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: releng/10.4/sys/fs/nfsserver/nfs_nfsdport.c 317914 2017-05-07 19:57:45Z rmacklem $");
36
37 #include <sys/capsicum.h>
38
39 /*
40 * Functions that perform the vfs operations required by the routines in
41 * nfsd_serv.c. It is hoped that this change will make the server more
42 * portable.
43 */
44
45 #include <fs/nfs/nfsport.h>
46 #include <sys/hash.h>
47 #include <sys/sysctl.h>
48 #include <nlm/nlm_prot.h>
49 #include <nlm/nlm.h>
50
51 FEATURE(nfsd, "NFSv4 server");
52
53 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
54 extern int nfsrv_useacl;
55 extern int newnfs_numnfsd;
56 extern struct mount nfsv4root_mnt;
57 extern struct nfsrv_stablefirst nfsrv_stablefirst;
58 extern void (*nfsd_call_servertimer)(void);
59 extern SVCPOOL *nfsrvd_pool;
60 extern struct nfsv4lock nfsd_suspend_lock;
61 extern struct nfsclienthashhead *nfsclienthash;
62 extern struct nfslockhashhead *nfslockhash;
63 extern struct nfssessionhash *nfssessionhash;
64 extern int nfsrv_sessionhashsize;
65 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
66 NFSDLOCKMUTEX;
67 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
68 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
69 struct mtx nfsrc_udpmtx;
70 struct mtx nfs_v4root_mutex;
71 struct nfsrvfh nfs_rootfh, nfs_pubfh;
72 int nfs_pubfhset = 0, nfs_rootfhset = 0;
73 struct proc *nfsd_master_proc = NULL;
74 int nfsd_debuglevel = 0;
75 static pid_t nfsd_master_pid = (pid_t)-1;
76 static char nfsd_master_comm[MAXCOMLEN + 1];
77 static struct timeval nfsd_master_start;
78 static uint32_t nfsv4_sysid = 0;
79
80 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
81 struct ucred *);
82
83 int nfsrv_enable_crossmntpt = 1;
84 static int nfs_commit_blks;
85 static int nfs_commit_miss;
86 extern int nfsrv_issuedelegs;
87 extern int nfsrv_dolocallocks;
88 extern int nfsd_enable_stringtouid;
89
90 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
91 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
92 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
93 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
94 0, "");
95 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
96 0, "");
97 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
98 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
99 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
100 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
101 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
102 0, "Debug level for new nfs server");
103 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
104 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
105
106 #define MAX_REORDERED_RPC 16
107 #define NUM_HEURISTIC 1031
108 #define NHUSE_INIT 64
109 #define NHUSE_INC 16
110 #define NHUSE_MAX 2048
111
112 static struct nfsheur {
113 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
114 off_t nh_nextoff; /* next offset for sequential detection */
115 int nh_use; /* use count for selection */
116 int nh_seqcount; /* heuristic */
117 } nfsheur[NUM_HEURISTIC];
118
119
120 /*
121 * Heuristic to detect sequential operation.
122 */
123 static struct nfsheur *
124 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
125 {
126 struct nfsheur *nh;
127 int hi, try;
128
129 /* Locate best candidate. */
130 try = 32;
131 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
132 nh = &nfsheur[hi];
133 while (try--) {
134 if (nfsheur[hi].nh_vp == vp) {
135 nh = &nfsheur[hi];
136 break;
137 }
138 if (nfsheur[hi].nh_use > 0)
139 --nfsheur[hi].nh_use;
140 hi = (hi + 1) % NUM_HEURISTIC;
141 if (nfsheur[hi].nh_use < nh->nh_use)
142 nh = &nfsheur[hi];
143 }
144
145 /* Initialize hint if this is a new file. */
146 if (nh->nh_vp != vp) {
147 nh->nh_vp = vp;
148 nh->nh_nextoff = uio->uio_offset;
149 nh->nh_use = NHUSE_INIT;
150 if (uio->uio_offset == 0)
151 nh->nh_seqcount = 4;
152 else
153 nh->nh_seqcount = 1;
154 }
155
156 /* Calculate heuristic. */
157 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
158 uio->uio_offset == nh->nh_nextoff) {
159 /* See comments in vfs_vnops.c:sequential_heuristic(). */
160 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
161 if (nh->nh_seqcount > IO_SEQMAX)
162 nh->nh_seqcount = IO_SEQMAX;
163 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
164 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
165 /* Probably a reordered RPC, leave seqcount alone. */
166 } else if (nh->nh_seqcount > 1) {
167 nh->nh_seqcount /= 2;
168 } else {
169 nh->nh_seqcount = 0;
170 }
171 nh->nh_use += NHUSE_INC;
172 if (nh->nh_use > NHUSE_MAX)
173 nh->nh_use = NHUSE_MAX;
174 return (nh);
175 }
176
177 /*
178 * Get attributes into nfsvattr structure.
179 */
180 int
181 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
182 struct thread *p, int vpislocked)
183 {
184 int error, lockedit = 0;
185
186 if (vpislocked == 0) {
187 /*
188 * When vpislocked == 0, the vnode is either exclusively
189 * locked by this thread or not locked by this thread.
190 * As such, shared lock it, if not exclusively locked.
191 */
192 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
193 lockedit = 1;
194 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
195 }
196 }
197 error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
198 if (lockedit != 0)
199 NFSVOPUNLOCK(vp, 0);
200
201 NFSEXITCODE(error);
202 return (error);
203 }
204
205 /*
206 * Get a file handle for a vnode.
207 */
208 int
209 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
210 {
211 int error;
212
213 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
214 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
215 error = VOP_VPTOFH(vp, &fhp->fh_fid);
216
217 NFSEXITCODE(error);
218 return (error);
219 }
220
221 /*
222 * Perform access checking for vnodes obtained from file handles that would
223 * refer to files already opened by a Unix client. You cannot just use
224 * vn_writechk() and VOP_ACCESSX() for two reasons.
225 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
226 * case.
227 * 2 - The owner is to be given access irrespective of mode bits for some
228 * operations, so that processes that chmod after opening a file don't
229 * break.
230 */
231 int
232 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
233 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
234 u_int32_t *supportedtypep)
235 {
236 struct vattr vattr;
237 int error = 0, getret = 0;
238
239 if (vpislocked == 0) {
240 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
241 error = EPERM;
242 goto out;
243 }
244 }
245 if (accmode & VWRITE) {
246 /* Just vn_writechk() changed to check rdonly */
247 /*
248 * Disallow write attempts on read-only file systems;
249 * unless the file is a socket or a block or character
250 * device resident on the file system.
251 */
252 if (NFSVNO_EXRDONLY(exp) ||
253 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
254 switch (vp->v_type) {
255 case VREG:
256 case VDIR:
257 case VLNK:
258 error = EROFS;
259 default:
260 break;
261 }
262 }
263 /*
264 * If there's shared text associated with
265 * the inode, try to free it up once. If
266 * we fail, we can't allow writing.
267 */
268 if (VOP_IS_TEXT(vp) && error == 0)
269 error = ETXTBSY;
270 }
271 if (error != 0) {
272 if (vpislocked == 0)
273 NFSVOPUNLOCK(vp, 0);
274 goto out;
275 }
276
277 /*
278 * Should the override still be applied when ACLs are enabled?
279 */
280 error = VOP_ACCESSX(vp, accmode, cred, p);
281 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
282 /*
283 * Try again with VEXPLICIT_DENY, to see if the test for
284 * deletion is supported.
285 */
286 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
287 if (error == 0) {
288 if (vp->v_type == VDIR) {
289 accmode &= ~(VDELETE | VDELETE_CHILD);
290 accmode |= VWRITE;
291 error = VOP_ACCESSX(vp, accmode, cred, p);
292 } else if (supportedtypep != NULL) {
293 *supportedtypep &= ~NFSACCESS_DELETE;
294 }
295 }
296 }
297
298 /*
299 * Allow certain operations for the owner (reads and writes
300 * on files that are already open).
301 */
302 if (override != NFSACCCHK_NOOVERRIDE &&
303 (error == EPERM || error == EACCES)) {
304 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
305 error = 0;
306 else if (override & NFSACCCHK_ALLOWOWNER) {
307 getret = VOP_GETATTR(vp, &vattr, cred);
308 if (getret == 0 && cred->cr_uid == vattr.va_uid)
309 error = 0;
310 }
311 }
312 if (vpislocked == 0)
313 NFSVOPUNLOCK(vp, 0);
314
315 out:
316 NFSEXITCODE(error);
317 return (error);
318 }
319
320 /*
321 * Set attribute(s) vnop.
322 */
323 int
324 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
325 struct thread *p, struct nfsexstuff *exp)
326 {
327 int error;
328
329 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
330 NFSEXITCODE(error);
331 return (error);
332 }
333
334 /*
335 * Set up nameidata for a lookup() call and do it.
336 */
337 int
338 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
339 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
340 struct vnode **retdirp)
341 {
342 struct componentname *cnp = &ndp->ni_cnd;
343 int i;
344 struct iovec aiov;
345 struct uio auio;
346 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
347 int error = 0, crossmnt;
348 char *cp;
349
350 *retdirp = NULL;
351 cnp->cn_nameptr = cnp->cn_pnbuf;
352 ndp->ni_strictrelative = 0;
353 /*
354 * Extract and set starting directory.
355 */
356 if (dp->v_type != VDIR) {
357 if (islocked)
358 vput(dp);
359 else
360 vrele(dp);
361 nfsvno_relpathbuf(ndp);
362 error = ENOTDIR;
363 goto out1;
364 }
365 if (islocked)
366 NFSVOPUNLOCK(dp, 0);
367 VREF(dp);
368 *retdirp = dp;
369 if (NFSVNO_EXRDONLY(exp))
370 cnp->cn_flags |= RDONLY;
371 ndp->ni_segflg = UIO_SYSSPACE;
372 crossmnt = 1;
373
374 if (nd->nd_flag & ND_PUBLOOKUP) {
375 ndp->ni_loopcnt = 0;
376 if (cnp->cn_pnbuf[0] == '/') {
377 vrele(dp);
378 /*
379 * Check for degenerate pathnames here, since lookup()
380 * panics on them.
381 */
382 for (i = 1; i < ndp->ni_pathlen; i++)
383 if (cnp->cn_pnbuf[i] != '/')
384 break;
385 if (i == ndp->ni_pathlen) {
386 error = NFSERR_ACCES;
387 goto out;
388 }
389 dp = rootvnode;
390 VREF(dp);
391 }
392 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
393 (nd->nd_flag & ND_NFSV4) == 0) {
394 /*
395 * Only cross mount points for NFSv4 when doing a
396 * mount while traversing the file system above
397 * the mount point, unless nfsrv_enable_crossmntpt is set.
398 */
399 cnp->cn_flags |= NOCROSSMOUNT;
400 crossmnt = 0;
401 }
402
403 /*
404 * Initialize for scan, set ni_startdir and bump ref on dp again
405 * because lookup() will dereference ni_startdir.
406 */
407
408 cnp->cn_thread = p;
409 ndp->ni_startdir = dp;
410 ndp->ni_rootdir = rootvnode;
411 ndp->ni_topdir = NULL;
412
413 if (!lockleaf)
414 cnp->cn_flags |= LOCKLEAF;
415 for (;;) {
416 cnp->cn_nameptr = cnp->cn_pnbuf;
417 /*
418 * Call lookup() to do the real work. If an error occurs,
419 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
420 * we do not have to dereference anything before returning.
421 * In either case ni_startdir will be dereferenced and NULLed
422 * out.
423 */
424 error = lookup(ndp);
425 if (error)
426 break;
427
428 /*
429 * Check for encountering a symbolic link. Trivial
430 * termination occurs if no symlink encountered.
431 */
432 if ((cnp->cn_flags & ISSYMLINK) == 0) {
433 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
434 nfsvno_relpathbuf(ndp);
435 if (ndp->ni_vp && !lockleaf)
436 NFSVOPUNLOCK(ndp->ni_vp, 0);
437 break;
438 }
439
440 /*
441 * Validate symlink
442 */
443 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
444 NFSVOPUNLOCK(ndp->ni_dvp, 0);
445 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
446 error = EINVAL;
447 goto badlink2;
448 }
449
450 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
451 error = ELOOP;
452 goto badlink2;
453 }
454 if (ndp->ni_pathlen > 1)
455 cp = uma_zalloc(namei_zone, M_WAITOK);
456 else
457 cp = cnp->cn_pnbuf;
458 aiov.iov_base = cp;
459 aiov.iov_len = MAXPATHLEN;
460 auio.uio_iov = &aiov;
461 auio.uio_iovcnt = 1;
462 auio.uio_offset = 0;
463 auio.uio_rw = UIO_READ;
464 auio.uio_segflg = UIO_SYSSPACE;
465 auio.uio_td = NULL;
466 auio.uio_resid = MAXPATHLEN;
467 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
468 if (error) {
469 badlink1:
470 if (ndp->ni_pathlen > 1)
471 uma_zfree(namei_zone, cp);
472 badlink2:
473 vrele(ndp->ni_dvp);
474 vput(ndp->ni_vp);
475 break;
476 }
477 linklen = MAXPATHLEN - auio.uio_resid;
478 if (linklen == 0) {
479 error = ENOENT;
480 goto badlink1;
481 }
482 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
483 error = ENAMETOOLONG;
484 goto badlink1;
485 }
486
487 /*
488 * Adjust or replace path
489 */
490 if (ndp->ni_pathlen > 1) {
491 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
492 uma_zfree(namei_zone, cnp->cn_pnbuf);
493 cnp->cn_pnbuf = cp;
494 } else
495 cnp->cn_pnbuf[linklen] = '\0';
496 ndp->ni_pathlen += linklen;
497
498 /*
499 * Cleanup refs for next loop and check if root directory
500 * should replace current directory. Normally ni_dvp
501 * becomes the new base directory and is cleaned up when
502 * we loop. Explicitly null pointers after invalidation
503 * to clarify operation.
