nfs_serv.c revision 1.82 1 /* $NetBSD: nfs_serv.c,v 1.82 2003/10/15 11:29:01 hannken Exp $ */
2
3 /*
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95
35 */
36
37 /*
38 * nfs version 2 and 3 server calls to vnode ops
39 * - these routines generally have 3 phases
40 * 1 - break down and validate rpc request in mbuf list
41 * 2 - do the vnode ops for the request
42 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
43 * 3 - build the rpc reply in an mbuf list
44 * nb:
45 * - do not mix the phases, since the nfsm_?? macros can return failures
46 * on a bad rpc or similar and do not do any vrele() or vput()'s
47 *
48 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
49 * error number iff error != 0 whereas
50 * returning an error from the server function implies a fatal error
51 * such as a badly constructed rpc request that should be dropped without
52 * a reply.
53 * For Version 3, nfsm_reply() does not return for the error case, since
54 * most version 3 rpcs return more than the status for error cases.
55 */
56
57 #include <sys/cdefs.h>
58 __KERNEL_RCSID(0, "$NetBSD: nfs_serv.c,v 1.82 2003/10/15 11:29:01 hannken Exp $");
59
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/proc.h>
63 #include <sys/file.h>
64 #include <sys/namei.h>
65 #include <sys/vnode.h>
66 #include <sys/mount.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/mbuf.h>
70 #include <sys/dirent.h>
71 #include <sys/stat.h>
72 #include <sys/kernel.h>
73 #include <ufs/ufs/dir.h>
74
75 #include <uvm/uvm.h>
76
77 #include <nfs/nfsproto.h>
78 #include <nfs/rpcv2.h>
79 #include <nfs/nfs.h>
80 #include <nfs/xdr_subs.h>
81 #include <nfs/nfsm_subs.h>
82 #include <nfs/nqnfs.h>
83 #include <nfs/nfs_var.h>
84
85 /* Global vars */
86 extern u_int32_t nfs_xdrneg1;
87 extern u_int32_t nfs_false, nfs_true;
88 extern enum vtype nv3tov_type[8];
89 extern struct nfsstats nfsstats;
90 extern const nfstype nfsv2_type[9];
91 extern const nfstype nfsv3_type[9];
92 int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
93 int nfsd_use_loan = 1; /* use page-loan for READ OP */
94
95 /*
96 * nfs v3 access service
97 */
98 int
99 nfsrv3_access(nfsd, slp, procp, mrq)
100 struct nfsrv_descript *nfsd;
101 struct nfssvc_sock *slp;
102 struct proc *procp;
103 struct mbuf **mrq;
104 {
105 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
106 struct mbuf *nam = nfsd->nd_nam;
107 caddr_t dpos = nfsd->nd_dpos;
108 struct ucred *cred = &nfsd->nd_cr;
109 struct vnode *vp;
110 nfsfh_t nfh;
111 fhandle_t *fhp;
112 u_int32_t *tl;
113 int32_t t1;
114 caddr_t bpos;
115 int error = 0, rdonly, cache = 0, getret;
116 char *cp2;
117 struct mbuf *mb, *mreq;
118 struct vattr va;
119 u_long inmode, testmode, outmode;
120 u_quad_t frev;
121
122 fhp = &nfh.fh_generic;
123 nfsm_srvmtofh(fhp);
124 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
125 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
126 (nfsd->nd_flag & ND_KERBAUTH), FALSE);
127 if (error) {
128 nfsm_reply(NFSX_UNSIGNED);
129 nfsm_srvpostop_attr(1, (struct vattr *)0);
130 return (0);
131 }
132 inmode = fxdr_unsigned(u_int32_t, *tl);
133 outmode = 0;
134 if ((inmode & NFSV3ACCESS_READ) &&
135 nfsrv_access(vp, VREAD, cred, rdonly, procp, 0) == 0)
136 outmode |= NFSV3ACCESS_READ;
137 if (vp->v_type != VDIR) {
138 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
139 if (testmode &&
140 nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
141 outmode |= testmode;
142 if ((inmode & NFSV3ACCESS_EXECUTE) &&
143 nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
144 outmode |= NFSV3ACCESS_EXECUTE;
145 } else {
146 testmode = inmode & (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
147 NFSV3ACCESS_DELETE);
148 if (testmode &&
149 nfsrv_access(vp, VWRITE, cred, rdonly, procp, 0) == 0)
150 outmode |= testmode;
151 if ((inmode & NFSV3ACCESS_LOOKUP) &&
152 nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0) == 0)
153 outmode |= NFSV3ACCESS_LOOKUP;
154 }
155 getret = VOP_GETATTR(vp, &va, cred, procp);
156 vput(vp);
157 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
158 nfsm_srvpostop_attr(getret, &va);
159 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
160 *tl = txdr_unsigned(outmode);
161 nfsm_srvdone;
162 }
163
164 /*
165 * nfs getattr service
166 */
167 int
168 nfsrv_getattr(nfsd, slp, procp, mrq)
169 struct nfsrv_descript *nfsd;
170 struct nfssvc_sock *slp;
171 struct proc *procp;
172 struct mbuf **mrq;
173 {
174 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
175 struct mbuf *nam = nfsd->nd_nam;
176 caddr_t dpos = nfsd->nd_dpos;
177 struct ucred *cred = &nfsd->nd_cr;
178 struct nfs_fattr *fp;
179 struct vattr va;
180 struct vnode *vp;
181 nfsfh_t nfh;
182 fhandle_t *fhp;
183 u_int32_t *tl;
184 int32_t t1;
185 caddr_t bpos;
186 int error = 0, rdonly, cache;
187 char *cp2;
188 struct mbuf *mb, *mreq;
189 u_quad_t frev;
190
191 fhp = &nfh.fh_generic;
192 nfsm_srvmtofh(fhp);
193 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
194 (nfsd->nd_flag & ND_KERBAUTH), FALSE);
195 if (error) {
196 nfsm_reply(0);
197 return (0);
198 }
199 nqsrv_getl(vp, ND_READ);
200 error = VOP_GETATTR(vp, &va, cred, procp);
201 vput(vp);
202 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
203 if (error)
204 return (0);
205 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
206 nfsm_srvfillattr(&va, fp);
207 nfsm_srvdone;
208 }
209
210 /*
211 * nfs setattr service
212 */
213 int
214 nfsrv_setattr(nfsd, slp, procp, mrq)
215 struct nfsrv_descript *nfsd;
216 struct nfssvc_sock *slp;
217 struct proc *procp;
218 struct mbuf **mrq;
219 {
220 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
221 struct mbuf *nam = nfsd->nd_nam;
222 caddr_t dpos = nfsd->nd_dpos;
223 struct ucred *cred = &nfsd->nd_cr;
224 struct vattr va, preat;
225 struct nfsv2_sattr *sp;
226 struct nfs_fattr *fp;
227 struct vnode *vp;
228 nfsfh_t nfh;
229 fhandle_t *fhp;
230 u_int32_t *tl;
231 int32_t t1;
232 caddr_t bpos;
233 int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
234 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
235 char *cp2;
236 struct mbuf *mb, *mreq;
237 u_quad_t frev;
238 struct timespec guard;
239 struct mount *mp = NULL;
240
241 fhp = &nfh.fh_generic;
242 nfsm_srvmtofh(fhp);
243 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
244 return (ESTALE);
245 vn_start_write(NULL, &mp, V_WAIT);
246 VATTR_NULL(&va);
247 if (v3) {
248 nfsm_srvsattr(&va);
249 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
250 gcheck = fxdr_unsigned(int, *tl);
251 if (gcheck) {
252 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
253 fxdr_nfsv3time(tl, &guard);
254 }
255 } else {
256 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
257 /*
258 * Nah nah nah nah na nah
259 * There is a bug in the Sun client that puts 0xffff in the mode
260 * field of sattr when it should put in 0xffffffff. The u_short
261 * doesn't sign extend.
262 * --> check the low order 2 bytes for 0xffff
263 */
264 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
265 va.va_mode = nfstov_mode(sp->sa_mode);
266 if (sp->sa_uid != nfs_xdrneg1)
267 va.va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
268 if (sp->sa_gid != nfs_xdrneg1)
269 va.va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
270 if (sp->sa_size != nfs_xdrneg1)
271 va.va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
272 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
273 #ifdef notyet
274 fxdr_nfsv2time(&sp->sa_atime, &va.va_atime);
275 #else
276 va.va_atime.tv_sec =
277 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
278 va.va_atime.tv_nsec = 0;
279 #endif
280 }
281 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
282 fxdr_nfsv2time(&sp->sa_mtime, &va.va_mtime);
283
284 }
285
286 /*
287 * Now that we have all the fields, lets do it.
288 */
289 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
290 (nfsd->nd_flag & ND_KERBAUTH), FALSE);
291 if (error) {
292 nfsm_reply(2 * NFSX_UNSIGNED);
293 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
294 vn_finished_write(mp, 0);
295 return (0);
296 }
297 nqsrv_getl(vp, ND_WRITE);
298 if (v3) {
299 error = preat_ret = VOP_GETATTR(vp, &preat, cred, procp);
300 if (!error && gcheck &&
301 (preat.va_ctime.tv_sec != guard.tv_sec ||
302 preat.va_ctime.tv_nsec != guard.tv_nsec))
303 error = NFSERR_NOT_SYNC;
304 if (error) {
305 vput(vp);
306 nfsm_reply(NFSX_WCCDATA(v3));
307 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
308 vn_finished_write(mp, 0);
309 return (0);
310 }
311 }
312
313 /*
314 * If the size is being changed write acces is required, otherwise
315 * just check for a read only file system.
316 */
317 if (va.va_size == ((u_quad_t)((quad_t) -1))) {
318 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
319 error = EROFS;
320 goto out;
321 }
322 } else {
323 if (vp->v_type == VDIR) {
324 error = EISDIR;
325 goto out;
326 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
327 procp, 0)) != 0)
328 goto out;
329 }
330 error = VOP_SETATTR(vp, &va, cred, procp);
331 postat_ret = VOP_GETATTR(vp, &va, cred, procp);
332 if (!error)
333 error = postat_ret;
334 out:
335 vput(vp);
336 vn_finished_write(mp, 0);
337 nfsm_reply(NFSX_WCCORFATTR(v3));
338 if (v3) {
339 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, &va);
340 return (0);
341 } else {
342 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
343 nfsm_srvfillattr(&va, fp);
344 }
345 nfsm_srvdone;
346 }
347
348 /*
349 * nfs lookup rpc
350 */
351 int
352 nfsrv_lookup(nfsd, slp, procp, mrq)
353 struct nfsrv_descript *nfsd;
354 struct nfssvc_sock *slp;
355 struct proc *procp;
356 struct mbuf **mrq;
357 {
358 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
359 struct mbuf *nam = nfsd->nd_nam;
360 caddr_t dpos = nfsd->nd_dpos;
361 struct ucred *cred = &nfsd->nd_cr;
362 struct nfs_fattr *fp;
363 struct nameidata nd, ind, *ndp = &nd;
364 struct vnode *vp, *dirp;
365 nfsfh_t nfh;
366 fhandle_t *fhp;
367 caddr_t cp;
368 u_int32_t *tl;
369 int32_t t1;
370 caddr_t bpos;
371 int error = 0, cache, dirattr_ret = 1;
372 uint32_t len;
373 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
374 char *cp2;
375 struct mbuf *mb, *mreq;
376 struct vattr va, dirattr;
377 u_quad_t frev;
378
379 fhp = &nfh.fh_generic;
380 nfsm_srvmtofh(fhp);
381 nfsm_srvnamesiz(len);
382
383 pubflag = nfs_ispublicfh(fhp);
384
385 nd.ni_cnd.cn_cred = cred;
386 nd.ni_cnd.cn_nameiop = LOOKUP;
387 nd.ni_cnd.cn_flags = LOCKLEAF | SAVESTART;
388 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
389 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
390
391 if (!error && pubflag) {
392 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
393 /*
394 * Setup call to lookup() to see if we can find
395 * the index file. Arguably, this doesn't belong
396 * in a kernel.. Ugh.
397 */
398 ind = nd;
399 VOP_UNLOCK(nd.ni_vp, 0);
400 ind.ni_pathlen = strlen(nfs_pub.np_index);
401 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
402 nfs_pub.np_index;
403 ind.ni_startdir = nd.ni_vp;
404 VREF(ind.ni_startdir);
405 error = lookup(&ind);
406 if (!error) {
407 /*
408 * Found an index file. Get rid of
409 * the old references.
410 */
411 if (dirp)
412 vrele(dirp);
413 dirp = nd.ni_vp;
414 vrele(nd.ni_startdir);
415 ndp = &ind;
416 } else
417 error = 0;
418 }
419 /*
420 * If the public filehandle was used, check that this lookup
421 * didn't result in a filehandle outside the publicly exported
422 * filesystem.
