This commit represents a major revamping of the clock interrupt and timebase
[dragonfly.git] / sys / vfs / nfs / nfs_serv.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * @(#)nfs_serv.c 8.8 (Berkeley) 7/31/95
37 * $FreeBSD: src/sys/nfs/nfs_serv.c,v 1.93.2.6 2002/12/29 18:19:53 dillon Exp $
bc0c094e 38 * $DragonFly: src/sys/vfs/nfs/nfs_serv.c,v 1.11 2003/10/09 22:27:26 dillon Exp $
984263bc
MD
39 */
40
41/*
42 * nfs version 2 and 3 server calls to vnode ops
43 * - these routines generally have 3 phases
44 * 1 - break down and validate rpc request in mbuf list
45 * 2 - do the vnode ops for the request
46 * (surprisingly ?? many are very similar to syscalls in vfs_syscalls.c)
47 * 3 - build the rpc reply in an mbuf list
48 * nb:
49 * - do not mix the phases, since the nfsm_?? macros can return failures
50 * on a bad rpc or similar and do not do any vrele() or vput()'s
51 *
52 * - the nfsm_reply() macro generates an nfs rpc reply with the nfs
53 * error number iff error != 0 whereas
54 * returning an error from the server function implies a fatal error
55 * such as a badly constructed rpc request that should be dropped without
56 * a reply.
57 * For Version 3, nfsm_reply() does not return for the error case, since
58 * most version 3 rpcs return more than the status for error cases.
59 *
60 * Other notes:
61 * Warning: always pay careful attention to resource cleanup on return
62 * and note that nfsm_*() macros can terminate a procedure on certain
63 * errors.
64 *
65 * lookup() and namei()
66 * may return garbage in various structural fields/return elements
67 * if an error is returned, and may garbage up nd.ni_dvp even if no
68 * error is returned and you did not request LOCKPARENT or WANTPARENT.
69 *
70 * We use the ni_cnd.cn_flags 'HASBUF' flag to track whether the name
71 * buffer has been freed or not.
72 */
73
74#include <sys/param.h>
75#include <sys/systm.h>
76#include <sys/proc.h>
77#include <sys/namei.h>
78#include <sys/unistd.h>
79#include <sys/vnode.h>
80#include <sys/mount.h>
81#include <sys/socket.h>
82#include <sys/socketvar.h>
83#include <sys/malloc.h>
84#include <sys/mbuf.h>
85#include <sys/dirent.h>
86#include <sys/stat.h>
87#include <sys/kernel.h>
88#include <sys/sysctl.h>
89#include <sys/buf.h>
90
91#include <vm/vm.h>
92#include <vm/vm_extern.h>
93#include <vm/vm_zone.h>
94#include <vm/vm_object.h>
95
3020e3be
MD
96#include <sys/buf2.h>
97
1f2de5d4
MD
98#include "nfsproto.h"
99#include "rpcv2.h"
100#include "nfs.h"
101#include "xdr_subs.h"
102#include "nfsm_subs.h"
103#include "nqnfs.h"
984263bc
MD
104
105#ifdef NFSRV_DEBUG
106#define nfsdbprintf(info) printf info
107#else
108#define nfsdbprintf(info)
109#endif
110
111#define MAX_COMMIT_COUNT (1024 * 1024)
112
113#define NUM_HEURISTIC 64
114#define NHUSE_INIT 64
115#define NHUSE_INC 16
116#define NHUSE_MAX 2048
117
118static struct nfsheur {
119 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
120 off_t nh_nextr; /* next offset for sequential detection */
121 int nh_use; /* use count for selection */
122 int nh_seqcount; /* heuristic */
123} nfsheur[NUM_HEURISTIC];
124
125nfstype nfsv3_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFSOCK,
126 NFFIFO, NFNON };
127#ifndef NFS_NOSERVER
128nfstype nfsv2_type[9] = { NFNON, NFREG, NFDIR, NFBLK, NFCHR, NFLNK, NFNON,
129 NFCHR, NFNON };
130/* Global vars */
131extern u_int32_t nfs_xdrneg1;
132extern u_int32_t nfs_false, nfs_true;
133extern enum vtype nv3tov_type[8];
134extern struct nfsstats nfsstats;
135
136int nfsrvw_procrastinate = NFS_GATHERDELAY * 1000;
137int nfsrvw_procrastinate_v3 = 0;
138
139static struct timeval nfsver = { 0 };
140
141SYSCTL_DECL(_vfs_nfs);
142
143static int nfs_async;
144SYSCTL_INT(_vfs_nfs, OID_AUTO, async, CTLFLAG_RW, &nfs_async, 0, "");
145static int nfs_commit_blks;
146static int nfs_commit_miss;
147SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks, 0, "");
148SYSCTL_INT(_vfs_nfs, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss, 0, "");
149
a6ee311a
RG
150static int nfsrv_access (struct vnode *,int,struct ucred *,int,
151 struct thread *, int);
152static void nfsrvw_coalesce (struct nfsrv_descript *,
153 struct nfsrv_descript *);
984263bc
MD
154
155/*
156 * Clear nameidata fields that are tested in nsfmout cleanup code prior
157 * to using first nfsm macro (that might jump to the cleanup code).
158 */
159
160static __inline
161void
162ndclear(struct nameidata *nd)
163{
164 nd->ni_cnd.cn_flags = 0;
165 nd->ni_vp = NULL;
166 nd->ni_dvp = NULL;
167 nd->ni_startdir = NULL;
168}
169
170/*
171 * nfs v3 access service
172 */
173int
dadab5e9
MD
174nfsrv3_access(struct nfsrv_descript *nfsd, struct nfssvc_sock *slp,
175 struct thread *td, struct mbuf **mrq)
984263bc
MD
176{
177 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
178 struct sockaddr *nam = nfsd->nd_nam;
179 caddr_t dpos = nfsd->nd_dpos;
180 struct ucred *cred = &nfsd->nd_cr;
181 struct vnode *vp = NULL;
182 nfsfh_t nfh;
183 fhandle_t *fhp;
dadab5e9
MD
184 u_int32_t *tl;
185 int32_t t1;
984263bc
MD
186 caddr_t bpos;
187 int error = 0, rdonly, cache, getret;
188 char *cp2;
189 struct mbuf *mb, *mreq, *mb2;
190 struct vattr vattr, *vap = &vattr;
191 u_long testmode, nfsmode;
192 u_quad_t frev;
193
194 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
195#ifndef nolint
196 cache = 0;
197#endif
198 fhp = &nfh.fh_generic;
199 nfsm_srvmtofh(fhp);
200 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
201 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
202 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
203 if (error) {
204 nfsm_reply(NFSX_UNSIGNED);
205 nfsm_srvpostop_attr(1, (struct vattr *)0);
206 error = 0;
207 goto nfsmout;
208 }
209 nfsmode = fxdr_unsigned(u_int32_t, *tl);
210 if ((nfsmode & NFSV3ACCESS_READ) &&
dadab5e9 211 nfsrv_access(vp, VREAD, cred, rdonly, td, 0))
984263bc
MD
212 nfsmode &= ~NFSV3ACCESS_READ;
213 if (vp->v_type == VDIR)
214 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND |
215 NFSV3ACCESS_DELETE);
216 else
217 testmode = (NFSV3ACCESS_MODIFY | NFSV3ACCESS_EXTEND);
218 if ((nfsmode & testmode) &&
dadab5e9 219 nfsrv_access(vp, VWRITE, cred, rdonly, td, 0))
984263bc
MD
220 nfsmode &= ~testmode;
221 if (vp->v_type == VDIR)
222 testmode = NFSV3ACCESS_LOOKUP;
223 else
224 testmode = NFSV3ACCESS_EXECUTE;
225 if ((nfsmode & testmode) &&
dadab5e9 226 nfsrv_access(vp, VEXEC, cred, rdonly, td, 0))
984263bc 227 nfsmode &= ~testmode;
3b568787 228 getret = VOP_GETATTR(vp, vap, td);
984263bc
MD
229 vput(vp);
230 vp = NULL;
231 nfsm_reply(NFSX_POSTOPATTR(1) + NFSX_UNSIGNED);
232 nfsm_srvpostop_attr(getret, vap);
233 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
234 *tl = txdr_unsigned(nfsmode);
235nfsmout:
236 if (vp)
237 vput(vp);
238 return(error);
239}
240
241/*
242 * nfs getattr service
243 */
244int
dadab5e9 245nfsrv_getattr(nfsd, slp, td, mrq)
984263bc
MD
246 struct nfsrv_descript *nfsd;
247 struct nfssvc_sock *slp;
dadab5e9 248 struct thread *td;
984263bc
MD
249 struct mbuf **mrq;
250{
251 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
252 struct sockaddr *nam = nfsd->nd_nam;
253 caddr_t dpos = nfsd->nd_dpos;
254 struct ucred *cred = &nfsd->nd_cr;
dadab5e9 255 struct nfs_fattr *fp;
984263bc 256 struct vattr va;
dadab5e9 257 struct vattr *vap = &va;
984263bc
MD
258 struct vnode *vp = NULL;
259 nfsfh_t nfh;
260 fhandle_t *fhp;
dadab5e9
MD
261 u_int32_t *tl;
262 int32_t t1;
984263bc
MD
263 caddr_t bpos;
264 int error = 0, rdonly, cache;
265 char *cp2;
266 struct mbuf *mb, *mb2, *mreq;
267 u_quad_t frev;
268
269 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
270 fhp = &nfh.fh_generic;
271 nfsm_srvmtofh(fhp);
272 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
273 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
274 if (error) {
275 nfsm_reply(0);
276 error = 0;
277 goto nfsmout;
278 }
279 nqsrv_getl(vp, ND_READ);
3b568787 280 error = VOP_GETATTR(vp, vap, td);
984263bc
MD
281 vput(vp);
282 vp = NULL;
283 nfsm_reply(NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
284 if (error) {
285 error = 0;
286 goto nfsmout;
287 }
288 nfsm_build(fp, struct nfs_fattr *, NFSX_FATTR(nfsd->nd_flag & ND_NFSV3));
289 nfsm_srvfillattr(vap, fp);
290 /* fall through */
291
292nfsmout:
293 if (vp)
294 vput(vp);
295 return(error);
296}
297
298/*
299 * nfs setattr service
300 */
301int
dadab5e9 302nfsrv_setattr(nfsd, slp, td, mrq)
984263bc
MD
303 struct nfsrv_descript *nfsd;
304 struct nfssvc_sock *slp;
dadab5e9 305 struct thread *td;
984263bc
MD
306 struct mbuf **mrq;
307{
308 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
309 struct sockaddr *nam = nfsd->nd_nam;
310 caddr_t dpos = nfsd->nd_dpos;
311 struct ucred *cred = &nfsd->nd_cr;
312 struct vattr va, preat;
dadab5e9
MD
313 struct vattr *vap = &va;
314 struct nfsv2_sattr *sp;
315 struct nfs_fattr *fp;
984263bc
MD
316 struct vnode *vp = NULL;
317 nfsfh_t nfh;
318 fhandle_t *fhp;
dadab5e9
MD
319 u_int32_t *tl;
320 int32_t t1;
984263bc
MD
321 caddr_t bpos;
322 int error = 0, rdonly, cache, preat_ret = 1, postat_ret = 1;
323 int v3 = (nfsd->nd_flag & ND_NFSV3), gcheck = 0;
324 char *cp2;
325 struct mbuf *mb, *mb2, *mreq;
326 u_quad_t frev;
327 struct timespec guard;
328
329 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
330 fhp = &nfh.fh_generic;
331 nfsm_srvmtofh(fhp);
332 VATTR_NULL(vap);
333 if (v3) {
334 nfsm_srvsattr(vap);
335 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
336 gcheck = fxdr_unsigned(int, *tl);
337 if (gcheck) {
338 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
339 fxdr_nfsv3time(tl, &guard);
340 }
341 } else {
342 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
343 /*
344 * Nah nah nah nah na nah
345 * There is a bug in the Sun client that puts 0xffff in the mode
346 * field of sattr when it should put in 0xffffffff. The u_short
347 * doesn't sign extend.
348 * --> check the low order 2 bytes for 0xffff
349 */
350 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
351 vap->va_mode = nfstov_mode(sp->sa_mode);
352 if (sp->sa_uid != nfs_xdrneg1)
353 vap->va_uid = fxdr_unsigned(uid_t, sp->sa_uid);
354 if (sp->sa_gid != nfs_xdrneg1)
355 vap->va_gid = fxdr_unsigned(gid_t, sp->sa_gid);
356 if (sp->sa_size != nfs_xdrneg1)
357 vap->va_size = fxdr_unsigned(u_quad_t, sp->sa_size);
358 if (sp->sa_atime.nfsv2_sec != nfs_xdrneg1) {
359#ifdef notyet
360 fxdr_nfsv2time(&sp->sa_atime, &vap->va_atime);
361#else
362 vap->va_atime.tv_sec =
363 fxdr_unsigned(int32_t, sp->sa_atime.nfsv2_sec);
364 vap->va_atime.tv_nsec = 0;
365#endif
366 }
367 if (sp->sa_mtime.nfsv2_sec != nfs_xdrneg1)
368 fxdr_nfsv2time(&sp->sa_mtime, &vap->va_mtime);
369
370 }
371
372 /*
373 * Now that we have all the fields, lets do it.
374 */
375 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam, &rdonly,
376 (nfsd->nd_flag & ND_KERBAUTH), TRUE);
377 if (error) {
378 nfsm_reply(2 * NFSX_UNSIGNED);
379 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
380 error = 0;
381 goto nfsmout;
382 }
383
384 /*
385 * vp now an active resource, pay careful attention to cleanup
386 */
387
388 nqsrv_getl(vp, ND_WRITE);
389 if (v3) {
3b568787 390 error = preat_ret = VOP_GETATTR(vp, &preat, td);
984263bc
MD
391 if (!error && gcheck &&
392 (preat.va_ctime.tv_sec != guard.tv_sec ||
393 preat.va_ctime.tv_nsec != guard.tv_nsec))
394 error = NFSERR_NOT_SYNC;
395 if (error) {
396 vput(vp);
397 vp = NULL;
398 nfsm_reply(NFSX_WCCDATA(v3));
399 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
400 error = 0;
401 goto nfsmout;
402 }
403 }
404
405 /*
406 * If the size is being changed write acces is required, otherwise
407 * just check for a read only file system.
408 */
409 if (vap->va_size == ((u_quad_t)((quad_t) -1))) {
410 if (rdonly || (vp->v_mount->mnt_flag & MNT_RDONLY)) {
411 error = EROFS;
412 goto out;
413 }
414 } else {
415 if (vp->v_type == VDIR) {
416 error = EISDIR;
417 goto out;
418 } else if ((error = nfsrv_access(vp, VWRITE, cred, rdonly,
dadab5e9 419 td, 0)) != 0){
984263bc 420 goto out;
dadab5e9 421 }
984263bc 422 }
dadab5e9 423 error = VOP_SETATTR(vp, vap, cred, td);
3b568787 424 postat_ret = VOP_GETATTR(vp, vap, td);
984263bc
MD
425 if (!error)
426 error = postat_ret;
427out:
428 vput(vp);
429 vp = NULL;
430 nfsm_reply(NFSX_WCCORFATTR(v3));
431 if (v3) {
432 nfsm_srvwcc_data(preat_ret, &preat, postat_ret, vap);
433 error = 0;
434 goto nfsmout;
435 } else {
436 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
437 nfsm_srvfillattr(vap, fp);
438 }
439 /* fall through */
440
441nfsmout:
442 if (vp)
443 vput(vp);
444 return(error);
445}
446
447/*
448 * nfs lookup rpc
449 */
450int
dadab5e9 451nfsrv_lookup(nfsd, slp, td, mrq)
984263bc
MD
452 struct nfsrv_descript *nfsd;
453 struct nfssvc_sock *slp;
dadab5e9 454 struct thread *td;
984263bc
MD
455 struct mbuf **mrq;
456{
457 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
458 struct sockaddr *nam = nfsd->nd_nam;
459 caddr_t dpos = nfsd->nd_dpos;
460 struct ucred *cred = &nfsd->nd_cr;
dadab5e9 461 struct nfs_fattr *fp;
984263bc
MD
462 struct nameidata nd, ind, *ndp = &nd;
463 struct vnode *vp, *dirp = NULL;
464 nfsfh_t nfh;
465 fhandle_t *fhp;
dadab5e9
MD
466 caddr_t cp;
467 u_int32_t *tl;
468 int32_t t1;
984263bc
MD
469 caddr_t bpos;
470 int error = 0, cache, len, dirattr_ret = 1;
471 int v3 = (nfsd->nd_flag & ND_NFSV3), pubflag;
472 char *cp2;
473 struct mbuf *mb, *mb2, *mreq;
474 struct vattr va, dirattr, *vap = &va;
475 u_quad_t frev;
476
477 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
478 ndclear(&nd);
479
480 fhp = &nfh.fh_generic;
481 nfsm_srvmtofh(fhp);
482 nfsm_srvnamesiz(len);
483
484 pubflag = nfs_ispublicfh(fhp);
485
486 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
487 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
488 nd.ni_cnd.cn_flags = CNP_LOCKLEAF | CNP_SAVESTART;
984263bc 489 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 490 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), pubflag);
984263bc
MD
491
492 /*
493 * namei failure, only dirp to cleanup. Clear out garbarge from
494 * structure in case macros jump to nfsmout.
495 */
496
497 if (error) {
498 if (dirp) {
499 if (v3)
3b568787 500 dirattr_ret = VOP_GETATTR(dirp, &dirattr, td);
984263bc
MD
501 vrele(dirp);
502 dirp = NULL;
503 }
504 nfsm_reply(NFSX_POSTOPATTR(v3));
505 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
506 error = 0;
507 goto nfsmout;
508 }
509
510 /*
511 * Locate index file for public filehandle
512 *
513 * error is 0 on entry and 0 on exit from this block.
