmount *nmp;
947
948 nmp = VFSTONFS(vp->v_mount);
949 if (nd->nd_flag & ND_NFSV4)
950 vers = NFS_VER4;
951 else if (nd->nd_flag & ND_NFSV3)
952 vers = NFS_VER3;
953 else
954 vers = NFS_VER2;
955 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
956 NFS_PROG, vers, NULL, 1, NULL, NULL);
957 return (ret);
958 }
959
960 /*
961 * fill in this bsden's variant of statfs using nfsstatfs.
962 */
963 void
964 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
965 {
966 struct statfs *sbp = (struct statfs *)statfs;
967
968 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
969 sbp->f_bsize = NFS_FABLKSIZE;
970 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
971 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
972 /*
973 * Although sf_abytes is uint64_t and f_bavail is int64_t,
974 * the value after dividing by NFS_FABLKSIZE is small
975 * enough that it will fit in 63bits, so it is ok to
976 * assign it to f_bavail without fear that it will become
977 * negative.
978 */
979 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
980 sbp->f_files = sfp->sf_tfiles;
981 /* Since f_ffree is int64_t, clip it to 63bits. */
982 if (sfp->sf_ffiles > INT64_MAX)
983 sbp->f_ffree = INT64_MAX;
984 else
985 sbp->f_ffree = sfp->sf_ffiles;
986 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
987 /*
988 * The type casts to (int32_t) ensure that this code is
989 * compatible with the old NFS client, in that it will
990 * propagate bit31 to the high order bits. This may or may
991 * not be correct for NFSv2, but since it is a legacy
992 * environment, I'd rather retain backwards compatibility.
993 */
994 sbp->f_bsize = (int32_t)sfp->sf_bsize;
995 sbp->f_blocks = (int32_t)sfp->sf_blocks;
996 sbp->f_bfree = (int32_t)sfp->sf_bfree;
997 sbp->f_bavail = (int32_t)sfp->sf_bavail;
998 sbp->f_files = 0;
999 sbp->f_ffree = 0;
1000 }
1001 }
1002
1003 /*
1004 * Use the fsinfo stuff to update the mount point.
1005 */
1006 void
1007 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1008 {
1009
1010 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1011 fsp->fs_wtpref >= NFS_FABLKSIZE)
1012 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1013 ~(NFS_FABLKSIZE - 1);
1014 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1015 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1016 if (nmp->nm_wsize == 0)
1017 nmp->nm_wsize = fsp->fs_wtmax;
1018 }
1019 if (nmp->nm_wsize < NFS_FABLKSIZE)
1020 nmp->nm_wsize = NFS_FABLKSIZE;
1021 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1022 fsp->fs_rtpref >= NFS_FABLKSIZE)
1023 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1024 ~(NFS_FABLKSIZE - 1);
1025 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1026 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1027 if (nmp->nm_rsize == 0)
1028 nmp->nm_rsize = fsp->fs_rtmax;
1029 }
1030 if (nmp->nm_rsize < NFS_FABLKSIZE)
1031 nmp->nm_rsize = NFS_FABLKSIZE;
1032 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1033 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1034 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1035 ~(NFS_DIRBLKSIZ - 1);
1036 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1037 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1038 if (nmp->nm_readdirsize == 0)
1039 nmp->nm_readdirsize = fsp->fs_rtmax;
1040 }
1041 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1042 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1043 if (fsp->fs_maxfilesize > 0 &&
1044 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1045 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1046 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1047 nmp->nm_state |= NFSSTA_GOTFSINFO;
1048 }
1049
1050 /*
1051 * Lookups source address which should be used to communicate with
1052 * @nmp and stores it inside @pdst.
1053 *
1054 * Returns 0 on success.
1055 */
1056 u_int8_t *
1057 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1058 {
1059 #if defined(INET6) || defined(INET)
1060 int error, fibnum;
1061
1062 fibnum = curthread->td_proc->p_fibnum;
1063 #endif
1064 #ifdef INET
1065 if (nmp->nm_nam->sa_family == AF_INET) {
1066 struct sockaddr_in *sin;
1067 struct nhop4_extended nh_ext;
1068
1069 sin = (struct sockaddr_in *)nmp->nm_nam;
1070 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1071 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1072 &nh_ext);
1073 CURVNET_RESTORE();
1074 if (error != 0)
1075 return (NULL);
1076
1077 if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1078 IN_LOOPBACKNET) {
1079 /* Ignore loopback addresses */
1080 return (NULL);
1081 }
1082
1083 *isinet6p = 0;
1084 *((struct in_addr *)paddr) = nh_ext.nh_src;
1085
1086 return (u_int8_t *)paddr;
1087 }
1088 #endif
1089 #ifdef INET6
1090 if (nmp->nm_nam->sa_family == AF_INET6) {
1091 struct sockaddr_in6 *sin6;
1092
1093 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1094
1095 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1096 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1097 sin6->sin6_scope_id, NULL, paddr, NULL);
1098 CURVNET_RESTORE();
1099 if (error != 0)
1100 return (NULL);
1101
1102 if (IN6_IS_ADDR_LOOPBACK(paddr))
1103 return (NULL);
1104
1105 /* Scope is embedded in */
1106 *isinet6p = 1;
1107
1108 return (u_int8_t *)paddr;
1109 }
1110 #endif
1111 return (NULL);
1112 }
1113
1114 /*
1115 * Copy NFS uid, gids from the cred structure.
