1 /* $NetBSD: nfs_nfsdport.c,v 1.1.1.1 2013/09/30 07:19:57 dholland Exp $ */
2 /*-
3 * Copyright (c) 1989, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Rick Macklem at The University of Guelph.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 */
34
35 #include <sys/cdefs.h>
36 /* __FBSDID("FreeBSD: head/sys/fs/nfsserver/nfs_nfsdport.c 255219 2013-09-05 00:09:56Z pjd "); */
37 __RCSID("$NetBSD: nfs_nfsdport.c,v 1.1.1.1 2013/09/30 07:19:57 dholland Exp $");
38
39 #include <sys/capability.h>
40
41 /*
42 * Functions that perform the vfs operations required by the routines in
43 * nfsd_serv.c. It is hoped that this change will make the server more
44 * portable.
45 */
46
47 #include <fs/nfs/nfsport.h>
48 #include <sys/hash.h>
49 #include <sys/sysctl.h>
50 #include <nlm/nlm_prot.h>
51 #include <nlm/nlm.h>
52
53 FEATURE(nfsd, "NFSv4 server");
54
55 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
56 extern int nfsrv_useacl;
57 extern int newnfs_numnfsd;
58 extern struct mount nfsv4root_mnt;
59 extern struct nfsrv_stablefirst nfsrv_stablefirst;
60 extern void (*nfsd_call_servertimer)(void);
61 extern SVCPOOL *nfsrvd_pool;
62 extern struct nfsv4lock nfsd_suspend_lock;
63 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
64 NFSDLOCKMUTEX;
65 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
66 struct mtx nfsrc_udpmtx;
67 struct mtx nfs_v4root_mutex;
68 struct nfsrvfh nfs_rootfh, nfs_pubfh;
69 int nfs_pubfhset = 0, nfs_rootfhset = 0;
70 struct proc *nfsd_master_proc = NULL;
71 static pid_t nfsd_master_pid = (pid_t)-1;
72 static char nfsd_master_comm[MAXCOMLEN + 1];
73 static struct timeval nfsd_master_start;
74 static uint32_t nfsv4_sysid = 0;
75
76 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
77 struct ucred *);
78
79 int nfsrv_enable_crossmntpt = 1;
80 static int nfs_commit_blks;
81 static int nfs_commit_miss;
82 extern int nfsrv_issuedelegs;
83 extern int nfsrv_dolocallocks;
84
85 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW, 0, "New NFS server");
86 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
87 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
88 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
89 0, "");
90 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
91 0, "");
92 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
93 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
94 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
95 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
96
97 #define MAX_REORDERED_RPC 16
98 #define NUM_HEURISTIC 1031
99 #define NHUSE_INIT 64
100 #define NHUSE_INC 16
101 #define NHUSE_MAX 2048
102
103 static struct nfsheur {
104 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
105 off_t nh_nextoff; /* next offset for sequential detection */
106 int nh_use; /* use count for selection */
107 int nh_seqcount; /* heuristic */
108 } nfsheur[NUM_HEURISTIC];
109
110
111 /*
112 * Heuristic to detect sequential operation.
113 */
114 static struct nfsheur *
nfsrv_sequential_heuristic(struct uio * uio,struct vnode * vp)115 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
116 {
117 struct nfsheur *nh;
118 int hi, try;
119
120 /* Locate best candidate. */
121 try = 32;
122 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
123 nh = &nfsheur[hi];
124 while (try--) {
125 if (nfsheur[hi].nh_vp == vp) {
126 nh = &nfsheur[hi];
127 break;
128 }
129 if (nfsheur[hi].nh_use > 0)
130 --nfsheur[hi].nh_use;
131 hi = (hi + 1) % NUM_HEURISTIC;
132 if (nfsheur[hi].nh_use < nh->nh_use)
133 nh = &nfsheur[hi];
134 }
135
136 /* Initialize hint if this is a new file. */
137 if (nh->nh_vp != vp) {
138 nh->nh_vp = vp;
139 nh->nh_nextoff = uio->uio_offset;
140 nh->nh_use = NHUSE_INIT;
141 if (uio->uio_offset == 0)
142 nh->nh_seqcount = 4;
143 else
144 nh->nh_seqcount = 1;
145 }
146
147 /* Calculate heuristic. */
148 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
149 uio->uio_offset == nh->nh_nextoff) {
150 /* See comments in vfs_vnops.c:sequential_heuristic(). */
151 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
152 if (nh->nh_seqcount > IO_SEQMAX)
153 nh->nh_seqcount = IO_SEQMAX;
154 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
155 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
156 /* Probably a reordered RPC, leave seqcount alone. */
157 } else if (nh->nh_seqcount > 1) {
158 nh->nh_seqcount /= 2;
159 } else {
160 nh->nh_seqcount = 0;
161 }
162 nh->nh_use += NHUSE_INC;
163 if (nh->nh_use > NHUSE_MAX)
164 nh->nh_use = NHUSE_MAX;
165 return (nh);
166 }
167
168 /*
169 * Get attributes into nfsvattr structure.
170 */
171 int
nfsvno_getattr(struct vnode * vp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p,int vpislocked)172 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
173 struct thread *p, int vpislocked)
174 {
175 int error, lockedit = 0;
176
177 if (vpislocked == 0) {
178 /*
179 * When vpislocked == 0, the vnode is either exclusively
180 * locked by this thread or not locked by this thread.
181 * As such, shared lock it, if not exclusively locked.
182 */
183 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
184 lockedit = 1;
185 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
186 }
187 }
188 error = VOP_GETATTR(vp, &nvap->na_vattr, cred);
189 if (lockedit != 0)
190 NFSVOPUNLOCK(vp, 0);
191
192 NFSEXITCODE(error);
193 return (error);
194 }
195
196 /*
197 * Get a file handle for a vnode.
198 */
199 int
nfsvno_getfh(struct vnode * vp,fhandle_t * fhp,struct thread * p)200 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
201 {
202 int error;
203
204 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
205 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
206 error = VOP_VPTOFH(vp, &fhp->fh_fid);
207
208 NFSEXITCODE(error);
209 return (error);
210 }
211
212 /*
213 * Perform access checking for vnodes obtained from file handles that would
214 * refer to files already opened by a Unix client. You cannot just use
215 * vn_writechk() and VOP_ACCESSX() for two reasons.
216 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
217 * case.
218 * 2 - The owner is to be given access irrespective of mode bits for some
219 * operations, so that processes that chmod after opening a file don't
220 * break.
221 */
222 int
nfsvno_accchk(struct vnode * vp,accmode_t accmode,struct ucred * cred,struct nfsexstuff * exp,struct thread * p,int override,int vpislocked,u_int32_t * supportedtypep)223 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
224 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
225 u_int32_t *supportedtypep)
226 {
227 struct vattr vattr;
228 int error = 0, getret = 0;
229
230 if (vpislocked == 0) {
231 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
232 error = EPERM;
233 goto out;
234 }
235 }
236 if (accmode & VWRITE) {
237 /* Just vn_writechk() changed to check rdonly */
238 /*
239 * Disallow write attempts on read-only file systems;
240 * unless the file is a socket or a block or character
241 * device resident on the file system.
242 */
243 if (NFSVNO_EXRDONLY(exp) ||
244 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
245 switch (vp->v_type) {
246 case VREG:
247 case VDIR:
248 case VLNK:
249 error = EROFS;
250 default:
251 break;
252 }
253 }
254 /*
255 * If there's shared text associated with
256 * the inode, try to free it up once. If
257 * we fail, we can't allow writing.
258 */
259 if (VOP_IS_TEXT(vp) && error == 0)
260 error = ETXTBSY;
261 }
262 if (error != 0) {
263 if (vpislocked == 0)
264 NFSVOPUNLOCK(vp, 0);
265 goto out;
266 }
267
268 /*
269 * Should the override still be applied when ACLs are enabled?
270 */
271 error = VOP_ACCESSX(vp, accmode, cred, p);
272 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
273 /*
274 * Try again with VEXPLICIT_DENY, to see if the test for
275 * deletion is supported.
276 */
277 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
278 if (error == 0) {
279 if (vp->v_type == VDIR) {
280 accmode &= ~(VDELETE | VDELETE_CHILD);
281 accmode |= VWRITE;
282 error = VOP_ACCESSX(vp, accmode, cred, p);
283 } else if (supportedtypep != NULL) {
284 *supportedtypep &= ~NFSACCESS_DELETE;
285 }
286 }
287 }
288
289 /*
290 * Allow certain operations for the owner (reads and writes
291 * on files that are already open).
292 */
293 if (override != NFSACCCHK_NOOVERRIDE &&
294 (error == EPERM || error == EACCES)) {
295 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
296 error = 0;
297 else if (override & NFSACCCHK_ALLOWOWNER) {
298 getret = VOP_GETATTR(vp, &vattr, cred);
299 if (getret == 0 && cred->cr_uid == vattr.va_uid)
300 error = 0;
301 }
302 }
303 if (vpislocked == 0)
304 NFSVOPUNLOCK(vp, 0);
305
306 out:
307 NFSEXITCODE(error);
308 return (error);
309 }
310
311 /*
312 * Set attribute(s) vnop.
313 */
314 int
nfsvno_setattr(struct vnode * vp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)315 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
316 struct thread *p, struct nfsexstuff *exp)
317 {
318 int error;
319
320 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
321 NFSEXITCODE(error);
322 return (error);
323 }
324
325 /*
326 * Set up nameidata for a lookup() call and do it.
327 */
328 int
nfsvno_namei(struct nfsrv_descript * nd,struct nameidata * ndp,struct vnode * dp,int islocked,struct nfsexstuff * exp,struct thread * p,struct vnode ** retdirp)329 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
330 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
331 struct vnode **retdirp)
332 {
333 struct componentname *cnp = &ndp->ni_cnd;
334 int i;
335 struct iovec aiov;
336 struct uio auio;
337 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
338 int error = 0, crossmnt;
339 char *cp;
340
341 *retdirp = NULL;
342 cnp->cn_nameptr = cnp->cn_pnbuf;
343 ndp->ni_strictrelative = 0;
344 /*
345 * Extract and set starting directory.
346 */
347 if (dp->v_type != VDIR) {
348 if (islocked)
349 vput(dp);
350 else
351 vrele(dp);
352 nfsvno_relpathbuf(ndp);
353 error = ENOTDIR;
354 goto out1;
355 }
356 if (islocked)
357 NFSVOPUNLOCK(dp, 0);
358 VREF(dp);
359 *retdirp = dp;
360 if (NFSVNO_EXRDONLY(exp))
361 cnp->cn_flags |= RDONLY;
362 ndp->ni_segflg = UIO_SYSSPACE;
363 crossmnt = 1;
364
365 if (nd->nd_flag & ND_PUBLOOKUP) {
366 ndp->ni_loopcnt = 0;
367 if (cnp->cn_pnbuf[0] == '/') {
368 vrele(dp);
369 /*
370 * Check for degenerate pathnames here, since lookup()
371 * panics on them.
372 */
373 for (i = 1; i < ndp->ni_pathlen; i++)
374 if (cnp->cn_pnbuf[i] != '/')
375 break;
376 if (i == ndp->ni_pathlen) {
377 error = NFSERR_ACCES;
378 goto out;
379 }
380 dp = rootvnode;
381 VREF(dp);
382 }
383 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
384 (nd->nd_flag & ND_NFSV4) == 0) {
385 /*
386 * Only cross mount points for NFSv4 when doing a
387 * mount while traversing the file system above
388 * the mount point, unless nfsrv_enable_crossmntpt is set.
389 */
390 cnp->cn_flags |= NOCROSSMOUNT;
391 crossmnt = 0;
392 }
393
394 /*
395 * Initialize for scan, set ni_startdir and bump ref on dp again
396 * because lookup() will dereference ni_startdir.
397 */
398
399 cnp->cn_thread = p;
400 ndp->ni_startdir = dp;
401 ndp->ni_rootdir = rootvnode;
402 ndp->ni_topdir = NULL;
403
404 if (!lockleaf)
405 cnp->cn_flags |= LOCKLEAF;
406 for (;;) {
407 cnp->cn_nameptr = cnp->cn_pnbuf;
408 /*
409 * Call lookup() to do the real work. If an error occurs,
410 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
411 * we do not have to dereference anything before returning.
412 * In either case ni_startdir will be dereferenced and NULLed
413 * out.
414 */
415 error = lookup(ndp);
416 if (error)
417 break;
418
419 /*
420 * Check for encountering a symbolic link. Trivial
421 * termination occurs if no symlink encountered.
422 */
423 if ((cnp->cn_flags & ISSYMLINK) == 0) {
424 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
425 nfsvno_relpathbuf(ndp);
426 if (ndp->ni_vp && !lockleaf)
427 NFSVOPUNLOCK(ndp->ni_vp, 0);
428 break;
429 }
430
431 /*
432 * Validate symlink
433 */
434 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
435 NFSVOPUNLOCK(ndp->ni_dvp, 0);
436 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
437 error = EINVAL;
438 goto badlink2;
439 }
440
441 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
442 error = ELOOP;
443 goto badlink2;
444 }
445 if (ndp->ni_pathlen > 1)
446 cp = uma_zalloc(namei_zone, M_WAITOK);
447 else
448 cp = cnp->cn_pnbuf;
449 aiov.iov_base = cp;
450 aiov.iov_len = MAXPATHLEN;
451 auio.uio_iov = &aiov;
452 auio.uio_iovcnt = 1;
453 auio.uio_offset = 0;
454 auio.uio_rw = UIO_READ;
455 auio.uio_segflg = UIO_SYSSPACE;
456 auio.uio_td = NULL;
457 auio.uio_resid = MAXPATHLEN;
458 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
459 if (error) {
460 badlink1:
461 if (ndp->ni_pathlen > 1)
462 uma_zfree(namei_zone, cp);
463 badlink2:
464 vrele(ndp->ni_dvp);
465 vput(ndp->ni_vp);
466 break;
467 }
468 linklen = MAXPATHLEN - auio.uio_resid;
469 if (linklen == 0) {
470 error = ENOENT;
471 goto badlink1;
472 }
473 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
474 error = ENAMETOOLONG;
475 goto badlink1;
476 }
477
478 /*
479 * Adjust or replace path
480 */
481 if (ndp->ni_pathlen > 1) {
482 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
483 uma_zfree(namei_zone, cnp->cn_pnbuf);
484 cnp->cn_pnbuf = cp;
485 } else
486 cnp->cn_pnbuf[linklen] = '\0';
487 ndp->ni_pathlen += linklen;
488
489 /*
490 * Cleanup refs for next loop and check if root directory
491 * should replace current directory. Normally ni_dvp
492 * becomes the new base directory and is cleaned up when
493 * we loop. Explicitly null pointers after invalidation
494 * to clarify operation.
