1 /*-
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/9.2/sys/nfsserver/nfs_serv.c 244658 2012-12-24 13:22:32Z kib $");
37
38 /*
39 * nfs version 2 and 3 server calls to vnode ops
40 * - these routines generally have 3 phases
41 * 1 - break down and validate rpc request in mbuf list
42 * 2 - do the vnode ops for the request
43 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
44 * 3 - build the rpc reply in an mbuf list
45 * nb:
46 * - do not mix the phases, since the nfsm_?? macros can return failures
47 * on a bad rpc or similar and do not do any vrele() or vput()'s
48 *
49 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
50 * error number iff error != 0 whereas
51 * returning an error from the server function implies a fatal error
52 * such as a badly constructed rpc request that should be dropped without
53 * a reply.
54 * For nfsm_reply(), the case where error == EBADRPC is treated
55 * specially; after constructing a reply, it does an immediate
56 * `goto nfsmout' to avoid getting any V3 post-op status appended.
57 *
58 * Other notes:
59 * Warning: always pay careful attention to resource cleanup on return
60 * and note that nfsm_*() macros can terminate a procedure on certain
61 * errors.
62 *
63 * lookup() and namei()
64 * may return garbage in various structural fields/return elements
65 * if an error is returned, and may garbage up nd.ni_dvp even if no
66 * error is returned and you did not request LOCKPARENT or WANTPARENT.
67 *
68 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
69 * buffer has been freed or not.
70 */
71
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/proc.h>
75 #include <sys/namei.h>
76 #include <sys/unistd.h>
77 #include <sys/vnode.h>
78 #include <sys/mount.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/malloc.h>
82 #include <sys/mbuf.h>
83 #include <sys/priv.h>
84 #include <sys/dirent.h>
85 #include <sys/stat.h>
86 #include <sys/kernel.h>
87 #include <sys/sysctl.h>
88 #include <sys/bio.h>
89 #include <sys/buf.h>
90
91 #include <vm/vm.h>
92 #include <vm/vm_extern.h>
93 #include <vm/vm_object.h>
94
95 #include <nfs/nfsproto.h>
96 #include <nfsserver/nfs.h>
97 #include <nfs/xdr_subs.h>
98 #include <nfsserver/nfsm_subs.h>
99
100 FEATURE(nfsserver, "NFS server");
101
102 #ifdef NFSRV_DEBUG
103 #define nfsdbprintf(info) printf info
104 #else
105 #define nfsdbprintf(info)
106 #endif
107
108 #define MAX_COMMIT_COUNT (1024 * 1024)
109
110 #define MAX_REORDERED_RPC 16
111 #define NUM_HEURISTIC 1031
112 #define NHUSE_INIT 64
113 #define NHUSE_INC 16
114 #define NHUSE_MAX 2048
115
116 static struct nfsheur {
117 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
118 off_t nh_nextoff; /* next offset for sequential detection */
119 int nh_use; /* use count for selection */
120 int nh_seqcount; /* heuristic */
121 } nfsheur[NUM_HEURISTIC];
122
123 /* Global vars */
124
125 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
126 int nfsrvw_procrastinate_v3 = 0;
127
128 static struct timeval nfsver = { 0 };
129
130 SYSCTL_NODE(_vfs, OID_AUTO, nfsrv, CTLFLAG_RW, 0, "NFS server");
131
132 static int nfs_async;
133 static int nfs_commit_blks;
134 static int nfs_commit_miss;
135 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0,
136 "Tell client that writes were synced even though they were not");
137 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0,
138 "Number of completed commits");
139 SYSCTL_INT(_vfs_nfsrv, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
140
141 struct nfsrvstats nfsrvstats;
142 SYSCTL_STRUCT(_vfs_nfsrv, NFS_NFSRVSTATS, nfsrvstats, CTLFLAG_RW,
143 &nfsrvstats, nfsrvstats, "S,nfsrvstats");
144
145 static int nfsrv_access(struct vnode *, accmode_t, struct ucred *,
146 int, int);
147
148 /*
149 * Clear nameidata fields that are tested in nsfmout cleanup code prior
150 * to using first nfsm macro (that might jump to the cleanup code).
151 */
152
153 static __inline void
154 ndclear(struct nameidata *nd)
155 {
156
157 nd->ni_cnd.cn_flags = 0;
158 nd->ni_vp = NULL;
159 nd->ni_dvp = NULL;
160 nd->ni_startdir = NULL;
161 nd->ni_strictrelative = 0;
162 }
163
164 /*
165 * Takes two vfslocked integers and returns with at most one
166 * reference to giant. The return value indicates whether giant
167 * is held by either lock. This simplifies nfsrv ops by allowing
168 * them to track only one vfslocked var.
169 */
170 static __inline int
171 nfsrv_lockedpair(int vfs1, int vfs2)
172 {
173
174 if (vfs1 && vfs2)
175 VFS_UNLOCK_GIANT(vfs2);
176
177 return (vfs1 | vfs2);
178 }
179
180 static __inline int
181 nfsrv_lockedpair_nd(int vfs1, struct nameidata *nd)
182 {
183 int vfs2;
184
185 vfs2 = NDHASGIANT(nd);
186
187 return nfsrv_lockedpair(vfs1, vfs2);
188 }
189
190 /*
191 * Heuristic to detect sequential operation.
192 */
193 static struct nfsheur *
194 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
195 {
196 struct nfsheur *nh;
197 int hi, try;
198
199 /* Locate best candidate. */
200 try = 32;
201 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
202 nh = &nfsheur[hi];
203 while (try--) {
204 if (nfsheur[hi].nh_vp == vp) {
205 nh = &nfsheur[hi];
206 break;
207 }
208 if (nfsheur[hi].nh_use > 0)
209 --nfsheur[hi].nh_use;
210 hi = (hi + 1) % NUM_HEURISTIC;
211 if (nfsheur[hi].nh_use < nh->nh_use)
212 nh = &nfsheur[hi];
213 }
214
215 /* Initialize hint if this is a new file. */
216 if (nh->nh_vp != vp) {
217 nh->nh_vp = vp;
218 nh->nh_nextoff = uio->uio_offset;
219 nh->nh_use = NHUSE_INIT;
220 if (uio->uio_offset == 0)
221 nh->nh_seqcount = 4;
222 else
223 nh->nh_seqcount = 1;
224 }
225
226 /* Calculate heuristic. */
227 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
228 uio->uio_offset == nh->nh_nextoff) {
229 /* See comments in vfs_vnops.c:sequential_heuristic(). */
230 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
231 if (nh->nh_seqcount > IO_SEQMAX)
232 nh->nh_seqcount = IO_SEQMAX;
233 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
234 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
235 /* Probably a reordered RPC, leave seqcount alone. */
236 } else if (nh->nh_seqcount > 1) {
237 nh->nh_seqcount /= 2;
238 } else {
239 nh->nh_seqcount = 0;
240 }
241 nh->nh_use += NHUSE_INC;
242 if (nh->nh_use > NHUSE_MAX)
243 nh->nh_use = NHUSE_MAX;
244 return (nh);
245 }
246
247 /*
248 * nfs v3 access service
249 */
250 int
251 nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
252 struct mbuf **mrq)
253 {
254 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
255 struct sockaddr *nam = nfsd->nd_nam;
256 caddr_t dpos = nfsd->nd_dpos;
257 struct ucred *cred = nfsd->nd_cr;
258 struct vnode *vp = NULL;
259 nfsfh_t nfh;
260 fhandle_t *fhp;
261 u_int32_t *tl;
262 caddr_t bpos;
263 int error = 0, rdonly, getret;
264 struct mbuf *mb, *mreq;
265 struct vattr vattr, *vap = &vattr;
266 u_long testmode, nfsmode;
267 int v3 = (nfsd->nd_flag & ND_NFSV3);
268 int vfslocked;
269
270 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
271 if (!v3)
272 panic("nfsrv3_access: v3 proc called on a v2 connection");
273 vfslocked = 0;
274 fhp = &nfh.fh_generic;
275 nfsm_srvmtofh(fhp);
276 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
277 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
278 if (error) {
279 nfsm_reply(NFSX_UNSIGNED);
280 nfsm_srvpostop_attr(1, NULL);
281 error = 0;
282 goto nfsmout;
283 }
284 nfsmode = fxdr_unsigned(u_int32_t, *tl);
285 if ((nfsmode & NFSV3ACCESS_READ) &&
286 nfsrv_access(vp, VREAD, cred, rdonly, 0))
287 nfsmode &= ~NFSV3ACCESS_READ;
288 if (vp->v_type == VDIR)
289 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
290 NFSV3ACCESS_DELETE);
291 else
292 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
293 if ((nfsmode & testmode) &&
294 nfsrv_access(vp, VWRITE, cred, rdonly, 0))
295 nfsmode &= ~testmode;
296 if (vp->v_type == VDIR)
297 testmode = NFSV3ACCESS_LOOKUP;
298 else
299 testmode = NFSV3ACCESS_EXECUTE;
300 if ((nfsmode & testmode) &&
301 nfsrv_access(vp, VEXEC, cred, rdonly, 0))
302 nfsmode &= ~testmode;
303 getret = VOP_GETATTR(vp, vap, cred);
304 vput(vp);
305 vp = NULL;
306 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
307 nfsm_srvpostop_attr(getret, vap);
308 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
309 *tl = txdr_unsigned(nfsmode);
310 nfsmout:
311 if (vp)
312 vput(vp);
313 VFS_UNLOCK_GIANT(vfslocked);
314 return(error);
315 }
316
317 /*
318 * nfs getattr service
319 */
320 int
321 nfsrv_getattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
322 struct mbuf **mrq)
323 {
324 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
325 struct sockaddr *nam = nfsd->nd_nam;
326 caddr_t dpos = nfsd->nd_dpos;
327 struct ucred *cred = nfsd->nd_cr;
328 struct nfs_fattr *fp;
329 struct vattr va;
330 struct vattr *vap = &va;
331 struct vnode *vp = NULL;
332 nfsfh_t nfh;
333 fhandle_t *fhp;
334 caddr_t bpos;
335 int error = 0, rdonly;
336 struct mbuf *mb, *mreq;
337 int vfslocked;
338
339 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
340 vfslocked = 0;
341 fhp = &nfh.fh_generic;
342 nfsm_srvmtofh(fhp);
343 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
344 if (error) {
345 nfsm_reply(0);
346 error = 0;
347 goto nfsmout;
348 }
349 error = VOP_GETATTR(vp, vap, cred);
350 vput(vp);
351 vp = NULL;
352 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
353 if (error) {
354 error = 0;
355 goto nfsmout;
356 }
357 fp = nfsm_build(struct nfs_fattr *,
358 NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
359 nfsm_srvfillattr(vap, fp);
360 /* fall through */
361
362 nfsmout:
363 if (vp)
364 vput(vp);
365 VFS_UNLOCK_GIANT(vfslocked);
366 return(error);
367 }
368
369 /*
370 * nfs setattr service
371 */
372 int
373 nfsrv_setattr(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
374 struct mbuf **mrq)
375 {
376 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
377 struct sockaddr *nam = nfsd->nd_nam;
378 caddr_t dpos = nfsd->nd_dpos;
379 struct ucred *cred = nfsd->nd_cr;
380 struct vattr va, preat;
381 struct vattr *vap = &va;
382 struct nfsv2_sattr *sp;
383 struct nfs_fattr *fp;
384 struct vnode *vp = NULL;
385 nfsfh_t nfh;
386 fhandle_t *fhp;
387 u_int32_t *tl;
388 caddr_t bpos;
389 int error = 0, rdonly, preat_ret = 1, postat_ret = 1;
390 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
391 struct mbuf *mb, *mreq;
392 struct timespec guard = { 0, 0 };
393 struct mount *mp = NULL;
394 int tvfslocked;
395 int vfslocked;
396
397 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
398 vfslocked = 0;
399 fhp = &nfh.fh_generic;
400 nfsm_srvmtofh(fhp);
401 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
402 error = ESTALE;
403 goto out;
404 }
405 vfslocked = VFS_LOCK_GIANT(mp);
406 (void) vn_start_write(NULL, &mp, V_WAIT);
407 vfs_rel(mp); /* The write holds a ref. */
408 VATTR_NULL(vap);
409 if (v3) {
410 nfsm_srvsattr(vap);
411 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
412 gcheck = fxdr_unsigned(int, *tl);
413 if (gcheck) {
414 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
415 fxdr_nfsv3time(tl, &guard);
416 }
417 } else {
418 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
419 /*
420 * Nah nah nah nah na nah
421 * There is a bug in the Sun client that puts 0xffff in the mode
422 * field of sattr when it should put in 0xffffffff. The u_short
423 * doesn't sign extend.
424 * --> check the low order 2 bytes for 0xffff
425 */
426 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
427 vap->va_mode = nfstov_mode(sp->sa_mode);
428 if (sp->sa_uid != nfsrv_nfs_xdrneg1)
429 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
430 if (sp->sa_gid != nfsrv_nfs_xdrneg1)
431 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
432 if (sp->sa_size != nfsrv_nfs_xdrneg1)
433 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
434 if (sp->sa_atime.nfsv2_sec != nfsrv_nfs_xdrneg1) {
435 #ifdef notyet
436 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
437 #else
438 vap->va_atime.tv_sec =
439 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
440 vap->va_atime.tv_nsec = 0;
441 #endif
442 }
443 if (sp->sa_mtime.nfsv2_sec != nfsrv_nfs_xdrneg1)
444 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
445
446 }
447
448 /*
449 * Now that we have all the fields, lets do it.
450 */
451 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly);
452 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
453 if (error) {
454 nfsm_reply(2 * NFSX_UNSIGNED);
455 if (v3)
456 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
457 error = 0;
458 goto nfsmout;
459 }
460
461 /*
462 * vp now an active resource, pay careful attention to cleanup
463 */
464 if (v3) {
465 error = preat_ret = VOP_GETATTR(vp, &preat, cred);
466 if (!error && gcheck &&
467 (preat.va_ctime.tv_sec != guard.tv_sec ||
468 preat.va_ctime.tv_nsec != guard.tv_nsec))
469 error = NFSERR_NOT_SYNC;
470 if (error) {
471 vput(vp);
472 vp = NULL;
473 nfsm_reply(NFSX_WCCDATA(v3));
474 if (v3)
475 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
476 error = 0;
477 goto nfsmout;
478 }
479 }
480
481 /*
482 * If the size is being changed write acces is required, otherwise
483 * just check for a read only filesystem.
484 */
485 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
486 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
487 error = EROFS;
488 goto out;
489 }
490 } else {
491 if (vp->v_type == VDIR) {
492 error = EISDIR;
493 goto out;
494 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
495 0)) != 0)
496 goto out;
497 }
498 error = VOP_SETATTR(vp, vap, cred);
499 postat_ret = VOP_GETATTR(vp, vap, cred);
500 if (!error)
501 error = postat_ret;
502 out:
503 if (vp != NULL)
504 vput(vp);
505
506 vp = NULL;
507 nfsm_reply(NFSX_WCCORFATTR(v3));
508 if (v3) {
509 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
510 } else if (!error) {
511 /* v2 non-error case. */
512 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
513 nfsm_srvfillattr(vap, fp);
514 }
515 error = 0;
516 /* fall through */
517
518 nfsmout:
519 if (vp)
520 vput(vp);
521 vn_finished_write(mp);
522 VFS_UNLOCK_GIANT(vfslocked);
523 return(error);
524 }
525
526 /*
527 * nfs lookup rpc
528 */
529 int
530 nfsrv_lookup(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
531 struct mbuf **mrq)
532 {
533 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
534 struct sockaddr *nam = nfsd->nd_nam;
535 caddr_t dpos = nfsd->nd_dpos;
536 struct ucred *cred = nfsd->nd_cr;
537 struct nfs_fattr *fp;
538 struct nameidata nd, ind, *ndp = &nd;
539 struct vnode *vp, *dirp = NULL;
540 nfsfh_t nfh;
541 fhandle_t *fhp;
542 caddr_t bpos;
543 int error = 0, len, dirattr_ret = 1;
544 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
545 struct mbuf *mb, *mreq;
546 struct vattr va, dirattr, *vap = &va;
547 int tvfslocked;
548 int vfslocked;
549
550 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
551 ndclear(&nd);
552 vfslocked = 0;
553
554 fhp = &nfh.fh_generic;
555 nfsm_srvmtofh(fhp);
556 nfsm_srvnamesiz(len);
557
558 pubflag = nfs_ispublicfh(fhp);
559
560 nd.ni_cnd.cn_cred = cred;
561 nd.ni_cnd.cn_nameiop = LOOKUP;
562 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART | MPSAFE;
563 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
564 &dirp, v3, &dirattr, &dirattr_ret, pubflag);
565 vfslocked = NDHASGIANT(&nd);
566
567 /*
568 * namei failure, only dirp to cleanup. Clear out garbarge from
569 * structure in case macros jump to nfsmout.