423 */
424
425 if (!error && ndp->ni_vp->v_mount != nfs_pub.np_mount) {
426 vput(nd.ni_vp);
427 error = EPERM;
428 }
429 }
430
431 if (dirp) {
432 if (v3)
433 dirattr_ret = VOP_GETATTR(dirp, &dirattr, cred,
434 procp);
435 vrele(dirp);
436 }
437
438 if (error) {
439 nfsm_reply(NFSX_POSTOPATTR(v3));
440 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
441 return (0);
442 }
443
444 nqsrv_getl(ndp->ni_startdir, ND_READ);
445 vrele(ndp->ni_startdir);
446 PNBUF_PUT(nd.ni_cnd.cn_pnbuf);
447 vp = ndp->ni_vp;
448 memset((caddr_t)fhp, 0, sizeof(nfh));
449 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
450 error = VFS_VPTOFH(vp, &fhp->fh_fid);
451 if (!error)
452 error = VOP_GETATTR(vp, &va, cred, procp);
453 vput(vp);
454 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
455 if (error) {
456 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
457 return (0);
458 }
459 nfsm_srvfhtom(fhp, v3);
460 if (v3) {
461 nfsm_srvpostop_attr(0, &va);
462 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
463 } else {
464 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
465 nfsm_srvfillattr(&va, fp);
466 }
467 nfsm_srvdone;
468 }
469
470 /*
471 * nfs readlink service
472 */
473 int
474 nfsrv_readlink(nfsd, slp, procp, mrq)
475 struct nfsrv_descript *nfsd;
476 struct nfssvc_sock *slp;
477 struct proc *procp;
478 struct mbuf **mrq;
479 {
480 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
481 struct mbuf *nam = nfsd->nd_nam;
482 caddr_t dpos = nfsd->nd_dpos;
483 struct ucred *cred = &nfsd->nd_cr;
484 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
485 struct iovec *ivp = iv;
486 struct mbuf *mp;
487 u_int32_t *tl;
488 int32_t t1;
489 caddr_t bpos;
490 int error = 0, rdonly, cache, i, padlen, getret;
491 uint32_t len;
492 int v3 = (nfsd->nd_flag & ND_NFSV3);
493 char *cp2;
494 struct mbuf *mb, *mp2 = NULL, *mp3 = NULL, *mreq;
495 struct vnode *vp;
496 struct vattr attr;
497 nfsfh_t nfh;
498 fhandle_t *fhp;
499 struct uio io, *uiop = &io;
500 u_quad_t frev;
501
502 fhp = &nfh.fh_generic;
503 nfsm_srvmtofh(fhp);
504 len = 0;
505 i = 0;
506 while (len < NFS_MAXPATHLEN) {
507 mp = m_get(M_WAIT, MT_DATA);
508 MCLAIM(mp, &nfs_mowner);
509 m_clget(mp, M_WAIT);
510 mp->m_len = NFSMSIZ(mp);
511 if (len == 0)
512 mp3 = mp2 = mp;
513 else {
514 mp2->m_next = mp;
515 mp2 = mp;
516 }
517 if ((len+mp->m_len) > NFS_MAXPATHLEN) {
518 mp->m_len = NFS_MAXPATHLEN-len;
519 len = NFS_MAXPATHLEN;
520 } else
521 len += mp->m_len;
522 ivp->iov_base = mtod(mp, caddr_t);
523 ivp->iov_len = mp->m_len;
524 i++;
525 ivp++;
526 }
527 uiop->uio_iov = iv;
528 uiop->uio_iovcnt = i;
529 uiop->uio_offset = 0;
530 uiop->uio_resid = len;
531 uiop->uio_rw = UIO_READ;
532 uiop->uio_segflg = UIO_SYSSPACE;
533 uiop->uio_procp = (struct proc *)0;
534 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
535 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
536 if (error) {
537 m_freem(mp3);
538 nfsm_reply(2 * NFSX_UNSIGNED);
539 nfsm_srvpostop_attr(1, (struct vattr *)0);
540 return (0);
541 }
542 if (vp->v_type != VLNK) {
543 if (v3)
544 error = EINVAL;
545 else
546 error = ENXIO;
547 goto out;
548 }
549 nqsrv_getl(vp, ND_READ);
550 error = VOP_READLINK(vp, uiop, cred);
551 out:
552 getret = VOP_GETATTR(vp, &attr, cred, procp);
553 vput(vp);
554 if (error)
555 m_freem(mp3);
556 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
557 if (v3) {
558 nfsm_srvpostop_attr(getret, &attr);
559 if (error)
560 return (0);
561 }
562 len -= uiop->uio_resid;
563 padlen = nfsm_padlen(len);
564 if (uiop->uio_resid || padlen)
565 nfs_zeropad(mp3, uiop->uio_resid, padlen);
566 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
567 *tl = txdr_unsigned(len);
568 mb->m_next = mp3;
569 nfsm_srvdone;
570 }
571
572 /*
573 * XXX UBC temp limit
574 * maximum number of pages we can do VOP_GETPAGES and loan-out at once.
575 * should be <= MAX_READ_AHEAD in genfs_vnops.c
576 */
577 #define NFSD_READ_GETPAGES_CHUNK 16
578
579 /*
580 * nfs read service
581 */
582 int
583 nfsrv_read(nfsd, slp, procp, mrq)
584 struct nfsrv_descript *nfsd;
585 struct nfssvc_sock *slp;
586 struct proc *procp;
587 struct mbuf **mrq;
588 {
589 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
590 struct mbuf *nam = nfsd->nd_nam;
591 caddr_t dpos = nfsd->nd_dpos;
592 struct ucred *cred = &nfsd->nd_cr;
593 struct mbuf *m;
594 struct nfs_fattr *fp;
595 u_int32_t *tl;
596 int32_t t1;
597 int i;
598 caddr_t bpos;
599 int error = 0, rdonly, cache, getret;
600 int v3 = (nfsd->nd_flag & ND_NFSV3);
601 uint32_t reqlen, len, cnt, left;
602 int padlen;
603 char *cp2;
604 struct mbuf *mb, *mreq;
605 struct vnode *vp;
606 nfsfh_t nfh;
607 fhandle_t *fhp;
608 struct uio io, *uiop = &io;
609 struct vattr va;
610 off_t off;
611 u_quad_t frev;
612
613 fhp = &nfh.fh_generic;
614 nfsm_srvmtofh(fhp);
615 if (v3) {
616 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
617 off = fxdr_hyper(tl);
618 } else {
619 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
620 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
621 }
622 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
623 reqlen = fxdr_unsigned(uint32_t, *tl);
624 reqlen = MIN(reqlen, NFS_SRVMAXDATA(nfsd));
625 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
626 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
627 if (error) {
628 nfsm_reply(2 * NFSX_UNSIGNED);
629 nfsm_srvpostop_attr(1, (struct vattr *)0);
630 return (0);
631 }
632 if (vp->v_type != VREG) {
633 if (v3)
634 error = EINVAL;
635 else
636 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
637 }
638 if (!error) {
639 nqsrv_getl(vp, ND_READ);
640 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, procp, 1)) != 0)
641 error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 1);
642 }
643 getret = VOP_GETATTR(vp, &va, cred, procp);
644 if (!error)
645 error = getret;
646 if (error) {
647 vput(vp);
648 nfsm_reply(NFSX_POSTOPATTR(v3));
649 nfsm_srvpostop_attr(getret, &va);
650 return (0);
651 }
652 if (off >= va.va_size)
653 cnt = 0;
654 else if ((off + reqlen) > va.va_size)
655 cnt = va.va_size - off;
656 else
657 cnt = reqlen;
658 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
659 if (v3) {
660 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
661 *tl++ = nfs_true;
662 fp = (struct nfs_fattr *)tl;
663 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
664 } else {
665 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
666 fp = (struct nfs_fattr *)tl;
667 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
668 }
669 len = left = cnt;
670 if (cnt > 0) {
671 if (nfsd_use_loan) {
672 struct vm_page **pgpp;
673 voff_t pgoff = trunc_page(off);
674 int orignpages, nleftpages;
675 vaddr_t lva, curlva;
676
677 orignpages = (round_page(off + cnt) - pgoff)
678 >> PAGE_SHIFT;
679 KASSERT(orignpages <= M_EXT_MAXPAGES); /* XXX */
680
681 lva = sokvaalloc(orignpages << PAGE_SHIFT, slp->ns_so);
682 if (lva == 0) {
683 /* fall back to VOP_READ */
684 goto loan_fail;
685 }
686
687 m = m_get(M_WAIT, MT_DATA);
688 pgpp = m->m_ext.ext_pgs;
689
690 curlva = lva;
691 nleftpages = orignpages;
692 while (nleftpages > 0) {
693 int npages = nleftpages;
694 if (npages > NFSD_READ_GETPAGES_CHUNK)
695 npages = NFSD_READ_GETPAGES_CHUNK;
696 again:
697 simple_lock(&vp->v_interlock);
698 error = VOP_GETPAGES(vp, pgoff, pgpp, &npages,
699 0, VM_PROT_READ, 0, PGO_SYNCIO);
700 if (error == EAGAIN) {
701 tsleep(&lbolt, PVM, "nfsread", 0);
702 goto again;
703 }
704 if (error) {
705 uvm_unloan(m->m_ext.ext_pgs,
706 orignpages - nleftpages,
707 UVM_LOAN_TOPAGE);
708 sokvafree(lva,
709 orignpages << PAGE_SHIFT);
710 m_free(m);
711 goto read_error;
712 }
713
714 /* loan and unbusy pages */
715 simple_lock(&vp->v_interlock);
716 for (i = 0; i < npages; i++) {
717 if (pgpp[i]->flags & PG_RELEASED) {
718 uvm_lock_pageq();
719 uvm_page_unbusy(pgpp, npages);
720 uvm_unlock_pageq();
721 simple_unlock(&vp->v_interlock);
722 continue;
723 }
724 }
725 uvm_loanuobjpages(pgpp, npages);
726 simple_unlock(&vp->v_interlock);
727
728 /* map pages */
729 for (i = 0; i < npages; i++) {
730 pmap_kenter_pa(curlva,
731 VM_PAGE_TO_PHYS(pgpp[i]),
732 VM_PROT_READ);
733 curlva += PAGE_SIZE;
734 }
735 nleftpages -= npages;
736 pgpp += npages;
737 pgoff += npages << PAGE_SHIFT;
738 }
739
740 lva += off & PAGE_MASK;
741
742 MCLAIM(m, &nfs_mowner);
743 MEXTADD(m, (void *)lva, cnt, M_MBUF, soloanfree,
744 slp->ns_so);
745 m->m_flags |= M_EXT_PAGES | M_EXT_ROMAP;
746 m->m_len = cnt;
747
748 pmap_update(pmap_kernel());
749 mb->m_next = m;
750 mb = m;
751 error = 0;
752 uiop->uio_resid = 0;
753 } else {
754 struct iovec *iv;
755 struct iovec *iv2;
756 struct mbuf *m2;
757 int siz;
758 loan_fail:
759 /*
760 * Generate the mbuf list with the uio_iov ref. to it.
761 */
762 i = 0;
763 m = m2 = mb;
764 while (left > 0) {
765 siz = min(M_TRAILINGSPACE(m), left);
766 if (siz > 0) {
767 left -= siz;
768 i++;
769 }
770 if (left > 0) {
771 m = m_get(M_WAIT, MT_DATA);
772 MCLAIM(m, &nfs_mowner);
773 m_clget(m, M_WAIT);
774 m->m_len = 0;
775 m2->m_next = m;
776 m2 = m;
777 }
778 }
779 iv = malloc(i * sizeof(struct iovec), M_TEMP, M_WAITOK);
780 uiop->uio_iov = iv2 = iv;
781 m = mb;
782 left = cnt;
783 i = 0;
784 while (left > 0) {
785 if (m == NULL)
786 panic("nfsrv_read iov");
787 siz = min(M_TRAILINGSPACE(m), left);
788 if (siz > 0) {
789 iv->iov_base = mtod(m, caddr_t) +
790 m->m_len;
791 iv->iov_len = siz;
792 m->m_len += siz;
793 left -= siz;
794 iv++;
795 i++;
796 }
797 m = m->m_next;
798 }
799 uiop->uio_iovcnt = i;
800 uiop->uio_offset = off;
801 uiop->uio_resid = cnt;
802 uiop->uio_rw = UIO_READ;
803 uiop->uio_segflg = UIO_SYSSPACE;
804 error = VOP_READ(vp, uiop, IO_NODELOCKED, cred);
805 off = uiop->uio_offset;
806 free((caddr_t)iv2, M_TEMP);
807 }
808 read_error:
809 if (error || (getret = VOP_GETATTR(vp, &va, cred, procp)) != 0){
810 if (!error)
811 error = getret;
812 m_freem(mreq);
813 vput(vp);
814 nfsm_reply(NFSX_POSTOPATTR(v3));
815 nfsm_srvpostop_attr(getret, &va);
816 return (0);
817 }
818 } else {
819 uiop->uio_resid = 0;
820 }
821 vput(vp);
822 nfsm_srvfillattr(&va, fp);
823 len -= uiop->uio_resid;
824 padlen = nfsm_padlen(len);
825 if (uiop->uio_resid || padlen)
826 nfs_zeropad(mb, uiop->uio_resid, padlen);
827 if (v3) {
828 *tl++ = txdr_unsigned(len);
829 if (len < reqlen)
830 *tl++ = nfs_true;
831 else
832 *tl++ = nfs_false;
833 }
834 *tl = txdr_unsigned(len);
835 nfsm_srvdone;
836 }
837
838 /*
839 * nfs write service
840 */
841 int
842 nfsrv_write(nfsd, slp, procp, mrq)
843 struct nfsrv_descript *nfsd;
844 struct nfssvc_sock *slp;
845 struct proc *procp;
846 struct mbuf **mrq;
847 {
848 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
849 struct mbuf *nam = nfsd->nd_nam;
850 caddr_t dpos = nfsd->nd_dpos;
851 struct ucred *cred = &nfsd->nd_cr;
852 struct iovec *ivp;
853 int i, cnt;
854 struct mbuf *mp;
855 struct nfs_fattr *fp;
856 struct iovec *iv;
857 struct vattr va, forat;
858 u_int32_t *tl;
859 int32_t t1;
860 caddr_t bpos;
861 int error = 0, rdonly, cache, len, forat_ret = 1;
862 int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
863 int stable = NFSV3WRITE_FILESYNC;
864 int v3 = (nfsd->nd_flag & ND_NFSV3);
865 char *cp2;
866 struct mbuf *mb, *mreq;
867 struct vnode *vp;
868 nfsfh_t nfh;
869 fhandle_t *fhp;
870 struct uio io, *uiop = &io;
871 off_t off;
872 u_quad_t frev;
873 struct mount *mntp = NULL;
874
875 if (mrep == NULL) {
876 *mrq = NULL;
877 return (0);
878 }
879 fhp = &nfh.fh_generic;
880 nfsm_srvmtofh(fhp);
881 if ((mntp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
882 return (ESTALE);
883 vn_start_write(NULL, &mntp, V_WAIT);
884 if (v3) {
885 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
886 off = fxdr_hyper(tl);
887 tl += 3;
888 stable = fxdr_unsigned(int, *tl++);
889 } else {
890 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
891 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
892 tl += 2;
893 }
894 retlen = len = fxdr_unsigned(int32_t, *tl);
895 cnt = i = 0;
896
897 /*
898 * For NFS Version 2, it is not obvious what a write of zero length
899 * should do, but I might as well be consistent with Version 3,
900 * which is to return ok so long as there are no permission problems.