514 */
515
516 if (pubflag) {
517 if (nd.ni_vp->v_type == VDIR && nfs_pub.np_index != NULL) {
518 /*
519 * Setup call to lookup() to see if we can find
520 * the index file. Arguably, this doesn't belong
521 * in a kernel.. Ugh. If an error occurs, do not
522 * try to install an index file and then clear the
523 * error.
524 *
525 * When we replace nd with ind and redirect ndp,
526 * maintenance of ni_startdir and ni_vp shift to
527 * ind and we have to clean them up in the old nd.
528 * However, the cnd resource continues to be maintained
529 * via the original nd. Confused? You aren't alone!
530 */
531 ind = nd;
dadab5e9 532 VOP_UNLOCK(nd.ni_vp, 0, td);
984263bc
MD
533 ind.ni_pathlen = strlen(nfs_pub.np_index);
534 ind.ni_cnd.cn_nameptr = ind.ni_cnd.cn_pnbuf =
535 nfs_pub.np_index;
536 ind.ni_startdir = nd.ni_vp;
537 VREF(ind.ni_startdir);
538
539 error = lookup(&ind);
540 ind.ni_dvp = NULL;
541
542 if (error == 0) {
543 /*
544 * Found an index file. Get rid of
545 * the old references. transfer nd.ni_vp'
546 */
547 if (dirp)
548 vrele(dirp);
549 dirp = nd.ni_vp;
550 nd.ni_vp = NULL;
551 vrele(nd.ni_startdir);
552 nd.ni_startdir = NULL;
553 ndp = &ind;
554 }
555 error = 0;
556 }
557 /*
558 * If the public filehandle was used, check that this lookup
559 * didn't result in a filehandle outside the publicly exported
560 * filesystem. We clear the poor vp here to avoid lockups due
561 * to NFS I/O.
562 */
563
564 if (ndp->ni_vp->v_mount != nfs_pub.np_mount) {
565 vput(nd.ni_vp);
566 nd.ni_vp = NULL;
567 error = EPERM;
568 }
569 }
570
571 if (dirp) {
572 if (v3)
3b568787 573 dirattr_ret = VOP_GETATTR(dirp, &dirattr, td);
984263bc
MD
574 vrele(dirp);
575 dirp = NULL;
576 }
577
578 /*
579 * Resources at this point:
580 * ndp->ni_vp may not be NULL
581 *
582 */
583
584 if (error) {
585 nfsm_reply(NFSX_POSTOPATTR(v3));
586 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
587 error = 0;
588 goto nfsmout;
589 }
590
591 nqsrv_getl(ndp->ni_startdir, ND_READ);
592
593 /*
594 * Clear out some resources prior to potentially blocking. This
595 * is not as critical as ni_dvp resources in other routines, but
596 * it helps.
597 */
598 vrele(ndp->ni_startdir);
599 ndp->ni_startdir = NULL;
600 NDFREE(&nd, NDF_ONLY_PNBUF);
601
602 /*
603 * Get underlying attribute, then release remaining resources ( for
604 * the same potential blocking reason ) and reply.
605 */
606 vp = ndp->ni_vp;
607 bzero((caddr_t)fhp, sizeof(nfh));
608 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
609 error = VFS_VPTOFH(vp, &fhp->fh_fid);
610 if (!error)
3b568787 611 error = VOP_GETATTR(vp, vap, td);
984263bc
MD
612
613 vput(vp);
614 ndp->ni_vp = NULL;
615 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPORFATTR(v3) + NFSX_POSTOPATTR(v3));
616 if (error) {
617 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
618 error = 0;
619 goto nfsmout;
620 }
621 nfsm_srvfhtom(fhp, v3);
622 if (v3) {
623 nfsm_srvpostop_attr(0, vap);
624 nfsm_srvpostop_attr(dirattr_ret, &dirattr);
625 } else {
626 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
627 nfsm_srvfillattr(vap, fp);
628 }
629
630nfsmout:
631 if (dirp)
632 vrele(dirp);
633 NDFREE(&nd, NDF_ONLY_PNBUF);
634 if (ndp->ni_startdir)
635 vrele(ndp->ni_startdir);
636 if (ndp->ni_vp)
637 vput(ndp->ni_vp);
638 return (error);
639}
640
641/*
642 * nfs readlink service
643 */
644int
dadab5e9 645nfsrv_readlink(nfsd, slp, td, mrq)
984263bc
MD
646 struct nfsrv_descript *nfsd;
647 struct nfssvc_sock *slp;
dadab5e9 648 struct thread *td;
984263bc
MD
649 struct mbuf **mrq;
650{
651 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
652 struct sockaddr *nam = nfsd->nd_nam;
653 caddr_t dpos = nfsd->nd_dpos;
654 struct ucred *cred = &nfsd->nd_cr;
655 struct iovec iv[(NFS_MAXPATHLEN+MLEN-1)/MLEN];
dadab5e9
MD
656 struct iovec *ivp = iv;
657 struct mbuf *mp;
658 u_int32_t *tl;
659 int32_t t1;
984263bc
MD
660 caddr_t bpos;
661 int error = 0, rdonly, cache, i, tlen, len, getret;
662 int v3 = (nfsd->nd_flag & ND_NFSV3);
663 char *cp2;
664 struct mbuf *mb, *mb2, *mp2, *mp3, *mreq;
665 struct vnode *vp = NULL;
666 struct vattr attr;
667 nfsfh_t nfh;
668 fhandle_t *fhp;
669 struct uio io, *uiop = &io;
670 u_quad_t frev;
671
672 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
673#ifndef nolint
674 mp2 = (struct mbuf *)0;
675#endif
676 mp3 = NULL;
677 fhp = &nfh.fh_generic;
678 nfsm_srvmtofh(fhp);
679 len = 0;
680 i = 0;
681 while (len < NFS_MAXPATHLEN) {
682 MGET(mp, M_WAIT, MT_DATA);
683 MCLGET(mp, M_WAIT);
684 mp->m_len = NFSMSIZ(mp);
685 if (len == 0)
686 mp3 = mp2 = mp;
687 else {
688 mp2->m_next = mp;
689 mp2 = mp;
690 }
691 if ((len+mp->m_len) > NFS_MAXPATHLEN) {
692 mp->m_len = NFS_MAXPATHLEN-len;
693 len = NFS_MAXPATHLEN;
694 } else
695 len += mp->m_len;
696 ivp->iov_base = mtod(mp, caddr_t);
697 ivp->iov_len = mp->m_len;
698 i++;
699 ivp++;
700 }
701 uiop->uio_iov = iv;
702 uiop->uio_iovcnt = i;
703 uiop->uio_offset = 0;
704 uiop->uio_resid = len;
705 uiop->uio_rw = UIO_READ;
706 uiop->uio_segflg = UIO_SYSSPACE;
dadab5e9 707 uiop->uio_td = NULL;
984263bc
MD
708 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
709 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
710 if (error) {
711 nfsm_reply(2 * NFSX_UNSIGNED);
712 nfsm_srvpostop_attr(1, (struct vattr *)0);
713 error = 0;
714 goto nfsmout;
715 }
716 if (vp->v_type != VLNK) {
717 if (v3)
718 error = EINVAL;
719 else
720 error = ENXIO;
721 goto out;
722 }
723 nqsrv_getl(vp, ND_READ);
724 error = VOP_READLINK(vp, uiop, cred);
725out:
3b568787 726 getret = VOP_GETATTR(vp, &attr, td);
984263bc
MD
727 vput(vp);
728 vp = NULL;
729 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_UNSIGNED);
730 if (v3) {
731 nfsm_srvpostop_attr(getret, &attr);
732 if (error) {
733 error = 0;
734 goto nfsmout;
735 }
736 }
737 if (uiop->uio_resid > 0) {
738 len -= uiop->uio_resid;
739 tlen = nfsm_rndup(len);
740 nfsm_adj(mp3, NFS_MAXPATHLEN-tlen, tlen-len);
741 }
742 nfsm_build(tl, u_int32_t *, NFSX_UNSIGNED);
743 *tl = txdr_unsigned(len);
744 mb->m_next = mp3;
745 mp3 = NULL;
746nfsmout:
747 if (mp3)
748 m_freem(mp3);
749 if (vp)
750 vput(vp);
751 return(error);
752}
753
754/*
755 * nfs read service
756 */
757int
dadab5e9 758nfsrv_read(nfsd, slp, td, mrq)
984263bc
MD
759 struct nfsrv_descript *nfsd;
760 struct nfssvc_sock *slp;
dadab5e9 761 struct thread *td;
984263bc
MD
762 struct mbuf **mrq;
763{
764 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
765 struct sockaddr *nam = nfsd->nd_nam;
766 caddr_t dpos = nfsd->nd_dpos;
767 struct ucred *cred = &nfsd->nd_cr;
dadab5e9 768 struct iovec *iv;
984263bc 769 struct iovec *iv2;
dadab5e9
MD
770 struct mbuf *m;
771 struct nfs_fattr *fp;
772 u_int32_t *tl;
773 int32_t t1;
774 int i;
984263bc
MD
775 caddr_t bpos;
776 int error = 0, rdonly, cache, cnt, len, left, siz, tlen, getret;
777 int v3 = (nfsd->nd_flag & ND_NFSV3), reqlen;
778 char *cp2;
779 struct mbuf *mb, *mb2, *mreq;
780 struct mbuf *m2;
781 struct vnode *vp = NULL;
782 nfsfh_t nfh;
783 fhandle_t *fhp;
784 struct uio io, *uiop = &io;
785 struct vattr va, *vap = &va;
786 struct nfsheur *nh;
787 off_t off;
788 int ioflag = 0;
789 u_quad_t frev;
790
791 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
792 fhp = &nfh.fh_generic;
793 nfsm_srvmtofh(fhp);
794 if (v3) {
795 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
796 off = fxdr_hyper(tl);
797 } else {
798 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
799 off = (off_t)fxdr_unsigned(u_int32_t, *tl);
800 }
801 nfsm_srvstrsiz(reqlen, NFS_SRVMAXDATA(nfsd));
802
803 /*
804 * Reference vp. If an error occurs, vp will be invalid, but we
805 * have to NULL it just in case. The macros might goto nfsmout
806 * as well.
807 */
808
809 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
810 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
811 if (error) {
812 vp = NULL;
813 nfsm_reply(2 * NFSX_UNSIGNED);
814 nfsm_srvpostop_attr(1, (struct vattr *)0);
815 error = 0;
816 goto nfsmout;
817 }
818
819 if (vp->v_type != VREG) {
820 if (v3)
821 error = EINVAL;
822 else
823 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
824 }
825 if (!error) {
826 nqsrv_getl(vp, ND_READ);
dadab5e9
MD
827 if ((error = nfsrv_access(vp, VREAD, cred, rdonly, td, 1)) != 0)
828 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 1);
984263bc 829 }
3b568787 830 getret = VOP_GETATTR(vp, vap, td);
984263bc
MD
831 if (!error)
832 error = getret;
833 if (error) {
834 vput(vp);
835 vp = NULL;
836 nfsm_reply(NFSX_POSTOPATTR(v3));
837 nfsm_srvpostop_attr(getret, vap);
838 error = 0;
839 goto nfsmout;
840 }
841
842 /*
843 * Calculate byte count to read
844 */
845
846 if (off >= vap->va_size)
847 cnt = 0;
848 else if ((off + reqlen) > vap->va_size)
849 cnt = vap->va_size - off;
850 else
851 cnt = reqlen;
852
853 /*
854 * Calculate seqcount for heuristic
855 */
856
857 {
858 int hi;
859 int try = 4;
860
861 /*
862 * Locate best candidate
863 */
864
865 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) & (NUM_HEURISTIC - 1);
866 nh = &nfsheur[hi];
867
868 while (try--) {
869 if (nfsheur[hi].nh_vp == vp) {
870 nh = &nfsheur[hi];
871 break;
872 }
873 if (nfsheur[hi].nh_use > 0)
874 --nfsheur[hi].nh_use;
875 hi = (hi + 1) & (NUM_HEURISTIC - 1);
876 if (nfsheur[hi].nh_use < nh->nh_use)
877 nh = &nfsheur[hi];
878 }
879
880 if (nh->nh_vp != vp) {
881 nh->nh_vp = vp;
882 nh->nh_nextr = off;
883 nh->nh_use = NHUSE_INIT;
884 if (off == 0)
885 nh->nh_seqcount = 4;
886 else
887 nh->nh_seqcount = 1;
888 }
889
890 /*
891 * Calculate heuristic
892 */
893
894 if ((off == 0 && nh->nh_seqcount > 0) || off == nh->nh_nextr) {
895 if (++nh->nh_seqcount > IO_SEQMAX)
896 nh->nh_seqcount = IO_SEQMAX;
897 } else if (nh->nh_seqcount > 1) {
898 nh->nh_seqcount = 1;
899 } else {
900 nh->nh_seqcount = 0;
901 }
902 nh->nh_use += NHUSE_INC;
903 if (nh->nh_use > NHUSE_MAX)
904 nh->nh_use = NHUSE_MAX;
905 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
906 }
907
908 nfsm_reply(NFSX_POSTOPORFATTR(v3) + 3 * NFSX_UNSIGNED+nfsm_rndup(cnt));
909 if (v3) {
910 nfsm_build(tl, u_int32_t *, NFSX_V3FATTR + 4 * NFSX_UNSIGNED);
911 *tl++ = nfs_true;
912 fp = (struct nfs_fattr *)tl;
913 tl += (NFSX_V3FATTR / sizeof (u_int32_t));
914 } else {
915 nfsm_build(tl, u_int32_t *, NFSX_V2FATTR + NFSX_UNSIGNED);
916 fp = (struct nfs_fattr *)tl;
917 tl += (NFSX_V2FATTR / sizeof (u_int32_t));
918 }
919 len = left = nfsm_rndup(cnt);
920 if (cnt > 0) {
921 /*
922 * Generate the mbuf list with the uio_iov ref. to it.
923 */
924 i = 0;
925 m = m2 = mb;
926 while (left > 0) {
927 siz = min(M_TRAILINGSPACE(m), left);
928 if (siz > 0) {
929 left -= siz;
930 i++;
931 }
932 if (left > 0) {
933 MGET(m, M_WAIT, MT_DATA);
934 MCLGET(m, M_WAIT);
935 m->m_len = 0;
936 m2->m_next = m;
937 m2 = m;
938 }
939 }
940 MALLOC(iv, struct iovec *, i * sizeof (struct iovec),
941 M_TEMP, M_WAITOK);
942 uiop->uio_iov = iv2 = iv;
943 m = mb;
944 left = len;
945 i = 0;
946 while (left > 0) {
947 if (m == NULL)
948 panic("nfsrv_read iov");
949 siz = min(M_TRAILINGSPACE(m), left);
950 if (siz > 0) {
951 iv->iov_base = mtod(m, caddr_t) + m->m_len;
952 iv->iov_len = siz;
953 m->m_len += siz;
954 left -= siz;
955 iv++;
956 i++;
957 }
958 m = m->m_next;
959 }
960 uiop->uio_iovcnt = i;
961 uiop->uio_offset = off;
962 uiop->uio_resid = len;
963 uiop->uio_rw = UIO_READ;
964 uiop->uio_segflg = UIO_SYSSPACE;
965 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
966 off = uiop->uio_offset;
967 nh->nh_nextr = off;
968 FREE((caddr_t)iv2, M_TEMP);
3b568787 969 if (error || (getret = VOP_GETATTR(vp, vap, td))) {
984263bc
MD
970 if (!error)
971 error = getret;
972 m_freem(mreq);
973 vput(vp);
974 vp = NULL;
975 nfsm_reply(NFSX_POSTOPATTR(v3));
976 nfsm_srvpostop_attr(getret, vap);
977 error = 0;
978 goto nfsmout;
979 }
980 } else {
981 uiop->uio_resid = 0;
982 }
983 vput(vp);
984 vp = NULL;
985 nfsm_srvfillattr(vap, fp);
986 tlen = len - uiop->uio_resid;
987 cnt = cnt < tlen ? cnt : tlen;
988 tlen = nfsm_rndup(cnt);
989 if (len != tlen || tlen != cnt)
990 nfsm_adj(mb, len - tlen, tlen - cnt);
991 if (v3) {
992 *tl++ = txdr_unsigned(cnt);
993 if (len < reqlen)
994 *tl++ = nfs_true;
995 else
996 *tl++ = nfs_false;
997 }
998 *tl = txdr_unsigned(cnt);
999nfsmout:
1000 if (vp)
1001 vput(vp);
1002 return(error);
1003}
1004
1005/*
1006 * nfs write service
1007 */
1008int
dadab5e9 1009nfsrv_write(nfsd, slp, td, mrq)
984263bc
MD
1010 struct nfsrv_descript *nfsd;
1011 struct nfssvc_sock *slp;
dadab5e9 1012 struct thread *td;
984263bc
MD
1013 struct mbuf **mrq;
1014{
1015 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1016 struct sockaddr *nam = nfsd->nd_nam;
1017 caddr_t dpos = nfsd->nd_dpos;
1018 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
1019 struct iovec *ivp;
1020 int i, cnt;
1021 struct mbuf *mp;
1022 struct nfs_fattr *fp;
984263bc
MD
1023 struct iovec *iv;
1024 struct vattr va, forat;
dadab5e9
MD
1025 struct vattr *vap = &va;
1026 u_int32_t *tl;
1027 int32_t t1;
984263bc
MD
1028 caddr_t bpos;
1029 int error = 0, rdonly, cache, len, forat_ret = 1;
1030 int ioflags, aftat_ret = 1, retlen, zeroing, adjust;
1031 int stable = NFSV3WRITE_FILESYNC;
1032 int v3 = (nfsd->nd_flag & ND_NFSV3);
1033 char *cp2;
1034 struct mbuf *mb, *mb2, *mreq;
1035 struct vnode *vp = NULL;
1036 nfsfh_t nfh;
1037 fhandle_t *fhp;
1038 struct uio io, *uiop = &io;
1039 off_t off;
1040 u_quad_t frev;
1041
1042 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1043 if (mrep == NULL) {
1044 *mrq = NULL;
1045 error = 0;
1046 goto nfsmout;
1047 }
1048 fhp = &nfh.fh_generic;
1049 nfsm_srvmtofh(fhp);
1050 if (v3) {
1051 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1052 off = fxdr_hyper(tl);
1053 tl += 3;
1054 stable = fxdr_unsigned(int, *tl++);
1055 } else {
1056 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1057 off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1058 tl += 2;
1059 if (nfs_async)
1060 stable = NFSV3WRITE_UNSTABLE;
1061 }
1062 retlen = len = fxdr_unsigned(int32_t, *tl);
1063 cnt = i = 0;
1064
1065 /*
1066 * For NFS Version 2, it is not obvious what a write of zero length
1067 * should do, but I might as well be consistent with Version 3,
1068 * which is to return ok so long as there are no permission problems.