570 */
571
572 if (error) {
573 if (dirp) {
574 vrele(dirp);
575 dirp = NULL;
576 }
577 nfsm_reply(NFSX_POSTOPATTR(v3));
578 if (v3)
579 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
580 error = 0;
581 goto nfsmout;
582 }
583
584 /*
585 * Locate index file for public filehandle
586 *
587 * error is 0 on entry and 0 on exit from this block.
588 */
589
590 if (pubflag) {
591 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
592 /*
593 * Setup call to lookup() to see if we can find
594 * the index file. Arguably, this doesn't belong
595 * in a kernel.. Ugh. If an error occurs, do not
596 * try to install an index file and then clear the
597 * error.
598 *
599 * When we replace nd with ind and redirect ndp,
600 * maintenance of ni_startdir and ni_vp shift to
601 * ind and we have to clean them up in the old nd.
602 * However, the cnd resource continues to be maintained
603 * via the original nd. Confused? You aren't alone!
604 */
605 ind = nd;
606 VOP_UNLOCK(nd.ni_vp, 0);
607 ind.ni_pathlen = strlen(nfs_pub.np_index);
608 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
609 nfs_pub.np_index;
610 ind.ni_startdir = nd.ni_vp;
611 VREF(ind.ni_startdir);
612 ind.ni_cnd.cn_flags &= ~GIANTHELD;
613 tvfslocked = VFS_LOCK_GIANT(ind.ni_startdir->v_mount);
614 if (tvfslocked)
615 nd.ni_cnd.cn_flags |= GIANTHELD;
616 error = lookup(&ind);
617 ind.ni_dvp = NULL;
618 vfslocked = nfsrv_lockedpair_nd(vfslocked, &ind);
619 ind.ni_cnd.cn_flags &= ~GIANTHELD;
620
621 if (error == 0) {
622 /*
623 * Found an index file. Get rid of
624 * the old references. transfer nd.ni_vp'
625 */
626 if (dirp)
627 vrele(dirp);
628 dirp = nd.ni_vp;
629 nd.ni_vp = NULL;
630 vrele(nd.ni_startdir);
631 nd.ni_startdir = NULL;
632 ndp = &ind;
633 }
634 error = 0;
635 }
636 /*
637 * If the public filehandle was used, check that this lookup
638 * didn't result in a filehandle outside the publicly exported
639 * filesystem. We clear the poor vp here to avoid lockups due
640 * to NFS I/O.
641 */
642
643 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
644 vput(nd.ni_vp);
645 nd.ni_vp = NULL;
646 error = EPERM;
647 }
648 }
649
650 /*
651 * Resources at this point:
652 * ndp->ni_vp may not be NULL
653 */
654
655 if (error) {
656 nfsm_reply(NFSX_POSTOPATTR(v3));
657 if (v3)
658 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
659 error = 0;
660 goto nfsmout;
661 }
662
663 /*
664 * Get underlying attribute, then release remaining resources ( for
665 * the same potential blocking reason ) and reply.
666 */
667 vp = ndp->ni_vp;
668 bzero((caddr_t)fhp, sizeof(nfh));
669 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
670 error = VOP_VPTOFH(vp, &fhp->fh_fid);
671 if (!error)
672 error = VOP_GETATTR(vp, vap, cred);
673
674 vput(vp);
675 vrele(ndp->ni_startdir);
676 vrele(dirp);
677 ndp->ni_vp = NULL;
678 ndp->ni_startdir = NULL;
679 dirp = NULL;
680 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
681 if (error) {
682 if (v3)
683 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
684 error = 0;
685 goto nfsmout;
686 }
687 nfsm_srvfhtom(fhp, v3);
688 if (v3) {
689 nfsm_srvpostop_attr(0, vap);
690 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
691 } else {
692 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
693 nfsm_srvfillattr(vap, fp);
694 }
695
696 nfsmout:
697 if (ndp->ni_vp || dirp || ndp->ni_startdir) {
698 if (ndp->ni_vp)
699 vput(ndp->ni_vp);
700 if (dirp)
701 vrele(dirp);
702 if (ndp->ni_startdir)
703 vrele(ndp->ni_startdir);
704 }
705 NDFREE(&nd, NDF_ONLY_PNBUF);
706 VFS_UNLOCK_GIANT(vfslocked);
707 return (error);
708 }
709
710 /*
711 * nfs readlink service
712 */
713 int
714 nfsrv_readlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
715 struct mbuf **mrq)
716 {
717 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
718 struct sockaddr *nam = nfsd->nd_nam;
719 caddr_t dpos = nfsd->nd_dpos;
720 struct ucred *cred = nfsd->nd_cr;
721 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
722 struct iovec *ivp = iv;
723 struct mbuf *mp;
724 u_int32_t *tl;
725 caddr_t bpos;
726 int error = 0, rdonly, i, tlen, len, getret;
727 int v3 = (nfsd->nd_flag & ND_NFSV3);
728 struct mbuf *mb, *mp3, *nmp, *mreq;
729 struct vnode *vp = NULL;
730 struct vattr attr;
731 nfsfh_t nfh;
732 fhandle_t *fhp;
733 struct uio io, *uiop = &io;
734 int vfslocked;
735
736 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
737 vfslocked = 0;
738 #ifndef nolint
739 mp = NULL;
740 #endif
741 mp3 = NULL;
742 fhp = &nfh.fh_generic;
743 nfsm_srvmtofh(fhp);
744 len = 0;
745 i = 0;
746 while (len < NFS_MAXPATHLEN) {
747 MGET(nmp, M_WAIT, MT_DATA);
748 MCLGET(nmp, M_WAIT);
749 nmp->m_len = NFSMSIZ(nmp);
750 if (len == 0)
751 mp3 = mp = nmp;
752 else {
753 mp->m_next = nmp;
754 mp = nmp;
755 }
756 if ((len + mp->m_len) > NFS_MAXPATHLEN) {
757 mp->m_len = NFS_MAXPATHLEN - len;
758 len = NFS_MAXPATHLEN;
759 } else
760 len += mp->m_len;
761 ivp->iov_base = mtod(mp, caddr_t);
762 ivp->iov_len = mp->m_len;
763 i++;
764 ivp++;
765 }
766 uiop->uio_iov = iv;
767 uiop->uio_iovcnt = i;
768 uiop->uio_offset = 0;
769 uiop->uio_resid = len;
770 uiop->uio_rw = UIO_READ;
771 uiop->uio_segflg = UIO_SYSSPACE;
772 uiop->uio_td = NULL;
773 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
774 if (error) {
775 nfsm_reply(2 * NFSX_UNSIGNED);
776 if (v3)
777 nfsm_srvpostop_attr(1, NULL);
778 error = 0;
779 goto nfsmout;
780 }
781 if (vp->v_type != VLNK) {
782 if (v3)
783 error = EINVAL;
784 else
785 error = ENXIO;
786 } else
787 error = VOP_READLINK(vp, uiop, cred);
788 getret = VOP_GETATTR(vp, &attr, cred);
789 vput(vp);
790 vp = NULL;
791 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
792 if (v3)
793 nfsm_srvpostop_attr(getret, &attr);
794 if (error) {
795 error = 0;
796 goto nfsmout;
797 }
798 if (uiop->uio_resid > 0) {
799 len -= uiop->uio_resid;
800 tlen = nfsm_rndup(len);
801 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
802 }
803 tl = nfsm_build(u_int32_t *, NFSX_UNSIGNED);
804 *tl = txdr_unsigned(len);
805 mb->m_next = mp3;
806 mp3 = NULL;
807 nfsmout:
808 if (mp3)
809 m_freem(mp3);
810 if (vp)
811 vput(vp);
812 VFS_UNLOCK_GIANT(vfslocked);
813 return(error);
814 }
815
816 /*
817 * nfs read service
818 */
819 int
820 nfsrv_read(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
821 struct mbuf **mrq)
822 {
823 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
824 struct sockaddr *nam = nfsd->nd_nam;
825 caddr_t dpos = nfsd->nd_dpos;
826 struct ucred *cred = nfsd->nd_cr;
827 struct iovec *iv;
828 struct iovec *iv2;
829 struct mbuf *m;
830 struct nfs_fattr *fp;
831 u_int32_t *tl;
832 int i;
833 caddr_t bpos;
834 int error = 0, rdonly, cnt, len, left, siz, tlen, getret;
835 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
836 struct mbuf *mb, *mreq;
837 struct mbuf *m2;
838 struct vnode *vp = NULL;
839 nfsfh_t nfh;
840 fhandle_t *fhp;
841 struct uio io, *uiop = &io;
842 struct vattr va, *vap = &va;
843 struct nfsheur *nh;
844 off_t off;
845 int ioflag = 0;
846 int vfslocked;
847
848
849 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
850 vfslocked = 0;
851 fhp = &nfh.fh_generic;
852 nfsm_srvmtofh(fhp);
853 if (v3) {
854 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
855 off = fxdr_hyper(tl);
856 } else {
857 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
858 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
859 }
860 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
861
862 /*
863 * Reference vp. If an error occurs, vp will be invalid, but we
864 * have to NULL it just in case. The macros might goto nfsmout
865 * as well.
866 */
867
868 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
869 if (error) {
870 vp = NULL;
871 nfsm_reply(2 * NFSX_UNSIGNED);
872 if (v3)
873 nfsm_srvpostop_attr(1, NULL);
874 error = 0;
875 goto nfsmout;
876 }
877
878 if (vp->v_type != VREG) {
879 if (v3)
880 error = EINVAL;
881 else
882 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
883 }
884 if (!error) {
885 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, 1)) != 0)
886 error = nfsrv_access(vp, VEXEC, cred, rdonly, 1);
887 }
888 getret = VOP_GETATTR(vp, vap, cred);
889 if (!error)
890 error = getret;
891 if (error) {
892 vput(vp);
893 vp = NULL;
894 nfsm_reply(NFSX_POSTOPATTR(v3));
895 if (v3)
896 nfsm_srvpostop_attr(getret, vap);
897 error = 0;
898 goto nfsmout;
899 }
900
901 /*
902 * Calculate byte count to read
903 */
904 if (off >= vap->va_size)
905 cnt = 0;
906 else if ((off + reqlen) > vap->va_size)
907 cnt = vap->va_size - off;
908 else
909 cnt = reqlen;
910
911 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
912 if (v3) {
913 tl = nfsm_build(u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
914 *tl++ = nfsrv_nfs_true;
915 fp = (struct nfs_fattr *)tl;
916 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
917 } else {
918 tl = nfsm_build(u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
919 fp = (struct nfs_fattr *)tl;
920 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
921 }
922 len = left = nfsm_rndup(cnt);
923 if (cnt > 0) {
924 /*
925 * Generate the mbuf list with the uio_iov ref. to it.
926 */
927 i = 0;
928 m = m2 = mb;
929 while (left > 0) {
930 siz = min(M_TRAILINGSPACE(m), left);
931 if (siz > 0) {
932 left -= siz;
933 i++;
934 }
935 if (left > 0) {
936 MGET(m, M_WAIT, MT_DATA);
937 MCLGET(m, M_WAIT);
938 m->m_len = 0;
939 m2->m_next = m;
940 m2 = m;
941 }
942 }
943 iv = malloc(i * sizeof (struct iovec),
944 M_TEMP, M_WAITOK);
945 uiop->uio_iov = iv2 = iv;
946 m = mb;
947 left = len;
948 i = 0;
949 while (left > 0) {
950 if (m == NULL)
951 panic("nfsrv_read iov");
952 siz = min(M_TRAILINGSPACE(m), left);
953 if (siz > 0) {
954 iv->iov_base = mtod(m, caddr_t) + m->m_len;
955 iv->iov_len = siz;
956 m->m_len += siz;
957 left -= siz;
958 iv++;
959 i++;
960 }
961 m = m->m_next;
962 }
963 uiop->uio_iovcnt = i;
964 uiop->uio_offset = off;
965 uiop->uio_resid = len;
966 uiop->uio_rw = UIO_READ;
967 uiop->uio_segflg = UIO_SYSSPACE;
968 nh = nfsrv_sequential_heuristic(uiop, vp);
969 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
970 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
971 if (error == 0)
972 nh->nh_nextoff = uiop->uio_offset;
973 free((caddr_t)iv2, M_TEMP);
974 if (error || (getret = VOP_GETATTR(vp, vap, cred))) {
975 if (!error)
976 error = getret;
977 m_freem(mreq);
978 vput(vp);
979 vp = NULL;
980 nfsm_reply(NFSX_POSTOPATTR(v3));
981 if (v3)
982 nfsm_srvpostop_attr(getret, vap);
983 error = 0;
984 goto nfsmout;
985 }
986 } else
987 uiop->uio_resid = 0;
988 vput(vp);
989 vp = NULL;
990 nfsm_srvfillattr(vap, fp);
991 tlen = len - uiop->uio_resid;
992 cnt = cnt < tlen ? cnt : tlen;
993 tlen = nfsm_rndup(cnt);
994 if (len != tlen || tlen != cnt)
995 nfsm_adj(mb, len - tlen, tlen - cnt);
996 if (v3) {
997 *tl++ = txdr_unsigned(cnt);
998 if (cnt < reqlen)
999 *tl++ = nfsrv_nfs_true;
1000 else
1001 *tl++ = nfsrv_nfs_false;
1002 }
1003 *tl = txdr_unsigned(cnt);
1004 nfsmout:
1005 if (vp)
1006 vput(vp);
1007 VFS_UNLOCK_GIANT(vfslocked);
1008 return(error);
1009 }
1010
1011 /*
1012 * nfs write service
1013 */
1014 int
1015 nfsrv_write(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1016 struct mbuf **mrq)
1017 {
1018 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1019 struct sockaddr *nam = nfsd->nd_nam;
1020 caddr_t dpos = nfsd->nd_dpos;
1021 struct ucred *cred = nfsd->nd_cr;
1022 struct iovec *ivp;
1023 int i, cnt;
1024 struct mbuf *mp;
1025 struct nfs_fattr *fp;
1026 struct iovec *iv;
1027 struct vattr va, forat;
1028 struct vattr *vap = &va;
1029 u_int32_t *tl;
1030 caddr_t bpos;
1031 int error = 0, rdonly, len, forat_ret = 1;
1032 int ioflags, aftat_ret = 1, retlen = 0, zeroing, adjust;
1033 int stable = NFSV3WRITE_FILESYNC;
1034 int v3 = (nfsd->nd_flag & ND_NFSV3);
1035 struct mbuf *mb, *mreq;
1036 struct vnode *vp = NULL;
1037 struct nfsheur *nh;
1038 nfsfh_t nfh;
1039 fhandle_t *fhp;
1040 struct uio io, *uiop = &io;
1041 off_t off;
1042 struct mount *mntp = NULL;
1043 int tvfslocked;
1044 int vfslocked;
1045
1046 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1047 vfslocked = 0;
1048 if (mrep == NULL) {
1049 *mrq = NULL;
1050 error = 0;
1051 goto nfsmout;
1052 }
1053 fhp = &nfh.fh_generic;
1054 nfsm_srvmtofh(fhp);
1055 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1056 error = ESTALE;
1057 goto ereply;
1058 }
1059 vfslocked = VFS_LOCK_GIANT(mntp);
1060 (void) vn_start_write(NULL, &mntp, V_WAIT);
1061 vfs_rel(mntp); /* The write holds a ref. */
1062 if (v3) {
1063 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
1064 off = fxdr_hyper(tl);
1065 tl += 3;
1066 stable = fxdr_unsigned(int, *tl++);
1067 } else {
1068 tl = nfsm_dissect_nonblock(u_int32_t *, 4 * NFSX_UNSIGNED);
1069 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1070 tl += 2;
1071 if (nfs_async)
1072 stable = NFSV3WRITE_UNSTABLE;
1073 }
1074 retlen = len = fxdr_unsigned(int32_t, *tl);
1075 cnt = i = 0;
1076
1077 /*
1078 * For NFS Version 2, it is not obvious what a write of zero length
1079 * should do, but I might as well be consistent with Version 3,
1080 * which is to return ok so long as there are no permission problems.