1116 */
1117 void
1118 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1119 {
1120 int i;
1121
1122 KASSERT(cr->cr_ngroups >= 0,
1123 ("newnfs_copyincred: negative cr_ngroups"));
1124 nfscr->nfsc_uid = cr->cr_uid;
1125 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1126 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1127 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1128 }
1129
1130
1131 /*
1132 * Do any client specific initialization.
1133 */
1134 void
1135 nfscl_init(void)
1136 {
1137 static int inited = 0;
1138
1139 if (inited)
1140 return;
1141 inited = 1;
1142 nfscl_inited = 1;
1143 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1144 }
1145
1146 /*
1147 * Check each of the attributes to be set, to ensure they aren't already
1148 * the correct value. Disable setting ones already correct.
1149 */
1150 int
1151 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1152 {
1153
1154 if (vap->va_mode != (mode_t)VNOVAL) {
1155 if (vap->va_mode == nvap->na_mode)
1156 vap->va_mode = (mode_t)VNOVAL;
1157 }
1158 if (vap->va_uid != (uid_t)VNOVAL) {
1159 if (vap->va_uid == nvap->na_uid)
1160 vap->va_uid = (uid_t)VNOVAL;
1161 }
1162 if (vap->va_gid != (gid_t)VNOVAL) {
1163 if (vap->va_gid == nvap->na_gid)
1164 vap->va_gid = (gid_t)VNOVAL;
1165 }
1166 if (vap->va_size != VNOVAL) {
1167 if (vap->va_size == nvap->na_size)
1168 vap->va_size = VNOVAL;
1169 }
1170
1171 /*
1172 * We are normally called with only a partially initialized
1173 * VAP. Since the NFSv3 spec says that server may use the
1174 * file attributes to store the verifier, the spec requires
1175 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1176 * in atime, but we can't really assume that all servers will
1177 * so we ensure that our SETATTR sets both atime and mtime.
1178 * Set the VA_UTIMES_NULL flag for this case, so that
1179 * the server's time will be used. This is needed to
1180 * work around a bug in some Solaris servers, where
1181 * setting the time TOCLIENT causes the Setattr RPC
1182 * to return NFS_OK, but not set va_mode.
1183 */
1184 if (vap->va_mtime.tv_sec == VNOVAL) {
1185 vfs_timestamp(&vap->va_mtime);
1186 vap->va_vaflags |= VA_UTIMES_NULL;
1187 }
1188 if (vap->va_atime.tv_sec == VNOVAL)
1189 vap->va_atime = vap->va_mtime;
1190 return (1);
1191 }
1192
1193 /*
1194 * Map nfsv4 errors to errno.h errors.
1195 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1196 * error should only be returned for the Open, Create and Setattr Ops.
1197 * As such, most calls can just pass in 0 for those arguments.
