1 /* $NetBSD: nfs_clport.c,v 1.2 2016/12/13 22:17:33 pgoyette 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 299413 2016-05-11 06:35:46Z kib "); */
37 __RCSID("$NetBSD: nfs_clport.c,v 1.2 2016/12/13 22:17:33 pgoyette Exp $");
38
39 #ifdef _KERNEL_OPT
40 #include "opt_inet.h"
41 #include "opt_inet6.h"
42 #endif
43
44 #include <sys/capsicum.h>
45
46 /*
47 * generally, I don't like #includes inside .h files, but it seems to
48 * be the easiest way to handle the port.
49 */
50 #include <sys/fail.h>
51
52 #include <sys/hash.h>
53 #include <sys/sysctl.h>
54 #include <fs/nfs/common/nfsport.h>
55
56 #include <netinet/in_fib.h>
57 #include <netinet/if_ether.h>
58 #include <netinet6/ip6_var.h>
59 #include <net/if_types.h>
60
61 #include <fs/nfs/client/nfs_kdtrace.h>
62
63 #ifdef KDTRACE_HOOKS
64 dtrace_nfsclient_attrcache_flush_probe_func_t
65 dtrace_nfscl_attrcache_flush_done_probe;
66 uint32_t nfscl_attrcache_flush_done_id;
67
68 dtrace_nfsclient_attrcache_get_hit_probe_func_t
69 dtrace_nfscl_attrcache_get_hit_probe;
70 uint32_t nfscl_attrcache_get_hit_id;
71
72 dtrace_nfsclient_attrcache_get_miss_probe_func_t
73 dtrace_nfscl_attrcache_get_miss_probe;
74 uint32_t nfscl_attrcache_get_miss_id;
75
76 dtrace_nfsclient_attrcache_load_probe_func_t
77 dtrace_nfscl_attrcache_load_done_probe;
78 uint32_t nfscl_attrcache_load_done_id;
79 #endif /* !KDTRACE_HOOKS */
80
81 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
82 extern struct vop_vector newnfs_vnodeops;
83 extern struct vop_vector newnfs_fifoops;
84 extern uma_zone_t newnfsnode_zone;
85 extern struct buf_ops buf_ops_newnfs;
86 extern int ncl_pbuf_freecnt;
87 extern short nfsv4_cbport;
88 extern int nfscl_enablecallb;
89 extern int nfs_numnfscbd;
90 extern int nfscl_inited;
91 struct mtx nfs_clstate_mutex;
92 struct mtx ncl_iod_mutex;
93 NFSDLOCKMUTEX;
94
95 extern void (*ncl_call_invalcaches)(struct vnode *);
96
97 SYSCTL_DECL(_vfs_nfs);
98 static int ncl_fileid_maxwarnings = 10;
99 SYSCTL_INT(_vfs_nfs, OID_AUTO, fileid_maxwarnings, CTLFLAG_RWTUN,
100 &ncl_fileid_maxwarnings, 0,
101 "Limit fileid corruption warnings; 0 is off; -1 is unlimited");
102 static volatile int ncl_fileid_nwarnings;
103
104 static void nfscl_warn_fileid(struct nfsmount *, struct nfsvattr *,
105 struct nfsvattr *);
106
107 /*
108 * Comparison function for vfs_hash functions.
109 */
110 int
newnfs_vncmpf(struct vnode * vp,void * arg)111 newnfs_vncmpf(struct vnode *vp, void *arg)
112 {
113 struct nfsfh *nfhp = (struct nfsfh *)arg;
114 struct nfsnode *np = VTONFS(vp);
115
116 if (np->n_fhp->nfh_len != nfhp->nfh_len ||
117 NFSBCMP(np->n_fhp->nfh_fh, nfhp->nfh_fh, nfhp->nfh_len))
118 return (1);
119 return (0);
120 }
121
122 /*
123 * Look up a vnode/nfsnode by file handle.
124 * Callers must check for mount points!!
125 * In all cases, a pointer to a
126 * nfsnode structure is returned.
127 * This variant takes a "struct nfsfh *" as second argument and uses
128 * that structure up, either by hanging off the nfsnode or FREEing it.
129 */
130 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)131 nfscl_nget(struct mount *mntp, struct vnode *dvp, struct nfsfh *nfhp,
132 struct componentname *cnp, struct thread *td, struct nfsnode **npp,
133 void *stuff, int lkflags)
134 {
135 struct nfsnode *np, *dnp;
136 struct vnode *vp, *nvp;
137 struct nfsv4node *newd, *oldd;
138 int error;
139 u_int hash;
140 struct nfsmount *nmp;
141
142 nmp = VFSTONFS(mntp);
143 dnp = VTONFS(dvp);
144 *npp = NULL;
145
146 hash = fnv_32_buf(nfhp->nfh_fh, nfhp->nfh_len, FNV1_32_INIT);
147
148 error = vfs_hash_get(mntp, hash, lkflags,
149 td, &nvp, newnfs_vncmpf, nfhp);
150 if (error == 0 && nvp != NULL) {
151 /*
152 * I believe there is a slight chance that vgonel() could
153 * get called on this vnode between when NFSVOPLOCK() drops
154 * the VI_LOCK() and vget() acquires it again, so that it
155 * hasn't yet had v_usecount incremented. If this were to
156 * happen, the VI_DOOMED flag would be set, so check for
157 * that here. Since we now have the v_usecount incremented,
158 * we should be ok until we vrele() it, if the VI_DOOMED
159 * flag isn't set now.
160 */
161 VI_LOCK(nvp);
162 if ((nvp->v_iflag & VI_DOOMED)) {
163 VI_UNLOCK(nvp);
164 vrele(nvp);
165 error = ENOENT;
166 } else {
167 VI_UNLOCK(nvp);
168 }
169 }
170 if (error) {
171 FREE((caddr_t)nfhp, M_NFSFH);
172 return (error);
173 }
174 if (nvp != NULL) {
175 np = VTONFS(nvp);
176 /*
177 * For NFSv4, check to see if it is the same name and
178 * replace the name, if it is different.
179 */
180 oldd = newd = NULL;
181 if ((nmp->nm_flag & NFSMNT_NFSV4) && np->n_v4 != NULL &&
182 nvp->v_type == VREG &&
183 (np->n_v4->n4_namelen != cnp->cn_namelen ||
184 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
185 cnp->cn_namelen) ||
186 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
187 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
188 dnp->n_fhp->nfh_len))) {
189 MALLOC(newd, struct nfsv4node *,
190 sizeof (struct nfsv4node) + dnp->n_fhp->nfh_len +
191 + cnp->cn_namelen - 1, M_NFSV4NODE, M_WAITOK);
192 NFSLOCKNODE(np);
193 if (newd != NULL && np->n_v4 != NULL && nvp->v_type == VREG
194 && (np->n_v4->n4_namelen != cnp->cn_namelen ||
195 NFSBCMP(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
196 cnp->cn_namelen) ||
197 dnp->n_fhp->nfh_len != np->n_v4->n4_fhlen ||
198 NFSBCMP(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
199 dnp->n_fhp->nfh_len))) {
200 oldd = np->n_v4;
201 np->n_v4 = newd;
202 newd = NULL;
203 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
204 np->n_v4->n4_namelen = cnp->cn_namelen;
205 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
206 dnp->n_fhp->nfh_len);
207 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
208 cnp->cn_namelen);
209 }
210 NFSUNLOCKNODE(np);
211 }
212 if (newd != NULL)
213 FREE((caddr_t)newd, M_NFSV4NODE);
214 if (oldd != NULL)
215 FREE((caddr_t)oldd, M_NFSV4NODE);
216 *npp = np;
217 FREE((caddr_t)nfhp, M_NFSFH);
218 return (0);
219 }
220 np = uma_zalloc(newnfsnode_zone, M_WAITOK | M_ZERO);
221
222 error = getnewvnode(nfs_vnode_tag, mntp, &newnfs_vnodeops, &nvp);
223 if (error) {
224 uma_zfree(newnfsnode_zone, np);
225 FREE((caddr_t)nfhp, M_NFSFH);
226 return (error);
227 }
228 vp = nvp;
229 KASSERT(vp->v_bufobj.bo_bsize != 0, ("nfscl_nget: bo_bsize == 0"));
230 vp->v_bufobj.bo_ops = &buf_ops_newnfs;
231 vp->v_data = np;
232 np->n_vnode = vp;
233 /*
234 * Initialize the mutex even if the vnode is going to be a loser.