1081 */
1082 if (len > 0) {
1083 zeroing = 1;
1084 mp = mrep;
1085 while (mp) {
1086 if (mp == md) {
1087 zeroing = 0;
1088 adjust = dpos - mtod(mp, caddr_t);
1089 mp->m_len -= adjust;
1090 if (mp->m_len > 0 && adjust > 0)
1091 mp->m_data += adjust;
1092 }
1093 if (zeroing)
1094 mp->m_len = 0;
1095 else if (mp->m_len > 0) {
1096 i += mp->m_len;
1097 if (i > len) {
1098 mp->m_len -= (i - len);
1099 zeroing = 1;
1100 }
1101 if (mp->m_len > 0)
1102 cnt++;
1103 }
1104 mp = mp->m_next;
1105 }
1106 }
1107 if (len > NFS_MAXDATA || len < 0 || i < len) {
1108 error = EIO;
1109 nfsm_reply(2 * NFSX_UNSIGNED);
1110 if (v3)
1111 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1112 error = 0;
1113 goto nfsmout;
1114 }
1115 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly);
1116 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
1117 if (error) {
1118 vp = NULL;
1119 nfsm_reply(2 * NFSX_UNSIGNED);
1120 if (v3)
1121 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1122 error = 0;
1123 goto nfsmout;
1124 }
1125 if (v3)
1126 forat_ret = VOP_GETATTR(vp, &forat, cred);
1127 if (vp->v_type != VREG) {
1128 if (v3)
1129 error = EINVAL;
1130 else
1131 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1132 }
1133 if (!error)
1134 error = nfsrv_access(vp, VWRITE, cred, rdonly, 1);
1135 if (error) {
1136 vput(vp);
1137 vp = NULL;
1138 nfsm_reply(NFSX_WCCDATA(v3));
1139 if (v3)
1140 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1141 error = 0;
1142 goto nfsmout;
1143 }
1144
1145 if (len > 0) {
1146 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
1147 M_WAITOK);
1148 uiop->uio_iov = iv = ivp;
1149 uiop->uio_iovcnt = cnt;
1150 mp = mrep;
1151 while (mp) {
1152 if (mp->m_len > 0) {
1153 ivp->iov_base = mtod(mp, caddr_t);
1154 ivp->iov_len = mp->m_len;
1155 ivp++;
1156 }
1157 mp = mp->m_next;
1158 }
1159
1160 /*
1161 * XXX
1162 * The IO_METASYNC flag indicates that all metadata (and not just
1163 * enough to ensure data integrity) mus be written to stable storage
1164 * synchronously.
1165 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1166 */
1167 if (stable == NFSV3WRITE_UNSTABLE)
1168 ioflags = IO_NODELOCKED;
1169 else if (stable == NFSV3WRITE_DATASYNC)
1170 ioflags = (IO_SYNC | IO_NODELOCKED);
1171 else
1172 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1173 uiop->uio_resid = len;
1174 uiop->uio_rw = UIO_WRITE;
1175 uiop->uio_segflg = UIO_SYSSPACE;
1176 uiop->uio_td = NULL;
1177 uiop->uio_offset = off;
1178 nh = nfsrv_sequential_heuristic(uiop, vp);
1179 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
1180 error = VOP_WRITE(vp, uiop, ioflags, cred);
1181 if (error == 0)
1182 nh->nh_nextoff = uiop->uio_offset;
1183 /* Unlocked write. */
1184 nfsrvstats.srvvop_writes++;
1185 free((caddr_t)iv, M_TEMP);
1186 }
1187 aftat_ret = VOP_GETATTR(vp, vap, cred);
1188 vput(vp);
1189 vp = NULL;
1190 if (!error)
1191 error = aftat_ret;
1192 ereply:
1193 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1194 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1195 if (v3) {
1196 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1197 if (error) {
1198 error = 0;
1199 goto nfsmout;
1200 }
1201 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
1202 *tl++ = txdr_unsigned(retlen);
1203 /*
1204 * If nfs_async is set, then pretend the write was FILESYNC.
1205 */
1206 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1207 *tl++ = txdr_unsigned(stable);
1208 else
1209 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1210 /*
1211 * Actually, there is no need to txdr these fields,
1212 * but it may make the values more human readable,
1213 * for debugging purposes.
1214 */
1215 if (nfsver.tv_sec == 0)
1216 nfsver = boottime;
1217 *tl++ = txdr_unsigned(nfsver.tv_sec);
1218 *tl = txdr_unsigned(nfsver.tv_usec);
1219 } else if (!error) {
1220 /* v2 non-error case. */
1221 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1222 nfsm_srvfillattr(vap, fp);
1223 }
1224 error = 0;
1225 nfsmout:
1226 if (vp)
1227 vput(vp);
1228 vn_finished_write(mntp);
1229 VFS_UNLOCK_GIANT(vfslocked);
1230 return(error);
1231 }
1232
1233 /*
1234 * nfs create service
1235 * now does a truncate to 0 length via. setattr if it already exists
1236 */
1237 int
1238 nfsrv_create(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1239 struct mbuf **mrq)
1240 {
1241 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1242 struct sockaddr *nam = nfsd->nd_nam;
1243 caddr_t dpos = nfsd->nd_dpos;
1244 struct ucred *cred = nfsd->nd_cr;
1245 struct nfs_fattr *fp;
1246 struct vattr va, dirfor, diraft;
1247 struct vattr *vap = &va;
1248 struct nfsv2_sattr *sp;
1249 u_int32_t *tl;
1250 struct nameidata nd;
1251 caddr_t bpos;
1252 int error = 0, rdev, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1253 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1254 struct mbuf *mb, *mreq;
1255 struct vnode *dirp = NULL;
1256 nfsfh_t nfh;
1257 fhandle_t *fhp;
1258 u_quad_t tempsize;
1259 struct timespec cverf;
1260 struct mount *mp = NULL;
1261 int tvfslocked;
1262 int vfslocked;
1263
1264 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1265 vfslocked = 0;
1266 #ifndef nolint
1267 rdev = 0;
1268 #endif
1269 ndclear(&nd);
1270
1271 fhp = &nfh.fh_generic;
1272 nfsm_srvmtofh(fhp);
1273 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1274 error = ESTALE;
1275 goto ereply;
1276 }
1277 vfslocked = VFS_LOCK_GIANT(mp);
1278 (void) vn_start_write(NULL, &mp, V_WAIT);
1279 vfs_rel(mp); /* The write holds a ref. */
1280 nfsm_srvnamesiz(len);
1281
1282 nd.ni_cnd.cn_cred = cred;
1283 nd.ni_cnd.cn_nameiop = CREATE;
1284 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1285
1286 /*
1287 * Call namei and do initial cleanup to get a few things
1288 * out of the way. If we get an initial error we cleanup
1289 * and return here to avoid special-casing the invalid nd
1290 * structure through the rest of the case. dirp may be
1291 * set even if an error occurs, but the nd structure will not
1292 * be valid at all if an error occurs so we have to invalidate it
1293 * prior to calling nfsm_reply ( which might goto nfsmout ).
1294 */
1295 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1296 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
1297 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1298 if (dirp && !v3) {
1299 vrele(dirp);
1300 dirp = NULL;
1301 }
1302 if (error) {
1303 nfsm_reply(NFSX_WCCDATA(v3));
1304 if (v3)
1305 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1306 error = 0;
1307 goto nfsmout;
1308 }
1309
1310 /*
1311 * No error. Continue. State:
1312 *
1313 * startdir is valid ( we release this immediately )
1314 * dirp may be valid
1315 * nd.ni_vp may be valid
1316 * nd.ni_dvp is valid
1317 *
1318 * The error state is set through the code and we may also do some
1319 * opportunistic releasing of vnodes to avoid holding locks through
1320 * NFS I/O. The cleanup at the end is a catch-all
1321 */
1322
1323 VATTR_NULL(vap);
1324 if (v3) {
1325 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1326 how = fxdr_unsigned(int, *tl);
1327 switch (how) {
1328 case NFSV3CREATE_GUARDED:
1329 if (nd.ni_vp) {
1330 error = EEXIST;
1331 break;
1332 }
1333 /* fall through */
1334 case NFSV3CREATE_UNCHECKED:
1335 nfsm_srvsattr(vap);
1336 break;
1337 case NFSV3CREATE_EXCLUSIVE:
1338 tl = nfsm_dissect_nonblock(u_int32_t *,
1339 NFSX_V3CREATEVERF);
1340 /* Unique bytes, endianness is not important. */
1341 cverf.tv_sec = (int32_t)tl[0];
1342 cverf.tv_nsec = tl[1];
1343 exclusive_flag = 1;
1344 break;
1345 };
1346 vap->va_type = VREG;
1347 } else {
1348 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
1349 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1350 if (vap->va_type == VNON)
1351 vap->va_type = VREG;
1352 vap->va_mode = nfstov_mode(sp->sa_mode);
1353 switch (vap->va_type) {
1354 case VREG:
1355 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1356 if (tsize != -1)
1357 vap->va_size = (u_quad_t)tsize;
1358 break;
1359 case VCHR:
1360 case VBLK:
1361 case VFIFO:
1362 rdev = fxdr_unsigned(long, sp->sa_size);
1363 break;
1364 default:
1365 break;
1366 };
1367 }
1368
1369 /*
1370 * Iff doesn't exist, create it
1371 * otherwise just truncate to 0 length
1372 * should I set the mode too ?
1373 *
1374 * The only possible error we can have at this point is EEXIST.
1375 * nd.ni_vp will also be non-NULL in that case.
1376 */
1377 if (nd.ni_vp == NULL) {
1378 if (vap->va_mode == (mode_t)VNOVAL)
1379 vap->va_mode = 0;
1380 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1381 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1382 if (error)
1383 NDFREE(&nd, NDF_ONLY_PNBUF);
1384 else {
1385 if (exclusive_flag) {
1386 exclusive_flag = 0;
1387 VATTR_NULL(vap);
1388 vap->va_atime = cverf;
1389 error = VOP_SETATTR(nd.ni_vp, vap,
1390 cred);
1391 }
1392 }
1393 } else if (vap->va_type == VCHR || vap->va_type == VBLK ||
1394 vap->va_type == VFIFO) {
1395 /*
1396 * NFSv2-specific code for creating device nodes
1397 * and fifos.
1398 *
1399 * Handle SysV FIFO node special cases. All other
1400 * devices require super user to access.
1401 */
1402 if (vap->va_type == VCHR && rdev == 0xffffffff)
1403 vap->va_type = VFIFO;
1404 if (vap->va_type != VFIFO &&
1405 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV,
1406 0))) {
1407 goto ereply;
1408 }
1409 vap->va_rdev = rdev;
1410 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1411 if (error) {
1412 NDFREE(&nd, NDF_ONLY_PNBUF);
1413 goto ereply;
1414 }
1415 vput(nd.ni_vp);
1416 nd.ni_vp = NULL;
1417
1418 /*
1419 * release dvp prior to lookup
1420 */
1421 vput(nd.ni_dvp);
1422 nd.ni_dvp = NULL;
1423 /*
1424 * Setup for lookup.
1425 *
1426 * Even though LOCKPARENT was cleared, ni_dvp may
1427 * be garbage.
1428 */
1429 nd.ni_cnd.cn_nameiop = LOOKUP;
1430 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1431 nd.ni_cnd.cn_thread = curthread;
1432 nd.ni_cnd.cn_cred = cred;
1433 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
1434 if (tvfslocked)
1435 nd.ni_cnd.cn_flags |= GIANTHELD;
1436 error = lookup(&nd);
1437 nd.ni_dvp = NULL;
1438 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1439 nd.ni_cnd.cn_flags &= ~GIANTHELD;
1440 if (error)
1441 goto ereply;
1442
1443 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1444 error = EINVAL;
1445 goto ereply;
1446 }
1447 } else {
1448 error = ENXIO;
1449 }
1450 } else {
1451 if (vap->va_size != -1) {
1452 error = nfsrv_access(nd.ni_vp, VWRITE,
1453 cred, (nd.ni_cnd.cn_flags & RDONLY), 0);
1454 if (!error) {
1455 tempsize = vap->va_size;
1456 VATTR_NULL(vap);
1457 vap->va_size = tempsize;
1458 error = VOP_SETATTR(nd.ni_vp, vap, cred);
1459 }
1460 }
1461 }
1462
1463 if (!error) {
1464 bzero((caddr_t)fhp, sizeof(nfh));
1465 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1466 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1467 if (!error)
1468 error = VOP_GETATTR(nd.ni_vp, vap, cred);
1469 }
1470 if (v3) {
1471 if (exclusive_flag && !error &&
1472 bcmp(&cverf, &vap->va_atime, sizeof (cverf)))
1473 error = EEXIST;
1474 if (dirp == nd.ni_dvp)
1475 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1476 else {
1477 /* Drop the other locks to avoid deadlock. */
1478 if (nd.ni_dvp) {
1479 if (nd.ni_dvp == nd.ni_vp)
1480 vrele(nd.ni_dvp);
1481 else
1482 vput(nd.ni_dvp);
1483 }
1484 if (nd.ni_vp)
1485 vput(nd.ni_vp);
1486 nd.ni_dvp = NULL;
1487 nd.ni_vp = NULL;
1488
1489 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1490 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1491 VOP_UNLOCK(dirp, 0);
1492 }
1493 }
1494 ereply:
1495 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1496 if (v3) {
1497 if (!error) {
1498 nfsm_srvpostop_fh(fhp);
1499 nfsm_srvpostop_attr(0, vap);
1500 }
1501 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1502 } else if (!error) {
1503 /* v2 non-error case. */
1504 nfsm_srvfhtom(fhp, v3);
1505 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
1506 nfsm_srvfillattr(vap, fp);
1507 }
1508 error = 0;
1509
1510 nfsmout:
1511 if (nd.ni_dvp) {
1512 if (nd.ni_dvp == nd.ni_vp)
1513 vrele(nd.ni_dvp);
1514 else
1515 vput(nd.ni_dvp);
1516 }
1517 if (nd.ni_vp)
1518 vput(nd.ni_vp);
1519 if (nd.ni_startdir) {
1520 vrele(nd.ni_startdir);
1521 nd.ni_startdir = NULL;
1522 }
1523 if (dirp)
1524 vrele(dirp);
1525 NDFREE(&nd, NDF_ONLY_PNBUF);
1526 vn_finished_write(mp);
1527 VFS_UNLOCK_GIANT(vfslocked);
1528 return (error);
1529 }
1530
1531 /*
1532 * nfs v3 mknod service
1533 */
1534 int
1535 nfsrv_mknod(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1536 struct mbuf **mrq)
1537 {
1538 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1539 struct sockaddr *nam = nfsd->nd_nam;
1540 caddr_t dpos = nfsd->nd_dpos;
1541 struct ucred *cred = nfsd->nd_cr;
1542 struct vattr va, dirfor, diraft;
1543 struct vattr *vap = &va;
1544 struct thread *td = curthread;
1545 u_int32_t *tl;
1546 struct nameidata nd;
1547 caddr_t bpos;
1548 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1549 u_int32_t major, minor;
1550 enum vtype vtyp;
1551 struct mbuf *mb, *mreq;
1552 struct vnode *vp, *dirp = NULL;
1553 nfsfh_t nfh;
1554 fhandle_t *fhp;
1555 struct mount *mp = NULL;
1556 int v3 = (nfsd->nd_flag & ND_NFSV3);
1557 int tvfslocked;
1558 int vfslocked;
1559
1560 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1561 vfslocked = 0;
1562 if (!v3)
1563 panic("nfsrv_mknod: v3 proc called on a v2 connection");
1564 ndclear(&nd);
1565
1566 fhp = &nfh.fh_generic;
1567 nfsm_srvmtofh(fhp);
1568 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1569 error = ESTALE;
1570 goto ereply;
1571 }
1572 vfslocked = VFS_LOCK_GIANT(mp);
1573 (void) vn_start_write(NULL, &mp, V_WAIT);
1574 vfs_rel(mp); /* The write holds a ref. */
1575 nfsm_srvnamesiz(len);
1576
1577 nd.ni_cnd.cn_cred = cred;
1578 nd.ni_cnd.cn_nameiop = CREATE;
1579 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | SAVESTART | MPSAFE;
1580
1581 /*
1582 * Handle nfs_namei() call. If an error occurs, the nd structure
1583 * is not valid. However, nfsm_*() routines may still jump to
1584 * nfsmout.