1069 */
1070 if (len > 0) {
1071 zeroing = 1;
1072 mp = mrep;
1073 while (mp) {
1074 if (mp == md) {
1075 zeroing = 0;
1076 adjust = dpos - mtod(mp, caddr_t);
1077 mp->m_len -= adjust;
1078 if (mp->m_len > 0 && adjust > 0)
1079 NFSMADV(mp, adjust);
1080 }
1081 if (zeroing)
1082 mp->m_len = 0;
1083 else if (mp->m_len > 0) {
1084 i += mp->m_len;
1085 if (i > len) {
1086 mp->m_len -= (i - len);
1087 zeroing = 1;
1088 }
1089 if (mp->m_len > 0)
1090 cnt++;
1091 }
1092 mp = mp->m_next;
1093 }
1094 }
1095 if (len > NFS_MAXDATA || len < 0 || i < len) {
1096 error = EIO;
1097 nfsm_reply(2 * NFSX_UNSIGNED);
1098 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1099 error = 0;
1100 goto nfsmout;
1101 }
1102 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
1103 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1104 if (error) {
1105 vp = NULL;
1106 nfsm_reply(2 * NFSX_UNSIGNED);
1107 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1108 error = 0;
1109 goto nfsmout;
1110 }
1111 if (v3)
3b568787 1112 forat_ret = VOP_GETATTR(vp, &forat, td);
984263bc
MD
1113 if (vp->v_type != VREG) {
1114 if (v3)
1115 error = EINVAL;
1116 else
1117 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1118 }
1119 if (!error) {
1120 nqsrv_getl(vp, ND_WRITE);
dadab5e9 1121 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
984263bc
MD
1122 }
1123 if (error) {
1124 vput(vp);
1125 vp = NULL;
1126 nfsm_reply(NFSX_WCCDATA(v3));
1127 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1128 error = 0;
1129 goto nfsmout;
1130 }
1131
1132 if (len > 0) {
1133 MALLOC(ivp, struct iovec *, cnt * sizeof (struct iovec), M_TEMP,
1134 M_WAITOK);
1135 uiop->uio_iov = iv = ivp;
1136 uiop->uio_iovcnt = cnt;
1137 mp = mrep;
1138 while (mp) {
1139 if (mp->m_len > 0) {
1140 ivp->iov_base = mtod(mp, caddr_t);
1141 ivp->iov_len = mp->m_len;
1142 ivp++;
1143 }
1144 mp = mp->m_next;
1145 }
1146
1147 /*
1148 * XXX
1149 * The IO_METASYNC flag indicates that all metadata (and not just
1150 * enough to ensure data integrity) mus be written to stable storage
1151 * synchronously.
1152 * (IO_METASYNC is not yet implemented in 4.4BSD-Lite.)
1153 */
1154 if (stable == NFSV3WRITE_UNSTABLE)
1155 ioflags = IO_NODELOCKED;
1156 else if (stable == NFSV3WRITE_DATASYNC)
1157 ioflags = (IO_SYNC | IO_NODELOCKED);
1158 else
1159 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1160 uiop->uio_resid = len;
1161 uiop->uio_rw = UIO_WRITE;
1162 uiop->uio_segflg = UIO_SYSSPACE;
dadab5e9 1163 uiop->uio_td = NULL;
984263bc
MD
1164 uiop->uio_offset = off;
1165 error = VOP_WRITE(vp, uiop, ioflags, cred);
1166 nfsstats.srvvop_writes++;
1167 FREE((caddr_t)iv, M_TEMP);
1168 }
3b568787 1169 aftat_ret = VOP_GETATTR(vp, vap, td);
984263bc
MD
1170 vput(vp);
1171 vp = NULL;
1172 if (!error)
1173 error = aftat_ret;
1174 nfsm_reply(NFSX_PREOPATTR(v3) + NFSX_POSTOPORFATTR(v3) +
1175 2 * NFSX_UNSIGNED + NFSX_WRITEVERF(v3));
1176 if (v3) {
1177 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, vap);
1178 if (error) {
1179 error = 0;
1180 goto nfsmout;
1181 }
1182 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1183 *tl++ = txdr_unsigned(retlen);
1184 /*
1185 * If nfs_async is set, then pretend the write was FILESYNC.
1186 */
1187 if (stable == NFSV3WRITE_UNSTABLE && !nfs_async)
1188 *tl++ = txdr_unsigned(stable);
1189 else
1190 *tl++ = txdr_unsigned(NFSV3WRITE_FILESYNC);
1191 /*
1192 * Actually, there is no need to txdr these fields,
1193 * but it may make the values more human readable,
1194 * for debugging purposes.
1195 */
1196 if (nfsver.tv_sec == 0)
1197 nfsver = boottime;
1198 *tl++ = txdr_unsigned(nfsver.tv_sec);
1199 *tl = txdr_unsigned(nfsver.tv_usec);
1200 } else {
1201 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1202 nfsm_srvfillattr(vap, fp);
1203 }
1204nfsmout:
1205 if (vp)
1206 vput(vp);
1207 return(error);
1208}
1209
1210/*
1211 * NFS write service with write gathering support. Called when
1212 * nfsrvw_procrastinate > 0.
1213 * See: Chet Juszczak, "Improving the Write Performance of an NFS Server",
1214 * in Proc. of the Winter 1994 Usenix Conference, pg. 247-259, San Franscisco,
1215 * Jan. 1994.
1216 */
1217int
dadab5e9 1218nfsrv_writegather(ndp, slp, td, mrq)
984263bc
MD
1219 struct nfsrv_descript **ndp;
1220 struct nfssvc_sock *slp;
dadab5e9 1221 struct thread *td;
984263bc
MD
1222 struct mbuf **mrq;
1223{
dadab5e9
MD
1224 struct iovec *ivp;
1225 struct mbuf *mp;
1226 struct nfsrv_descript *wp, *nfsd, *owp, *swp;
1227 struct nfs_fattr *fp;
1228 int i;
984263bc
MD
1229 struct iovec *iov;
1230 struct nfsrvw_delayhash *wpp;
1231 struct ucred *cred;
1232 struct vattr va, forat;
dadab5e9
MD
1233 u_int32_t *tl;
1234 int32_t t1;
984263bc
MD
1235 caddr_t bpos, dpos;
1236 int error = 0, rdonly, cache, len, forat_ret = 1;
1237 int ioflags, aftat_ret = 1, s, adjust, v3, zeroing;
1238 char *cp2;
1239 struct mbuf *mb, *mb2, *mreq, *mrep, *md;
1240 struct vnode *vp = NULL;
1241 struct uio io, *uiop = &io;
1242 u_quad_t frev, cur_usec;
1243
1244 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1245#ifndef nolint
1246 i = 0;
1247 len = 0;
1248#endif
1249 *mrq = NULL;
1250 if (*ndp) {
1251 nfsd = *ndp;
1252 *ndp = NULL;
1253 mrep = nfsd->nd_mrep;
1254 md = nfsd->nd_md;
1255 dpos = nfsd->nd_dpos;
1256 cred = &nfsd->nd_cr;
1257 v3 = (nfsd->nd_flag & ND_NFSV3);
1258 LIST_INIT(&nfsd->nd_coalesce);
1259 nfsd->nd_mreq = NULL;
1260 nfsd->nd_stable = NFSV3WRITE_FILESYNC;
1261 cur_usec = nfs_curusec();
1262 nfsd->nd_time = cur_usec +
1263 (v3 ? nfsrvw_procrastinate_v3 : nfsrvw_procrastinate);
1264
1265 /*
1266 * Now, get the write header..
1267 */
1268 nfsm_srvmtofh(&nfsd->nd_fh);
1269 if (v3) {
1270 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1271 nfsd->nd_off = fxdr_hyper(tl);
1272 tl += 3;
1273 nfsd->nd_stable = fxdr_unsigned(int, *tl++);
1274 } else {
1275 nfsm_dissect(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1276 nfsd->nd_off = (off_t)fxdr_unsigned(u_int32_t, *++tl);
1277 tl += 2;
1278 if (nfs_async)
1279 nfsd->nd_stable = NFSV3WRITE_UNSTABLE;
1280 }
1281 len = fxdr_unsigned(int32_t, *tl);
1282 nfsd->nd_len = len;
1283 nfsd->nd_eoff = nfsd->nd_off + len;
1284
1285 /*
1286 * Trim the header out of the mbuf list and trim off any trailing
1287 * junk so that the mbuf list has only the write data.
1288 */
1289 zeroing = 1;
1290 i = 0;
1291 mp = mrep;
1292 while (mp) {
1293 if (mp == md) {
1294 zeroing = 0;
1295 adjust = dpos - mtod(mp, caddr_t);
1296 mp->m_len -= adjust;
1297 if (mp->m_len > 0 && adjust > 0)
1298 NFSMADV(mp, adjust);
1299 }
1300 if (zeroing)
1301 mp->m_len = 0;
1302 else {
1303 i += mp->m_len;
1304 if (i > len) {
1305 mp->m_len -= (i - len);
1306 zeroing = 1;
1307 }
1308 }
1309 mp = mp->m_next;
1310 }
1311 if (len > NFS_MAXDATA || len < 0 || i < len) {
1312nfsmout:
1313 m_freem(mrep);
1314 error = EIO;
1315 nfsm_writereply(2 * NFSX_UNSIGNED, v3);
1316 if (v3)
1317 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1318 nfsd->nd_mreq = mreq;
1319 nfsd->nd_mrep = NULL;
1320 nfsd->nd_time = 0;
1321 }
1322
1323 /*
1324 * Add this entry to the hash and time queues.
1325 */
1326 s = splsoftclock();
1327 owp = NULL;
1328 wp = slp->ns_tq.lh_first;
1329 while (wp && wp->nd_time < nfsd->nd_time) {
1330 owp = wp;
1331 wp = wp->nd_tq.le_next;
1332 }
1333 NFS_DPF(WG, ("Q%03x", nfsd->nd_retxid & 0xfff));
1334 if (owp) {
1335 LIST_INSERT_AFTER(owp, nfsd, nd_tq);
1336 } else {
1337 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1338 }
1339 if (nfsd->nd_mrep) {
1340 wpp = NWDELAYHASH(slp, nfsd->nd_fh.fh_fid.fid_data);
1341 owp = NULL;
1342 wp = wpp->lh_first;
1343 while (wp &&
1344 bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1345 owp = wp;
1346 wp = wp->nd_hash.le_next;
1347 }
1348 while (wp && wp->nd_off < nfsd->nd_off &&
1349 !bcmp((caddr_t)&nfsd->nd_fh,(caddr_t)&wp->nd_fh,NFSX_V3FH)) {
1350 owp = wp;
1351 wp = wp->nd_hash.le_next;
1352 }
1353 if (owp) {
1354 LIST_INSERT_AFTER(owp, nfsd, nd_hash);
1355
1356 /*
1357 * Search the hash list for overlapping entries and
1358 * coalesce.
1359 */
1360 for(; nfsd && NFSW_CONTIG(owp, nfsd); nfsd = wp) {
1361 wp = nfsd->nd_hash.le_next;
1362 if (NFSW_SAMECRED(owp, nfsd))
1363 nfsrvw_coalesce(owp, nfsd);
1364 }
1365 } else {
1366 LIST_INSERT_HEAD(wpp, nfsd, nd_hash);
1367 }
1368 }
1369 splx(s);
1370 }
1371
1372 /*
1373 * Now, do VOP_WRITE()s for any one(s) that need to be done now
1374 * and generate the associated reply mbuf list(s).
1375 */
1376loop1:
1377 cur_usec = nfs_curusec();
1378 s = splsoftclock();
1379 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = owp) {
1380 owp = nfsd->nd_tq.le_next;
1381 if (nfsd->nd_time > cur_usec)
1382 break;
1383 if (nfsd->nd_mreq)
1384 continue;
1385 NFS_DPF(WG, ("P%03x", nfsd->nd_retxid & 0xfff));
1386 LIST_REMOVE(nfsd, nd_tq);
1387 LIST_REMOVE(nfsd, nd_hash);
1388 splx(s);
1389 mrep = nfsd->nd_mrep;
1390 nfsd->nd_mrep = NULL;
1391 cred = &nfsd->nd_cr;
1392 v3 = (nfsd->nd_flag & ND_NFSV3);
1393 forat_ret = aftat_ret = 1;
1394 error = nfsrv_fhtovp(&nfsd->nd_fh, 1, &vp, cred, slp,
1395 nfsd->nd_nam, &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
1396 if (!error) {
1397 if (v3)
3b568787 1398 forat_ret = VOP_GETATTR(vp, &forat, td);
984263bc
MD
1399 if (vp->v_type != VREG) {
1400 if (v3)
1401 error = EINVAL;
1402 else
1403 error = (vp->v_type == VDIR) ? EISDIR : EACCES;
1404 }
1405 } else {
1406 vp = NULL;
1407 }
1408 if (!error) {
1409 nqsrv_getl(vp, ND_WRITE);
dadab5e9 1410 error = nfsrv_access(vp, VWRITE, cred, rdonly, td, 1);
984263bc
MD
1411 }
1412
1413 if (nfsd->nd_stable == NFSV3WRITE_UNSTABLE)
1414 ioflags = IO_NODELOCKED;
1415 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC)
1416 ioflags = (IO_SYNC | IO_NODELOCKED);
1417 else
1418 ioflags = (IO_METASYNC | IO_SYNC | IO_NODELOCKED);
1419 uiop->uio_rw = UIO_WRITE;
1420 uiop->uio_segflg = UIO_SYSSPACE;
dadab5e9 1421 uiop->uio_td = NULL;
984263bc
MD
1422 uiop->uio_offset = nfsd->nd_off;
1423 uiop->uio_resid = nfsd->nd_eoff - nfsd->nd_off;
1424 if (uiop->uio_resid > 0) {
1425 mp = mrep;
1426 i = 0;
1427 while (mp) {
1428 if (mp->m_len > 0)
1429 i++;
1430 mp = mp->m_next;
1431 }
1432 uiop->uio_iovcnt = i;
1433 MALLOC(iov, struct iovec *, i * sizeof (struct iovec),
1434 M_TEMP, M_WAITOK);
1435 uiop->uio_iov = ivp = iov;
1436 mp = mrep;
1437 while (mp) {
1438 if (mp->m_len > 0) {
1439 ivp->iov_base = mtod(mp, caddr_t);
1440 ivp->iov_len = mp->m_len;
1441 ivp++;
1442 }
1443 mp = mp->m_next;
1444 }
1445 if (!error) {
1446 error = VOP_WRITE(vp, uiop, ioflags, cred);
1447 nfsstats.srvvop_writes++;
1448 }
1449 FREE((caddr_t)iov, M_TEMP);
1450 }
1451 m_freem(mrep);
1452 if (vp) {
3b568787 1453 aftat_ret = VOP_GETATTR(vp, &va, td);
984263bc
MD
1454 vput(vp);
1455 vp = NULL;
1456 }
1457
1458 /*
1459 * Loop around generating replies for all write rpcs that have
1460 * now been completed.
1461 */
1462 swp = nfsd;
1463 do {
1464 NFS_DPF(WG, ("R%03x", nfsd->nd_retxid & 0xfff));
1465 if (error) {
1466 nfsm_writereply(NFSX_WCCDATA(v3), v3);
1467 if (v3) {
1468 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1469 }
1470 } else {
1471 nfsm_writereply(NFSX_PREOPATTR(v3) +
1472 NFSX_POSTOPORFATTR(v3) + 2 * NFSX_UNSIGNED +
1473 NFSX_WRITEVERF(v3), v3);
1474 if (v3) {
1475 nfsm_srvwcc_data(forat_ret, &forat, aftat_ret, &va);
1476 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1477 *tl++ = txdr_unsigned(nfsd->nd_len);
1478 *tl++ = txdr_unsigned(swp->nd_stable);
1479 /*
1480 * Actually, there is no need to txdr these fields,
1481 * but it may make the values more human readable,
1482 * for debugging purposes.
1483 */
1484 if (nfsver.tv_sec == 0)
1485 nfsver = boottime;
1486 *tl++ = txdr_unsigned(nfsver.tv_sec);
1487 *tl = txdr_unsigned(nfsver.tv_usec);
1488 } else {
1489 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1490 nfsm_srvfillattr(&va, fp);
1491 }
1492 }
1493 nfsd->nd_mreq = mreq;
1494 if (nfsd->nd_mrep)
1495 panic("nfsrv_write: nd_mrep not free");
1496
1497 /*
1498 * Done. Put it at the head of the timer queue so that
1499 * the final phase can return the reply.
1500 */
1501 s = splsoftclock();
1502 if (nfsd != swp) {
1503 nfsd->nd_time = 0;
1504 LIST_INSERT_HEAD(&slp->ns_tq, nfsd, nd_tq);
1505 }
1506 nfsd = swp->nd_coalesce.lh_first;
1507 if (nfsd) {
1508 LIST_REMOVE(nfsd, nd_tq);
1509 }
1510 splx(s);
1511 } while (nfsd);
1512 s = splsoftclock();
1513 swp->nd_time = 0;
1514 LIST_INSERT_HEAD(&slp->ns_tq, swp, nd_tq);
1515 splx(s);
1516 goto loop1;
1517 }
1518 splx(s);
1519
1520 /*
1521 * Search for a reply to return.