235 * This simplifies the logic in reclaim, which can then unconditionally
236 * destroy the mutex (in the case of the loser, or if hash_insert
237 * happened to return an error no special casing is needed).
238 */
239 mtx_init(&np->n_mtx, "NEWNFSnode lock", NULL, MTX_DEF | MTX_DUPOK);
240
241 /*
242 * Are we getting the root? If so, make sure the vnode flags
243 * are correct
244 */
245 if ((nfhp->nfh_len == nmp->nm_fhsize) &&
246 !bcmp(nfhp->nfh_fh, nmp->nm_fh, nfhp->nfh_len)) {
247 if (vp->v_type == VNON)
248 vp->v_type = VDIR;
249 vp->v_vflag |= VV_ROOT;
250 }
251
252 np->n_fhp = nfhp;
253 /*
254 * For NFSv4, we have to attach the directory file handle and
255 * file name, so that Open Ops can be done later.
256 */
257 if (nmp->nm_flag & NFSMNT_NFSV4) {
258 MALLOC(np->n_v4, struct nfsv4node *, sizeof (struct nfsv4node)
259 + dnp->n_fhp->nfh_len + cnp->cn_namelen - 1, M_NFSV4NODE,
260 M_WAITOK);
261 np->n_v4->n4_fhlen = dnp->n_fhp->nfh_len;
262 np->n_v4->n4_namelen = cnp->cn_namelen;
263 NFSBCOPY(dnp->n_fhp->nfh_fh, np->n_v4->n4_data,
264 dnp->n_fhp->nfh_len);
265 NFSBCOPY(cnp->cn_nameptr, NFS4NODENAME(np->n_v4),
266 cnp->cn_namelen);
267 } else {
268 np->n_v4 = NULL;
269 }
270
271 /*
272 * NFS supports recursive and shared locking.
273 */
274 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_NOWITNESS, NULL);
275 VN_LOCK_AREC(vp);
276 VN_LOCK_ASHARE(vp);
277 error = insmntque(vp, mntp);
278 if (error != 0) {
279 *npp = NULL;
280 mtx_destroy(&np->n_mtx);
281 FREE((caddr_t)nfhp, M_NFSFH);
282 if (np->n_v4 != NULL)
283 FREE((caddr_t)np->n_v4, M_NFSV4NODE);
284 uma_zfree(newnfsnode_zone, np);
285 return (error);
286 }
287 error = vfs_hash_insert(vp, hash, lkflags,
288 td, &nvp, newnfs_vncmpf, nfhp);
289 if (error)
290 return (error);
291 if (nvp != NULL) {
292 *npp = VTONFS(nvp);
293 /* vfs_hash_insert() vput()'s the losing vnode */
294 return (0);
295 }
296 *npp = np;
297
298 return (0);
299 }
300
301 /*
302 * Another variant of nfs_nget(). This one is only used by reopen. It
303 * takes almost the same args as nfs_nget(), but only succeeds if an entry
304 * exists in the cache. (Since files should already be "open" with a
305 * vnode ref cnt on the node when reopen calls this, it should always
306 * succeed.)
307 * Also, don't get a vnode lock, since it may already be locked by some
308 * other process that is handling it. This is ok, since all other threads
309 * on the client are blocked by the nfsc_lock being exclusively held by the
310 * caller of this function.
311 */
312 int
nfscl_ngetreopen(struct mount * mntp,u_int8_t * fhp,int fhsize,struct thread * td,struct nfsnode ** npp)313 nfscl_ngetreopen(struct mount *mntp, u_int8_t *fhp, int fhsize,
314 struct thread *td, struct nfsnode **npp)
315 {
316 struct vnode *nvp;
317 u_int hash;
318 struct nfsfh *nfhp;
319 int error;
320
321 *npp = NULL;
322 /* For forced dismounts, just return error. */
323 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
324 return (EINTR);
325 MALLOC(nfhp, struct nfsfh *, sizeof (struct nfsfh) + fhsize,
326 M_NFSFH, M_WAITOK);
327 bcopy(fhp, &nfhp->nfh_fh[0], fhsize);
328 nfhp->nfh_len = fhsize;
329
330 hash = fnv_32_buf(fhp, fhsize, FNV1_32_INIT);
331
332 /*
333 * First, try to get the vnode locked, but don't block for the lock.
334 */
335 error = vfs_hash_get(mntp, hash, (LK_EXCLUSIVE | LK_NOWAIT), td, &nvp,
336 newnfs_vncmpf, nfhp);
337 if (error == 0 && nvp != NULL) {
338 NFSVOPUNLOCK(nvp, 0);
339 } else if (error == EBUSY) {
340 /*
341 * It is safe so long as a vflush() with
342 * FORCECLOSE has not been done. Since the Renew thread is
343 * stopped and the MNTK_UNMOUNTF flag is set before doing
344 * a vflush() with FORCECLOSE, we should be ok here.
345 */
346 if ((mntp->mnt_kern_flag & MNTK_UNMOUNTF))
347 error = EINTR;
348 else {
349 vfs_hash_ref(mntp, hash, td, &nvp, newnfs_vncmpf, nfhp);
350 if (nvp == NULL) {
351 error = ENOENT;
352 } else if ((nvp->v_iflag & VI_DOOMED) != 0) {
353 error = ENOENT;
354 vrele(nvp);
355 } else {
356 error = 0;
357 }
358 }
359 }
360 FREE(nfhp, M_NFSFH);
361 if (error)
362 return (error);
363 if (nvp != NULL) {
364 *npp = VTONFS(nvp);
365 return (0);
366 }
367 return (EINVAL);
368 }
369
370 static void
nfscl_warn_fileid(struct nfsmount * nmp,struct nfsvattr * oldnap,struct nfsvattr * newnap)371 nfscl_warn_fileid(struct nfsmount *nmp, struct nfsvattr *oldnap,
372 struct nfsvattr *newnap)
373 {
374 int off;
375
376 if (ncl_fileid_maxwarnings >= 0 &&
377 ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
378 return;
379 off = 0;
380 if (ncl_fileid_maxwarnings >= 0) {
381 if (++ncl_fileid_nwarnings >= ncl_fileid_maxwarnings)
382 off = 1;
383 }
384
385 printf("newnfs: server '%s' error: fileid changed. "
386 "fsid %jx:%jx: expected fileid %#jx, got %#jx. "
387 "(BROKEN NFS SERVER OR MIDDLEWARE)\n",
388 nmp->nm_com.nmcom_hostname,
389 (uintmax_t)nmp->nm_fsid[0],
390 (uintmax_t)nmp->nm_fsid[1],
391 (uintmax_t)oldnap->na_fileid,
392 (uintmax_t)newnap->na_fileid);
393
394 if (off)
395 printf("newnfs: Logged %d times about fileid corruption; "
396 "going quiet to avoid spamming logs excessively. (Limit "
397 "is: %d).\n", ncl_fileid_nwarnings,
398 ncl_fileid_maxwarnings);
399 }
400
401 /*
402 * Load the attribute cache (that lives in the nfsnode entry) with
403 * the attributes of the second argument and
404 * Iff vaper not NULL
405 * copy the attributes to *vaper
406 * Similar to nfs_loadattrcache(), except the attributes are passed in
407 * instead of being parsed out of the mbuf list.