1585 */
1586
1587 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1588 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
1589 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1590 if (error) {
1591 nfsm_reply(NFSX_WCCDATA(1));
1592 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1593 error = 0;
1594 goto nfsmout;
1595 }
1596 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
1597 vtyp = nfsv3tov_type(*tl);
1598 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1599 error = NFSERR_BADTYPE;
1600 goto out;
1601 }
1602 VATTR_NULL(vap);
1603 nfsm_srvsattr(vap);
1604 if (vtyp == VCHR || vtyp == VBLK) {
1605 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
1606 major = fxdr_unsigned(u_int32_t, *tl++);
1607 minor = fxdr_unsigned(u_int32_t, *tl);
1608 vap->va_rdev = makedev(major, minor);
1609 }
1610
1611 /*
1612 * Iff doesn't exist, create it.
1613 */
1614 if (nd.ni_vp) {
1615 error = EEXIST;
1616 goto out;
1617 }
1618 vap->va_type = vtyp;
1619 if (vap->va_mode == (mode_t)VNOVAL)
1620 vap->va_mode = 0;
1621 if (vtyp == VSOCK) {
1622 vrele(nd.ni_startdir);
1623 nd.ni_startdir = NULL;
1624 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1625 if (error)
1626 NDFREE(&nd, NDF_ONLY_PNBUF);
1627 } else {
1628 if (vtyp != VFIFO && (error = priv_check_cred(cred,
1629 PRIV_VFS_MKNOD_DEV, 0)))
1630 goto out;
1631 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
1632 if (error) {
1633 NDFREE(&nd, NDF_ONLY_PNBUF);
1634 goto out;
1635 }
1636 vput(nd.ni_vp);
1637 nd.ni_vp = NULL;
1638
1639 /*
1640 * Release dvp prior to lookup
1641 */
1642 vput(nd.ni_dvp);
1643 nd.ni_dvp = NULL;
1644
1645 nd.ni_cnd.cn_nameiop = LOOKUP;
1646 nd.ni_cnd.cn_flags &= ~(LOCKPARENT);
1647 nd.ni_cnd.cn_thread = td;
1648 nd.ni_cnd.cn_cred = td->td_ucred;
1649 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
1650 if (tvfslocked)
1651 nd.ni_cnd.cn_flags |= GIANTHELD;
1652 error = lookup(&nd);
1653 nd.ni_dvp = NULL;
1654 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1655 nd.ni_cnd.cn_flags &= ~GIANTHELD;
1656
1657 if (error)
1658 goto out;
1659 if (nd.ni_cnd.cn_flags & ISSYMLINK)
1660 error = EINVAL;
1661 }
1662
1663 /*
1664 * send response, cleanup, return.
1665 */
1666 out:
1667 vp = nd.ni_vp;
1668 if (!error) {
1669 bzero((caddr_t)fhp, sizeof(nfh));
1670 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1671 error = VOP_VPTOFH(vp, &fhp->fh_fid);
1672 if (!error)
1673 error = VOP_GETATTR(vp, vap, cred);
1674 }
1675 if (nd.ni_dvp) {
1676 if (nd.ni_dvp == nd.ni_vp)
1677 vrele(nd.ni_dvp);
1678 else
1679 vput(nd.ni_dvp);
1680 nd.ni_dvp = NULL;
1681 }
1682 if (vp) {
1683 vput(vp);
1684 vp = NULL;
1685 nd.ni_vp = NULL;
1686 }
1687 if (nd.ni_startdir) {
1688 vrele(nd.ni_startdir);
1689 nd.ni_startdir = NULL;
1690 }
1691 NDFREE(&nd, NDF_ONLY_PNBUF);
1692 if (dirp) {
1693 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1694 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1695 vput(dirp);
1696 }
1697 ereply:
1698 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1699 if (v3) {
1700 if (!error) {
1701 nfsm_srvpostop_fh(fhp);
1702 nfsm_srvpostop_attr(0, vap);
1703 }
1704 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1705 }
1706 vn_finished_write(mp);
1707 VFS_UNLOCK_GIANT(vfslocked);
1708 return (0);
1709 nfsmout:
1710 if (nd.ni_dvp) {
1711 if (nd.ni_dvp == nd.ni_vp)
1712 vrele(nd.ni_dvp);
1713 else
1714 vput(nd.ni_dvp);
1715 }
1716 if (nd.ni_vp)
1717 vput(nd.ni_vp);
1718 if (dirp)
1719 vrele(dirp);
1720 if (nd.ni_startdir)
1721 vrele(nd.ni_startdir);
1722 NDFREE(&nd, NDF_ONLY_PNBUF);
1723 vn_finished_write(mp);
1724 VFS_UNLOCK_GIANT(vfslocked);
1725 return (error);
1726 }
1727
1728 /*
1729 * nfs remove service
1730 */
1731 int
1732 nfsrv_remove(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1733 struct mbuf **mrq)
1734 {
1735 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1736 struct sockaddr *nam = nfsd->nd_nam;
1737 caddr_t dpos = nfsd->nd_dpos;
1738 struct ucred *cred = nfsd->nd_cr;
1739 struct nameidata nd;
1740 caddr_t bpos;
1741 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
1742 int v3 = (nfsd->nd_flag & ND_NFSV3);
1743 struct mbuf *mb, *mreq;
1744 struct vnode *dirp;
1745 struct vattr dirfor, diraft;
1746 nfsfh_t nfh;
1747 fhandle_t *fhp;
1748 struct mount *mp = NULL;
1749 int vfslocked;
1750
1751 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1752 ndclear(&nd);
1753 vfslocked = 0;
1754
1755 fhp = &nfh.fh_generic;
1756 nfsm_srvmtofh(fhp);
1757 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
1758 error = ESTALE;
1759 goto ereply;
1760 }
1761 vfslocked = VFS_LOCK_GIANT(mp);
1762 (void) vn_start_write(NULL, &mp, V_WAIT);
1763 vfs_rel(mp); /* The write holds a ref. */
1764 nfsm_srvnamesiz(len);
1765
1766 nd.ni_cnd.cn_cred = cred;
1767 nd.ni_cnd.cn_nameiop = DELETE;
1768 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
1769 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
1770 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
1771 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
1772 if (dirp && !v3) {
1773 vrele(dirp);
1774 dirp = NULL;
1775 }
1776 if (error == 0) {
1777 if (nd.ni_vp->v_type == VDIR) {
1778 error = EPERM; /* POSIX */
1779 goto out;
1780 }
1781 /*
1782 * The root of a mounted filesystem cannot be deleted.
1783 */
1784 if (nd.ni_vp->v_vflag & VV_ROOT) {
1785 error = EBUSY;
1786 goto out;
1787 }
1788 out:
1789 if (!error) {
1790 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1791 NDFREE(&nd, NDF_ONLY_PNBUF);
1792 }
1793 }
1794 if (dirp && v3) {
1795 if (dirp == nd.ni_dvp)
1796 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1797 else {
1798 /* Drop the other locks to avoid deadlock. */
1799 if (nd.ni_dvp) {
1800 if (nd.ni_dvp == nd.ni_vp)
1801 vrele(nd.ni_dvp);
1802 else
1803 vput(nd.ni_dvp);
1804 }
1805 if (nd.ni_vp)
1806 vput(nd.ni_vp);
1807 nd.ni_dvp = NULL;
1808 nd.ni_vp = NULL;
1809
1810 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
1811 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
1812 VOP_UNLOCK(dirp, 0);
1813 }
1814 vrele(dirp);
1815 dirp = NULL;
1816 }
1817 ereply:
1818 nfsm_reply(NFSX_WCCDATA(v3));
1819 if (v3)
1820 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1821 error = 0;
1822 nfsmout:
1823 NDFREE(&nd, NDF_ONLY_PNBUF);
1824 if (nd.ni_dvp) {
1825 if (nd.ni_dvp == nd.ni_vp)
1826 vrele(nd.ni_dvp);
1827 else
1828 vput(nd.ni_dvp);
1829 }
1830 if (nd.ni_vp)
1831 vput(nd.ni_vp);
1832 vn_finished_write(mp);
1833 VFS_UNLOCK_GIANT(vfslocked);
1834 return(error);
1835 }
1836
1837 /*
1838 * nfs rename service
1839 */
1840 int
1841 nfsrv_rename(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
1842 struct mbuf **mrq)
1843 {
1844 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1845 struct sockaddr *nam = nfsd->nd_nam;
1846 caddr_t dpos = nfsd->nd_dpos;
1847 struct ucred *cred = nfsd->nd_cr;
1848 caddr_t bpos;
1849 int error = 0, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
1850 int tdirfor_ret = 1, tdiraft_ret = 1;
1851 int v3 = (nfsd->nd_flag & ND_NFSV3);
1852 struct mbuf *mb, *mreq;
1853 struct nameidata fromnd, tond;
1854 struct vnode *fvp, *tvp, *tdvp, *fdirp = NULL;
1855 struct vnode *tdirp = NULL;
1856 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1857 nfsfh_t fnfh, tnfh;
1858 fhandle_t *ffhp, *tfhp;
1859 uid_t saved_uid;
1860 struct mount *mp = NULL;
1861 int vfslocked;
1862
1863 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1864 vfslocked = 0;
1865 #ifndef nolint
1866 fvp = NULL;
1867 #endif
1868 ffhp = &fnfh.fh_generic;
1869 tfhp = &tnfh.fh_generic;
1870
1871 /*
1872 * Clear fields incase goto nfsmout occurs from macro.
1873 */
1874
1875 ndclear(&fromnd);
1876 ndclear(&tond);
1877
1878 nfsm_srvmtofh(ffhp);
1879 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL) {
1880 error = ESTALE;
1881 goto out1;
1882 }
1883 vfslocked = VFS_LOCK_GIANT(mp);
1884 (void) vn_start_write(NULL, &mp, V_WAIT);
1885 vfs_rel(mp); /* The write holds a ref. */
1886 nfsm_srvnamesiz(len);
1887 /*
1888 * Remember our original uid so that we can reset cr_uid before
1889 * the second nfs_namei() call, in case it is remapped.
1890 */
1891 saved_uid = cred->cr_uid;
1892 fromnd.ni_cnd.cn_cred = cred;
1893 fromnd.ni_cnd.cn_nameiop = DELETE;
1894 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART | MPSAFE;
1895 error = nfs_namei(&fromnd, nfsd, ffhp, len, slp, nam, &md,
1896 &dpos, &fdirp, v3, &fdirfor, &fdirfor_ret, FALSE);
1897 vfslocked = nfsrv_lockedpair_nd(vfslocked, &fromnd);
1898 if (fdirp && !v3) {
1899 vrele(fdirp);
1900 fdirp = NULL;
1901 }
1902 if (error) {
1903 nfsm_reply(2 * NFSX_WCCDATA(v3));
1904 if (v3) {
1905 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1906 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1907 }
1908 error = 0;
1909 goto nfsmout;
1910 }
1911 fvp = fromnd.ni_vp;
1912 nfsm_srvmtofh(tfhp);
1913 nfsm_srvnamesiz(len2);
1914 cred->cr_uid = saved_uid;
1915 tond.ni_cnd.cn_cred = cred;
1916 tond.ni_cnd.cn_nameiop = RENAME;
1917 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART | MPSAFE;
1918 error = nfs_namei(&tond, nfsd, tfhp, len2, slp, nam, &md,
1919 &dpos, &tdirp, v3, &tdirfor, &tdirfor_ret, FALSE);
1920 vfslocked = nfsrv_lockedpair_nd(vfslocked, &tond);
1921 if (tdirp && !v3) {
1922 vrele(tdirp);
1923 tdirp = NULL;
1924 }
1925 if (error)
1926 goto out1;
1927
1928 tdvp = tond.ni_dvp;
1929 tvp = tond.ni_vp;
1930 if (tvp != NULL) {
1931 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1932 if (v3)
1933 error = EEXIST;
1934 else
1935 error = EISDIR;
1936 goto out;
1937 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1938 if (v3)
1939 error = EEXIST;
1940 else
1941 error = ENOTDIR;
1942 goto out;
1943 }
1944 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1945 if (v3)
1946 error = EXDEV;
1947 else
1948 error = ENOTEMPTY;
1949 goto out;
1950 }
1951 }
1952 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1953 if (v3)
1954 error = EXDEV;
1955 else
1956 error = ENOTEMPTY;
1957 goto out;
1958 }
1959 if (fvp->v_mount != tdvp->v_mount) {
1960 if (v3)
1961 error = EXDEV;
1962 else
1963 error = ENOTEMPTY;
1964 goto out;
1965 }
1966 if (fvp == tdvp) {
1967 if (v3)
1968 error = EINVAL;
1969 else
1970 error = ENOTEMPTY;
1971 }
1972 /*
1973 * If source is the same as the destination (that is the
1974 * same vnode with the same name in the same directory),
1975 * then there is nothing to do.
1976 */
1977 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
1978 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
1979 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
1980 fromnd.ni_cnd.cn_namelen))
1981 error = -1;
1982 out:
1983 if (!error) {
1984 /*
1985 * The VOP_RENAME function releases all vnode references &
1986 * locks prior to returning so we need to clear the pointers
1987 * to bypass cleanup code later on.