1522 */
1523 s = splsoftclock();
1524 for (nfsd = slp->ns_tq.lh_first; nfsd; nfsd = nfsd->nd_tq.le_next)
1525 if (nfsd->nd_mreq) {
1526 NFS_DPF(WG, ("X%03x", nfsd->nd_retxid & 0xfff));
1527 LIST_REMOVE(nfsd, nd_tq);
1528 *mrq = nfsd->nd_mreq;
1529 *ndp = nfsd;
1530 break;
1531 }
1532 splx(s);
1533 return (0);
1534}
1535
1536/*
1537 * Coalesce the write request nfsd into owp. To do this we must:
1538 * - remove nfsd from the queues
1539 * - merge nfsd->nd_mrep into owp->nd_mrep
1540 * - update the nd_eoff and nd_stable for owp
1541 * - put nfsd on owp's nd_coalesce list
1542 * NB: Must be called at splsoftclock().
1543 */
1544static void
dadab5e9 1545nfsrvw_coalesce(struct nfsrv_descript *owp, struct nfsrv_descript *nfsd)
984263bc 1546{
dadab5e9
MD
1547 int overlap;
1548 struct mbuf *mp;
984263bc
MD
1549 struct nfsrv_descript *p;
1550
1551 NFS_DPF(WG, ("C%03x-%03x",
1552 nfsd->nd_retxid & 0xfff, owp->nd_retxid & 0xfff));
1553 LIST_REMOVE(nfsd, nd_hash);
1554 LIST_REMOVE(nfsd, nd_tq);
1555 if (owp->nd_eoff < nfsd->nd_eoff) {
1556 overlap = owp->nd_eoff - nfsd->nd_off;
1557 if (overlap < 0)
1558 panic("nfsrv_coalesce: bad off");
1559 if (overlap > 0)
1560 m_adj(nfsd->nd_mrep, overlap);
1561 mp = owp->nd_mrep;
1562 while (mp->m_next)
1563 mp = mp->m_next;
1564 mp->m_next = nfsd->nd_mrep;
1565 owp->nd_eoff = nfsd->nd_eoff;
1566 } else
1567 m_freem(nfsd->nd_mrep);
1568 nfsd->nd_mrep = NULL;
1569 if (nfsd->nd_stable == NFSV3WRITE_FILESYNC)
1570 owp->nd_stable = NFSV3WRITE_FILESYNC;
1571 else if (nfsd->nd_stable == NFSV3WRITE_DATASYNC &&
1572 owp->nd_stable == NFSV3WRITE_UNSTABLE)
1573 owp->nd_stable = NFSV3WRITE_DATASYNC;
1574 LIST_INSERT_HEAD(&owp->nd_coalesce, nfsd, nd_tq);
1575
1576 /*
1577 * If nfsd had anything else coalesced into it, transfer them
1578 * to owp, otherwise their replies will never get sent.
1579 */
1580 for (p = nfsd->nd_coalesce.lh_first; p;
1581 p = nfsd->nd_coalesce.lh_first) {
1582 LIST_REMOVE(p, nd_tq);
1583 LIST_INSERT_HEAD(&owp->nd_coalesce, p, nd_tq);
1584 }
1585}
1586
1587/*
1588 * nfs create service
1589 * now does a truncate to 0 length via. setattr if it already exists
1590 */
1591int
dadab5e9 1592nfsrv_create(nfsd, slp, td, mrq)
984263bc
MD
1593 struct nfsrv_descript *nfsd;
1594 struct nfssvc_sock *slp;
dadab5e9 1595 struct thread *td;
984263bc
MD
1596 struct mbuf **mrq;
1597{
1598 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1599 struct sockaddr *nam = nfsd->nd_nam;
1600 caddr_t dpos = nfsd->nd_dpos;
1601 struct ucred *cred = &nfsd->nd_cr;
dadab5e9 1602 struct nfs_fattr *fp;
984263bc 1603 struct vattr va, dirfor, diraft;
dadab5e9
MD
1604 struct vattr *vap = &va;
1605 struct nfsv2_sattr *sp;
1606 u_int32_t *tl;
984263bc 1607 struct nameidata nd;
dadab5e9 1608 int32_t t1;
984263bc
MD
1609 caddr_t bpos;
1610 int error = 0, rdev, cache, len, tsize, dirfor_ret = 1, diraft_ret = 1;
1611 int v3 = (nfsd->nd_flag & ND_NFSV3), how, exclusive_flag = 0;
1612 caddr_t cp;
1613 char *cp2;
1614 struct mbuf *mb, *mb2, *mreq;
1615 struct vnode *dirp = (struct vnode *)0;
1616 nfsfh_t nfh;
1617 fhandle_t *fhp;
1618 u_quad_t frev, tempsize;
1619 u_char cverf[NFSX_V3CREATEVERF];
1620
1621 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1622#ifndef nolint
1623 rdev = 0;
1624#endif
1625 ndclear(&nd);
1626
1627 fhp = &nfh.fh_generic;
1628 nfsm_srvmtofh(fhp);
1629 nfsm_srvnamesiz(len);
1630
1631 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
1632 nd.ni_cnd.cn_nameiop = NAMEI_CREATE;
1633 nd.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_LOCKLEAF | CNP_SAVESTART;
984263bc
MD
1634
1635 /*
1636 * Call namei and do initial cleanup to get a few things
1637 * out of the way. If we get an initial error we cleanup
1638 * and return here to avoid special-casing the invalid nd
1639 * structure through the rest of the case. dirp may be
1640 * set even if an error occurs, but the nd structure will not
1641 * be valid at all if an error occurs so we have to invalidate it
1642 * prior to calling nfsm_reply ( which might goto nfsmout ).
1643 */
1644 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 1645 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
1646 if (dirp) {
1647 if (v3) {
3b568787 1648 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
1649 } else {
1650 vrele(dirp);
1651 dirp = NULL;
1652 }
1653 }
1654 if (error) {
1655 nfsm_reply(NFSX_WCCDATA(v3));
1656 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1657 error = 0;
1658 goto nfsmout;
1659 }
1660
1661 /*
1662 * No error. Continue. State:
1663 *
1664 * startdir is valid ( we release this immediately )
1665 * dirp may be valid
1666 * nd.ni_vp may be valid
1667 * nd.ni_dvp is valid
1668 *
1669 * The error state is set through the code and we may also do some
1670 * opportunistic releasing of vnodes to avoid holding locks through
1671 * NFS I/O. The cleanup at the end is a catch-all
1672 */
1673
1674 VATTR_NULL(vap);
1675 if (v3) {
1676 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1677 how = fxdr_unsigned(int, *tl);
1678 switch (how) {
1679 case NFSV3CREATE_GUARDED:
1680 if (nd.ni_vp) {
1681 error = EEXIST;
1682 break;
1683 }
1684 /* fall through */
1685 case NFSV3CREATE_UNCHECKED:
1686 nfsm_srvsattr(vap);
1687 break;
1688 case NFSV3CREATE_EXCLUSIVE:
1689 nfsm_dissect(cp, caddr_t, NFSX_V3CREATEVERF);
1690 bcopy(cp, cverf, NFSX_V3CREATEVERF);
1691 exclusive_flag = 1;
1692 break;
1693 };
1694 vap->va_type = VREG;
1695 } else {
1696 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1697 vap->va_type = IFTOVT(fxdr_unsigned(u_int32_t, sp->sa_mode));
1698 if (vap->va_type == VNON)
1699 vap->va_type = VREG;
1700 vap->va_mode = nfstov_mode(sp->sa_mode);
1701 switch (vap->va_type) {
1702 case VREG:
1703 tsize = fxdr_unsigned(int32_t, sp->sa_size);
1704 if (tsize != -1)
1705 vap->va_size = (u_quad_t)tsize;
1706 break;
1707 case VCHR:
1708 case VBLK:
1709 case VFIFO:
1710 rdev = fxdr_unsigned(long, sp->sa_size);
1711 break;
1712 default:
1713 break;
1714 };
1715 }
1716
1717 /*
1718 * Iff doesn't exist, create it
1719 * otherwise just truncate to 0 length
1720 * should I set the mode too ?
1721 *
1722 * The only possible error we can have at this point is EEXIST.
1723 * nd.ni_vp will also be non-NULL in that case.
1724 */
1725 if (nd.ni_vp == NULL) {
1726 if (vap->va_mode == (mode_t)VNOVAL)
1727 vap->va_mode = 0;
1728 if (vap->va_type == VREG || vap->va_type == VSOCK) {
1729 nqsrv_getl(nd.ni_dvp, ND_WRITE);
bc0c094e
MD
1730 error = VOP_CREATE(nd.ni_dvp, NCPNULL, &nd.ni_vp,
1731 &nd.ni_cnd, vap);
984263bc
MD
1732 if (error)
1733 NDFREE(&nd, NDF_ONLY_PNBUF);
1734 else {
1735 nfsrv_object_create(nd.ni_vp);
1736 if (exclusive_flag) {
1737 exclusive_flag = 0;
1738 VATTR_NULL(vap);
1739 bcopy(cverf, (caddr_t)&vap->va_atime,
1740 NFSX_V3CREATEVERF);
1741 error = VOP_SETATTR(nd.ni_vp, vap, cred,
dadab5e9 1742 td);
984263bc
MD
1743 }
1744 }
1745 } else if (
1746 vap->va_type == VCHR ||
1747 vap->va_type == VBLK ||
1748 vap->va_type == VFIFO
1749 ) {
1750 /*
1751 * Handle SysV FIFO node special cases. All other
1752 * devices require super user to access.
1753 */
1754 if (vap->va_type == VCHR && rdev == 0xffffffff)
1755 vap->va_type = VFIFO;
1756 if (vap->va_type != VFIFO &&
dadab5e9 1757 (error = suser_cred(cred, 0))) {
984263bc
MD
1758 goto nfsmreply0;
1759 }
1760 vap->va_rdev = rdev;
1761 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1762
bc0c094e
MD
1763 error = VOP_MKNOD(nd.ni_dvp, NCPNULL, &nd.ni_vp,
1764 &nd.ni_cnd, vap);
984263bc
MD
1765 if (error) {
1766 NDFREE(&nd, NDF_ONLY_PNBUF);
1767 goto nfsmreply0;
1768 }
1769 vput(nd.ni_vp);
1770 nd.ni_vp = NULL;
1771
1772 /*
1773 * release dvp prior to lookup
1774 */
1775 vput(nd.ni_dvp);
1776 nd.ni_dvp = NULL;
1777
1778 /*
1779 * Setup for lookup.
1780 *
1781 * Even though LOCKPARENT was cleared, ni_dvp may
1782 * be garbage.
1783 */
2b69e610
MD
1784 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1785 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
dadab5e9 1786 nd.ni_cnd.cn_td = td;
984263bc
MD
1787 nd.ni_cnd.cn_cred = cred;
1788
1789 error = lookup(&nd);
1790 nd.ni_dvp = NULL;
1791
1792 if (error != 0) {
1793 nfsm_reply(0);
1794 /* fall through on certain errors */
1795 }
1796 nfsrv_object_create(nd.ni_vp);
2b69e610 1797 if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK) {
984263bc
MD
1798 error = EINVAL;
1799 goto nfsmreply0;
1800 }
1801 } else {
1802 error = ENXIO;
1803 }
1804 } else {
1805 if (vap->va_size != -1) {
1806 error = nfsrv_access(nd.ni_vp, VWRITE, cred,
2b69e610 1807 (nd.ni_cnd.cn_flags & CNP_RDONLY), td, 0);
984263bc
MD
1808 if (!error) {
1809 nqsrv_getl(nd.ni_vp, ND_WRITE);
1810 tempsize = vap->va_size;
1811 VATTR_NULL(vap);
1812 vap->va_size = tempsize;
dadab5e9 1813 error = VOP_SETATTR(nd.ni_vp, vap, cred, td);
984263bc
MD
1814 }
1815 }
1816 }
1817
1818 if (!error) {
1819 bzero((caddr_t)fhp, sizeof(nfh));
1820 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
1821 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
1822 if (!error)
3b568787 1823 error = VOP_GETATTR(nd.ni_vp, vap, td);
984263bc
MD
1824 }
1825 if (v3) {
1826 if (exclusive_flag && !error &&
1827 bcmp(cverf, (caddr_t)&vap->va_atime, NFSX_V3CREATEVERF))
1828 error = EEXIST;
3b568787 1829 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
1830 vrele(dirp);
1831 dirp = NULL;
1832 }
1833 nfsm_reply(NFSX_SRVFH(v3) + NFSX_FATTR(v3) + NFSX_WCCDATA(v3));
1834 if (v3) {
1835 if (!error) {
1836 nfsm_srvpostop_fh(fhp);
1837 nfsm_srvpostop_attr(0, vap);
1838 }
1839 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1840 error = 0;
1841 } else {
1842 nfsm_srvfhtom(fhp, v3);
1843 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
1844 nfsm_srvfillattr(vap, fp);
1845 }
1846 goto nfsmout;
1847
1848nfsmreply0:
1849 nfsm_reply(0);
1850 error = 0;
1851 /* fall through */
1852
1853nfsmout:
1854 if (nd.ni_startdir) {
1855 vrele(nd.ni_startdir);
1856 nd.ni_startdir = NULL;
1857 }
1858 if (dirp)
1859 vrele(dirp);
1860 NDFREE(&nd, NDF_ONLY_PNBUF);
1861 if (nd.ni_dvp) {
1862 if (nd.ni_dvp == nd.ni_vp)
1863 vrele(nd.ni_dvp);
1864 else
1865 vput(nd.ni_dvp);
1866 }
1867 if (nd.ni_vp)
1868 vput(nd.ni_vp);
1869 return (error);
1870}
1871
1872/*
1873 * nfs v3 mknod service
1874 */
1875int
dadab5e9 1876nfsrv_mknod(nfsd, slp, td, mrq)
984263bc
MD
1877 struct nfsrv_descript *nfsd;
1878 struct nfssvc_sock *slp;
dadab5e9 1879 struct thread *td;
984263bc
MD
1880 struct mbuf **mrq;
1881{
1882 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
1883 struct sockaddr *nam = nfsd->nd_nam;
1884 caddr_t dpos = nfsd->nd_dpos;
1885 struct ucred *cred = &nfsd->nd_cr;
1886 struct vattr va, dirfor, diraft;
dadab5e9
MD
1887 struct vattr *vap = &va;
1888 u_int32_t *tl;
984263bc 1889 struct nameidata nd;
dadab5e9 1890 int32_t t1;
984263bc
MD
1891 caddr_t bpos;
1892 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
1893 u_int32_t major, minor;
1894 enum vtype vtyp;
1895 char *cp2;
1896 struct mbuf *mb, *mb2, *mreq;
1897 struct vnode *vp, *dirp = (struct vnode *)0;
1898 nfsfh_t nfh;
1899 fhandle_t *fhp;
1900 u_quad_t frev;
1901
1902 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
1903 ndclear(&nd);
1904
1905 fhp = &nfh.fh_generic;
1906 nfsm_srvmtofh(fhp);
1907 nfsm_srvnamesiz(len);
1908
1909 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
1910 nd.ni_cnd.cn_nameiop = NAMEI_CREATE;
1911 nd.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_LOCKLEAF | CNP_SAVESTART;
984263bc
MD
1912
1913 /*
1914 * Handle nfs_namei() call. If an error occurs, the nd structure
1915 * is not valid. However, nfsm_*() routines may still jump to
1916 * nfsmout.
1917 */
1918
1919 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 1920 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc 1921 if (dirp)
3b568787 1922 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
1923 if (error) {
1924 nfsm_reply(NFSX_WCCDATA(1));
1925 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
1926 error = 0;
1927 goto nfsmout;
1928 }
1929 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
1930 vtyp = nfsv3tov_type(*tl);
1931 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1932 error = NFSERR_BADTYPE;
1933 goto out;
1934 }
1935 VATTR_NULL(vap);
1936 nfsm_srvsattr(vap);
1937 if (vtyp == VCHR || vtyp == VBLK) {
1938 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1939 major = fxdr_unsigned(u_int32_t, *tl++);
1940 minor = fxdr_unsigned(u_int32_t, *tl);
1941 vap->va_rdev = makeudev(major, minor);
1942 }
1943
1944 /*
1945 * Iff doesn't exist, create it.
1946 */
1947 if (nd.ni_vp) {
1948 error = EEXIST;
1949 goto out;
1950 }
1951 vap->va_type = vtyp;
1952 if (vap->va_mode == (mode_t)VNOVAL)
1953 vap->va_mode = 0;
1954 if (vtyp == VSOCK) {
1955 vrele(nd.ni_startdir);
1956 nd.ni_startdir = NULL;
1957 nqsrv_getl(nd.ni_dvp, ND_WRITE);
bc0c094e
MD
1958 error = VOP_CREATE(nd.ni_dvp, NCPNULL, &nd.ni_vp,
1959 &nd.ni_cnd, vap);
984263bc
MD
1960 if (error)
1961 NDFREE(&nd, NDF_ONLY_PNBUF);
1962 } else {
dadab5e9 1963 if (vtyp != VFIFO && (error = suser_cred(cred, 0)))
984263bc
MD
1964 goto out;
1965 nqsrv_getl(nd.ni_dvp, ND_WRITE);
1966
bc0c094e
MD
1967 error = VOP_MKNOD(nd.ni_dvp, NCPNULL, &nd.ni_vp,
1968 &nd.ni_cnd, vap);
984263bc
MD
1969 if (error) {
1970 NDFREE(&nd, NDF_ONLY_PNBUF);
1971 goto out;
1972 }
1973 vput(nd.ni_vp);
1974 nd.ni_vp = NULL;
1975
1976 /*
1977 * Release dvp prior to lookup
1978 */
1979 vput(nd.ni_dvp);
1980 nd.ni_dvp = NULL;
1981
dadab5e9 1982 KKASSERT(td->td_proc);
2b69e610
MD
1983 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
1984 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT);
dadab5e9
MD
1985 nd.ni_cnd.cn_td = td;
1986 nd.ni_cnd.cn_cred = td->td_proc->p_ucred;
984263bc
MD
1987
1988 error = lookup(&nd);
1989 nd.ni_dvp = NULL;
1990
1991 if (error)
1992 goto out;
2b69e610 1993 if (nd.ni_cnd.cn_flags & CNP_ISSYMLINK)
984263bc
MD
1994 error = EINVAL;
1995 }
1996
1997 /*
1998 * send response, cleanup, return.