408 */
409 int
nfscl_loadattrcache(struct vnode ** vpp,struct nfsvattr * nap,void * nvaper,void * stuff,int writeattr,int dontshrink)410 nfscl_loadattrcache(struct vnode **vpp, struct nfsvattr *nap, void *nvaper,
411 void *stuff, int writeattr, int dontshrink)
412 {
413 struct vnode *vp = *vpp;
414 struct vattr *vap, *nvap = &nap->na_vattr, *vaper = nvaper;
415 struct nfsnode *np;
416 struct nfsmount *nmp;
417 struct timespec mtime_save;
418 u_quad_t nsize;
419 int setnsize, error, force_fid_err;
420
421 error = 0;
422 setnsize = 0;
423 nsize = 0;
424
425 /*
426 * If v_type == VNON it is a new node, so fill in the v_type,
427 * n_mtime fields. Check to see if it represents a special
428 * device, and if so, check for a possible alias. Once the
429 * correct vnode has been obtained, fill in the rest of the
430 * information.
431 */
432 np = VTONFS(vp);
433 NFSLOCKNODE(np);
434 if (vp->v_type != nvap->va_type) {
435 vp->v_type = nvap->va_type;
436 if (vp->v_type == VFIFO)
437 vp->v_op = &newnfs_fifoops;
438 np->n_mtime = nvap->va_mtime;
439 }
440 nmp = VFSTONFS(vp->v_mount);
441 vap = &np->n_vattr.na_vattr;
442 mtime_save = vap->va_mtime;
443 if (writeattr) {
444 np->n_vattr.na_filerev = nap->na_filerev;
445 np->n_vattr.na_size = nap->na_size;
446 np->n_vattr.na_mtime = nap->na_mtime;
447 np->n_vattr.na_ctime = nap->na_ctime;
448 np->n_vattr.na_fsid = nap->na_fsid;
449 np->n_vattr.na_mode = nap->na_mode;
450 } else {
451 force_fid_err = 0;
452 KFAIL_POINT_ERROR(DEBUG_FP, nfscl_force_fileid_warning,
453 force_fid_err);
454 /*
455 * BROKEN NFS SERVER OR MIDDLEWARE
456 *
457 * Certain NFS servers (certain old proprietary filers ca.
458 * 2006) or broken middleboxes (e.g. WAN accelerator products)
459 * will respond to GETATTR requests with results for a
460 * different fileid.
461 *
462 * The WAN accelerator we've observed not only serves stale
463 * cache results for a given file, it also occasionally serves
464 * results for wholly different files. This causes surprising
465 * problems; for example the cached size attribute of a file
466 * may truncate down and then back up, resulting in zero
467 * regions in file contents read by applications. We observed
468 * this reliably with Clang and .c files during parallel build.
469 * A pcap revealed packet fragmentation and GETATTR RPC
470 * responses with wholly wrong fileids.
471 */
472 if ((np->n_vattr.na_fileid != 0 &&
473 np->n_vattr.na_fileid != nap->na_fileid) ||
474 force_fid_err) {
475 nfscl_warn_fileid(nmp, &np->n_vattr, nap);
476 error = EIDRM;
477 goto out;
478 }
479 NFSBCOPY((caddr_t)nap, (caddr_t)&np->n_vattr,
480 sizeof (struct nfsvattr));
481 }
482
483 /*
484 * For NFSv4, if the node's fsid is not equal to the mount point's
485 * fsid, return the low order 32bits of the node's fsid. This
486 * allows getcwd(3) to work. There is a chance that the fsid might
487 * be the same as a local fs, but since this is in an NFS mount
488 * point, I don't think that will cause any problems?
489 */
490 if (NFSHASNFSV4(nmp) && NFSHASHASSETFSID(nmp) &&
491 (nmp->nm_fsid[0] != np->n_vattr.na_filesid[0] ||
492 nmp->nm_fsid[1] != np->n_vattr.na_filesid[1])) {
493 /*
494 * va_fsid needs to be set to some value derived from
495 * np->n_vattr.na_filesid that is not equal
496 * vp->v_mount->mnt_stat.f_fsid[0], so that it changes
497 * from the value used for the top level server volume
498 * in the mounted subtree.
499 */
500 if (vp->v_mount->mnt_stat.f_fsid.val[0] !=
501 (uint32_t)np->n_vattr.na_filesid[0])
502 vap->va_fsid = (uint32_t)np->n_vattr.na_filesid[0];
503 else
504 vap->va_fsid = (uint32_t)hash32_buf(
505 np->n_vattr.na_filesid, 2 * sizeof(uint64_t), 0);
506 } else
507 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
508 np->n_attrstamp = time_second;
509 if (vap->va_size != np->n_size) {
510 if (vap->va_type == VREG) {
511 if (dontshrink && vap->va_size < np->n_size) {
512 /*
513 * We've been told not to shrink the file;
514 * zero np->n_attrstamp to indicate that
515 * the attributes are stale.
516 */
517 vap->va_size = np->n_size;
518 np->n_attrstamp = 0;
519 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
520 vnode_pager_setsize(vp, np->n_size);
521 } else if (np->n_flag & NMODIFIED) {
522 /*
523 * We've modified the file: Use the larger
524 * of our size, and the server's size.
525 */
526 if (vap->va_size < np->n_size) {
527 vap->va_size = np->n_size;
528 } else {
529 np->n_size = vap->va_size;
530 np->n_flag |= NSIZECHANGED;
531 }
532 vnode_pager_setsize(vp, np->n_size);
533 } else if (vap->va_size < np->n_size) {
534 /*
535 * When shrinking the size, the call to
536 * vnode_pager_setsize() cannot be done
537 * with the mutex held, so delay it until
538 * after the mtx_unlock call.
539 */
540 nsize = np->n_size = vap->va_size;
541 np->n_flag |= NSIZECHANGED;
542 setnsize = 1;
543 } else {
544 np->n_size = vap->va_size;
545 np->n_flag |= NSIZECHANGED;
546 vnode_pager_setsize(vp, np->n_size);
547 }
548 } else {
549 np->n_size = vap->va_size;
550 }
551 }
552 /*
553 * The following checks are added to prevent a race between (say)
554 * a READDIR+ and a WRITE.
555 * READDIR+, WRITE requests sent out.
556 * READDIR+ resp, WRITE resp received on client.