495 */
496 vput(ndp->ni_vp);
497 ndp->ni_vp = NULL;
498
499 if (cnp->cn_pnbuf[0] == '/') {
500 vrele(ndp->ni_dvp);
501 ndp->ni_dvp = ndp->ni_rootdir;
502 VREF(ndp->ni_dvp);
503 }
504 ndp->ni_startdir = ndp->ni_dvp;
505 ndp->ni_dvp = NULL;
506 }
507 if (!lockleaf)
508 cnp->cn_flags &= ~LOCKLEAF;
509
510 out:
511 if (error) {
512 nfsvno_relpathbuf(ndp);
513 ndp->ni_vp = NULL;
514 ndp->ni_dvp = NULL;
515 ndp->ni_startdir = NULL;
516 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
517 ndp->ni_dvp = NULL;
518 }
519
520 out1:
521 NFSEXITCODE2(error, nd);
522 return (error);
523 }
524
525 /*
526 * Set up a pathname buffer and return a pointer to it and, optionally
527 * set a hash pointer.
528 */
529 void
nfsvno_setpathbuf(struct nameidata * ndp,char ** bufpp,u_long ** hashpp)530 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
531 {
532 struct componentname *cnp = &ndp->ni_cnd;
533
534 cnp->cn_flags |= (NOMACCHECK | HASBUF);
535 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
536 if (hashpp != NULL)
537 *hashpp = NULL;
538 *bufpp = cnp->cn_pnbuf;
539 }
540
541 /*
542 * Release the above path buffer, if not released by nfsvno_namei().
543 */
544 void
nfsvno_relpathbuf(struct nameidata * ndp)545 nfsvno_relpathbuf(struct nameidata *ndp)
546 {
547
548 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
549 panic("nfsrelpath");
550 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
551 ndp->ni_cnd.cn_flags &= ~HASBUF;
552 }
553
554 /*
555 * Readlink vnode op into an mbuf list.
556 */
557 int
nfsvno_readlink(struct vnode * vp,struct ucred * cred,struct thread * p,struct mbuf ** mpp,struct mbuf ** mpendp,int * lenp)558 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
559 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
560 {
561 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
562 struct iovec *ivp = iv;
563 struct uio io, *uiop = &io;
564 struct mbuf *mp, *mp2 = NULL, *mp3 = NULL;
565 int i, len, tlen, error = 0;
566
567 len = 0;
568 i = 0;
569 while (len < NFS_MAXPATHLEN) {
570 NFSMGET(mp);
571 MCLGET(mp, M_WAITOK);
572 mp->m_len = NFSMSIZ(mp);
573 if (len == 0) {
574 mp3 = mp2 = mp;
575 } else {
576 mp2->m_next = mp;
577 mp2 = mp;
578 }
579 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
580 mp->m_len = NFS_MAXPATHLEN - len;
581 len = NFS_MAXPATHLEN;
582 } else {
583 len += mp->m_len;
584 }
585 ivp->iov_base = mtod(mp, caddr_t);
586 ivp->iov_len = mp->m_len;
587 i++;
588 ivp++;
589 }
590 uiop->uio_iov = iv;
591 uiop->uio_iovcnt = i;
592 uiop->uio_offset = 0;
593 uiop->uio_resid = len;
594 uiop->uio_rw = UIO_READ;
595 uiop->uio_segflg = UIO_SYSSPACE;
596 uiop->uio_td = NULL;
597 error = VOP_READLINK(vp, uiop, cred);
598 if (error) {
599 m_freem(mp3);
600 *lenp = 0;
601 goto out;
602 }
603 if (uiop->uio_resid > 0) {
604 len -= uiop->uio_resid;
605 tlen = NFSM_RNDUP(len);
606 nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen, tlen - len);
607 }
608 *lenp = len;
609 *mpp = mp3;
610 *mpendp = mp;
611
612 out:
613 NFSEXITCODE(error);
614 return (error);
615 }
616
617 /*
618 * Read vnode op call into mbuf list.
619 */
620 int
nfsvno_read(struct vnode * vp,off_t off,int cnt,struct ucred * cred,struct thread * p,struct mbuf ** mpp,struct mbuf ** mpendp)621 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
622 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
623 {
624 struct mbuf *m;
625 int i;
626 struct iovec *iv;
627 struct iovec *iv2;
628 int error = 0, len, left, siz, tlen, ioflag = 0;
629 struct mbuf *m2 = NULL, *m3;
630 struct uio io, *uiop = &io;
631 struct nfsheur *nh;
632
633 len = left = NFSM_RNDUP(cnt);
634 m3 = NULL;
635 /*
636 * Generate the mbuf list with the uio_iov ref. to it.
637 */
638 i = 0;
639 while (left > 0) {
640 NFSMGET(m);
641 MCLGET(m, M_WAITOK);
642 m->m_len = 0;
643 siz = min(M_TRAILINGSPACE(m), left);
644 left -= siz;
645 i++;
646 if (m3)
647 m2->m_next = m;
648 else
649 m3 = m;
650 m2 = m;
651 }
652 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
653 M_TEMP, M_WAITOK);
654 uiop->uio_iov = iv2 = iv;
655 m = m3;
656 left = len;
657 i = 0;
658 while (left > 0) {
659 if (m == NULL)
660 panic("nfsvno_read iov");
661 siz = min(M_TRAILINGSPACE(m), left);
662 if (siz > 0) {
663 iv->iov_base = mtod(m, caddr_t) + m->m_len;
664 iv->iov_len = siz;
665 m->m_len += siz;
666 left -= siz;
667 iv++;
668 i++;
669 }
670 m = m->m_next;
671 }
672 uiop->uio_iovcnt = i;
673 uiop->uio_offset = off;
674 uiop->uio_resid = len;
675 uiop->uio_rw = UIO_READ;
676 uiop->uio_segflg = UIO_SYSSPACE;
677 nh = nfsrv_sequential_heuristic(uiop, vp);
678 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
679 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
680 FREE((caddr_t)iv2, M_TEMP);
681 if (error) {
682 m_freem(m3);
683 *mpp = NULL;
684 goto out;
685 }
686 nh->nh_nextoff = uiop->uio_offset;
687 tlen = len - uiop->uio_resid;
688 cnt = cnt < tlen ? cnt : tlen;
689 tlen = NFSM_RNDUP(cnt);
690 if (tlen == 0) {
691 m_freem(m3);
692 m3 = NULL;
693 } else if (len != tlen || tlen != cnt)
694 nfsrv_adj(m3, len - tlen, tlen - cnt);
695 *mpp = m3;
696 *mpendp = m2;
697
698 out:
699 NFSEXITCODE(error);
700 return (error);
701 }
702
703 /*
704 * Write vnode op from an mbuf list.
705 */
706 int
nfsvno_write(struct vnode * vp,off_t off,int retlen,int cnt,int stable,struct mbuf * mp,char * cp,struct ucred * cred,struct thread * p)707 nfsvno_write(struct vnode *vp, off_t off, int retlen, int cnt, int stable,
708 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
709 {
710 struct iovec *ivp;
711 int i, len;
712 struct iovec *iv;
713 int ioflags, error;
714 struct uio io, *uiop = &io;
715 struct nfsheur *nh;
716
717 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
718 M_WAITOK);
719 uiop->uio_iov = iv = ivp;
720 uiop->uio_iovcnt = cnt;
721 i = mtod(mp, caddr_t) + mp->m_len - cp;
722 len = retlen;
723 while (len > 0) {
724 if (mp == NULL)
725 panic("nfsvno_write");
726 if (i > 0) {
727 i = min(i, len);
728 ivp->iov_base = cp;
729 ivp->iov_len = i;
730 ivp++;
731 len -= i;
732 }
733 mp = mp->m_next;
734 if (mp) {
735 i = mp->m_len;
736 cp = mtod(mp, caddr_t);
737 }
738 }
739
740 if (stable == NFSWRITE_UNSTABLE)
741 ioflags = IO_NODELOCKED;
742 else
743 ioflags = (IO_SYNC | IO_NODELOCKED);
744 uiop->uio_resid = retlen;
745 uiop->uio_rw = UIO_WRITE;
746 uiop->uio_segflg = UIO_SYSSPACE;
747 NFSUIOPROC(uiop, p);
748 uiop->uio_offset = off;
749 nh = nfsrv_sequential_heuristic(uiop, vp);
750 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
751 error = VOP_WRITE(vp, uiop, ioflags, cred);
752 if (error == 0)
753 nh->nh_nextoff = uiop->uio_offset;
754 FREE((caddr_t)iv, M_TEMP);
755
756 NFSEXITCODE(error);
757 return (error);
758 }
759
760 /*
761 * Common code for creating a regular file (plus special files for V2).
762 */
763 int
nfsvno_createsub(struct nfsrv_descript * nd,struct nameidata * ndp,struct vnode ** vpp,struct nfsvattr * nvap,int * exclusive_flagp,int32_t * cverf,NFSDEV_T rdev,struct thread * p,struct nfsexstuff * exp)764 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
765 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
766 int32_t *cverf, NFSDEV_T rdev, struct thread *p, struct nfsexstuff *exp)
767 {
768 u_quad_t tempsize;
769 int error;
770
771 error = nd->nd_repstat;
772 if (!error && ndp->ni_vp == NULL) {
773 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
774 vrele(ndp->ni_startdir);
775 error = VOP_CREATE(ndp->ni_dvp,
776 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
777 vput(ndp->ni_dvp);
778 nfsvno_relpathbuf(ndp);
779 if (!error) {
780 if (*exclusive_flagp) {
781 *exclusive_flagp = 0;
782 NFSVNO_ATTRINIT(nvap);
783 nvap->na_atime.tv_sec = cverf[0];
784 nvap->na_atime.tv_nsec = cverf[1];
785 error = VOP_SETATTR(ndp->ni_vp,
786 &nvap->na_vattr, nd->nd_cred);
787 }
788 }
789 /*
790 * NFS V2 Only. nfsrvd_mknod() does this for V3.
791 * (This implies, just get out on an error.)
792 */
793 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
794 nvap->na_type == VFIFO) {
795 if (nvap->na_type == VCHR && rdev == 0xffffffff)
796 nvap->na_type = VFIFO;
797 if (nvap->na_type != VFIFO &&
798 (error = priv_check_cred(nd->nd_cred,
799 PRIV_VFS_MKNOD_DEV, 0))) {
800 vrele(ndp->ni_startdir);
801 nfsvno_relpathbuf(ndp);
802 vput(ndp->ni_dvp);
803 goto out;
804 }
805 nvap->na_rdev = rdev;
806 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
807 &ndp->ni_cnd, &nvap->na_vattr);
808 vput(ndp->ni_dvp);
809 nfsvno_relpathbuf(ndp);
810 vrele(ndp->ni_startdir);
811 if (error)
812 goto out;
813 } else {
814 vrele(ndp->ni_startdir);
815 nfsvno_relpathbuf(ndp);
816 vput(ndp->ni_dvp);
817 error = ENXIO;
818 goto out;
819 }
820 *vpp = ndp->ni_vp;
821 } else {
822 /*
823 * Handle cases where error is already set and/or
824 * the file exists.
825 * 1 - clean up the lookup
826 * 2 - iff !error and na_size set, truncate it
827 */
828 vrele(ndp->ni_startdir);
829 nfsvno_relpathbuf(ndp);
830 *vpp = ndp->ni_vp;
831 if (ndp->ni_dvp == *vpp)
832 vrele(ndp->ni_dvp);
833 else
834 vput(ndp->ni_dvp);
835 if (!error && nvap->na_size != VNOVAL) {
836 error = nfsvno_accchk(*vpp, VWRITE,
837 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
838 NFSACCCHK_VPISLOCKED, NULL);
839 if (!error) {
840 tempsize = nvap->na_size;
841 NFSVNO_ATTRINIT(nvap);
842 nvap->na_size = tempsize;
843 error = VOP_SETATTR(*vpp,
844 &nvap->na_vattr, nd->nd_cred);
845 }
846 }
847 if (error)
848 vput(*vpp);
849 }
850
851 out:
852 NFSEXITCODE(error);
853 return (error);
854 }
855
856 /*
857 * Do a mknod vnode op.
858 */
859 int
nfsvno_mknod(struct nameidata * ndp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p)860 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
861 struct thread *p)
862 {
863 int error = 0;
864 enum vtype vtyp;
865
866 vtyp = nvap->na_type;
867 /*
868 * Iff doesn't exist, create it.
