1 /* $NetBSD: nfs_clport.c,v 1.1.1.1 2013/09/30 07:19:24 dholland Exp $ */
2 /*-
3 * Copyright (c) 1989, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Rick Macklem at The University of Guelph.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 */
34
35 #include <sys/cdefs.h>
36 /* __FBSDID("FreeBSD: head/sys/fs/nfsclient/nfs_clport.c 255219 2013-09-05 00:09:56Z pjd "); */
37 __RCSID("$NetBSD: nfs_clport.c,v 1.1.1.1 2013/09/30 07:19:24 dholland Exp $");
38
39 #include "opt_inet6.h"
40 #include "opt_kdtrace.h"
41
42 #include <sys/capability.h>
43
44 /*
45 * generally, I don't like #includes inside .h files, but it seems to
46 * be the easiest way to handle the port.
47 */
48 #include <sys/hash.h>
49 #include <fs/nfs/nfsport.h>
50 #include <netinet/if_ether.h>
51 #include <net/if_types.h>
52
53 #include <fs/nfsclient/nfs_kdtrace.h>
54
55 #ifdef KDTRACE_HOOKS
56 dtrace_nfsclient_attrcache_flush_probe_func_t
57 dtrace_nfscl_attrcache_flush_done_probe;
58 uint32_t nfscl_attrcache_flush_done_id;
59
60 dtrace_nfsclient_attrcache_get_hit_probe_func_t
61 dtrace_nfscl_attrcache_get_hit_probe;
62 uint32_t nfscl_attrcache_get_hit_id;
63
64 dtrace_nfsclient_attrcache_get_miss_probe_func_t
65 dtrace_nfscl_attrcache_get_miss_probe;
66 uint32_t nfscl_attrcache_get_miss_id;
67
68 dtrace_nfsclient_attrcache_load_probe_func_t
69 dtrace_nfscl_attrcache_load_done_probe;
70 uint32_t nfscl_attrcache_load_done_id;
71 #endif /* !KDTRACE_HOOKS */
72
73 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
74 extern struct vop_vector newnfs_vnodeops;
75 extern struct vop_vector newnfs_fifoops;
76 extern uma_zone_t newnfsnode_zone;
77 extern struct buf_ops buf_ops_newnfs;
78 extern int ncl_pbuf_freecnt;
79 extern short nfsv4_cbport;
80 extern int nfscl_enablecallb;
81 extern int nfs_numnfscbd;
82 extern int nfscl_inited;
83 struct mtx nfs_clstate_mutex;
84 struct mtx ncl_iod_mutex;
85 NFSDLOCKMUTEX;
86
87 extern void (*ncl_call_invalcaches)(struct vnode *);
88
89 /*
90 * Comparison function for vfs_hash functions.
91 */
92 int
newnfs_vncmpf(struct vnode * vp,void * arg)93 newnfs_vncmpf(struct vnode *vp, void *arg)
94 {
95 struct nfsfh *nfhp = (struct nfsfh *)arg;
96 struct nfsnode *np = VTONFS(vp);
97
98 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
99 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
100 return (1);
101 return (0);
102 }
103
104 /*
105 * Look up a vnode/nfsnode by file handle.
106 * Callers must check for mount points!!
107 * In all cases, a pointer to a
108 * nfsnode structure is returned.
109 * This variant takes a "struct nfsfh *" as second argument and uses
110 * that structure up, either by hanging off the nfsnode or FREEing it.
111 */
112 int
nfscl_nget(struct mount * mntp,struct vnode * dvp,struct nfsfh * nfhp,struct componentname * cnp,struct thread * td,struct nfsnode ** npp,void * stuff,int lkflags)113 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
114 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
115 void *stuff, int lkflags)
116 {
117 struct nfsnode *np, *dnp;
118 struct vnode *vp, *nvp;
119 struct nfsv4node *newd, *oldd;
120 int error;
121 u_int hash;
122 struct nfsmount *nmp;
123
124 nmp = VFSTONFS(mntp);
125 dnp = VTONFS(dvp);
126 *npp = NULL;
127
128 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
129
130 error = vfs_hash_get(mntp, hash, lkflags,
131 td, &nvp, newnfs_vncmpf, nfhp);
132 if (error == 0 && nvp != NULL) {
133 /*
134 * I believe there is a slight chance that vgonel() could
135 * get called on this vnode between when NFSVOPLOCK() drops
136 * the VI_LOCK() and vget() acquires it again, so that it
137 * hasn't yet had v_usecount incremented. If this were to
138 * happen, the VI_DOOMED flag would be set, so check for
139 * that here. Since we now have the v_usecount incremented,
140 * we should be ok until we vrele() it, if the VI_DOOMED
141 * flag isn't set now.
142 */
143 VI_LOCK(nvp);
144 if ((nvp->v_iflag & VI_DOOMED)) {
145 VI_UNLOCK(nvp);
146 vrele(nvp);
147 error = ENOENT;
148 } else {
149 VI_UNLOCK(nvp);
150 }
151 }
152 if (error) {
153 FREE((caddr_t)nfhp, M_NFSFH);
154 return (error);
155 }
156 if (nvp != NULL) {
157 np = VTONFS(nvp);
158 /*
159 * For NFSv4, check to see if it is the same name and
160 * replace the name, if it is different.