557 * However, the WRITE resp was handled before the READDIR+ resp
558 * causing the post op attrs from the write to be loaded first
559 * and the attrs from the READDIR+ to be loaded later. If this
560 * happens, we have stale attrs loaded into the attrcache.
561 * We detect this by for the mtime moving back. We invalidate the
562 * attrcache when this happens.
563 */
564 if (timespeccmp(&mtime_save, &vap->va_mtime, >)) {
565 /* Size changed or mtime went backwards */
566 np->n_attrstamp = 0;
567 KDTRACE_NFS_ATTRCACHE_FLUSH_DONE(vp);
568 }
569 if (vaper != NULL) {
570 NFSBCOPY((caddr_t)vap, (caddr_t)vaper, sizeof(*vap));
571 if (np->n_flag & NCHG) {
572 if (np->n_flag & NACC)
573 vaper->va_atime = np->n_atim;
574 if (np->n_flag & NUPD)
575 vaper->va_mtime = np->n_mtim;
576 }
577 }
578
579 out:
580 #ifdef KDTRACE_HOOKS
581 if (np->n_attrstamp != 0)
582 KDTRACE_NFS_ATTRCACHE_LOAD_DONE(vp, vap, error);
583 #endif
584 NFSUNLOCKNODE(np);
585 if (setnsize)
586 vnode_pager_setsize(vp, nsize);
587 return (error);
588 }
589
590 /*
591 * Fill in the client id name. For these bytes:
592 * 1 - they must be unique
593 * 2 - they should be persistent across client reboots
594 * 1 is more critical than 2
595 * Use the mount point's unique id plus either the uuid or, if that
596 * isn't set, random junk.
597 */
598 void
nfscl_fillclid(u_int64_t clval,char * uuid,u_int8_t * cp,u_int16_t idlen)599 nfscl_fillclid(u_int64_t clval, char *uuid, u_int8_t *cp, u_int16_t idlen)
600 {
601 int uuidlen;
602
603 /*
604 * First, put in the 64bit mount point identifier.
605 */
606 if (idlen >= sizeof (u_int64_t)) {
607 NFSBCOPY((caddr_t)&clval, cp, sizeof (u_int64_t));
608 cp += sizeof (u_int64_t);
609 idlen -= sizeof (u_int64_t);
610 }
611
612 /*
613 * If uuid is non-zero length, use it.
614 */
615 uuidlen = strlen(uuid);
616 if (uuidlen > 0 && idlen >= uuidlen) {
617 NFSBCOPY(uuid, cp, uuidlen);
618 cp += uuidlen;
619 idlen -= uuidlen;
620 }
621
622 /*
623 * This only normally happens if the uuid isn't set.
624 */
625 while (idlen > 0) {
626 *cp++ = (u_int8_t)(arc4random() % 256);
627 idlen--;
628 }
629 }
630
631 /*
632 * Fill in a lock owner name. For now, pid + the process's creation time.
633 */
634 void
nfscl_filllockowner(void * id,u_int8_t * cp,int flags)635 nfscl_filllockowner(void *id, u_int8_t *cp, int flags)
636 {
637 union {
638 u_int32_t lval;
639 u_int8_t cval[4];
640 } tl;
641 struct proc *p;
642
643 if (id == NULL) {
644 printf("NULL id\n");
645 bzero(cp, NFSV4CL_LOCKNAMELEN);
646 return;
647 }
648 if ((flags & F_POSIX) != 0) {
649 p = (struct proc *)id;
650 tl.lval = p->p_pid;
651 *cp++ = tl.cval[0];
652 *cp++ = tl.cval[1];
653 *cp++ = tl.cval[2];
654 *cp++ = tl.cval[3];
655 tl.lval = p->p_stats->p_start.tv_sec;
656 *cp++ = tl.cval[0];
657 *cp++ = tl.cval[1];
658 *cp++ = tl.cval[2];
659 *cp++ = tl.cval[3];
660 tl.lval = p->p_stats->p_start.tv_usec;
661 *cp++ = tl.cval[0];
662 *cp++ = tl.cval[1];
663 *cp++ = tl.cval[2];
664 *cp = tl.cval[3];
665 } else if ((flags & F_FLOCK) != 0) {
666 bcopy(&id, cp, sizeof(id));
667 bzero(&cp[sizeof(id)], NFSV4CL_LOCKNAMELEN - sizeof(id));
668 } else {
669 printf("nfscl_filllockowner: not F_POSIX or F_FLOCK\n");
670 bzero(cp, NFSV4CL_LOCKNAMELEN);
671 }
672 }
673
674 /*
675 * Find the parent process for the thread passed in as an argument.
676 * If none exists, return NULL, otherwise return a thread for the parent.
677 * (Can be any of the threads, since it is only used for td->td_proc.)
678 */
679 NFSPROC_T *
nfscl_getparent(struct thread * td)680 nfscl_getparent(struct thread *td)
681 {
682 struct proc *p;
683 struct thread *ptd;
684
685 if (td == NULL)
686 return (NULL);
687 p = td->td_proc;
688 if (p->p_pid == 0)
689 return (NULL);
690 p = p->p_pptr;
691 if (p == NULL)
692 return (NULL);
693 ptd = TAILQ_FIRST(&p->p_threads);
694 return (ptd);
695 }
696
697 /*
698 * Start up the renew kernel thread.
699 */
700 static void
start_nfscl(void * arg)701 start_nfscl(void *arg)
702 {
703 struct nfsclclient *clp;
704 struct thread *td;
705
706 clp = (struct nfsclclient *)arg;
707 td = TAILQ_FIRST(&clp->nfsc_renewthread->p_threads);
708 nfscl_renewthread(clp, td);
709 kproc_exit(0);
710 }
711
712 void
nfscl_start_renewthread(struct nfsclclient * clp)713 nfscl_start_renewthread(struct nfsclclient *clp)
714 {
715
716 kproc_create(start_nfscl, (void *)clp, &clp->nfsc_renewthread, 0, 0,
717 "nfscl");
718 }
719
720 /*
721 * Handle wcc_data.
722 * For NFSv4, it assumes that nfsv4_wccattr() was used to set up the getattr
723 * as the first Op after PutFH.
724 * (For NFSv4, the postop attributes are after the Op, so they can't be
725 * parsed here. A separate call to nfscl_postop_attr() is required.)
726 */
727 int
nfscl_wcc_data(struct nfsrv_descript * nd,struct vnode * vp,struct nfsvattr * nap,int * flagp,int * wccflagp,void * stuff)728 nfscl_wcc_data(struct nfsrv_descript *nd, struct vnode *vp,
729 struct nfsvattr *nap, int *flagp, int *wccflagp, void *stuff)
730 {
731 u_int32_t *tl;
732 struct nfsnode *np = VTONFS(vp);
733 struct nfsvattr nfsva;
734 int error = 0;
735
736 if (wccflagp != NULL)
737 *wccflagp = 0;
738 if (nd->nd_flag & ND_NFSV3) {
739 *flagp = 0;
740 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
741 if (*tl == newnfs_true) {
742 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
743 if (wccflagp != NULL) {
744 mtx_lock(&np->n_mtx);
745 *wccflagp = (np->n_mtime.tv_sec ==
746 fxdr_unsigned(u_int32_t, *(tl + 2)) &&
747 np->n_mtime.tv_nsec ==
748 fxdr_unsigned(u_int32_t, *(tl + 3)));
749 mtx_unlock(&np->n_mtx);
750 }
751 }
752 error = nfscl_postop_attr(nd, nap, flagp, stuff);
753 } else if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR))
754 == (ND_NFSV4 | ND_V4WCCATTR)) {
755 error = nfsv4_loadattr(nd, NULL, &nfsva, NULL,
756 NULL, 0, NULL, NULL, NULL, NULL, NULL, 0,
757 NULL, NULL, NULL, NULL, NULL);
758 if (error)
759 return (error);
760 /*
761 * Get rid of Op# and status for next op.