1198 */
1199 int
1200 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1201 {
1202 struct proc *p;
1203
1204 if (error < 10000 || error >= NFSERR_STALEWRITEVERF)
1205 return (error);
1206 if (td != NULL)
1207 p = td->td_proc;
1208 else
1209 p = NULL;
1210 switch (error) {
1211 case NFSERR_BADOWNER:
1212 tprintf(p, LOG_INFO,
1213 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1214 uid, gid);
1215 return (EPERM);
1216 case NFSERR_BADNAME:
1217 case NFSERR_BADCHAR:
1218 printf("nfsv4 char/name not handled by server\n");
1219 return (ENOENT);
1220 case NFSERR_STALECLIENTID:
1221 case NFSERR_STALESTATEID:
1222 case NFSERR_EXPIRED:
1223 case NFSERR_BADSTATEID:
1224 case NFSERR_BADSESSION:
1225 printf("nfsv4 recover err returned %d\n", error);
1226 return (EIO);
1227 case NFSERR_BADHANDLE:
1228 case NFSERR_SERVERFAULT:
1229 case NFSERR_BADTYPE:
1230 case NFSERR_FHEXPIRED:
1231 case NFSERR_RESOURCE:
1232 case NFSERR_MOVED:
1233 case NFSERR_NOFILEHANDLE:
1234 case NFSERR_MINORVERMISMATCH:
1235 case NFSERR_OLDSTATEID:
1236 case NFSERR_BADSEQID:
1237 case NFSERR_LEASEMOVED:
1238 case NFSERR_RECLAIMBAD:
1239 case NFSERR_BADXDR:
1240 case NFSERR_OPILLEGAL:
1241 printf("nfsv4 client/server protocol prob err=%d\n",
1242 error);
1243 return (EIO);
1244 default:
1245 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1246 return (EIO);
1247 };
1248 }
1249
1250 /*
1251 * Check to see if the process for this owner exists. Return 1 if it doesn't
1252 * and 0 otherwise.
1253 */
1254 int
1255 nfscl_procdoesntexist(u_int8_t *own)
1256 {
1257 union {
1258 u_int32_t lval;
1259 u_int8_t cval[4];
1260 } tl;
1261 struct proc *p;
1262 pid_t pid;
1263 int i, ret = 0;
1264
1265 /* For the single open_owner of all 0 bytes, just return 0. */
1266 for (i = 0; i < NFSV4CL_LOCKNAMELEN; i++)
1267 if (own[i] != 0)
1268 break;
1269 if (i == NFSV4CL_LOCKNAMELEN)
1270 return (0);
1271
1272 tl.cval[0] = *own++;
1273 tl.cval[1] = *own++;
1274 tl.cval[2] = *own++;
1275 tl.cval[3] = *own++;
1276 pid = tl.lval;
1277 p = pfind_locked(pid);
1278 if (p == NULL)
1279 return (1);
1280 if (p->p_stats == NULL) {
1281 PROC_UNLOCK(p);
1282 return (0);
1283 }
1284 tl.cval[0] = *own++;
1285 tl.cval[1] = *own++;
1286 tl.cval[2] = *own++;
1287 tl.cval[3] = *own++;
1288 if (tl.lval != p->p_stats->p_start.tv_sec) {
1289 ret = 1;
1290 } else {
1291 tl.cval[0] = *own++;
1292 tl.cval[1] = *own++;
1293 tl.cval[2] = *own++;
1294 tl.cval[3] = *own;
1295 if (tl.lval != p->p_stats->p_start.tv_usec)
1296 ret = 1;
1297 }
1298 PROC_UNLOCK(p);
1299 return (ret);
1300 }
1301
1302 /*
1303 * - nfs pseudo system call for the client
1304 */
1305 /*
1306 * MPSAFE
1307 */
1308 static int
1309 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1310 {
1311 struct file *fp;
1312 struct nfscbd_args nfscbdarg;
1313 struct nfsd_nfscbd_args nfscbdarg2;
1314 struct nameidata nd;
1315 struct nfscl_dumpmntopts dumpmntopts;
1316 cap_rights_t rights;
1317 char *buf;
1318 int error;
1319 struct mount *mp;
1320 struct nfsmount *nmp;
1321
1322 if (uap->flag & NFSSVC_CBADDSOCK) {
1323 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1324 if (error)
1325 return (error);
1326 /*
1327 * Since we don't know what rights might be required,
1328 * pretend that we need them all. It is better to be too
1329 * careful than too reckless.