869 */
870 if (ndp->ni_vp) {
871 vrele(ndp->ni_startdir);
872 nfsvno_relpathbuf(ndp);
873 vput(ndp->ni_dvp);
874 vrele(ndp->ni_vp);
875 error = EEXIST;
876 goto out;
877 }
878 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
879 vrele(ndp->ni_startdir);
880 nfsvno_relpathbuf(ndp);
881 vput(ndp->ni_dvp);
882 error = NFSERR_BADTYPE;
883 goto out;
884 }
885 if (vtyp == VSOCK) {
886 vrele(ndp->ni_startdir);
887 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
888 &ndp->ni_cnd, &nvap->na_vattr);
889 vput(ndp->ni_dvp);
890 nfsvno_relpathbuf(ndp);
891 } else {
892 if (nvap->na_type != VFIFO &&
893 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV, 0))) {
894 vrele(ndp->ni_startdir);
895 nfsvno_relpathbuf(ndp);
896 vput(ndp->ni_dvp);
897 goto out;
898 }
899 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
900 &ndp->ni_cnd, &nvap->na_vattr);
901 vput(ndp->ni_dvp);
902 nfsvno_relpathbuf(ndp);
903 vrele(ndp->ni_startdir);
904 /*
905 * Since VOP_MKNOD returns the ni_vp, I can't
906 * see any reason to do the lookup.
907 */
908 }
909
910 out:
911 NFSEXITCODE(error);
912 return (error);
913 }
914
915 /*
916 * Mkdir vnode op.
917 */
918 int
nfsvno_mkdir(struct nameidata * ndp,struct nfsvattr * nvap,uid_t saved_uid,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)919 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
920 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
921 {
922 int error = 0;
923
924 if (ndp->ni_vp != NULL) {
925 if (ndp->ni_dvp == ndp->ni_vp)
926 vrele(ndp->ni_dvp);
927 else
928 vput(ndp->ni_dvp);
929 vrele(ndp->ni_vp);
930 nfsvno_relpathbuf(ndp);
931 error = EEXIST;
932 goto out;
933 }
934 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
935 &nvap->na_vattr);
936 vput(ndp->ni_dvp);
937 nfsvno_relpathbuf(ndp);
938
939 out:
940 NFSEXITCODE(error);
941 return (error);
942 }
943
944 /*
945 * symlink vnode op.
946 */
947 int
nfsvno_symlink(struct nameidata * ndp,struct nfsvattr * nvap,char * pathcp,int pathlen,int not_v2,uid_t saved_uid,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)948 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
949 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
950 struct nfsexstuff *exp)
951 {
952 int error = 0;
953
954 if (ndp->ni_vp) {
955 vrele(ndp->ni_startdir);
956 nfsvno_relpathbuf(ndp);
957 if (ndp->ni_dvp == ndp->ni_vp)
958 vrele(ndp->ni_dvp);
959 else
960 vput(ndp->ni_dvp);
961 vrele(ndp->ni_vp);
962 error = EEXIST;
963 goto out;
964 }
965
966 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
967 &nvap->na_vattr, pathcp);
968 vput(ndp->ni_dvp);
969 vrele(ndp->ni_startdir);
970 nfsvno_relpathbuf(ndp);
971 /*
972 * Although FreeBSD still had the lookup code in
973 * it for 7/current, there doesn't seem to be any
974 * point, since VOP_SYMLINK() returns the ni_vp.
975 * Just vput it for v2.
976 */
977 if (!not_v2 && !error)
978 vput(ndp->ni_vp);
979
980 out:
981 NFSEXITCODE(error);
982 return (error);
983 }
984
985 /*
986 * Parse symbolic link arguments.
987 * This function has an ugly side effect. It will MALLOC() an area for
988 * the symlink and set iov_base to point to it, only if it succeeds.
989 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
990 * be FREE'd later.
991 */
992 int
nfsvno_getsymlink(struct nfsrv_descript * nd,struct nfsvattr * nvap,struct thread * p,char ** pathcpp,int * lenp)993 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
994 struct thread *p, char **pathcpp, int *lenp)
995 {
996 u_int32_t *tl;
997 char *pathcp = NULL;
998 int error = 0, len;
999 struct nfsv2_sattr *sp;
1000
1001 *pathcpp = NULL;
1002 *lenp = 0;
1003 if ((nd->nd_flag & ND_NFSV3) &&
1004 (error = nfsrv_sattr(nd, nvap, NULL, NULL, p)))
1005 goto nfsmout;
1006 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1007 len = fxdr_unsigned(int, *tl);
1008 if (len > NFS_MAXPATHLEN || len <= 0) {
1009 error = EBADRPC;
1010 goto nfsmout;
1011 }
1012 MALLOC(pathcp, caddr_t, len + 1, M_TEMP, M_WAITOK);
1013 error = nfsrv_mtostr(nd, pathcp, len);
1014 if (error)
1015 goto nfsmout;
1016 if (nd->nd_flag & ND_NFSV2) {
1017 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1018 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1019 }
1020 *pathcpp = pathcp;
1021 *lenp = len;
1022 NFSEXITCODE2(0, nd);
1023 return (0);
1024 nfsmout:
1025 if (pathcp)
1026 free(pathcp, M_TEMP);
1027 NFSEXITCODE2(error, nd);
1028 return (error);
1029 }
1030
1031 /*
1032 * Remove a non-directory object.
1033 */
1034 int
nfsvno_removesub(struct nameidata * ndp,int is_v4,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1035 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1036 struct thread *p, struct nfsexstuff *exp)
1037 {
1038 struct vnode *vp;
1039 int error = 0;
1040
1041 vp = ndp->ni_vp;
1042 if (vp->v_type == VDIR)
1043 error = NFSERR_ISDIR;
1044 else if (is_v4)
1045 error = nfsrv_checkremove(vp, 1, p);
1046 if (!error)
1047 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1048 if (ndp->ni_dvp == vp)
1049 vrele(ndp->ni_dvp);
1050 else
1051 vput(ndp->ni_dvp);
1052 vput(vp);
1053 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1054 nfsvno_relpathbuf(ndp);
1055 NFSEXITCODE(error);
1056 return (error);
1057 }
1058
1059 /*
1060 * Remove a directory.
1061 */
1062 int
nfsvno_rmdirsub(struct nameidata * ndp,int is_v4,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1063 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1064 struct thread *p, struct nfsexstuff *exp)
1065 {
1066 struct vnode *vp;
1067 int error = 0;
1068
1069 vp = ndp->ni_vp;
1070 if (vp->v_type != VDIR) {
1071 error = ENOTDIR;
1072 goto out;
1073 }
1074 /*
1075 * No rmdir "." please.
1076 */
1077 if (ndp->ni_dvp == vp) {
1078 error = EINVAL;
1079 goto out;
1080 }
1081 /*
1082 * The root of a mounted filesystem cannot be deleted.
1083 */
1084 if (vp->v_vflag & VV_ROOT)
1085 error = EBUSY;
1086 out:
1087 if (!error)
1088 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1089 if (ndp->ni_dvp == vp)
1090 vrele(ndp->ni_dvp);
1091 else
1092 vput(ndp->ni_dvp);
1093 vput(vp);
1094 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1095 nfsvno_relpathbuf(ndp);
1096 NFSEXITCODE(error);
1097 return (error);
1098 }
1099
1100 /*
1101 * Rename vnode op.
1102 */
1103 int
nfsvno_rename(struct nameidata * fromndp,struct nameidata * tondp,u_int32_t ndstat,u_int32_t ndflag,struct ucred * cred,struct thread * p)1104 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1105 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1106 {
1107 struct vnode *fvp, *tvp, *tdvp;
1108 int error = 0;
1109
1110 fvp = fromndp->ni_vp;
1111 if (ndstat) {
1112 vrele(fromndp->ni_dvp);
1113 vrele(fvp);
1114 error = ndstat;
1115 goto out1;
1116 }
1117 tdvp = tondp->ni_dvp;
1118 tvp = tondp->ni_vp;
1119 if (tvp != NULL) {
1120 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1121 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1122 goto out;
1123 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1124 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1125 goto out;
1126 }
1127 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1128 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1129 goto out;
1130 }
1131
1132 /*
1133 * A rename to '.' or '..' results in a prematurely
1134 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1135 * here.
1136 */
1137 if ((tondp->ni_cnd.cn_namelen == 1 &&
1138 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1139 (tondp->ni_cnd.cn_namelen == 2 &&
1140 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1141 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1142 error = EINVAL;
1143 goto out;
1144 }
1145 }
1146 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1147 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1148 goto out;
1149 }
1150 if (fvp->v_mount != tdvp->v_mount) {
1151 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1152 goto out;
1153 }
1154 if (fvp == tdvp) {
1155 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1156 goto out;
1157 }
1158 if (fvp == tvp) {
1159 /*
1160 * If source and destination are the same, there is nothing to
1161 * do. Set error to -1 to indicate this.
1162 */
1163 error = -1;
1164 goto out;
1165 }
1166 if (ndflag & ND_NFSV4) {
1167 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1168 error = nfsrv_checkremove(fvp, 0, p);
1169 NFSVOPUNLOCK(fvp, 0);
1170 } else
1171 error = EPERM;
1172 if (tvp && !error)
1173 error = nfsrv_checkremove(tvp, 1, p);
1174 } else {
1175 /*
1176 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1177 * that the NFSv4 client won't be confused by the rename.
1178 * Since nfsd_recalldelegation() can only be called on an
1179 * unlocked vnode at this point and fvp is the file that will
1180 * still exist after the rename, just do fvp.
1181 */
1182 nfsd_recalldelegation(fvp, p);
1183 }
1184 out:
1185 if (!error) {
1186 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1187 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1188 &tondp->ni_cnd);
1189 } else {
1190 if (tdvp == tvp)
1191 vrele(tdvp);
1192 else
1193 vput(tdvp);
1194 if (tvp)
1195 vput(tvp);
1196 vrele(fromndp->ni_dvp);
1197 vrele(fvp);
1198 if (error == -1)
1199 error = 0;
1200 }
1201 vrele(tondp->ni_startdir);
1202 nfsvno_relpathbuf(tondp);
1203 out1:
1204 vrele(fromndp->ni_startdir);
1205 nfsvno_relpathbuf(fromndp);
1206 NFSEXITCODE(error);
1207 return (error);
1208 }
1209
1210 /*
1211 * Link vnode op.
1212 */
1213 int
nfsvno_link(struct nameidata * ndp,struct vnode * vp,struct ucred * cred,struct thread * p,struct nfsexstuff * exp)1214 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1215 struct thread *p, struct nfsexstuff *exp)
1216 {
1217 struct vnode *xp;
1218 int error = 0;
1219
1220 xp = ndp->ni_vp;
1221 if (xp != NULL) {
1222 error = EEXIST;
1223 } else {
1224 xp = ndp->ni_dvp;
1225 if (vp->v_mount != xp->v_mount)
1226 error = EXDEV;
1227 }
1228 if (!error) {
1229 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1230 if ((vp->v_iflag & VI_DOOMED) == 0)
1231 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1232 else
1233 error = EPERM;
1234 if (ndp->ni_dvp == vp)
1235 vrele(ndp->ni_dvp);
1236 else
1237 vput(ndp->ni_dvp);
1238 NFSVOPUNLOCK(vp, 0);
1239 } else {
1240 if (ndp->ni_dvp == ndp->ni_vp)
1241 vrele(ndp->ni_dvp);
1242 else
1243 vput(ndp->ni_dvp);
1244 if (ndp->ni_vp)
1245 vrele(ndp->ni_vp);
1246 }
1247 nfsvno_relpathbuf(ndp);
1248 NFSEXITCODE(error);
1249 return (error);
1250 }
1251
1252 /*
1253 * Do the fsync() appropriate for the commit.
1254 */
1255 int
nfsvno_fsync(struct vnode * vp,u_int64_t off,int cnt,struct ucred * cred,struct thread * td)1256 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1257 struct thread *td)
1258 {
1259 int error = 0;
1260
1261 /*
1262 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1263 * file is done. At this time VOP_FSYNC does not accept offset and
1264 * byte count parameters so call VOP_FSYNC the whole file for now.
1265 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1266 */
1267 if (cnt == 0 || cnt > MAX_COMMIT_COUNT) {
1268 /*
1269 * Give up and do the whole thing
1270 */
1271 if (vp->v_object &&
1272 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1273 VM_OBJECT_WLOCK(vp->v_object);
1274 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1275 VM_OBJECT_WUNLOCK(vp->v_object);
1276 }
1277 error = VOP_FSYNC(vp, MNT_WAIT, td);
1278 } else {
1279 /*
1280 * Locate and synchronously write any buffers that fall
1281 * into the requested range. Note: we are assuming that
1282 * f_iosize is a power of 2.
1283 */
1284 int iosize = vp->v_mount->mnt_stat.f_iosize;
1285 int iomask = iosize - 1;
1286 struct bufobj *bo;
1287 daddr_t lblkno;
1288
1289 /*
1290 * Align to iosize boundry, super-align to page boundry.
1291 */
1292 if (off & iomask) {
1293 cnt += off & iomask;
1294 off &= ~(u_quad_t)iomask;
1295 }
1296 if (off & PAGE_MASK) {
1297 cnt += off & PAGE_MASK;
1298 off &= ~(u_quad_t)PAGE_MASK;
1299 }
1300 lblkno = off / iosize;
1301
1302 if (vp->v_object &&
1303 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
1304 VM_OBJECT_WLOCK(vp->v_object);
1305 vm_object_page_clean(vp->v_object, off, off + cnt,
1306 OBJPC_SYNC);
1307 VM_OBJECT_WUNLOCK(vp->v_object);
1308 }
1309
1310 bo = &vp->v_bufobj;
1311 BO_LOCK(bo);
1312 while (cnt > 0) {
1313 struct buf *bp;
1314
1315 /*
1316 * If we have a buffer and it is marked B_DELWRI we
1317 * have to lock and write it. Otherwise the prior
1318 * write is assumed to have already been committed.
1319 *
1320 * gbincore() can return invalid buffers now so we
1321 * have to check that bit as well (though B_DELWRI
1322 * should not be set if B_INVAL is set there could be
1323 * a race here since we haven't locked the buffer).