161 */
162 oldd = newd = NULL;
163 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
164 nvp->v_type == VREG &&
165 (np->n_v4->n4_namelen != cnp->cn_namelen ||
166 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
167 cnp->cn_namelen) ||
168 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
169 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
170 dnp->n_fhp->nfh_len))) {
171 MALLOC(newd, struct nfsv4node *,
172 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
173 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
174 NFSLOCKNODE(np);
175 if (newd != NULL && np->n_v4 != NULL && 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 oldd = np->n_v4;
183 np->n_v4 = newd;
184 newd = NULL;
185 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
186 np->n_v4->n4_namelen = cnp->cn_namelen;
187 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
188 dnp->n_fhp->nfh_len);
189 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
190 cnp->cn_namelen);
191 }
192 NFSUNLOCKNODE(np);
193 }
194 if (newd != NULL)
195 FREE((caddr_t)newd, M_NFSV4NODE);
196 if (oldd != NULL)
197 FREE((caddr_t)oldd, M_NFSV4NODE);
198 *npp = np;
199 FREE((caddr_t)nfhp, M_NFSFH);
200 return (0);
201 }
202 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
203
204 error = getnewvnode("newnfs", mntp, &newnfs_vnodeops, &nvp);
205 if (error) {
206 uma_zfree(newnfsnode_zone, np);
207 FREE((caddr_t)nfhp, M_NFSFH);
208 return (error);
209 }
210 vp = nvp;
211 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
212 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
213 vp->v_data = np;
214 np->n_vnode = vp;
215 /*
216 * Initialize the mutex even if the vnode is going to be a loser.
217 * This simplifies the logic in reclaim, which can then unconditionally
218 * destroy the mutex (in the case of the loser, or if hash_insert
219 * happened to return an error no special casing is needed).
220 */
221 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
222
223 /*
224 * Are we getting the root? If so, make sure the vnode flags
225 * are correct
226 */
227 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
228 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
229 if (vp->v_type == VNON)
230 vp->v_type = VDIR;
231 vp->v_vflag |= VV_ROOT;
232 }
233
234 np->n_fhp = nfhp;
235 /*
236 * For NFSv4, we have to attach the directory file handle and
237 * file name, so that Open Ops can be done later.
238 */
239 if (nmp->nm_flag & NFSMNT_NFSV4) {
240 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
241 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
242 M_WAITOK);
243 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
244 np->n_v4->n4_namelen = cnp->cn_namelen;
245 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
246 dnp->n_fhp->nfh_len);
247 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
248 cnp->cn_namelen);
249 } else {
250 np->n_v4 = NULL;
251 }
252
253 /*
254 * NFS supports recursive and shared locking.
255 */
256 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
257 VN_LOCK_AREC(vp);
258 VN_LOCK_ASHARE(vp);
259 error = insmntque(vp, mntp);
260 if (error != 0) {
261 *npp = NULL;
262 mtx_destroy(&np->n_mtx);
263 FREE((caddr_t)nfhp, M_NFSFH);
264 if (np->n_v4 != NULL)
265 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
266 uma_zfree(newnfsnode_zone, np);
267 return (error);
268 }
269 error = vfs_hash_insert(vp, hash, lkflags,
270 td, &nvp, newnfs_vncmpf, nfhp);
271 if (error)
272 return (error);
273 if (nvp != NULL) {
274 *npp = VTONFS(nvp);
275 /* vfs_hash_insert() vput()'s the losing vnode */
276 return (0);
277 }
278 *npp = np;
279
280 return (0);
281 }
282
283 /*
284 * Anothe variant of nfs_nget(). This one is only used by reopen. It
285 * takes almost the same args as nfs_nget(), but only succeeds if an entry
286 * exists in the cache. (Since files should already be "open" with a
287 * vnode ref cnt on the node when reopen calls this, it should always
288 * succeed.)
289 * Also, don't get a vnode lock, since it may already be locked by some
290 * other process that is handling it. This is ok, since all other threads
291 * on the client are blocked by the nfsc_lock being exclusively held by the
292 * caller of this function.
293 */
294 int
nfscl_ngetreopen(struct mount * mntp,u_int8_t * fhp,int fhsize,struct thread * td,struct nfsnode ** npp)295 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
296 struct thread *td, struct nfsnode **npp)
297 {
298 struct vnode *nvp;
299 u_int hash;
300 struct nfsfh *nfhp;
301 int error;
302
303 *npp = NULL;
304 /* For forced dismounts, just return error. */
305 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
306 return (EINTR);
307 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
308 M_NFSFH, M_WAITOK);
309 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
310 nfhp->nfh_len = fhsize;
311
312 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
313
314 /*
315 * First, try to get the vnode locked, but don't block for the lock.
316 */
317 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
318 newnfs_vncmpf, nfhp);
319 if (error == 0 && nvp != NULL) {
320 NFSVOPUNLOCK(nvp, 0);
321 } else if (error == EBUSY) {
322 /*
323 * The LK_EXCLOTHER lock type tells nfs_lock1() to not try
324 * and lock the vnode, but just get a v_usecount on it.
325 * LK_NOWAIT is set so that when vget() returns ENOENT,
326 * vfs_hash_get() fails instead of looping.
327 * If this succeeds, it is safe so long as a vflush() with
328 * FORCECLOSE has not been done. Since the Renew thread is
329 * stopped and the MNTK_UNMOUNTF flag is set before doing
330 * a vflush() with FORCECLOSE, we should be ok here.
331 */
332 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
333 error = EINTR;
334 else
335 error = vfs_hash_get(mntp, hash,
336 (LK_EXCLOTHER | LK_NOWAIT), td, &nvp,
337 newnfs_vncmpf, nfhp);
338 }
339 FREE(nfhp, M_NFSFH);
340 if (error)
341 return (error);
342 if (nvp != NULL) {
343 *npp = VTONFS(nvp);
344 return (0);
345 }
346 return (EINVAL);
347 }
348
349 /*
350 * Load the attribute cache (that lives in the nfsnode entry) with
351 * the attributes of the second argument and
352 * Iff vaper not NULL
353 * copy the attributes to *vaper
354 * Similar to nfs_loadattrcache(), except the attributes are passed in
355 * instead of being parsed out of the mbuf list.