1988 */
1989 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
1990 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
1991 fromnd.ni_dvp = NULL;
1992 fromnd.ni_vp = NULL;
1993 tond.ni_dvp = NULL;
1994 tond.ni_vp = NULL;
1995 if (error) {
1996 NDFREE(&fromnd, NDF_ONLY_PNBUF);
1997 NDFREE(&tond, NDF_ONLY_PNBUF);
1998 }
1999 } else {
2000 if (error == -1)
2001 error = 0;
2002 }
2003 /* fall through */
2004 out1:
2005 nfsm_reply(2 * NFSX_WCCDATA(v3));
2006 if (v3) {
2007 /* Release existing locks to prevent deadlock. */
2008 if (tond.ni_dvp) {
2009 if (tond.ni_dvp == tond.ni_vp)
2010 vrele(tond.ni_dvp);
2011 else
2012 vput(tond.ni_dvp);
2013 }
2014 if (tond.ni_vp)
2015 vput(tond.ni_vp);
2016 tond.ni_dvp = NULL;
2017 tond.ni_vp = NULL;
2018
2019 if (fdirp) {
2020 vn_lock(fdirp, LK_EXCLUSIVE | LK_RETRY);
2021 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred);
2022 VOP_UNLOCK(fdirp, 0);
2023 }
2024 if (tdirp) {
2025 vn_lock(tdirp, LK_EXCLUSIVE | LK_RETRY);
2026 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred);
2027 VOP_UNLOCK(tdirp, 0);
2028 }
2029 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2030 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2031 }
2032 error = 0;
2033 /* fall through */
2034
2035 nfsmout:
2036 /*
2037 * Clear out tond related fields
2038 */
2039 if (tond.ni_dvp) {
2040 if (tond.ni_dvp == tond.ni_vp)
2041 vrele(tond.ni_dvp);
2042 else
2043 vput(tond.ni_dvp);
2044 }
2045 if (tond.ni_vp)
2046 vput(tond.ni_vp);
2047 if (tdirp)
2048 vrele(tdirp);
2049 if (tond.ni_startdir)
2050 vrele(tond.ni_startdir);
2051 NDFREE(&tond, NDF_ONLY_PNBUF);
2052 /*
2053 * Clear out fromnd related fields
2054 */
2055 if (fdirp)
2056 vrele(fdirp);
2057 if (fromnd.ni_startdir)
2058 vrele(fromnd.ni_startdir);
2059 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2060 if (fromnd.ni_dvp)
2061 vrele(fromnd.ni_dvp);
2062 if (fromnd.ni_vp)
2063 vrele(fromnd.ni_vp);
2064
2065 vn_finished_write(mp);
2066 VFS_UNLOCK_GIANT(vfslocked);
2067 return (error);
2068 }
2069
2070 /*
2071 * nfs link service
2072 */
2073 int
2074 nfsrv_link(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2075 struct mbuf **mrq)
2076 {
2077 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2078 struct sockaddr *nam = nfsd->nd_nam;
2079 caddr_t dpos = nfsd->nd_dpos;
2080 struct ucred *cred = nfsd->nd_cr;
2081 struct nameidata nd;
2082 caddr_t bpos;
2083 int error = 0, rdonly, len, dirfor_ret = 1, diraft_ret = 1;
2084 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2085 struct mbuf *mb, *mreq;
2086 struct vnode *vp = NULL, *xp, *dirp = NULL;
2087 struct vattr dirfor, diraft, at;
2088 nfsfh_t nfh, dnfh;
2089 fhandle_t *fhp, *dfhp;
2090 struct mount *mp = NULL;
2091 int tvfslocked;
2092 int vfslocked;
2093
2094 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2095 ndclear(&nd);
2096 vfslocked = 0;
2097
2098 fhp = &nfh.fh_generic;
2099 dfhp = &dnfh.fh_generic;
2100 nfsm_srvmtofh(fhp);
2101 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2102 error = ESTALE;
2103 goto ereply;
2104 }
2105 vfslocked = VFS_LOCK_GIANT(mp);
2106 (void) vn_start_write(NULL, &mp, V_WAIT);
2107 vfs_rel(mp); /* The write holds a ref. */
2108 nfsm_srvmtofh(dfhp);
2109 nfsm_srvnamesiz(len);
2110
2111 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly);
2112 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
2113 if (error) {
2114 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2115 if (v3) {
2116 nfsm_srvpostop_attr(getret, &at);
2117 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2118 }
2119 vp = NULL;
2120 error = 0;
2121 goto nfsmout;
2122 }
2123 if (v3)
2124 getret = VOP_GETATTR(vp, &at, cred);
2125 if (vp->v_type == VDIR) {
2126 error = EPERM; /* POSIX */
2127 goto out1;
2128 }
2129 VOP_UNLOCK(vp, 0);
2130 nd.ni_cnd.cn_cred = cred;
2131 nd.ni_cnd.cn_nameiop = CREATE;
2132 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE | MPSAFE;
2133 error = nfs_namei(&nd, nfsd, dfhp, len, slp, nam, &md, &dpos,
2134 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
2135 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2136 if (dirp && !v3) {
2137 vrele(dirp);
2138 dirp = NULL;
2139 }
2140 if (error) {
2141 vrele(vp);
2142 vp = NULL;
2143 goto out2;
2144 }
2145 xp = nd.ni_vp;
2146 if (xp != NULL) {
2147 error = EEXIST;
2148 vrele(vp);
2149 vp = NULL;
2150 goto out2;
2151 }
2152 xp = nd.ni_dvp;
2153 if (vp->v_mount != xp->v_mount) {
2154 error = EXDEV;
2155 vrele(vp);
2156 vp = NULL;
2157 goto out2;
2158 }
2159 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2160 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2161 NDFREE(&nd, NDF_ONLY_PNBUF);
2162 /* fall through */
2163
2164 out1:
2165 if (v3)
2166 getret = VOP_GETATTR(vp, &at, cred);
2167 out2:
2168 if (dirp) {
2169 if (dirp == nd.ni_dvp)
2170 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2171 else {
2172 /* Release existing locks to prevent deadlock. */
2173 if (nd.ni_dvp) {
2174 if (nd.ni_dvp == nd.ni_vp)
2175 vrele(nd.ni_dvp);
2176 else
2177 vput(nd.ni_dvp);
2178 }
2179 if (nd.ni_vp)
2180 vrele(nd.ni_vp);
2181 nd.ni_dvp = NULL;
2182 nd.ni_vp = NULL;
2183
2184 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2185 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2186 VOP_UNLOCK(dirp, 0);
2187 }
2188 }
2189 ereply:
2190 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2191 if (v3) {
2192 nfsm_srvpostop_attr(getret, &at);
2193 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2194 }
2195 error = 0;
2196 /* fall through */
2197
2198 nfsmout:
2199 NDFREE(&nd, NDF_ONLY_PNBUF);
2200 if (vp)
2201 vput(vp);
2202 if (nd.ni_dvp) {
2203 if (nd.ni_dvp == nd.ni_vp)
2204 vrele(nd.ni_dvp);
2205 else
2206 vput(nd.ni_dvp);
2207 }
2208 if (dirp)
2209 vrele(dirp);
2210 if (nd.ni_vp)
2211 vrele(nd.ni_vp);
2212 vn_finished_write(mp);
2213 VFS_UNLOCK_GIANT(vfslocked);
2214 return(error);
2215 }
2216
2217 /*
2218 * nfs symbolic link service
2219 */
2220 int
2221 nfsrv_symlink(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2222 struct mbuf **mrq)
2223 {
2224 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2225 struct sockaddr *nam = nfsd->nd_nam;
2226 caddr_t dpos = nfsd->nd_dpos;
2227 struct ucred *cred = nfsd->nd_cr;
2228 struct vattr va, dirfor, diraft;
2229 struct nameidata nd;
2230 struct vattr *vap = &va;
2231 struct nfsv2_sattr *sp;
2232 char *bpos, *pathcp = NULL;
2233 struct uio io;
2234 struct iovec iv;
2235 int error = 0, len, len2, dirfor_ret = 1, diraft_ret = 1;
2236 int v3 = (nfsd->nd_flag & ND_NFSV3);
2237 struct mbuf *mb, *mreq;
2238 struct vnode *dirp = NULL;
2239 nfsfh_t nfh;
2240 fhandle_t *fhp;
2241 struct mount *mp = NULL;
2242 int tvfslocked;
2243 int vfslocked;
2244
2245 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2246 ndclear(&nd);
2247 vfslocked = 0;
2248
2249 fhp = &nfh.fh_generic;
2250 nfsm_srvmtofh(fhp);
2251 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2252 error = ESTALE;
2253 goto out;
2254 }
2255 vfslocked = VFS_LOCK_GIANT(mp);
2256 (void) vn_start_write(NULL, &mp, V_WAIT);
2257 vfs_rel(mp); /* The write holds a ref. */
2258 nfsm_srvnamesiz(len);
2259 nd.ni_cnd.cn_cred = cred;
2260 nd.ni_cnd.cn_nameiop = CREATE;
2261 nd.ni_cnd.cn_flags = LOCKPARENT | SAVESTART | MPSAFE;
2262 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2263 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
2264 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2265 if (error == 0) {
2266 VATTR_NULL(vap);
2267 if (v3)
2268 nfsm_srvsattr(vap);
2269 nfsm_srvpathsiz(len2);
2270 }
2271 if (dirp && !v3) {
2272 vrele(dirp);
2273 dirp = NULL;
2274 }
2275 if (error)
2276 goto out;
2277 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2278 iv.iov_base = pathcp;
2279 iv.iov_len = len2;
2280 io.uio_resid = len2;
2281 io.uio_offset = 0;
2282 io.uio_iov = &iv;
2283 io.uio_iovcnt = 1;
2284 io.uio_segflg = UIO_SYSSPACE;
2285 io.uio_rw = UIO_READ;
2286 io.uio_td = NULL;
2287 nfsm_mtouio(&io, len2);
2288 if (!v3) {
2289 sp = nfsm_dissect_nonblock(struct nfsv2_sattr *, NFSX_V2SATTR);
2290 vap->va_mode = nfstov_mode(sp->sa_mode);
2291 }
2292 *(pathcp + len2) = '\0';
2293 if (nd.ni_vp) {
2294 error = EEXIST;
2295 goto out;
2296 }
2297
2298 /*
2299 * issue symlink op. SAVESTART is set so the underlying path component
2300 * is only freed by the VOP if an error occurs.
2301 */
2302 if (vap->va_mode == (mode_t)VNOVAL)
2303 vap->va_mode = 0;
2304 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
2305 if (error)
2306 NDFREE(&nd, NDF_ONLY_PNBUF);
2307 else
2308 vput(nd.ni_vp);
2309 nd.ni_vp = NULL;
2310 /*
2311 * releases directory prior to potential lookup op.
2312 */
2313 vput(nd.ni_dvp);
2314 nd.ni_dvp = NULL;
2315
2316 if (error == 0) {
2317 if (v3) {
2318 /*
2319 * Issue lookup. Leave SAVESTART set so we can easily free
2320 * the name buffer later on.
2321 *
2322 * since LOCKPARENT is not set, ni_dvp will be garbage on
2323 * return whether an error occurs or not.
2324 */
2325 nd.ni_cnd.cn_nameiop = LOOKUP;
2326 nd.ni_cnd.cn_flags &= ~(LOCKPARENT | FOLLOW);
2327 nd.ni_cnd.cn_flags |= (NOFOLLOW | LOCKLEAF);
2328 nd.ni_cnd.cn_thread = curthread;
2329 nd.ni_cnd.cn_cred = cred;
2330 tvfslocked = VFS_LOCK_GIANT(nd.ni_startdir->v_mount);
2331 if (tvfslocked)
2332 nd.ni_cnd.cn_flags |= GIANTHELD;
2333 error = lookup(&nd);
2334 nd.ni_dvp = NULL;
2335 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2336 nd.ni_cnd.cn_flags &= ~GIANTHELD;
2337
2338 if (error == 0) {
2339 bzero((caddr_t)fhp, sizeof(nfh));
2340 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2341 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2342 if (!error)
2343 error = VOP_GETATTR(nd.ni_vp, vap, cred);
2344 vput(nd.ni_vp);
2345 nd.ni_vp = NULL;
2346 }
2347 }
2348 }
2349 out:
2350 /*
2351 * These releases aren't strictly required, does even doing them
2352 * make any sense? XXX can nfsm_reply() block?
2353 */
2354 if (pathcp) {
2355 free(pathcp, M_TEMP);
2356 pathcp = NULL;
2357 }
2358 if (dirp) {
2359 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2360 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2361 VOP_UNLOCK(dirp, 0);
2362 }
2363 if (nd.ni_startdir) {
2364 vrele(nd.ni_startdir);
2365 nd.ni_startdir = NULL;
2366 }
2367 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2368 if (v3) {
2369 if (!error) {
2370 nfsm_srvpostop_fh(fhp);
2371 nfsm_srvpostop_attr(0, vap);
2372 }
2373 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2374 }
2375 error = 0;
2376 /* fall through */
2377
2378 nfsmout:
2379 NDFREE(&nd, NDF_ONLY_PNBUF);
2380 if (nd.ni_dvp) {
2381 if (nd.ni_dvp == nd.ni_vp)
2382 vrele(nd.ni_dvp);
2383 else
2384 vput(nd.ni_dvp);
2385 }
2386 if (nd.ni_vp)
2387 vrele(nd.ni_vp);
2388 if (nd.ni_startdir)
2389 vrele(nd.ni_startdir);
2390 if (dirp)
2391 vrele(dirp);
2392 if (pathcp)
2393 free(pathcp, M_TEMP);
2394
2395 vn_finished_write(mp);
2396 VFS_UNLOCK_GIANT(vfslocked);
2397 return (error);
2398 }
2399
2400 /*
2401 * nfs mkdir service
2402 */
2403 int
2404 nfsrv_mkdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2405 struct mbuf **mrq)
2406 {
2407 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2408 struct sockaddr *nam = nfsd->nd_nam;
2409 caddr_t dpos = nfsd->nd_dpos;
2410 struct ucred *cred = nfsd->nd_cr;
2411 struct vattr va, dirfor, diraft;
2412 struct vattr *vap = &va;
2413 struct nfs_fattr *fp;
2414 struct nameidata nd;
2415 u_int32_t *tl;
2416 caddr_t bpos;
2417 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2418 int v3 = (nfsd->nd_flag & ND_NFSV3);
2419 struct mbuf *mb, *mreq;
2420 struct vnode *dirp = NULL;
2421 int vpexcl = 0;
2422 nfsfh_t nfh;
2423 fhandle_t *fhp;
2424 struct mount *mp = NULL;
2425 int vfslocked;
2426
2427 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2428 ndclear(&nd);
2429 vfslocked = 0;
2430
2431 fhp = &nfh.fh_generic;
2432 nfsm_srvmtofh(fhp);
2433 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2434 error = ESTALE;
2435 goto out;
2436 }
2437 vfslocked = VFS_LOCK_GIANT(mp);
2438 (void) vn_start_write(NULL, &mp, V_WAIT);
2439 vfs_rel(mp); /* The write holds a ref. */
2440 nfsm_srvnamesiz(len);
2441 nd.ni_cnd.cn_cred = cred;
2442 nd.ni_cnd.cn_nameiop = CREATE;
2443 nd.ni_cnd.cn_flags = LOCKPARENT | MPSAFE;
2444
2445 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2446 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
2447 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2448 if (dirp && !v3) {
2449 vrele(dirp);
2450 dirp = NULL;
2451 }
2452 if (error) {
2453 nfsm_reply(NFSX_WCCDATA(v3));
2454 if (v3)
2455 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2456 error = 0;
2457 goto nfsmout;
2458 }
2459 VATTR_NULL(vap);
2460 if (v3) {
2461 nfsm_srvsattr(vap);
2462 } else {
2463 tl = nfsm_dissect_nonblock(u_int32_t *, NFSX_UNSIGNED);
2464 vap->va_mode = nfstov_mode(*tl++);
2465 }
2466
2467 /*
2468 * At this point nd.ni_dvp is referenced and exclusively locked and
2469 * nd.ni_vp, if it exists, is referenced but not locked.
2470 */
2471
2472 vap->va_type = VDIR;
2473 if (nd.ni_vp != NULL) {
2474 NDFREE(&nd, NDF_ONLY_PNBUF);
2475 error = EEXIST;
2476 goto out;
2477 }
2478
2479 /*
2480 * Issue mkdir op. Since SAVESTART is not set, the pathname
2481 * component is freed by the VOP call. This will fill-in
2482 * nd.ni_vp, reference, and exclusively lock it.