1324 */
1325 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1326 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1327 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1328 BO_LOCK(bo);
1329 continue; /* retry */
1330 }
1331 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1332 B_DELWRI) {
1333 bremfree(bp);
1334 bp->b_flags &= ~B_ASYNC;
1335 bwrite(bp);
1336 ++nfs_commit_miss;
1337 } else
1338 BUF_UNLOCK(bp);
1339 BO_LOCK(bo);
1340 }
1341 ++nfs_commit_blks;
1342 if (cnt < iosize)
1343 break;
1344 cnt -= iosize;
1345 ++lblkno;
1346 }
1347 BO_UNLOCK(bo);
1348 }
1349 NFSEXITCODE(error);
1350 return (error);
1351 }
1352
1353 /*
1354 * Statfs vnode op.
1355 */
1356 int
nfsvno_statfs(struct vnode * vp,struct statfs * sf)1357 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1358 {
1359 int error;
1360
1361 error = VFS_STATFS(vp->v_mount, sf);
1362 if (error == 0) {
1363 /*
1364 * Since NFS handles these values as unsigned on the
1365 * wire, there is no way to represent negative values,
1366 * so set them to 0. Without this, they will appear
1367 * to be very large positive values for clients like
1368 * Solaris10.
1369 */
1370 if (sf->f_bavail < 0)
1371 sf->f_bavail = 0;
1372 if (sf->f_ffree < 0)
1373 sf->f_ffree = 0;
1374 }
1375 NFSEXITCODE(error);
1376 return (error);
1377 }
1378
1379 /*
1380 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1381 * must handle nfsrv_opencheck() calls after any other access checks.
1382 */
1383 void
nfsvno_open(struct nfsrv_descript * nd,struct nameidata * ndp,nfsquad_t clientid,nfsv4stateid_t * stateidp,struct nfsstate * stp,int * exclusive_flagp,struct nfsvattr * nvap,int32_t * cverf,int create,NFSACL_T * aclp,nfsattrbit_t * attrbitp,struct ucred * cred,struct thread * p,struct nfsexstuff * exp,struct vnode ** vpp)1384 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1385 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1386 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1387 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred, struct thread *p,
1388 struct nfsexstuff *exp, struct vnode **vpp)
1389 {
1390 struct vnode *vp = NULL;
1391 u_quad_t tempsize;
1392 struct nfsexstuff nes;
1393
1394 if (ndp->ni_vp == NULL)
1395 nd->nd_repstat = nfsrv_opencheck(clientid,
1396 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1397 if (!nd->nd_repstat) {
1398 if (ndp->ni_vp == NULL) {
1399 vrele(ndp->ni_startdir);
1400 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1401 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1402 vput(ndp->ni_dvp);
1403 nfsvno_relpathbuf(ndp);
1404 if (!nd->nd_repstat) {
1405 if (*exclusive_flagp) {
1406 *exclusive_flagp = 0;
1407 NFSVNO_ATTRINIT(nvap);
1408 nvap->na_atime.tv_sec = cverf[0];
1409 nvap->na_atime.tv_nsec = cverf[1];
1410 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1411 &nvap->na_vattr, cred);
1412 } else {
1413 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1414 aclp, p, attrbitp, exp);
1415 }
1416 }
1417 vp = ndp->ni_vp;
1418 } else {
1419 if (ndp->ni_startdir)
1420 vrele(ndp->ni_startdir);
1421 nfsvno_relpathbuf(ndp);
1422 vp = ndp->ni_vp;
1423 if (create == NFSV4OPEN_CREATE) {
1424 if (ndp->ni_dvp == vp)
1425 vrele(ndp->ni_dvp);
1426 else
1427 vput(ndp->ni_dvp);
1428 }
1429 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1430 if (ndp->ni_cnd.cn_flags & RDONLY)
1431 NFSVNO_SETEXRDONLY(&nes);
1432 else
1433 NFSVNO_EXINIT(&nes);
1434 nd->nd_repstat = nfsvno_accchk(vp,
1435 VWRITE, cred, &nes, p,
1436 NFSACCCHK_NOOVERRIDE,
1437 NFSACCCHK_VPISLOCKED, NULL);
1438 nd->nd_repstat = nfsrv_opencheck(clientid,
1439 stateidp, stp, vp, nd, p, nd->nd_repstat);
1440 if (!nd->nd_repstat) {
1441 tempsize = nvap->na_size;
1442 NFSVNO_ATTRINIT(nvap);
1443 nvap->na_size = tempsize;
1444 nd->nd_repstat = VOP_SETATTR(vp,
1445 &nvap->na_vattr, cred);
1446 }
1447 } else if (vp->v_type == VREG) {
1448 nd->nd_repstat = nfsrv_opencheck(clientid,
1449 stateidp, stp, vp, nd, p, nd->nd_repstat);
1450 }
1451 }
1452 } else {
1453 if (ndp->ni_cnd.cn_flags & HASBUF)
1454 nfsvno_relpathbuf(ndp);
1455 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1456 vrele(ndp->ni_startdir);
1457 if (ndp->ni_dvp == ndp->ni_vp)
1458 vrele(ndp->ni_dvp);
1459 else
1460 vput(ndp->ni_dvp);
1461 if (ndp->ni_vp)
1462 vput(ndp->ni_vp);
1463 }
1464 }
1465 *vpp = vp;
1466
1467 NFSEXITCODE2(0, nd);
1468 }
1469
1470 /*
1471 * Updates the file rev and sets the mtime and ctime
1472 * to the current clock time, returning the va_filerev and va_Xtime
1473 * values.
1474 */
1475 void
nfsvno_updfilerev(struct vnode * vp,struct nfsvattr * nvap,struct ucred * cred,struct thread * p)1476 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1477 struct ucred *cred, struct thread *p)
1478 {
1479 struct vattr va;
1480
1481 VATTR_NULL(&va);
1482 vfs_timestamp(&va.va_mtime);
1483 (void) VOP_SETATTR(vp, &va, cred);
1484 (void) nfsvno_getattr(vp, nvap, cred, p, 1);
1485 }
1486
1487 /*
1488 * Glue routine to nfsv4_fillattr().
1489 */
1490 int
nfsvno_fillattr(struct nfsrv_descript * nd,struct mount * mp,struct vnode * vp,struct nfsvattr * nvap,fhandle_t * fhp,int rderror,nfsattrbit_t * attrbitp,struct ucred * cred,struct thread * p,int isdgram,int reterr,int supports_nfsv4acls,int at_root,uint64_t mounted_on_fileno)1491 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1492 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1493 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1494 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1495 {
1496 int error;
1497
1498 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1499 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1500 mounted_on_fileno);
1501 NFSEXITCODE2(0, nd);
1502 return (error);
1503 }
1504
1505 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1506 /*
1507 * nfs readdir service
1508 * - mallocs what it thinks is enough to read
1509 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1510 * - calls VOP_READDIR()
1511 * - loops around building the reply
1512 * if the output generated exceeds count break out of loop
1513 * The NFSM_CLGET macro is used here so that the reply will be packed
1514 * tightly in mbuf clusters.
1515 * - it trims out records with d_fileno == 0
1516 * this doesn't matter for Unix clients, but they might confuse clients
1517 * for other os'.
1518 * - it trims out records with d_type == DT_WHT
1519 * these cannot be seen through NFS (unless we extend the protocol)
1520 * The alternate call nfsrvd_readdirplus() does lookups as well.
1521 * PS: The NFS protocol spec. does not clarify what the "count" byte
1522 * argument is a count of.. just name strings and file id's or the
1523 * entire reply rpc or ...
1524 * I tried just file name and id sizes and it confused the Sun client,
1525 * so I am using the full rpc size now. The "paranoia.." comment refers
1526 * to including the status longwords that are not a part of the dir.
1527 * "entry" structures, but are in the rpc.
1528 */
1529 int
nfsrvd_readdir(struct nfsrv_descript * nd,int isdgram,struct vnode * vp,struct thread * p,struct nfsexstuff * exp)1530 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1531 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1532 {
1533 struct dirent *dp;
1534 u_int32_t *tl;
1535 int dirlen;
1536 char *cpos, *cend, *rbuf;
1537 struct nfsvattr at;
1538 int nlen, error = 0, getret = 1;
1539 int siz, cnt, fullsiz, eofflag, ncookies;
1540 u_int64_t off, toff, verf;
1541 u_long *cookies = NULL, *cookiep;
1542 struct uio io;
1543 struct iovec iv;
1544 int not_zfs;
1545
1546 if (nd->nd_repstat) {
1547 nfsrv_postopattr(nd, getret, &at);
1548 goto out;
1549 }
1550 if (nd->nd_flag & ND_NFSV2) {
1551 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1552 off = fxdr_unsigned(u_quad_t, *tl++);
1553 } else {
1554 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1555 off = fxdr_hyper(tl);
1556 tl += 2;
1557 verf = fxdr_hyper(tl);
1558 tl += 2;
1559 }
1560 toff = off;
1561 cnt = fxdr_unsigned(int, *tl);
1562 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1563 cnt = NFS_SRVMAXDATA(nd);
1564 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1565 fullsiz = siz;
1566 if (nd->nd_flag & ND_NFSV3) {
1567 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred,
1568 p, 1);
1569 #if 0
1570 /*
1571 * va_filerev is not sufficient as a cookie verifier,
1572 * since it is not supposed to change when entries are
1573 * removed/added unless that offset cookies returned to
1574 * the client are no longer valid.
1575 */
1576 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1577 nd->nd_repstat = NFSERR_BAD_COOKIE;
1578 #endif
1579 }
1580 if (!nd->nd_repstat && vp->v_type != VDIR)
1581 nd->nd_repstat = NFSERR_NOTDIR;
1582 if (nd->nd_repstat == 0 && cnt == 0) {
1583 if (nd->nd_flag & ND_NFSV2)
1584 /* NFSv2 does not have NFSERR_TOOSMALL */
1585 nd->nd_repstat = EPERM;
1586 else
1587 nd->nd_repstat = NFSERR_TOOSMALL;
1588 }
1589 if (!nd->nd_repstat)
1590 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1591 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1592 NFSACCCHK_VPISLOCKED, NULL);
1593 if (nd->nd_repstat) {
1594 vput(vp);
1595 if (nd->nd_flag & ND_NFSV3)
1596 nfsrv_postopattr(nd, getret, &at);
1597 goto out;
1598 }
1599 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1600 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1601 again:
1602 eofflag = 0;
1603 if (cookies) {
1604 free((caddr_t)cookies, M_TEMP);
1605 cookies = NULL;
1606 }
1607
1608 iv.iov_base = rbuf;
1609 iv.iov_len = siz;
1610 io.uio_iov = &iv;
1611 io.uio_iovcnt = 1;
1612 io.uio_offset = (off_t)off;
1613 io.uio_resid = siz;
1614 io.uio_segflg = UIO_SYSSPACE;
1615 io.uio_rw = UIO_READ;
1616 io.uio_td = NULL;
1617 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1618 &cookies);
1619 off = (u_int64_t)io.uio_offset;
1620 if (io.uio_resid)
1621 siz -= io.uio_resid;
1622
1623 if (!cookies && !nd->nd_repstat)
1624 nd->nd_repstat = NFSERR_PERM;
1625 if (nd->nd_flag & ND_NFSV3) {
1626 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1627 if (!nd->nd_repstat)
1628 nd->nd_repstat = getret;
1629 }
1630
1631 /*
1632 * Handles the failed cases. nd->nd_repstat == 0 past here.
1633 */
1634 if (nd->nd_repstat) {
1635 vput(vp);
1636 free((caddr_t)rbuf, M_TEMP);
1637 if (cookies)
1638 free((caddr_t)cookies, M_TEMP);
1639 if (nd->nd_flag & ND_NFSV3)
1640 nfsrv_postopattr(nd, getret, &at);
1641 goto out;
1642 }
1643 /*
1644 * If nothing read, return eof
1645 * rpc reply
1646 */
1647 if (siz == 0) {
1648 vput(vp);
1649 if (nd->nd_flag & ND_NFSV2) {
1650 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1651 } else {
1652 nfsrv_postopattr(nd, getret, &at);
1653 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1654 txdr_hyper(at.na_filerev, tl);
1655 tl += 2;
1656 }
1657 *tl++ = newnfs_false;
1658 *tl = newnfs_true;
1659 FREE((caddr_t)rbuf, M_TEMP);
1660 FREE((caddr_t)cookies, M_TEMP);
1661 goto out;
1662 }
1663
1664 /*
1665 * Check for degenerate cases of nothing useful read.
1666 * If so go try again
1667 */
1668 cpos = rbuf;
1669 cend = rbuf + siz;
1670 dp = (struct dirent *)cpos;
1671 cookiep = cookies;
1672
1673 /*
1674 * For some reason FreeBSD's ufs_readdir() chooses to back the
1675 * directory offset up to a block boundary, so it is necessary to
1676 * skip over the records that precede the requested offset. This
1677 * requires the assumption that file offset cookies monotonically
1678 * increase.
1679 * Since the offset cookies don't monotonically increase for ZFS,
1680 * this is not done when ZFS is the file system.
1681 */
1682 while (cpos < cend && ncookies > 0 &&
1683 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1684 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
1685 cpos += dp->d_reclen;
1686 dp = (struct dirent *)cpos;
1687 cookiep++;
1688 ncookies--;
1689 }
1690 if (cpos >= cend || ncookies == 0) {
1691 siz = fullsiz;
1692 toff = off;
1693 goto again;
1694 }
1695 vput(vp);
1696
1697 /*
1698 * dirlen is the size of the reply, including all XDR and must
1699 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
1700 * if the XDR should be included in "count", but to be safe, we do.
1701 * (Include the two booleans at the end of the reply in dirlen now.)
1702 */
1703 if (nd->nd_flag & ND_NFSV3) {
1704 nfsrv_postopattr(nd, getret, &at);
1705 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1706 txdr_hyper(at.na_filerev, tl);
1707 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
1708 } else {
1709 dirlen = 2 * NFSX_UNSIGNED;
1710 }
1711
1712 /* Loop through the records and build reply */
1713 while (cpos < cend && ncookies > 0) {
1714 nlen = dp->d_namlen;
1715 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
1716 nlen <= NFS_MAXNAMLEN) {
1717 if (nd->nd_flag & ND_NFSV3)
1718 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1719 else
1720 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
1721 if (dirlen > cnt) {
1722 eofflag = 0;
1723 break;
1724 }
1725
1726 /*
1727 * Build the directory record xdr from
1728 * the dirent entry.