1999 */
2000out:
2001 if (nd.ni_startdir) {
2002 vrele(nd.ni_startdir);
2003 nd.ni_startdir = NULL;
2004 }
2005 NDFREE(&nd, NDF_ONLY_PNBUF);
2006 if (nd.ni_dvp) {
2007 if (nd.ni_dvp == nd.ni_vp)
2008 vrele(nd.ni_dvp);
2009 else
2010 vput(nd.ni_dvp);
2011 nd.ni_dvp = NULL;
2012 }
2013 vp = nd.ni_vp;
2014 if (!error) {
2015 bzero((caddr_t)fhp, sizeof(nfh));
2016 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
2017 error = VFS_VPTOFH(vp, &fhp->fh_fid);
2018 if (!error)
3b568787 2019 error = VOP_GETATTR(vp, vap, td);
984263bc
MD
2020 }
2021 if (vp) {
2022 vput(vp);
2023 vp = NULL;
2024 nd.ni_vp = NULL;
2025 }
3b568787 2026 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2027 if (dirp) {
2028 vrele(dirp);
2029 dirp = NULL;
2030 }
2031 nfsm_reply(NFSX_SRVFH(1) + NFSX_POSTOPATTR(1) + NFSX_WCCDATA(1));
2032 if (!error) {
2033 nfsm_srvpostop_fh(fhp);
2034 nfsm_srvpostop_attr(0, vap);
2035 }
2036 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2037 return (0);
2038nfsmout:
2039 if (dirp)
2040 vrele(dirp);
2041 if (nd.ni_startdir)
2042 vrele(nd.ni_startdir);
2043 NDFREE(&nd, NDF_ONLY_PNBUF);
2044 if (nd.ni_dvp) {
2045 if (nd.ni_dvp == nd.ni_vp)
2046 vrele(nd.ni_dvp);
2047 else
2048 vput(nd.ni_dvp);
2049 }
2050 if (nd.ni_vp)
2051 vput(nd.ni_vp);
2052 return (error);
2053}
2054
2055/*
2056 * nfs remove service
2057 */
2058int
dadab5e9 2059nfsrv_remove(nfsd, slp, td, mrq)
984263bc
MD
2060 struct nfsrv_descript *nfsd;
2061 struct nfssvc_sock *slp;
dadab5e9 2062 struct thread *td;
984263bc
MD
2063 struct mbuf **mrq;
2064{
2065 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2066 struct sockaddr *nam = nfsd->nd_nam;
2067 caddr_t dpos = nfsd->nd_dpos;
2068 struct ucred *cred = &nfsd->nd_cr;
2069 struct nameidata nd;
dadab5e9
MD
2070 u_int32_t *tl;
2071 int32_t t1;
984263bc
MD
2072 caddr_t bpos;
2073 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2074 int v3 = (nfsd->nd_flag & ND_NFSV3);
2075 char *cp2;
2076 struct mbuf *mb, *mreq;
2077 struct vnode *dirp;
2078 struct vattr dirfor, diraft;
2079 nfsfh_t nfh;
2080 fhandle_t *fhp;
2081 u_quad_t frev;
2082
2083 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2084 ndclear(&nd);
2085
2086 fhp = &nfh.fh_generic;
2087 nfsm_srvmtofh(fhp);
2088 nfsm_srvnamesiz(len);
2089
2090 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
2091 nd.ni_cnd.cn_nameiop = NAMEI_DELETE;
2092 nd.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_LOCKLEAF;
984263bc 2093 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 2094 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2095 if (dirp) {
2096 if (v3) {
3b568787 2097 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
2098 } else {
2099 vrele(dirp);
2100 dirp = NULL;
2101 }
2102 }
2103 if (error == 0) {
2104 if (nd.ni_vp->v_type == VDIR) {
2105 error = EPERM; /* POSIX */
2106 goto out;
2107 }
2108 /*
2109 * The root of a mounted filesystem cannot be deleted.
2110 */
2111 if (nd.ni_vp->v_flag & VROOT) {
2112 error = EBUSY;
2113 goto out;
2114 }
2115out:
2116 if (!error) {
2117 nqsrv_getl(nd.ni_dvp, ND_WRITE);
2118 nqsrv_getl(nd.ni_vp, ND_WRITE);
bc0c094e 2119 error = VOP_REMOVE(nd.ni_dvp, NCPNULL, nd.ni_vp, &nd.ni_cnd);
984263bc
MD
2120 NDFREE(&nd, NDF_ONLY_PNBUF);
2121 }
2122 }
2123 if (dirp && v3) {
3b568787 2124 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2125 vrele(dirp);
2126 dirp = NULL;
2127 }
2128 nfsm_reply(NFSX_WCCDATA(v3));
2129 if (v3) {
2130 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2131 error = 0;
2132 }
2133nfsmout:
2134 NDFREE(&nd, NDF_ONLY_PNBUF);
2135 if (nd.ni_dvp) {
2136 if (nd.ni_dvp == nd.ni_vp)
2137 vrele(nd.ni_dvp);
2138 else
2139 vput(nd.ni_dvp);
2140 }
2141 if (nd.ni_vp)
2142 vput(nd.ni_vp);
2143 return(error);
2144}
2145
2146/*
2147 * nfs rename service
2148 */
2149int
dadab5e9 2150nfsrv_rename(nfsd, slp, td, mrq)
984263bc
MD
2151 struct nfsrv_descript *nfsd;
2152 struct nfssvc_sock *slp;
dadab5e9 2153 struct thread *td;
984263bc
MD
2154 struct mbuf **mrq;
2155{
2156 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2157 struct sockaddr *nam = nfsd->nd_nam;
2158 caddr_t dpos = nfsd->nd_dpos;
2159 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
2160 u_int32_t *tl;
2161 int32_t t1;
984263bc
MD
2162 caddr_t bpos;
2163 int error = 0, cache, len, len2, fdirfor_ret = 1, fdiraft_ret = 1;
2164 int tdirfor_ret = 1, tdiraft_ret = 1;
2165 int v3 = (nfsd->nd_flag & ND_NFSV3);
2166 char *cp2;
2167 struct mbuf *mb, *mreq;
2168 struct nameidata fromnd, tond;
2169 struct vnode *fvp, *tvp, *tdvp, *fdirp = (struct vnode *)0;
2170 struct vnode *tdirp = (struct vnode *)0;
2171 struct vattr fdirfor, fdiraft, tdirfor, tdiraft;
2172 nfsfh_t fnfh, tnfh;
2173 fhandle_t *ffhp, *tfhp;
2174 u_quad_t frev;
2175 uid_t saved_uid;
2176
2177 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2178#ifndef nolint
2179 fvp = (struct vnode *)0;
2180#endif
2181 ffhp = &fnfh.fh_generic;
2182 tfhp = &tnfh.fh_generic;
2183
2184 /*
2185 * Clear fields incase goto nfsmout occurs from macro.
2186 */
2187
2188 ndclear(&fromnd);
2189 ndclear(&tond);
2190
2191 nfsm_srvmtofh(ffhp);
2192 nfsm_srvnamesiz(len);
2193 /*
2194 * Remember our original uid so that we can reset cr_uid before
2195 * the second nfs_namei() call, in case it is remapped.
2196 */
2197 saved_uid = cred->cr_uid;
2198 fromnd.ni_cnd.cn_cred = cred;
2b69e610
MD
2199 fromnd.ni_cnd.cn_nameiop = NAMEI_DELETE;
2200 fromnd.ni_cnd.cn_flags = CNP_WANTPARENT | CNP_SAVESTART;
984263bc 2201 error = nfs_namei(&fromnd, ffhp, len, slp, nam, &md,
dadab5e9 2202 &dpos, &fdirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2203 if (fdirp) {
2204 if (v3) {
3b568787 2205 fdirfor_ret = VOP_GETATTR(fdirp, &fdirfor, td);
984263bc
MD
2206 } else {
2207 vrele(fdirp);
2208 fdirp = NULL;
2209 }
2210 }
2211 if (error) {
2212 nfsm_reply(2 * NFSX_WCCDATA(v3));
2213 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2214 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2215 error = 0;
2216 goto nfsmout;
2217 }
2218 fvp = fromnd.ni_vp;
2219 nfsm_srvmtofh(tfhp);
2220 nfsm_strsiz(len2, NFS_MAXNAMLEN);
2221 cred->cr_uid = saved_uid;
2222 tond.ni_cnd.cn_cred = cred;
2b69e610
MD
2223 tond.ni_cnd.cn_nameiop = NAMEI_RENAME;
2224 tond.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_LOCKLEAF |
2225 CNP_NOCACHE | CNP_SAVESTART;
984263bc 2226 error = nfs_namei(&tond, tfhp, len2, slp, nam, &md,
dadab5e9 2227 &dpos, &tdirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2228 if (tdirp) {
2229 if (v3) {
3b568787 2230 tdirfor_ret = VOP_GETATTR(tdirp, &tdirfor, td);
984263bc
MD
2231 } else {
2232 vrele(tdirp);
2233 tdirp = NULL;
2234 }
2235 }
2236 if (error)
2237 goto out1;
2238
2239 tdvp = tond.ni_dvp;
2240 tvp = tond.ni_vp;
2241 if (tvp != NULL) {
2242 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
2243 if (v3)
2244 error = EEXIST;
2245 else
2246 error = EISDIR;
2247 goto out;
2248 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
2249 if (v3)
2250 error = EEXIST;
2251 else
2252 error = ENOTDIR;
2253 goto out;
2254 }
2255 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
2256 if (v3)
2257 error = EXDEV;
2258 else
2259 error = ENOTEMPTY;
2260 goto out;
2261 }
2262 }
2263 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
2264 if (v3)
2265 error = EXDEV;
2266 else
2267 error = ENOTEMPTY;
2268 goto out;
2269 }
2270 if (fvp->v_mount != tdvp->v_mount) {
2271 if (v3)
2272 error = EXDEV;
2273 else
2274 error = ENOTEMPTY;
2275 goto out;
2276 }
2277 if (fvp == tdvp) {
2278 if (v3)
2279 error = EINVAL;
2280 else
2281 error = ENOTEMPTY;
2282 }
2283 /*
2284 * If source is the same as the destination (that is the
2285 * same vnode with the same name in the same directory),
2286 * then there is nothing to do.
2287 */
2288 if (fvp == tvp && fromnd.ni_dvp == tdvp &&
2289 fromnd.ni_cnd.cn_namelen == tond.ni_cnd.cn_namelen &&
2290 !bcmp(fromnd.ni_cnd.cn_nameptr, tond.ni_cnd.cn_nameptr,
2291 fromnd.ni_cnd.cn_namelen))
2292 error = -1;
2293out:
2294 if (!error) {
2295 /*
2296 * The VOP_RENAME function releases all vnode references &
2297 * locks prior to returning so we need to clear the pointers
2298 * to bypass cleanup code later on.
2299 */
2300 nqsrv_getl(fromnd.ni_dvp, ND_WRITE);
2301 nqsrv_getl(tdvp, ND_WRITE);
2302 if (tvp) {
2303 nqsrv_getl(tvp, ND_WRITE);
2304 }
bc0c094e
MD
2305 error = VOP_RENAME(fromnd.ni_dvp, NCPNULL, fromnd.ni_vp, &fromnd.ni_cnd,
2306 tond.ni_dvp, NCPNULL, tond.ni_vp, &tond.ni_cnd);
984263bc
MD
2307 fromnd.ni_dvp = NULL;
2308 fromnd.ni_vp = NULL;
2309 tond.ni_dvp = NULL;
2310 tond.ni_vp = NULL;
2311 if (error) {
2b69e610
MD
2312 fromnd.ni_cnd.cn_flags &= ~CNP_HASBUF;
2313 tond.ni_cnd.cn_flags &= ~CNP_HASBUF;
984263bc
MD
2314 }
2315 } else {
2316 if (error == -1)
2317 error = 0;
2318 }
2319 /* fall through */
2320
2321out1:
2322 if (fdirp)
3b568787 2323 fdiraft_ret = VOP_GETATTR(fdirp, &fdiraft, td);
984263bc 2324 if (tdirp)
3b568787 2325 tdiraft_ret = VOP_GETATTR(tdirp, &tdiraft, td);
984263bc
MD
2326 nfsm_reply(2 * NFSX_WCCDATA(v3));
2327 if (v3) {
2328 nfsm_srvwcc_data(fdirfor_ret, &fdirfor, fdiraft_ret, &fdiraft);
2329 nfsm_srvwcc_data(tdirfor_ret, &tdirfor, tdiraft_ret, &tdiraft);
2330 }
2331 error = 0;
2332 /* fall through */
2333
2334nfsmout:
2335 /*
2336 * Clear out tond related fields
2337 */
2338 if (tdirp)
2339 vrele(tdirp);
2340 if (tond.ni_startdir)
2341 vrele(tond.ni_startdir);
2342 NDFREE(&tond, NDF_ONLY_PNBUF);
2343 if (tond.ni_dvp) {
2344 if (tond.ni_dvp == tond.ni_vp)
2345 vrele(tond.ni_dvp);
2346 else
2347 vput(tond.ni_dvp);
2348 }
2349 if (tond.ni_vp)
2350 vput(tond.ni_vp);
2351
2352 /*
2353 * Clear out fromnd related fields
2354 */
2355 if (fdirp)
2356 vrele(fdirp);
2357 if (fromnd.ni_startdir)
2358 vrele(fromnd.ni_startdir);
2359 NDFREE(&fromnd, NDF_ONLY_PNBUF);
2360 if (fromnd.ni_dvp)
2361 vrele(fromnd.ni_dvp);
2362 if (fromnd.ni_vp)
2363 vrele(fromnd.ni_vp);
2364
2365 return (error);
2366}
2367
2368/*
2369 * nfs link service
2370 */
2371int
dadab5e9 2372nfsrv_link(nfsd, slp, td, mrq)
984263bc
MD
2373 struct nfsrv_descript *nfsd;
2374 struct nfssvc_sock *slp;
dadab5e9 2375 struct thread *td;
984263bc
MD
2376 struct mbuf **mrq;
2377{
2378 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2379 struct sockaddr *nam = nfsd->nd_nam;
2380 caddr_t dpos = nfsd->nd_dpos;
2381 struct ucred *cred = &nfsd->nd_cr;
2382 struct nameidata nd;
dadab5e9
MD
2383 u_int32_t *tl;
2384 int32_t t1;
984263bc
MD
2385 caddr_t bpos;
2386 int error = 0, rdonly, cache, len, dirfor_ret = 1, diraft_ret = 1;
2387 int getret = 1, v3 = (nfsd->nd_flag & ND_NFSV3);
2388 char *cp2;
2389 struct mbuf *mb, *mreq;
2390 struct vnode *vp = NULL, *xp, *dirp = (struct vnode *)0;
2391 struct vattr dirfor, diraft, at;
2392 nfsfh_t nfh, dnfh;
2393 fhandle_t *fhp, *dfhp;
2394 u_quad_t frev;
2395
2396 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2397 ndclear(&nd);
2398
2399 fhp = &nfh.fh_generic;
2400 dfhp = &dnfh.fh_generic;
2401 nfsm_srvmtofh(fhp);
2402 nfsm_srvmtofh(dfhp);
2403 nfsm_srvnamesiz(len);
2404
2405 error = nfsrv_fhtovp(fhp, FALSE, &vp, cred, slp, nam,
2406 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2407 if (error) {
2408 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2409 nfsm_srvpostop_attr(getret, &at);
2410 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2411 vp = NULL;
2412 error = 0;
2413 goto nfsmout;
2414 }
2415 if (vp->v_type == VDIR) {
2416 error = EPERM; /* POSIX */
2417 goto out1;
2418 }
2419 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
2420 nd.ni_cnd.cn_nameiop = NAMEI_CREATE;
2421 nd.ni_cnd.cn_flags = CNP_LOCKPARENT;
984263bc 2422 error = nfs_namei(&nd, dfhp, len, slp, nam, &md, &dpos,
dadab5e9 2423 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2424 if (dirp) {
2425 if (v3) {
3b568787 2426 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
2427 } else {
2428 vrele(dirp);
2429 dirp = NULL;
2430 }
2431 }
2432 if (error)
2433 goto out1;
2434
2435 xp = nd.ni_vp;
2436 if (xp != NULL) {
2437 error = EEXIST;
2438 goto out;
2439 }
2440 xp = nd.ni_dvp;
2441 if (vp->v_mount != xp->v_mount)
2442 error = EXDEV;
2443out:
2444 if (!error) {
2445 nqsrv_getl(vp, ND_WRITE);
2446 nqsrv_getl(xp, ND_WRITE);
bc0c094e 2447 error = VOP_LINK(nd.ni_dvp, NCPNULL, vp, &nd.ni_cnd);
984263bc
MD
2448 NDFREE(&nd, NDF_ONLY_PNBUF);
2449 }
2450 /* fall through */
2451
2452out1:
2453 if (v3)
3b568787 2454 getret = VOP_GETATTR(vp, &at, td);
984263bc 2455 if (dirp)
3b568787 2456 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2457 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2458 if (v3) {
2459 nfsm_srvpostop_attr(getret, &at);
2460 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2461 error = 0;
2462 }
2463 /* fall through */
2464
2465nfsmout:
2466 NDFREE(&nd, NDF_ONLY_PNBUF);
2467 if (dirp)
2468 vrele(dirp);
2469 if (vp)
2470 vrele(vp);
2471 if (nd.ni_dvp) {
2472 if (nd.ni_dvp == nd.ni_vp)
2473 vrele(nd.ni_dvp);
2474 else
2475 vput(nd.ni_dvp);
2476 }
2477 if (nd.ni_vp)
2478 vrele(nd.ni_vp);
2479 return(error);
2480}
2481
2482/*
2483 * nfs symbolic link service
2484 */
2485int
dadab5e9 2486nfsrv_symlink(nfsd, slp, td, mrq)
984263bc
MD
2487 struct nfsrv_descript *nfsd;
2488 struct nfssvc_sock *slp;
dadab5e9 2489 struct thread *td;
984263bc
MD
2490 struct mbuf **mrq;
2491{
2492 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2493 struct sockaddr *nam = nfsd->nd_nam;
2494 caddr_t dpos = nfsd->nd_dpos;
2495 struct ucred *cred = &nfsd->nd_cr;
2496 struct vattr va, dirfor, diraft;
2497 struct nameidata nd;
dadab5e9
MD
2498 struct vattr *vap = &va;
2499 u_int32_t *tl;
2500 int32_t t1;
984263bc
MD
2501 struct nfsv2_sattr *sp;
2502 char *bpos, *pathcp = (char *)0, *cp2;
2503 struct uio io;
2504 struct iovec iv;
2505 int error = 0, cache, len, len2, dirfor_ret = 1, diraft_ret = 1;
2506 int v3 = (nfsd->nd_flag & ND_NFSV3);
2507 struct mbuf *mb, *mreq, *mb2;
2508 struct vnode *dirp = (struct vnode *)0;
2509 nfsfh_t nfh;
2510 fhandle_t *fhp;
2511 u_quad_t frev;
2512
2513 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2514 ndclear(&nd);
2515
2516 fhp = &nfh.fh_generic;
2517 nfsm_srvmtofh(fhp);
2518 nfsm_srvnamesiz(len);
2519 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
2520 nd.ni_cnd.cn_nameiop = NAMEI_CREATE;
2521 nd.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_SAVESTART;
984263bc 2522 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 2523 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2524 if (dirp) {
2525 if (v3) {
3b568787 2526 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
2527 } else {
2528 vrele(dirp);
2529 dirp = NULL;
2530 }
2531 }
2532 if (error)
2533 goto out;
2534
2535 VATTR_NULL(vap);
2536 if (v3)
2537 nfsm_srvsattr(vap);
2538 nfsm_strsiz(len2, NFS_MAXPATHLEN);
2539 MALLOC(pathcp, caddr_t, len2 + 1, M_TEMP, M_WAITOK);
2540 iv.iov_base = pathcp;
2541 iv.iov_len = len2;
2542 io.uio_resid = len2;
2543 io.uio_offset = 0;
2544 io.uio_iov = &iv;
2545 io.uio_iovcnt = 1;
2546 io.uio_segflg = UIO_SYSSPACE;
2547 io.uio_rw = UIO_READ;
dadab5e9 2548 io.uio_td = NULL;
984263bc
MD
2549 nfsm_mtouio(&io, len2);
2550 if (!v3) {
2551 nfsm_dissect(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2552 vap->va_mode = nfstov_mode(sp->sa_mode);
2553 }
2554 *(pathcp + len2) = '\0';
2555 if (nd.ni_vp) {
2556 error = EEXIST;
2557 goto out;
2558 }
2559
2560 /*
2561 * issue symlink op. SAVESTART is set so the underlying path component
2562 * is only freed by the VOP if an error occurs.