356 */
357 int
nfscl_loadattrcache(struct vnode ** vpp,struct nfsvattr * nap,void * nvaper,void * stuff,int writeattr,int dontshrink)358 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
359 void *stuff, int writeattr, int dontshrink)
360 {
361 struct vnode *vp = *vpp;
362 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
363 struct nfsnode *np;
364 struct nfsmount *nmp;
365 struct timespec mtime_save;
366 u_quad_t nsize;
367 int setnsize;
368
369 /*
370 * If v_type == VNON it is a new node, so fill in the v_type,
371 * n_mtime fields. Check to see if it represents a special
372 * device, and if so, check for a possible alias. Once the
373 * correct vnode has been obtained, fill in the rest of the
374 * information.
375 */
376 np = VTONFS(vp);
377 NFSLOCKNODE(np);
378 if (vp->v_type != nvap->va_type) {
379 vp->v_type = nvap->va_type;
380 if (vp->v_type == VFIFO)
381 vp->v_op = &newnfs_fifoops;
382 np->n_mtime = nvap->va_mtime;
383 }
384 nmp = VFSTONFS(vp->v_mount);
385 vap = &np->n_vattr.na_vattr;
386 mtime_save = vap->va_mtime;
387 if (writeattr) {
388 np->n_vattr.na_filerev = nap->na_filerev;
389 np->n_vattr.na_size = nap->na_size;
390 np->n_vattr.na_mtime = nap->na_mtime;
391 np->n_vattr.na_ctime = nap->na_ctime;
392 np->n_vattr.na_fsid = nap->na_fsid;
393 np->n_vattr.na_mode = nap->na_mode;
394 } else {
395 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
396 sizeof (struct nfsvattr));
397 }
398
399 /*
400 * For NFSv4, if the node's fsid is not equal to the mount point's
401 * fsid, return the low order 32bits of the node's fsid. This
402 * allows getcwd(3) to work. There is a chance that the fsid might
403 * be the same as a local fs, but since this is in an NFS mount
404 * point, I don't think that will cause any problems?
405 */
406 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
407 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
408 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
409 /*
410 * va_fsid needs to be set to some value derived from
411 * np->n_vattr.na_filesid that is not equal
412 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
413 * from the value used for the top level server volume
414 * in the mounted subtree.
415 */
416 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
417 (uint32_t)np->n_vattr.na_filesid[0])
418 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
419 else
420 vap->va_fsid = (uint32_t)hash32_buf(
421 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
422 } else
423 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
424 np->n_attrstamp = time_second;
425 setnsize = 0;
426 nsize = 0;
427 if (vap->va_size != np->n_size) {
428 if (vap->va_type == VREG) {
429 if (dontshrink && vap->va_size < np->n_size) {
430 /*
431 * We've been told not to shrink the file;
432 * zero np->n_attrstamp to indicate that
433 * the attributes are stale.
434 */
435 vap->va_size = np->n_size;
436 np->n_attrstamp = 0;
437 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
438 vnode_pager_setsize(vp, np->n_size);
439 } else if (np->n_flag & NMODIFIED) {
440 /*
441 * We've modified the file: Use the larger
442 * of our size, and the server's size.
443 */
444 if (vap->va_size < np->n_size) {
445 vap->va_size = np->n_size;
446 } else {
447 np->n_size = vap->va_size;
448 np->n_flag |= NSIZECHANGED;
449 }
450 vnode_pager_setsize(vp, np->n_size);
451 } else if (vap->va_size < np->n_size) {
452 /*
453 * When shrinking the size, the call to
454 * vnode_pager_setsize() cannot be done
455 * with the mutex held, so delay it until
456 * after the mtx_unlock call.
457 */
458 nsize = np->n_size = vap->va_size;
459 np->n_flag |= NSIZECHANGED;
460 setnsize = 1;
461 } else {
462 np->n_size = vap->va_size;
463 np->n_flag |= NSIZECHANGED;
464 vnode_pager_setsize(vp, np->n_size);
465 }
466 } else {
467 np->n_size = vap->va_size;
468 }
469 }
470 /*
471 * The following checks are added to prevent a race between (say)
472 * a READDIR+ and a WRITE.
473 * READDIR+, WRITE requests sent out.
474 * READDIR+ resp, WRITE resp received on client.
475 * However, the WRITE resp was handled before the READDIR+ resp
476 * causing the post op attrs from the write to be loaded first
477 * and the attrs from the READDIR+ to be loaded later. If this
478 * happens, we have stale attrs loaded into the attrcache.
479 * We detect this by for the mtime moving back. We invalidate the
480 * attrcache when this happens.
481 */
482 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
483 /* Size changed or mtime went backwards */
484 np->n_attrstamp = 0;
485 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
486 }
487 if (vaper != NULL) {
488 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
489 if (np->n_flag & NCHG) {
490 if (np->n_flag & NACC)
491 vaper->va_atime = np->n_atim;
492 if (np->n_flag & NUPD)
493 vaper->va_mtime = np->n_mtim;
494 }
495 }
496 #ifdef KDTRACE_HOOKS
497 if (np->n_attrstamp != 0)
498 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, 0);
499 #endif
500 NFSUNLOCKNODE(np);
501 if (setnsize)
502 vnode_pager_setsize(vp, nsize);
503 return (0);
504 }
505
506 /*
507 * Fill in the client id name. For these bytes:
508 * 1 - they must be unique
509 * 2 - they should be persistent across client reboots
510 * 1 is more critical than 2
511 * Use the mount point's unique id plus either the uuid or, if that
512 * isn't set, random junk.