1729 */
1730 if (nd->nd_flag & ND_NFSV3) {
1731 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1732 *tl++ = newnfs_true;
1733 *tl++ = 0;
1734 } else {
1735 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1736 *tl++ = newnfs_true;
1737 }
1738 *tl = txdr_unsigned(dp->d_fileno);
1739 (void) nfsm_strtom(nd, dp->d_name, nlen);
1740 if (nd->nd_flag & ND_NFSV3) {
1741 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1742 *tl++ = 0;
1743 } else
1744 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
1745 *tl = txdr_unsigned(*cookiep);
1746 }
1747 cpos += dp->d_reclen;
1748 dp = (struct dirent *)cpos;
1749 cookiep++;
1750 ncookies--;
1751 }
1752 if (cpos < cend)
1753 eofflag = 0;
1754 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1755 *tl++ = newnfs_false;
1756 if (eofflag)
1757 *tl = newnfs_true;
1758 else
1759 *tl = newnfs_false;
1760 FREE((caddr_t)rbuf, M_TEMP);
1761 FREE((caddr_t)cookies, M_TEMP);
1762
1763 out:
1764 NFSEXITCODE2(0, nd);
1765 return (0);
1766 nfsmout:
1767 vput(vp);
1768 NFSEXITCODE2(error, nd);
1769 return (error);
1770 }
1771
1772 /*
1773 * Readdirplus for V3 and Readdir for V4.
1774 */
1775 int
nfsrvd_readdirplus(struct nfsrv_descript * nd,int isdgram,struct vnode * vp,struct thread * p,struct nfsexstuff * exp)1776 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
1777 struct vnode *vp, struct thread *p, struct nfsexstuff *exp)
1778 {
1779 struct dirent *dp;
1780 u_int32_t *tl;
1781 int dirlen;
1782 char *cpos, *cend, *rbuf;
1783 struct vnode *nvp;
1784 fhandle_t nfh;
1785 struct nfsvattr nva, at, *nvap = &nva;
1786 struct mbuf *mb0, *mb1;
1787 struct nfsreferral *refp;
1788 int nlen, r, error = 0, getret = 1, usevget = 1;
1789 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
1790 caddr_t bpos0, bpos1;
1791 u_int64_t off, toff, verf;
1792 u_long *cookies = NULL, *cookiep;
1793 nfsattrbit_t attrbits, rderrbits, savbits;
1794 struct uio io;
1795 struct iovec iv;
1796 struct componentname cn;
1797 int at_root, needs_unbusy, not_zfs, supports_nfsv4acls;
1798 struct mount *mp, *new_mp;
1799 uint64_t mounted_on_fileno;
1800
1801 if (nd->nd_repstat) {
1802 nfsrv_postopattr(nd, getret, &at);
1803 goto out;
1804 }
1805 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
1806 off = fxdr_hyper(tl);
1807 toff = off;
1808 tl += 2;
1809 verf = fxdr_hyper(tl);
1810 tl += 2;
1811 siz = fxdr_unsigned(int, *tl++);
1812 cnt = fxdr_unsigned(int, *tl);
1813
1814 /*
1815 * Use the server's maximum data transfer size as the upper bound
1816 * on reply datalen.
1817 */
1818 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1819 cnt = NFS_SRVMAXDATA(nd);
1820
1821 /*
1822 * siz is a "hint" of how much directory information (name, fileid,
1823 * cookie) should be in the reply. At least one client "hints" 0,
1824 * so I set it to cnt for that case. I also round it up to the
1825 * next multiple of DIRBLKSIZ.
1826 */
1827 if (siz <= 0)
1828 siz = cnt;
1829 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1830
1831 if (nd->nd_flag & ND_NFSV4) {
1832 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
1833 if (error)
1834 goto nfsmout;
1835 NFSSET_ATTRBIT(&savbits, &attrbits);
1836 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits);
1837 NFSZERO_ATTRBIT(&rderrbits);
1838 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
1839 } else {
1840 NFSZERO_ATTRBIT(&attrbits);
1841 }
1842 fullsiz = siz;
1843 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1844 if (!nd->nd_repstat) {
1845 if (off && verf != at.na_filerev) {
1846 /*
1847 * va_filerev is not sufficient as a cookie verifier,
1848 * since it is not supposed to change when entries are
1849 * removed/added unless that offset cookies returned to
1850 * the client are no longer valid.
1851 */
1852 #if 0
1853 if (nd->nd_flag & ND_NFSV4) {
1854 nd->nd_repstat = NFSERR_NOTSAME;
1855 } else {
1856 nd->nd_repstat = NFSERR_BAD_COOKIE;
1857 }
1858 #endif
1859 } else if ((nd->nd_flag & ND_NFSV4) && off == 0 && verf != 0) {
1860 nd->nd_repstat = NFSERR_BAD_COOKIE;
1861 }
1862 }
1863 if (!nd->nd_repstat && vp->v_type != VDIR)
1864 nd->nd_repstat = NFSERR_NOTDIR;
1865 if (!nd->nd_repstat && cnt == 0)
1866 nd->nd_repstat = NFSERR_TOOSMALL;
1867 if (!nd->nd_repstat)
1868 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1869 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1870 NFSACCCHK_VPISLOCKED, NULL);
1871 if (nd->nd_repstat) {
1872 vput(vp);
1873 if (nd->nd_flag & ND_NFSV3)
1874 nfsrv_postopattr(nd, getret, &at);
1875 goto out;
1876 }
1877 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs");
1878
1879 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
1880 again:
1881 eofflag = 0;
1882 if (cookies) {
1883 free((caddr_t)cookies, M_TEMP);
1884 cookies = NULL;
1885 }
1886
1887 iv.iov_base = rbuf;
1888 iv.iov_len = siz;
1889 io.uio_iov = &iv;
1890 io.uio_iovcnt = 1;
1891 io.uio_offset = (off_t)off;
1892 io.uio_resid = siz;
1893 io.uio_segflg = UIO_SYSSPACE;
1894 io.uio_rw = UIO_READ;
1895 io.uio_td = NULL;
1896 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1897 &cookies);
1898 off = (u_int64_t)io.uio_offset;
1899 if (io.uio_resid)
1900 siz -= io.uio_resid;
1901
1902 getret = nfsvno_getattr(vp, &at, nd->nd_cred, p, 1);
1903
1904 if (!cookies && !nd->nd_repstat)
1905 nd->nd_repstat = NFSERR_PERM;
1906 if (!nd->nd_repstat)
1907 nd->nd_repstat = getret;
1908 if (nd->nd_repstat) {
1909 vput(vp);
1910 if (cookies)
1911 free((caddr_t)cookies, M_TEMP);
1912 free((caddr_t)rbuf, M_TEMP);
1913 if (nd->nd_flag & ND_NFSV3)
1914 nfsrv_postopattr(nd, getret, &at);
1915 goto out;
1916 }
1917 /*
1918 * If nothing read, return eof
1919 * rpc reply
1920 */
1921 if (siz == 0) {
1922 vput(vp);
1923 if (nd->nd_flag & ND_NFSV3)
1924 nfsrv_postopattr(nd, getret, &at);
1925 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1926 txdr_hyper(at.na_filerev, tl);
1927 tl += 2;
1928 *tl++ = newnfs_false;
1929 *tl = newnfs_true;
1930 free((caddr_t)cookies, M_TEMP);
1931 free((caddr_t)rbuf, M_TEMP);
1932 goto out;
1933 }
1934
1935 /*
1936 * Check for degenerate cases of nothing useful read.
1937 * If so go try again
1938 */
1939 cpos = rbuf;
1940 cend = rbuf + siz;
1941 dp = (struct dirent *)cpos;
1942 cookiep = cookies;
1943
1944 /*
1945 * For some reason FreeBSD's ufs_readdir() chooses to back the
1946 * directory offset up to a block boundary, so it is necessary to
1947 * skip over the records that precede the requested offset. This
1948 * requires the assumption that file offset cookies monotonically
1949 * increase.
1950 * Since the offset cookies don't monotonically increase for ZFS,
1951 * this is not done when ZFS is the file system.
1952 */
1953 while (cpos < cend && ncookies > 0 &&
1954 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
1955 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff) ||
1956 ((nd->nd_flag & ND_NFSV4) &&
1957 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1958 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
1959 cpos += dp->d_reclen;
1960 dp = (struct dirent *)cpos;
1961 cookiep++;
1962 ncookies--;
1963 }
1964 if (cpos >= cend || ncookies == 0) {
1965 siz = fullsiz;
1966 toff = off;
1967 goto again;
1968 }
1969
1970 /*
1971 * Busy the file system so that the mount point won't go away
1972 * and, as such, VFS_VGET() can be used safely.
1973 */
1974 mp = vp->v_mount;
1975 vfs_ref(mp);
1976 NFSVOPUNLOCK(vp, 0);
1977 nd->nd_repstat = vfs_busy(mp, 0);
1978 vfs_rel(mp);
1979 if (nd->nd_repstat != 0) {
1980 vrele(vp);
1981 free(cookies, M_TEMP);
1982 free(rbuf, M_TEMP);
1983 if (nd->nd_flag & ND_NFSV3)
1984 nfsrv_postopattr(nd, getret, &at);
1985 goto out;
1986 }
1987
1988 /*
1989 * Save this position, in case there is an error before one entry
1990 * is created.
1991 */
1992 mb0 = nd->nd_mb;
1993 bpos0 = nd->nd_bpos;
1994
1995 /*
1996 * Fill in the first part of the reply.
1997 * dirlen is the reply length in bytes and cannot exceed cnt.
1998 * (Include the two booleans at the end of the reply in dirlen now,
1999 * so we recognize when we have exceeded cnt.)
2000 */
2001 if (nd->nd_flag & ND_NFSV3) {
2002 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2003 nfsrv_postopattr(nd, getret, &at);
2004 } else {
2005 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2006 }
2007 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2008 txdr_hyper(at.na_filerev, tl);
2009
2010 /*
2011 * Save this position, in case there is an empty reply needed.
2012 */
2013 mb1 = nd->nd_mb;
2014 bpos1 = nd->nd_bpos;
2015
2016 /* Loop through the records and build reply */
2017 entrycnt = 0;
2018 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2019 nlen = dp->d_namlen;
2020 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2021 nlen <= NFS_MAXNAMLEN &&
2022 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2023 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2024 || (nlen == 1 && dp->d_name[0] != '.'))) {
2025 /*
2026 * Save the current position in the reply, in case
2027 * this entry exceeds cnt.
2028 */
2029 mb1 = nd->nd_mb;
2030 bpos1 = nd->nd_bpos;
2031
2032 /*
2033 * For readdir_and_lookup get the vnode using
2034 * the file number.
2035 */
2036 nvp = NULL;
2037 refp = NULL;
2038 r = 0;
2039 at_root = 0;
2040 needs_unbusy = 0;
2041 new_mp = mp;
2042 mounted_on_fileno = (uint64_t)dp->d_fileno;
2043 if ((nd->nd_flag & ND_NFSV3) ||
2044 NFSNONZERO_ATTRBIT(&savbits)) {
2045 if (nd->nd_flag & ND_NFSV4)
2046 refp = nfsv4root_getreferral(NULL,
2047 vp, dp->d_fileno);
2048 if (refp == NULL) {
2049 if (usevget)
2050 r = VFS_VGET(mp, dp->d_fileno,
2051 LK_SHARED, &nvp);
2052 else
2053 r = EOPNOTSUPP;
2054 if (r == EOPNOTSUPP) {
2055 if (usevget) {
2056 usevget = 0;
2057 cn.cn_nameiop = LOOKUP;
2058 cn.cn_lkflags =
2059 LK_SHARED |
2060 LK_RETRY;
2061 cn.cn_cred =
2062 nd->nd_cred;
2063 cn.cn_thread = p;
2064 }
2065 cn.cn_nameptr = dp->d_name;
2066 cn.cn_namelen = nlen;
2067 cn.cn_flags = ISLASTCN |
2068 NOFOLLOW | LOCKLEAF;
2069 if (nlen == 2 &&
2070 dp->d_name[0] == '.' &&
2071 dp->d_name[1] == '.')
2072 cn.cn_flags |=
2073 ISDOTDOT;
2074 if (NFSVOPLOCK(vp, LK_SHARED)
2075 != 0) {
2076 nd->nd_repstat = EPERM;
2077 break;
2078 }
2079 if ((vp->v_vflag & VV_ROOT) != 0
2080 && (cn.cn_flags & ISDOTDOT)
2081 != 0) {
2082 vref(vp);
2083 nvp = vp;
2084 r = 0;
2085 } else {
2086 r = VOP_LOOKUP(vp, &nvp,
2087 &cn);
2088 if (vp != nvp)
2089 NFSVOPUNLOCK(vp,
2090 0);
2091 }
2092 }
2093
2094 /*
2095 * For NFSv4, check to see if nvp is
2096 * a mount point and get the mount
2097 * point vnode, as required.