2563 */
2564 if (vap->va_mode == (mode_t)VNOVAL)
2565 vap->va_mode = 0;
2566 nqsrv_getl(nd.ni_dvp, ND_WRITE);
bc0c094e 2567 error = VOP_SYMLINK(nd.ni_dvp, NCPNULL, &nd.ni_vp, &nd.ni_cnd, vap, pathcp);
984263bc
MD
2568 if (error)
2569 NDFREE(&nd, NDF_ONLY_PNBUF);
2570 else
2571 vput(nd.ni_vp);
2572 nd.ni_vp = NULL;
2573 /*
2574 * releases directory prior to potential lookup op.
2575 */
2576 vput(nd.ni_dvp);
2577 nd.ni_dvp = NULL;
2578
2579 if (error == 0) {
2580 if (v3) {
2581 /*
2582 * Issue lookup. Leave SAVESTART set so we can easily free
2583 * the name buffer later on.
2584 *
2585 * since LOCKPARENT is not set, ni_dvp will be garbage on
2586 * return whether an error occurs or not.
2587 */
2b69e610
MD
2588 nd.ni_cnd.cn_nameiop = NAMEI_LOOKUP;
2589 nd.ni_cnd.cn_flags &= ~(CNP_LOCKPARENT | CNP_FOLLOW);
6fc0e645 2590 nd.ni_cnd.cn_flags |= CNP_LOCKLEAF;
dadab5e9 2591 nd.ni_cnd.cn_td = td;
984263bc
MD
2592 nd.ni_cnd.cn_cred = cred;
2593
2594 error = lookup(&nd);
2595 nd.ni_dvp = NULL;
2596
2597 if (error == 0) {
2598 bzero((caddr_t)fhp, sizeof(nfh));
2599 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2600 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2601 if (!error)
3b568787 2602 error = VOP_GETATTR(nd.ni_vp, vap, td);
984263bc
MD
2603 vput(nd.ni_vp);
2604 nd.ni_vp = NULL;
2605 }
2606 }
2607 }
2608out:
2609 /*
2610 * These releases aren't strictly required, does even doing them
2611 * make any sense? XXX can nfsm_reply() block?
2612 */
2613 if (pathcp) {
2614 FREE(pathcp, M_TEMP);
2615 pathcp = NULL;
2616 }
2617 if (dirp) {
3b568787 2618 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2619 vrele(dirp);
2620 dirp = NULL;
2621 }
2622 if (nd.ni_startdir) {
2623 vrele(nd.ni_startdir);
2624 nd.ni_startdir = NULL;
2625 }
2626 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2627 if (v3) {
2628 if (!error) {
2629 nfsm_srvpostop_fh(fhp);
2630 nfsm_srvpostop_attr(0, vap);
2631 }
2632 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2633 }
2634 error = 0;
2635 /* fall through */
2636
2637nfsmout:
2638 NDFREE(&nd, NDF_ONLY_PNBUF);
2639 if (nd.ni_dvp) {
2640 if (nd.ni_dvp == nd.ni_vp)
2641 vrele(nd.ni_dvp);
2642 else
2643 vput(nd.ni_dvp);
2644 }
2645 if (nd.ni_vp)
2646 vrele(nd.ni_vp);
2647 if (nd.ni_startdir)
2648 vrele(nd.ni_startdir);
2649 if (dirp)
2650 vrele(dirp);
2651 if (pathcp)
2652 FREE(pathcp, M_TEMP);
2653
2654 return (error);
2655}
2656
2657/*
2658 * nfs mkdir service
2659 */
2660int
dadab5e9 2661nfsrv_mkdir(nfsd, slp, td, mrq)
984263bc
MD
2662 struct nfsrv_descript *nfsd;
2663 struct nfssvc_sock *slp;
dadab5e9 2664 struct thread *td;
984263bc
MD
2665 struct mbuf **mrq;
2666{
2667 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2668 struct sockaddr *nam = nfsd->nd_nam;
2669 caddr_t dpos = nfsd->nd_dpos;
2670 struct ucred *cred = &nfsd->nd_cr;
2671 struct vattr va, dirfor, diraft;
dadab5e9
MD
2672 struct vattr *vap = &va;
2673 struct nfs_fattr *fp;
984263bc 2674 struct nameidata nd;
dadab5e9
MD
2675 caddr_t cp;
2676 u_int32_t *tl;
2677 int32_t t1;
984263bc
MD
2678 caddr_t bpos;
2679 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2680 int v3 = (nfsd->nd_flag & ND_NFSV3);
2681 char *cp2;
2682 struct mbuf *mb, *mb2, *mreq;
2683 struct vnode *dirp = NULL;
2684 int vpexcl = 0;
2685 nfsfh_t nfh;
2686 fhandle_t *fhp;
2687 u_quad_t frev;
2688
2689 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2690 ndclear(&nd);
2691
2692 fhp = &nfh.fh_generic;
2693 nfsm_srvmtofh(fhp);
2694 nfsm_srvnamesiz(len);
2695 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
2696 nd.ni_cnd.cn_nameiop = NAMEI_CREATE;
2697 nd.ni_cnd.cn_flags = CNP_LOCKPARENT;
984263bc
MD
2698
2699 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 2700 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2701 if (dirp) {
2702 if (v3) {
3b568787 2703 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
2704 } else {
2705 vrele(dirp);
2706 dirp = NULL;
2707 }
2708 }
2709 if (error) {
2710 nfsm_reply(NFSX_WCCDATA(v3));
2711 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2712 error = 0;
2713 goto nfsmout;
2714 }
2715 VATTR_NULL(vap);
2716 if (v3) {
2717 nfsm_srvsattr(vap);
2718 } else {
2719 nfsm_dissect(tl, u_int32_t *, NFSX_UNSIGNED);
2720 vap->va_mode = nfstov_mode(*tl++);
2721 }
2722
2723 /*
2724 * At this point nd.ni_dvp is referenced and exclusively locked and
2725 * nd.ni_vp, if it exists, is referenced but not locked.
2726 */
2727
2728 vap->va_type = VDIR;
2729 if (nd.ni_vp != NULL) {
2730 NDFREE(&nd, NDF_ONLY_PNBUF);
2731 error = EEXIST;
2732 goto out;
2733 }
2734
2735 /*
2736 * Issue mkdir op. Since SAVESTART is not set, the pathname
2737 * component is freed by the VOP call. This will fill-in
2738 * nd.ni_vp, reference, and exclusively lock it.
2739 */
2740 if (vap->va_mode == (mode_t)VNOVAL)
2741 vap->va_mode = 0;
2742 nqsrv_getl(nd.ni_dvp, ND_WRITE);
bc0c094e 2743 error = VOP_MKDIR(nd.ni_dvp, NCPNULL, &nd.ni_vp, &nd.ni_cnd, vap);
984263bc
MD
2744 NDFREE(&nd, NDF_ONLY_PNBUF);
2745 vpexcl = 1;
2746
2747 vput(nd.ni_dvp);
2748 nd.ni_dvp = NULL;
2749
2750 if (!error) {
2751 bzero((caddr_t)fhp, sizeof(nfh));
2752 fhp->fh_fsid = nd.ni_vp->v_mount->mnt_stat.f_fsid;
2753 error = VFS_VPTOFH(nd.ni_vp, &fhp->fh_fid);
2754 if (!error)
3b568787 2755 error = VOP_GETATTR(nd.ni_vp, vap, td);
984263bc
MD
2756 }
2757out:
2758 if (dirp)
3b568787 2759 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2760 nfsm_reply(NFSX_SRVFH(v3) + NFSX_POSTOPATTR(v3) + NFSX_WCCDATA(v3));
2761 if (v3) {
2762 if (!error) {
2763 nfsm_srvpostop_fh(fhp);
2764 nfsm_srvpostop_attr(0, vap);
2765 }
2766 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2767 } else {
2768 nfsm_srvfhtom(fhp, v3);
2769 nfsm_build(fp, struct nfs_fattr *, NFSX_V2FATTR);
2770 nfsm_srvfillattr(vap, fp);
2771 }
2772 error = 0;
2773 /* fall through */
2774
2775nfsmout:
2776 if (dirp)
2777 vrele(dirp);
2778 if (nd.ni_dvp) {
2779 NDFREE(&nd, NDF_ONLY_PNBUF);
2780 if (nd.ni_dvp == nd.ni_vp && vpexcl)
2781 vrele(nd.ni_dvp);
2782 else
2783 vput(nd.ni_dvp);
2784 }
2785 if (nd.ni_vp) {
2786 if (vpexcl)
2787 vput(nd.ni_vp);
2788 else
2789 vrele(nd.ni_vp);
2790 }
2791 return (error);
2792}
2793
2794/*
2795 * nfs rmdir service
2796 */
2797int
dadab5e9 2798nfsrv_rmdir(nfsd, slp, td, mrq)
984263bc
MD
2799 struct nfsrv_descript *nfsd;
2800 struct nfssvc_sock *slp;
dadab5e9 2801 struct thread *td;
984263bc
MD
2802 struct mbuf **mrq;
2803{
2804 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2805 struct sockaddr *nam = nfsd->nd_nam;
2806 caddr_t dpos = nfsd->nd_dpos;
2807 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
2808 u_int32_t *tl;
2809 int32_t t1;
984263bc
MD
2810 caddr_t bpos;
2811 int error = 0, cache, len, dirfor_ret = 1, diraft_ret = 1;
2812 int v3 = (nfsd->nd_flag & ND_NFSV3);
2813 char *cp2;
2814 struct mbuf *mb, *mreq;
2815 struct vnode *vp, *dirp = (struct vnode *)0;
2816 struct vattr dirfor, diraft;
2817 nfsfh_t nfh;
2818 fhandle_t *fhp;
2819 struct nameidata nd;
2820 u_quad_t frev;
2821
2822 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2823 ndclear(&nd);
2824
2825 fhp = &nfh.fh_generic;
2826 nfsm_srvmtofh(fhp);
2827 nfsm_srvnamesiz(len);
2828 nd.ni_cnd.cn_cred = cred;
2b69e610
MD
2829 nd.ni_cnd.cn_nameiop = NAMEI_DELETE;
2830 nd.ni_cnd.cn_flags = CNP_LOCKPARENT | CNP_LOCKLEAF;
984263bc 2831 error = nfs_namei(&nd, fhp, len, slp, nam, &md, &dpos,
dadab5e9 2832 &dirp, td, (nfsd->nd_flag & ND_KERBAUTH), FALSE);
984263bc
MD
2833 if (dirp) {
2834 if (v3) {
3b568787 2835 dirfor_ret = VOP_GETATTR(dirp, &dirfor, td);
984263bc
MD
2836 } else {
2837 vrele(dirp);
2838 dirp = NULL;
2839 }
2840 }
2841 if (error) {
2842 nfsm_reply(NFSX_WCCDATA(v3));
2843 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2844 error = 0;
2845 goto nfsmout;
2846 }
2847 vp = nd.ni_vp;
2848 if (vp->v_type != VDIR) {
2849 error = ENOTDIR;
2850 goto out;
2851 }
2852 /*
2853 * No rmdir "." please.
2854 */
2855 if (nd.ni_dvp == vp) {
2856 error = EINVAL;
2857 goto out;
2858 }
2859 /*
2860 * The root of a mounted filesystem cannot be deleted.
2861 */
2862 if (vp->v_flag & VROOT)
2863 error = EBUSY;
2864out:
2865 /*
2866 * Issue or abort op. Since SAVESTART is not set, path name
2867 * component is freed by the VOP after either.
2868 */
2869 if (!error) {
2870 nqsrv_getl(nd.ni_dvp, ND_WRITE);
2871 nqsrv_getl(vp, ND_WRITE);
bc0c094e 2872 error = VOP_RMDIR(nd.ni_dvp, NCPNULL, nd.ni_vp, &nd.ni_cnd);
984263bc
MD
2873 }
2874 NDFREE(&nd, NDF_ONLY_PNBUF);
2875
2876 if (dirp)
3b568787 2877 diraft_ret = VOP_GETATTR(dirp, &diraft, td);
984263bc
MD
2878 nfsm_reply(NFSX_WCCDATA(v3));
2879 if (v3) {
2880 nfsm_srvwcc_data(dirfor_ret, &dirfor, diraft_ret, &diraft);
2881 error = 0;
2882 }
2883 /* fall through */
2884
2885nfsmout:
2886 NDFREE(&nd, NDF_ONLY_PNBUF);
2887 if (dirp)
2888 vrele(dirp);
2889 if (nd.ni_dvp) {
2890 if (nd.ni_dvp == nd.ni_vp)
2891 vrele(nd.ni_dvp);
2892 else
2893 vput(nd.ni_dvp);
2894 }
2895 if (nd.ni_vp)
2896 vput(nd.ni_vp);
2897
2898 return(error);
2899}
2900
2901/*
2902 * nfs readdir service
2903 * - mallocs what it thinks is enough to read
2904 * count rounded up to a multiple of NFS_DIRBLKSIZ <= NFS_MAXREADDIR
2905 * - calls VOP_READDIR()
2906 * - loops around building the reply
2907 * if the output generated exceeds count break out of loop
2908 * The nfsm_clget macro is used here so that the reply will be packed
2909 * tightly in mbuf clusters.
2910 * - it only knows that it has encountered eof when the VOP_READDIR()
2911 * reads nothing
2912 * - as such one readdir rpc will return eof false although you are there
2913 * and then the next will return eof
2914 * - it trims out records with d_fileno == 0
2915 * this doesn't matter for Unix clients, but they might confuse clients
2916 * for other os'.
2917 * NB: It is tempting to set eof to true if the VOP_READDIR() reads less
2918 * than requested, but this may not apply to all filesystems. For
2919 * example, client NFS does not { although it is never remote mounted
2920 * anyhow }
2921 * The alternate call nfsrv_readdirplus() does lookups as well.
2922 * PS: The NFS protocol spec. does not clarify what the "count" byte
2923 * argument is a count of.. just name strings and file id's or the
2924 * entire reply rpc or ...
2925 * I tried just file name and id sizes and it confused the Sun client,
2926 * so I am using the full rpc size now. The "paranoia.." comment refers
2927 * to including the status longwords that are not a part of the dir.
2928 * "entry" structures, but are in the rpc.