513 */
514 void
nfscl_fillclid(u_int64_t clval,char * uuid,u_int8_t * cp,u_int16_t idlen)515 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
516 {
517 int uuidlen;
518
519 /*
520 * First, put in the 64bit mount point identifier.
521 */
522 if (idlen >= sizeof (u_int64_t)) {
523 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
524 cp += sizeof (u_int64_t);
525 idlen -= sizeof (u_int64_t);
526 }
527
528 /*
529 * If uuid is non-zero length, use it.
530 */
531 uuidlen = strlen(uuid);
532 if (uuidlen > 0 && idlen >= uuidlen) {
533 NFSBCOPY(uuid, cp, uuidlen);
534 cp += uuidlen;
535 idlen -= uuidlen;
536 }
537
538 /*
539 * This only normally happens if the uuid isn't set.
540 */
541 while (idlen > 0) {
542 *cp++ = (u_int8_t)(arc4random() % 256);
543 idlen--;
544 }
545 }
546
547 /*
548 * Fill in a lock owner name. For now, pid + the process's creation time.
549 */
550 void
nfscl_filllockowner(void * id,u_int8_t * cp,int flags)551 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
552 {
553 union {
554 u_int32_t lval;
555 u_int8_t cval[4];
556 } tl;
557 struct proc *p;
558
559 if (id == NULL) {
560 printf("NULL id\n");
561 bzero(cp, NFSV4CL_LOCKNAMELEN);
562 return;
563 }
564 if ((flags & F_POSIX) != 0) {
565 p = (struct proc *)id;
566 tl.lval = p->p_pid;
567 *cp++ = tl.cval[0];
568 *cp++ = tl.cval[1];
569 *cp++ = tl.cval[2];
570 *cp++ = tl.cval[3];
571 tl.lval = p->p_stats->p_start.tv_sec;
572 *cp++ = tl.cval[0];
573 *cp++ = tl.cval[1];
574 *cp++ = tl.cval[2];
575 *cp++ = tl.cval[3];
576 tl.lval = p->p_stats->p_start.tv_usec;
577 *cp++ = tl.cval[0];
578 *cp++ = tl.cval[1];
579 *cp++ = tl.cval[2];
580 *cp = tl.cval[3];
581 } else if ((flags & F_FLOCK) != 0) {
582 bcopy(&id, cp, sizeof(id));
583 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
584 } else {
585 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
586 bzero(cp, NFSV4CL_LOCKNAMELEN);
587 }
588 }
589
590 /*
591 * Find the parent process for the thread passed in as an argument.
592 * If none exists, return NULL, otherwise return a thread for the parent.
593 * (Can be any of the threads, since it is only used for td->td_proc.)
594 */
595 NFSPROC_T *
nfscl_getparent(struct thread * td)596 nfscl_getparent(struct thread *td)
597 {
598 struct proc *p;
599 struct thread *ptd;
600
601 if (td == NULL)
602 return (NULL);
603 p = td->td_proc;
604 if (p->p_pid == 0)
605 return (NULL);
606 p = p->p_pptr;
607 if (p == NULL)
608 return (NULL);
609 ptd = TAILQ_FIRST(&p->p_threads);
610 return (ptd);
611 }
612
613 /*
614 * Start up the renew kernel thread.
615 */
616 static void
start_nfscl(void * arg)617 start_nfscl(void *arg)
618 {
619 struct nfsclclient *clp;
620 struct thread *td;
621
622 clp = (struct nfsclclient *)arg;
623 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
624 nfscl_renewthread(clp, td);
625 kproc_exit(0);
626 }
627
628 void
nfscl_start_renewthread(struct nfsclclient * clp)629 nfscl_start_renewthread(struct nfsclclient *clp)
630 {
631
632 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
633 "nfscl");
634 }
635
636 /*
637 * Handle wcc_data.
638 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
639 * as the first Op after PutFH.
640 * (For NFSv4, the postop attributes are after the Op, so they can't be
641 * parsed here. A separate call to nfscl_postop_attr() is required.)
642 */
643 int
nfscl_wcc_data(struct nfsrv_descript * nd,struct vnode * vp,struct nfsvattr * nap,int * flagp,int * wccflagp,void * stuff)644 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
645 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
646 {
647 u_int32_t *tl;
648 struct nfsnode *np = VTONFS(vp);
649 struct nfsvattr nfsva;
650 int error = 0;
651
652 if (wccflagp != NULL)
653 *wccflagp = 0;
654 if (nd->nd_flag & ND_NFSV3) {
655 *flagp = 0;
656 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
657 if (*tl == newnfs_true) {
658 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
659 if (wccflagp != NULL) {
660 mtx_lock(&np->n_mtx);
661 *wccflagp = (np->n_mtime.tv_sec ==
662 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
663 np->n_mtime.tv_nsec ==
664 fxdr_unsigned(u_int32_t, *(tl + 3)));
665 mtx_unlock(&np->n_mtx);
666 }
667 }
668 error = nfscl_postop_attr(nd, nap, flagp, stuff);
669 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
670 == (ND_NFSV4 | ND_V4WCCATTR)) {
671 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
672 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
673 NULL, NULL, NULL, NULL, NULL);
674 if (error)
675 return (error);
676 /*
677 * Get rid of Op# and status for next op.