901 */
902 if (len > 0) {
903 zeroing = 1;
904 mp = mrep;
905 while (mp) {
906 if (mp == md) {
907 zeroing = 0;
908 adjust = dpos - mtod(mp, caddr_t);
909 mp->m_len -= adjust;
910 if (mp->m_len > 0 && adjust > 0)
911 NFSMADV(mp, adjust);
912 }
913 if (zeroing)
914 mp->m_len = 0;
915 else if (mp->m_len > 0) {
916 i += mp->m_len;
917 if (i > len) {
918 mp->m_len -= (i - len);
919 zeroing = 1;
920 }
921 if (mp->m_len > 0)
922 cnt++;
923 }
924 mp = mp->m_next;
925 }
926 }
927 if (len > NFS_MAXDATA || len < 0 || i < len) {
928 error = EIO;
929 nfsm_reply(2 * NFSX_UNSIGNED);
930 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
931 vn_finished_write(mntp, 0);
932 return (0);
933 }
934 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
935 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
936 if (error) {
937 nfsm_reply(2 * NFSX_UNSIGNED);
938 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
939 vn_finished_write(mntp, 0);
940 return (0);
941 }
942 if (v3)
943 forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
944 if (vp->v_type != VREG) {
945 if (v3)
946 error = EINVAL;
947 else
948 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
949 }
950 if (!error) {
951 nqsrv_getl(vp, ND_WRITE);
952 error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
953 }
954 if (error) {
955 vput(vp);
956 nfsm_reply(NFSX_WCCDATA(v3));
957 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
958 vn_finished_write(mntp, 0);
959 return (0);
960 }
961
962 if (len > 0) {
963 ivp = malloc(cnt * sizeof (struct iovec), M_TEMP, M_WAITOK);
964 uiop->uio_iov = iv = ivp;
965 uiop->uio_iovcnt = cnt;
966 mp = mrep;
967 while (mp) {
968 if (mp->m_len > 0) {
969 ivp->iov_base = mtod(mp, caddr_t);
970 ivp->iov_len = mp->m_len;
971 ivp++;
972 }
973 mp = mp->m_next;
974 }
975
976 /*
977 * XXX
978 * The IO_METASYNC flag indicates that all metadata (and not
979 * just enough to ensure data integrity) must be written to
980 * stable storage synchronously.
981 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
982 */
983 if (stable == NFSV3WRITE_UNSTABLE)
984 ioflags = IO_NODELOCKED;
985 else if (stable == NFSV3WRITE_DATASYNC)
986 ioflags = (IO_SYNC | IO_NODELOCKED);
987 else
988 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
989 uiop->uio_resid = len;
990 uiop->uio_rw = UIO_WRITE;
991 uiop->uio_segflg = UIO_SYSSPACE;
992 uiop->uio_procp = (struct proc *)0;
993 uiop->uio_offset = off;
994 error = VOP_WRITE(vp, uiop, ioflags, cred);
995 nfsstats.srvvop_writes++;
996 free(iv, M_TEMP);
997 }
998 aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
999 vput(vp);
1000 vn_finished_write(mntp, 0);
1001 if (!error)
1002 error = aftat_ret;
1003 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1004 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1005 if (v3) {
1006 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1007 if (error)
1008 return (0);
1009 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1010 *tl++ = txdr_unsigned(retlen);
1011 if (stable == NFSV3WRITE_UNSTABLE)
1012 *tl++ = txdr_unsigned(stable);
1013 else
1014 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1015 /*
1016 * Actually, there is no need to txdr these fields,
1017 * but it may make the values more human readable,
1018 * for debugging purposes.
1019 */
1020 *tl++ = txdr_unsigned(boottime.tv_sec);
1021 *tl = txdr_unsigned(boottime.tv_usec);
1022 } else {
1023 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1024 nfsm_srvfillattr(&va, fp);
1025 }
1026 nfsm_srvdone;
1027 }
1028
1029 /*
1030 * NFS write service with write gathering support. Called when
1031 * nfsrvw_procrastinate > 0.
1032 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1033 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1034 * Jan. 1994.
1035 */
1036 int
1037 nfsrv_writegather(ndp, slp, procp, mrq)
1038 struct nfsrv_descript **ndp;
1039 struct nfssvc_sock *slp;
1040 struct proc *procp;
1041 struct mbuf **mrq;
1042 {
1043 struct iovec *ivp;
1044 struct mbuf *mp;
1045 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1046 struct nfs_fattr *fp;
1047 int i = 0;
1048 struct iovec *iov;
1049 struct nfsrvw_delayhash *wpp;
1050 struct ucred *cred;
1051 struct vattr va, forat;
1052 u_int32_t *tl;
1053 int32_t t1;
1054 caddr_t bpos, dpos;
1055 int error = 0, rdonly, cache, len = 0, forat_ret = 1;
1056 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1057 char *cp2;
1058 struct mbuf *mb, *mreq, *mrep, *md;
1059 struct vnode *vp;
1060 struct uio io, *uiop = &io;
1061 u_quad_t frev, cur_usec;
1062 struct mount *mntp = NULL;
1063
1064 *mrq = NULL;
1065 if (*ndp) {
1066 nfsd = *ndp;
1067 *ndp = NULL;
1068 mrep = nfsd->nd_mrep;
1069 md = nfsd->nd_md;
1070 dpos = nfsd->nd_dpos;
1071 cred = &nfsd->nd_cr;
1072 v3 = (nfsd->nd_flag & ND_NFSV3);
1073 LIST_INIT(&nfsd->nd_coalesce);
1074 nfsd->nd_mreq = NULL;
1075 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1076 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1077 nfsd->nd_time = cur_usec + nfsrvw_procrastinate;
1078
1079 /*
1080 * Now, get the write header..
1081 */
1082 nfsm_srvmtofh(&nfsd->nd_fh);
1083 if (v3) {
1084 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1085 nfsd->nd_off = fxdr_hyper(tl);
1086 tl += 3;
1087 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1088 } else {
1089 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1090 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1091 tl += 2;
1092 }
1093 len = fxdr_unsigned(int32_t, *tl);
1094 nfsd->nd_len = len;
1095 nfsd->nd_eoff = nfsd->nd_off + len;
1096
1097 /*
1098 * Trim the header out of the mbuf list and trim off any trailing
1099 * junk so that the mbuf list has only the write data.
1100 */
1101 zeroing = 1;
1102 i = 0;
1103 mp = mrep;
1104 while (mp) {
1105 if (mp == md) {
1106 zeroing = 0;
1107 adjust = dpos - mtod(mp, caddr_t);
1108 mp->m_len -= adjust;
1109 if (mp->m_len > 0 && adjust > 0)
1110 NFSMADV(mp, adjust);
1111 }
1112 if (zeroing)
1113 mp->m_len = 0;
1114 else {
1115 i += mp->m_len;
1116 if (i > len) {
1117 mp->m_len -= (i - len);
1118 zeroing = 1;
1119 }
1120 }
1121 mp = mp->m_next;
1122 }
1123 if (len > NFS_MAXDATA || len < 0 || i < len) {
1124 nfsmout:
1125 m_freem(mrep);
1126 error = EIO;
1127 nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1128 if (v3)
1129 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1130 nfsd->nd_mreq = mreq;
1131 nfsd->nd_mrep = NULL;
1132 nfsd->nd_time = 0;
1133 }
1134
1135 /*
1136 * Add this entry to the hash and time queues.
1137 */
1138 s = splsoftclock();
1139 owp = NULL;
1140 wp = LIST_FIRST(&slp->ns_tq);
1141 while (wp && wp->nd_time < nfsd->nd_time) {
1142 owp = wp;
1143 wp = LIST_NEXT(wp, nd_tq);
1144 }
1145 if (owp) {
1146 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1147 } else {
1148 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1149 }
1150 if (nfsd->nd_mrep) {
1151 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1152 owp = NULL;
1153 wp = LIST_FIRST(wpp);
1154 while (wp &&
1155 memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1156 owp = wp;
1157 wp = LIST_NEXT(wp, nd_hash);
1158 }
1159 while (wp && wp->nd_off < nfsd->nd_off &&
1160 !memcmp((caddr_t)&nfsd->nd_fh, (caddr_t)&wp->nd_fh, NFSX_V3FH)) {
1161 owp = wp;
1162 wp = LIST_NEXT(wp, nd_hash);
1163 }
1164 if (owp) {
1165 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1166
1167 /*
1168 * Search the hash list for overlapping entries and
1169 * coalesce.
1170 */
1171 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1172 wp = LIST_NEXT(nfsd, nd_hash);
1173 if (NFSW_SAMECRED(owp, nfsd))
1174 nfsrvw_coalesce(owp, nfsd);
1175 }
1176 } else {
1177 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1178 }
1179 }
1180 splx(s);
1181 }
1182
1183 /*
1184 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1185 * and generate the associated reply mbuf list(s).
1186 */
1187 loop1:
1188 cur_usec = (u_quad_t)time.tv_sec * 1000000 + (u_quad_t)time.tv_usec;
1189 s = splsoftclock();
1190 for (nfsd = LIST_FIRST(&slp->ns_tq); nfsd; nfsd = owp) {
1191 owp = LIST_NEXT(nfsd, nd_tq);
1192 if (nfsd->nd_time > cur_usec)
1193 break;
1194 if (nfsd->nd_mreq)
1195 continue;
1196 LIST_REMOVE(nfsd, nd_tq);
1197 LIST_REMOVE(nfsd, nd_hash);
1198 splx(s);
1199 mrep = nfsd->nd_mrep;
1200 nfsd->nd_mrep = NULL;
1201 cred = &nfsd->nd_cr;
1202 v3 = (nfsd->nd_flag & ND_NFSV3);
1203 forat_ret = aftat_ret = 1;
1204 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1205 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH),
1206 FALSE);
1207 if (!error) {
1208 if (v3)
1209 forat_ret = VOP_GETATTR(vp, &forat, cred, procp);
1210 if (vp->v_type != VREG) {
1211 if (v3)
1212 error = EINVAL;
1213 else
1214 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1215 }
1216 } else
1217 vp = NULL;
1218 if (!error) {
1219 nqsrv_getl(vp, ND_WRITE);
1220 error = nfsrv_access(vp, VWRITE, cred, rdonly, procp, 1);
1221 }
1222
1223 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1224 ioflags = IO_NODELOCKED;
1225 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1226 ioflags = (IO_SYNC | IO_NODELOCKED);
1227 else
1228 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1229 uiop->uio_rw = UIO_WRITE;
1230 uiop->uio_segflg = UIO_SYSSPACE;
1231 uiop->uio_procp = (struct proc *)0;
1232 uiop->uio_offset = nfsd->nd_off;
1233 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1234 if (uiop->uio_resid > 0) {
1235 mp = mrep;
1236 i = 0;
1237 while (mp) {
1238 if (mp->m_len > 0)
1239 i++;
1240 mp = mp->m_next;
1241 }
1242 uiop->uio_iovcnt = i;
1243 iov = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
1244 uiop->uio_iov = ivp = iov;
1245 mp = mrep;
1246 while (mp) {
1247 if (mp->m_len > 0) {
1248 ivp->iov_base = mtod(mp, caddr_t);
1249 ivp->iov_len = mp->m_len;
1250 ivp++;
1251 }
1252 mp = mp->m_next;
1253 }
1254 if (!error) {
1255 if (vn_start_write(vp, &mntp, V_NOWAIT) != 0) {
1256 VOP_UNLOCK(vp, 0);
1257 error = vn_start_write(NULL, &mntp, V_WAIT);
1258 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1259 }
1260 if (!error) {
1261 error = VOP_WRITE(vp, uiop, ioflags, cred);
1262 nfsstats.srvvop_writes++;
1263 vn_finished_write(mntp, 0);
1264 }
1265 }
1266 free((caddr_t)iov, M_TEMP);
1267 }
1268 m_freem(mrep);
1269 if (vp) {
1270 aftat_ret = VOP_GETATTR(vp, &va, cred, procp);
1271 vput(vp);
1272 }
1273
1274 /*
1275 * Loop around generating replies for all write rpcs that have
1276 * now been completed.
1277 */
1278 swp = nfsd;
1279 do {
1280 if (error) {
1281 nfsm_writereply(NFSX_WCCDATA(v3), v3);
1282 if (v3) {
1283 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1284 }
1285 } else {
1286 nfsm_writereply(NFSX_PREOPATTR(v3) +
1287 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1288 NFSX_WRITEVERF(v3), v3);
1289 if (v3) {
1290 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1291 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1292 *tl++ = txdr_unsigned(nfsd->nd_len);
1293 *tl++ = txdr_unsigned(swp->nd_stable);
1294 /*
1295 * Actually, there is no need to txdr these fields,
1296 * but it may make the values more human readable,
1297 * for debugging purposes.
1298 */
1299 *tl++ = txdr_unsigned(boottime.tv_sec);
1300 *tl = txdr_unsigned(boottime.tv_usec);
1301 } else {
1302 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1303 nfsm_srvfillattr(&va, fp);
1304 }
1305 }
1306 nfsd->nd_mreq = mreq;
1307 if (nfsd->nd_mrep)
1308 panic("nfsrv_write: nd_mrep not free");
1309
1310 /*
1311 * Done. Put it at the head of the timer queue so that
1312 * the final phase can return the reply.
1313 */
1314 s = splsoftclock();
1315 if (nfsd != swp) {
1316 nfsd->nd_time = 0;
1317 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1318 }
1319 nfsd = LIST_FIRST(&swp->nd_coalesce);
1320 if (nfsd) {
1321 LIST_REMOVE(nfsd, nd_tq);
1322 }
1323 splx(s);
1324 } while (nfsd);
1325 s = splsoftclock();
1326 swp->nd_time = 0;
1327 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1328 splx(s);
1329 goto loop1;
1330 }
1331 splx(s);
1332
1333 /*
1334 * Search for a reply to return.
1335 */
1336 s = splsoftclock();
1337 LIST_FOREACH(nfsd, &slp->ns_tq, nd_tq) {
1338 if (nfsd->nd_mreq) {
1339 LIST_REMOVE(nfsd, nd_tq);
1340 *mrq = nfsd->nd_mreq;
1341 *ndp = nfsd;
1342 break;
1343 }
1344 }
1345 splx(s);
1346 return (0);
1347 }
1348
1349 /*
1350 * Coalesce the write request nfsd into owp. To do this we must:
1351 * - remove nfsd from the queues
1352 * - merge nfsd->nd_mrep into owp->nd_mrep
1353 * - update the nd_eoff and nd_stable for owp
1354 * - put nfsd on owp's nd_coalesce list
1355 * NB: Must be called at splsoftclock().
1356 */
1357 void
1358 nfsrvw_coalesce(owp, nfsd)
1359 struct nfsrv_descript *owp;
1360 struct nfsrv_descript *nfsd;
1361 {
1362 int overlap;
1363 struct mbuf *mp;
1364 struct nfsrv_descript *m;
1365
1366 LIST_REMOVE(nfsd, nd_hash);
1367 LIST_REMOVE(nfsd, nd_tq);
1368 if (owp->nd_eoff < nfsd->nd_eoff) {
1369 overlap = owp->nd_eoff - nfsd->nd_off;
1370 if (overlap < 0)
1371 panic("nfsrv_coalesce: bad off");
1372 if (overlap > 0)
1373 m_adj(nfsd->nd_mrep, overlap);
1374 mp = owp->nd_mrep;
1375 while (mp->m_next)
1376 mp = mp->m_next;
1377 mp->m_next = nfsd->nd_mrep;
1378 owp->nd_eoff = nfsd->nd_eoff;
1379 } else
1380 m_freem(nfsd->nd_mrep);
1381 nfsd->nd_mrep = NULL;
1382 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1383 owp->nd_stable = NFSV3WRITE_FILESYNC;
1384 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1385 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1386 owp->nd_stable = NFSV3WRITE_DATASYNC;
1387 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1388 /*
1389 * nfsd might hold coalesce elements! Move them to owp.