2098 */
2099 if (r == 0 &&
2100 nfsrv_enable_crossmntpt != 0 &&
2101 (nd->nd_flag & ND_NFSV4) != 0 &&
2102 nvp->v_type == VDIR &&
2103 nvp->v_mountedhere != NULL) {
2104 new_mp = nvp->v_mountedhere;
2105 r = vfs_busy(new_mp, 0);
2106 vput(nvp);
2107 nvp = NULL;
2108 if (r == 0) {
2109 r = VFS_ROOT(new_mp,
2110 LK_SHARED, &nvp);
2111 needs_unbusy = 1;
2112 if (r == 0)
2113 at_root = 1;
2114 }
2115 }
2116 }
2117 if (!r) {
2118 if (refp == NULL &&
2119 ((nd->nd_flag & ND_NFSV3) ||
2120 NFSNONZERO_ATTRBIT(&attrbits))) {
2121 r = nfsvno_getfh(nvp, &nfh, p);
2122 if (!r)
2123 r = nfsvno_getattr(nvp, nvap,
2124 nd->nd_cred, p, 1);
2125 }
2126 } else {
2127 nvp = NULL;
2128 }
2129 if (r) {
2130 if (!NFSISSET_ATTRBIT(&attrbits,
2131 NFSATTRBIT_RDATTRERROR)) {
2132 if (nvp != NULL)
2133 vput(nvp);
2134 if (needs_unbusy != 0)
2135 vfs_unbusy(new_mp);
2136 nd->nd_repstat = r;
2137 break;
2138 }
2139 }
2140 }
2141
2142 /*
2143 * Build the directory record xdr
2144 */
2145 if (nd->nd_flag & ND_NFSV3) {
2146 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2147 *tl++ = newnfs_true;
2148 *tl++ = 0;
2149 *tl = txdr_unsigned(dp->d_fileno);
2150 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2151 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2152 *tl++ = 0;
2153 *tl = txdr_unsigned(*cookiep);
2154 nfsrv_postopattr(nd, 0, nvap);
2155 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2156 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2157 if (nvp != NULL)
2158 vput(nvp);
2159 } else {
2160 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2161 *tl++ = newnfs_true;
2162 *tl++ = 0;
2163 *tl = txdr_unsigned(*cookiep);
2164 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2165 if (nvp != NULL) {
2166 supports_nfsv4acls =
2167 nfs_supportsnfsv4acls(nvp);
2168 NFSVOPUNLOCK(nvp, 0);
2169 } else
2170 supports_nfsv4acls = 0;
2171 if (refp != NULL) {
2172 dirlen += nfsrv_putreferralattr(nd,
2173 &savbits, refp, 0,
2174 &nd->nd_repstat);
2175 if (nd->nd_repstat) {
2176 if (nvp != NULL)
2177 vrele(nvp);
2178 if (needs_unbusy != 0)
2179 vfs_unbusy(new_mp);
2180 break;
2181 }
2182 } else if (r) {
2183 dirlen += nfsvno_fillattr(nd, new_mp,
2184 nvp, nvap, &nfh, r, &rderrbits,
2185 nd->nd_cred, p, isdgram, 0,
2186 supports_nfsv4acls, at_root,
2187 mounted_on_fileno);
2188 } else {
2189 dirlen += nfsvno_fillattr(nd, new_mp,
2190 nvp, nvap, &nfh, r, &attrbits,
2191 nd->nd_cred, p, isdgram, 0,
2192 supports_nfsv4acls, at_root,
2193 mounted_on_fileno);
2194 }
2195 if (nvp != NULL)
2196 vrele(nvp);
2197 dirlen += (3 * NFSX_UNSIGNED);
2198 }
2199 if (needs_unbusy != 0)
2200 vfs_unbusy(new_mp);
2201 if (dirlen <= cnt)
2202 entrycnt++;
2203 }
2204 cpos += dp->d_reclen;
2205 dp = (struct dirent *)cpos;
2206 cookiep++;
2207 ncookies--;
2208 }
2209 vrele(vp);
2210 vfs_unbusy(mp);
2211
2212 /*
2213 * If dirlen > cnt, we must strip off the last entry. If that
2214 * results in an empty reply, report NFSERR_TOOSMALL.
2215 */
2216 if (dirlen > cnt || nd->nd_repstat) {
2217 if (!nd->nd_repstat && entrycnt == 0)
2218 nd->nd_repstat = NFSERR_TOOSMALL;
2219 if (nd->nd_repstat)
2220 newnfs_trimtrailing(nd, mb0, bpos0);
2221 else
2222 newnfs_trimtrailing(nd, mb1, bpos1);
2223 eofflag = 0;
2224 } else if (cpos < cend)
2225 eofflag = 0;
2226 if (!nd->nd_repstat) {
2227 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2228 *tl++ = newnfs_false;
2229 if (eofflag)
2230 *tl = newnfs_true;
2231 else
2232 *tl = newnfs_false;
2233 }
2234 FREE((caddr_t)cookies, M_TEMP);
2235 FREE((caddr_t)rbuf, M_TEMP);
2236
2237 out:
2238 NFSEXITCODE2(0, nd);
2239 return (0);
2240 nfsmout:
2241 vput(vp);
2242 NFSEXITCODE2(error, nd);
2243 return (error);
2244 }
2245
2246 /*
2247 * Get the settable attributes out of the mbuf list.
2248 * (Return 0 or EBADRPC)
2249 */
2250 int
nfsrv_sattr(struct nfsrv_descript * nd,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,NFSACL_T * aclp,struct thread * p)2251 nfsrv_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2252 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2253 {
2254 u_int32_t *tl;
2255 struct nfsv2_sattr *sp;
2256 int error = 0, toclient = 0;
2257
2258 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2259 case ND_NFSV2:
2260 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2261 /*
2262 * Some old clients didn't fill in the high order 16bits.
2263 * --> check the low order 2 bytes for 0xffff
2264 */
2265 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2266 nvap->na_mode = nfstov_mode(sp->sa_mode);
2267 if (sp->sa_uid != newnfs_xdrneg1)
2268 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2269 if (sp->sa_gid != newnfs_xdrneg1)
2270 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2271 if (sp->sa_size != newnfs_xdrneg1)
2272 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2273 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2274 #ifdef notyet
2275 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2276 #else
2277 nvap->na_atime.tv_sec =
2278 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2279 nvap->na_atime.tv_nsec = 0;
2280 #endif
2281 }
2282 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2283 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2284 break;
2285 case ND_NFSV3:
2286 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2287 if (*tl == newnfs_true) {
2288 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2289 nvap->na_mode = nfstov_mode(*tl);
2290 }
2291 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2292 if (*tl == newnfs_true) {
2293 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2294 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2295 }
2296 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2297 if (*tl == newnfs_true) {
2298 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2299 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2300 }
2301 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2302 if (*tl == newnfs_true) {
2303 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2304 nvap->na_size = fxdr_hyper(tl);
2305 }
2306 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2307 switch (fxdr_unsigned(int, *tl)) {
2308 case NFSV3SATTRTIME_TOCLIENT:
2309 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2310 fxdr_nfsv3time(tl, &nvap->na_atime);
2311 toclient = 1;
2312 break;
2313 case NFSV3SATTRTIME_TOSERVER:
2314 vfs_timestamp(&nvap->na_atime);
2315 nvap->na_vaflags |= VA_UTIMES_NULL;
2316 break;
2317 };
2318 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2319 switch (fxdr_unsigned(int, *tl)) {
2320 case NFSV3SATTRTIME_TOCLIENT:
2321 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2322 fxdr_nfsv3time(tl, &nvap->na_mtime);
2323 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2324 break;
2325 case NFSV3SATTRTIME_TOSERVER:
2326 vfs_timestamp(&nvap->na_mtime);
2327 if (!toclient)
2328 nvap->na_vaflags |= VA_UTIMES_NULL;
2329 break;
2330 };
2331 break;
2332 case ND_NFSV4:
2333 error = nfsv4_sattr(nd, nvap, attrbitp, aclp, p);
2334 };
2335 nfsmout:
2336 NFSEXITCODE2(error, nd);
2337 return (error);
2338 }
2339
2340 /*
2341 * Handle the setable attributes for V4.
2342 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2343 */
2344 int
nfsv4_sattr(struct nfsrv_descript * nd,struct nfsvattr * nvap,nfsattrbit_t * attrbitp,NFSACL_T * aclp,struct thread * p)2345 nfsv4_sattr(struct nfsrv_descript *nd, struct nfsvattr *nvap,
2346 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2347 {
2348 u_int32_t *tl;
2349 int attrsum = 0;
2350 int i, j;
2351 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2352 int toclient = 0;
2353 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2354 uid_t uid;
2355 gid_t gid;
2356
2357 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2358 if (error)
2359 goto nfsmout;
2360 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2361 attrsize = fxdr_unsigned(int, *tl);
2362
2363 /*
2364 * Loop around getting the setable attributes. If an unsupported
2365 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2366 */
2367 if (retnotsup) {
2368 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2369 bitpos = NFSATTRBIT_MAX;
2370 } else {
2371 bitpos = 0;
2372 }
2373 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2374 if (attrsum > attrsize) {
2375 error = NFSERR_BADXDR;
2376 goto nfsmout;
2377 }
2378 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2379 switch (bitpos) {
2380 case NFSATTRBIT_SIZE:
2381 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2382 nvap->na_size = fxdr_hyper(tl);
2383 attrsum += NFSX_HYPER;
2384 break;
2385 case NFSATTRBIT_ACL:
2386 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2387 p);
2388 if (error)
2389 goto nfsmout;
2390 if (aceerr && !nd->nd_repstat)
2391 nd->nd_repstat = aceerr;
2392 attrsum += aclsize;
2393 break;
2394 case NFSATTRBIT_ARCHIVE:
2395 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2396 if (!nd->nd_repstat)
2397 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2398 attrsum += NFSX_UNSIGNED;
2399 break;
2400 case NFSATTRBIT_HIDDEN:
2401 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2402 if (!nd->nd_repstat)
2403 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2404 attrsum += NFSX_UNSIGNED;
2405 break;
2406 case NFSATTRBIT_MIMETYPE:
2407 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2408 i = fxdr_unsigned(int, *tl);
2409 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2410 if (error)
2411 goto nfsmout;
2412 if (!nd->nd_repstat)
2413 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2414 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2415 break;
2416 case NFSATTRBIT_MODE:
2417 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2418 nvap->na_mode = nfstov_mode(*tl);
2419 attrsum += NFSX_UNSIGNED;
2420 break;
2421 case NFSATTRBIT_OWNER:
2422 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2423 j = fxdr_unsigned(int, *tl);
2424 if (j < 0) {
2425 error = NFSERR_BADXDR;
2426 goto nfsmout;
2427 }
2428 if (j > NFSV4_SMALLSTR)
2429 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2430 else
2431 cp = namestr;
2432 error = nfsrv_mtostr(nd, cp, j);
2433 if (error) {
2434 if (j > NFSV4_SMALLSTR)
2435 free(cp, M_NFSSTRING);
2436 goto nfsmout;
2437 }
2438 if (!nd->nd_repstat) {
2439 nd->nd_repstat = nfsv4_strtouid(nd, cp, j, &uid,
2440 p);
2441 if (!nd->nd_repstat)
2442 nvap->na_uid = uid;
2443 }
2444 if (j > NFSV4_SMALLSTR)
2445 free(cp, M_NFSSTRING);
2446 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2447 break;
2448 case NFSATTRBIT_OWNERGROUP:
2449 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2450 j = fxdr_unsigned(int, *tl);
2451 if (j < 0) {
2452 error = NFSERR_BADXDR;
2453 goto nfsmout;
2454 }
2455 if (j > NFSV4_SMALLSTR)
2456 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2457 else
2458 cp = namestr;
2459 error = nfsrv_mtostr(nd, cp, j);
2460 if (error) {
2461 if (j > NFSV4_SMALLSTR)
2462 free(cp, M_NFSSTRING);
2463 goto nfsmout;
2464 }
2465 if (!nd->nd_repstat) {
2466 nd->nd_repstat = nfsv4_strtogid(nd, cp, j, &gid,
2467 p);
2468 if (!nd->nd_repstat)
2469 nvap->na_gid = gid;
2470 }
2471 if (j > NFSV4_SMALLSTR)
2472 free(cp, M_NFSSTRING);
2473 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2474 break;
2475 case NFSATTRBIT_SYSTEM:
2476 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2477 if (!nd->nd_repstat)
2478 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2479 attrsum += NFSX_UNSIGNED;
2480 break;
2481 case NFSATTRBIT_TIMEACCESSSET:
2482 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2483 attrsum += NFSX_UNSIGNED;
2484 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2485 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2486 fxdr_nfsv4time(tl, &nvap->na_atime);
2487 toclient = 1;
2488 attrsum += NFSX_V4TIME;
2489 } else {
2490 vfs_timestamp(&nvap->na_atime);
2491 nvap->na_vaflags |= VA_UTIMES_NULL;
2492 }
2493 break;
2494 case NFSATTRBIT_TIMEBACKUP:
2495 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2496 if (!nd->nd_repstat)
2497 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2498 attrsum += NFSX_V4TIME;
2499 break;
2500 case NFSATTRBIT_TIMECREATE:
2501 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2502 if (!nd->nd_repstat)
2503 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2504 attrsum += NFSX_V4TIME;
2505 break;
2506 case NFSATTRBIT_TIMEMODIFYSET:
2507 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2508 attrsum += NFSX_UNSIGNED;
2509 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2510 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2511 fxdr_nfsv4time(tl, &nvap->na_mtime);
2512 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2513 attrsum += NFSX_V4TIME;
2514 } else {
2515 vfs_timestamp(&nvap->na_mtime);
2516 if (!toclient)
2517 nvap->na_vaflags |= VA_UTIMES_NULL;
2518 }
2519 break;
2520 default:
2521 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2522 /*
2523 * set bitpos so we drop out of the loop.
2524 */
2525 bitpos = NFSATTRBIT_MAX;
2526 break;
2527 };
2528 }
2529
2530 /*
2531 * some clients pad the attrlist, so we need to skip over the
2532 * padding.
2533 */
2534 if (attrsum > attrsize) {
2535 error = NFSERR_BADXDR;
2536 } else {
2537 attrsize = NFSM_RNDUP(attrsize);
2538 if (attrsum < attrsize)
2539 error = nfsm_advance(nd, attrsize - attrsum, -1);
2540 }
2541 nfsmout:
2542 NFSEXITCODE2(error, nd);
2543 return (error);
2544 }
2545
2546 /*
2547 * Check/setup export credentials.