678 */
679 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
680 if (*++tl)
681 nd->nd_flag |= ND_NOMOREDATA;
682 if (wccflagp != NULL &&
683 nfsva.na_vattr.va_mtime.tv_sec != 0) {
684 mtx_lock(&np->n_mtx);
685 *wccflagp = (np->n_mtime.tv_sec ==
686 nfsva.na_vattr.va_mtime.tv_sec &&
687 np->n_mtime.tv_nsec ==
688 nfsva.na_vattr.va_mtime.tv_sec);
689 mtx_unlock(&np->n_mtx);
690 }
691 }
692 nfsmout:
693 return (error);
694 }
695
696 /*
697 * Get postop attributes.
698 */
699 int
nfscl_postop_attr(struct nfsrv_descript * nd,struct nfsvattr * nap,int * retp,void * stuff)700 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
701 void *stuff)
702 {
703 u_int32_t *tl;
704 int error = 0;
705
706 *retp = 0;
707 if (nd->nd_flag & ND_NOMOREDATA)
708 return (error);
709 if (nd->nd_flag & ND_NFSV3) {
710 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
711 *retp = fxdr_unsigned(int, *tl);
712 } else if (nd->nd_flag & ND_NFSV4) {
713 /*
714 * For NFSv4, the postop attr are at the end, so no point
715 * in looking if nd_repstat != 0.
716 */
717 if (!nd->nd_repstat) {
718 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
719 if (*(tl + 1))
720 /* should never happen since nd_repstat != 0 */
721 nd->nd_flag |= ND_NOMOREDATA;
722 else
723 *retp = 1;
724 }
725 } else if (!nd->nd_repstat) {
726 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
727 *retp = 1;
728 }
729 if (*retp) {
730 error = nfsm_loadattr(nd, nap);
731 if (error)
732 *retp = 0;
733 }
734 nfsmout:
735 return (error);
736 }
737
738 /*
739 * Fill in the setable attributes. The full argument indicates whether
740 * to fill in them all or just mode and time.
741 */
742 void
nfscl_fillsattr(struct nfsrv_descript * nd,struct vattr * vap,struct vnode * vp,int flags,u_int32_t rdev)743 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
744 struct vnode *vp, int flags, u_int32_t rdev)
745 {
746 u_int32_t *tl;
747 struct nfsv2_sattr *sp;
748 nfsattrbit_t attrbits;
749
750 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
751 case ND_NFSV2:
752 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
753 if (vap->va_mode == (mode_t)VNOVAL)
754 sp->sa_mode = newnfs_xdrneg1;
755 else
756 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
757 if (vap->va_uid == (uid_t)VNOVAL)
758 sp->sa_uid = newnfs_xdrneg1;
759 else
760 sp->sa_uid = txdr_unsigned(vap->va_uid);
761 if (vap->va_gid == (gid_t)VNOVAL)
762 sp->sa_gid = newnfs_xdrneg1;
763 else
764 sp->sa_gid = txdr_unsigned(vap->va_gid);
765 if (flags & NFSSATTR_SIZE0)
766 sp->sa_size = 0;
767 else if (flags & NFSSATTR_SIZENEG1)
768 sp->sa_size = newnfs_xdrneg1;
769 else if (flags & NFSSATTR_SIZERDEV)
770 sp->sa_size = txdr_unsigned(rdev);
771 else
772 sp->sa_size = txdr_unsigned(vap->va_size);
773 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
774 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
775 break;
776 case ND_NFSV3:
777 if (vap->va_mode != (mode_t)VNOVAL) {
778 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
779 *tl++ = newnfs_true;
780 *tl = txdr_unsigned(vap->va_mode);
781 } else {
782 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
783 *tl = newnfs_false;
784 }
785 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
786 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
787 *tl++ = newnfs_true;
788 *tl = txdr_unsigned(vap->va_uid);
789 } else {
790 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
791 *tl = newnfs_false;
792 }
793 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
794 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
795 *tl++ = newnfs_true;
796 *tl = txdr_unsigned(vap->va_gid);
797 } else {
798 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
799 *tl = newnfs_false;
800 }
801 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
802 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
803 *tl++ = newnfs_true;
804 txdr_hyper(vap->va_size, tl);
805 } else {
806 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
807 *tl = newnfs_false;
808 }
809 if (vap->va_atime.tv_sec != VNOVAL) {
810 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
811 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
812 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
813 txdr_nfsv3time(&vap->va_atime, tl);
814 } else {
815 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
816 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
817 }
818 } else {
819 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
820 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
821 }
822 if (vap->va_mtime.tv_sec != VNOVAL) {
823 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
824 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
825 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
826 txdr_nfsv3time(&vap->va_mtime, tl);
827 } else {
828 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
829 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
830 }
831 } else {
832 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
833 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
834 }
835 break;
836 case ND_NFSV4:
837 NFSZERO_ATTRBIT(&attrbits);
838 if (vap->va_mode != (mode_t)VNOVAL)
839 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
840 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
841 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
842 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
843 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
844 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
845 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
846 if (vap->va_atime.tv_sec != VNOVAL)
847 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
848 if (vap->va_mtime.tv_sec != VNOVAL)
849 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
850 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
851 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
852 break;
853 };
854 }
855
856 /*
857 * nfscl_request() - mostly a wrapper for newnfs_request().
858 */
859 int
nfscl_request(struct nfsrv_descript * nd,struct vnode * vp,NFSPROC_T * p,struct ucred * cred,void * stuff)860 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
861 struct ucred *cred, void *stuff)
862 {
863 int ret, vers;
864 struct nfsmount *nmp;
865
866 nmp = VFSTONFS(vp->v_mount);
867 if (nd->nd_flag & ND_NFSV4)
868 vers = NFS_VER4;
869 else if (nd->nd_flag & ND_NFSV3)
870 vers = NFS_VER3;
871 else
872 vers = NFS_VER2;
873 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
874 NFS_PROG, vers, NULL, 1, NULL, NULL);
875 return (ret);
876 }
877
878 /*
879 * fill in this bsden's variant of statfs using nfsstatfs.