1390 * Otherwise, requests may be lost and clients will be stuck.
1391 */
1392 while ((m = LIST_FIRST(&nfsd->nd_coalesce)) != NULL) {
1393 LIST_REMOVE(m, nd_tq);
1394 LIST_INSERT_HEAD(&owp->nd_coalesce, m, nd_tq);
1395 }
1396 }
1397
1398 /*
1399 * nfs create service
1400 * now does a truncate to 0 length via. setattr if it already exists
1401 */
1402 int
1403 nfsrv_create(nfsd, slp, procp, mrq)
1404 struct nfsrv_descript *nfsd;
1405 struct nfssvc_sock *slp;
1406 struct proc *procp;
1407 struct mbuf **mrq;
1408 {
1409 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1410 struct mbuf *nam = nfsd->nd_nam;
1411 caddr_t dpos = nfsd->nd_dpos;
1412 struct ucred *cred = &nfsd->nd_cr;
1413 struct nfs_fattr *fp;
1414 struct vattr va, dirfor, diraft;
1415 struct nfsv2_sattr *sp;
1416 u_int32_t *tl;
1417 struct nameidata nd;
1418 caddr_t cp;
1419 int32_t t1;
1420 caddr_t bpos;
1421 int error = 0, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1422 int rdev = 0;
1423 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1424 char *cp2;
1425 struct mbuf *mb, *mreq;
1426 struct vnode *vp = NULL, *dirp = NULL;
1427 nfsfh_t nfh;
1428 fhandle_t *fhp;
1429 u_quad_t frev, tempsize;
1430 u_char cverf[NFSX_V3CREATEVERF];
1431 struct mount *mp = NULL;
1432
1433 nd.ni_cnd.cn_nameiop = 0;
1434 fhp = &nfh.fh_generic;
1435 nfsm_srvmtofh(fhp);
1436 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
1437 return (ESTALE);
1438 vn_start_write(NULL, &mp, V_WAIT);
1439 nfsm_srvnamesiz(len);
1440 nd.ni_cnd.cn_cred = cred;
1441 nd.ni_cnd.cn_nameiop = CREATE;
1442 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1443 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1444 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1445 if (dirp) {
1446 if (v3)
1447 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1448 procp);
1449 else {
1450 vrele(dirp);
1451 dirp = (struct vnode *)0;
1452 }
1453 }
1454 if (error) {
1455 nfsm_reply(NFSX_WCCDATA(v3));
1456 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1457 if (dirp)
1458 vrele(dirp);
1459 vn_finished_write(mp, 0);
1460 return (0);
1461 }
1462 VATTR_NULL(&va);
1463 if (v3) {
1464 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1465 how = fxdr_unsigned(int, *tl);
1466 switch (how) {
1467 case NFSV3CREATE_GUARDED:
1468 if (nd.ni_vp) {
1469 error = EEXIST;
1470 break;
1471 }
1472 case NFSV3CREATE_UNCHECKED:
1473 nfsm_srvsattr(&va);
1474 break;
1475 case NFSV3CREATE_EXCLUSIVE:
1476 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1477 memcpy(cverf, cp, NFSX_V3CREATEVERF);
1478 exclusive_flag = 1;
1479 if (nd.ni_vp == NULL)
1480 va.va_mode = 0;
1481 break;
1482 };
1483 va.va_type = VREG;
1484 } else {
1485 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1486 va.va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1487 if (va.va_type == VNON)
1488 va.va_type = VREG;
1489 va.va_mode = nfstov_mode(sp->sa_mode);
1490 switch (va.va_type) {
1491 case VREG:
1492 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1493 if (tsize != -1)
1494 va.va_size = (u_quad_t)tsize;
1495 break;
1496 case VCHR:
1497 case VBLK:
1498 case VFIFO:
1499 rdev = fxdr_unsigned(int32_t, sp->sa_size);
1500 break;
1501 default:
1502 break;
1503 };
1504 }
1505
1506 /*
1507 * Iff doesn't exist, create it
1508 * otherwise just truncate to 0 length
1509 * should I set the mode too ??
1510 */
1511 if (nd.ni_vp == NULL) {
1512 if (va.va_type == VREG || va.va_type == VSOCK) {
1513 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1514 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1515 if (!error) {
1516 if (exclusive_flag) {
1517 exclusive_flag = 0;
1518 VATTR_NULL(&va);
1519 memcpy((caddr_t)&va.va_atime, cverf,
1520 NFSX_V3CREATEVERF);
1521 error = VOP_SETATTR(nd.ni_vp, &va, cred,
1522 procp);
1523 }
1524 }
1525 } else if (va.va_type == VCHR || va.va_type == VBLK ||
1526 va.va_type == VFIFO) {
1527 if (va.va_type == VCHR && rdev == 0xffffffff)
1528 va.va_type = VFIFO;
1529 if (va.va_type != VFIFO &&
1530 (error = suser(cred, (u_short *)0))) {
1531 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1532 vput(nd.ni_dvp);
1533 nfsm_reply(0);
1534 vn_finished_write(mp, 0);
1535 return (error);
1536 } else
1537 va.va_rdev = (dev_t)rdev;
1538 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1539 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd,
1540 &va);
1541 if (error) {
1542 nfsm_reply(0);
1543 }
1544 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1545 vrele(nd.ni_dvp);
1546 vput(nd.ni_vp);
1547 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1548 error = EINVAL;
1549 nfsm_reply(0);
1550 }
1551 } else {
1552 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1553 vput(nd.ni_dvp);
1554 error = ENXIO;
1555 }
1556 vp = nd.ni_vp;
1557 } else {
1558 vp = nd.ni_vp;
1559 if (nd.ni_dvp == vp)
1560 vrele(nd.ni_dvp);
1561 else
1562 vput(nd.ni_dvp);
1563 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1564 if (!error && va.va_size != -1) {
1565 error = nfsrv_access(vp, VWRITE, cred,
1566 (nd.ni_cnd.cn_flags & RDONLY), procp, 0);
1567 if (!error) {
1568 nqsrv_getl(vp, ND_WRITE);
1569 tempsize = va.va_size;
1570 VATTR_NULL(&va);
1571 va.va_size = tempsize;
1572 error = VOP_SETATTR(vp, &va, cred,
1573 procp);
1574 }
1575 }
1576 if (error)
1577 vput(vp);
1578 }
1579 if (!error) {
1580 memset((caddr_t)fhp, 0, sizeof(nfh));
1581 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1582 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1583 if (!error)
1584 error = VOP_GETATTR(vp, &va, cred, procp);
1585 vput(vp);
1586 }
1587 if (v3) {
1588 if (exclusive_flag && !error &&
1589 memcmp(cverf, (caddr_t)&va.va_atime, NFSX_V3CREATEVERF))
1590 error = EEXIST;
1591 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1592 vrele(dirp);
1593 }
1594 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1595 if (v3) {
1596 if (!error) {
1597 nfsm_srvpostop_fh(fhp);
1598 nfsm_srvpostop_attr(0, &va);
1599 }
1600 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1601 } else {
1602 nfsm_srvfhtom(fhp, v3);
1603 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1604 nfsm_srvfillattr(&va, fp);
1605 }
1606 vn_finished_write(mp, 0);
1607 return (0);
1608 nfsmout:
1609 if (dirp)
1610 vrele(dirp);
1611 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1612 if (nd.ni_dvp == nd.ni_vp)
1613 vrele(nd.ni_dvp);
1614 else
1615 vput(nd.ni_dvp);
1616 if (nd.ni_vp)
1617 vput(nd.ni_vp);
1618 vn_finished_write(mp, 0);
1619 return (error);
1620 }
1621
1622 /*
1623 * nfs v3 mknod service
1624 */
1625 int
1626 nfsrv_mknod(nfsd, slp, procp, mrq)
1627 struct nfsrv_descript *nfsd;
1628 struct nfssvc_sock *slp;
1629 struct proc *procp;
1630 struct mbuf **mrq;
1631 {
1632 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1633 struct mbuf *nam = nfsd->nd_nam;
1634 caddr_t dpos = nfsd->nd_dpos;
1635 struct ucred *cred = &nfsd->nd_cr;
1636 struct vattr va, dirfor, diraft;
1637 u_int32_t *tl;
1638 struct nameidata nd;
1639 int32_t t1;
1640 caddr_t bpos;
1641 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1642 u_int32_t major, minor;
1643 enum vtype vtyp;
1644 char *cp2;
1645 struct mbuf *mb, *mreq;
1646 struct vnode *vp, *dirp = (struct vnode *)0;
1647 nfsfh_t nfh;
1648 fhandle_t *fhp;
1649 u_quad_t frev;
1650 struct mount *mp = NULL;
1651
1652 nd.ni_cnd.cn_nameiop = 0;
1653 fhp = &nfh.fh_generic;
1654 nfsm_srvmtofh(fhp);
1655 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
1656 return (ESTALE);
1657 vn_start_write(NULL, &mp, V_WAIT);
1658 nfsm_srvnamesiz(len);
1659 nd.ni_cnd.cn_cred = cred;
1660 nd.ni_cnd.cn_nameiop = CREATE;
1661 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1662 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1663 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1664 if (dirp)
1665 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred, procp);
1666 if (error) {
1667 nfsm_reply(NFSX_WCCDATA(1));
1668 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1669 if (dirp)
1670 vrele(dirp);
1671 vn_finished_write(mp, 0);
1672 return (0);
1673 }
1674 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1675 vtyp = nfsv3tov_type(*tl);
1676 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1677 error = NFSERR_BADTYPE;
1678 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1679 if (nd.ni_dvp == nd.ni_vp)
1680 vrele(nd.ni_dvp);
1681 else
1682 vput(nd.ni_dvp);
1683 if (nd.ni_vp)
1684 vput(nd.ni_vp);
1685 goto out;
1686 }
1687 VATTR_NULL(&va);
1688 nfsm_srvsattr(&va);
1689 if (vtyp == VCHR || vtyp == VBLK) {
1690 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1691 major = fxdr_unsigned(u_int32_t, *tl++);
1692 minor = fxdr_unsigned(u_int32_t, *tl);
1693 va.va_rdev = makedev(major, minor);
1694 }
1695
1696 /*
1697 * Iff doesn't exist, create it.
1698 */
1699 if (nd.ni_vp) {
1700 error = EEXIST;
1701 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1702 if (nd.ni_dvp == nd.ni_vp)
1703 vrele(nd.ni_dvp);
1704 else
1705 vput(nd.ni_dvp);
1706 vput(nd.ni_vp);
1707 goto out;
1708 }
1709 va.va_type = vtyp;
1710 if (vtyp == VSOCK) {
1711 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1712 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1713 } else {
1714 if (va.va_type != VFIFO &&
1715 (error = suser(cred, (u_short *)0))) {
1716 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1717 vput(nd.ni_dvp);
1718 goto out;
1719 }
1720 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1721 error = VOP_MKNOD(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
1722 if (error) {
1723 goto out;
1724 }
1725 if (error)
1726 goto out;
1727 if (nd.ni_cnd.cn_flags & ISSYMLINK) {
1728 vput(nd.ni_vp);
1729 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1730 error = EINVAL;
1731 }
1732 }
1733 out:
1734 vp = nd.ni_vp;
1735 if (!error) {
1736 memset((caddr_t)fhp, 0, sizeof(nfh));
1737 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
1738 error = VFS_VPTOFH(vp, &fhp->fh_fid);
1739 if (!error)
1740 error = VOP_GETATTR(vp, &va, cred, procp);
1741 vput(vp);
1742 }
1743 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1744 vrele(dirp);
1745 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
1746 if (!error) {
1747 nfsm_srvpostop_fh(fhp);
1748 nfsm_srvpostop_attr(0, &va);
1749 }
1750 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1751 vn_finished_write(mp, 0);
1752 return (0);
1753 nfsmout:
1754 if (dirp)
1755 vrele(dirp);
1756 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1757 if (nd.ni_dvp == nd.ni_vp)
1758 vrele(nd.ni_dvp);
1759 else
1760 vput(nd.ni_dvp);
1761 if (nd.ni_vp)
1762 vput(nd.ni_vp);
1763 vn_finished_write(mp, 0);
1764 return (error);
1765 }
1766
1767 /*
1768 * nfs remove service
1769 */
1770 int
1771 nfsrv_remove(nfsd, slp, procp, mrq)
1772 struct nfsrv_descript *nfsd;
1773 struct nfssvc_sock *slp;
1774 struct proc *procp;
1775 struct mbuf **mrq;
1776 {
1777 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1778 struct mbuf *nam = nfsd->nd_nam;
1779 caddr_t dpos = nfsd->nd_dpos;
1780 struct ucred *cred = &nfsd->nd_cr;
1781 struct nameidata nd;
1782 u_int32_t *tl;
1783 int32_t t1;
1784 caddr_t bpos;
1785 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1786 int v3 = (nfsd->nd_flag & ND_NFSV3);
1787 char *cp2;
1788 struct mbuf *mb, *mreq;
1789 struct vnode *vp, *dirp;
1790 struct vattr dirfor, diraft;
1791 nfsfh_t nfh;
1792 fhandle_t *fhp;
1793 u_quad_t frev;
1794 struct mount *mp = NULL;
1795
1796 #ifndef nolint
1797 vp = (struct vnode *)0;
1798 #endif
1799 fhp = &nfh.fh_generic;
1800 nfsm_srvmtofh(fhp);
1801 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
1802 return (ESTALE);
1803 vn_start_write(NULL, &mp, V_WAIT);
1804 nfsm_srvnamesiz(len);
1805 nd.ni_cnd.cn_cred = cred;
1806 nd.ni_cnd.cn_nameiop = DELETE;
1807 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
1808 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
1809 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1810 if (dirp) {
1811 if (v3)
1812 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
1813 procp);
1814 else
1815 vrele(dirp);
1816 }
1817 if (!error) {
1818 vp = nd.ni_vp;
1819 if (vp->v_type == VDIR &&
1820 (error = suser(cred, (u_short *)0)) != 0)
1821 goto out;
1822 /*
1823 * The root of a mounted filesystem cannot be deleted.