1330 */
1331 error = fget(td, nfscbdarg.sock,
1332 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1333 if (error)
1334 return (error);
1335 if (fp->f_type != DTYPE_SOCKET) {
1336 fdrop(fp, td);
1337 return (EPERM);
1338 }
1339 error = nfscbd_addsock(fp);
1340 fdrop(fp, td);
1341 if (!error && nfscl_enablecallb == 0) {
1342 nfsv4_cbport = nfscbdarg.port;
1343 nfscl_enablecallb = 1;
1344 }
1345 } else if (uap->flag & NFSSVC_NFSCBD) {
1346 if (uap->argp == NULL)
1347 return (EINVAL);
1348 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1349 sizeof(nfscbdarg2));
1350 if (error)
1351 return (error);
1352 error = nfscbd_nfsd(td, &nfscbdarg2);
1353 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1354 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1355 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1356 dumpmntopts.ndmnt_blen > 1024))
1357 error = EINVAL;
1358 if (error == 0)
1359 error = nfsrv_lookupfilename(&nd,
1360 dumpmntopts.ndmnt_fname, td);
1361 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1362 "nfs") != 0) {
1363 vput(nd.ni_vp);
1364 error = EINVAL;
1365 }
1366 if (error == 0) {
1367 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1368 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1369 dumpmntopts.ndmnt_blen);
1370 vput(nd.ni_vp);
1371 error = copyout(buf, dumpmntopts.ndmnt_buf,
1372 dumpmntopts.ndmnt_blen);
1373 free(buf, M_TEMP);
1374 }
1375 } else if (uap->flag & NFSSVC_FORCEDISM) {
1376 buf = malloc(MNAMELEN + 1, M_TEMP, M_WAITOK);
1377 error = copyinstr(uap->argp, buf, MNAMELEN + 1, NULL);
1378 if (error == 0) {
1379 nmp = NULL;
1380 mtx_lock(&mountlist_mtx);
1381 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
1382 if (strcmp(mp->mnt_stat.f_mntonname, buf) ==
1383 0 && strcmp(mp->mnt_stat.f_fstypename,
1384 "nfs") == 0 && mp->mnt_data != NULL) {
1385 nmp = VFSTONFS(mp);
1386 mtx_lock(&nmp->nm_mtx);
1387 if ((nmp->nm_privflag &
1388 NFSMNTP_FORCEDISM) == 0) {
1389 nmp->nm_privflag |=
1390 (NFSMNTP_FORCEDISM |
1391 NFSMNTP_CANCELRPCS);
1392 mtx_unlock(&nmp->nm_mtx);
1393 } else {
1394 mtx_unlock(&nmp->nm_mtx);
1395 nmp = NULL;
1396 }
1397 break;
1398 }
1399 }
1400 mtx_unlock(&mountlist_mtx);
1401
1402 if (nmp != NULL) {
1403 /*
1404 * Call newnfs_nmcancelreqs() to cause
1405 * any RPCs in progress on the mount point to
1406 * fail.
1407 * This will cause any process waiting for an
1408 * RPC to complete while holding a vnode lock
1409 * on the mounted-on vnode (such as "df" or
1410 * a non-forced "umount") to fail.
1411 * This will unlock the mounted-on vnode so
1412 * a forced dismount can succeed.
1413 * Then clear NFSMNTP_CANCELRPCS and wakeup(),
1414 * so that nfs_unmount() can complete.
1415 */
1416 newnfs_nmcancelreqs(nmp);
1417 mtx_lock(&nmp->nm_mtx);
1418 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
1419 wakeup(nmp);
1420 mtx_unlock(&nmp->nm_mtx);
1421 } else
1422 error = EINVAL;
1423 }
1424 free(buf, M_TEMP);
1425 } else {
1426 error = EINVAL;
1427 }
1428 return (error);
1429 }
1430
1431 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1432
1433 /*
1434 * Called once to initialize data structures...
1435 */
1436 static int
1437 nfscl_modevent(module_t mod, int type, void *data)
1438 {
1439 int error = 0;
1440 static int loaded = 0;
1441
1442 switch (type) {
1443 case MOD_LOAD:
1444 if (loaded)
1445 return (0);
1446 newnfs_portinit();
1447 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1448 nfscl_init();
1449 NFSD_LOCK();
1450 nfsrvd_cbinit(0);
1451 NFSD_UNLOCK();
1452 ncl_call_invalcaches = ncl_invalcaches;
1453 nfsd_call_nfscl = nfssvc_nfscl;
1454 loaded = 1;
1455 break;
1456
1457 case MOD_UNLOAD:
1458 if (nfs_numnfscbd != 0) {
1459 error = EBUSY;
1460 break;
1461 }
1462
1463 /*
1464 * XXX: Unloading of nfscl module is unsupported.
1465 */
1466 #if 0
1467 ncl_call_invalcaches = NULL;
1468 nfsd_call_nfscl = NULL;
1469 /* and get rid of the mutexes */
1470 mtx_destroy(&ncl_iod_mutex);
1471 loaded = 0;
1472 break;
1473 #else
1474 /* FALLTHROUGH */
1475 #endif
1476 default:
1477 error = EOPNOTSUPP;
1478 break;
1479 }
1480 return error;
1481 }
1482 static moduledata_t nfscl_mod = {
1483 "nfscl",
1484 nfscl_modevent,
1485 NULL,
1486 };
1487 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1488
1489 /* So that loader and kldload(2) can find us, wherever we are.. */
1490 MODULE_VERSION(nfscl, 1);
1491 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1492 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1493 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1494 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1495
Cache object: 1c0af549e2bd7fdbabb8ec5cdfe6b967
|