880 */
881 void
nfscl_loadsbinfo(struct nfsmount * nmp,struct nfsstatfs * sfp,void * statfs)882 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
883 {
884 struct statfs *sbp = (struct statfs *)statfs;
885
886 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
887 sbp->f_bsize = NFS_FABLKSIZE;
888 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
889 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
890 /*
891 * Although sf_abytes is uint64_t and f_bavail is int64_t,
892 * the value after dividing by NFS_FABLKSIZE is small
893 * enough that it will fit in 63bits, so it is ok to
894 * assign it to f_bavail without fear that it will become
895 * negative.
896 */
897 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
898 sbp->f_files = sfp->sf_tfiles;
899 /* Since f_ffree is int64_t, clip it to 63bits. */
900 if (sfp->sf_ffiles > INT64_MAX)
901 sbp->f_ffree = INT64_MAX;
902 else
903 sbp->f_ffree = sfp->sf_ffiles;
904 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
905 /*
906 * The type casts to (int32_t) ensure that this code is
907 * compatible with the old NFS client, in that it will
908 * propagate bit31 to the high order bits. This may or may
909 * not be correct for NFSv2, but since it is a legacy
910 * environment, I'd rather retain backwards compatibility.
911 */
912 sbp->f_bsize = (int32_t)sfp->sf_bsize;
913 sbp->f_blocks = (int32_t)sfp->sf_blocks;
914 sbp->f_bfree = (int32_t)sfp->sf_bfree;
915 sbp->f_bavail = (int32_t)sfp->sf_bavail;
916 sbp->f_files = 0;
917 sbp->f_ffree = 0;
918 }
919 }
920
921 /*
922 * Use the fsinfo stuff to update the mount point.
923 */
924 void
nfscl_loadfsinfo(struct nfsmount * nmp,struct nfsfsinfo * fsp)925 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
926 {
927
928 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
929 fsp->fs_wtpref >= NFS_FABLKSIZE)
930 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
931 ~(NFS_FABLKSIZE - 1);
932 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
933 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
934 if (nmp->nm_wsize == 0)
935 nmp->nm_wsize = fsp->fs_wtmax;
936 }
937 if (nmp->nm_wsize < NFS_FABLKSIZE)
938 nmp->nm_wsize = NFS_FABLKSIZE;
939 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
940 fsp->fs_rtpref >= NFS_FABLKSIZE)
941 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
942 ~(NFS_FABLKSIZE - 1);
943 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
944 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
945 if (nmp->nm_rsize == 0)
946 nmp->nm_rsize = fsp->fs_rtmax;
947 }
948 if (nmp->nm_rsize < NFS_FABLKSIZE)
949 nmp->nm_rsize = NFS_FABLKSIZE;
950 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
951 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
952 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
953 ~(NFS_DIRBLKSIZ - 1);
954 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
955 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
956 if (nmp->nm_readdirsize == 0)
957 nmp->nm_readdirsize = fsp->fs_rtmax;
958 }
959 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
960 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
961 if (fsp->fs_maxfilesize > 0 &&
962 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
963 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
964 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
965 nmp->nm_state |= NFSSTA_GOTFSINFO;
966 }
967
968 /*
969 * Get a pointer to my IP addrress and return it.
970 * Return NULL if you can't find one.
971 */
972 u_int8_t *
nfscl_getmyip(struct nfsmount * nmp,int * isinet6p)973 nfscl_getmyip(struct nfsmount *nmp, int *isinet6p)
974 {
975 struct sockaddr_in sad, *sin;
976 struct rtentry *rt;
977 u_int8_t *retp = NULL;
978 static struct in_addr laddr;
979
980 *isinet6p = 0;
981 /*
982 * Loop up a route for the destination address.
983 */
984 if (nmp->nm_nam->sa_family == AF_INET) {
985 bzero(&sad, sizeof (sad));
986 sin = (struct sockaddr_in *)nmp->nm_nam;
987 sad.sin_family = AF_INET;
988 sad.sin_len = sizeof (struct sockaddr_in);
989 sad.sin_addr.s_addr = sin->sin_addr.s_addr;
990 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
991 rt = rtalloc1_fib((struct sockaddr *)&sad, 0, 0UL,
992 curthread->td_proc->p_fibnum);
993 if (rt != NULL) {
994 if (rt->rt_ifp != NULL &&
995 rt->rt_ifa != NULL &&
996 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
997 rt->rt_ifa->ifa_addr->sa_family == AF_INET) {
998 sin = (struct sockaddr_in *)
999 rt->rt_ifa->ifa_addr;
1000 laddr.s_addr = sin->sin_addr.s_addr;
1001 retp = (u_int8_t *)&laddr;
1002 }
1003 RTFREE_LOCKED(rt);
1004 }
1005 CURVNET_RESTORE();
1006 #ifdef INET6
1007 } else if (nmp->nm_nam->sa_family == AF_INET6) {
1008 struct sockaddr_in6 sad6, *sin6;
1009 static struct in6_addr laddr6;
1010
1011 bzero(&sad6, sizeof (sad6));
1012 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1013 sad6.sin6_family = AF_INET6;
1014 sad6.sin6_len = sizeof (struct sockaddr_in6);
1015 sad6.sin6_addr = sin6->sin6_addr;
1016 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1017 rt = rtalloc1_fib((struct sockaddr *)&sad6, 0, 0UL,
1018 curthread->td_proc->p_fibnum);
1019 if (rt != NULL) {
1020 if (rt->rt_ifp != NULL &&
1021 rt->rt_ifa != NULL &&
1022 ((rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) &&
1023 rt->rt_ifa->ifa_addr->sa_family == AF_INET6) {
1024 sin6 = (struct sockaddr_in6 *)
1025 rt->rt_ifa->ifa_addr;
1026 laddr6 = sin6->sin6_addr;
1027 retp = (u_int8_t *)&laddr6;
1028 *isinet6p = 1;
1029 }
1030 RTFREE_LOCKED(rt);
1031 }
1032 CURVNET_RESTORE();
1033 #endif
1034 }
1035 return (retp);
1036 }
1037
1038 /*
1039 * Copy NFS uid, gids from the cred structure.