1824 */
1825 if (vp->v_flag & VROOT) {
1826 error = EBUSY;
1827 goto out;
1828 }
1829 out:
1830 if (!error) {
1831 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1832 nqsrv_getl(vp, ND_WRITE);
1833 error = VOP_REMOVE(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
1834 } else {
1835 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1836 if (nd.ni_dvp == vp)
1837 vrele(nd.ni_dvp);
1838 else
1839 vput(nd.ni_dvp);
1840 vput(vp);
1841 }
1842 }
1843 if (dirp && v3) {
1844 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
1845 vrele(dirp);
1846 }
1847 nfsm_reply(NFSX_WCCDATA(v3));
1848 if (v3) {
1849 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1850 vn_finished_write(mp, 0);
1851 return (0);
1852 }
1853 vn_finished_write(mp, 0);
1854 nfsm_srvdone;
1855 }
1856
1857 /*
1858 * nfs rename service
1859 */
1860 int
1861 nfsrv_rename(nfsd, slp, procp, mrq)
1862 struct nfsrv_descript *nfsd;
1863 struct nfssvc_sock *slp;
1864 struct proc *procp;
1865 struct mbuf **mrq;
1866 {
1867 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1868 struct mbuf *nam = nfsd->nd_nam;
1869 caddr_t dpos = nfsd->nd_dpos;
1870 struct ucred *cred = &nfsd->nd_cr;
1871 u_int32_t *tl;
1872 int32_t t1;
1873 caddr_t bpos;
1874 int error = 0, cache, fdirfor_ret = 1, fdiraft_ret = 1;
1875 uint32_t len, len2;
1876 int tdirfor_ret = 1, tdiraft_ret = 1;
1877 int v3 = (nfsd->nd_flag & ND_NFSV3);
1878 char *cp2;
1879 struct mbuf *mb, *mreq;
1880 struct nameidata fromnd, tond;
1881 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
1882 struct vnode *tdirp = (struct vnode *)0;
1883 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
1884 nfsfh_t fnfh, tnfh;
1885 fhandle_t *ffhp, *tfhp;
1886 u_quad_t frev;
1887 uid_t saved_uid;
1888 struct mount *mp = NULL;
1889
1890 #ifndef nolint
1891 fvp = (struct vnode *)0;
1892 #endif
1893 ffhp = &fnfh.fh_generic;
1894 tfhp = &tnfh.fh_generic;
1895 fromnd.ni_cnd.cn_nameiop = 0;
1896 tond.ni_cnd.cn_nameiop = 0;
1897 nfsm_srvmtofh(ffhp);
1898 if ((mp = vfs_getvfs(&ffhp->fh_fsid)) == NULL)
1899 return (ESTALE);
1900 vn_start_write(NULL, &mp, V_WAIT);
1901 nfsm_srvnamesiz(len);
1902 /*
1903 * Remember our original uid so that we can reset cr_uid before
1904 * the second nfs_namei() call, in case it is remapped.
1905 */
1906 saved_uid = cred->cr_uid;
1907 fromnd.ni_cnd.cn_cred = cred;
1908 fromnd.ni_cnd.cn_nameiop = DELETE;
1909 fromnd.ni_cnd.cn_flags = WANTPARENT | SAVESTART;
1910 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
1911 &dpos, &fdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1912 if (fdirp) {
1913 if (v3)
1914 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, cred,
1915 procp);
1916 else {
1917 vrele(fdirp);
1918 fdirp = (struct vnode *)0;
1919 }
1920 }
1921 if (error) {
1922 nfsm_reply(2 * NFSX_WCCDATA(v3));
1923 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
1924 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
1925 if (fdirp)
1926 vrele(fdirp);
1927 vn_finished_write(mp, 0);
1928 return (0);
1929 }
1930 fvp = fromnd.ni_vp;
1931 nfsm_srvmtofh(tfhp);
1932 if (v3) {
1933 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
1934 len2 = fxdr_unsigned(uint32_t, *tl);
1935 /* len2 will be checked by nfs_namei */
1936 }
1937 else {
1938 /* NFSv2 */
1939 nfsm_strsiz(len2, NFS_MAXNAMLEN);
1940 }
1941 cred->cr_uid = saved_uid;
1942 tond.ni_cnd.cn_cred = cred;
1943 tond.ni_cnd.cn_nameiop = RENAME;
1944 tond.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | NOCACHE | SAVESTART;
1945 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
1946 &dpos, &tdirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
1947 if (tdirp) {
1948 if (v3)
1949 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, cred,
1950 procp);
1951 else {
1952 vrele(tdirp);
1953 tdirp = (struct vnode *)0;
1954 }
1955 }
1956 if (error) {
1957 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
1958 vrele(fromnd.ni_dvp);
1959 vrele(fvp);
1960 goto out1;
1961 }
1962 tdvp = tond.ni_dvp;
1963 tvp = tond.ni_vp;
1964 if (tvp != NULL) {
1965 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1966 if (v3)
1967 error = EEXIST;
1968 else
1969 error = EISDIR;
1970 goto out;
1971 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1972 if (v3)
1973 error = EEXIST;
1974 else
1975 error = ENOTDIR;
1976 goto out;
1977 }
1978 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1979 if (v3)
1980 error = EXDEV;
1981 else
1982 error = ENOTEMPTY;
1983 goto out;
1984 }
1985 }
1986 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1987 if (v3)
1988 error = EXDEV;
1989 else
1990 error = ENOTEMPTY;
1991 goto out;
1992 }
1993 if (fvp->v_mount != tdvp->v_mount) {
1994 if (v3)
1995 error = EXDEV;
1996 else
1997 error = ENOTEMPTY;
1998 goto out;
1999 }
2000 if (fvp == tdvp) {
2001 if (v3)
2002 error = EINVAL;
2003 else
2004 error = ENOTEMPTY;
2005 }
2006 /*
2007 * If source is the same as the destination (that is the
2008 * same vnode with the same name in the same directory),
2009 * then there is nothing to do.
2010 */
2011 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2012 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2013 !memcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2014 fromnd.ni_cnd.cn_namelen))
2015 error = -1;
2016 out:
2017 if (!error) {
2018 nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2019 nqsrv_getl(tdvp, ND_WRITE);
2020 if (tvp) {
2021 nqsrv_getl(tvp, ND_WRITE);
2022 }
2023 error = VOP_RENAME(fromnd.ni_dvp, fromnd.ni_vp, &fromnd.ni_cnd,
2024 tond.ni_dvp, tond.ni_vp, &tond.ni_cnd);
2025 } else {
2026 VOP_ABORTOP(tond.ni_dvp, &tond.ni_cnd);
2027 if (tdvp == tvp)
2028 vrele(tdvp);
2029 else
2030 vput(tdvp);
2031 if (tvp)
2032 vput(tvp);
2033 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2034 vrele(fromnd.ni_dvp);
2035 vrele(fvp);
2036 if (error == -1)
2037 error = 0;
2038 }
2039 vrele(tond.ni_startdir);
2040 PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2041 out1:
2042 if (fdirp) {
2043 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, cred, procp);
2044 vrele(fdirp);
2045 }
2046 if (tdirp) {
2047 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, cred, procp);
2048 vrele(tdirp);
2049 }
2050 vrele(fromnd.ni_startdir);
2051 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2052 nfsm_reply(2 * NFSX_WCCDATA(v3));
2053 if (v3) {
2054 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2055 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2056 }
2057 vn_finished_write(mp, 0);
2058 return (0);
2059
2060 nfsmout:
2061 if (fdirp)
2062 vrele(fdirp);
2063 if (tdirp)
2064 vrele(tdirp);
2065 if (tond.ni_cnd.cn_nameiop) {
2066 vrele(tond.ni_startdir);
2067 PNBUF_PUT(tond.ni_cnd.cn_pnbuf);
2068 }
2069 if (fromnd.ni_cnd.cn_nameiop) {
2070 vrele(fromnd.ni_startdir);
2071 PNBUF_PUT(fromnd.ni_cnd.cn_pnbuf);
2072 VOP_ABORTOP(fromnd.ni_dvp, &fromnd.ni_cnd);
2073 vrele(fromnd.ni_dvp);
2074 vrele(fvp);
2075 }
2076 return (error);
2077 }
2078
2079 /*
2080 * nfs link service
2081 */
2082 int
2083 nfsrv_link(nfsd, slp, procp, mrq)
2084 struct nfsrv_descript *nfsd;
2085 struct nfssvc_sock *slp;
2086 struct proc *procp;
2087 struct mbuf **mrq;
2088 {
2089 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2090 struct mbuf *nam = nfsd->nd_nam;
2091 caddr_t dpos = nfsd->nd_dpos;
2092 struct ucred *cred = &nfsd->nd_cr;
2093 struct nameidata nd;
2094 u_int32_t *tl;
2095 int32_t t1;
2096 caddr_t bpos;
2097 int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
2098 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2099 char *cp2;
2100 struct mbuf *mb, *mreq;
2101 struct vnode *vp, *xp, *dirp = (struct vnode *)0;
2102 struct vattr dirfor, diraft, at;
2103 nfsfh_t nfh, dnfh;
2104 fhandle_t *fhp, *dfhp;
2105 u_quad_t frev;
2106 struct mount *mp = NULL;
2107
2108 fhp = &nfh.fh_generic;
2109 dfhp = &dnfh.fh_generic;
2110 nfsm_srvmtofh(fhp);
2111 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
2112 return (ESTALE);
2113 vn_start_write(NULL, &mp, V_WAIT);
2114 nfsm_srvmtofh(dfhp);
2115 nfsm_srvnamesiz(len);
2116 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
2117 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2118 if (error) {
2119 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2120 nfsm_srvpostop_attr(getret, &at);
2121 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2122 vn_finished_write(mp, 0);
2123 return (0);
2124 }
2125 if (vp->v_type == VDIR && (error = suser(cred, (u_short *)0)) != 0)
2126 goto out1;
2127 nd.ni_cnd.cn_cred = cred;
2128 nd.ni_cnd.cn_nameiop = CREATE;
2129 nd.ni_cnd.cn_flags = LOCKPARENT;
2130 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
2131 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2132 if (dirp) {
2133 if (v3)
2134 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2135 procp);
2136 else {
2137 vrele(dirp);
2138 dirp = (struct vnode *)0;
2139 }
2140 }
2141 if (error)
2142 goto out1;
2143 xp = nd.ni_vp;
2144 if (xp != NULL) {
2145 error = EEXIST;
2146 goto out;
2147 }
2148 xp = nd.ni_dvp;
2149 if (vp->v_mount != xp->v_mount)
2150 error = EXDEV;
2151 out:
2152 if (!error) {
2153 nqsrv_getl(vp, ND_WRITE);
2154 nqsrv_getl(xp, ND_WRITE);
2155 error = VOP_LINK(nd.ni_dvp, vp, &nd.ni_cnd);
2156 } else {
2157 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2158 if (nd.ni_dvp == nd.ni_vp)
2159 vrele(nd.ni_dvp);
2160 else
2161 vput(nd.ni_dvp);
2162 if (nd.ni_vp)
2163 vrele(nd.ni_vp);
2164 }
2165 out1:
2166 if (v3)
2167 getret = VOP_GETATTR(vp, &at, cred, procp);
2168 if (dirp) {
2169 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2170 vrele(dirp);
2171 }
2172 vrele(vp);
2173 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2174 if (v3) {
2175 nfsm_srvpostop_attr(getret, &at);
2176 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2177 vn_finished_write(mp, 0);
2178 return (0);
2179 }
2180 vn_finished_write(mp, 0);
2181 nfsm_srvdone;
2182 }
2183
2184 /*
2185 * nfs symbolic link service
2186 */
2187 int
2188 nfsrv_symlink(nfsd, slp, procp, mrq)
2189 struct nfsrv_descript *nfsd;
2190 struct nfssvc_sock *slp;
2191 struct proc *procp;
2192 struct mbuf **mrq;
2193 {
2194 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2195 struct mbuf *nam = nfsd->nd_nam;
2196 caddr_t dpos = nfsd->nd_dpos;
2197 struct ucred *cred = &nfsd->nd_cr;
2198 struct vattr va, dirfor, diraft;
2199 struct nameidata nd;
2200 u_int32_t *tl;
2201 int32_t t1;
2202 struct nfsv2_sattr *sp;
2203 char *bpos, *pathcp = NULL, *cp2;
2204 struct uio io;
2205 struct iovec iv;
2206 int error = 0, cache, dirfor_ret = 1, diraft_ret = 1;
2207 uint32_t len, len2;
2208 int v3 = (nfsd->nd_flag & ND_NFSV3);
2209 struct mbuf *mb, *mreq;
2210 struct vnode *dirp = (struct vnode *)0;
2211 nfsfh_t nfh;
2212 fhandle_t *fhp;
2213 u_quad_t frev;
2214 struct mount *mp = NULL;
2215
2216 nd.ni_cnd.cn_nameiop = 0;
2217 fhp = &nfh.fh_generic;
2218 nfsm_srvmtofh(fhp);
2219 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
2220 return (ESTALE);
2221 vn_start_write(NULL, &mp, V_WAIT);
2222 nfsm_srvnamesiz(len);
2223 nd.ni_cnd.cn_cred = cred;
2224 nd.ni_cnd.cn_nameiop = CREATE;
2225 nd.ni_cnd.cn_flags = LOCKPARENT;
2226 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2227 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2228 if (dirp) {
2229 if (v3)
2230 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2231 procp);
2232 else {
2233 vrele(dirp);
2234 dirp = (struct vnode *)0;
2235 }
2236 }
2237 if (error)
2238 goto out;
2239 VATTR_NULL(&va);
2240 if (v3) {
2241 nfsm_srvsattr(&va);
2242 nfsm_dissect(tl, uint32_t *, NFSX_UNSIGNED);
2243 len2 = fxdr_unsigned(uint32_t, *tl);
2244 if (len2 > PATH_MAX) {
2245 /* XXX should check _PC_NO_TRUNC */
2246 error = ENAMETOOLONG;
2247 goto abortop;
2248 }
2249 }
2250 else {
2251 /* NFSv2 */
2252 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2253 }
2254 pathcp = malloc(len2 + 1, M_TEMP, M_WAITOK);
2255 iv.iov_base = pathcp;
2256 iv.iov_len = len2;
2257 io.uio_resid = len2;
2258 io.uio_offset = 0;
2259 io.uio_iov = &iv;
2260 io.uio_iovcnt = 1;
2261 io.uio_segflg = UIO_SYSSPACE;
2262 io.uio_rw = UIO_READ;
2263 io.uio_procp = (struct proc *)0;
2264 nfsm_mtouio(&io, len2);
2265 if (!v3) {
2266 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2267 va.va_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
2268 }
2269 *(pathcp + len2) = '\0';
2270 if (nd.ni_vp) {
2271 error = EEXIST;
2272 abortop:
2273 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2274 if (nd.ni_dvp == nd.ni_vp)
2275 vrele(nd.ni_dvp);
2276 else
2277 vput(nd.ni_dvp);
2278 if (nd.ni_vp)
2279 vrele(nd.ni_vp);
2280 goto out;
2281 }
2282 nqsrv_getl(nd.ni_dvp, ND_WRITE);
2283 error = VOP_SYMLINK(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va, pathcp);
2284 if (!error) {
2285 if (v3) {
2286 memset((caddr_t)fhp, 0, sizeof(nfh));