2548 */
2549 int
nfsd_excred(struct nfsrv_descript * nd,struct nfsexstuff * exp,struct ucred * credanon)2550 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2551 struct ucred *credanon)
2552 {
2553 int error = 0;
2554
2555 /*
2556 * Check/setup credentials.
2557 */
2558 if (nd->nd_flag & ND_GSS)
2559 exp->nes_exflag &= ~MNT_EXPORTANON;
2560
2561 /*
2562 * Check to see if the operation is allowed for this security flavor.
2563 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
2564 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
2565 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
2566 */
2567 if (nfsvno_testexp(nd, exp) &&
2568 nd->nd_procnum != NFSV4OP_SECINFO &&
2569 nd->nd_procnum != NFSPROC_FSINFO) {
2570 if (nd->nd_flag & ND_NFSV4)
2571 error = NFSERR_WRONGSEC;
2572 else
2573 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2574 goto out;
2575 }
2576
2577 /*
2578 * Check to see if the file system is exported V4 only.
2579 */
2580 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
2581 error = NFSERR_PROGNOTV4;
2582 goto out;
2583 }
2584
2585 /*
2586 * Now, map the user credentials.
2587 * (Note that ND_AUTHNONE will only be set for an NFSv3
2588 * Fsinfo RPC. If set for anything else, this code might need
2589 * to change.)
2590 */
2591 if (NFSVNO_EXPORTED(exp) &&
2592 ((!(nd->nd_flag & ND_GSS) && nd->nd_cred->cr_uid == 0) ||
2593 NFSVNO_EXPORTANON(exp) ||
2594 (nd->nd_flag & ND_AUTHNONE))) {
2595 nd->nd_cred->cr_uid = credanon->cr_uid;
2596 nd->nd_cred->cr_gid = credanon->cr_gid;
2597 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
2598 credanon->cr_groups);
2599 }
2600
2601 out:
2602 NFSEXITCODE2(error, nd);
2603 return (error);
2604 }
2605
2606 /*
2607 * Check exports.
2608 */
2609 int
nfsvno_checkexp(struct mount * mp,struct sockaddr * nam,struct nfsexstuff * exp,struct ucred ** credp)2610 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
2611 struct ucred **credp)
2612 {
2613 int i, error, *secflavors;
2614
2615 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2616 &exp->nes_numsecflavor, &secflavors);
2617 if (error) {
2618 if (nfs_rootfhset) {
2619 exp->nes_exflag = 0;
2620 exp->nes_numsecflavor = 0;
2621 error = 0;
2622 }
2623 } else {
2624 /* Copy the security flavors. */
2625 for (i = 0; i < exp->nes_numsecflavor; i++)
2626 exp->nes_secflavors[i] = secflavors[i];
2627 }
2628 NFSEXITCODE(error);
2629 return (error);
2630 }
2631
2632 /*
2633 * Get a vnode for a file handle and export stuff.
2634 */
2635 int
nfsvno_fhtovp(struct mount * mp,fhandle_t * fhp,struct sockaddr * nam,int lktype,struct vnode ** vpp,struct nfsexstuff * exp,struct ucred ** credp)2636 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
2637 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
2638 struct ucred **credp)
2639 {
2640 int i, error, *secflavors;
2641
2642 *credp = NULL;
2643 exp->nes_numsecflavor = 0;
2644 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
2645 if (error != 0)
2646 /* Make sure the server replies ESTALE to the client. */
2647 error = ESTALE;
2648 if (nam && !error) {
2649 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
2650 &exp->nes_numsecflavor, &secflavors);
2651 if (error) {
2652 if (nfs_rootfhset) {
2653 exp->nes_exflag = 0;
2654 exp->nes_numsecflavor = 0;
2655 error = 0;
2656 } else {
2657 vput(*vpp);
2658 }
2659 } else {
2660 /* Copy the security flavors. */
2661 for (i = 0; i < exp->nes_numsecflavor; i++)
2662 exp->nes_secflavors[i] = secflavors[i];
2663 }
2664 }
2665 NFSEXITCODE(error);
2666 return (error);
2667 }
2668
2669 /*
2670 * nfsd_fhtovp() - convert a fh to a vnode ptr
2671 * - look up fsid in mount list (if not found ret error)
2672 * - get vp and export rights by calling nfsvno_fhtovp()
2673 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
2674 * for AUTH_SYS
2675 * - if mpp != NULL, return the mount point so that it can
2676 * be used for vn_finished_write() by the caller
2677 */
2678 void
nfsd_fhtovp(struct nfsrv_descript * nd,struct nfsrvfh * nfp,int lktype,struct vnode ** vpp,struct nfsexstuff * exp,struct mount ** mpp,int startwrite,struct thread * p)2679 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
2680 struct vnode **vpp, struct nfsexstuff *exp,
2681 struct mount **mpp, int startwrite, struct thread *p)
2682 {
2683 struct mount *mp;
2684 struct ucred *credanon;
2685 fhandle_t *fhp;
2686
2687 fhp = (fhandle_t *)nfp->nfsrvfh_data;
2688 /*
2689 * Check for the special case of the nfsv4root_fh.
2690 */
2691 mp = vfs_busyfs(&fhp->fh_fsid);
2692 if (mpp != NULL)
2693 *mpp = mp;
2694 if (mp == NULL) {
2695 *vpp = NULL;
2696 nd->nd_repstat = ESTALE;
2697 goto out;
2698 }
2699
2700 if (startwrite) {
2701 vn_start_write(NULL, mpp, V_WAIT);
2702 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
2703 lktype = LK_EXCLUSIVE;
2704 }
2705 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
2706 &credanon);
2707 vfs_unbusy(mp);
2708
2709 /*
2710 * For NFSv4 without a pseudo root fs, unexported file handles
2711 * can be returned, so that Lookup works everywhere.
2712 */
2713 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
2714 !(nd->nd_flag & ND_NFSV4)) {
2715 vput(*vpp);
2716 nd->nd_repstat = EACCES;
2717 }
2718
2719 /*
2720 * Personally, I've never seen any point in requiring a
2721 * reserved port#, since only in the rare case where the
2722 * clients are all boxes with secure system priviledges,
2723 * does it provide any enhanced security, but... some people
2724 * believe it to be useful and keep putting this code back in.
2725 * (There is also some "security checker" out there that
2726 * complains if the nfs server doesn't enforce this.)
2727 * However, note the following:
2728 * RFC3530 (NFSv4) specifies that a reserved port# not be
2729 * required.
2730 * RFC2623 recommends that, if a reserved port# is checked for,
2731 * that there be a way to turn that off--> ifdef'd.
2732 */
2733 #ifdef NFS_REQRSVPORT
2734 if (!nd->nd_repstat) {
2735 struct sockaddr_in *saddr;
2736 struct sockaddr_in6 *saddr6;
2737
2738 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
2739 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
2740 if (!(nd->nd_flag & ND_NFSV4) &&
2741 ((saddr->sin_family == AF_INET &&
2742 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
2743 (saddr6->sin6_family == AF_INET6 &&
2744 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
2745 vput(*vpp);
2746 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
2747 }
2748 }
2749 #endif /* NFS_REQRSVPORT */
2750
2751 /*
2752 * Check/setup credentials.
2753 */
2754 if (!nd->nd_repstat) {
2755 nd->nd_saveduid = nd->nd_cred->cr_uid;
2756 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
2757 if (nd->nd_repstat)
2758 vput(*vpp);
2759 }
2760 if (credanon != NULL)
2761 crfree(credanon);
2762 if (nd->nd_repstat) {
2763 if (startwrite)
2764 vn_finished_write(mp);
2765 *vpp = NULL;
2766 if (mpp != NULL)
2767 *mpp = NULL;
2768 }
2769
2770 out:
2771 NFSEXITCODE2(0, nd);
2772 }
2773
2774 /*
2775 * glue for fp.
2776 */
2777 static int
fp_getfvp(struct thread * p,int fd,struct file ** fpp,struct vnode ** vpp)2778 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
2779 {
2780 struct filedesc *fdp;
2781 struct file *fp;
2782 int error = 0;
2783
2784 fdp = p->td_proc->p_fd;
2785 if (fd < 0 || fd >= fdp->fd_nfiles ||
2786 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
2787 error = EBADF;
2788 goto out;
2789 }
2790 *fpp = fp;
2791
2792 out:
2793 NFSEXITCODE(error);
2794 return (error);
2795 }
2796
2797 /*
2798 * Called from nfssvc() to update the exports list. Just call
2799 * vfs_export(). This has to be done, since the v4 root fake fs isn't
2800 * in the mount list.
2801 */
2802 int
nfsrv_v4rootexport(void * argp,struct ucred * cred,struct thread * p)2803 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
2804 {
2805 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
2806 int error = 0;
2807 struct nameidata nd;
2808 fhandle_t fh;
2809
2810 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
2811 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
2812 nfs_rootfhset = 0;
2813 else if (error == 0) {
2814 if (nfsexargp->fspec == NULL) {
2815 error = EPERM;
2816 goto out;
2817 }
2818 /*
2819 * If fspec != NULL, this is the v4root path.
2820 */
2821 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
2822 nfsexargp->fspec, p);
2823 if ((error = namei(&nd)) != 0)
2824 goto out;
2825 error = nfsvno_getfh(nd.ni_vp, &fh, p);
2826 vrele(nd.ni_vp);
2827 if (!error) {
2828 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
2829 NFSBCOPY((caddr_t)&fh,
2830 nfs_rootfh.nfsrvfh_data,
2831 sizeof (fhandle_t));
2832 nfs_rootfhset = 1;
2833 }
2834 }
2835
2836 out:
2837 NFSEXITCODE(error);
2838 return (error);
2839 }
2840
2841 /*
2842 * Get the tcp socket sequence numbers we need.
2843 * (Maybe this should be moved to the tcp sources?)
2844 */
2845 int
nfsrv_getsocksndseq(struct socket * so,tcp_seq * maxp,tcp_seq * unap)2846 nfsrv_getsocksndseq(struct socket *so, tcp_seq *maxp, tcp_seq *unap)
2847 {
2848 struct inpcb *inp;
2849 struct tcpcb *tp;
2850 int error = 0;
2851
2852 inp = sotoinpcb(so);
2853 KASSERT(inp != NULL, ("nfsrv_getsocksndseq: inp == NULL"));
2854 INP_RLOCK(inp);
2855 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
2856 INP_RUNLOCK(inp);
2857 error = EPIPE;
2858 goto out;
2859 }
2860 tp = intotcpcb(inp);
2861 if (tp->t_state != TCPS_ESTABLISHED) {
2862 INP_RUNLOCK(inp);
2863 error = EPIPE;
2864 goto out;
2865 }
2866 *maxp = tp->snd_max;
2867 *unap = tp->snd_una;
2868 INP_RUNLOCK(inp);
2869
2870 out:
2871 NFSEXITCODE(error);
2872 return (error);
2873 }
2874
2875 /*
2876 * This function needs to test to see if the system is near its limit
2877 * for memory allocation via malloc() or mget() and return True iff
2878 * either of these resources are near their limit.
2879 * XXX (For now, this is just a stub.)
2880 */
2881 int nfsrv_testmalloclimit = 0;
2882 int
nfsrv_mallocmget_limit(void)2883 nfsrv_mallocmget_limit(void)
2884 {
2885 static int printmesg = 0;
2886 static int testval = 1;
2887
2888 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
2889 if ((printmesg++ % 100) == 0)
2890 printf("nfsd: malloc/mget near limit\n");
2891 return (1);
2892 }
2893 return (0);
2894 }
2895
2896 /*
2897 * BSD specific initialization of a mount point.
2898 */
2899 void
nfsd_mntinit(void)2900 nfsd_mntinit(void)
2901 {
2902 static int inited = 0;
2903
2904 if (inited)
2905 return;
2906 inited = 1;
2907 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
2908 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
2909 TAILQ_INIT(&nfsv4root_mnt.mnt_activevnodelist);
2910 nfsv4root_mnt.mnt_export = NULL;
2911 TAILQ_INIT(&nfsv4root_opt);
2912 TAILQ_INIT(&nfsv4root_newopt);
2913 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
2914 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
2915 nfsv4root_mnt.mnt_nvnodelistsize = 0;
2916 nfsv4root_mnt.mnt_activevnodelistsize = 0;
2917 }
2918
2919 /*
2920 * Get a vnode for a file handle, without checking exports, etc.
2921 */
2922 struct vnode *
nfsvno_getvp(fhandle_t * fhp)2923 nfsvno_getvp(fhandle_t *fhp)
2924 {
2925 struct mount *mp;
2926 struct vnode *vp;
2927 int error;
2928
2929 mp = vfs_busyfs(&fhp->fh_fsid);
2930 if (mp == NULL)
2931 return (NULL);
2932 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
2933 vfs_unbusy(mp);
2934 if (error)
2935 return (NULL);
2936 return (vp);
2937 }
2938
2939 /*
2940 * Do a local VOP_ADVLOCK().
2941 */
2942 int
nfsvno_advlock(struct vnode * vp,int ftype,u_int64_t first,u_int64_t end,struct thread * td)2943 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
2944 u_int64_t end, struct thread *td)
2945 {
2946 int error = 0;
2947 struct flock fl;
2948 u_int64_t tlen;
2949
2950 if (nfsrv_dolocallocks == 0)
2951 goto out;
2952
2953 /* Check for VI_DOOMED here, so that VOP_ADVLOCK() isn't performed. */
2954 if ((vp->v_iflag & VI_DOOMED) != 0) {
2955 error = EPERM;
2956 goto out;
2957 }
2958
2959 fl.l_whence = SEEK_SET;
2960 fl.l_type = ftype;
2961 fl.l_start = (off_t)first;
2962 if (end == NFS64BITSSET) {
2963 fl.l_len = 0;
2964 } else {
2965 tlen = end - first;
2966 fl.l_len = (off_t)tlen;
2967 }
2968 /*
2969 * For FreeBSD8, the l_pid and l_sysid must be set to the same
2970 * values for all calls, so that all locks will be held by the
2971 * nfsd server. (The nfsd server handles conflicts between the
2972 * various clients.)