504 */
505 vput(ndp->ni_vp);
506 ndp->ni_vp = NULL;
507
508 if (cnp->cn_pnbuf[0] == '/') {
509 vrele(ndp->ni_dvp);
510 ndp->ni_dvp = ndp->ni_rootdir;
511 VREF(ndp->ni_dvp);
512 }
513 ndp->ni_startdir = ndp->ni_dvp;
514 ndp->ni_dvp = NULL;
515 }
516 if (!lockleaf)
517 cnp->cn_flags &= ~LOCKLEAF;
518
519 out:
520 if (error) {
521 nfsvno_relpathbuf(ndp);
522 ndp->ni_vp = NULL;
523 ndp->ni_dvp = NULL;
524 ndp->ni_startdir = NULL;
525 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
526 ndp->ni_dvp = NULL;
527 }
528
529 out1:
530 NFSEXITCODE2(error, nd);
531 return (error);
532 }
533
534 /*
535 * Set up a pathname buffer and return a pointer to it and, optionally
536 * set a hash pointer.
537 */
538 void
539 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
540 {
541 struct componentname *cnp = &ndp->ni_cnd;
542
543 cnp->cn_flags |= (NOMACCHECK | HASBUF);
544 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
545 if (hashpp != NULL)
546 *hashpp = NULL;
547 *bufpp = cnp->cn_pnbuf;
548 }
549
550 /*
551 * Release the above path buffer, if not released by nfsvno_namei().
552 */
553 void
554 nfsvno_relpathbuf(struct nameidata *ndp)
555 {
556
557 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
558 panic("nfsrelpath");
559 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
560 ndp->ni_cnd.cn_flags &= ~HASBUF;
561 }
562
563 /*
564 * Readlink vnode op into an mbuf list.
565 */
566 int
567 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
568 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
569 {
570 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
571 struct iovec *ivp = iv;
572 struct uio io, *uiop = &io;
573 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
574 int i, len, tlen, error = 0;
575
576 len = 0;
577 i = 0;
578 while (len < NFS_MAXPATHLEN) {
579 NFSMGET(mp);
580 MCLGET(mp, M_WAITOK);
581 mp->m_len = NFSMSIZ(mp);
582 if (len == 0) {
583 mp3 = mp2 = mp;
584 } else {
585 mp2->m_next = mp;
586 mp2 = mp;
587 }
588 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
589 mp->m_len = NFS_MAXPATHLEN - len;
590 len = NFS_MAXPATHLEN;
591 } else {
592 len += mp->m_len;
593 }
594 ivp->iov_base = mtod(mp, caddr_t);
595 ivp->iov_len = mp->m_len;
596 i++;
597 ivp++;
598 }
599 uiop->uio_iov = iv;
600 uiop->uio_iovcnt = i;
601 uiop->uio_offset = 0;
602 uiop->uio_resid = len;
603 uiop->uio_rw = UIO_READ;
604 uiop->uio_segflg = UIO_SYSSPACE;
605 uiop->uio_td = NULL;
606 error = VOP_READLINK(vp, uiop, cred);
607 if (error) {
608 m_freem(mp3);
609 *lenp = 0;
610 goto out;
611 }
612 if (uiop->uio_resid > 0) {
613 len -= uiop->uio_resid;
614 tlen = NFSM_RNDUP(len);
615 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
616 }
617 *lenp = len;
618 *mpp = mp3;
619 *mpendp = mp;
620
621 out:
622 NFSEXITCODE(error);
623 return (error);
624 }
625
626 /*
627 * Read vnode op call into mbuf list.
628 */
629 int
630 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
631 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
632 {
633 struct mbuf *m;
634 int i;
635 struct iovec *iv;
636 struct iovec *iv2;
637 int error = 0, len, left, siz, tlen, ioflag = 0;
638 struct mbuf *m2 = NULL, *m3;
639 struct uio io, *uiop = &io;
640 struct nfsheur *nh;
641
642 len = left = NFSM_RNDUP(cnt);
643 m3 = NULL;
644 /*
645 * Generate the mbuf list with the uio_iov ref. to it.
646 */
647 i = 0;
648 while (left > 0) {
649 NFSMGET(m);
650 MCLGET(m, M_WAITOK);
651 m->m_len = 0;
652 siz = min(M_TRAILINGSPACE(m), left);
653 left -= siz;
654 i++;
655 if (m3)
656 m2->m_next = m;
657 else
658 m3 = m;
659 m2 = m;
660 }
661 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
662 M_TEMP, M_WAITOK);
663 uiop->uio_iov = iv2 = iv;
664 m = m3;
665 left = len;
666 i = 0;
667 while (left > 0) {
668 if (m == NULL)
669 panic("nfsvno_read iov");
670 siz = min(M_TRAILINGSPACE(m), left);
671 if (siz > 0) {
672 iv->iov_base = mtod(m, caddr_t) + m->m_len;
673 iv->iov_len = siz;
674 m->m_len += siz;
675 left -= siz;
676 iv++;
677 i++;
678 }
679 m = m->m_next;
680 }
681 uiop->uio_iovcnt = i;
682 uiop->uio_offset = off;
683 uiop->uio_resid = len;
684 uiop->uio_rw = UIO_READ;
685 uiop->uio_segflg = UIO_SYSSPACE;
686 uiop->uio_td = NULL;
687 nh = nfsrv_sequential_heuristic(uiop, vp);
688 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
689 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
690 FREE((caddr_t)iv2, M_TEMP);
691 if (error) {
692 m_freem(m3);
693 *mpp = NULL;
694 goto out;
695 }
696 nh->nh_nextoff = uiop->uio_offset;
697 tlen = len - uiop->uio_resid;
698 cnt = cnt < tlen ? cnt : tlen;
699 tlen = NFSM_RNDUP(cnt);
700 if (tlen == 0) {
701 m_freem(m3);
702 m3 = NULL;
703 } else if (len != tlen || tlen != cnt)
704 nfsrv_adj(m3, len - tlen, tlen - cnt);
705 *mpp = m3;
706 *mpendp = m2;
707
708 out:
709 NFSEXITCODE(error);
710 return (error);
711 }
712
713 /*
714 * Write vnode op from an mbuf list.
715 */
716 int
717 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
718 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
719 {
720 struct iovec *ivp;
721 int i, len;
722 struct iovec *iv;
723 int ioflags, error;
724 struct uio io, *uiop = &io;
725 struct nfsheur *nh;
726
727 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
728 M_WAITOK);
729 uiop->uio_iov = iv = ivp;
730 uiop->uio_iovcnt = cnt;
731 i = mtod(mp, caddr_t) + mp->m_len - cp;
732 len = retlen;
733 while (len > 0) {
734 if (mp == NULL)
735 panic("nfsvno_write");
736 if (i > 0) {
737 i = min(i, len);
738 ivp->iov_base = cp;
739 ivp->iov_len = i;
740 ivp++;
741 len -= i;
742 }
743 mp = mp->m_next;
744 if (mp) {
745 i = mp->m_len;
746 cp = mtod(mp, caddr_t);
747 }
748 }
749
750 if (stable == NFSWRITE_UNSTABLE)
751 ioflags = IO_NODELOCKED;
752 else
753 ioflags = (IO_SYNC | IO_NODELOCKED);
754 uiop->uio_resid = retlen;
755 uiop->uio_rw = UIO_WRITE;
756 uiop->uio_segflg = UIO_SYSSPACE;
757 NFSUIOPROC(uiop, p);
758 uiop->uio_offset = off;
759 nh = nfsrv_sequential_heuristic(uiop, vp);
760 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
761 error = VOP_WRITE(vp, uiop, ioflags, cred);
762 if (error == 0)
763 nh->nh_nextoff = uiop->uio_offset;
764 FREE((caddr_t)iv, M_TEMP);
765
766 NFSEXITCODE(error);
767 return (error);
768 }
769
770 /*
771 * Common code for creating a regular file (plus special files for V2).
772 */
773 int
774 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
775 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
776 int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
777 {
778 u_quad_t tempsize;
779 int error;
780
781 error = nd->nd_repstat;
782 if (!error && ndp->ni_vp == NULL) {
783 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
784 vrele(ndp->ni_startdir);
785 error = VOP_CREATE(ndp->ni_dvp,
786 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
787 vput(ndp->ni_dvp);
788 nfsvno_relpathbuf(ndp);
789 if (!error) {
790 if (*exclusive_flagp) {
791 *exclusive_flagp = 0;
792 NFSVNO_ATTRINIT(nvap);
793 nvap->na_atime.tv_sec = cverf[0];
794 nvap->na_atime.tv_nsec = cverf[1];
795 error = VOP_SETATTR(ndp->ni_vp,
796 &nvap->na_vattr, nd->nd_cred);
797 if (error != 0) {
798 vput(ndp->ni_vp);
799 ndp->ni_vp = NULL;
800 error = NFSERR_NOTSUPP;
801 }
802 }
803 }
804 /*
805 * NFS V2 Only. nfsrvd_mknod() does this for V3.
806 * (This implies, just get out on an error.)
807 */
808 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
809 nvap->na_type == VFIFO) {
810 if (nvap->na_type == VCHR && rdev == 0xffffffff)
811 nvap->na_type = VFIFO;
812 if (nvap->na_type != VFIFO &&
813 (error = priv_check_cred(nd->nd_cred,
814 PRIV_VFS_MKNOD_DEV, 0))) {
815 vrele(ndp->ni_startdir);
816 nfsvno_relpathbuf(ndp);
817 vput(ndp->ni_dvp);
818 goto out;
819 }
820 nvap->na_rdev = rdev;
821 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
822 &ndp->ni_cnd, &nvap->na_vattr);
823 vput(ndp->ni_dvp);
824 nfsvno_relpathbuf(ndp);
825 vrele(ndp->ni_startdir);
826 if (error)
827 goto out;
828 } else {
829 vrele(ndp->ni_startdir);
830 nfsvno_relpathbuf(ndp);
831 vput(ndp->ni_dvp);
832 error = ENXIO;
833 goto out;
834 }
835 *vpp = ndp->ni_vp;
836 } else {
837 /*
838 * Handle cases where error is already set and/or
839 * the file exists.
840 * 1 - clean up the lookup
841 * 2 - iff !error and na_size set, truncate it
842 */
843 vrele(ndp->ni_startdir);
844 nfsvno_relpathbuf(ndp);
845 *vpp = ndp->ni_vp;
846 if (ndp->ni_dvp == *vpp)
847 vrele(ndp->ni_dvp);
848 else
849 vput(ndp->ni_dvp);
850 if (!error && nvap->na_size != VNOVAL) {
851 error = nfsvno_accchk(*vpp, VWRITE,
852 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
853 NFSACCCHK_VPISLOCKED, NULL);
854 if (!error) {
855 tempsize = nvap->na_size;
856 NFSVNO_ATTRINIT(nvap);
857 nvap->na_size = tempsize;
858 error = VOP_SETATTR(*vpp,
859 &nvap->na_vattr, nd->nd_cred);
860 }
861 }
862 if (error)
863 vput(*vpp);
864 }
865
866 out:
867 NFSEXITCODE(error);
868 return (error);
869 }
870
871 /*
872 * Do a mknod vnode op.
873 */
874 int
875 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
876 struct thread *p)
877 {
878 int error = 0;
879 enum vtype vtyp;
880
881 vtyp = nvap->na_type;
882 /*
883 * Iff doesn't exist, create it.