2287 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2288 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2289 if (!error)
2290 error = VOP_GETATTR(nd.ni_vp, &va, cred,
2291 procp);
2292 vput(nd.ni_vp);
2293 } else {
2294 vput(nd.ni_vp);
2295 }
2296 }
2297 out:
2298 if (pathcp)
2299 free(pathcp, M_TEMP);
2300 if (dirp) {
2301 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2302 vrele(dirp);
2303 }
2304 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2305 if (v3) {
2306 if (!error) {
2307 nfsm_srvpostop_fh(fhp);
2308 nfsm_srvpostop_attr(0, &va);
2309 }
2310 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2311 }
2312 vn_finished_write(mp, 0);
2313 return (0);
2314 nfsmout:
2315 if (dirp)
2316 vrele(dirp);
2317 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2318 if (nd.ni_dvp == nd.ni_vp)
2319 vrele(nd.ni_dvp);
2320 else
2321 vput(nd.ni_dvp);
2322 if (nd.ni_vp)
2323 vrele(nd.ni_vp);
2324 if (pathcp)
2325 free(pathcp, M_TEMP);
2326 vn_finished_write(mp, 0);
2327 return (error);
2328 }
2329
2330 /*
2331 * nfs mkdir service
2332 */
2333 int
2334 nfsrv_mkdir(nfsd, slp, procp, mrq)
2335 struct nfsrv_descript *nfsd;
2336 struct nfssvc_sock *slp;
2337 struct proc *procp;
2338 struct mbuf **mrq;
2339 {
2340 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2341 struct mbuf *nam = nfsd->nd_nam;
2342 caddr_t dpos = nfsd->nd_dpos;
2343 struct ucred *cred = &nfsd->nd_cr;
2344 struct vattr va, dirfor, diraft;
2345 struct nfs_fattr *fp;
2346 struct nameidata nd;
2347 caddr_t cp;
2348 u_int32_t *tl;
2349 int32_t t1;
2350 caddr_t bpos;
2351 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2352 int v3 = (nfsd->nd_flag & ND_NFSV3);
2353 char *cp2;
2354 struct mbuf *mb, *mreq;
2355 struct vnode *vp, *dirp = (struct vnode *)0;
2356 nfsfh_t nfh;
2357 fhandle_t *fhp;
2358 u_quad_t frev;
2359 struct mount *mp = NULL;
2360
2361 fhp = &nfh.fh_generic;
2362 nfsm_srvmtofh(fhp);
2363 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
2364 return (ESTALE);
2365 vn_start_write(NULL, &mp, V_WAIT);
2366 nfsm_srvnamesiz(len);
2367 nd.ni_cnd.cn_cred = cred;
2368 nd.ni_cnd.cn_nameiop = CREATE;
2369 nd.ni_cnd.cn_flags = LOCKPARENT;
2370 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2371 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2372 if (dirp) {
2373 if (v3)
2374 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2375 procp);
2376 else {
2377 vrele(dirp);
2378 dirp = (struct vnode *)0;
2379 }
2380 }
2381 if (error) {
2382 nfsm_reply(NFSX_WCCDATA(v3));
2383 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2384 if (dirp)
2385 vrele(dirp);
2386 vn_finished_write(mp, 0);
2387 return (0);
2388 }
2389 VATTR_NULL(&va);
2390 if (v3) {
2391 nfsm_srvsattr(&va);
2392 } else {
2393 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2394 va.va_mode = nfstov_mode(*tl++);
2395 }
2396 va.va_type = VDIR;
2397 vp = nd.ni_vp;
2398 if (vp != NULL) {
2399 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2400 if (nd.ni_dvp == vp)
2401 vrele(nd.ni_dvp);
2402 else
2403 vput(nd.ni_dvp);
2404 vrele(vp);
2405 error = EEXIST;
2406 goto out;
2407 }
2408 nqsrv_getl(nd.ni_dvp, ND_WRITE);
2409 error = VOP_MKDIR(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &va);
2410 if (!error) {
2411 vp = nd.ni_vp;
2412 memset((caddr_t)fhp, 0, sizeof(nfh));
2413 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2414 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2415 if (!error)
2416 error = VOP_GETATTR(vp, &va, cred, procp);
2417 vput(vp);
2418 }
2419 out:
2420 if (dirp) {
2421 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2422 vrele(dirp);
2423 }
2424 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2425 if (v3) {
2426 if (!error) {
2427 nfsm_srvpostop_fh(fhp);
2428 nfsm_srvpostop_attr(0, &va);
2429 }
2430 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2431 } else {
2432 nfsm_srvfhtom(fhp, v3);
2433 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2434 nfsm_srvfillattr(&va, fp);
2435 }
2436 vn_finished_write(mp, 0);
2437 return (0);
2438 nfsmout:
2439 if (dirp)
2440 vrele(dirp);
2441 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2442 if (nd.ni_dvp == nd.ni_vp)
2443 vrele(nd.ni_dvp);
2444 else
2445 vput(nd.ni_dvp);
2446 if (nd.ni_vp)
2447 vrele(nd.ni_vp);
2448 vn_finished_write(mp, 0);
2449 return (error);
2450 }
2451
2452 /*
2453 * nfs rmdir service
2454 */
2455 int
2456 nfsrv_rmdir(nfsd, slp, procp, mrq)
2457 struct nfsrv_descript *nfsd;
2458 struct nfssvc_sock *slp;
2459 struct proc *procp;
2460 struct mbuf **mrq;
2461 {
2462 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2463 struct mbuf *nam = nfsd->nd_nam;
2464 caddr_t dpos = nfsd->nd_dpos;
2465 struct ucred *cred = &nfsd->nd_cr;
2466 u_int32_t *tl;
2467 int32_t t1;
2468 caddr_t bpos;
2469 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2470 int v3 = (nfsd->nd_flag & ND_NFSV3);
2471 char *cp2;
2472 struct mbuf *mb, *mreq;
2473 struct vnode *vp, *dirp = (struct vnode *)0;
2474 struct vattr dirfor, diraft;
2475 nfsfh_t nfh;
2476 fhandle_t *fhp;
2477 struct nameidata nd;
2478 u_quad_t frev;
2479 struct mount *mp = NULL;
2480
2481 fhp = &nfh.fh_generic;
2482 nfsm_srvmtofh(fhp);
2483 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
2484 return (ESTALE);
2485 vn_start_write(NULL, &mp, V_WAIT);
2486 nfsm_srvnamesiz(len);
2487 nd.ni_cnd.cn_cred = cred;
2488 nd.ni_cnd.cn_nameiop = DELETE;
2489 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF;
2490 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
2491 &dirp, procp, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2492 if (dirp) {
2493 if (v3)
2494 dirfor_ret = VOP_GETATTR(dirp, &dirfor, cred,
2495 procp);
2496 else {
2497 vrele(dirp);
2498 dirp = (struct vnode *)0;
2499 }
2500 }
2501 if (error) {
2502 nfsm_reply(NFSX_WCCDATA(v3));
2503 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2504 if (dirp)
2505 vrele(dirp);
2506 vn_finished_write(mp, 0);
2507 return (0);
2508 }
2509 vp = nd.ni_vp;
2510 if (vp->v_type != VDIR) {
2511 error = ENOTDIR;
2512 goto out;
2513 }
2514 /*
2515 * No rmdir "." please.
2516 */
2517 if (nd.ni_dvp == vp) {
2518 error = EINVAL;
2519 goto out;
2520 }
2521 /*
2522 * The root of a mounted filesystem cannot be deleted.
2523 */
2524 if (vp->v_flag & VROOT)
2525 error = EBUSY;
2526 out:
2527 if (!error) {
2528 nqsrv_getl(nd.ni_dvp, ND_WRITE);
2529 nqsrv_getl(vp, ND_WRITE);
2530 error = VOP_RMDIR(nd.ni_dvp, nd.ni_vp, &nd.ni_cnd);
2531 } else {
2532 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
2533 if (nd.ni_dvp == nd.ni_vp)
2534 vrele(nd.ni_dvp);
2535 else
2536 vput(nd.ni_dvp);
2537 vput(vp);
2538 }
2539 if (dirp) {
2540 diraft_ret = VOP_GETATTR(dirp, &diraft, cred, procp);
2541 vrele(dirp);
2542 }
2543 nfsm_reply(NFSX_WCCDATA(v3));
2544 if (v3) {
2545 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2546 vn_finished_write(mp, 0);
2547 return (0);
2548 }
2549 vn_finished_write(mp, 0);
2550 nfsm_srvdone;
2551 }
2552
2553 /*
2554 * nfs readdir service
2555 * - mallocs what it thinks is enough to read
2556 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2557 * - calls VOP_READDIR()
2558 * - loops around building the reply
2559 * if the output generated exceeds count break out of loop
2560 * The nfsm_clget macro is used here so that the reply will be packed
2561 * tightly in mbuf clusters.
2562 * - it only knows that it has encountered eof when the VOP_READDIR()
2563 * reads nothing
2564 * - as such one readdir rpc will return eof false although you are there
2565 * and then the next will return eof
2566 * - it trims out records with d_fileno == 0
2567 * this doesn't matter for Unix clients, but they might confuse clients
2568 * for other os'.
2569 * - it trims out records with d_type == DT_WHT
2570 * these cannot be seen through NFS (unless we extend the protocol)
2571 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2572 * than requested, but this may not apply to all filesystems. For
2573 * example, client NFS does not { although it is never remote mounted
2574 * anyhow }
2575 * The alternate call nfsrv_readdirplus() does lookups as well.
2576 * PS: The NFS protocol spec. does not clarify what the "count" byte
2577 * argument is a count of.. just name strings and file id's or the
2578 * entire reply rpc or ...
2579 * I tried just file name and id sizes and it confused the Sun client,
2580 * so I am using the full rpc size now. The "paranoia.." comment refers
2581 * to including the status longwords that are not a part of the dir.
2582 * "entry" structures, but are in the rpc.
2583 */
2584 struct flrep {
2585 nfsuint64 fl_off;
2586 u_int32_t fl_postopok;
2587 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2588 u_int32_t fl_fhok;
2589 u_int32_t fl_fhsize;
2590 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2591 };
2592
2593 int
2594 nfsrv_readdir(nfsd, slp, procp, mrq)
2595 struct nfsrv_descript *nfsd;
2596 struct nfssvc_sock *slp;
2597 struct proc *procp;
2598 struct mbuf **mrq;
2599 {
2600 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2601 struct mbuf *nam = nfsd->nd_nam;
2602 caddr_t dpos = nfsd->nd_dpos;
2603 struct ucred *cred = &nfsd->nd_cr;
2604 char *bp, *be;
2605 struct mbuf *mp;
2606 struct dirent *dp;
2607 caddr_t cp;
2608 u_int32_t *tl;
2609 int32_t t1;
2610 caddr_t bpos;
2611 struct mbuf *mb, *mreq, *mp2;
2612 char *cpos, *cend, *cp2, *rbuf;
2613 struct vnode *vp;
2614 struct vattr at;
2615 nfsfh_t nfh;
2616 fhandle_t *fhp;
2617 struct uio io;
2618 struct iovec iv;
2619 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2620 int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2621 int v3 = (nfsd->nd_flag & ND_NFSV3);
2622 u_quad_t frev, off, toff, verf;
2623 off_t *cookies = NULL, *cookiep;
2624 nfsuint64 jar;
2625
2626 fhp = &nfh.fh_generic;
2627 nfsm_srvmtofh(fhp);
2628 if (v3) {
2629 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2630 toff = fxdr_hyper(tl);
2631 tl += 2;
2632 verf = fxdr_hyper(tl);
2633 tl += 2;
2634 } else {
2635 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2636 toff = fxdr_unsigned(u_quad_t, *tl++);
2637 }
2638 off = toff;
2639 cnt = fxdr_unsigned(int, *tl);
2640 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2641 xfer = NFS_SRVMAXDATA(nfsd);
2642 if (siz > xfer)
2643 siz = xfer;
2644 fullsiz = siz;
2645 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2646 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2647 if (!error && vp->v_type != VDIR) {
2648 error = ENOTDIR;
2649 vput(vp);
2650 }
2651 if (error) {
2652 nfsm_reply(NFSX_UNSIGNED);
2653 nfsm_srvpostop_attr(getret, &at);
2654 return (0);
2655 }
2656 nqsrv_getl(vp, ND_READ);
2657 if (v3) {
2658 error = getret = VOP_GETATTR(vp, &at, cred, procp);
2659 #ifdef NFS3_STRICTVERF
2660 /*
2661 * XXX This check is too strict for Solaris 2.5 clients.
2662 */
2663 if (!error && toff && verf != at.va_filerev)
2664 error = NFSERR_BAD_COOKIE;
2665 #endif
2666 }
2667 if (!error)
2668 error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2669 if (error) {
2670 vput(vp);
2671 nfsm_reply(NFSX_POSTOPATTR(v3));
2672 nfsm_srvpostop_attr(getret, &at);
2673 return (0);
2674 }
2675 VOP_UNLOCK(vp, 0);
2676 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2677 again:
2678 iv.iov_base = rbuf;
2679 iv.iov_len = fullsiz;
2680 io.uio_iov = &iv;
2681 io.uio_iovcnt = 1;
2682 io.uio_offset = (off_t)off;
2683 io.uio_resid = fullsiz;
2684 io.uio_segflg = UIO_SYSSPACE;
2685 io.uio_rw = UIO_READ;
2686 io.uio_procp = (struct proc *)0;
2687 eofflag = 0;
2688 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2689
2690 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2691
2692 off = (off_t)io.uio_offset;
2693 if (!cookies && !error)
2694 error = NFSERR_PERM;
2695 if (v3) {
2696 getret = VOP_GETATTR(vp, &at, cred, procp);
2697 if (!error)
2698 error = getret;
2699 }
2700
2701 VOP_UNLOCK(vp, 0);
2702 if (error) {
2703 vrele(vp);
2704 free((caddr_t)rbuf, M_TEMP);
2705 if (cookies)
2706 free((caddr_t)cookies, M_TEMP);
2707 nfsm_reply(NFSX_POSTOPATTR(v3));
2708 nfsm_srvpostop_attr(getret, &at);
2709 return (0);
2710 }
2711 if (io.uio_resid) {
2712 siz -= io.uio_resid;
2713
2714 /*
2715 * If nothing read, return eof
2716 * rpc reply
2717 */
2718 if (siz == 0) {
2719 vrele(vp);
2720 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
2721 2 * NFSX_UNSIGNED);
2722 if (v3) {
2723 nfsm_srvpostop_attr(getret, &at);
2724 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2725 txdr_hyper(at.va_filerev, tl);
2726 tl += 2;
2727 } else
2728 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2729 *tl++ = nfs_false;
2730 *tl = nfs_true;
2731 free((caddr_t)rbuf, M_TEMP);
2732 free((caddr_t)cookies, M_TEMP);
2733 return (0);
2734 }
2735 }
2736
2737 /*
2738 * Check for degenerate cases of nothing useful read.