2483 */
2484 if (vap->va_mode == (mode_t)VNOVAL)
2485 vap->va_mode = 0;
2486 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, vap);
2487 NDFREE(&nd, NDF_ONLY_PNBUF);
2488 vpexcl = 1;
2489
2490 vput(nd.ni_dvp);
2491 nd.ni_dvp = NULL;
2492
2493 if (!error) {
2494 bzero((caddr_t)fhp, sizeof(nfh));
2495 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2496 error = VOP_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2497 if (!error)
2498 error = VOP_GETATTR(nd.ni_vp, vap, cred);
2499 }
2500 out:
2501 if (dirp) {
2502 if (dirp == nd.ni_dvp) {
2503 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2504 } else {
2505 /* Release existing locks to prevent deadlock. */
2506 if (nd.ni_dvp) {
2507 NDFREE(&nd, NDF_ONLY_PNBUF);
2508 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2509 vrele(nd.ni_dvp);
2510 else
2511 vput(nd.ni_dvp);
2512 }
2513 if (nd.ni_vp) {
2514 if (vpexcl)
2515 vput(nd.ni_vp);
2516 else
2517 vrele(nd.ni_vp);
2518 }
2519 nd.ni_dvp = NULL;
2520 nd.ni_vp = NULL;
2521 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2522 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2523 VOP_UNLOCK(dirp, 0);
2524 }
2525 }
2526 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2527 if (v3) {
2528 if (!error) {
2529 nfsm_srvpostop_fh(fhp);
2530 nfsm_srvpostop_attr(0, vap);
2531 }
2532 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2533 } else if (!error) {
2534 /* v2 non-error case. */
2535 nfsm_srvfhtom(fhp, v3);
2536 fp = nfsm_build(struct nfs_fattr *, NFSX_V2FATTR);
2537 nfsm_srvfillattr(vap, fp);
2538 }
2539 error = 0;
2540 /* fall through */
2541
2542 nfsmout:
2543 if (nd.ni_dvp) {
2544 NDFREE(&nd, NDF_ONLY_PNBUF);
2545 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2546 vrele(nd.ni_dvp);
2547 else
2548 vput(nd.ni_dvp);
2549 }
2550 if (nd.ni_vp) {
2551 if (vpexcl)
2552 vput(nd.ni_vp);
2553 else
2554 vrele(nd.ni_vp);
2555 }
2556 if (dirp)
2557 vrele(dirp);
2558 vn_finished_write(mp);
2559 VFS_UNLOCK_GIANT(vfslocked);
2560 return (error);
2561 }
2562
2563 /*
2564 * nfs rmdir service
2565 */
2566 int
2567 nfsrv_rmdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2568 struct mbuf **mrq)
2569 {
2570 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2571 struct sockaddr *nam = nfsd->nd_nam;
2572 caddr_t dpos = nfsd->nd_dpos;
2573 struct ucred *cred = nfsd->nd_cr;
2574 caddr_t bpos;
2575 int error = 0, len, dirfor_ret = 1, diraft_ret = 1;
2576 int v3 = (nfsd->nd_flag & ND_NFSV3);
2577 struct mbuf *mb, *mreq;
2578 struct vnode *vp, *dirp = NULL;
2579 struct vattr dirfor, diraft;
2580 nfsfh_t nfh;
2581 fhandle_t *fhp;
2582 struct nameidata nd;
2583 struct mount *mp = NULL;
2584 int vfslocked;
2585
2586 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2587 ndclear(&nd);
2588 vfslocked = 0;
2589
2590 fhp = &nfh.fh_generic;
2591 nfsm_srvmtofh(fhp);
2592 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
2593 error = ESTALE;
2594 goto out;
2595 }
2596 vfslocked = VFS_LOCK_GIANT(mp);
2597 (void) vn_start_write(NULL, &mp, V_WAIT);
2598 vfs_rel(mp); /* The write holds a ref. */
2599 nfsm_srvnamesiz(len);
2600 nd.ni_cnd.cn_cred = cred;
2601 nd.ni_cnd.cn_nameiop = DELETE;
2602 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | MPSAFE;
2603 error = nfs_namei(&nd, nfsd, fhp, len, slp, nam, &md, &dpos,
2604 &dirp, v3, &dirfor, &dirfor_ret, FALSE);
2605 vfslocked = nfsrv_lockedpair_nd(vfslocked, &nd);
2606 if (dirp && !v3) {
2607 vrele(dirp);
2608 dirp = NULL;
2609 }
2610 if (error) {
2611 nfsm_reply(NFSX_WCCDATA(v3));
2612 if (v3)
2613 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2614 error = 0;
2615 goto nfsmout;
2616 }
2617 vp = nd.ni_vp;
2618 if (vp->v_type != VDIR) {
2619 error = ENOTDIR;
2620 goto out;
2621 }
2622 /*
2623 * No rmdir "." please.
2624 */
2625 if (nd.ni_dvp == vp) {
2626 error = EINVAL;
2627 goto out;
2628 }
2629 /*
2630 * The root of a mounted filesystem cannot be deleted.
2631 */
2632 if (vp->v_vflag & VV_ROOT)
2633 error = EBUSY;
2634 out:
2635 /*
2636 * Issue or abort op. Since SAVESTART is not set, path name
2637 * component is freed by the VOP after either.
2638 */
2639 if (!error)
2640 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2641 NDFREE(&nd, NDF_ONLY_PNBUF);
2642
2643 if (dirp) {
2644 if (dirp == nd.ni_dvp)
2645 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2646 else {
2647 /* Release existing locks to prevent deadlock. */
2648 if (nd.ni_dvp) {
2649 if (nd.ni_dvp == nd.ni_vp)
2650 vrele(nd.ni_dvp);
2651 else
2652 vput(nd.ni_dvp);
2653 }
2654 if (nd.ni_vp)
2655 vput(nd.ni_vp);
2656 nd.ni_dvp = NULL;
2657 nd.ni_vp = NULL;
2658 vn_lock(dirp, LK_EXCLUSIVE | LK_RETRY);
2659 diraft_ret = VOP_GETATTR(dirp, &diraft, cred);
2660 VOP_UNLOCK(dirp, 0);
2661 }
2662 }
2663 nfsm_reply(NFSX_WCCDATA(v3));
2664 error = 0;
2665 if (v3)
2666 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2667 /* fall through */
2668
2669 nfsmout:
2670 NDFREE(&nd, NDF_ONLY_PNBUF);
2671 if (nd.ni_dvp) {
2672 if (nd.ni_dvp == nd.ni_vp)
2673 vrele(nd.ni_dvp);
2674 else
2675 vput(nd.ni_dvp);
2676 }
2677 if (nd.ni_vp)
2678 vput(nd.ni_vp);
2679 if (dirp)
2680 vrele(dirp);
2681
2682 vn_finished_write(mp);
2683 VFS_UNLOCK_GIANT(vfslocked);
2684 return(error);
2685 }
2686
2687 /*
2688 * nfs readdir service
2689 * - mallocs what it thinks is enough to read
2690 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2691 * - calls VOP_READDIR()
2692 * - loops around building the reply
2693 * if the output generated exceeds count break out of loop
2694 * The nfsm_clget macro is used here so that the reply will be packed
2695 * tightly in mbuf clusters.
2696 * - it only knows that it has encountered eof when the VOP_READDIR()
2697 * reads nothing
2698 * - as such one readdir rpc will return eof false although you are there
2699 * and then the next will return eof
2700 * - it trims out records with d_fileno == 0
2701 * this doesn't matter for Unix clients, but they might confuse clients
2702 * for other os'.
2703 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2704 * than requested, but this may not apply to all filesystems. For
2705 * example, client NFS does not { although it is never remote mounted
2706 * anyhow }
2707 * The alternate call nfsrv_readdirplus() does lookups as well.
2708 * PS: The NFS protocol spec. does not clarify what the "count" byte
2709 * argument is a count of.. just name strings and file id's or the
2710 * entire reply rpc or ...
2711 * I tried just file name and id sizes and it confused the Sun client,
2712 * so I am using the full rpc size now. The "paranoia.." comment refers
2713 * to including the status longwords that are not a part of the dir.
2714 * "entry" structures, but are in the rpc.
2715 */
2716 struct flrep {
2717 nfsuint64 fl_off;
2718 u_int32_t fl_postopok;
2719 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2720 u_int32_t fl_fhok;
2721 u_int32_t fl_fhsize;
2722 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2723 };
2724
2725 int
2726 nfsrv_readdir(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
2727 struct mbuf **mrq)
2728 {
2729 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2730 struct sockaddr *nam = nfsd->nd_nam;
2731 caddr_t dpos = nfsd->nd_dpos;
2732 struct ucred *cred = nfsd->nd_cr;
2733 char *bp, *be;
2734 struct mbuf *mp;
2735 struct dirent *dp;
2736 caddr_t cp;
2737 u_int32_t *tl;
2738 caddr_t bpos;
2739 struct mbuf *mb, *mreq;
2740 char *cpos, *cend, *rbuf;
2741 struct vnode *vp = NULL;
2742 struct vattr at;
2743 nfsfh_t nfh;
2744 fhandle_t *fhp;
2745 struct uio io;
2746 struct iovec iv;
2747 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2748 int siz, cnt, fullsiz, eofflag, rdonly, ncookies;
2749 int v3 = (nfsd->nd_flag & ND_NFSV3);
2750 u_quad_t off, toff, verf;
2751 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2752 int vfslocked, not_zfs;
2753
2754 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2755 vfslocked = 0;
2756 fhp = &nfh.fh_generic;
2757 nfsm_srvmtofh(fhp);
2758 if (v3) {
2759 tl = nfsm_dissect_nonblock(u_int32_t *, 5 * NFSX_UNSIGNED);
2760 toff = fxdr_hyper(tl);
2761 tl += 2;
2762 verf = fxdr_hyper(tl);
2763 tl += 2;
2764 } else {
2765 tl = nfsm_dissect_nonblock(u_int32_t *, 2 * NFSX_UNSIGNED);
2766 toff = fxdr_unsigned(u_quad_t, *tl++);
2767 verf = 0; /* shut up gcc */
2768 }
2769 off = toff;
2770 cnt = fxdr_unsigned(int, *tl);
2771 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2772 xfer = NFS_SRVMAXDATA(nfsd);
2773 if (cnt > xfer)
2774 cnt = xfer;
2775 if (siz > xfer)
2776 siz = xfer;
2777 fullsiz = siz;
2778 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
2779 if (!error && vp->v_type != VDIR) {
2780 error = ENOTDIR;
2781 vput(vp);
2782 vp = NULL;
2783 }
2784 if (error) {
2785 nfsm_reply(NFSX_UNSIGNED);
2786 if (v3)
2787 nfsm_srvpostop_attr(getret, &at);
2788 error = 0;
2789 goto nfsmout;
2790 }
2791
2792 /*
2793 * Obtain lock on vnode for this section of the code
2794 */
2795 if (v3) {
2796 error = getret = VOP_GETATTR(vp, &at, cred);
2797 #if 0
2798 /*
2799 * XXX This check may be too strict for Solaris 2.5 clients.
2800 */
2801 if (!error && toff && verf && verf != at.va_filerev)
2802 error = NFSERR_BAD_COOKIE;
2803 #endif
2804 }
2805 if (!error)
2806 error = nfsrv_access(vp, VEXEC, cred, rdonly, 0);
2807 if (error) {
2808 vput(vp);
2809 vp = NULL;
2810 nfsm_reply(NFSX_POSTOPATTR(v3));
2811 if (v3)
2812 nfsm_srvpostop_attr(getret, &at);
2813 error = 0;
2814 goto nfsmout;
2815 }
2816 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") != 0;
2817 VOP_UNLOCK(vp, 0);
2818
2819 /*
2820 * end section. Allocate rbuf and continue
2821 */
2822 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2823 again:
2824 iv.iov_base = rbuf;
2825 iv.iov_len = fullsiz;
2826 io.uio_iov = &iv;
2827 io.uio_iovcnt = 1;
2828 io.uio_offset = (off_t)off;
2829 io.uio_resid = fullsiz;
2830 io.uio_segflg = UIO_SYSSPACE;
2831 io.uio_rw = UIO_READ;
2832 io.uio_td = NULL;
2833 eofflag = 0;
2834 if (cookies) {
2835 free((caddr_t)cookies, M_TEMP);
2836 cookies = NULL;
2837 }
2838 vn_lock(vp, LK_SHARED | LK_RETRY);
2839 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
2840 off = (off_t)io.uio_offset;
2841 if (!cookies && !error)
2842 error = NFSERR_PERM;
2843 if (v3) {
2844 getret = VOP_GETATTR(vp, &at, cred);
2845 if (!error)
2846 error = getret;
2847 }
2848 VOP_UNLOCK(vp, 0);
2849 if (error) {
2850 vrele(vp);
2851 vp = NULL;
2852 free((caddr_t)rbuf, M_TEMP);
2853 if (cookies)
2854 free((caddr_t)cookies, M_TEMP);
2855 nfsm_reply(NFSX_POSTOPATTR(v3));
2856 if (v3)
2857 nfsm_srvpostop_attr(getret, &at);
2858 error = 0;
2859 goto nfsmout;
2860 }
2861 if (io.uio_resid) {
2862 siz -= io.uio_resid;
2863
2864 /*
2865 * If nothing read, return eof
2866 * rpc reply
2867 */
2868 if (siz == 0) {
2869 vrele(vp);
2870 vp = NULL;
2871 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2872 2 * NFSX_UNSIGNED);
2873 if (v3) {
2874 nfsm_srvpostop_attr(getret, &at);
2875 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
2876 txdr_hyper(at.va_filerev, tl);
2877 tl += 2;
2878 } else
2879 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2880 *tl++ = nfsrv_nfs_false;
2881 *tl = nfsrv_nfs_true;
2882 free((caddr_t)rbuf, M_TEMP);
2883 free((caddr_t)cookies, M_TEMP);
2884 error = 0;
2885 goto nfsmout;
2886 }
2887 }
2888
2889 /*
2890 * Check for degenerate cases of nothing useful read.
2891 * If so go try again
2892 */
2893 cpos = rbuf;
2894 cend = rbuf + siz;
2895 dp = (struct dirent *)cpos;
2896 cookiep = cookies;
2897 /*
2898 * For some reason FreeBSD's ufs_readdir() chooses to back the
2899 * directory offset up to a block boundary, so it is necessary to
2900 * skip over the records that precede the requested offset. This
2901 * requires the assumption that file offset cookies monotonically
2902 * increase.
2903 * Since the offset cookies don't monotonically increase for ZFS,
2904 * this is not done when ZFS is the file system.
2905 */
2906 while (cpos < cend && ncookies > 0 &&
2907 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2908 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
2909 cpos += dp->d_reclen;
2910 dp = (struct dirent *)cpos;
2911 cookiep++;
2912 ncookies--;
2913 }
2914 if (cpos >= cend || ncookies == 0) {
2915 toff = off;
2916 siz = fullsiz;
2917 goto again;
2918 }
2919
2920 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
2921 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2922 if (v3) {
2923 nfsm_srvpostop_attr(getret, &at);
2924 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
2925 txdr_hyper(at.va_filerev, tl);
2926 }
2927 mp = mb;
2928 bp = bpos;
2929 be = bp + M_TRAILINGSPACE(mp);
2930
2931 /* Loop through the records and build reply */
2932 while (cpos < cend && ncookies > 0) {
2933 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2934 nlen = dp->d_namlen;
2935 rem = nfsm_rndup(nlen) - nlen;
2936 len += (4 * NFSX_UNSIGNED + nlen + rem);
2937 if (v3)
2938 len += 2 * NFSX_UNSIGNED;
2939 if (len > cnt) {
2940 eofflag = 0;
2941 break;
2942 }
2943 /*
2944 * Build the directory record xdr from
2945 * the dirent entry.