884 */
885 if (ndp->ni_vp) {
886 vrele(ndp->ni_startdir);
887 nfsvno_relpathbuf(ndp);
888 vput(ndp->ni_dvp);
889 vrele(ndp->ni_vp);
890 error = EEXIST;
891 goto out;
892 }
893 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
894 vrele(ndp->ni_startdir);
895 nfsvno_relpathbuf(ndp);
896 vput(ndp->ni_dvp);
897 error = NFSERR_BADTYPE;
898 goto out;
899 }
900 if (vtyp == VSOCK) {
901 vrele(ndp->ni_startdir);
902 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
903 &ndp->ni_cnd, &nvap->na_vattr);
904 vput(ndp->ni_dvp);
905 nfsvno_relpathbuf(ndp);
906 } else {
907 if (nvap->na_type != VFIFO &&
908 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
909 vrele(ndp->ni_startdir);
910 nfsvno_relpathbuf(ndp);
911 vput(ndp->ni_dvp);
912 goto out;
913 }
914 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
915 &ndp->ni_cnd, &nvap->na_vattr);
916 vput(ndp->ni_dvp);
917 nfsvno_relpathbuf(ndp);
918 vrele(ndp->ni_startdir);
919 /*
920 * Since VOP_MKNOD returns the ni_vp, I can't
921 * see any reason to do the lookup.
922 */
923 }
924
925 out:
926 NFSEXITCODE(error);
927 return (error);
928 }
929
930 /*
931 * Mkdir vnode op.
932 */
933 int
934 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
935 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
936 {
937 int error = 0;
938
939 if (ndp->ni_vp != NULL) {
940 if (ndp->ni_dvp == ndp->ni_vp)
941 vrele(ndp->ni_dvp);
942 else
943 vput(ndp->ni_dvp);
944 vrele(ndp->ni_vp);
945 nfsvno_relpathbuf(ndp);
946 error = EEXIST;
947 goto out;
948 }
949 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
950 &nvap->na_vattr);
951 vput(ndp->ni_dvp);
952 nfsvno_relpathbuf(ndp);
953
954 out:
955 NFSEXITCODE(error);
956 return (error);
957 }
958
959 /*
960 * symlink vnode op.
961 */
962 int
963 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
964 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
965 struct nfsexstuff *exp)
966 {
967 int error = 0;
968
969 if (ndp->ni_vp) {
970 vrele(ndp->ni_startdir);
971 nfsvno_relpathbuf(ndp);
972 if (ndp->ni_dvp == ndp->ni_vp)
973 vrele(ndp->ni_dvp);
974 else
975 vput(ndp->ni_dvp);
976 vrele(ndp->ni_vp);
977 error = EEXIST;
978 goto out;
979 }
980
981 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
982 &nvap->na_vattr, pathcp);
983 vput(ndp->ni_dvp);
984 vrele(ndp->ni_startdir);
985 nfsvno_relpathbuf(ndp);
986 /*
987 * Although FreeBSD still had the lookup code in
988 * it for 7/current, there doesn't seem to be any
989 * point, since VOP_SYMLINK() returns the ni_vp.
990 * Just vput it for v2.
991 */
992 if (!not_v2 && !error)
993 vput(ndp->ni_vp);
994
995 out:
996 NFSEXITCODE(error);
997 return (error);
998 }
999
1000 /*
1001 * Parse symbolic link arguments.
1002 * This function has an ugly side effect. It will MALLOC() an area for
1003 * the symlink and set iov_base to point to it, only if it succeeds.
1004 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1005 * be FREE'd later.
1006 */
1007 int
1008 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1009 struct thread *p, char **pathcpp, int *lenp)
1010 {
1011 u_int32_t *tl;
1012 char *pathcp = NULL;
1013 int error = 0, len;
1014 struct nfsv2_sattr *sp;
1015
1016 *pathcpp = NULL;
1017 *lenp = 0;
1018 if ((nd->nd_flag & ND_NFSV3) &&
1019 (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1020 goto nfsmout;
1021 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1022 len = fxdr_unsigned(int, *tl);
1023 if (len > NFS_MAXPATHLEN || len <= 0) {
1024 error = EBADRPC;
1025 goto nfsmout;
1026 }
1027 MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1028 error = nfsrv_mtostr(nd, pathcp, len);
1029 if (error)
1030 goto nfsmout;
1031 if (nd->nd_flag & ND_NFSV2) {
1032 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1033 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1034 }
1035 *pathcpp = pathcp;
1036 *lenp = len;
1037 NFSEXITCODE2(0, nd);
1038 return (0);
1039 nfsmout:
1040 if (pathcp)
1041 free(pathcp, M_TEMP);
1042 NFSEXITCODE2(error, nd);
1043 return (error);
1044 }
1045
1046 /*
1047 * Remove a non-directory object.
1048 */
1049 int
1050 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1051 struct thread *p, struct nfsexstuff *exp)
1052 {
1053 struct vnode *vp;
1054 int error = 0;
1055
1056 vp = ndp->ni_vp;
1057 if (vp->v_type == VDIR)
1058 error = NFSERR_ISDIR;
1059 else if (is_v4)
1060 error = nfsrv_checkremove(vp, 1, p);
1061 if (!error)
1062 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1063 if (ndp->ni_dvp == vp)
1064 vrele(ndp->ni_dvp);
1065 else
1066 vput(ndp->ni_dvp);
1067 vput(vp);
1068 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1069 nfsvno_relpathbuf(ndp);
1070 NFSEXITCODE(error);
1071 return (error);
1072 }
1073
1074 /*
1075 * Remove a directory.
1076 */
1077 int
1078 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1079 struct thread *p, struct nfsexstuff *exp)
1080 {
1081 struct vnode *vp;
1082 int error = 0;
1083
1084 vp = ndp->ni_vp;
1085 if (vp->v_type != VDIR) {
1086 error = ENOTDIR;
1087 goto out;
1088 }
1089 /*
1090 * No rmdir "." please.
1091 */
1092 if (ndp->ni_dvp == vp) {
1093 error = EINVAL;
1094 goto out;
1095 }
1096 /*
1097 * The root of a mounted filesystem cannot be deleted.
1098 */
1099 if (vp->v_vflag & VV_ROOT)
1100 error = EBUSY;
1101 out:
1102 if (!error)
1103 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1104 if (ndp->ni_dvp == vp)
1105 vrele(ndp->ni_dvp);
1106 else
1107 vput(ndp->ni_dvp);
1108 vput(vp);
1109 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1110 nfsvno_relpathbuf(ndp);
1111 NFSEXITCODE(error);
1112 return (error);
1113 }
1114
1115 /*
1116 * Rename vnode op.
1117 */
1118 int
1119 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1120 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1121 {
1122 struct vnode *fvp, *tvp, *tdvp;
1123 int error = 0;
1124
1125 fvp = fromndp->ni_vp;
1126 if (ndstat) {
1127 vrele(fromndp->ni_dvp);
1128 vrele(fvp);
1129 error = ndstat;
1130 goto out1;
1131 }
1132 tdvp = tondp->ni_dvp;
1133 tvp = tondp->ni_vp;
1134 if (tvp != NULL) {
1135 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1136 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1137 goto out;
1138 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1139 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1140 goto out;
1141 }
1142 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1143 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1144 goto out;
1145 }
1146
1147 /*
1148 * A rename to '.' or '..' results in a prematurely
1149 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1150 * here.
1151 */
1152 if ((tondp->ni_cnd.cn_namelen == 1 &&
1153 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1154 (tondp->ni_cnd.cn_namelen == 2 &&
1155 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1156 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1157 error = EINVAL;
1158 goto out;
1159 }
1160 }
1161 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1162 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1163 goto out;
1164 }
1165 if (fvp->v_mount != tdvp->v_mount) {
1166 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1167 goto out;
1168 }
1169 if (fvp == tdvp) {
1170 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1171 goto out;
1172 }
1173 if (fvp == tvp) {
1174 /*
1175 * If source and destination are the same, there is nothing to
1176 * do. Set error to -1 to indicate this.
1177 */
1178 error = -1;
1179 goto out;
1180 }
1181 if (ndflag & ND_NFSV4) {
1182 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1183 error = nfsrv_checkremove(fvp, 0, p);
1184 NFSVOPUNLOCK(fvp, 0);
1185 } else
1186 error = EPERM;
1187 if (tvp && !error)
1188 error = nfsrv_checkremove(tvp, 1, p);
1189 } else {
1190 /*
1191 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1192 * that the NFSv4 client won't be confused by the rename.
1193 * Since nfsd_recalldelegation() can only be called on an
1194 * unlocked vnode at this point and fvp is the file that will
1195 * still exist after the rename, just do fvp.
1196 */
1197 nfsd_recalldelegation(fvp, p);
1198 }
1199 out:
1200 if (!error) {
1201 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1202 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1203 &tondp->ni_cnd);
1204 } else {
1205 if (tdvp == tvp)
1206 vrele(tdvp);
1207 else
1208 vput(tdvp);
1209 if (tvp)
1210 vput(tvp);
1211 vrele(fromndp->ni_dvp);
1212 vrele(fvp);
1213 if (error == -1)
1214 error = 0;
1215 }
1216 vrele(tondp->ni_startdir);
1217 nfsvno_relpathbuf(tondp);
1218 out1:
1219 vrele(fromndp->ni_startdir);
1220 nfsvno_relpathbuf(fromndp);
1221 NFSEXITCODE(error);
1222 return (error);
1223 }
1224
1225 /*
1226 * Link vnode op.
1227 */
1228 int
1229 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1230 struct thread *p, struct nfsexstuff *exp)
1231 {
1232 struct vnode *xp;
1233 int error = 0;
1234
1235 xp = ndp->ni_vp;
1236 if (xp != NULL) {
1237 error = EEXIST;
1238 } else {
1239 xp = ndp->ni_dvp;
1240 if (vp->v_mount != xp->v_mount)
1241 error = EXDEV;
1242 }
1243 if (!error) {
1244 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1245 if ((vp->v_iflag & VI_DOOMED) == 0)
1246 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1247 else
1248 error = EPERM;
1249 if (ndp->ni_dvp == vp)
1250 vrele(ndp->ni_dvp);
1251 else
1252 vput(ndp->ni_dvp);
1253 NFSVOPUNLOCK(vp, 0);
1254 } else {
1255 if (ndp->ni_dvp == ndp->ni_vp)
1256 vrele(ndp->ni_dvp);
1257 else
1258 vput(ndp->ni_dvp);
1259 if (ndp->ni_vp)
1260 vrele(ndp->ni_vp);
1261 }
1262 nfsvno_relpathbuf(ndp);
1263 NFSEXITCODE(error);
1264 return (error);
1265 }
1266
1267 /*
1268 * Do the fsync() appropriate for the commit.
1269 */
1270 int
1271 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1272 struct thread *td)
1273 {
1274 int error = 0;
1275
1276 /*
1277 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1278 * file is done. At this time VOP_FSYNC does not accept offset and
1279 * byte count parameters so call VOP_FSYNC the whole file for now.
1280 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1281 * File systems that do not use the buffer cache (as indicated
1282 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1283 */
1284 if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1285 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1286 /*
1287 * Give up and do the whole thing
1288 */
1289 if (vp->v_object &&
1290 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1291 VM_OBJECT_WLOCK(vp->v_object);
1292 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1293 VM_OBJECT_WUNLOCK(vp->v_object);
1294 }
1295 error = VOP_FSYNC(vp, MNT_WAIT, td);
1296 } else {
1297 /*
1298 * Locate and synchronously write any buffers that fall
1299 * into the requested range. Note: we are assuming that
1300 * f_iosize is a power of 2.
1301 */
1302 int iosize = vp->v_mount->mnt_stat.f_iosize;
1303 int iomask = iosize - 1;
1304 struct bufobj *bo;
1305 daddr_t lblkno;
1306
1307 /*
1308 * Align to iosize boundry, super-align to page boundry.
1309 */
1310 if (off & iomask) {
1311 cnt += off & iomask;
1312 off &= ~(u_quad_t)iomask;
1313 }
1314 if (off & PAGE_MASK) {
1315 cnt += off & PAGE_MASK;
1316 off &= ~(u_quad_t)PAGE_MASK;
1317 }
1318 lblkno = off / iosize;
1319
1320 if (vp->v_object &&
1321 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1322 VM_OBJECT_WLOCK(vp->v_object);
1323 vm_object_page_clean(vp->v_object, off, off + cnt,
1324 OBJPC_SYNC);
1325 VM_OBJECT_WUNLOCK(vp->v_object);
1326 }
1327
1328 bo = &vp->v_bufobj;
1329 BO_LOCK(bo);
1330 while (cnt > 0) {
1331 struct buf *bp;
1332
1333 /*
1334 * If we have a buffer and it is marked B_DELWRI we
1335 * have to lock and write it. Otherwise the prior
1336 * write is assumed to have already been committed.
1337 *
1338 * gbincore() can return invalid buffers now so we
1339 * have to check that bit as well (though B_DELWRI
1340 * should not be set if B_INVAL is set there could be
1341 * a race here since we haven't locked the buffer).
1342 */
1343 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1344 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1345 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1346 BO_LOCK(bo);
1347 continue; /* retry */
1348 }
1349 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1350 B_DELWRI) {
1351 bremfree(bp);
1352 bp->b_flags &= ~B_ASYNC;
1353 bwrite(bp);
1354 ++nfs_commit_miss;
1355 } else
1356 BUF_UNLOCK(bp);
1357 BO_LOCK(bo);
1358 }
1359 ++nfs_commit_blks;
1360 if (cnt < iosize)
1361 break;
1362 cnt -= iosize;
1363 ++lblkno;
1364 }
1365 BO_UNLOCK(bo);
1366 }
1367 NFSEXITCODE(error);
1368 return (error);
1369 }
1370
1371 /*
1372 * Statfs vnode op.