762 */
763 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
764 if (*++tl)
765 nd->nd_flag |= ND_NOMOREDATA;
766 if (wccflagp != NULL &&
767 nfsva.na_vattr.va_mtime.tv_sec != 0) {
768 mtx_lock(&np->n_mtx);
769 *wccflagp = (np->n_mtime.tv_sec ==
770 nfsva.na_vattr.va_mtime.tv_sec &&
771 np->n_mtime.tv_nsec ==
772 nfsva.na_vattr.va_mtime.tv_sec);
773 mtx_unlock(&np->n_mtx);
774 }
775 }
776 nfsmout:
777 return (error);
778 }
779
780 /*
781 * Get postop attributes.
782 */
783 int
nfscl_postop_attr(struct nfsrv_descript * nd,struct nfsvattr * nap,int * retp,void * stuff)784 nfscl_postop_attr(struct nfsrv_descript *nd, struct nfsvattr *nap, int *retp,
785 void *stuff)
786 {
787 u_int32_t *tl;
788 int error = 0;
789
790 *retp = 0;
791 if (nd->nd_flag & ND_NOMOREDATA)
792 return (error);
793 if (nd->nd_flag & ND_NFSV3) {
794 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
795 *retp = fxdr_unsigned(int, *tl);
796 } else if (nd->nd_flag & ND_NFSV4) {
797 /*
798 * For NFSv4, the postop attr are at the end, so no point
799 * in looking if nd_repstat != 0.
800 */
801 if (!nd->nd_repstat) {
802 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
803 if (*(tl + 1))
804 /* should never happen since nd_repstat != 0 */
805 nd->nd_flag |= ND_NOMOREDATA;
806 else
807 *retp = 1;
808 }
809 } else if (!nd->nd_repstat) {
810 /* For NFSv2, the attributes are here iff nd_repstat == 0 */
811 *retp = 1;
812 }
813 if (*retp) {
814 error = nfsm_loadattr(nd, nap);
815 if (error)
816 *retp = 0;
817 }
818 nfsmout:
819 return (error);
820 }
821
822 /*
823 * Fill in the setable attributes. The full argument indicates whether
824 * to fill in them all or just mode and time.
825 */
826 void
nfscl_fillsattr(struct nfsrv_descript * nd,struct vattr * vap,struct vnode * vp,int flags,u_int32_t rdev)827 nfscl_fillsattr(struct nfsrv_descript *nd, struct vattr *vap,
828 struct vnode *vp, int flags, u_int32_t rdev)
829 {
830 u_int32_t *tl;
831 struct nfsv2_sattr *sp;
832 nfsattrbit_t attrbits;
833
834 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
835 case ND_NFSV2:
836 NFSM_BUILD(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
837 if (vap->va_mode == (mode_t)VNOVAL)
838 sp->sa_mode = newnfs_xdrneg1;
839 else
840 sp->sa_mode = vtonfsv2_mode(vap->va_type, vap->va_mode);
841 if (vap->va_uid == (uid_t)VNOVAL)
842 sp->sa_uid = newnfs_xdrneg1;
843 else
844 sp->sa_uid = txdr_unsigned(vap->va_uid);
845 if (vap->va_gid == (gid_t)VNOVAL)
846 sp->sa_gid = newnfs_xdrneg1;
847 else
848 sp->sa_gid = txdr_unsigned(vap->va_gid);
849 if (flags & NFSSATTR_SIZE0)
850 sp->sa_size = 0;
851 else if (flags & NFSSATTR_SIZENEG1)
852 sp->sa_size = newnfs_xdrneg1;
853 else if (flags & NFSSATTR_SIZERDEV)
854 sp->sa_size = txdr_unsigned(rdev);
855 else
856 sp->sa_size = txdr_unsigned(vap->va_size);
857 txdr_nfsv2time(&vap->va_atime, &sp->sa_atime);
858 txdr_nfsv2time(&vap->va_mtime, &sp->sa_mtime);
859 break;
860 case ND_NFSV3:
861 if (vap->va_mode != (mode_t)VNOVAL) {
862 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
863 *tl++ = newnfs_true;
864 *tl = txdr_unsigned(vap->va_mode);
865 } else {
866 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
867 *tl = newnfs_false;
868 }
869 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL) {
870 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
871 *tl++ = newnfs_true;
872 *tl = txdr_unsigned(vap->va_uid);
873 } else {
874 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
875 *tl = newnfs_false;
876 }
877 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL) {
878 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
879 *tl++ = newnfs_true;
880 *tl = txdr_unsigned(vap->va_gid);
881 } else {
882 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
883 *tl = newnfs_false;
884 }
885 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL) {
886 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
887 *tl++ = newnfs_true;
888 txdr_hyper(vap->va_size, tl);
889 } else {
890 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
891 *tl = newnfs_false;
892 }
893 if (vap->va_atime.tv_sec != VNOVAL) {
894 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
895 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
896 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
897 txdr_nfsv3time(&vap->va_atime, tl);
898 } else {
899 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
900 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
901 }
902 } else {
903 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
904 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
905 }
906 if (vap->va_mtime.tv_sec != VNOVAL) {
907 if ((vap->va_vaflags & VA_UTIMES_NULL) == 0) {
908 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
909 *tl++ = txdr_unsigned(NFSV3SATTRTIME_TOCLIENT);
910 txdr_nfsv3time(&vap->va_mtime, tl);
911 } else {
912 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
913 *tl = txdr_unsigned(NFSV3SATTRTIME_TOSERVER);
914 }
915 } else {
916 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
917 *tl = txdr_unsigned(NFSV3SATTRTIME_DONTCHANGE);
918 }
919 break;
920 case ND_NFSV4:
921 NFSZERO_ATTRBIT(&attrbits);
922 if (vap->va_mode != (mode_t)VNOVAL)
923 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_MODE);
924 if ((flags & NFSSATTR_FULL) && vap->va_uid != (uid_t)VNOVAL)
925 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNER);
926 if ((flags & NFSSATTR_FULL) && vap->va_gid != (gid_t)VNOVAL)
927 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_OWNERGROUP);
928 if ((flags & NFSSATTR_FULL) && vap->va_size != VNOVAL)
929 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
930 if (vap->va_atime.tv_sec != VNOVAL)
931 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESSSET);
932 if (vap->va_mtime.tv_sec != VNOVAL)
933 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFYSET);
934 (void) nfsv4_fillattr(nd, vp->v_mount, vp, NULL, vap, NULL, 0,
935 &attrbits, NULL, NULL, 0, 0, 0, 0, (uint64_t)0);
936 break;
937 }
938 }
939
940 /*
941 * nfscl_request() - mostly a wrapper for newnfs_request().