2946 */
2947 nfsm_clget;
2948 *tl = nfsrv_nfs_true;
2949 bp += NFSX_UNSIGNED;
2950 if (v3) {
2951 nfsm_clget;
2952 *tl = 0;
2953 bp += NFSX_UNSIGNED;
2954 }
2955 nfsm_clget;
2956 *tl = txdr_unsigned(dp->d_fileno);
2957 bp += NFSX_UNSIGNED;
2958 nfsm_clget;
2959 *tl = txdr_unsigned(nlen);
2960 bp += NFSX_UNSIGNED;
2961
2962 /* And loop around copying the name */
2963 xfer = nlen;
2964 cp = dp->d_name;
2965 while (xfer > 0) {
2966 nfsm_clget;
2967 if ((bp+xfer) > be)
2968 tsiz = be-bp;
2969 else
2970 tsiz = xfer;
2971 bcopy(cp, bp, tsiz);
2972 bp += tsiz;
2973 xfer -= tsiz;
2974 if (xfer > 0)
2975 cp += tsiz;
2976 }
2977 /* And null pad to an int32_t boundary. */
2978 for (i = 0; i < rem; i++)
2979 *bp++ = '\0';
2980 nfsm_clget;
2981
2982 /* Finish off the record */
2983 if (v3) {
2984 *tl = 0;
2985 bp += NFSX_UNSIGNED;
2986 nfsm_clget;
2987 }
2988 *tl = txdr_unsigned(*cookiep);
2989 bp += NFSX_UNSIGNED;
2990 }
2991 cpos += dp->d_reclen;
2992 dp = (struct dirent *)cpos;
2993 cookiep++;
2994 ncookies--;
2995 }
2996 vrele(vp);
2997 vp = NULL;
2998 nfsm_clget;
2999 *tl = nfsrv_nfs_false;
3000 bp += NFSX_UNSIGNED;
3001 nfsm_clget;
3002 if (eofflag)
3003 *tl = nfsrv_nfs_true;
3004 else
3005 *tl = nfsrv_nfs_false;
3006 bp += NFSX_UNSIGNED;
3007 if (mp != mb) {
3008 if (bp < be)
3009 mp->m_len = bp - mtod(mp, caddr_t);
3010 } else
3011 mp->m_len += bp - bpos;
3012 free((caddr_t)rbuf, M_TEMP);
3013 free((caddr_t)cookies, M_TEMP);
3014
3015 nfsmout:
3016 if (vp)
3017 vrele(vp);
3018 VFS_UNLOCK_GIANT(vfslocked);
3019 return(error);
3020 }
3021
3022 int
3023 nfsrv_readdirplus(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3024 struct mbuf **mrq)
3025 {
3026 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3027 struct sockaddr *nam = nfsd->nd_nam;
3028 caddr_t dpos = nfsd->nd_dpos;
3029 struct ucred *cred = nfsd->nd_cr;
3030 char *bp, *be;
3031 struct mbuf *mp;
3032 struct dirent *dp;
3033 caddr_t cp;
3034 u_int32_t *tl;
3035 caddr_t bpos;
3036 struct mbuf *mb, *mreq;
3037 char *cpos, *cend, *rbuf;
3038 struct vnode *vp = NULL, *nvp;
3039 struct flrep fl;
3040 nfsfh_t nfh;
3041 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3042 struct uio io;
3043 struct iovec iv;
3044 struct vattr va, at, *vap = &va;
3045 struct nfs_fattr *fp;
3046 int len, nlen, rem, xfer, tsiz, i, error = 0, error1, getret = 1;
3047 int vp_locked;
3048 int siz, cnt, fullsiz, eofflag, rdonly, dirlen, ncookies;
3049 u_quad_t off, toff, verf;
3050 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3051 int v3 = (nfsd->nd_flag & ND_NFSV3);
3052 int usevget = 1, vfslocked;
3053 struct componentname cn;
3054 struct mount *mntp = NULL;
3055 int not_zfs;
3056
3057 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3058 vfslocked = 0;
3059 vp_locked = 0;
3060 if (!v3)
3061 panic("nfsrv_readdirplus: v3 proc called on a v2 connection");
3062 fhp = &nfh.fh_generic;
3063 nfsm_srvmtofh(fhp);
3064 tl = nfsm_dissect_nonblock(u_int32_t *, 6 * NFSX_UNSIGNED);
3065 toff = fxdr_hyper(tl);
3066 tl += 2;
3067 verf = fxdr_hyper(tl);
3068 tl += 2;
3069 siz = fxdr_unsigned(int, *tl++);
3070 cnt = fxdr_unsigned(int, *tl);
3071 off = toff;
3072 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3073 xfer = NFS_SRVMAXDATA(nfsd);
3074 if (cnt > xfer)
3075 cnt = xfer;
3076 if (siz > xfer)
3077 siz = xfer;
3078 fullsiz = siz;
3079 error = nfsrv_fhtovp(fhp, NFSRV_FLAG_BUSY, &vp, &vfslocked, nfsd, slp,
3080 nam, &rdonly);
3081 if (!error) {
3082 vp_locked = 1;
3083 mntp = vp->v_mount;
3084 if (vp->v_type != VDIR) {
3085 error = ENOTDIR;
3086 vput(vp);
3087 vp = NULL;
3088 vp_locked = 0;
3089 }
3090 }
3091 if (error) {
3092 nfsm_reply(NFSX_UNSIGNED);
3093 nfsm_srvpostop_attr(getret, &at);
3094 error = 0;
3095 goto nfsmout;
3096 }
3097 error = getret = VOP_GETATTR(vp, &at, cred);
3098 #if 0
3099 /*
3100 * XXX This check may be too strict for Solaris 2.5 clients.
3101 */
3102 if (!error && toff && verf && verf != at.va_filerev)
3103 error = NFSERR_BAD_COOKIE;
3104 #endif
3105 if (!error)
3106 error = nfsrv_access(vp, VEXEC, cred, rdonly, 0);
3107 if (error) {
3108 vput(vp);
3109 vp_locked = 0;
3110 vp = NULL;
3111 nfsm_reply(NFSX_V3POSTOPATTR);
3112 nfsm_srvpostop_attr(getret, &at);
3113 error = 0;
3114 goto nfsmout;
3115 }
3116 not_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") != 0;
3117 VOP_UNLOCK(vp, 0);
3118 vp_locked = 0;
3119 rbuf = malloc(siz, M_TEMP, M_WAITOK);
3120 again:
3121 iv.iov_base = rbuf;
3122 iv.iov_len = fullsiz;
3123 io.uio_iov = &iv;
3124 io.uio_iovcnt = 1;
3125 io.uio_offset = (off_t)off;
3126 io.uio_resid = fullsiz;
3127 io.uio_segflg = UIO_SYSSPACE;
3128 io.uio_rw = UIO_READ;
3129 io.uio_td = NULL;
3130 eofflag = 0;
3131 vp_locked = 1;
3132 if (cookies) {
3133 free((caddr_t)cookies, M_TEMP);
3134 cookies = NULL;
3135 }
3136 vn_lock(vp, LK_SHARED | LK_RETRY);
3137 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3138 off = (u_quad_t)io.uio_offset;
3139 getret = VOP_GETATTR(vp, &at, cred);
3140 VOP_UNLOCK(vp, 0);
3141 vp_locked = 0;
3142 if (!cookies && !error)
3143 error = NFSERR_PERM;
3144 if (!error)
3145 error = getret;
3146 if (error) {
3147 vrele(vp);
3148 vp = NULL;
3149 if (cookies)
3150 free((caddr_t)cookies, M_TEMP);
3151 free((caddr_t)rbuf, M_TEMP);
3152 nfsm_reply(NFSX_V3POSTOPATTR);
3153 nfsm_srvpostop_attr(getret, &at);
3154 error = 0;
3155 goto nfsmout;
3156 }
3157 if (io.uio_resid) {
3158 siz -= io.uio_resid;
3159
3160 /*
3161 * If nothing read, return eof
3162 * rpc reply
3163 */
3164 if (siz == 0) {
3165 vrele(vp);
3166 vp = NULL;
3167 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3168 2 * NFSX_UNSIGNED);
3169 nfsm_srvpostop_attr(getret, &at);
3170 tl = nfsm_build(u_int32_t *, 4 * NFSX_UNSIGNED);
3171 txdr_hyper(at.va_filerev, tl);
3172 tl += 2;
3173 *tl++ = nfsrv_nfs_false;
3174 *tl = nfsrv_nfs_true;
3175 free((caddr_t)cookies, M_TEMP);
3176 free((caddr_t)rbuf, M_TEMP);
3177 error = 0;
3178 goto nfsmout;
3179 }
3180 }
3181
3182 /*
3183 * Check for degenerate cases of nothing useful read.
3184 * If so go try again
3185 */
3186 cpos = rbuf;
3187 cend = rbuf + siz;
3188 dp = (struct dirent *)cpos;
3189 cookiep = cookies;
3190 /*
3191 * For some reason FreeBSD's ufs_readdir() chooses to back the
3192 * directory offset up to a block boundary, so it is necessary to
3193 * skip over the records that precede the requested offset. This
3194 * requires the assumption that file offset cookies monotonically
3195 * increase.
3196 * Since the offset cookies don't monotonically increase for ZFS,
3197 * this is not done when ZFS is the file system.
3198 */
3199 while (cpos < cend && ncookies > 0 &&
3200 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3201 (not_zfs != 0 && ((u_quad_t)(*cookiep)) <= toff))) {
3202 cpos += dp->d_reclen;
3203 dp = (struct dirent *)cpos;
3204 cookiep++;
3205 ncookies--;
3206 }
3207 if (cpos >= cend || ncookies == 0) {
3208 toff = off;
3209 siz = fullsiz;
3210 goto again;
3211 }
3212
3213 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3214 2 * NFSX_UNSIGNED;
3215 nfsm_reply(cnt);
3216 nfsm_srvpostop_attr(getret, &at);
3217 tl = nfsm_build(u_int32_t *, 2 * NFSX_UNSIGNED);
3218 txdr_hyper(at.va_filerev, tl);
3219 mp = mb;
3220 bp = bpos;
3221 be = bp + M_TRAILINGSPACE(mp);
3222
3223 /* Loop through the records and build reply */
3224 while (cpos < cend && ncookies > 0) {
3225 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3226 nlen = dp->d_namlen;
3227 rem = nfsm_rndup(nlen)-nlen;
3228
3229 if (usevget) {
3230 /*
3231 * For readdir_and_lookup get the vnode using
3232 * the file number.
3233 */
3234 error = VFS_VGET(mntp, dp->d_fileno, LK_SHARED,
3235 &nvp);
3236 if (error != 0 && error != EOPNOTSUPP) {
3237 error = 0;
3238 goto invalid;
3239 } else if (error == EOPNOTSUPP) {
3240 /*
3241 * VFS_VGET() not supported?
3242 * Let's switch to VOP_LOOKUP().
3243 */
3244 error = 0;
3245 usevget = 0;
3246 cn.cn_nameiop = LOOKUP;
3247 cn.cn_flags = ISLASTCN | NOFOLLOW | \
3248 LOCKSHARED | LOCKLEAF | MPSAFE;
3249 cn.cn_lkflags = LK_SHARED | LK_RETRY;
3250 cn.cn_cred = cred;
3251 cn.cn_thread = curthread;
3252 }
3253 }
3254 if (!usevget) {
3255 cn.cn_nameptr = dp->d_name;
3256 cn.cn_namelen = dp->d_namlen;
3257 if (dp->d_namlen == 2 &&
3258 dp->d_name[0] == '.' &&
3259 dp->d_name[1] == '.') {
3260 cn.cn_flags |= ISDOTDOT;
3261 } else {
3262 cn.cn_flags &= ~ISDOTDOT;
3263 }
3264 if (!vp_locked) {
3265 vn_lock(vp, LK_SHARED | LK_RETRY);
3266 vp_locked = 1;
3267 }
3268 if ((vp->v_vflag & VV_ROOT) != 0 &&
3269 (cn.cn_flags & ISDOTDOT) != 0) {
3270 vref(vp);
3271 nvp = vp;
3272 } else if (VOP_LOOKUP(vp, &nvp, &cn) != 0)
3273 goto invalid;
3274 }
3275
3276 bzero((caddr_t)nfhp, NFSX_V3FH);
3277 nfhp->fh_fsid = nvp->v_mount->mnt_stat.f_fsid;
3278 if ((error1 = VOP_VPTOFH(nvp, &nfhp->fh_fid)) == 0)
3279 error1 = VOP_GETATTR(nvp, vap, cred);
3280 if (!usevget && vp == nvp)
3281 vunref(nvp);
3282 else
3283 vput(nvp);
3284 nvp = NULL;
3285 if (error1 != 0)
3286 goto invalid;
3287
3288 /*
3289 * If either the dircount or maxcount will be
3290 * exceeded, get out now. Both of these lengths
3291 * are calculated conservatively, including all
3292 * XDR overheads.
3293 */
3294 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3295 NFSX_V3POSTOPATTR);
3296 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3297 if (len > cnt || dirlen > fullsiz) {
3298 eofflag = 0;
3299 break;
3300 }
3301
3302 /*
3303 * Build the directory record xdr from
3304 * the dirent entry.
3305 */
3306 fp = (struct nfs_fattr *)&fl.fl_fattr;
3307 nfsm_srvfillattr(vap, fp);
3308 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3309 fl.fl_fhok = nfsrv_nfs_true;
3310 fl.fl_postopok = nfsrv_nfs_true;
3311 fl.fl_off.nfsuquad[0] = 0;
3312 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3313
3314 nfsm_clget;
3315 *tl = nfsrv_nfs_true;
3316 bp += NFSX_UNSIGNED;
3317 nfsm_clget;
3318 *tl = 0;
3319 bp += NFSX_UNSIGNED;
3320 nfsm_clget;
3321 *tl = txdr_unsigned(dp->d_fileno);
3322 bp += NFSX_UNSIGNED;
3323 nfsm_clget;
3324 *tl = txdr_unsigned(nlen);
3325 bp += NFSX_UNSIGNED;
3326
3327 /* And loop around copying the name */
3328 xfer = nlen;
3329 cp = dp->d_name;
3330 while (xfer > 0) {
3331 nfsm_clget;
3332 if ((bp + xfer) > be)
3333 tsiz = be - bp;
3334 else
3335 tsiz = xfer;
3336 bcopy(cp, bp, tsiz);
3337 bp += tsiz;
3338 xfer -= tsiz;
3339 if (xfer > 0)
3340 cp += tsiz;
3341 }
3342 /* And null pad to an int32_t boundary. */
3343 for (i = 0; i < rem; i++)
3344 *bp++ = '\0';
3345
3346 /*
3347 * Now copy the flrep structure out.
3348 */
3349 xfer = sizeof (struct flrep);
3350 cp = (caddr_t)&fl;
3351 while (xfer > 0) {
3352 nfsm_clget;
3353 if ((bp + xfer) > be)
3354 tsiz = be - bp;
3355 else
3356 tsiz = xfer;
3357 bcopy(cp, bp, tsiz);
3358 bp += tsiz;
3359 xfer -= tsiz;
3360 if (xfer > 0)
3361 cp += tsiz;
3362 }
3363 }
3364 invalid:
3365 cpos += dp->d_reclen;
3366 dp = (struct dirent *)cpos;
3367 cookiep++;
3368 ncookies--;
3369 }
3370 if (!usevget && vp_locked)
3371 vput(vp);
3372 else
3373 vrele(vp);
3374 vp = NULL;
3375 nfsm_clget;
3376 *tl = nfsrv_nfs_false;
3377 bp += NFSX_UNSIGNED;
3378 nfsm_clget;
3379 if (eofflag)
3380 *tl = nfsrv_nfs_true;
3381 else
3382 *tl = nfsrv_nfs_false;
3383 bp += NFSX_UNSIGNED;
3384 if (mp != mb) {
3385 if (bp < be)
3386 mp->m_len = bp - mtod(mp, caddr_t);
3387 } else
3388 mp->m_len += bp - bpos;
3389 free((caddr_t)cookies, M_TEMP);
3390 free((caddr_t)rbuf, M_TEMP);
3391 nfsmout:
3392 if (vp)
3393 vrele(vp);
3394 if (mntp)
3395 vfs_unbusy(mntp);
3396 VFS_UNLOCK_GIANT(vfslocked);
3397 return(error);
3398 }
3399
3400 /*
3401 * nfs commit service
3402 */
3403 int
3404 nfsrv_commit(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3405 struct mbuf **mrq)
3406 {
3407 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3408 struct sockaddr *nam = nfsd->nd_nam;
3409 caddr_t dpos = nfsd->nd_dpos;
3410 struct ucred *cred = nfsd->nd_cr;
3411 struct vattr bfor, aft;
3412 struct vnode *vp = NULL;
3413 nfsfh_t nfh;
3414 fhandle_t *fhp;
3415 u_int32_t *tl;
3416 caddr_t bpos;
3417 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt;
3418 struct mbuf *mb, *mreq;
3419 u_quad_t off;
3420 struct mount *mp = NULL;
3421 int v3 = (nfsd->nd_flag & ND_NFSV3);
3422 int tvfslocked;
3423 int vfslocked;
3424
3425 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3426 vfslocked = 0;
3427 if (!v3)
3428 panic("nfsrv_commit: v3 proc called on a v2 connection");
3429 fhp = &nfh.fh_generic;
3430 nfsm_srvmtofh(fhp);
3431 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL) {
3432 error = ESTALE;
3433 goto ereply;
3434 }
3435 vfslocked = VFS_LOCK_GIANT(mp);
3436 (void) vn_start_write(NULL, &mp, V_WAIT);
3437 vfs_rel(mp); /* The write holds a ref. */
3438 tl = nfsm_dissect_nonblock(u_int32_t *, 3 * NFSX_UNSIGNED);
3439
3440 /*
3441 * XXX At this time VOP_FSYNC() does not accept offset and byte
3442 * count parameters, so these arguments are useless (someday maybe).