1040 */
1041 void
newnfs_copyincred(struct ucred * cr,struct nfscred * nfscr)1042 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1043 {
1044 int i;
1045
1046 KASSERT(cr->cr_ngroups >= 0,
1047 ("newnfs_copyincred: negative cr_ngroups"));
1048 nfscr->nfsc_uid = cr->cr_uid;
1049 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1050 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1051 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1052 }
1053
1054
1055 /*
1056 * Do any client specific initialization.
1057 */
1058 void
nfscl_init(void)1059 nfscl_init(void)
1060 {
1061 static int inited = 0;
1062
1063 if (inited)
1064 return;
1065 inited = 1;
1066 nfscl_inited = 1;
1067 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1068 }
1069
1070 /*
1071 * Check each of the attributes to be set, to ensure they aren't already
1072 * the correct value. Disable setting ones already correct.
1073 */
1074 int
nfscl_checksattr(struct vattr * vap,struct nfsvattr * nvap)1075 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1076 {
1077
1078 if (vap->va_mode != (mode_t)VNOVAL) {
1079 if (vap->va_mode == nvap->na_mode)
1080 vap->va_mode = (mode_t)VNOVAL;
1081 }
1082 if (vap->va_uid != (uid_t)VNOVAL) {
1083 if (vap->va_uid == nvap->na_uid)
1084 vap->va_uid = (uid_t)VNOVAL;
1085 }
1086 if (vap->va_gid != (gid_t)VNOVAL) {
1087 if (vap->va_gid == nvap->na_gid)
1088 vap->va_gid = (gid_t)VNOVAL;
1089 }
1090 if (vap->va_size != VNOVAL) {
1091 if (vap->va_size == nvap->na_size)
1092 vap->va_size = VNOVAL;
1093 }
1094
1095 /*
1096 * We are normally called with only a partially initialized
1097 * VAP. Since the NFSv3 spec says that server may use the
1098 * file attributes to store the verifier, the spec requires
1099 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1100 * in atime, but we can't really assume that all servers will
1101 * so we ensure that our SETATTR sets both atime and mtime.
1102 */
1103 if (vap->va_mtime.tv_sec == VNOVAL)
1104 vfs_timestamp(&vap->va_mtime);
1105 if (vap->va_atime.tv_sec == VNOVAL)
1106 vap->va_atime = vap->va_mtime;
1107 return (1);
1108 }
1109
1110 /*
1111 * Map nfsv4 errors to errno.h errors.
1112 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1113 * error should only be returned for the Open, Create and Setattr Ops.
1114 * As such, most calls can just pass in 0 for those arguments.
1115 */
1116 APPLESTATIC int
nfscl_maperr(struct thread * td,int error,uid_t uid,gid_t gid)1117 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1118 {
1119 struct proc *p;
1120
1121 if (error < 10000)
1122 return (error);
1123 if (td != NULL)
1124 p = td->td_proc;
1125 else
1126 p = NULL;
1127 switch (error) {
1128 case NFSERR_BADOWNER:
1129 tprintf(p, LOG_INFO,
1130 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1131 uid, gid);
1132 return (EPERM);
1133 case NFSERR_BADNAME:
1134 case NFSERR_BADCHAR:
1135 printf("nfsv4 char/name not handled by server\n");
1136 return (ENOENT);
1137 case NFSERR_STALECLIENTID:
1138 case NFSERR_STALESTATEID:
1139 case NFSERR_EXPIRED:
1140 case NFSERR_BADSTATEID:
1141 case NFSERR_BADSESSION:
1142 printf("nfsv4 recover err returned %d\n", error);
1143 return (EIO);
1144 case NFSERR_BADHANDLE:
1145 case NFSERR_SERVERFAULT:
1146 case NFSERR_BADTYPE:
1147 case NFSERR_FHEXPIRED:
1148 case NFSERR_RESOURCE:
1149 case NFSERR_MOVED:
1150 case NFSERR_NOFILEHANDLE:
1151 case NFSERR_MINORVERMISMATCH:
1152 case NFSERR_OLDSTATEID:
1153 case NFSERR_BADSEQID:
1154 case NFSERR_LEASEMOVED:
1155 case NFSERR_RECLAIMBAD:
1156 case NFSERR_BADXDR:
1157 case NFSERR_OPILLEGAL:
1158 printf("nfsv4 client/server protocol prob err=%d\n",
1159 error);
1160 return (EIO);
1161 default:
1162 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1163 return (EIO);
1164 };
1165 }
1166
1167 /*
1168 * Check to see if the process for this owner exists. Return 1 if it doesn't
1169 * and 0 otherwise.