1373 */
1374 int
1375 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1376 {
1377 int error;
1378
1379 error = VFS_STATFS(vp->v_mount, sf);
1380 if (error == 0) {
1381 /*
1382 * Since NFS handles these values as unsigned on the
1383 * wire, there is no way to represent negative values,
1384 * so set them to 0. Without this, they will appear
1385 * to be very large positive values for clients like
1386 * Solaris10.
1387 */
1388 if (sf->f_bavail < 0)
1389 sf->f_bavail = 0;
1390 if (sf->f_ffree < 0)
1391 sf->f_ffree = 0;
1392 }
1393 NFSEXITCODE(error);
1394 return (error);
1395 }
1396
1397 /*
1398 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1399 * must handle nfsrv_opencheck() calls after any other access checks.
1400 */
1401 void
1402 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1403 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1404 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1405 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1406 struct nfsexstuff *exp, struct vnode **vpp)
1407 {
1408 struct vnode *vp = NULL;
1409 u_quad_t tempsize;
1410 struct nfsexstuff nes;
1411
1412 if (ndp->ni_vp == NULL)
1413 nd->nd_repstat = nfsrv_opencheck(clientid,
1414 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1415 if (!nd->nd_repstat) {
1416 if (ndp->ni_vp == NULL) {
1417 vrele(ndp->ni_startdir);
1418 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1419 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1420 vput(ndp->ni_dvp);
1421 nfsvno_relpathbuf(ndp);
1422 if (!nd->nd_repstat) {
1423 if (*exclusive_flagp) {
1424 *exclusive_flagp = 0;
1425 NFSVNO_ATTRINIT(nvap);
1426 nvap->na_atime.tv_sec = cverf[0];
1427 nvap->na_atime.tv_nsec = cverf[1];
1428 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1429 &nvap->na_vattr, cred);
1430 if (nd->nd_repstat != 0) {
1431 vput(ndp->ni_vp);
1432 ndp->ni_vp = NULL;
1433 nd->nd_repstat = NFSERR_NOTSUPP;
1434 } else
1435 NFSSETBIT_ATTRBIT(attrbitp,
1436 NFSATTRBIT_TIMEACCESS);
1437 } else {
1438 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1439 aclp, p, attrbitp, exp);
1440 }
1441 }
1442 vp = ndp->ni_vp;
1443 } else {
1444 if (ndp->ni_startdir)
1445 vrele(ndp->ni_startdir);
1446 nfsvno_relpathbuf(ndp);
1447 vp = ndp->ni_vp;
1448 if (create == NFSV4OPEN_CREATE) {
1449 if (ndp->ni_dvp == vp)
1450 vrele(ndp->ni_dvp);
1451 else
1452 vput(ndp->ni_dvp);
1453 }
1454 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1455 if (ndp->ni_cnd.cn_flags & RDONLY)
1456 NFSVNO_SETEXRDONLY(&nes);
1457 else
1458 NFSVNO_EXINIT(&nes);
1459 nd->nd_repstat = nfsvno_accchk(vp,
1460 VWRITE, cred, &nes, p,
1461 NFSACCCHK_NOOVERRIDE,
1462 NFSACCCHK_VPISLOCKED, NULL);
1463 nd->nd_repstat = nfsrv_opencheck(clientid,
1464 stateidp, stp, vp, nd, p, nd->nd_repstat);
1465 if (!nd->nd_repstat) {
1466 tempsize = nvap->na_size;
1467 NFSVNO_ATTRINIT(nvap);
1468 nvap->na_size = tempsize;
1469 nd->nd_repstat = VOP_SETATTR(vp,
1470 &nvap->na_vattr, cred);
1471 }
1472 } else if (vp->v_type == VREG) {
1473 nd->nd_repstat = nfsrv_opencheck(clientid,
1474 stateidp, stp, vp, nd, p, nd->nd_repstat);
1475 }
1476 }
1477 } else {
1478 if (ndp->ni_cnd.cn_flags & HASBUF)
1479 nfsvno_relpathbuf(ndp);
1480 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1481 vrele(ndp->ni_startdir);
1482 if (ndp->ni_dvp == ndp->ni_vp)
1483 vrele(ndp->ni_dvp);
1484 else
1485 vput(ndp->ni_dvp);
1486 if (ndp->ni_vp)
1487 vput(ndp->ni_vp);
1488 }
1489 }
1490 *vpp = vp;
1491
1492 NFSEXITCODE2(0, nd);
1493 }
1494
1495 /*
1496 * Updates the file rev and sets the mtime and ctime
1497 * to the current clock time, returning the va_filerev and va_Xtime
1498 * values.
1499 * Return ESTALE to indicate the vnode is VI_DOOMED.
1500 */
1501 int
1502 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1503 struct ucred *cred, struct thread *p)
1504 {
1505 struct vattr va;
1506
1507 VATTR_NULL(&va);
1508 vfs_timestamp(&va.va_mtime);
1509 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1510 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1511 if ((vp->v_iflag & VI_DOOMED) != 0)
1512 return (ESTALE);
1513 }
1514 (void) VOP_SETATTR(vp, &va, cred);
1515 (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1516 return (0);
1517 }
1518
1519 /*
1520 * Glue routine to nfsv4_fillattr().
1521 */
1522 int
1523 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1524 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1525 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1526 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1527 {
1528 int error;
1529
1530 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1531 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1532 mounted_on_fileno);
1533 NFSEXITCODE2(0, nd);
1534 return (error);
1535 }
1536
1537 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1538 /*
1539 * nfs readdir service
1540 * - mallocs what it thinks is enough to read
1541 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1542 * - calls VOP_READDIR()
1543 * - loops around building the reply
1544 * if the output generated exceeds count break out of loop
1545 * The NFSM_CLGET macro is used here so that the reply will be packed
1546 * tightly in mbuf clusters.
1547 * - it trims out records with d_fileno == 0
1548 * this doesn't matter for Unix clients, but they might confuse clients
1549 * for other os'.
1550 * - it trims out records with d_type == DT_WHT
1551 * these cannot be seen through NFS (unless we extend the protocol)
1552 * The alternate call nfsrvd_readdirplus() does lookups as well.
1553 * PS: The NFS protocol spec. does not clarify what the "count" byte
1554 * argument is a count of.. just name strings and file id's or the
1555 * entire reply rpc or ...
1556 * I tried just file name and id sizes and it confused the Sun client,
1557 * so I am using the full rpc size now. The "paranoia.." comment refers
1558 * to including the status longwords that are not a part of the dir.
1559 * "entry" structures, but are in the rpc.
1560 */
1561 int
1562 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1563 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1564 {
1565 struct dirent *dp;
1566 u_int32_t *tl;
1567 int dirlen;
1568 char *cpos, *cend, *rbuf;
1569 struct nfsvattr at;
1570 int nlen, error = 0, getret = 1;
1571 int siz, cnt, fullsiz, eofflag, ncookies;
1572 u_int64_t off, toff, verf;
1573 u_long *cookies = NULL, *cookiep;
1574 struct uio io;
1575 struct iovec iv;
1576 int is_ufs;
1577
1578 if (nd->nd_repstat) {
1579 nfsrv_postopattr(nd, getret, &at);
1580 goto out;
1581 }
1582 if (nd->nd_flag & ND_NFSV2) {
1583 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1584 off = fxdr_unsigned(u_quad_t, *tl++);
1585 } else {
1586 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1587 off = fxdr_hyper(tl);
1588 tl += 2;
1589 verf = fxdr_hyper(tl);
1590 tl += 2;
1591 }
1592 toff = off;
1593 cnt = fxdr_unsigned(int, *tl);
1594 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1595 cnt = NFS_SRVMAXDATA(nd);
1596 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1597 fullsiz = siz;
1598 if (nd->nd_flag & ND_NFSV3) {
1599 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1600 p, 1);
1601 #if 0
1602 /*
1603 * va_filerev is not sufficient as a cookie verifier,
1604 * since it is not supposed to change when entries are
1605 * removed/added unless that offset cookies returned to
1606 * the client are no longer valid.
1607 */
1608 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1609 nd->nd_repstat = NFSERR_BAD_COOKIE;
1610 #endif
1611 }
1612 if (!nd->nd_repstat && vp->v_type != VDIR)
1613 nd->nd_repstat = NFSERR_NOTDIR;
1614 if (nd->nd_repstat == 0 && cnt == 0) {
1615 if (nd->nd_flag & ND_NFSV2)
1616 /* NFSv2 does not have NFSERR_TOOSMALL */
1617 nd->nd_repstat = EPERM;
1618 else
1619 nd->nd_repstat = NFSERR_TOOSMALL;
1620 }
1621 if (!nd->nd_repstat)
1622 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1623 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1624 NFSACCCHK_VPISLOCKED, NULL);
1625 if (nd->nd_repstat) {
1626 vput(vp);
1627 if (nd->nd_flag & ND_NFSV3)
1628 nfsrv_postopattr(nd, getret, &at);
1629 goto out;
1630 }
1631 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1632 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1633 again:
1634 eofflag = 0;
1635 if (cookies) {
1636 free((caddr_t)cookies, M_TEMP);
1637 cookies = NULL;
1638 }
1639
1640 iv.iov_base = rbuf;
1641 iv.iov_len = siz;
1642 io.uio_iov = &iv;
1643 io.uio_iovcnt = 1;
1644 io.uio_offset = (off_t)off;
1645 io.uio_resid = siz;
1646 io.uio_segflg = UIO_SYSSPACE;
1647 io.uio_rw = UIO_READ;
1648 io.uio_td = NULL;
1649 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1650 &cookies);
1651 off = (u_int64_t)io.uio_offset;
1652 if (io.uio_resid)
1653 siz -= io.uio_resid;
1654
1655 if (!cookies && !nd->nd_repstat)
1656 nd->nd_repstat = NFSERR_PERM;
1657 if (nd->nd_flag & ND_NFSV3) {
1658 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1659 if (!nd->nd_repstat)
1660 nd->nd_repstat = getret;
1661 }
1662
1663 /*
1664 * Handles the failed cases. nd->nd_repstat == 0 past here.
1665 */
1666 if (nd->nd_repstat) {
1667 vput(vp);
1668 free((caddr_t)rbuf, M_TEMP);
1669 if (cookies)
1670 free((caddr_t)cookies, M_TEMP);
1671 if (nd->nd_flag & ND_NFSV3)
1672 nfsrv_postopattr(nd, getret, &at);
1673 goto out;
1674 }
1675 /*
1676 * If nothing read, return eof
1677 * rpc reply
1678 */
1679 if (siz == 0) {
1680 vput(vp);
1681 if (nd->nd_flag & ND_NFSV2) {
1682 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1683 } else {
1684 nfsrv_postopattr(nd, getret, &at);
1685 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1686 txdr_hyper(at.na_filerev, tl);
1687 tl += 2;
1688 }
1689 *tl++ = newnfs_false;
1690 *tl = newnfs_true;
1691 FREE((caddr_t)rbuf, M_TEMP);
1692 FREE((caddr_t)cookies, M_TEMP);
1693 goto out;
1694 }
1695
1696 /*
1697 * Check for degenerate cases of nothing useful read.
1698 * If so go try again
1699 */
1700 cpos = rbuf;
1701 cend = rbuf + siz;
1702 dp = (struct dirent *)cpos;
1703 cookiep = cookies;
1704
1705 /*
1706 * For some reason FreeBSD's ufs_readdir() chooses to back the
1707 * directory offset up to a block boundary, so it is necessary to
1708 * skip over the records that precede the requested offset. This
1709 * requires the assumption that file offset cookies monotonically
1710 * increase.
1711 */
1712 while (cpos < cend && ncookies > 0 &&
1713 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1714 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
1715 cpos += dp->d_reclen;
1716 dp = (struct dirent *)cpos;
1717 cookiep++;
1718 ncookies--;
1719 }
1720 if (cpos >= cend || ncookies == 0) {
1721 siz = fullsiz;
1722 toff = off;
1723 goto again;
1724 }
1725 vput(vp);
1726
1727 /*
1728 * dirlen is the size of the reply, including all XDR and must
1729 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1730 * if the XDR should be included in "count", but to be safe, we do.
1731 * (Include the two booleans at the end of the reply in dirlen now.)
1732 */
1733 if (nd->nd_flag & ND_NFSV3) {
1734 nfsrv_postopattr(nd, getret, &at);
1735 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1736 txdr_hyper(at.na_filerev, tl);
1737 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1738 } else {
1739 dirlen = 2 * NFSX_UNSIGNED;
1740 }
1741
1742 /* Loop through the records and build reply */
1743 while (cpos < cend && ncookies > 0) {
1744 nlen = dp->d_namlen;
1745 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1746 nlen <= NFS_MAXNAMLEN) {
1747 if (nd->nd_flag & ND_NFSV3)
1748 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1749 else
1750 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1751 if (dirlen > cnt) {
1752 eofflag = 0;
1753 break;
1754 }
1755
1756 /*
1757 * Build the directory record xdr from
1758 * the dirent entry.