3443 */
3444 off = fxdr_hyper(tl);
3445 tl += 2;
3446 cnt = fxdr_unsigned(int, *tl);
3447 error = nfsrv_fhtovp(fhp, 0, &vp, &tvfslocked, nfsd, slp, nam, &rdonly);
3448 vfslocked = nfsrv_lockedpair(vfslocked, tvfslocked);
3449 if (error) {
3450 nfsm_reply(2 * NFSX_UNSIGNED);
3451 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3452 error = 0;
3453 goto nfsmout;
3454 }
3455 for_ret = VOP_GETATTR(vp, &bfor, cred);
3456
3457 if (cnt > MAX_COMMIT_COUNT) {
3458 /*
3459 * Give up and do the whole thing
3460 */
3461 if (vp->v_object &&
3462 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3463 VM_OBJECT_LOCK(vp->v_object);
3464 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3465 VM_OBJECT_UNLOCK(vp->v_object);
3466 }
3467 error = VOP_FSYNC(vp, MNT_WAIT, curthread);
3468 } else {
3469 /*
3470 * Locate and synchronously write any buffers that fall
3471 * into the requested range. Note: we are assuming that
3472 * f_iosize is a power of 2.
3473 */
3474 int iosize = vp->v_mount->mnt_stat.f_iosize;
3475 int iomask = iosize - 1;
3476 struct bufobj *bo;
3477 daddr_t lblkno;
3478
3479 /*
3480 * Align to iosize boundry, super-align to page boundry.
3481 */
3482 if (off & iomask) {
3483 cnt += off & iomask;
3484 off &= ~(u_quad_t)iomask;
3485 }
3486 if (off & PAGE_MASK) {
3487 cnt += off & PAGE_MASK;
3488 off &= ~(u_quad_t)PAGE_MASK;
3489 }
3490 lblkno = off / iosize;
3491
3492 if (vp->v_object &&
3493 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3494 VM_OBJECT_LOCK(vp->v_object);
3495 vm_object_page_clean(vp->v_object, off, off + cnt,
3496 OBJPC_SYNC);
3497 VM_OBJECT_UNLOCK(vp->v_object);
3498 }
3499
3500 bo = &vp->v_bufobj;
3501 BO_LOCK(bo);
3502 while (cnt > 0) {
3503 struct buf *bp;
3504
3505 /*
3506 * If we have a buffer and it is marked B_DELWRI we
3507 * have to lock and write it. Otherwise the prior
3508 * write is assumed to have already been committed.
3509 *
3510 * gbincore() can return invalid buffers now so we
3511 * have to check that bit as well (though B_DELWRI
3512 * should not be set if B_INVAL is set there could be
3513 * a race here since we haven't locked the buffer).
3514 */
3515 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
3516 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
3517 LK_INTERLOCK, BO_MTX(bo)) == ENOLCK) {
3518 BO_LOCK(bo);
3519 continue; /* retry */
3520 }
3521 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
3522 B_DELWRI) {
3523 bremfree(bp);
3524 bp->b_flags &= ~B_ASYNC;
3525 bwrite(bp);
3526 ++nfs_commit_miss;
3527 } else
3528 BUF_UNLOCK(bp);
3529 BO_LOCK(bo);
3530 }
3531 ++nfs_commit_blks;
3532 if (cnt < iosize)
3533 break;
3534 cnt -= iosize;
3535 ++lblkno;
3536 }
3537 BO_UNLOCK(bo);
3538 }
3539
3540 aft_ret = VOP_GETATTR(vp, &aft, cred);
3541 vput(vp);
3542 vp = NULL;
3543 ereply:
3544 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3545 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3546 if (!error) {
3547 tl = nfsm_build(u_int32_t *, NFSX_V3WRITEVERF);
3548 if (nfsver.tv_sec == 0)
3549 nfsver = boottime;
3550 *tl++ = txdr_unsigned(nfsver.tv_sec);
3551 *tl = txdr_unsigned(nfsver.tv_usec);
3552 } else {
3553 error = 0;
3554 }
3555 nfsmout:
3556 if (vp)
3557 vput(vp);
3558 vn_finished_write(mp);
3559 VFS_UNLOCK_GIANT(vfslocked);
3560 return(error);
3561 }
3562
3563 /*
3564 * nfs statfs service
3565 */
3566 int
3567 nfsrv_statfs(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3568 struct mbuf **mrq)
3569 {
3570 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3571 struct sockaddr *nam = nfsd->nd_nam;
3572 caddr_t dpos = nfsd->nd_dpos;
3573 struct ucred *cred = nfsd->nd_cr;
3574 struct statfs *sf;
3575 struct nfs_statfs *sfp;
3576 caddr_t bpos;
3577 int error = 0, rdonly, getret = 1;
3578 int v3 = (nfsd->nd_flag & ND_NFSV3);
3579 struct mbuf *mb, *mreq;
3580 struct vnode *vp = NULL;
3581 struct vattr at;
3582 nfsfh_t nfh;
3583 fhandle_t *fhp;
3584 struct statfs statfs;
3585 u_quad_t tval;
3586 int vfslocked;
3587
3588 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3589 vfslocked = 0;
3590 fhp = &nfh.fh_generic;
3591 nfsm_srvmtofh(fhp);
3592 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
3593 if (error) {
3594 nfsm_reply(NFSX_UNSIGNED);
3595 if (v3)
3596 nfsm_srvpostop_attr(getret, &at);
3597 error = 0;
3598 goto nfsmout;
3599 }
3600 sf = &statfs;
3601 error = VFS_STATFS(vp->v_mount, sf);
3602 getret = VOP_GETATTR(vp, &at, cred);
3603 vput(vp);
3604 vp = NULL;
3605 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3606 if (v3)
3607 nfsm_srvpostop_attr(getret, &at);
3608 if (error) {
3609 error = 0;
3610 goto nfsmout;
3611 }
3612 sfp = nfsm_build(struct nfs_statfs *, NFSX_STATFS(v3));
3613 if (v3) {
3614 tval = (u_quad_t)sf->f_blocks;
3615 tval *= (u_quad_t)sf->f_bsize;
3616 txdr_hyper(tval, &sfp->sf_tbytes);
3617 tval = (u_quad_t)sf->f_bfree;
3618 tval *= (u_quad_t)sf->f_bsize;
3619 txdr_hyper(tval, &sfp->sf_fbytes);
3620 /*
3621 * Don't send negative values for available space,
3622 * since this field is unsigned in the NFS protocol.
3623 * Otherwise, the client would see absurdly high
3624 * numbers for free space.
3625 */
3626 if (sf->f_bavail < 0)
3627 tval = 0;
3628 else
3629 tval = (u_quad_t)sf->f_bavail;
3630 tval *= (u_quad_t)sf->f_bsize;
3631 txdr_hyper(tval, &sfp->sf_abytes);
3632 sfp->sf_tfiles.nfsuquad[0] = 0;
3633 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3634 sfp->sf_ffiles.nfsuquad[0] = 0;
3635 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3636 sfp->sf_afiles.nfsuquad[0] = 0;
3637 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3638 sfp->sf_invarsec = 0;
3639 } else {
3640 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3641 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3642 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3643 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3644 if (sf->f_bavail < 0)
3645 sfp->sf_bavail = 0;
3646 else
3647 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3648 }
3649 nfsmout:
3650 if (vp)
3651 vput(vp);
3652 VFS_UNLOCK_GIANT(vfslocked);
3653 return(error);
3654 }
3655
3656 /*
3657 * nfs fsinfo service
3658 */
3659 int
3660 nfsrv_fsinfo(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3661 struct mbuf **mrq)
3662 {
3663 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3664 struct sockaddr *nam = nfsd->nd_nam;
3665 caddr_t dpos = nfsd->nd_dpos;
3666 struct ucred *cred = nfsd->nd_cr;
3667 struct nfsv3_fsinfo *sip;
3668 caddr_t bpos;
3669 int error = 0, rdonly, getret = 1, pref;
3670 struct mbuf *mb, *mreq;
3671 struct vnode *vp = NULL;
3672 struct vattr at;
3673 nfsfh_t nfh;
3674 fhandle_t *fhp;
3675 u_quad_t maxfsize;
3676 struct statfs sb;
3677 int v3 = (nfsd->nd_flag & ND_NFSV3);
3678 int vfslocked;
3679
3680 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3681 if (!v3)
3682 panic("nfsrv_fsinfo: v3 proc called on a v2 connection");
3683 fhp = &nfh.fh_generic;
3684 vfslocked = 0;
3685 nfsm_srvmtofh(fhp);
3686 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
3687 if (error) {
3688 nfsm_reply(NFSX_UNSIGNED);
3689 nfsm_srvpostop_attr(getret, &at);
3690 error = 0;
3691 goto nfsmout;
3692 }
3693
3694 /* XXX Try to make a guess on the max file size. */
3695 VFS_STATFS(vp->v_mount, &sb);
3696 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3697
3698 getret = VOP_GETATTR(vp, &at, cred);
3699 vput(vp);
3700 vp = NULL;
3701 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3702 nfsm_srvpostop_attr(getret, &at);
3703 sip = nfsm_build(struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3704
3705 /*
3706 * XXX
3707 * There should be filesystem VFS OP(s) to get this information.
3708 * For now, assume ufs.
3709 */
3710 pref = NFS_SRVMAXDATA(nfsd);
3711 sip->fs_rtmax = txdr_unsigned(pref);
3712 sip->fs_rtpref = txdr_unsigned(pref);
3713 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3714 sip->fs_wtmax = txdr_unsigned(pref);
3715 sip->fs_wtpref = txdr_unsigned(pref);
3716 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3717 sip->fs_dtpref = txdr_unsigned(pref);
3718 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3719 sip->fs_timedelta.nfsv3_sec = 0;
3720 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3721 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3722 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3723 NFSV3FSINFO_CANSETTIME);
3724 nfsmout:
3725 if (vp)
3726 vput(vp);
3727 VFS_UNLOCK_GIANT(vfslocked);
3728 return(error);
3729 }
3730
3731 /*
3732 * nfs pathconf service
3733 */
3734 int
3735 nfsrv_pathconf(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3736 struct mbuf **mrq)
3737 {
3738 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3739 struct sockaddr *nam = nfsd->nd_nam;
3740 caddr_t dpos = nfsd->nd_dpos;
3741 struct ucred *cred = nfsd->nd_cr;
3742 struct nfsv3_pathconf *pc;
3743 caddr_t bpos;
3744 int error = 0, rdonly, getret = 1;
3745 register_t linkmax, namemax, chownres, notrunc;
3746 struct mbuf *mb, *mreq;
3747 struct vnode *vp = NULL;
3748 struct vattr at;
3749 nfsfh_t nfh;
3750 fhandle_t *fhp;
3751 int v3 = (nfsd->nd_flag & ND_NFSV3);
3752 int vfslocked;
3753
3754 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3755 if (!v3)
3756 panic("nfsrv_pathconf: v3 proc called on a v2 connection");
3757 vfslocked = 0;
3758 fhp = &nfh.fh_generic;
3759 nfsm_srvmtofh(fhp);
3760 error = nfsrv_fhtovp(fhp, 0, &vp, &vfslocked, nfsd, slp, nam, &rdonly);
3761 if (error) {
3762 nfsm_reply(NFSX_UNSIGNED);
3763 nfsm_srvpostop_attr(getret, &at);
3764 error = 0;
3765 goto nfsmout;
3766 }
3767 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3768 if (!error)
3769 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3770 if (!error)
3771 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3772 if (!error)
3773 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3774 getret = VOP_GETATTR(vp, &at, cred);
3775 vput(vp);
3776 vp = NULL;
3777 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3778 nfsm_srvpostop_attr(getret, &at);
3779 if (error) {
3780 error = 0;
3781 goto nfsmout;
3782 }
3783 pc = nfsm_build(struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3784
3785 pc->pc_linkmax = txdr_unsigned(linkmax);
3786 pc->pc_namemax = txdr_unsigned(namemax);
3787 pc->pc_notrunc = txdr_unsigned(notrunc);
3788 pc->pc_chownrestricted = txdr_unsigned(chownres);
3789
3790 /*
3791 * These should probably be supported by VOP_PATHCONF(), but
3792 * until msdosfs is exportable (why would you want to?), the
3793 * Unix defaults should be ok.
3794 */
3795 pc->pc_caseinsensitive = nfsrv_nfs_false;
3796 pc->pc_casepreserving = nfsrv_nfs_true;
3797 nfsmout:
3798 if (vp)
3799 vput(vp);
3800 VFS_UNLOCK_GIANT(vfslocked);
3801 return(error);
3802 }
3803
3804 /*
3805 * Null operation, used by clients to ping server
3806 */
3807 /* ARGSUSED */
3808 int
3809 nfsrv_null(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3810 struct mbuf **mrq)
3811 {
3812 struct mbuf *mrep = nfsd->nd_mrep;
3813 caddr_t bpos;
3814 int error = NFSERR_RETVOID;
3815 struct mbuf *mb, *mreq;
3816
3817 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3818 nfsm_reply(0);
3819 nfsmout:
3820 return (error);
3821 }
3822
3823 /*
3824 * No operation, used for obsolete procedures
3825 */
3826 /* ARGSUSED */
3827 int
3828 nfsrv_noop(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
3829 struct mbuf **mrq)
3830 {
3831 struct mbuf *mrep = nfsd->nd_mrep;
3832 caddr_t bpos;
3833 int error;
3834 struct mbuf *mb, *mreq;
3835
3836 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3837 if (nfsd->nd_repstat)
3838 error = nfsd->nd_repstat;
3839 else
3840 error = EPROCUNAVAIL;
3841 nfsm_reply(0);
3842 error = 0;
3843 nfsmout:
3844 return (error);
3845 }
3846
3847 /*
3848 * Perform access checking for vnodes obtained from file handles that would
3849 * refer to files already opened by a Unix client. You cannot just use
3850 * vn_writechk() and VOP_ACCESS() for two reasons.
3851 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
3852 * case.
3853 * 2 - The owner is to be given access irrespective of mode bits for some
3854 * operations, so that processes that chmod after opening a file don't
3855 * break. I don't like this because it opens a security hole, but since
3856 * the nfs server opens a security hole the size of a barn door anyhow,
3857 * what the heck.
3858 *
3859 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3860 * will return EPERM instead of EACCES. EPERM is always an error.
3861 */
3862 static int
3863 nfsrv_access(struct vnode *vp, accmode_t accmode, struct ucred *cred,
3864 int rdonly, int override)
3865 {
3866 struct vattr vattr;
3867 int error;
3868
3869 VFS_ASSERT_GIANT(vp->v_mount);
3870
3871 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3872
3873 if (accmode & VWRITE) {
3874 /* Just vn_writechk() changed to check rdonly */
3875 /*
3876 * Disallow write attempts on read-only filesystems;
3877 * unless the file is a socket or a block or character
3878 * device resident on the filesystem.
3879 */
3880 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3881 switch (vp->v_type) {
3882 case VREG:
3883 case VDIR:
3884 case VLNK:
3885 return (EROFS);
3886 default:
3887 break;
3888 }
3889 }
3890 /*
3891 * If there's shared text associated with
3892 * the inode, we can't allow writing.
3893 */
3894 if (VOP_IS_TEXT(vp))
3895 return (ETXTBSY);
3896 }
3897
3898 error = VOP_GETATTR(vp, &vattr, cred);
3899 if (error)
3900 return (error);
3901 error = VOP_ACCESS(vp, accmode, cred, curthread);
3902 /*
3903 * Allow certain operations for the owner (reads and writes
3904 * on files that are already open).
3905 */
3906 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3907 error = 0;
3908 return (error);
3909 }
Cache object: 2465d1ede738d4d69f12d7f03714e29e
|