942 */
943 int
nfscl_request(struct nfsrv_descript * nd,struct vnode * vp,NFSPROC_T * p,struct ucred * cred,void * stuff)944 nfscl_request(struct nfsrv_descript *nd, struct vnode *vp, NFSPROC_T *p,
945 struct ucred *cred, void *stuff)
946 {
947 int ret, vers;
948 struct nfsmount *nmp;
949
950 nmp = VFSTONFS(vp->v_mount);
951 if (nd->nd_flag & ND_NFSV4)
952 vers = NFS_VER4;
953 else if (nd->nd_flag & ND_NFSV3)
954 vers = NFS_VER3;
955 else
956 vers = NFS_VER2;
957 ret = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, vp, p, cred,
958 NFS_PROG, vers, NULL, 1, NULL, NULL);
959 return (ret);
960 }
961
962 /*
963 * fill in this bsden's variant of statfs using nfsstatfs.
964 */
965 void
nfscl_loadsbinfo(struct nfsmount * nmp,struct nfsstatfs * sfp,void * statfs)966 nfscl_loadsbinfo(struct nfsmount *nmp, struct nfsstatfs *sfp, void *statfs)
967 {
968 struct statfs *sbp = (struct statfs *)statfs;
969
970 if (nmp->nm_flag & (NFSMNT_NFSV3 | NFSMNT_NFSV4)) {
971 sbp->f_bsize = NFS_FABLKSIZE;
972 sbp->f_blocks = sfp->sf_tbytes / NFS_FABLKSIZE;
973 sbp->f_bfree = sfp->sf_fbytes / NFS_FABLKSIZE;
974 /*
975 * Although sf_abytes is uint64_t and f_bavail is int64_t,
976 * the value after dividing by NFS_FABLKSIZE is small
977 * enough that it will fit in 63bits, so it is ok to
978 * assign it to f_bavail without fear that it will become
979 * negative.
980 */
981 sbp->f_bavail = sfp->sf_abytes / NFS_FABLKSIZE;
982 sbp->f_files = sfp->sf_tfiles;
983 /* Since f_ffree is int64_t, clip it to 63bits. */
984 if (sfp->sf_ffiles > INT64_MAX)
985 sbp->f_ffree = INT64_MAX;
986 else
987 sbp->f_ffree = sfp->sf_ffiles;
988 } else if ((nmp->nm_flag & NFSMNT_NFSV4) == 0) {
989 /*
990 * The type casts to (int32_t) ensure that this code is
991 * compatible with the old NFS client, in that it will
992 * propagate bit31 to the high order bits. This may or may
993 * not be correct for NFSv2, but since it is a legacy
994 * environment, I'd rather retain backwards compatibility.
995 */
996 sbp->f_bsize = (int32_t)sfp->sf_bsize;
997 sbp->f_blocks = (int32_t)sfp->sf_blocks;
998 sbp->f_bfree = (int32_t)sfp->sf_bfree;
999 sbp->f_bavail = (int32_t)sfp->sf_bavail;
1000 sbp->f_files = 0;
1001 sbp->f_ffree = 0;
1002 }
1003 }
1004
1005 /*
1006 * Use the fsinfo stuff to update the mount point.
1007 */
1008 void
nfscl_loadfsinfo(struct nfsmount * nmp,struct nfsfsinfo * fsp)1009 nfscl_loadfsinfo(struct nfsmount *nmp, struct nfsfsinfo *fsp)
1010 {
1011
1012 if ((nmp->nm_wsize == 0 || fsp->fs_wtpref < nmp->nm_wsize) &&
1013 fsp->fs_wtpref >= NFS_FABLKSIZE)
1014 nmp->nm_wsize = (fsp->fs_wtpref + NFS_FABLKSIZE - 1) &
1015 ~(NFS_FABLKSIZE - 1);
1016 if (fsp->fs_wtmax < nmp->nm_wsize && fsp->fs_wtmax > 0) {
1017 nmp->nm_wsize = fsp->fs_wtmax & ~(NFS_FABLKSIZE - 1);
1018 if (nmp->nm_wsize == 0)
1019 nmp->nm_wsize = fsp->fs_wtmax;
1020 }
1021 if (nmp->nm_wsize < NFS_FABLKSIZE)
1022 nmp->nm_wsize = NFS_FABLKSIZE;
1023 if ((nmp->nm_rsize == 0 || fsp->fs_rtpref < nmp->nm_rsize) &&
1024 fsp->fs_rtpref >= NFS_FABLKSIZE)
1025 nmp->nm_rsize = (fsp->fs_rtpref + NFS_FABLKSIZE - 1) &
1026 ~(NFS_FABLKSIZE - 1);
1027 if (fsp->fs_rtmax < nmp->nm_rsize && fsp->fs_rtmax > 0) {
1028 nmp->nm_rsize = fsp->fs_rtmax & ~(NFS_FABLKSIZE - 1);
1029 if (nmp->nm_rsize == 0)
1030 nmp->nm_rsize = fsp->fs_rtmax;
1031 }
1032 if (nmp->nm_rsize < NFS_FABLKSIZE)
1033 nmp->nm_rsize = NFS_FABLKSIZE;
1034 if ((nmp->nm_readdirsize == 0 || fsp->fs_dtpref < nmp->nm_readdirsize)
1035 && fsp->fs_dtpref >= NFS_DIRBLKSIZ)
1036 nmp->nm_readdirsize = (fsp->fs_dtpref + NFS_DIRBLKSIZ - 1) &
1037 ~(NFS_DIRBLKSIZ - 1);
1038 if (fsp->fs_rtmax < nmp->nm_readdirsize && fsp->fs_rtmax > 0) {
1039 nmp->nm_readdirsize = fsp->fs_rtmax & ~(NFS_DIRBLKSIZ - 1);
1040 if (nmp->nm_readdirsize == 0)
1041 nmp->nm_readdirsize = fsp->fs_rtmax;
1042 }
1043 if (nmp->nm_readdirsize < NFS_DIRBLKSIZ)
1044 nmp->nm_readdirsize = NFS_DIRBLKSIZ;
1045 if (fsp->fs_maxfilesize > 0 &&
1046 fsp->fs_maxfilesize < nmp->nm_maxfilesize)
1047 nmp->nm_maxfilesize = fsp->fs_maxfilesize;
1048 nmp->nm_mountp->mnt_stat.f_iosize = newnfs_iosize(nmp);
1049 nmp->nm_state |= NFSSTA_GOTFSINFO;
1050 }
1051
1052 /*
1053 * Lookups source address which should be used to communicate with
1054 * @nmp and stores it inside @pdst.
1055 *
1056 * Returns 0 on success.