1759 */
1760 if (nd->nd_flag & ND_NFSV3) {
1761 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1762 *tl++ = newnfs_true;
1763 *tl++ = 0;
1764 } else {
1765 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1766 *tl++ = newnfs_true;
1767 }
1768 *tl = txdr_unsigned(dp->d_fileno);
1769 (void) nfsm_strtom(nd, dp->d_name, nlen);
1770 if (nd->nd_flag & ND_NFSV3) {
1771 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1772 *tl++ = 0;
1773 } else
1774 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1775 *tl = txdr_unsigned(*cookiep);
1776 }
1777 cpos += dp->d_reclen;
1778 dp = (struct dirent *)cpos;
1779 cookiep++;
1780 ncookies--;
1781 }
1782 if (cpos < cend)
1783 eofflag = 0;
1784 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1785 *tl++ = newnfs_false;
1786 if (eofflag)
1787 *tl = newnfs_true;
1788 else
1789 *tl = newnfs_false;
1790 FREE((caddr_t)rbuf, M_TEMP);
1791 FREE((caddr_t)cookies, M_TEMP);
1792
1793 out:
1794 NFSEXITCODE2(0, nd);
1795 return (0);
1796 nfsmout:
1797 vput(vp);
1798 NFSEXITCODE2(error, nd);
1799 return (error);
1800 }
1801
1802 /*
1803 * Readdirplus for V3 and Readdir for V4.
1804 */
1805 int
1806 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1807 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1808 {
1809 struct dirent *dp;
1810 u_int32_t *tl;
1811 int dirlen;
1812 char *cpos, *cend, *rbuf;
1813 struct vnode *nvp;
1814 fhandle_t nfh;
1815 struct nfsvattr nva, at, *nvap = &nva;
1816 struct mbuf *mb0, *mb1;
1817 struct nfsreferral *refp;
1818 int nlen, r, error = 0, getret = 1, usevget = 1;
1819 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1820 caddr_t bpos0, bpos1;
1821 u_int64_t off, toff, verf;
1822 u_long *cookies = NULL, *cookiep;
1823 nfsattrbit_t attrbits, rderrbits, savbits;
1824 struct uio io;
1825 struct iovec iv;
1826 struct componentname cn;
1827 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
1828 struct mount *mp, *new_mp;
1829 uint64_t mounted_on_fileno;
1830
1831 if (nd->nd_repstat) {
1832 nfsrv_postopattr(nd, getret, &at);
1833 goto out;
1834 }
1835 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1836 off = fxdr_hyper(tl);
1837 toff = off;
1838 tl += 2;
1839 verf = fxdr_hyper(tl);
1840 tl += 2;
1841 siz = fxdr_unsigned(int, *tl++);
1842 cnt = fxdr_unsigned(int, *tl);
1843
1844 /*
1845 * Use the server's maximum data transfer size as the upper bound
1846 * on reply datalen.
1847 */
1848 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1849 cnt = NFS_SRVMAXDATA(nd);
1850
1851 /*
1852 * siz is a "hint" of how much directory information (name, fileid,
1853 * cookie) should be in the reply. At least one client "hints" 0,
1854 * so I set it to cnt for that case. I also round it up to the
1855 * next multiple of DIRBLKSIZ.
1856 */
1857 if (siz <= 0)
1858 siz = cnt;
1859 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1860
1861 if (nd->nd_flag & ND_NFSV4) {
1862 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1863 if (error)
1864 goto nfsmout;
1865 NFSSET_ATTRBIT(&savbits, &attrbits);
1866 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1867 NFSZERO_ATTRBIT(&rderrbits);
1868 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1869 } else {
1870 NFSZERO_ATTRBIT(&attrbits);
1871 }
1872 fullsiz = siz;
1873 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1874 if (!nd->nd_repstat) {
1875 if (off && verf != at.na_filerev) {
1876 /*
1877 * va_filerev is not sufficient as a cookie verifier,
1878 * since it is not supposed to change when entries are
1879 * removed/added unless that offset cookies returned to
1880 * the client are no longer valid.
1881 */
1882 #if 0
1883 if (nd->nd_flag & ND_NFSV4) {
1884 nd->nd_repstat = NFSERR_NOTSAME;
1885 } else {
1886 nd->nd_repstat = NFSERR_BAD_COOKIE;
1887 }
1888 #endif
1889 } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1890 nd->nd_repstat = NFSERR_BAD_COOKIE;
1891 }
1892 }
1893 if (!nd->nd_repstat && vp->v_type != VDIR)
1894 nd->nd_repstat = NFSERR_NOTDIR;
1895 if (!nd->nd_repstat && cnt == 0)
1896 nd->nd_repstat = NFSERR_TOOSMALL;
1897 if (!nd->nd_repstat)
1898 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1899 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1900 NFSACCCHK_VPISLOCKED, NULL);
1901 if (nd->nd_repstat) {
1902 vput(vp);
1903 if (nd->nd_flag & ND_NFSV3)
1904 nfsrv_postopattr(nd, getret, &at);
1905 goto out;
1906 }
1907 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1908 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
1909
1910 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1911 again:
1912 eofflag = 0;
1913 if (cookies) {
1914 free((caddr_t)cookies, M_TEMP);
1915 cookies = NULL;
1916 }
1917
1918 iv.iov_base = rbuf;
1919 iv.iov_len = siz;
1920 io.uio_iov = &iv;
1921 io.uio_iovcnt = 1;
1922 io.uio_offset = (off_t)off;
1923 io.uio_resid = siz;
1924 io.uio_segflg = UIO_SYSSPACE;
1925 io.uio_rw = UIO_READ;
1926 io.uio_td = NULL;
1927 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1928 &cookies);
1929 off = (u_int64_t)io.uio_offset;
1930 if (io.uio_resid)
1931 siz -= io.uio_resid;
1932
1933 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1934
1935 if (!cookies && !nd->nd_repstat)
1936 nd->nd_repstat = NFSERR_PERM;
1937 if (!nd->nd_repstat)
1938 nd->nd_repstat = getret;
1939 if (nd->nd_repstat) {
1940 vput(vp);
1941 if (cookies)
1942 free((caddr_t)cookies, M_TEMP);
1943 free((caddr_t)rbuf, M_TEMP);
1944 if (nd->nd_flag & ND_NFSV3)
1945 nfsrv_postopattr(nd, getret, &at);
1946 goto out;
1947 }
1948 /*
1949 * If nothing read, return eof
1950 * rpc reply
1951 */
1952 if (siz == 0) {
1953 vput(vp);
1954 if (nd->nd_flag & ND_NFSV3)
1955 nfsrv_postopattr(nd, getret, &at);
1956 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1957 txdr_hyper(at.na_filerev, tl);
1958 tl += 2;
1959 *tl++ = newnfs_false;
1960 *tl = newnfs_true;
1961 free((caddr_t)cookies, M_TEMP);
1962 free((caddr_t)rbuf, M_TEMP);
1963 goto out;
1964 }
1965
1966 /*
1967 * Check for degenerate cases of nothing useful read.
1968 * If so go try again
1969 */
1970 cpos = rbuf;
1971 cend = rbuf + siz;
1972 dp = (struct dirent *)cpos;
1973 cookiep = cookies;
1974
1975 /*
1976 * For some reason FreeBSD's ufs_readdir() chooses to back the
1977 * directory offset up to a block boundary, so it is necessary to
1978 * skip over the records that precede the requested offset. This
1979 * requires the assumption that file offset cookies monotonically
1980 * increase.
1981 */
1982 while (cpos < cend && ncookies > 0 &&
1983 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1984 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
1985 ((nd->nd_flag & ND_NFSV4) &&
1986 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1987 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1988 cpos += dp->d_reclen;
1989 dp = (struct dirent *)cpos;
1990 cookiep++;
1991 ncookies--;
1992 }
1993 if (cpos >= cend || ncookies == 0) {
1994 siz = fullsiz;
1995 toff = off;
1996 goto again;
1997 }
1998
1999 /*
2000 * Busy the file system so that the mount point won't go away
2001 * and, as such, VFS_VGET() can be used safely.
2002 */
2003 mp = vp->v_mount;
2004 vfs_ref(mp);
2005 NFSVOPUNLOCK(vp, 0);
2006 nd->nd_repstat = vfs_busy(mp, 0);
2007 vfs_rel(mp);
2008 if (nd->nd_repstat != 0) {
2009 vrele(vp);
2010 free(cookies, M_TEMP);
2011 free(rbuf, M_TEMP);
2012 if (nd->nd_flag & ND_NFSV3)
2013 nfsrv_postopattr(nd, getret, &at);
2014 goto out;
2015 }
2016
2017 /*
2018 * For now ZFS requires VOP_LOOKUP as a workaround. Until ino_t is changed
2019 * to 64 bit type a ZFS filesystem with over 1 billion files in it
2020 * will suffer from 64bit -> 32bit truncation.
2021 */
2022 if (is_zfs == 1)
2023 usevget = 0;
2024
2025 cn.cn_nameiop = LOOKUP;
2026 cn.cn_lkflags = LK_SHARED | LK_RETRY;
2027 cn.cn_cred = nd->nd_cred;
2028 cn.cn_thread = p;
2029
2030 /*
2031 * Save this position, in case there is an error before one entry
2032 * is created.
2033 */
2034 mb0 = nd->nd_mb;
2035 bpos0 = nd->nd_bpos;
2036
2037 /*
2038 * Fill in the first part of the reply.
2039 * dirlen is the reply length in bytes and cannot exceed cnt.
2040 * (Include the two booleans at the end of the reply in dirlen now,
2041 * so we recognize when we have exceeded cnt.)
2042 */
2043 if (nd->nd_flag & ND_NFSV3) {
2044 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2045 nfsrv_postopattr(nd, getret, &at);
2046 } else {
2047 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2048 }
2049 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2050 txdr_hyper(at.na_filerev, tl);
2051
2052 /*
2053 * Save this position, in case there is an empty reply needed.
2054 */
2055 mb1 = nd->nd_mb;
2056 bpos1 = nd->nd_bpos;
2057
2058 /* Loop through the records and build reply */
2059 entrycnt = 0;
2060 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2061 nlen = dp->d_namlen;
2062 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2063 nlen <= NFS_MAXNAMLEN &&
2064 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2065 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2066 || (nlen == 1 && dp->d_name[0] != '.'))) {
2067 /*
2068 * Save the current position in the reply, in case
2069 * this entry exceeds cnt.
2070 */
2071 mb1 = nd->nd_mb;
2072 bpos1 = nd->nd_bpos;
2073
2074 /*
2075 * For readdir_and_lookup get the vnode using
2076 * the file number.
2077 */
2078 nvp = NULL;
2079 refp = NULL;
2080 r = 0;
2081 at_root = 0;
2082 needs_unbusy = 0;
2083 new_mp = mp;
2084 mounted_on_fileno = (uint64_t)dp->d_fileno;
2085 if ((nd->nd_flag & ND_NFSV3) ||
2086 NFSNONZERO_ATTRBIT(&savbits)) {
2087 if (nd->nd_flag & ND_NFSV4)
2088 refp = nfsv4root_getreferral(NULL,
2089 vp, dp->d_fileno);
2090 if (refp == NULL) {
2091 if (usevget)
2092 r = VFS_VGET(mp, dp->d_fileno,
2093 LK_SHARED, &nvp);
2094 else
2095 r = EOPNOTSUPP;
2096 if (r == EOPNOTSUPP) {
2097 usevget = 0;
2098 cn.cn_nameptr = dp->d_name;
2099 cn.cn_namelen = nlen;
2100 cn.cn_flags = ISLASTCN |
2101 NOFOLLOW | LOCKLEAF;
2102 if (nlen == 2 &&
2103 dp->d_name[0] == '.' &&
2104 dp->d_name[1] == '.')
2105 cn.cn_flags |=
2106 ISDOTDOT;
2107 if (NFSVOPLOCK(vp, LK_SHARED)
2108 != 0) {
2109 nd->nd_repstat = EPERM;
2110 break;
2111 }
2112 if ((vp->v_vflag & VV_ROOT) != 0
2113 && (cn.cn_flags & ISDOTDOT)
2114 != 0) {
2115 vref(vp);
2116 nvp = vp;
2117 r = 0;
2118 } else {
2119 r = VOP_LOOKUP(vp, &nvp,
2120 &cn);
2121 if (vp != nvp)
2122 NFSVOPUNLOCK(vp,
2123 0);
2124 }
2125 }
2126
2127 /*
2128 * For NFSv4, check to see if nvp is
2129 * a mount point and get the mount
2130 * point vnode, as required.