2739 * If so go try again
2740 */
2741 cpos = rbuf;
2742 cend = rbuf + siz;
2743 dp = (struct dirent *)cpos;
2744 cookiep = cookies;
2745
2746 while (cpos < cend && ncookies > 0 &&
2747 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
2748 cpos += dp->d_reclen;
2749 dp = (struct dirent *)cpos;
2750 cookiep++;
2751 ncookies--;
2752 }
2753 if (cpos >= cend || ncookies == 0) {
2754 toff = off;
2755 siz = fullsiz;
2756 goto again;
2757 }
2758
2759 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
2760 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
2761 if (v3) {
2762 nfsm_srvpostop_attr(getret, &at);
2763 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2764 txdr_hyper(at.va_filerev, tl);
2765 }
2766 mp = mp2 = mb;
2767 bp = bpos;
2768 be = bp + M_TRAILINGSPACE(mp);
2769
2770 /* Loop through the records and build reply */
2771 while (cpos < cend && ncookies > 0) {
2772 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
2773 nlen = dp->d_namlen;
2774 rem = nfsm_rndup(nlen)-nlen;
2775 len += (4 * NFSX_UNSIGNED + nlen + rem);
2776 if (v3)
2777 len += 2 * NFSX_UNSIGNED;
2778 if (len > cnt) {
2779 eofflag = 0;
2780 break;
2781 }
2782 /*
2783 * Build the directory record xdr from
2784 * the dirent entry.
2785 */
2786 nfsm_clget;
2787 *tl = nfs_true;
2788 bp += NFSX_UNSIGNED;
2789 if (v3) {
2790 nfsm_clget;
2791 *tl = 0;
2792 bp += NFSX_UNSIGNED;
2793 }
2794 nfsm_clget;
2795 *tl = txdr_unsigned(dp->d_fileno);
2796 bp += NFSX_UNSIGNED;
2797 nfsm_clget;
2798 *tl = txdr_unsigned(nlen);
2799 bp += NFSX_UNSIGNED;
2800
2801 /* And loop around copying the name */
2802 xfer = nlen;
2803 cp = dp->d_name;
2804 while (xfer > 0) {
2805 nfsm_clget;
2806 if ((bp+xfer) > be)
2807 tsiz = be-bp;
2808 else
2809 tsiz = xfer;
2810 memcpy(bp, cp, tsiz);
2811 bp += tsiz;
2812 xfer -= tsiz;
2813 if (xfer > 0)
2814 cp += tsiz;
2815 }
2816 /* And null pad to an int32_t boundary */
2817 for (i = 0; i < rem; i++)
2818 *bp++ = '\0';
2819 nfsm_clget;
2820
2821 /* Finish off the record */
2822 txdr_hyper(*cookiep, &jar);
2823 if (v3) {
2824 *tl = jar.nfsuquad[0];
2825 bp += NFSX_UNSIGNED;
2826 nfsm_clget;
2827 }
2828 *tl = jar.nfsuquad[1];
2829 bp += NFSX_UNSIGNED;
2830 }
2831 cpos += dp->d_reclen;
2832 dp = (struct dirent *)cpos;
2833 cookiep++;
2834 ncookies--;
2835 }
2836 vrele(vp);
2837 nfsm_clget;
2838 *tl = nfs_false;
2839 bp += NFSX_UNSIGNED;
2840 nfsm_clget;
2841 if (eofflag)
2842 *tl = nfs_true;
2843 else
2844 *tl = nfs_false;
2845 bp += NFSX_UNSIGNED;
2846 if (mp != mb) {
2847 if (bp < be)
2848 mp->m_len = bp - mtod(mp, caddr_t);
2849 } else
2850 mp->m_len += bp - bpos;
2851 free((caddr_t)rbuf, M_TEMP);
2852 free((caddr_t)cookies, M_TEMP);
2853 nfsm_srvdone;
2854 }
2855
2856 int
2857 nfsrv_readdirplus(nfsd, slp, procp, mrq)
2858 struct nfsrv_descript *nfsd;
2859 struct nfssvc_sock *slp;
2860 struct proc *procp;
2861 struct mbuf **mrq;
2862 {
2863 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2864 struct mbuf *nam = nfsd->nd_nam;
2865 caddr_t dpos = nfsd->nd_dpos;
2866 struct ucred *cred = &nfsd->nd_cr;
2867 char *bp, *be;
2868 struct mbuf *mp;
2869 struct dirent *dp;
2870 caddr_t cp;
2871 u_int32_t *tl;
2872 int32_t t1;
2873 caddr_t bpos;
2874 struct mbuf *mb, *mreq, *mp2;
2875 char *cpos, *cend, *cp2, *rbuf;
2876 struct vnode *vp, *nvp;
2877 struct flrep fl;
2878 nfsfh_t nfh;
2879 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
2880 struct uio io;
2881 struct iovec iv;
2882 struct vattr va, at, *vap = &va;
2883 struct nfs_fattr *fp;
2884 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2885 int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
2886 u_quad_t frev, off, toff, verf;
2887 off_t *cookies = NULL, *cookiep;
2888
2889 fhp = &nfh.fh_generic;
2890 nfsm_srvmtofh(fhp);
2891 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2892 toff = fxdr_hyper(tl);
2893 tl += 2;
2894 verf = fxdr_hyper(tl);
2895 tl += 2;
2896 siz = fxdr_unsigned(int, *tl++);
2897 cnt = fxdr_unsigned(int, *tl);
2898 off = toff;
2899 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2900 xfer = NFS_SRVMAXDATA(nfsd);
2901 if (siz > xfer)
2902 siz = xfer;
2903 fullsiz = siz;
2904 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2905 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
2906 if (!error && vp->v_type != VDIR) {
2907 error = ENOTDIR;
2908 vput(vp);
2909 }
2910 if (error) {
2911 nfsm_reply(NFSX_UNSIGNED);
2912 nfsm_srvpostop_attr(getret, &at);
2913 return (0);
2914 }
2915 error = getret = VOP_GETATTR(vp, &at, cred, procp);
2916 #ifdef NFS3_STRICTVERF
2917 /*
2918 * XXX This check is too strict for Solaris 2.5 clients.
2919 */
2920 if (!error && toff && verf != at.va_filerev)
2921 error = NFSERR_BAD_COOKIE;
2922 #endif
2923 if (!error) {
2924 nqsrv_getl(vp, ND_READ);
2925 error = nfsrv_access(vp, VEXEC, cred, rdonly, procp, 0);
2926 }
2927 if (error) {
2928 vput(vp);
2929 nfsm_reply(NFSX_V3POSTOPATTR);
2930 nfsm_srvpostop_attr(getret, &at);
2931 return (0);
2932 }
2933 VOP_UNLOCK(vp, 0);
2934
2935 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2936 again:
2937 iv.iov_base = rbuf;
2938 iv.iov_len = fullsiz;
2939 io.uio_iov = &iv;
2940 io.uio_iovcnt = 1;
2941 io.uio_offset = (off_t)off;
2942 io.uio_resid = fullsiz;
2943 io.uio_segflg = UIO_SYSSPACE;
2944 io.uio_rw = UIO_READ;
2945 io.uio_procp = (struct proc *)0;
2946 eofflag = 0;
2947
2948 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2949
2950 error = VOP_READDIR(vp, &io, cred, &eofflag, &cookies, &ncookies);
2951
2952 off = (u_quad_t)io.uio_offset;
2953 getret = VOP_GETATTR(vp, &at, cred, procp);
2954
2955 VOP_UNLOCK(vp, 0);
2956
2957 /*
2958 * If the VGET operation doesn't work for this filesystem,
2959 * we can't support readdirplus. Returning NOTSUPP should
2960 * make clients fall back to plain readdir.
2961 * There's no need to check for VPTOFH as well, we wouldn't
2962 * even be here otherwise.
2963 */
2964 if (!getret) {
2965 if ((getret = VFS_VGET(vp->v_mount, at.va_fileid, &nvp)))
2966 getret = (getret == EOPNOTSUPP) ?
2967 NFSERR_NOTSUPP : NFSERR_IO;
2968 else
2969 vput(nvp);
2970 }
2971
2972 if (!cookies && !error)
2973 error = NFSERR_PERM;
2974 if (!error)
2975 error = getret;
2976 if (error) {
2977 vrele(vp);
2978 if (cookies)
2979 free((caddr_t)cookies, M_TEMP);
2980 free((caddr_t)rbuf, M_TEMP);
2981 nfsm_reply(NFSX_V3POSTOPATTR);
2982 nfsm_srvpostop_attr(getret, &at);
2983 return (0);
2984 }
2985 if (io.uio_resid) {
2986 siz -= io.uio_resid;
2987
2988 /*
2989 * If nothing read, return eof
2990 * rpc reply
2991 */
2992 if (siz == 0) {
2993 vrele(vp);
2994 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
2995 2 * NFSX_UNSIGNED);
2996 nfsm_srvpostop_attr(getret, &at);
2997 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2998 txdr_hyper(at.va_filerev, tl);
2999 tl += 2;
3000 *tl++ = nfs_false;
3001 *tl = nfs_true;
3002 free((caddr_t)cookies, M_TEMP);
3003 free((caddr_t)rbuf, M_TEMP);
3004 return (0);
3005 }
3006 }
3007
3008 /*
3009 * Check for degenerate cases of nothing useful read.
3010 * If so go try again
3011 */
3012 cpos = rbuf;
3013 cend = rbuf + siz;
3014 dp = (struct dirent *)cpos;
3015 cookiep = cookies;
3016
3017 while (cpos < cend && ncookies > 0 &&
3018 (dp->d_fileno == 0 || dp->d_type == DT_WHT)) {
3019 cpos += dp->d_reclen;
3020 dp = (struct dirent *)cpos;
3021 cookiep++;
3022 ncookies--;
3023 }
3024 if (cpos >= cend || ncookies == 0) {
3025 toff = off;
3026 siz = fullsiz;
3027 goto again;
3028 }
3029
3030 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3031 nfsm_reply(cnt);
3032 nfsm_srvpostop_attr(getret, &at);
3033 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3034 txdr_hyper(at.va_filerev, tl);
3035 mp = mp2 = mb;
3036 bp = bpos;
3037 be = bp + M_TRAILINGSPACE(mp);
3038
3039 /* Loop through the records and build reply */
3040 while (cpos < cend && ncookies > 0) {
3041 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3042 nlen = dp->d_namlen;
3043 rem = nfsm_rndup(nlen)-nlen;
3044
3045 /*
3046 * For readdir_and_lookup get the vnode using
3047 * the file number.
3048 */
3049 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3050 goto invalid;
3051 memset((caddr_t)nfhp, 0, NFSX_V3FH);
3052 nfhp->fh_fsid =
3053 nvp->v_mount->mnt_stat.f_fsid;
3054 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3055 vput(nvp);
3056 goto invalid;
3057 }
3058 if (VOP_GETATTR(nvp, vap, cred, procp)) {
3059 vput(nvp);
3060 goto invalid;
3061 }
3062 vput(nvp);
3063
3064 /*
3065 * If either the dircount or maxcount will be
3066 * exceeded, get out now. Both of these lengths
3067 * are calculated conservatively, including all
3068 * XDR overheads.
3069 */
3070 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3071 NFSX_V3POSTOPATTR);
3072 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3073 if (len > cnt || dirlen > fullsiz) {
3074 eofflag = 0;
3075 break;
3076 }
3077
3078 /*
3079 * Build the directory record xdr from
3080 * the dirent entry.
3081 */
3082 fp = (struct nfs_fattr *)&fl.fl_fattr;
3083 nfsm_srvfillattr(vap, fp);
3084 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3085 fl.fl_fhok = nfs_true;
3086 fl.fl_postopok = nfs_true;
3087 txdr_hyper(*cookiep, fl.fl_off.nfsuquad);
3088
3089 nfsm_clget;
3090 *tl = nfs_true;
3091 bp += NFSX_UNSIGNED;
3092 nfsm_clget;
3093 *tl = 0;
3094 bp += NFSX_UNSIGNED;
3095 nfsm_clget;
3096 *tl = txdr_unsigned(dp->d_fileno);
3097 bp += NFSX_UNSIGNED;
3098 nfsm_clget;
3099 *tl = txdr_unsigned(nlen);
3100 bp += NFSX_UNSIGNED;
3101
3102 /* And loop around copying the name */
3103 xfer = nlen;
3104 cp = dp->d_name;
3105 while (xfer > 0) {
3106 nfsm_clget;
3107 if ((bp + xfer) > be)
3108 tsiz = be - bp;
3109 else
3110 tsiz = xfer;
3111 memcpy(bp, cp, tsiz);
3112 bp += tsiz;
3113 xfer -= tsiz;
3114 if (xfer > 0)
3115 cp += tsiz;
3116 }
3117 /* And null pad to an int32_t boundary */
3118 for (i = 0; i < rem; i++)
3119 *bp++ = '\0';
3120
3121 /*
3122 * Now copy the flrep structure out.