1057 */
1058 u_int8_t *
nfscl_getmyip(struct nfsmount * nmp,struct in6_addr * paddr,int * isinet6p)1059 nfscl_getmyip(struct nfsmount *nmp, struct in6_addr *paddr, int *isinet6p)
1060 {
1061 #if defined(INET6) || defined(INET)
1062 int error, fibnum;
1063
1064 fibnum = curthread->td_proc->p_fibnum;
1065 #endif
1066 #ifdef INET
1067 if (nmp->nm_nam->sa_family == AF_INET) {
1068 struct sockaddr_in *sin;
1069 struct nhop4_extended nh_ext;
1070
1071 sin = (struct sockaddr_in *)nmp->nm_nam;
1072 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1073 error = fib4_lookup_nh_ext(fibnum, sin->sin_addr, 0, 0,
1074 &nh_ext);
1075 CURVNET_RESTORE();
1076 if (error != 0)
1077 return (NULL);
1078
1079 if ((ntohl(nh_ext.nh_src.s_addr) >> IN_CLASSA_NSHIFT) ==
1080 IN_LOOPBACKNET) {
1081 /* Ignore loopback addresses */
1082 return (NULL);
1083 }
1084
1085 *isinet6p = 0;
1086 *((struct in_addr *)paddr) = nh_ext.nh_src;
1087
1088 return (u_int8_t *)paddr;
1089 }
1090 #endif
1091 #ifdef INET6
1092 if (nmp->nm_nam->sa_family == AF_INET6) {
1093 struct sockaddr_in6 *sin6;
1094
1095 sin6 = (struct sockaddr_in6 *)nmp->nm_nam;
1096
1097 CURVNET_SET(CRED_TO_VNET(nmp->nm_sockreq.nr_cred));
1098 error = in6_selectsrc_addr(fibnum, &sin6->sin6_addr,
1099 sin6->sin6_scope_id, NULL, paddr, NULL);
1100 CURVNET_RESTORE();
1101 if (error != 0)
1102 return (NULL);
1103
1104 if (IN6_IS_ADDR_LOOPBACK(paddr))
1105 return (NULL);
1106
1107 /* Scope is embedded in */
1108 *isinet6p = 1;
1109
1110 return (u_int8_t *)paddr;
1111 }
1112 #endif
1113 return (NULL);
1114 }
1115
1116 /*
1117 * Copy NFS uid, gids from the cred structure.
1118 */
1119 void
newnfs_copyincred(struct ucred * cr,struct nfscred * nfscr)1120 newnfs_copyincred(struct ucred *cr, struct nfscred *nfscr)
1121 {
1122 int i;
1123
1124 KASSERT(cr->cr_ngroups >= 0,
1125 ("newnfs_copyincred: negative cr_ngroups"));
1126 nfscr->nfsc_uid = cr->cr_uid;
1127 nfscr->nfsc_ngroups = MIN(cr->cr_ngroups, NFS_MAXGRPS + 1);
1128 for (i = 0; i < nfscr->nfsc_ngroups; i++)
1129 nfscr->nfsc_groups[i] = cr->cr_groups[i];
1130 }
1131
1132
1133 /*
1134 * Do any client specific initialization.
1135 */
1136 void
nfscl_init(void)1137 nfscl_init(void)
1138 {
1139 static int inited = 0;
1140
1141 if (inited)
1142 return;
1143 inited = 1;
1144 nfscl_inited = 1;
1145 ncl_pbuf_freecnt = nswbuf / 2 + 1;
1146 }
1147
1148 /*
1149 * Check each of the attributes to be set, to ensure they aren't already
1150 * the correct value. Disable setting ones already correct.
1151 */
1152 int
nfscl_checksattr(struct vattr * vap,struct nfsvattr * nvap)1153 nfscl_checksattr(struct vattr *vap, struct nfsvattr *nvap)
1154 {
1155
1156 if (vap->va_mode != (mode_t)VNOVAL) {
1157 if (vap->va_mode == nvap->na_mode)
1158 vap->va_mode = (mode_t)VNOVAL;
1159 }
1160 if (vap->va_uid != (uid_t)VNOVAL) {
1161 if (vap->va_uid == nvap->na_uid)
1162 vap->va_uid = (uid_t)VNOVAL;
1163 }
1164 if (vap->va_gid != (gid_t)VNOVAL) {
1165 if (vap->va_gid == nvap->na_gid)
1166 vap->va_gid = (gid_t)VNOVAL;
1167 }
1168 if (vap->va_size != VNOVAL) {
1169 if (vap->va_size == nvap->na_size)
1170 vap->va_size = VNOVAL;
1171 }
1172
1173 /*
1174 * We are normally called with only a partially initialized
1175 * VAP. Since the NFSv3 spec says that server may use the
1176 * file attributes to store the verifier, the spec requires
1177 * us to do a SETATTR RPC. FreeBSD servers store the verifier
1178 * in atime, but we can't really assume that all servers will
1179 * so we ensure that our SETATTR sets both atime and mtime.
1180 * Set the VA_UTIMES_NULL flag for this case, so that
1181 * the server's time will be used. This is needed to
1182 * work around a bug in some Solaris servers, where
1183 * setting the time TOCLIENT causes the Setattr RPC
1184 * to return NFS_OK, but not set va_mode.
1185 */
1186 if (vap->va_mtime.tv_sec == VNOVAL) {
1187 vfs_timestamp(&vap->va_mtime);
1188 vap->va_vaflags |= VA_UTIMES_NULL;
1189 }
1190 if (vap->va_atime.tv_sec == VNOVAL)
1191 vap->va_atime = vap->va_mtime;
1192 return (1);
1193 }
1194
1195 /*
1196 * Map nfsv4 errors to errno.h errors.
1197 * The uid and gid arguments are only used for NFSERR_BADOWNER and that
1198 * error should only be returned for the Open, Create and Setattr Ops.
1199 * As such, most calls can just pass in 0 for those arguments.
1200 */
1201 APPLESTATIC int
nfscl_maperr(struct thread * td,int error,uid_t uid,gid_t gid)1202 nfscl_maperr(struct thread *td, int error, uid_t uid, gid_t gid)
1203 {
1204 struct proc *p;
1205
1206 if (error < 10000)
1207 return (error);
1208 if (td != NULL)
1209 p = td->td_proc;
1210 else
1211 p = NULL;
1212 switch (error) {
1213 case NFSERR_BADOWNER:
1214 tprintf(p, LOG_INFO,
1215 "No name and/or group mapping for uid,gid:(%d,%d)\n",
1216 uid, gid);
1217 return (EPERM);
1218 case NFSERR_BADNAME:
1219 case NFSERR_BADCHAR:
1220 printf("nfsv4 char/name not handled by server\n");
1221 return (ENOENT);
1222 case NFSERR_STALECLIENTID:
1223 case NFSERR_STALESTATEID:
1224 case NFSERR_EXPIRED:
1225 case NFSERR_BADSTATEID:
1226 case NFSERR_BADSESSION:
1227 printf("nfsv4 recover err returned %d\n", error);
1228 return (EIO);
1229 case NFSERR_BADHANDLE:
1230 case NFSERR_SERVERFAULT:
1231 case NFSERR_BADTYPE:
1232 case NFSERR_FHEXPIRED:
1233 case NFSERR_RESOURCE:
1234 case NFSERR_MOVED:
1235 case NFSERR_NOFILEHANDLE:
1236 case NFSERR_MINORVERMISMATCH:
1237 case NFSERR_OLDSTATEID:
1238 case NFSERR_BADSEQID:
1239 case NFSERR_LEASEMOVED:
1240 case NFSERR_RECLAIMBAD:
1241 case NFSERR_BADXDR:
1242 case NFSERR_OPILLEGAL:
1243 printf("nfsv4 client/server protocol prob err=%d\n",
1244 error);
1245 return (EIO);
1246 default:
1247 tprintf(p, LOG_INFO, "nfsv4 err=%d\n", error);
1248 return (EIO);
1249 };
1250 }
1251
1252 /*
1253 * Check to see if the process for this owner exists. Return 1 if it doesn't
1254 * and 0 otherwise.