2131 */
2132 if (r == 0 &&
2133 nfsrv_enable_crossmntpt != 0 &&
2134 (nd->nd_flag & ND_NFSV4) != 0 &&
2135 nvp->v_type == VDIR &&
2136 nvp->v_mountedhere != NULL) {
2137 new_mp = nvp->v_mountedhere;
2138 r = vfs_busy(new_mp, 0);
2139 vput(nvp);
2140 nvp = NULL;
2141 if (r == 0) {
2142 r = VFS_ROOT(new_mp,
2143 LK_SHARED, &nvp);
2144 needs_unbusy = 1;
2145 if (r == 0)
2146 at_root = 1;
2147 }
2148 }
2149 }
2150 if (!r) {
2151 if (refp == NULL &&
2152 ((nd->nd_flag & ND_NFSV3) ||
2153 NFSNONZERO_ATTRBIT(&attrbits))) {
2154 r = nfsvno_getfh(nvp, &nfh, p);
2155 if (!r)
2156 r = nfsvno_getattr(nvp, nvap,
2157 nd->nd_cred, p, 1);
2158 if (r == 0 && is_zfs == 1 &&
2159 nfsrv_enable_crossmntpt != 0 &&
2160 (nd->nd_flag & ND_NFSV4) != 0 &&
2161 nvp->v_type == VDIR &&
2162 vp->v_mount != nvp->v_mount) {
2163 /*
2164 * For a ZFS snapshot, there is a
2165 * pseudo mount that does not set
2166 * v_mountedhere, so it needs to
2167 * be detected via a different
2168 * mount structure.
2169 */
2170 at_root = 1;
2171 if (new_mp == mp)
2172 new_mp = nvp->v_mount;
2173 }
2174 }
2175 } else {
2176 nvp = NULL;
2177 }
2178 if (r) {
2179 if (!NFSISSET_ATTRBIT(&attrbits,
2180 NFSATTRBIT_RDATTRERROR)) {
2181 if (nvp != NULL)
2182 vput(nvp);
2183 if (needs_unbusy != 0)
2184 vfs_unbusy(new_mp);
2185 nd->nd_repstat = r;
2186 break;
2187 }
2188 }
2189 }
2190
2191 /*
2192 * Build the directory record xdr
2193 */
2194 if (nd->nd_flag & ND_NFSV3) {
2195 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2196 *tl++ = newnfs_true;
2197 *tl++ = 0;
2198 *tl = txdr_unsigned(dp->d_fileno);
2199 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2200 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2201 *tl++ = 0;
2202 *tl = txdr_unsigned(*cookiep);
2203 nfsrv_postopattr(nd, 0, nvap);
2204 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2205 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2206 if (nvp != NULL)
2207 vput(nvp);
2208 } else {
2209 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2210 *tl++ = newnfs_true;
2211 *tl++ = 0;
2212 *tl = txdr_unsigned(*cookiep);
2213 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2214 if (nvp != NULL) {
2215 supports_nfsv4acls =
2216 nfs_supportsnfsv4acls(nvp);
2217 NFSVOPUNLOCK(nvp, 0);
2218 } else
2219 supports_nfsv4acls = 0;
2220 if (refp != NULL) {
2221 dirlen += nfsrv_putreferralattr(nd,
2222 &savbits, refp, 0,
2223 &nd->nd_repstat);
2224 if (nd->nd_repstat) {
2225 if (nvp != NULL)
2226 vrele(nvp);
2227 if (needs_unbusy != 0)
2228 vfs_unbusy(new_mp);
2229 break;
2230 }
2231 } else if (r) {
2232 dirlen += nfsvno_fillattr(nd, new_mp,
2233 nvp, nvap, &nfh, r, &rderrbits,
2234 nd->nd_cred, p, isdgram, 0,
2235 supports_nfsv4acls, at_root,
2236 mounted_on_fileno);
2237 } else {
2238 dirlen += nfsvno_fillattr(nd, new_mp,
2239 nvp, nvap, &nfh, r, &attrbits,
2240 nd->nd_cred, p, isdgram, 0,
2241 supports_nfsv4acls, at_root,
2242 mounted_on_fileno);
2243 }
2244 if (nvp != NULL)
2245 vrele(nvp);
2246 dirlen += (3 * NFSX_UNSIGNED);
2247 }
2248 if (needs_unbusy != 0)
2249 vfs_unbusy(new_mp);
2250 if (dirlen <= cnt)
2251 entrycnt++;
2252 }
2253 cpos += dp->d_reclen;
2254 dp = (struct dirent *)cpos;
2255 cookiep++;
2256 ncookies--;
2257 }
2258 vrele(vp);
2259 vfs_unbusy(mp);
2260
2261 /*
2262 * If dirlen > cnt, we must strip off the last entry. If that
2263 * results in an empty reply, report NFSERR_TOOSMALL.
2264 */
2265 if (dirlen > cnt || nd->nd_repstat) {
2266 if (!nd->nd_repstat && entrycnt == 0)
2267 nd->nd_repstat = NFSERR_TOOSMALL;
2268 if (nd->nd_repstat) {
2269 newnfs_trimtrailing(nd, mb0, bpos0);
2270 if (nd->nd_flag & ND_NFSV3)
2271 nfsrv_postopattr(nd, getret, &at);
2272 } else
2273 newnfs_trimtrailing(nd, mb1, bpos1);
2274 eofflag = 0;
2275 } else if (cpos < cend)
2276 eofflag = 0;
2277 if (!nd->nd_repstat) {
2278 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2279 *tl++ = newnfs_false;
2280 if (eofflag)
2281 *tl = newnfs_true;
2282 else
2283 *tl = newnfs_false;
2284 }
2285 FREE((caddr_t)cookies, M_TEMP);
2286 FREE((caddr_t)rbuf, M_TEMP);
2287
2288 out:
2289 NFSEXITCODE2(0, nd);
2290 return (0);
2291 nfsmout:
2292 vput(vp);
2293 NFSEXITCODE2(error, nd);
2294 return (error);
2295 }
2296
2297 /*
2298 * Get the settable attributes out of the mbuf list.
2299 * (Return 0 or EBADRPC)
2300 */
2301 int
2302 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2303 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2304 {
2305 u_int32_t *tl;
2306 struct nfsv2_sattr *sp;
2307 int error = 0, toclient = 0;
2308
2309 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2310 case ND_NFSV2:
2311 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2312 /*
2313 * Some old clients didn't fill in the high order 16bits.
2314 * --> check the low order 2 bytes for 0xffff
2315 */
2316 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2317 nvap->na_mode = nfstov_mode(sp->sa_mode);
2318 if (sp->sa_uid != newnfs_xdrneg1)
2319 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2320 if (sp->sa_gid != newnfs_xdrneg1)
2321 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2322 if (sp->sa_size != newnfs_xdrneg1)
2323 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2324 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2325 #ifdef notyet
2326 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2327 #else
2328 nvap->na_atime.tv_sec =
2329 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2330 nvap->na_atime.tv_nsec = 0;
2331 #endif
2332 }
2333 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2334 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2335 break;
2336 case ND_NFSV3:
2337 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2338 if (*tl == newnfs_true) {
2339 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2340 nvap->na_mode = nfstov_mode(*tl);
2341 }
2342 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2343 if (*tl == newnfs_true) {
2344 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2345 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2346 }
2347 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2348 if (*tl == newnfs_true) {
2349 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2350 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2351 }
2352 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2353 if (*tl == newnfs_true) {
2354 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2355 nvap->na_size = fxdr_hyper(tl);
2356 }
2357 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2358 switch (fxdr_unsigned(int, *tl)) {
2359 case NFSV3SATTRTIME_TOCLIENT:
2360 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2361 fxdr_nfsv3time(tl, &nvap->na_atime);
2362 toclient = 1;
2363 break;
2364 case NFSV3SATTRTIME_TOSERVER:
2365 vfs_timestamp(&nvap->na_atime);
2366 nvap->na_vaflags |= VA_UTIMES_NULL;
2367 break;
2368 };
2369 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2370 switch (fxdr_unsigned(int, *tl)) {
2371 case NFSV3SATTRTIME_TOCLIENT:
2372 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2373 fxdr_nfsv3time(tl, &nvap->na_mtime);
2374 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2375 break;
2376 case NFSV3SATTRTIME_TOSERVER:
2377 vfs_timestamp(&nvap->na_mtime);
2378 if (!toclient)
2379 nvap->na_vaflags |= VA_UTIMES_NULL;
2380 break;
2381 };
2382 break;
2383 case ND_NFSV4:
2384 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2385 };
2386 nfsmout:
2387 NFSEXITCODE2(error, nd);
2388 return (error);
2389 }
2390
2391 /*
2392 * Handle the setable attributes for V4.
2393 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2394 */
2395 int
2396 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2397 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2398 {
2399 u_int32_t *tl;
2400 int attrsum = 0;
2401 int i, j;
2402 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2403 int toclient = 0;
2404 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2405 uid_t uid;
2406 gid_t gid;
2407
2408 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2409 if (error)
2410 goto nfsmout;
2411 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2412 attrsize = fxdr_unsigned(int, *tl);
2413
2414 /*
2415 * Loop around getting the setable attributes. If an unsupported
2416 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2417 */
2418 if (retnotsup) {
2419 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2420 bitpos = NFSATTRBIT_MAX;
2421 } else {
2422 bitpos = 0;
2423 }
2424 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2425 if (attrsum > attrsize) {
2426 error = NFSERR_BADXDR;
2427 goto nfsmout;
2428 }
2429 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2430 switch (bitpos) {
2431 case NFSATTRBIT_SIZE:
2432 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2433 nvap->na_size = fxdr_hyper(tl);
2434 attrsum += NFSX_HYPER;
2435 break;
2436 case NFSATTRBIT_ACL:
2437 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2438 p);
2439 if (error)
2440 goto nfsmout;
2441 if (aceerr && !nd->nd_repstat)
2442 nd->nd_repstat = aceerr;
2443 attrsum += aclsize;
2444 break;
2445 case NFSATTRBIT_ARCHIVE:
2446 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2447 if (!nd->nd_repstat)
2448 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2449 attrsum += NFSX_UNSIGNED;
2450 break;
2451 case NFSATTRBIT_HIDDEN:
2452 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2453 if (!nd->nd_repstat)
2454 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2455 attrsum += NFSX_UNSIGNED;
2456 break;
2457 case NFSATTRBIT_MIMETYPE:
2458 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2459 i = fxdr_unsigned(int, *tl);
2460 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2461 if (error)
2462 goto nfsmout;
2463 if (!nd->nd_repstat)
2464 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2465 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2466 break;
2467 case NFSATTRBIT_MODE:
2468 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2469 nvap->na_mode = nfstov_mode(*tl);
2470 attrsum += NFSX_UNSIGNED;
2471 break;
2472 case NFSATTRBIT_OWNER:
2473 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2474 j = fxdr_unsigned(int, *tl);
2475 if (j < 0) {
2476 error = NFSERR_BADXDR;
2477 goto nfsmout;
2478 }
2479 if (j > NFSV4_SMALLSTR)
2480 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2481 else
2482 cp = namestr;
2483 error = nfsrv_mtostr(nd, cp, j);
2484 if (error) {
2485 if (j > NFSV4_SMALLSTR)
2486 free(cp, M_NFSSTRING);
2487 goto nfsmout;
2488 }
2489 if (!nd->nd_repstat) {
2490 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2491 p);
2492 if (!nd->nd_repstat)
2493 nvap->na_uid = uid;
2494 }
2495 if (j > NFSV4_SMALLSTR)
2496 free(cp, M_NFSSTRING);
2497 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2498 break;
2499 case NFSATTRBIT_OWNERGROUP:
2500 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2501 j = fxdr_unsigned(int, *tl);
2502 if (j < 0) {
2503 error = NFSERR_BADXDR;
2504 goto nfsmout;
2505 }
2506 if (j > NFSV4_SMALLSTR)
2507 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2508 else
2509 cp = namestr;
2510 error = nfsrv_mtostr(nd, cp, j);
2511 if (error) {
2512 if (j > NFSV4_SMALLSTR)
2513 free(cp, M_NFSSTRING);
2514 goto nfsmout;
2515 }
2516 if (!nd->nd_repstat) {
2517 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2518 p);
2519 if (!nd->nd_repstat)
2520 nvap->na_gid = gid;
2521 }
2522 if (j > NFSV4_SMALLSTR)
2523 free(cp, M_NFSSTRING);
2524 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2525 break;
2526 case NFSATTRBIT_SYSTEM:
2527 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2528 if (!nd->nd_repstat)
2529 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2530 attrsum += NFSX_UNSIGNED;
2531 break;
2532 case NFSATTRBIT_TIMEACCESSSET:
2533 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2534 attrsum += NFSX_UNSIGNED;
2535 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2536 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2537 fxdr_nfsv4time(tl, &nvap->na_atime);
2538 toclient = 1;
2539 attrsum += NFSX_V4TIME;
2540 } else {
2541 vfs_timestamp(&nvap->na_atime);
2542 nvap->na_vaflags |= VA_UTIMES_NULL;
2543 }
2544 break;
2545 case NFSATTRBIT_TIMEBACKUP:
2546 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2547 if (!nd->nd_repstat)
2548 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2549 attrsum += NFSX_V4TIME;
2550 break;
2551 case NFSATTRBIT_TIMECREATE:
2552 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2553 if (!nd->nd_repstat)
2554 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2555 attrsum += NFSX_V4TIME;
2556 break;
2557 case NFSATTRBIT_TIMEMODIFYSET:
2558 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2559 attrsum += NFSX_UNSIGNED;
2560 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2561 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2562 fxdr_nfsv4time(tl, &nvap->na_mtime);
2563 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2564 attrsum += NFSX_V4TIME;
2565 } else {
2566 vfs_timestamp(&nvap->na_mtime);
2567 if (!toclient)
2568 nvap->na_vaflags |= VA_UTIMES_NULL;
2569 }
2570 break;
2571 default:
2572 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2573 /*
2574 * set bitpos so we drop out of the loop.