2929 */
2930struct flrep {
2931 nfsuint64 fl_off;
2932 u_int32_t fl_postopok;
2933 u_int32_t fl_fattr[NFSX_V3FATTR / sizeof (u_int32_t)];
2934 u_int32_t fl_fhok;
2935 u_int32_t fl_fhsize;
2936 u_int32_t fl_nfh[NFSX_V3FH / sizeof (u_int32_t)];
2937};
2938
2939int
dadab5e9 2940nfsrv_readdir(nfsd, slp, td, mrq)
984263bc
MD
2941 struct nfsrv_descript *nfsd;
2942 struct nfssvc_sock *slp;
dadab5e9 2943 struct thread *td;
984263bc
MD
2944 struct mbuf **mrq;
2945{
2946 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
2947 struct sockaddr *nam = nfsd->nd_nam;
2948 caddr_t dpos = nfsd->nd_dpos;
2949 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
2950 char *bp, *be;
2951 struct mbuf *mp;
2952 struct dirent *dp;
2953 caddr_t cp;
2954 u_int32_t *tl;
2955 int32_t t1;
984263bc
MD
2956 caddr_t bpos;
2957 struct mbuf *mb, *mb2, *mreq, *mp2;
2958 char *cpos, *cend, *cp2, *rbuf;
2959 struct vnode *vp = NULL;
2960 struct vattr at;
2961 nfsfh_t nfh;
2962 fhandle_t *fhp;
2963 struct uio io;
2964 struct iovec iv;
2965 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
2966 int siz, cnt, fullsiz, eofflag, rdonly, cache, ncookies;
2967 int v3 = (nfsd->nd_flag & ND_NFSV3);
2968 u_quad_t frev, off, toff, verf;
2969 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
2970
2971 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
2972 fhp = &nfh.fh_generic;
2973 nfsm_srvmtofh(fhp);
2974 if (v3) {
2975 nfsm_dissect(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
2976 toff = fxdr_hyper(tl);
2977 tl += 2;
2978 verf = fxdr_hyper(tl);
2979 tl += 2;
2980 } else {
2981 nfsm_dissect(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2982 toff = fxdr_unsigned(u_quad_t, *tl++);
2983 verf = 0; /* shut up gcc */
2984 }
2985 off = toff;
2986 cnt = fxdr_unsigned(int, *tl);
2987 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2988 xfer = NFS_SRVMAXDATA(nfsd);
2989 if (cnt > xfer)
2990 cnt = xfer;
2991 if (siz > xfer)
2992 siz = xfer;
2993 fullsiz = siz;
2994 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
2995 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
2996 if (!error && vp->v_type != VDIR) {
2997 error = ENOTDIR;
2998 vput(vp);
2999 vp = NULL;
3000 }
3001 if (error) {
3002 nfsm_reply(NFSX_UNSIGNED);
3003 nfsm_srvpostop_attr(getret, &at);
3004 error = 0;
3005 goto nfsmout;
3006 }
3007
3008 /*
3009 * Obtain lock on vnode for this section of the code
3010 */
3011
3012 nqsrv_getl(vp, ND_READ);
3013 if (v3) {
3b568787 3014 error = getret = VOP_GETATTR(vp, &at, td);
984263bc
MD
3015#if 0
3016 /*
3017 * XXX This check may be too strict for Solaris 2.5 clients.
3018 */
3019 if (!error && toff && verf && verf != at.va_filerev)
3020 error = NFSERR_BAD_COOKIE;
3021#endif
3022 }
3023 if (!error)
dadab5e9 3024 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
984263bc
MD
3025 if (error) {
3026 vput(vp);
3027 vp = NULL;
3028 nfsm_reply(NFSX_POSTOPATTR(v3));
3029 nfsm_srvpostop_attr(getret, &at);
3030 error = 0;
3031 goto nfsmout;
3032 }
dadab5e9 3033 VOP_UNLOCK(vp, 0, td);
984263bc
MD
3034
3035 /*
3036 * end section. Allocate rbuf and continue
3037 */
3038 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3039again:
3040 iv.iov_base = rbuf;
3041 iv.iov_len = fullsiz;
3042 io.uio_iov = &iv;
3043 io.uio_iovcnt = 1;
3044 io.uio_offset = (off_t)off;
3045 io.uio_resid = fullsiz;
3046 io.uio_segflg = UIO_SYSSPACE;
3047 io.uio_rw = UIO_READ;
dadab5e9 3048 io.uio_td = NULL;
984263bc 3049 eofflag = 0;
dadab5e9 3050 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
3051 if (cookies) {
3052 free((caddr_t)cookies, M_TEMP);
3053 cookies = NULL;
3054 }
3055 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3056 off = (off_t)io.uio_offset;
3057 if (!cookies && !error)
3058 error = NFSERR_PERM;
3059 if (v3) {
3b568787 3060 getret = VOP_GETATTR(vp, &at, td);
984263bc
MD
3061 if (!error)
3062 error = getret;
3063 }
dadab5e9 3064 VOP_UNLOCK(vp, 0, td);
984263bc
MD
3065 if (error) {
3066 vrele(vp);
3067 vp = NULL;
3068 free((caddr_t)rbuf, M_TEMP);
3069 if (cookies)
3070 free((caddr_t)cookies, M_TEMP);
3071 nfsm_reply(NFSX_POSTOPATTR(v3));
3072 nfsm_srvpostop_attr(getret, &at);
3073 error = 0;
3074 goto nfsmout;
3075 }
3076 if (io.uio_resid) {
3077 siz -= io.uio_resid;
3078
3079 /*
3080 * If nothing read, return eof
3081 * rpc reply
3082 */
3083 if (siz == 0) {
3084 vrele(vp);
3085 vp = NULL;
3086 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) +
3087 2 * NFSX_UNSIGNED);
3088 if (v3) {
3089 nfsm_srvpostop_attr(getret, &at);
3090 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3091 txdr_hyper(at.va_filerev, tl);
3092 tl += 2;
3093 } else
3094 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3095 *tl++ = nfs_false;
3096 *tl = nfs_true;
3097 FREE((caddr_t)rbuf, M_TEMP);
3098 FREE((caddr_t)cookies, M_TEMP);
3099 error = 0;
3100 goto nfsmout;
3101 }
3102 }
3103
3104 /*
3105 * Check for degenerate cases of nothing useful read.
3106 * If so go try again
3107 */
3108 cpos = rbuf;
3109 cend = rbuf + siz;
3110 dp = (struct dirent *)cpos;
3111 cookiep = cookies;
3112 /*
3113 * For some reason FreeBSD's ufs_readdir() chooses to back the
3114 * directory offset up to a block boundary, so it is necessary to
3115 * skip over the records that preceed the requested offset. This
3116 * requires the assumption that file offset cookies monotonically
3117 * increase.
3118 */
3119 while (cpos < cend && ncookies > 0 &&
3120 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3121 ((u_quad_t)(*cookiep)) <= toff)) {
3122 cpos += dp->d_reclen;
3123 dp = (struct dirent *)cpos;
3124 cookiep++;
3125 ncookies--;
3126 }
3127 if (cpos >= cend || ncookies == 0) {
3128 toff = off;
3129 siz = fullsiz;
3130 goto again;
3131 }
3132
3133 len = 3 * NFSX_UNSIGNED; /* paranoia, probably can be 0 */
3134 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_COOKIEVERF(v3) + siz);
3135 if (v3) {
3136 nfsm_srvpostop_attr(getret, &at);
3137 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3138 txdr_hyper(at.va_filerev, tl);
3139 }
3140 mp = mp2 = mb;
3141 bp = bpos;
3142 be = bp + M_TRAILINGSPACE(mp);
3143
3144 /* Loop through the records and build reply */
3145 while (cpos < cend && ncookies > 0) {
3146 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3147 nlen = dp->d_namlen;
3148 rem = nfsm_rndup(nlen) - nlen;
3149 len += (4 * NFSX_UNSIGNED + nlen + rem);
3150 if (v3)
3151 len += 2 * NFSX_UNSIGNED;
3152 if (len > cnt) {
3153 eofflag = 0;
3154 break;
3155 }
3156 /*
3157 * Build the directory record xdr from
3158 * the dirent entry.
3159 */
3160 nfsm_clget;
3161 *tl = nfs_true;
3162 bp += NFSX_UNSIGNED;
3163 if (v3) {
3164 nfsm_clget;
3165 *tl = 0;
3166 bp += NFSX_UNSIGNED;
3167 }
3168 nfsm_clget;
3169 *tl = txdr_unsigned(dp->d_fileno);
3170 bp += NFSX_UNSIGNED;
3171 nfsm_clget;
3172 *tl = txdr_unsigned(nlen);
3173 bp += NFSX_UNSIGNED;
3174
3175 /* And loop around copying the name */
3176 xfer = nlen;
3177 cp = dp->d_name;
3178 while (xfer > 0) {
3179 nfsm_clget;
3180 if ((bp+xfer) > be)
3181 tsiz = be-bp;
3182 else
3183 tsiz = xfer;
3184 bcopy(cp, bp, tsiz);
3185 bp += tsiz;
3186 xfer -= tsiz;
3187 if (xfer > 0)
3188 cp += tsiz;
3189 }
3190 /* And null pad to a int32_t boundary */
3191 for (i = 0; i < rem; i++)
3192 *bp++ = '\0';
3193 nfsm_clget;
3194
3195 /* Finish off the record */
3196 if (v3) {
3197 *tl = 0;
3198 bp += NFSX_UNSIGNED;
3199 nfsm_clget;
3200 }
3201 *tl = txdr_unsigned(*cookiep);
3202 bp += NFSX_UNSIGNED;
3203 }
3204 cpos += dp->d_reclen;
3205 dp = (struct dirent *)cpos;
3206 cookiep++;
3207 ncookies--;
3208 }
3209 vrele(vp);
3210 vp = NULL;
3211 nfsm_clget;
3212 *tl = nfs_false;
3213 bp += NFSX_UNSIGNED;
3214 nfsm_clget;
3215 if (eofflag)
3216 *tl = nfs_true;
3217 else
3218 *tl = nfs_false;
3219 bp += NFSX_UNSIGNED;
3220 if (mp != mb) {
3221 if (bp < be)
3222 mp->m_len = bp - mtod(mp, caddr_t);
3223 } else
3224 mp->m_len += bp - bpos;
3225 FREE((caddr_t)rbuf, M_TEMP);
3226 FREE((caddr_t)cookies, M_TEMP);
3227
3228nfsmout:
3229 if (vp)
3230 vrele(vp);
3231 return(error);
3232}
3233
3234int
dadab5e9 3235nfsrv_readdirplus(nfsd, slp, td, mrq)
984263bc
MD
3236 struct nfsrv_descript *nfsd;
3237 struct nfssvc_sock *slp;
dadab5e9 3238 struct thread *td;
984263bc
MD
3239 struct mbuf **mrq;
3240{
3241 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3242 struct sockaddr *nam = nfsd->nd_nam;
3243 caddr_t dpos = nfsd->nd_dpos;
3244 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
3245 char *bp, *be;
3246 struct mbuf *mp;
3247 struct dirent *dp;
3248 caddr_t cp;
3249 u_int32_t *tl;
3250 int32_t t1;
984263bc
MD
3251 caddr_t bpos;
3252 struct mbuf *mb, *mb2, *mreq, *mp2;
3253 char *cpos, *cend, *cp2, *rbuf;
3254 struct vnode *vp = NULL, *nvp;
3255 struct flrep fl;
3256 nfsfh_t nfh;
3257 fhandle_t *fhp, *nfhp = (fhandle_t *)fl.fl_nfh;
3258 struct uio io;
3259 struct iovec iv;
3260 struct vattr va, at, *vap = &va;
3261 struct nfs_fattr *fp;
3262 int len, nlen, rem, xfer, tsiz, i, error = 0, getret = 1;
3263 int siz, cnt, fullsiz, eofflag, rdonly, cache, dirlen, ncookies;
3264 u_quad_t frev, off, toff, verf;
3265 u_long *cookies = NULL, *cookiep; /* needs to be int64_t or off_t */
3266
3267 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3268 fhp = &nfh.fh_generic;
3269 nfsm_srvmtofh(fhp);
3270 nfsm_dissect(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
3271 toff = fxdr_hyper(tl);
3272 tl += 2;
3273 verf = fxdr_hyper(tl);
3274 tl += 2;
3275 siz = fxdr_unsigned(int, *tl++);
3276 cnt = fxdr_unsigned(int, *tl);
3277 off = toff;
3278 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
3279 xfer = NFS_SRVMAXDATA(nfsd);
3280 if (cnt > xfer)
3281 cnt = xfer;
3282 if (siz > xfer)
3283 siz = xfer;
3284 fullsiz = siz;
3285 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3286 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3287 if (!error && vp->v_type != VDIR) {
3288 error = ENOTDIR;
3289 vput(vp);
3290 vp = NULL;
3291 }
3292 if (error) {
3293 nfsm_reply(NFSX_UNSIGNED);
3294 nfsm_srvpostop_attr(getret, &at);
3295 error = 0;
3296 goto nfsmout;
3297 }
3b568787 3298 error = getret = VOP_GETATTR(vp, &at, td);
984263bc
MD
3299#if 0
3300 /*
3301 * XXX This check may be too strict for Solaris 2.5 clients.
3302 */
3303 if (!error && toff && verf && verf != at.va_filerev)
3304 error = NFSERR_BAD_COOKIE;
3305#endif
3306 if (!error) {
3307 nqsrv_getl(vp, ND_READ);
dadab5e9 3308 error = nfsrv_access(vp, VEXEC, cred, rdonly, td, 0);
984263bc
MD
3309 }
3310 if (error) {
3311 vput(vp);
3312 vp = NULL;
3313 nfsm_reply(NFSX_V3POSTOPATTR);
3314 nfsm_srvpostop_attr(getret, &at);
3315 error = 0;
3316 goto nfsmout;
3317 }
dadab5e9 3318 VOP_UNLOCK(vp, 0, td);
984263bc
MD
3319 MALLOC(rbuf, caddr_t, siz, M_TEMP, M_WAITOK);
3320again:
3321 iv.iov_base = rbuf;
3322 iv.iov_len = fullsiz;
3323 io.uio_iov = &iv;
3324 io.uio_iovcnt = 1;
3325 io.uio_offset = (off_t)off;
3326 io.uio_resid = fullsiz;
3327 io.uio_segflg = UIO_SYSSPACE;
3328 io.uio_rw = UIO_READ;
dadab5e9 3329 io.uio_td = NULL;
984263bc 3330 eofflag = 0;
dadab5e9 3331 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
3332 if (cookies) {
3333 free((caddr_t)cookies, M_TEMP);
3334 cookies = NULL;
3335 }
3336 error = VOP_READDIR(vp, &io, cred, &eofflag, &ncookies, &cookies);
3337 off = (u_quad_t)io.uio_offset;
3b568787 3338 getret = VOP_GETATTR(vp, &at, td);
dadab5e9 3339 VOP_UNLOCK(vp, 0, td);
984263bc
MD
3340 if (!cookies && !error)
3341 error = NFSERR_PERM;
3342 if (!error)
3343 error = getret;
3344 if (error) {
3345 vrele(vp);
3346 vp = NULL;
3347 if (cookies)
3348 free((caddr_t)cookies, M_TEMP);
3349 free((caddr_t)rbuf, M_TEMP);
3350 nfsm_reply(NFSX_V3POSTOPATTR);
3351 nfsm_srvpostop_attr(getret, &at);
3352 error = 0;
3353 goto nfsmout;
3354 }
3355 if (io.uio_resid) {
3356 siz -= io.uio_resid;
3357
3358 /*
3359 * If nothing read, return eof
3360 * rpc reply
3361 */
3362 if (siz == 0) {
3363 vrele(vp);
3364 vp = NULL;
3365 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF +
3366 2 * NFSX_UNSIGNED);
3367 nfsm_srvpostop_attr(getret, &at);
3368 nfsm_build(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3369 txdr_hyper(at.va_filerev, tl);
3370 tl += 2;
3371 *tl++ = nfs_false;
3372 *tl = nfs_true;
3373 FREE((caddr_t)cookies, M_TEMP);
3374 FREE((caddr_t)rbuf, M_TEMP);
3375 error = 0;
3376 goto nfsmout;
3377 }
3378 }
3379
3380 /*
3381 * Check for degenerate cases of nothing useful read.
3382 * If so go try again
3383 */
3384 cpos = rbuf;
3385 cend = rbuf + siz;
3386 dp = (struct dirent *)cpos;
3387 cookiep = cookies;
3388 /*
3389 * For some reason FreeBSD's ufs_readdir() chooses to back the
3390 * directory offset up to a block boundary, so it is necessary to
3391 * skip over the records that preceed the requested offset. This
3392 * requires the assumption that file offset cookies monotonically
3393 * increase.
3394 */
3395 while (cpos < cend && ncookies > 0 &&
3396 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
3397 ((u_quad_t)(*cookiep)) <= toff)) {
3398 cpos += dp->d_reclen;
3399 dp = (struct dirent *)cpos;
3400 cookiep++;
3401 ncookies--;
3402 }
3403 if (cpos >= cend || ncookies == 0) {
3404 toff = off;
3405 siz = fullsiz;
3406 goto again;
3407 }
3408
3409 /*
3410 * Probe one of the directory entries to see if the filesystem
3411 * supports VGET.
3412 */
3413 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp) == EOPNOTSUPP) {
3414 error = NFSERR_NOTSUPP;
3415 vrele(vp);
3416 vp = NULL;
3417 free((caddr_t)cookies, M_TEMP);
3418 free((caddr_t)rbuf, M_TEMP);
3419 nfsm_reply(NFSX_V3POSTOPATTR);
3420 nfsm_srvpostop_attr(getret, &at);
3421 error = 0;
3422 goto nfsmout;
3423 }
3424 vput(nvp);
3425 nvp = NULL;
3426
3427 dirlen = len = NFSX_V3POSTOPATTR + NFSX_V3COOKIEVERF + 2 * NFSX_UNSIGNED;
3428 nfsm_reply(cnt);
3429 nfsm_srvpostop_attr(getret, &at);
3430 nfsm_build(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
3431 txdr_hyper(at.va_filerev, tl);
3432 mp = mp2 = mb;
3433 bp = bpos;
3434 be = bp + M_TRAILINGSPACE(mp);
3435
3436 /* Loop through the records and build reply */
3437 while (cpos < cend && ncookies > 0) {
3438 if (dp->d_fileno != 0 && dp->d_type != DT_WHT) {
3439 nlen = dp->d_namlen;
3440 rem = nfsm_rndup(nlen)-nlen;
3441
3442 /*
3443 * For readdir_and_lookup get the vnode using
3444 * the file number.