3123 */
3124 xfer = sizeof (struct flrep);
3125 cp = (caddr_t)&fl;
3126 while (xfer > 0) {
3127 nfsm_clget;
3128 if ((bp + xfer) > be)
3129 tsiz = be - bp;
3130 else
3131 tsiz = xfer;
3132 memcpy(bp, cp, tsiz);
3133 bp += tsiz;
3134 xfer -= tsiz;
3135 if (xfer > 0)
3136 cp += tsiz;
3137 }
3138 }
3139 invalid:
3140 cpos += dp->d_reclen;
3141 dp = (struct dirent *)cpos;
3142 cookiep++;
3143 ncookies--;
3144 }
3145 vrele(vp);
3146 nfsm_clget;
3147 *tl = nfs_false;
3148 bp += NFSX_UNSIGNED;
3149 nfsm_clget;
3150 if (eofflag)
3151 *tl = nfs_true;
3152 else
3153 *tl = nfs_false;
3154 bp += NFSX_UNSIGNED;
3155 if (mp != mb) {
3156 if (bp < be)
3157 mp->m_len = bp - mtod(mp, caddr_t);
3158 } else
3159 mp->m_len += bp - bpos;
3160 free((caddr_t)cookies, M_TEMP);
3161 free((caddr_t)rbuf, M_TEMP);
3162 nfsm_srvdone;
3163 }
3164
3165 /*
3166 * nfs commit service
3167 */
3168 int
3169 nfsrv_commit(nfsd, slp, procp, mrq)
3170 struct nfsrv_descript *nfsd;
3171 struct nfssvc_sock *slp;
3172 struct proc *procp;
3173 struct mbuf **mrq;
3174 {
3175 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3176 struct mbuf *nam = nfsd->nd_nam;
3177 caddr_t dpos = nfsd->nd_dpos;
3178 struct ucred *cred = &nfsd->nd_cr;
3179 struct vattr bfor, aft;
3180 struct vnode *vp;
3181 nfsfh_t nfh;
3182 fhandle_t *fhp;
3183 u_int32_t *tl;
3184 int32_t t1;
3185 caddr_t bpos;
3186 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
3187 char *cp2;
3188 struct mbuf *mb, *mreq;
3189 u_quad_t frev, off, end;
3190 struct mount *mp = NULL;
3191
3192 #ifndef nolint
3193 cache = 0;
3194 #endif
3195 fhp = &nfh.fh_generic;
3196 nfsm_srvmtofh(fhp);
3197 if ((mp = vfs_getvfs(&fhp->fh_fsid)) == NULL)
3198 return (ESTALE);
3199 vn_start_write(NULL, &mp, V_WAIT);
3200 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3201
3202 off = fxdr_hyper(tl);
3203 tl += 2;
3204 cnt = fxdr_unsigned(int, *tl);
3205 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3206 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3207 if (error) {
3208 nfsm_reply(2 * NFSX_UNSIGNED);
3209 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3210 vn_finished_write(mp, 0);
3211 return (0);
3212 }
3213 for_ret = VOP_GETATTR(vp, &bfor, cred, procp);
3214 end = (cnt > 0) ? off + cnt : vp->v_size;
3215 if (end < off || end > vp->v_size)
3216 end = vp->v_size;
3217 if (off < vp->v_size)
3218 error = VOP_FSYNC(vp, cred, FSYNC_WAIT, off, end, procp);
3219 /* else error == 0, from nfsrv_fhtovp() */
3220 aft_ret = VOP_GETATTR(vp, &aft, cred, procp);
3221 vput(vp);
3222 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3223 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3224 if (!error) {
3225 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3226 *tl++ = txdr_unsigned(boottime.tv_sec);
3227 *tl = txdr_unsigned(boottime.tv_usec);
3228 } else {
3229 vn_finished_write(mp, 0);
3230 return (0);
3231 }
3232 vn_finished_write(mp, 0);
3233 nfsm_srvdone;
3234 }
3235
3236 /*
3237 * nfs statfs service
3238 */
3239 int
3240 nfsrv_statfs(nfsd, slp, procp, mrq)
3241 struct nfsrv_descript *nfsd;
3242 struct nfssvc_sock *slp;
3243 struct proc *procp;
3244 struct mbuf **mrq;
3245 {
3246 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3247 struct mbuf *nam = nfsd->nd_nam;
3248 caddr_t dpos = nfsd->nd_dpos;
3249 struct ucred *cred = &nfsd->nd_cr;
3250 struct statfs *sf;
3251 struct nfs_statfs *sfp;
3252 u_int32_t *tl;
3253 int32_t t1;
3254 caddr_t bpos;
3255 int error = 0, rdonly, cache, getret = 1;
3256 int v3 = (nfsd->nd_flag & ND_NFSV3);
3257 char *cp2;
3258 struct mbuf *mb, *mreq;
3259 struct vnode *vp;
3260 struct vattr at;
3261 nfsfh_t nfh;
3262 fhandle_t *fhp;
3263 struct statfs statfs;
3264 u_quad_t frev, tval;
3265
3266 #ifndef nolint
3267 cache = 0;
3268 #endif
3269 fhp = &nfh.fh_generic;
3270 nfsm_srvmtofh(fhp);
3271 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3272 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3273 if (error) {
3274 nfsm_reply(NFSX_UNSIGNED);
3275 nfsm_srvpostop_attr(getret, &at);
3276 return (0);
3277 }
3278 sf = &statfs;
3279 error = VFS_STATFS(vp->v_mount, sf, procp);
3280 getret = VOP_GETATTR(vp, &at, cred, procp);
3281 vput(vp);
3282 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3283 if (v3)
3284 nfsm_srvpostop_attr(getret, &at);
3285 if (error)
3286 return (0);
3287 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3288 if (v3) {
3289 tval = (u_quad_t)((quad_t)sf->f_blocks * (quad_t)sf->f_bsize);
3290 txdr_hyper(tval, &sfp->sf_tbytes);
3291 tval = (u_quad_t)((quad_t)sf->f_bfree * (quad_t)sf->f_bsize);
3292 txdr_hyper(tval, &sfp->sf_fbytes);
3293 tval = (u_quad_t)((quad_t)sf->f_bavail * (quad_t)sf->f_bsize);
3294 txdr_hyper(tval, &sfp->sf_abytes);
3295 tval = (u_quad_t)sf->f_files;
3296 txdr_hyper(tval, &sfp->sf_tfiles);
3297 tval = (u_quad_t)sf->f_ffree;
3298 txdr_hyper(tval, &sfp->sf_ffiles);
3299 txdr_hyper(tval, &sfp->sf_afiles);
3300 sfp->sf_invarsec = 0;
3301 } else {
3302 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3303 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3304 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3305 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3306 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3307 }
3308 nfsm_srvdone;
3309 }
3310
3311 /*
3312 * nfs fsinfo service
3313 */
3314 int
3315 nfsrv_fsinfo(nfsd, slp, procp, mrq)
3316 struct nfsrv_descript *nfsd;
3317 struct nfssvc_sock *slp;
3318 struct proc *procp;
3319 struct mbuf **mrq;
3320 {
3321 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3322 struct mbuf *nam = nfsd->nd_nam;
3323 caddr_t dpos = nfsd->nd_dpos;
3324 struct ucred *cred = &nfsd->nd_cr;
3325 u_int32_t *tl;
3326 struct nfsv3_fsinfo *sip;
3327 int32_t t1;
3328 caddr_t bpos;
3329 int error = 0, rdonly, cache, getret = 1;
3330 uint32_t maxdata;
3331 char *cp2;
3332 struct mbuf *mb, *mreq;
3333 struct vnode *vp;
3334 struct vattr at;
3335 nfsfh_t nfh;
3336 fhandle_t *fhp;
3337 u_quad_t frev, maxfsize;
3338 struct statfs sb;
3339
3340 #ifndef nolint
3341 cache = 0;
3342 #endif
3343 fhp = &nfh.fh_generic;
3344 nfsm_srvmtofh(fhp);
3345 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3346 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3347 if (error) {
3348 nfsm_reply(NFSX_UNSIGNED);
3349 nfsm_srvpostop_attr(getret, &at);
3350 return (0);
3351 }
3352
3353 /* XXX Try to make a guess on the max file size. */
3354 VFS_STATFS(vp->v_mount, &sb, (struct proc *)0);
3355 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3356
3357 getret = VOP_GETATTR(vp, &at, cred, procp);
3358 vput(vp);
3359 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3360 nfsm_srvpostop_attr(getret, &at);
3361 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3362
3363 /*
3364 * XXX
3365 * There should be file system VFS OP(s) to get this information.
3366 * For now, assume ufs.
3367 */
3368 if (slp->ns_so->so_type == SOCK_DGRAM)
3369 maxdata = NFS_MAXDGRAMDATA;
3370 else
3371 maxdata = NFS_MAXDATA;
3372 sip->fs_rtmax = txdr_unsigned(maxdata);
3373 sip->fs_rtpref = txdr_unsigned(maxdata);
3374 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3375 sip->fs_wtmax = txdr_unsigned(maxdata);
3376 sip->fs_wtpref = txdr_unsigned(maxdata);
3377 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3378 sip->fs_dtpref = txdr_unsigned(maxdata);
3379 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3380 sip->fs_timedelta.nfsv3_sec = 0;
3381 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3382 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3383 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3384 NFSV3FSINFO_CANSETTIME);
3385 nfsm_srvdone;
3386 }
3387
3388 /*
3389 * nfs pathconf service
3390 */
3391 int
3392 nfsrv_pathconf(nfsd, slp, procp, mrq)
3393 struct nfsrv_descript *nfsd;
3394 struct nfssvc_sock *slp;
3395 struct proc *procp;
3396 struct mbuf **mrq;
3397 {
3398 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3399 struct mbuf *nam = nfsd->nd_nam;
3400 caddr_t dpos = nfsd->nd_dpos;
3401 struct ucred *cred = &nfsd->nd_cr;
3402 u_int32_t *tl;
3403 struct nfsv3_pathconf *pc;
3404 int32_t t1;
3405 caddr_t bpos;
3406 int error = 0, rdonly, cache, getret = 1;
3407 register_t linkmax, namemax, chownres, notrunc;
3408 char *cp2;
3409 struct mbuf *mb, *mreq;
3410 struct vnode *vp;
3411 struct vattr at;
3412 nfsfh_t nfh;
3413 fhandle_t *fhp;
3414 u_quad_t frev;
3415
3416 #ifndef nolint
3417 cache = 0;
3418 #endif
3419 fhp = &nfh.fh_generic;
3420 nfsm_srvmtofh(fhp);
3421 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3422 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
3423 if (error) {
3424 nfsm_reply(NFSX_UNSIGNED);
3425 nfsm_srvpostop_attr(getret, &at);
3426 return (0);
3427 }
3428 error = VOP_PATHCONF(vp, _PC_LINK_MAX, &linkmax);
3429 if (!error)
3430 error = VOP_PATHCONF(vp, _PC_NAME_MAX, &namemax);
3431 if (!error)
3432 error = VOP_PATHCONF(vp, _PC_CHOWN_RESTRICTED, &chownres);
3433 if (!error)
3434 error = VOP_PATHCONF(vp, _PC_NO_TRUNC, ¬runc);
3435 getret = VOP_GETATTR(vp, &at, cred, procp);
3436 vput(vp);
3437 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3PATHCONF);
3438 nfsm_srvpostop_attr(getret, &at);
3439 if (error)
3440 return (0);
3441 nfsm_build(pc, struct nfsv3_pathconf *, NFSX_V3PATHCONF);
3442
3443 pc->pc_linkmax = txdr_unsigned(linkmax);
3444 pc->pc_namemax = txdr_unsigned(namemax);
3445 pc->pc_notrunc = txdr_unsigned(notrunc);
3446 pc->pc_chownrestricted = txdr_unsigned(chownres);
3447
3448 /*
3449 * These should probably be supported by VOP_PATHCONF(), but
3450 * until msdosfs is exportable (why would you want to?), the
3451 * Unix defaults should be ok.
3452 */
3453 pc->pc_caseinsensitive = nfs_false;
3454 pc->pc_casepreserving = nfs_true;
3455 nfsm_srvdone;
3456 }
3457
3458 /*
3459 * Null operation, used by clients to ping server
3460 */
3461 /* ARGSUSED */
3462 int
3463 nfsrv_null(nfsd, slp, procp, mrq)
3464 struct nfsrv_descript *nfsd;
3465 struct nfssvc_sock *slp;
3466 struct proc *procp;
3467 struct mbuf **mrq;
3468 {
3469 struct mbuf *mrep = nfsd->nd_mrep;
3470 caddr_t bpos;
3471 int error = NFSERR_RETVOID, cache = 0;
3472 struct mbuf *mb, *mreq;
3473 u_quad_t frev;
3474
3475 nfsm_reply(0);
3476 return (0);
3477 }
3478
3479 /*
3480 * No operation, used for obsolete procedures
3481 */
3482 /* ARGSUSED */
3483 int
3484 nfsrv_noop(nfsd, slp, procp, mrq)
3485 struct nfsrv_descript *nfsd;
3486 struct nfssvc_sock *slp;
3487 struct proc *procp;
3488 struct mbuf **mrq;
3489 {
3490 struct mbuf *mrep = nfsd->nd_mrep;
3491 caddr_t bpos;
3492 int error, cache = 0;
3493 struct mbuf *mb, *mreq;
3494 u_quad_t frev;
3495
3496 if (nfsd->nd_repstat)
3497 error = nfsd->nd_repstat;
3498 else
3499 error = EPROCUNAVAIL;
3500 nfsm_reply(0);
3501 return (0);
3502 }
3503
3504 /*
3505 * Perform access checking for vnodes obtained from file handles that would
3506 * refer to files already opened by a Unix client. You cannot just use
3507 * vn_writechk() and VOP_ACCESS() for two reasons.
3508 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write case
3509 * 2 - The owner is to be given access irrespective of mode bits for some
3510 * operations, so that processes that chmod after opening a file don't
3511 * break. I don't like this because it opens a security hole, but since
3512 * the nfs server opens a security hole the size of a barn door anyhow,
3513 * what the heck.
3514 *
3515 * The exception to rule 2 is EPERM. If a file is IMMUTABLE, VOP_ACCESS()
3516 * will return EPERM instead of EACCESS. EPERM is always an error.
3517 */
3518 int
3519 nfsrv_access(vp, flags, cred, rdonly, p, override)
3520 struct vnode *vp;
3521 int flags;
3522 struct ucred *cred;
3523 int rdonly;
3524 struct proc *p;
3525 {
3526 struct vattr vattr;
3527 int error;
3528 if (flags & VWRITE) {
3529 /* Just vn_writechk() changed to check rdonly */
3530 /*
3531 * Disallow write attempts on read-only file systems;
3532 * unless the file is a socket or a block or character
3533 * device resident on the file system.
3534 */
3535 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
3536 switch (vp->v_type) {
3537 case VREG:
3538 case VDIR:
3539 case VLNK:
3540 return (EROFS);
3541 default:
3542 break;
3543 }
3544 }
3545
3546 /*
3547 * If the vnode is in use as a process's text,
3548 * we can't allow writing.
3549 */
3550 if (vp->v_flag & VTEXT)
3551 return (ETXTBSY);
3552 }
3553 error = VOP_GETATTR(vp, &vattr, cred, p);
3554 if (error)
3555 return (error);
3556 error = VOP_ACCESS(vp, flags, cred, p);
3557 /*
3558 * Allow certain operations for the owner (reads and writes
3559 * on files that are already open).
3560 */
3561 if (override && error == EACCES && cred->cr_uid == vattr.va_uid)
3562 error = 0;
3563 return error;
3564 }
3565