2575 */
2576 bitpos = NFSATTRBIT_MAX;
2577 break;
2578 };
2579 }
2580
2581 /*
2582 * some clients pad the attrlist, so we need to skip over the
2583 * padding.
2584 */
2585 if (attrsum > attrsize) {
2586 error = NFSERR_BADXDR;
2587 } else {
2588 attrsize = NFSM_RNDUP(attrsize);
2589 if (attrsum < attrsize)
2590 error = nfsm_advance(nd, attrsize - attrsum, -1);
2591 }
2592 nfsmout:
2593 NFSEXITCODE2(error, nd);
2594 return (error);
2595 }
2596
2597 /*
2598 * Check/setup export credentials.
2599 */
2600 int
2601 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2602 struct ucred *credanon)
2603 {
2604 int error = 0;
2605
2606 /*
2607 * Check/setup credentials.
2608 */
2609 if (nd->nd_flag & ND_GSS)
2610 exp->nes_exflag &= ~MNT_EXPORTANON;
2611
2612 /*
2613 * Check to see if the operation is allowed for this security flavor.
2614 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2615 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2616 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2617 */
2618 if (nfsvno_testexp(nd, exp) &&
2619 nd->nd_procnum != NFSV4OP_SECINFO &&
2620 nd->nd_procnum != NFSPROC_FSINFO) {
2621 if (nd->nd_flag & ND_NFSV4)
2622 error = NFSERR_WRONGSEC;
2623 else
2624 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2625 goto out;
2626 }
2627
2628 /*
2629 * Check to see if the file system is exported V4 only.
2630 */
2631 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2632 error = NFSERR_PROGNOTV4;
2633 goto out;
2634 }
2635
2636 /*
2637 * Now, map the user credentials.
2638 * (Note that ND_AUTHNONE will only be set for an NFSv3
2639 * Fsinfo RPC. If set for anything else, this code might need
2640 * to change.)
2641 */
2642 if (NFSVNO_EXPORTED(exp)) {
2643 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
2644 NFSVNO_EXPORTANON(exp) ||
2645 (nd->nd_flag & ND_AUTHNONE) != 0) {
2646 nd->nd_cred->cr_uid = credanon->cr_uid;
2647 nd->nd_cred->cr_gid = credanon->cr_gid;
2648 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2649 credanon->cr_groups);
2650 } else if ((nd->nd_flag & ND_GSS) == 0) {
2651 /*
2652 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
2653 * if there is a replacement credential with a group
2654 * list set up by "nfsuserd -manage-gids".
2655 * If there is no replacement, nfsrv_getgrpscred()
2656 * simply returns its argument.
2657 */
2658 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
2659 }
2660 }
2661
2662 out:
2663 NFSEXITCODE2(error, nd);
2664 return (error);
2665 }
2666
2667 /*
2668 * Check exports.
2669 */
2670 int
2671 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2672 struct ucred **credp)
2673 {
2674 int i, error, *secflavors;
2675
2676 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2677 &exp->nes_numsecflavor, &secflavors);
2678 if (error) {
2679 if (nfs_rootfhset) {
2680 exp->nes_exflag = 0;
2681 exp->nes_numsecflavor = 0;
2682 error = 0;
2683 }
2684 } else {
2685 /* Copy the security flavors. */
2686 for (i = 0; i < exp->nes_numsecflavor; i++)
2687 exp->nes_secflavors[i] = secflavors[i];
2688 }
2689 NFSEXITCODE(error);
2690 return (error);
2691 }
2692
2693 /*
2694 * Get a vnode for a file handle and export stuff.
2695 */
2696 int
2697 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2698 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2699 struct ucred **credp)
2700 {
2701 int i, error, *secflavors;
2702
2703 *credp = NULL;
2704 exp->nes_numsecflavor = 0;
2705 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2706 if (error != 0)
2707 /* Make sure the server replies ESTALE to the client. */
2708 error = ESTALE;
2709 if (nam && !error) {
2710 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2711 &exp->nes_numsecflavor, &secflavors);
2712 if (error) {
2713 if (nfs_rootfhset) {
2714 exp->nes_exflag = 0;
2715 exp->nes_numsecflavor = 0;
2716 error = 0;
2717 } else {
2718 vput(*vpp);
2719 }
2720 } else {
2721 /* Copy the security flavors. */
2722 for (i = 0; i < exp->nes_numsecflavor; i++)
2723 exp->nes_secflavors[i] = secflavors[i];
2724 }
2725 }
2726 NFSEXITCODE(error);
2727 return (error);
2728 }
2729
2730 /*
2731 * nfsd_fhtovp() - convert a fh to a vnode ptr
2732 * - look up fsid in mount list (if not found ret error)
2733 * - get vp and export rights by calling nfsvno_fhtovp()
2734 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2735 * for AUTH_SYS
2736 * - if mpp != NULL, return the mount point so that it can
2737 * be used for vn_finished_write() by the caller
2738 */
2739 void
2740 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2741 struct vnode **vpp, struct nfsexstuff *exp,
2742 struct mount **mpp, int startwrite, struct thread *p)
2743 {
2744 struct mount *mp;
2745 struct ucred *credanon;
2746 fhandle_t *fhp;
2747
2748 fhp = (fhandle_t *)nfp->nfsrvfh_data;
2749 /*
2750 * Check for the special case of the nfsv4root_fh.
2751 */
2752 mp = vfs_busyfs(&fhp->fh_fsid);
2753 if (mpp != NULL)
2754 *mpp = mp;
2755 if (mp == NULL) {
2756 *vpp = NULL;
2757 nd->nd_repstat = ESTALE;
2758 goto out;
2759 }
2760
2761 if (startwrite) {
2762 vn_start_write(NULL, mpp, V_WAIT);
2763 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2764 lktype = LK_EXCLUSIVE;
2765 }
2766 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2767 &credanon);
2768 vfs_unbusy(mp);
2769
2770 /*
2771 * For NFSv4 without a pseudo root fs, unexported file handles
2772 * can be returned, so that Lookup works everywhere.
2773 */
2774 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2775 !(nd->nd_flag & ND_NFSV4)) {
2776 vput(*vpp);
2777 nd->nd_repstat = EACCES;
2778 }
2779
2780 /*
2781 * Personally, I've never seen any point in requiring a
2782 * reserved port#, since only in the rare case where the
2783 * clients are all boxes with secure system priviledges,
2784 * does it provide any enhanced security, but... some people
2785 * believe it to be useful and keep putting this code back in.
2786 * (There is also some "security checker" out there that
2787 * complains if the nfs server doesn't enforce this.)
2788 * However, note the following:
2789 * RFC3530 (NFSv4) specifies that a reserved port# not be
2790 * required.
2791 * RFC2623 recommends that, if a reserved port# is checked for,
2792 * that there be a way to turn that off--> ifdef'd.
2793 */
2794 #ifdef NFS_REQRSVPORT
2795 if (!nd->nd_repstat) {
2796 struct sockaddr_in *saddr;
2797 struct sockaddr_in6 *saddr6;
2798
2799 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2800 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2801 if (!(nd->nd_flag & ND_NFSV4) &&
2802 ((saddr->sin_family == AF_INET &&
2803 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2804 (saddr6->sin6_family == AF_INET6 &&
2805 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2806 vput(*vpp);
2807 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2808 }
2809 }
2810 #endif /* NFS_REQRSVPORT */
2811
2812 /*
2813 * Check/setup credentials.
2814 */
2815 if (!nd->nd_repstat) {
2816 nd->nd_saveduid = nd->nd_cred->cr_uid;
2817 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2818 if (nd->nd_repstat)
2819 vput(*vpp);
2820 }
2821 if (credanon != NULL)
2822 crfree(credanon);
2823 if (nd->nd_repstat) {
2824 if (startwrite)
2825 vn_finished_write(mp);
2826 *vpp = NULL;
2827 if (mpp != NULL)
2828 *mpp = NULL;
2829 }
2830
2831 out:
2832 NFSEXITCODE2(0, nd);
2833 }
2834
2835 /*
2836 * glue for fp.
2837 */
2838 static int
2839 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2840 {
2841 struct filedesc *fdp;
2842 struct file *fp;
2843 int error = 0;
2844
2845 fdp = p->td_proc->p_fd;
2846 if (fd < 0 || fd >= fdp->fd_nfiles ||
2847 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2848 error = EBADF;
2849 goto out;
2850 }
2851 *fpp = fp;
2852
2853 out:
2854 NFSEXITCODE(error);
2855 return (error);
2856 }
2857
2858 /*
2859 * Called from nfssvc() to update the exports list. Just call
2860 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2861 * in the mount list.
2862 */
2863 int
2864 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2865 {
2866 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2867 int error = 0;
2868 struct nameidata nd;
2869 fhandle_t fh;
2870
2871 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2872 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2873 nfs_rootfhset = 0;
2874 else if (error == 0) {
2875 if (nfsexargp->fspec == NULL) {
2876 error = EPERM;
2877 goto out;
2878 }
2879 /*
2880 * If fspec != NULL, this is the v4root path.
2881 */
2882 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2883 nfsexargp->fspec, p);
2884 if ((error = namei(&nd)) != 0)
2885 goto out;
2886 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2887 vrele(nd.ni_vp);
2888 if (!error) {
2889 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2890 NFSBCOPY((caddr_t)&fh,
2891 nfs_rootfh.nfsrvfh_data,
2892 sizeof (fhandle_t));
2893 nfs_rootfhset = 1;
2894 }
2895 }
2896
2897 out:
2898 NFSEXITCODE(error);
2899 return (error);
2900 }
2901
2902 /*
2903 * This function needs to test to see if the system is near its limit
2904 * for memory allocation via malloc() or mget() and return True iff
2905 * either of these resources are near their limit.
2906 * XXX (For now, this is just a stub.)
2907 */
2908 int nfsrv_testmalloclimit = 0;
2909 int
2910 nfsrv_mallocmget_limit(void)
2911 {
2912 static int printmesg = 0;
2913 static int testval = 1;
2914
2915 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2916 if ((printmesg++ % 100) == 0)
2917 printf("nfsd: malloc/mget near limit\n");
2918 return (1);
2919 }
2920 return (0);
2921 }
2922
2923 /*
2924 * BSD specific initialization of a mount point.
2925 */
2926 void
2927 nfsd_mntinit(void)
2928 {
2929 static int inited = 0;
2930
2931 if (inited)
2932 return;
2933 inited = 1;
2934 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2935 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2936 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2937 nfsv4root_mnt.mnt_export = NULL;
2938 TAILQ_INIT(&nfsv4root_opt);
2939 TAILQ_INIT(&nfsv4root_newopt);
2940 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2941 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2942 nfsv4root_mnt.mnt_nvnodelistsize = 0;
2943 nfsv4root_mnt.mnt_activevnodelistsize = 0;
2944 }
2945
2946 /*
2947 * Get a vnode for a file handle, without checking exports, etc.
2948 */
2949 struct vnode *
2950 nfsvno_getvp(fhandle_t *fhp)
2951 {
2952 struct mount *mp;
2953 struct vnode *vp;
2954 int error;
2955
2956 mp = vfs_busyfs(&fhp->fh_fsid);
2957 if (mp == NULL)
2958 return (NULL);
2959 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2960 vfs_unbusy(mp);
2961 if (error)
2962 return (NULL);
2963 return (vp);
2964 }
2965
2966 /*
2967 * Do a local VOP_ADVLOCK().
2968 */
2969 int
2970 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2971 u_int64_t end, struct thread *td)
2972 {
2973 int error = 0;
2974 struct flock fl;
2975 u_int64_t tlen;
2976
2977 if (nfsrv_dolocallocks == 0)
2978 goto out;
2979 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
2980
2981 fl.l_whence = SEEK_SET;
2982 fl.l_type = ftype;
2983 fl.l_start = (off_t)first;
2984 if (end == NFS64BITSSET) {
2985 fl.l_len = 0;
2986 } else {
2987 tlen = end - first;
2988 fl.l_len = (off_t)tlen;
2989 }
2990 /*
2991 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2992 * values for all calls, so that all locks will be held by the
2993 * nfsd server. (The nfsd server handles conflicts between the
2994 * various clients.)