1255 */
1256 int
nfscl_procdoesntexist(u_int8_t * own)1257 nfscl_procdoesntexist(u_int8_t *own)
1258 {
1259 union {
1260 u_int32_t lval;
1261 u_int8_t cval[4];
1262 } tl;
1263 struct proc *p;
1264 pid_t pid;
1265 int ret = 0;
1266
1267 tl.cval[0] = *own++;
1268 tl.cval[1] = *own++;
1269 tl.cval[2] = *own++;
1270 tl.cval[3] = *own++;
1271 pid = tl.lval;
1272 p = pfind_locked(pid);
1273 if (p == NULL)
1274 return (1);
1275 if (p->p_stats == NULL) {
1276 PROC_UNLOCK(p);
1277 return (0);
1278 }
1279 tl.cval[0] = *own++;
1280 tl.cval[1] = *own++;
1281 tl.cval[2] = *own++;
1282 tl.cval[3] = *own++;
1283 if (tl.lval != p->p_stats->p_start.tv_sec) {
1284 ret = 1;
1285 } else {
1286 tl.cval[0] = *own++;
1287 tl.cval[1] = *own++;
1288 tl.cval[2] = *own++;
1289 tl.cval[3] = *own;
1290 if (tl.lval != p->p_stats->p_start.tv_usec)
1291 ret = 1;
1292 }
1293 PROC_UNLOCK(p);
1294 return (ret);
1295 }
1296
1297 /*
1298 * - nfs pseudo system call for the client
1299 */
1300 /*
1301 * MPSAFE
1302 */
1303 static int
nfssvc_nfscl(struct thread * td,struct nfssvc_args * uap)1304 nfssvc_nfscl(struct thread *td, struct nfssvc_args *uap)
1305 {
1306 struct file *fp;
1307 struct nfscbd_args nfscbdarg;
1308 struct nfsd_nfscbd_args nfscbdarg2;
1309 struct nameidata nd;
1310 struct nfscl_dumpmntopts dumpmntopts;
1311 cap_rights_t rights;
1312 char *buf;
1313 int error;
1314
1315 if (uap->flag & NFSSVC_CBADDSOCK) {
1316 error = copyin(uap->argp, (caddr_t)&nfscbdarg, sizeof(nfscbdarg));
1317 if (error)
1318 return (error);
1319 /*
1320 * Since we don't know what rights might be required,
1321 * pretend that we need them all. It is better to be too
1322 * careful than too reckless.
1323 */
1324 error = fget(td, nfscbdarg.sock,
1325 cap_rights_init(&rights, CAP_SOCK_CLIENT), &fp);
1326 if (error)
1327 return (error);
1328 if (fp->f_type != DTYPE_SOCKET) {
1329 fdrop(fp, td);
1330 return (EPERM);
1331 }
1332 error = nfscbd_addsock(fp);
1333 fdrop(fp, td);
1334 if (!error && nfscl_enablecallb == 0) {
1335 nfsv4_cbport = nfscbdarg.port;
1336 nfscl_enablecallb = 1;
1337 }
1338 } else if (uap->flag & NFSSVC_NFSCBD) {
1339 if (uap->argp == NULL)
1340 return (EINVAL);
1341 error = copyin(uap->argp, (caddr_t)&nfscbdarg2,
1342 sizeof(nfscbdarg2));
1343 if (error)
1344 return (error);
1345 error = nfscbd_nfsd(td, &nfscbdarg2);
1346 } else if (uap->flag & NFSSVC_DUMPMNTOPTS) {
1347 error = copyin(uap->argp, &dumpmntopts, sizeof(dumpmntopts));
1348 if (error == 0 && (dumpmntopts.ndmnt_blen < 256 ||
1349 dumpmntopts.ndmnt_blen > 1024))
1350 error = EINVAL;
1351 if (error == 0)
1352 error = nfsrv_lookupfilename(&nd,
1353 dumpmntopts.ndmnt_fname, td);
1354 if (error == 0 && strcmp(nd.ni_vp->v_mount->mnt_vfc->vfc_name,
1355 "nfs") != 0) {
1356 vput(nd.ni_vp);
1357 error = EINVAL;
1358 }
1359 if (error == 0) {
1360 buf = malloc(dumpmntopts.ndmnt_blen, M_TEMP, M_WAITOK);
1361 nfscl_retopts(VFSTONFS(nd.ni_vp->v_mount), buf,
1362 dumpmntopts.ndmnt_blen);
1363 vput(nd.ni_vp);
1364 error = copyout(buf, dumpmntopts.ndmnt_buf,
1365 dumpmntopts.ndmnt_blen);
1366 free(buf, M_TEMP);
1367 }
1368 } else {
1369 error = EINVAL;
1370 }
1371 return (error);
1372 }
1373
1374 extern int (*nfsd_call_nfscl)(struct thread *, struct nfssvc_args *);
1375
1376 /*
1377 * Called once to initialize data structures...
1378 */
1379 static int
nfscl_modevent(module_t mod,int type,void * data)1380 nfscl_modevent(module_t mod, int type, void *data)
1381 {
1382 int error = 0;
1383 static int loaded = 0;
1384
1385 switch (type) {
1386 case MOD_LOAD:
1387 if (loaded)
1388 return (0);
1389 newnfs_portinit();
1390 mtx_init(&nfs_clstate_mutex, "nfs_clstate_mutex", NULL,
1391 MTX_DEF);
1392 mtx_init(&ncl_iod_mutex, "ncl_iod_mutex", NULL, MTX_DEF);
1393 nfscl_init();
1394 NFSD_LOCK();
1395 nfsrvd_cbinit(0);
1396 NFSD_UNLOCK();
1397 ncl_call_invalcaches = ncl_invalcaches;
1398 nfsd_call_nfscl = nfssvc_nfscl;
1399 loaded = 1;
1400 break;
1401
1402 case MOD_UNLOAD:
1403 if (nfs_numnfscbd != 0) {
1404 error = EBUSY;
1405 break;
1406 }
1407
1408 /*
1409 * XXX: Unloading of nfscl module is unsupported.
1410 */
1411 #if 0
1412 ncl_call_invalcaches = NULL;
1413 nfsd_call_nfscl = NULL;
1414 /* and get rid of the mutexes */
1415 mtx_destroy(&nfs_clstate_mutex);
1416 mtx_destroy(&ncl_iod_mutex);
1417 loaded = 0;
1418 break;
1419 #else
1420 /* FALLTHROUGH */
1421 #endif
1422 default:
1423 error = EOPNOTSUPP;
1424 break;
1425 }
1426 return error;
1427 }
1428 static moduledata_t nfscl_mod = {
1429 "nfscl",
1430 nfscl_modevent,
1431 NULL,
1432 };
1433 DECLARE_MODULE(nfscl, nfscl_mod, SI_SUB_VFS, SI_ORDER_FIRST);
1434
1435 /* So that loader and kldload(2) can find us, wherever we are.. */
1436 MODULE_VERSION(nfscl, 1);
1437 MODULE_DEPEND(nfscl, nfscommon, 1, 1, 1);
1438 MODULE_DEPEND(nfscl, krpc, 1, 1, 1);
1439 MODULE_DEPEND(nfscl, nfssvc, 1, 1, 1);
1440 MODULE_DEPEND(nfscl, nfslock, 1, 1, 1);
1441
1442