1170 */
1171 int
nfscl_procdoesntexist(u_int8_t * own)1172 nfscl_procdoesntexist(u_int8_t *own)
1173 {
1174 union {
1175 u_int32_t lval;
1176 u_int8_t cval[4];
1177 } tl;
1178 struct proc *p;
1179 pid_t pid;
1180 int ret = 0;
1181
1182 tl.cval[0] = *own++;
1183 tl.cval[1] = *own++;
1184 tl.cval[2] = *own++;
1185 tl.cval[3] = *own++;
1186 pid = tl.lval;
1187 p = pfind_locked(pid);
1188 if (p == NULL)
1189 return (1);
1190 if (p->p_stats == NULL) {
1191 PROC_UNLOCK(p);
1192 return (0);
1193 }
1194 tl.cval[0] = *own++;
1195 tl.cval[1] = *own++;
1196 tl.cval[2] = *own++;
1197 tl.cval[3] = *own++;
1198 if (tl.lval != p->p_stats->p_start.tv_sec) {
1199 ret = 1;
1200 } else {
1201 tl.cval[0] = *own++;
1202 tl.cval[1] = *own++;
1203 tl.cval[2] = *own++;
1204 tl.cval[3] = *own;
1205 if (tl.lval != p->p_stats->p_start.tv_usec)
1206 ret = 1;
1207 }
1208 PROC_UNLOCK(p);
1209 return (ret);
1210 }
1211
1212 /*
1213 * - nfs pseudo system call for the client
1214 */
1215 /*
1216 * MPSAFE
1217 */
1218 static int
nfssvc_nfscl(struct thread * td,struct nfssvc_args * uap)1219 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1220 {
1221 struct file *fp;
1222 struct nfscbd_args nfscbdarg;
1223 struct nfsd_nfscbd_args nfscbdarg2;
1224 struct nameidata nd;
1225 struct nfscl_dumpmntopts dumpmntopts;
1226 cap_rights_t rights;
1227 char *buf;
1228 int error;
1229
1230 if (uap->flag & NFSSVC_CBADDSOCK) {
1231 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1232 if (error)
1233 return (error);
1234 /*
1235 * Since we don't know what rights might be required,
1236 * pretend that we need them all. It is better to be too
1237 * careful than too reckless.
1238 */
1239 error = fget(td, nfscbdarg.sock,
1240 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1241 if (error)
1242 return (error);
1243 if (fp->f_type != DTYPE_SOCKET) {
1244 fdrop(fp, td);
1245 return (EPERM);
1246 }
1247 error = nfscbd_addsock(fp);
1248 fdrop(fp, td);
1249 if (!error && nfscl_enablecallb == 0) {
1250 nfsv4_cbport = nfscbdarg.port;
1251 nfscl_enablecallb = 1;
1252 }
1253 } else if (uap->flag & NFSSVC_NFSCBD) {
1254 if (uap->argp == NULL)
1255 return (EINVAL);
1256 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1257 sizeof(nfscbdarg2));
1258 if (error)
1259 return (error);
1260 error = nfscbd_nfsd(td, &nfscbdarg2);
1261 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1262 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1263 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1264 dumpmntopts.ndmnt_blen > 1024))
1265 error = EINVAL;
1266 if (error == 0)
1267 error = nfsrv_lookupfilename(&nd,
1268 dumpmntopts.ndmnt_fname, td);
1269 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1270 "nfs") != 0) {
1271 vput(nd.ni_vp);
1272 error = EINVAL;
1273 }
1274 if (error == 0) {
1275 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1276 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1277 dumpmntopts.ndmnt_blen);
1278 vput(nd.ni_vp);
1279 error = copyout(buf, dumpmntopts.ndmnt_buf,
1280 dumpmntopts.ndmnt_blen);
1281 free(buf, M_TEMP);
1282 }
1283 } else {
1284 error = EINVAL;
1285 }
1286 return (error);
1287 }
1288
1289 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1290
1291 /*
1292 * Called once to initialize data structures...
1293 */
1294 static int
nfscl_modevent(module_t mod,int type,void * data)1295 nfscl_modevent(module_t mod, int type, void *data)
1296 {
1297 int error = 0;
1298 static int loaded = 0;
1299
1300 switch (type) {
1301 case MOD_LOAD:
1302 if (loaded)
1303 return (0);
1304 newnfs_portinit();
1305 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1306 MTX_DEF);
1307 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1308 nfscl_init();
1309 NFSD_LOCK();
1310 nfsrvd_cbinit(0);
1311 NFSD_UNLOCK();
1312 ncl_call_invalcaches = ncl_invalcaches;
1313 nfsd_call_nfscl = nfssvc_nfscl;
1314 loaded = 1;
1315 break;
1316
1317 case MOD_UNLOAD:
1318 if (nfs_numnfscbd != 0) {
1319 error = EBUSY;
1320 break;
1321 }
1322
1323 /*
1324 * XXX: Unloading of nfscl module is unsupported.
1325 */
1326 #if 0
1327 ncl_call_invalcaches = NULL;
1328 nfsd_call_nfscl = NULL;
1329 /* and get rid of the mutexes */
1330 mtx_destroy(&nfs_clstate_mutex);
1331 mtx_destroy(&ncl_iod_mutex);
1332 loaded = 0;
1333 break;
1334 #else
1335 /* FALLTHROUGH */
1336 #endif
1337 default:
1338 error = EOPNOTSUPP;
1339 break;
1340 }
1341 return error;
1342 }
1343 static moduledata_t nfscl_mod = {
1344 "nfscl",
1345 nfscl_modevent,
1346 NULL,
1347 };
1348 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1349
1350 /* So that loader and kldload(2) can find us, wherever we are.. */
1351 MODULE_VERSION(nfscl, 1);
1352 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1353 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1354 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1355 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1356
1357