2995 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2996 * bytes, so it can't be put in l_sysid.
2997 */
2998 if (nfsv4_sysid == 0)
2999 nfsv4_sysid = nlm_acquire_next_sysid();
3000 fl.l_pid = (pid_t)0;
3001 fl.l_sysid = (int)nfsv4_sysid;
3002
3003 if (ftype == F_UNLCK)
3004 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3005 (F_POSIX | F_REMOTE));
3006 else
3007 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3008 (F_POSIX | F_REMOTE));
3009
3010 out:
3011 NFSEXITCODE(error);
3012 return (error);
3013 }
3014
3015 /*
3016 * Check the nfsv4 root exports.
3017 */
3018 int
3019 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3020 {
3021 struct ucred *credanon;
3022 int exflags, error = 0, numsecflavor, *secflavors, i;
3023
3024 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3025 &credanon, &numsecflavor, &secflavors);
3026 if (error) {
3027 error = NFSERR_PROGUNAVAIL;
3028 goto out;
3029 }
3030 if (credanon != NULL)
3031 crfree(credanon);
3032 for (i = 0; i < numsecflavor; i++) {
3033 if (secflavors[i] == AUTH_SYS)
3034 nd->nd_flag |= ND_EXAUTHSYS;
3035 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3036 nd->nd_flag |= ND_EXGSS;
3037 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3038 nd->nd_flag |= ND_EXGSSINTEGRITY;
3039 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3040 nd->nd_flag |= ND_EXGSSPRIVACY;
3041 }
3042
3043 out:
3044 NFSEXITCODE(error);
3045 return (error);
3046 }
3047
3048 /*
3049 * Nfs server psuedo system call for the nfsd's
3050 */
3051 /*
3052 * MPSAFE
3053 */
3054 static int
3055 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3056 {
3057 struct file *fp;
3058 struct nfsd_addsock_args sockarg;
3059 struct nfsd_nfsd_args nfsdarg;
3060 cap_rights_t rights;
3061 int error;
3062
3063 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3064 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3065 if (error)
3066 goto out;
3067 /*
3068 * Since we don't know what rights might be required,
3069 * pretend that we need them all. It is better to be too
3070 * careful than too reckless.
3071 */
3072 error = fget(td, sockarg.sock,
3073 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3074 if (error != 0)
3075 goto out;
3076 if (fp->f_type != DTYPE_SOCKET) {
3077 fdrop(fp, td);
3078 error = EPERM;
3079 goto out;
3080 }
3081 error = nfsrvd_addsock(fp);
3082 fdrop(fp, td);
3083 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3084 if (uap->argp == NULL) {
3085 error = EINVAL;
3086 goto out;
3087 }
3088 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3089 sizeof (nfsdarg));
3090 if (error)
3091 goto out;
3092 error = nfsrvd_nfsd(td, &nfsdarg);
3093 } else {
3094 error = nfssvc_srvcall(td, uap, td->td_ucred);
3095 }
3096
3097 out:
3098 NFSEXITCODE(error);
3099 return (error);
3100 }
3101
3102 static int
3103 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3104 {
3105 struct nfsex_args export;
3106 struct file *fp = NULL;
3107 int stablefd, len;
3108 struct nfsd_clid adminrevoke;
3109 struct nfsd_dumplist dumplist;
3110 struct nfsd_dumpclients *dumpclients;
3111 struct nfsd_dumplocklist dumplocklist;
3112 struct nfsd_dumplocks *dumplocks;
3113 struct nameidata nd;
3114 vnode_t vp;
3115 int error = EINVAL, igotlock;
3116 struct proc *procp;
3117 static int suspend_nfsd = 0;
3118
3119 if (uap->flag & NFSSVC_PUBLICFH) {
3120 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3121 sizeof (fhandle_t));
3122 error = copyin(uap->argp,
3123 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3124 if (!error)
3125 nfs_pubfhset = 1;
3126 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3127 error = copyin(uap->argp,(caddr_t)&export,
3128 sizeof (struct nfsex_args));
3129 if (!error)
3130 error = nfsrv_v4rootexport(&export, cred, p);
3131 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3132 nfs_pubfhset = 0;
3133 error = 0;
3134 } else if (uap->flag & NFSSVC_STABLERESTART) {
3135 error = copyin(uap->argp, (caddr_t)&stablefd,
3136 sizeof (int));
3137 if (!error)
3138 error = fp_getfvp(p, stablefd, &fp, &vp);
3139 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3140 error = EBADF;
3141 if (!error && newnfs_numnfsd != 0)
3142 error = EPERM;
3143 if (!error) {
3144 nfsrv_stablefirst.nsf_fp = fp;
3145 nfsrv_setupstable(p);
3146 }
3147 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3148 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3149 sizeof (struct nfsd_clid));
3150 if (!error)
3151 error = nfsrv_adminrevoke(&adminrevoke, p);
3152 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3153 error = copyin(uap->argp, (caddr_t)&dumplist,
3154 sizeof (struct nfsd_dumplist));
3155 if (!error && (dumplist.ndl_size < 1 ||
3156 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3157 error = EPERM;
3158 if (!error) {
3159 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3160 dumpclients = (struct nfsd_dumpclients *)malloc(len,
3161 M_TEMP, M_WAITOK);
3162 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3163 error = copyout(dumpclients,
3164 CAST_USER_ADDR_T(dumplist.ndl_list), len);
3165 free((caddr_t)dumpclients, M_TEMP);
3166 }
3167 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3168 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3169 sizeof (struct nfsd_dumplocklist));
3170 if (!error && (dumplocklist.ndllck_size < 1 ||
3171 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3172 error = EPERM;
3173 if (!error)
3174 error = nfsrv_lookupfilename(&nd,
3175 dumplocklist.ndllck_fname, p);
3176 if (!error) {
3177 len = sizeof (struct nfsd_dumplocks) *
3178 dumplocklist.ndllck_size;
3179 dumplocks = (struct nfsd_dumplocks *)malloc(len,
3180 M_TEMP, M_WAITOK);
3181 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3182 dumplocklist.ndllck_size, p);
3183 vput(nd.ni_vp);
3184 error = copyout(dumplocks,
3185 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3186 free((caddr_t)dumplocks, M_TEMP);
3187 }
3188 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3189 procp = p->td_proc;
3190 PROC_LOCK(procp);
3191 nfsd_master_pid = procp->p_pid;
3192 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3193 nfsd_master_start = procp->p_stats->p_start;
3194 nfsd_master_proc = procp;
3195 PROC_UNLOCK(procp);
3196 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3197 NFSLOCKV4ROOTMUTEX();
3198 if (suspend_nfsd == 0) {
3199 /* Lock out all nfsd threads */
3200 do {
3201 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3202 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3203 } while (igotlock == 0 && suspend_nfsd == 0);
3204 suspend_nfsd = 1;
3205 }
3206 NFSUNLOCKV4ROOTMUTEX();
3207 error = 0;
3208 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3209 NFSLOCKV4ROOTMUTEX();
3210 if (suspend_nfsd != 0) {
3211 nfsv4_unlock(&nfsd_suspend_lock, 0);
3212 suspend_nfsd = 0;
3213 }
3214 NFSUNLOCKV4ROOTMUTEX();
3215 error = 0;
3216 }
3217
3218 NFSEXITCODE(error);
3219 return (error);
3220 }
3221
3222 /*
3223 * Check exports.
3224 * Returns 0 if ok, 1 otherwise.
3225 */
3226 int
3227 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3228 {
3229 int i;
3230
3231 /*
3232 * This seems odd, but allow the case where the security flavor
3233 * list is empty. This happens when NFSv4 is traversing non-exported
3234 * file systems. Exported file systems should always have a non-empty
3235 * security flavor list.
3236 */
3237 if (exp->nes_numsecflavor == 0)
3238 return (0);
3239
3240 for (i = 0; i < exp->nes_numsecflavor; i++) {
3241 /*
3242 * The tests for privacy and integrity must be first,
3243 * since ND_GSS is set for everything but AUTH_SYS.
3244 */
3245 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3246 (nd->nd_flag & ND_GSSPRIVACY))
3247 return (0);
3248 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3249 (nd->nd_flag & ND_GSSINTEGRITY))
3250 return (0);
3251 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3252 (nd->nd_flag & ND_GSS))
3253 return (0);
3254 if (exp->nes_secflavors[i] == AUTH_SYS &&
3255 (nd->nd_flag & ND_GSS) == 0)
3256 return (0);
3257 }
3258 return (1);
3259 }
3260
3261 /*
3262 * Calculate a hash value for the fid in a file handle.
3263 */
3264 uint32_t
3265 nfsrv_hashfh(fhandle_t *fhp)
3266 {
3267 uint32_t hashval;
3268
3269 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3270 return (hashval);
3271 }
3272
3273 /*
3274 * Calculate a hash value for the sessionid.
3275 */
3276 uint32_t
3277 nfsrv_hashsessionid(uint8_t *sessionid)
3278 {
3279 uint32_t hashval;
3280
3281 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3282 return (hashval);
3283 }
3284
3285 /*
3286 * Signal the userland master nfsd to backup the stable restart file.
3287 */
3288 void
3289 nfsrv_backupstable(void)
3290 {
3291 struct proc *procp;
3292
3293 if (nfsd_master_proc != NULL) {
3294 procp = pfind(nfsd_master_pid);
3295 /* Try to make sure it is the correct process. */
3296 if (procp == nfsd_master_proc &&
3297 procp->p_stats->p_start.tv_sec ==
3298 nfsd_master_start.tv_sec &&
3299 procp->p_stats->p_start.tv_usec ==
3300 nfsd_master_start.tv_usec &&
3301 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3302 kern_psignal(procp, SIGUSR2);
3303 else
3304 nfsd_master_proc = NULL;
3305
3306 if (procp != NULL)
3307 PROC_UNLOCK(procp);
3308 }
3309 }
3310
3311 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3312
3313 /*
3314 * Called once to initialize data structures...
3315 */
3316 static int
3317 nfsd_modevent(module_t mod, int type, void *data)
3318 {
3319 int error = 0, i;
3320 static int loaded = 0;
3321
3322 switch (type) {
3323 case MOD_LOAD:
3324 if (loaded)
3325 goto out;
3326 newnfs_portinit();
3327 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3328 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
3329 MTX_DEF);
3330 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
3331 MTX_DEF);
3332 }
3333 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
3334 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
3335 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
3336 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3337 nfsrvd_initcache();
3338 nfsd_init();
3339 NFSD_LOCK();
3340 nfsrvd_init(0);
3341 NFSD_UNLOCK();
3342 nfsd_mntinit();
3343 #ifdef VV_DISABLEDELEG
3344 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3345 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3346 #endif
3347 nfsd_call_servertimer = nfsrv_servertimer;
3348 nfsd_call_nfsd = nfssvc_nfsd;
3349 loaded = 1;
3350 break;
3351
3352 case MOD_UNLOAD:
3353 if (newnfs_numnfsd != 0) {
3354 error = EBUSY;
3355 break;
3356 }
3357
3358 #ifdef VV_DISABLEDELEG
3359 vn_deleg_ops.vndeleg_recall = NULL;
3360 vn_deleg_ops.vndeleg_disable = NULL;
3361 #endif
3362 nfsd_call_servertimer = NULL;
3363 nfsd_call_nfsd = NULL;
3364
3365 /* Clean out all NFSv4 state. */
3366 nfsrv_throwawayallstate(curthread);
3367
3368 /* Clean the NFS server reply cache */
3369 nfsrvd_cleancache();
3370
3371 /* Free up the krpc server pool. */
3372 if (nfsrvd_pool != NULL)
3373 svcpool_destroy(nfsrvd_pool);
3374
3375 /* and get rid of the locks */
3376 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3377 mtx_destroy(&nfsrchash_table[i].mtx);
3378 mtx_destroy(&nfsrcahash_table[i].mtx);
3379 }
3380 mtx_destroy(&nfsrc_udpmtx);
3381 mtx_destroy(&nfs_v4root_mutex);
3382 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3383 for (i = 0; i < nfsrv_sessionhashsize; i++)
3384 mtx_destroy(&nfssessionhash[i].mtx);
3385 lockdestroy(&nfsv4root_mnt.mnt_explock);
3386 free(nfsclienthash, M_NFSDCLIENT);
3387 free(nfslockhash, M_NFSDLOCKFILE);
3388 free(nfssessionhash, M_NFSDSESSION);
3389 loaded = 0;
3390 break;
3391 default:
3392 error = EOPNOTSUPP;
3393 break;
3394 }
3395
3396 out:
3397 NFSEXITCODE(error);
3398 return (error);
3399 }
3400 static moduledata_t nfsd_mod = {
3401 "nfsd",
3402 nfsd_modevent,
3403 NULL,
3404 };
3405 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3406
3407 /* So that loader and kldload(2) can find us, wherever we are.. */
3408 MODULE_VERSION(nfsd, 1);
3409 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3410 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3411 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3412 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3413 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3414
Cache object: afcaec5051821c9f3282bcbc8d240b26
|