2973 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
2974 * bytes, so it can't be put in l_sysid.
2975 */
2976 if (nfsv4_sysid == 0)
2977 nfsv4_sysid = nlm_acquire_next_sysid();
2978 fl.l_pid = (pid_t)0;
2979 fl.l_sysid = (int)nfsv4_sysid;
2980
2981 NFSVOPUNLOCK(vp, 0);
2982 if (ftype == F_UNLCK)
2983 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
2984 (F_POSIX | F_REMOTE));
2985 else
2986 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
2987 (F_POSIX | F_REMOTE));
2988 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2989
2990 out:
2991 NFSEXITCODE(error);
2992 return (error);
2993 }
2994
2995 /*
2996 * Check the nfsv4 root exports.
2997 */
2998 int
nfsvno_v4rootexport(struct nfsrv_descript * nd)2999 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3000 {
3001 struct ucred *credanon;
3002 int exflags, error = 0, numsecflavor, *secflavors, i;
3003
3004 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3005 &credanon, &numsecflavor, &secflavors);
3006 if (error) {
3007 error = NFSERR_PROGUNAVAIL;
3008 goto out;
3009 }
3010 if (credanon != NULL)
3011 crfree(credanon);
3012 for (i = 0; i < numsecflavor; i++) {
3013 if (secflavors[i] == AUTH_SYS)
3014 nd->nd_flag |= ND_EXAUTHSYS;
3015 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3016 nd->nd_flag |= ND_EXGSS;
3017 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3018 nd->nd_flag |= ND_EXGSSINTEGRITY;
3019 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3020 nd->nd_flag |= ND_EXGSSPRIVACY;
3021 }
3022
3023 out:
3024 NFSEXITCODE(error);
3025 return (error);
3026 }
3027
3028 /*
3029 * Nfs server psuedo system call for the nfsd's
3030 */
3031 /*
3032 * MPSAFE
3033 */
3034 static int
nfssvc_nfsd(struct thread * td,struct nfssvc_args * uap)3035 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3036 {
3037 struct file *fp;
3038 struct nfsd_addsock_args sockarg;
3039 struct nfsd_nfsd_args nfsdarg;
3040 cap_rights_t rights;
3041 int error;
3042
3043 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3044 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3045 if (error)
3046 goto out;
3047 /*
3048 * Since we don't know what rights might be required,
3049 * pretend that we need them all. It is better to be too
3050 * careful than too reckless.
3051 */
3052 error = fget(td, sockarg.sock,
3053 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3054 if (error != 0)
3055 goto out;
3056 if (fp->f_type != DTYPE_SOCKET) {
3057 fdrop(fp, td);
3058 error = EPERM;
3059 goto out;
3060 }
3061 error = nfsrvd_addsock(fp);
3062 fdrop(fp, td);
3063 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3064 if (uap->argp == NULL) {
3065 error = EINVAL;
3066 goto out;
3067 }
3068 error = copyin(uap->argp, (caddr_t)&nfsdarg,
3069 sizeof (nfsdarg));
3070 if (error)
3071 goto out;
3072 error = nfsrvd_nfsd(td, &nfsdarg);
3073 } else {
3074 error = nfssvc_srvcall(td, uap, td->td_ucred);
3075 }
3076
3077 out:
3078 NFSEXITCODE(error);
3079 return (error);
3080 }
3081
3082 static int
nfssvc_srvcall(struct thread * p,struct nfssvc_args * uap,struct ucred * cred)3083 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3084 {
3085 struct nfsex_args export;
3086 struct file *fp = NULL;
3087 int stablefd, len;
3088 struct nfsd_clid adminrevoke;
3089 struct nfsd_dumplist dumplist;
3090 struct nfsd_dumpclients *dumpclients;
3091 struct nfsd_dumplocklist dumplocklist;
3092 struct nfsd_dumplocks *dumplocks;
3093 struct nameidata nd;
3094 vnode_t vp;
3095 int error = EINVAL, igotlock;
3096 struct proc *procp;
3097 static int suspend_nfsd = 0;
3098
3099 if (uap->flag & NFSSVC_PUBLICFH) {
3100 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3101 sizeof (fhandle_t));
3102 error = copyin(uap->argp,
3103 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3104 if (!error)
3105 nfs_pubfhset = 1;
3106 } else if (uap->flag & NFSSVC_V4ROOTEXPORT) {
3107 error = copyin(uap->argp,(caddr_t)&export,
3108 sizeof (struct nfsex_args));
3109 if (!error)
3110 error = nfsrv_v4rootexport(&export, cred, p);
3111 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3112 nfs_pubfhset = 0;
3113 error = 0;
3114 } else if (uap->flag & NFSSVC_STABLERESTART) {
3115 error = copyin(uap->argp, (caddr_t)&stablefd,
3116 sizeof (int));
3117 if (!error)
3118 error = fp_getfvp(p, stablefd, &fp, &vp);
3119 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3120 error = EBADF;
3121 if (!error && newnfs_numnfsd != 0)
3122 error = EPERM;
3123 if (!error) {
3124 nfsrv_stablefirst.nsf_fp = fp;
3125 nfsrv_setupstable(p);
3126 }
3127 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3128 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3129 sizeof (struct nfsd_clid));
3130 if (!error)
3131 error = nfsrv_adminrevoke(&adminrevoke, p);
3132 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3133 error = copyin(uap->argp, (caddr_t)&dumplist,
3134 sizeof (struct nfsd_dumplist));
3135 if (!error && (dumplist.ndl_size < 1 ||
3136 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3137 error = EPERM;
3138 if (!error) {
3139 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3140 dumpclients = (struct nfsd_dumpclients *)malloc(len,
3141 M_TEMP, M_WAITOK);
3142 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3143 error = copyout(dumpclients,
3144 CAST_USER_ADDR_T(dumplist.ndl_list), len);
3145 free((caddr_t)dumpclients, M_TEMP);
3146 }
3147 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3148 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3149 sizeof (struct nfsd_dumplocklist));
3150 if (!error && (dumplocklist.ndllck_size < 1 ||
3151 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3152 error = EPERM;
3153 if (!error)
3154 error = nfsrv_lookupfilename(&nd,
3155 dumplocklist.ndllck_fname, p);
3156 if (!error) {
3157 len = sizeof (struct nfsd_dumplocks) *
3158 dumplocklist.ndllck_size;
3159 dumplocks = (struct nfsd_dumplocks *)malloc(len,
3160 M_TEMP, M_WAITOK);
3161 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3162 dumplocklist.ndllck_size, p);
3163 vput(nd.ni_vp);
3164 error = copyout(dumplocks,
3165 CAST_USER_ADDR_T(dumplocklist.ndllck_list), len);
3166 free((caddr_t)dumplocks, M_TEMP);
3167 }
3168 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3169 procp = p->td_proc;
3170 PROC_LOCK(procp);
3171 nfsd_master_pid = procp->p_pid;
3172 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3173 nfsd_master_start = procp->p_stats->p_start;
3174 nfsd_master_proc = procp;
3175 PROC_UNLOCK(procp);
3176 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3177 NFSLOCKV4ROOTMUTEX();
3178 if (suspend_nfsd == 0) {
3179 /* Lock out all nfsd threads */
3180 do {
3181 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3182 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3183 } while (igotlock == 0 && suspend_nfsd == 0);
3184 suspend_nfsd = 1;
3185 }
3186 NFSUNLOCKV4ROOTMUTEX();
3187 error = 0;
3188 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3189 NFSLOCKV4ROOTMUTEX();
3190 if (suspend_nfsd != 0) {
3191 nfsv4_unlock(&nfsd_suspend_lock, 0);
3192 suspend_nfsd = 0;
3193 }
3194 NFSUNLOCKV4ROOTMUTEX();
3195 error = 0;
3196 }
3197
3198 NFSEXITCODE(error);
3199 return (error);
3200 }
3201
3202 /*
3203 * Check exports.
3204 * Returns 0 if ok, 1 otherwise.
3205 */
3206 int
nfsvno_testexp(struct nfsrv_descript * nd,struct nfsexstuff * exp)3207 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3208 {
3209 int i;
3210
3211 /*
3212 * This seems odd, but allow the case where the security flavor
3213 * list is empty. This happens when NFSv4 is traversing non-exported
3214 * file systems. Exported file systems should always have a non-empty
3215 * security flavor list.
3216 */
3217 if (exp->nes_numsecflavor == 0)
3218 return (0);
3219
3220 for (i = 0; i < exp->nes_numsecflavor; i++) {
3221 /*
3222 * The tests for privacy and integrity must be first,
3223 * since ND_GSS is set for everything but AUTH_SYS.
3224 */
3225 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3226 (nd->nd_flag & ND_GSSPRIVACY))
3227 return (0);
3228 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3229 (nd->nd_flag & ND_GSSINTEGRITY))
3230 return (0);
3231 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3232 (nd->nd_flag & ND_GSS))
3233 return (0);
3234 if (exp->nes_secflavors[i] == AUTH_SYS &&
3235 (nd->nd_flag & ND_GSS) == 0)
3236 return (0);
3237 }
3238 return (1);
3239 }
3240
3241 /*
3242 * Calculate a hash value for the fid in a file handle.
3243 */
3244 uint32_t
nfsrv_hashfh(fhandle_t * fhp)3245 nfsrv_hashfh(fhandle_t *fhp)
3246 {
3247 uint32_t hashval;
3248
3249 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3250 return (hashval);
3251 }
3252
3253 /*
3254 * Signal the userland master nfsd to backup the stable restart file.
3255 */
3256 void
nfsrv_backupstable(void)3257 nfsrv_backupstable(void)
3258 {
3259 struct proc *procp;
3260
3261 if (nfsd_master_proc != NULL) {
3262 procp = pfind(nfsd_master_pid);
3263 /* Try to make sure it is the correct process. */
3264 if (procp == nfsd_master_proc &&
3265 procp->p_stats->p_start.tv_sec ==
3266 nfsd_master_start.tv_sec &&
3267 procp->p_stats->p_start.tv_usec ==
3268 nfsd_master_start.tv_usec &&
3269 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3270 kern_psignal(procp, SIGUSR2);
3271 else
3272 nfsd_master_proc = NULL;
3273
3274 if (procp != NULL)
3275 PROC_UNLOCK(procp);
3276 }
3277 }
3278
3279 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
3280
3281 /*
3282 * Called once to initialize data structures...
3283 */
3284 static int
nfsd_modevent(module_t mod,int type,void * data)3285 nfsd_modevent(module_t mod, int type, void *data)
3286 {
3287 int error = 0, i;
3288 static int loaded = 0;
3289
3290 switch (type) {
3291 case MOD_LOAD:
3292 if (loaded)
3293 goto out;
3294 newnfs_portinit();
3295 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
3296 snprintf(nfsrchash_table[i].lock_name,
3297 sizeof(nfsrchash_table[i].lock_name), "nfsrc_tcp%d",
3298 i);
3299 mtx_init(&nfsrchash_table[i].mtx,
3300 nfsrchash_table[i].lock_name, NULL, MTX_DEF);
3301 }
3302 mtx_init(&nfsrc_udpmtx, "nfs_udpcache_mutex", NULL, MTX_DEF);
3303 mtx_init(&nfs_v4root_mutex, "nfs_v4root_mutex", NULL, MTX_DEF);
3304 mtx_init(&nfsv4root_mnt.mnt_mtx, "struct mount mtx", NULL,
3305 MTX_DEF);
3306 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
3307 nfsrvd_initcache();
3308 nfsd_init();
3309 NFSD_LOCK();
3310 nfsrvd_init(0);
3311 NFSD_UNLOCK();
3312 nfsd_mntinit();
3313 #ifdef VV_DISABLEDELEG
3314 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
3315 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
3316 #endif
3317 nfsd_call_servertimer = nfsrv_servertimer;
3318 nfsd_call_nfsd = nfssvc_nfsd;
3319 loaded = 1;
3320 break;
3321
3322 case MOD_UNLOAD:
3323 if (newnfs_numnfsd != 0) {
3324 error = EBUSY;
3325 break;
3326 }
3327
3328 #ifdef VV_DISABLEDELEG
3329 vn_deleg_ops.vndeleg_recall = NULL;
3330 vn_deleg_ops.vndeleg_disable = NULL;
3331 #endif
3332 nfsd_call_servertimer = NULL;
3333 nfsd_call_nfsd = NULL;
3334
3335 /* Clean out all NFSv4 state. */
3336 nfsrv_throwawayallstate(curthread);
3337
3338 /* Clean the NFS server reply cache */
3339 nfsrvd_cleancache();
3340
3341 /* Free up the krpc server pool. */
3342 if (nfsrvd_pool != NULL)
3343 svcpool_destroy(nfsrvd_pool);
3344
3345 /* and get rid of the locks */
3346 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++)
3347 mtx_destroy(&nfsrchash_table[i].mtx);
3348 mtx_destroy(&nfsrc_udpmtx);
3349 mtx_destroy(&nfs_v4root_mutex);
3350 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
3351 lockdestroy(&nfsv4root_mnt.mnt_explock);
3352 loaded = 0;
3353 break;
3354 default:
3355 error = EOPNOTSUPP;
3356 break;
3357 }
3358
3359 out:
3360 NFSEXITCODE(error);
3361 return (error);
3362 }
3363 static moduledata_t nfsd_mod = {
3364 "nfsd",
3365 nfsd_modevent,
3366 NULL,
3367 };
3368 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
3369
3370 /* So that loader and kldload(2) can find us, wherever we are.. */
3371 MODULE_VERSION(nfsd, 1);
3372 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
3373 MODULE_DEPEND(nfsd, nfslock, 1, 1, 1);
3374 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
3375 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
3376 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);
3377
3378