3445 */
3446 if (VFS_VGET(vp->v_mount, dp->d_fileno, &nvp))
3447 goto invalid;
3448 bzero((caddr_t)nfhp, NFSX_V3FH);
3449 nfhp->fh_fsid =
3450 nvp->v_mount->mnt_stat.f_fsid;
3451 if (VFS_VPTOFH(nvp, &nfhp->fh_fid)) {
3452 vput(nvp);
3453 nvp = NULL;
3454 goto invalid;
3455 }
3b568787 3456 if (VOP_GETATTR(nvp, vap, td)) {
984263bc
MD
3457 vput(nvp);
3458 nvp = NULL;
3459 goto invalid;
3460 }
3461 vput(nvp);
3462 nvp = NULL;
3463
3464 /*
3465 * If either the dircount or maxcount will be
3466 * exceeded, get out now. Both of these lengths
3467 * are calculated conservatively, including all
3468 * XDR overheads.
3469 */
3470 len += (8 * NFSX_UNSIGNED + nlen + rem + NFSX_V3FH +
3471 NFSX_V3POSTOPATTR);
3472 dirlen += (6 * NFSX_UNSIGNED + nlen + rem);
3473 if (len > cnt || dirlen > fullsiz) {
3474 eofflag = 0;
3475 break;
3476 }
3477
3478 /*
3479 * Build the directory record xdr from
3480 * the dirent entry.
3481 */
3482 fp = (struct nfs_fattr *)&fl.fl_fattr;
3483 nfsm_srvfillattr(vap, fp);
3484 fl.fl_fhsize = txdr_unsigned(NFSX_V3FH);
3485 fl.fl_fhok = nfs_true;
3486 fl.fl_postopok = nfs_true;
3487 fl.fl_off.nfsuquad[0] = 0;
3488 fl.fl_off.nfsuquad[1] = txdr_unsigned(*cookiep);
3489
3490 nfsm_clget;
3491 *tl = nfs_true;
3492 bp += NFSX_UNSIGNED;
3493 nfsm_clget;
3494 *tl = 0;
3495 bp += NFSX_UNSIGNED;
3496 nfsm_clget;
3497 *tl = txdr_unsigned(dp->d_fileno);
3498 bp += NFSX_UNSIGNED;
3499 nfsm_clget;
3500 *tl = txdr_unsigned(nlen);
3501 bp += NFSX_UNSIGNED;
3502
3503 /* And loop around copying the name */
3504 xfer = nlen;
3505 cp = dp->d_name;
3506 while (xfer > 0) {
3507 nfsm_clget;
3508 if ((bp + xfer) > be)
3509 tsiz = be - bp;
3510 else
3511 tsiz = xfer;
3512 bcopy(cp, bp, tsiz);
3513 bp += tsiz;
3514 xfer -= tsiz;
3515 if (xfer > 0)
3516 cp += tsiz;
3517 }
3518 /* And null pad to a int32_t boundary */
3519 for (i = 0; i < rem; i++)
3520 *bp++ = '\0';
3521
3522 /*
3523 * Now copy the flrep structure out.
3524 */
3525 xfer = sizeof (struct flrep);
3526 cp = (caddr_t)&fl;
3527 while (xfer > 0) {
3528 nfsm_clget;
3529 if ((bp + xfer) > be)
3530 tsiz = be - bp;
3531 else
3532 tsiz = xfer;
3533 bcopy(cp, bp, tsiz);
3534 bp += tsiz;
3535 xfer -= tsiz;
3536 if (xfer > 0)
3537 cp += tsiz;
3538 }
3539 }
3540invalid:
3541 cpos += dp->d_reclen;
3542 dp = (struct dirent *)cpos;
3543 cookiep++;
3544 ncookies--;
3545 }
3546 vrele(vp);
3547 vp = NULL;
3548 nfsm_clget;
3549 *tl = nfs_false;
3550 bp += NFSX_UNSIGNED;
3551 nfsm_clget;
3552 if (eofflag)
3553 *tl = nfs_true;
3554 else
3555 *tl = nfs_false;
3556 bp += NFSX_UNSIGNED;
3557 if (mp != mb) {
3558 if (bp < be)
3559 mp->m_len = bp - mtod(mp, caddr_t);
3560 } else
3561 mp->m_len += bp - bpos;
3562 FREE((caddr_t)cookies, M_TEMP);
3563 FREE((caddr_t)rbuf, M_TEMP);
3564nfsmout:
3565 if (vp)
3566 vrele(vp);
3567 return(error);
3568}
3569
3570/*
3571 * nfs commit service
3572 */
3573int
dadab5e9 3574nfsrv_commit(nfsd, slp, td, mrq)
984263bc
MD
3575 struct nfsrv_descript *nfsd;
3576 struct nfssvc_sock *slp;
dadab5e9 3577 struct thread *td;
984263bc
MD
3578 struct mbuf **mrq;
3579{
3580 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3581 struct sockaddr *nam = nfsd->nd_nam;
3582 caddr_t dpos = nfsd->nd_dpos;
3583 struct ucred *cred = &nfsd->nd_cr;
3584 struct vattr bfor, aft;
3585 struct vnode *vp = NULL;
3586 nfsfh_t nfh;
3587 fhandle_t *fhp;
dadab5e9
MD
3588 u_int32_t *tl;
3589 int32_t t1;
984263bc
MD
3590 caddr_t bpos;
3591 int error = 0, rdonly, for_ret = 1, aft_ret = 1, cnt, cache;
3592 char *cp2;
3593 struct mbuf *mb, *mb2, *mreq;
3594 u_quad_t frev, off;
3595
3596 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3597#ifndef nolint
3598 cache = 0;
3599#endif
3600 fhp = &nfh.fh_generic;
3601 nfsm_srvmtofh(fhp);
3602 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
3603
3604 /*
3605 * XXX At this time VOP_FSYNC() does not accept offset and byte
3606 * count parameters, so these arguments are useless (someday maybe).
3607 */
3608 off = fxdr_hyper(tl);
3609 tl += 2;
3610 cnt = fxdr_unsigned(int, *tl);
3611 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3612 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3613 if (error) {
3614 nfsm_reply(2 * NFSX_UNSIGNED);
3615 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3616 error = 0;
3617 goto nfsmout;
3618 }
3b568787 3619 for_ret = VOP_GETATTR(vp, &bfor, td);
984263bc
MD
3620
3621 if (cnt > MAX_COMMIT_COUNT) {
3622 /*
3623 * Give up and do the whole thing
3624 */
3625 if (vp->v_object &&
3626 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3627 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
3628 }
3b568787 3629 error = VOP_FSYNC(vp, MNT_WAIT, td);
984263bc
MD
3630 } else {
3631 /*
3632 * Locate and synchronously write any buffers that fall
3633 * into the requested range. Note: we are assuming that
3634 * f_iosize is a power of 2.
3635 */
3636 int iosize = vp->v_mount->mnt_stat.f_iosize;
3637 int iomask = iosize - 1;
3638 int s;
3639 daddr_t lblkno;
3640
3641 /*
3642 * Align to iosize boundry, super-align to page boundry.
3643 */
3644 if (off & iomask) {
3645 cnt += off & iomask;
3646 off &= ~(u_quad_t)iomask;
3647 }
3648 if (off & PAGE_MASK) {
3649 cnt += off & PAGE_MASK;
3650 off &= ~(u_quad_t)PAGE_MASK;
3651 }
3652 lblkno = off / iosize;
3653
3654 if (vp->v_object &&
3655 (vp->v_object->flags & OBJ_MIGHTBEDIRTY)) {
3656 vm_object_page_clean(vp->v_object, off / PAGE_SIZE, (cnt + PAGE_MASK) / PAGE_SIZE, OBJPC_SYNC);
3657 }
3658
3659 s = splbio();
3660 while (cnt > 0) {
3661 struct buf *bp;
3662
3663 /*
3664 * If we have a buffer and it is marked B_DELWRI we
3665 * have to lock and write it. Otherwise the prior
3666 * write is assumed to have already been committed.
3667 */
3668 if ((bp = gbincore(vp, lblkno)) != NULL && (bp->b_flags & B_DELWRI)) {
3669 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) {
3670 BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL);
3671 continue; /* retry */
3672 }
3673 bremfree(bp);
3674 bp->b_flags &= ~B_ASYNC;
3675 VOP_BWRITE(bp->b_vp, bp);
3676 ++nfs_commit_miss;
3677 }
3678 ++nfs_commit_blks;
3679 if (cnt < iosize)
3680 break;
3681 cnt -= iosize;
3682 ++lblkno;
3683 }
3684 splx(s);
3685 }
3686
3b568787 3687 aft_ret = VOP_GETATTR(vp, &aft, td);
984263bc
MD
3688 vput(vp);
3689 vp = NULL;
3690 nfsm_reply(NFSX_V3WCCDATA + NFSX_V3WRITEVERF);
3691 nfsm_srvwcc_data(for_ret, &bfor, aft_ret, &aft);
3692 if (!error) {
3693 nfsm_build(tl, u_int32_t *, NFSX_V3WRITEVERF);
3694 if (nfsver.tv_sec == 0)
3695 nfsver = boottime;
3696 *tl++ = txdr_unsigned(nfsver.tv_sec);
3697 *tl = txdr_unsigned(nfsver.tv_usec);
3698 } else {
3699 error = 0;
3700 }
3701nfsmout:
3702 if (vp)
3703 vput(vp);
3704 return(error);
3705}
3706
3707/*
3708 * nfs statfs service
3709 */
3710int
dadab5e9 3711nfsrv_statfs(nfsd, slp, td, mrq)
984263bc
MD
3712 struct nfsrv_descript *nfsd;
3713 struct nfssvc_sock *slp;
dadab5e9 3714 struct thread *td;
984263bc
MD
3715 struct mbuf **mrq;
3716{
3717 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3718 struct sockaddr *nam = nfsd->nd_nam;
3719 caddr_t dpos = nfsd->nd_dpos;
3720 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
3721 struct statfs *sf;
3722 struct nfs_statfs *sfp;
3723 u_int32_t *tl;
3724 int32_t t1;
984263bc
MD
3725 caddr_t bpos;
3726 int error = 0, rdonly, cache, getret = 1;
3727 int v3 = (nfsd->nd_flag & ND_NFSV3);
3728 char *cp2;
3729 struct mbuf *mb, *mb2, *mreq;
3730 struct vnode *vp = NULL;
3731 struct vattr at;
3732 nfsfh_t nfh;
3733 fhandle_t *fhp;
3734 struct statfs statfs;
3735 u_quad_t frev, tval;
3736
3737 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3738#ifndef nolint
3739 cache = 0;
3740#endif
3741 fhp = &nfh.fh_generic;
3742 nfsm_srvmtofh(fhp);
3743 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3744 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3745 if (error) {
3746 nfsm_reply(NFSX_UNSIGNED);
3747 nfsm_srvpostop_attr(getret, &at);
3748 error = 0;
3749 goto nfsmout;
3750 }
3751 sf = &statfs;
dadab5e9 3752 error = VFS_STATFS(vp->v_mount, sf, td);
3b568787 3753 getret = VOP_GETATTR(vp, &at, td);
984263bc
MD
3754 vput(vp);
3755 vp = NULL;
3756 nfsm_reply(NFSX_POSTOPATTR(v3) + NFSX_STATFS(v3));
3757 if (v3)
3758 nfsm_srvpostop_attr(getret, &at);
3759 if (error) {
3760 error = 0;
3761 goto nfsmout;
3762 }
3763 nfsm_build(sfp, struct nfs_statfs *, NFSX_STATFS(v3));
3764 if (v3) {
3765 tval = (u_quad_t)sf->f_blocks;
3766 tval *= (u_quad_t)sf->f_bsize;
3767 txdr_hyper(tval, &sfp->sf_tbytes);
3768 tval = (u_quad_t)sf->f_bfree;
3769 tval *= (u_quad_t)sf->f_bsize;
3770 txdr_hyper(tval, &sfp->sf_fbytes);
3771 tval = (u_quad_t)sf->f_bavail;
3772 tval *= (u_quad_t)sf->f_bsize;
3773 txdr_hyper(tval, &sfp->sf_abytes);
3774 sfp->sf_tfiles.nfsuquad[0] = 0;
3775 sfp->sf_tfiles.nfsuquad[1] = txdr_unsigned(sf->f_files);
3776 sfp->sf_ffiles.nfsuquad[0] = 0;
3777 sfp->sf_ffiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3778 sfp->sf_afiles.nfsuquad[0] = 0;
3779 sfp->sf_afiles.nfsuquad[1] = txdr_unsigned(sf->f_ffree);
3780 sfp->sf_invarsec = 0;
3781 } else {
3782 sfp->sf_tsize = txdr_unsigned(NFS_MAXDGRAMDATA);
3783 sfp->sf_bsize = txdr_unsigned(sf->f_bsize);
3784 sfp->sf_blocks = txdr_unsigned(sf->f_blocks);
3785 sfp->sf_bfree = txdr_unsigned(sf->f_bfree);
3786 sfp->sf_bavail = txdr_unsigned(sf->f_bavail);
3787 }
3788nfsmout:
3789 if (vp)
3790 vput(vp);
3791 return(error);
3792}
3793
3794/*
3795 * nfs fsinfo service
3796 */
3797int
dadab5e9 3798nfsrv_fsinfo(nfsd, slp, td, mrq)
984263bc
MD
3799 struct nfsrv_descript *nfsd;
3800 struct nfssvc_sock *slp;
dadab5e9 3801 struct thread *td;
984263bc
MD
3802 struct mbuf **mrq;
3803{
3804 struct mbuf *mrep = nfsd->nd_mrep, *md = nfsd->nd_md;
3805 struct sockaddr *nam = nfsd->nd_nam;
3806 caddr_t dpos = nfsd->nd_dpos;
3807 struct ucred *cred = &nfsd->nd_cr;
dadab5e9
MD
3808 u_int32_t *tl;
3809 struct nfsv3_fsinfo *sip;
3810 int32_t t1;
984263bc
MD
3811 caddr_t bpos;
3812 int error = 0, rdonly, cache, getret = 1, pref;
3813 char *cp2;
3814 struct mbuf *mb, *mb2, *mreq;
3815 struct vnode *vp = NULL;
3816 struct vattr at;
3817 nfsfh_t nfh;
3818 fhandle_t *fhp;
3819 u_quad_t frev, maxfsize;
3820 struct statfs sb;
3821
3822 nfsdbprintf(("%s %d\n", __FILE__, __LINE__));
3823#ifndef nolint
3824 cache = 0;
3825#endif
3826 fhp = &nfh.fh_generic;
3827 nfsm_srvmtofh(fhp);
3828 error = nfsrv_fhtovp(fhp, 1, &vp, cred, slp, nam,
3829 &rdonly, (nfsd->nd_flag & ND_KERBAUTH), TRUE);
3830 if (error) {
3831 nfsm_reply(NFSX_UNSIGNED);
3832 nfsm_srvpostop_attr(getret, &at);
3833 error = 0;
3834 goto nfsmout;
3835 }
3836
3837 /* XXX Try to make a guess on the max file size. */
dadab5e9 3838 VFS_STATFS(vp->v_mount, &sb, td);
984263bc
MD
3839 maxfsize = (u_quad_t)0x80000000 * sb.f_bsize - 1;
3840
3b568787 3841 getret = VOP_GETATTR(vp, &at, td);
984263bc
MD
3842 vput(vp);
3843 vp = NULL;
3844 nfsm_reply(NFSX_V3POSTOPATTR + NFSX_V3FSINFO);
3845 nfsm_srvpostop_attr(getret, &at);
3846 nfsm_build(sip, struct nfsv3_fsinfo *, NFSX_V3FSINFO);
3847
3848 /*
3849 * XXX
3850 * There should be file system VFS OP(s) to get this information.
3851 * For now, assume ufs.
3852 */
3853 if (slp->ns_so->so_type == SOCK_DGRAM)
3854 pref = NFS_MAXDGRAMDATA;
3855 else
3856 pref = NFS_MAXDATA;
3857 sip->fs_rtmax = txdr_unsigned(NFS_MAXDATA);
3858 sip->fs_rtpref = txdr_unsigned(pref);
3859 sip->fs_rtmult = txdr_unsigned(NFS_FABLKSIZE);
3860 sip->fs_wtmax = txdr_unsigned(NFS_MAXDATA);
3861 sip->fs_wtpref = txdr_unsigned(pref);
3862 sip->fs_wtmult = txdr_unsigned(NFS_FABLKSIZE);
3863 sip->fs_dtpref = txdr_unsigned(pref);
3864 txdr_hyper(maxfsize, &sip->fs_maxfilesize);
3865 sip->fs_timedelta.nfsv3_sec = 0;
3866 sip->fs_timedelta.nfsv3_nsec = txdr_unsigned(1);
3867 sip->fs_properties = txdr_unsigned(NFSV3FSINFO_LINK |
3868 NFSV3FSINFO_SYMLINK | NFSV3FSINFO_HOMOGENEOUS |
3869 NFSV3FSINFO_CANSETTIME);
3870nfsmout:
3871 if (vp)
3872 vput(vp);
3873 return(error);
3874}
3875
3876/*
3877 * nfs pathconf service
3878 */
3879int
dadab5e9 3880nfsrv_pathconf(nfsd, slp, td, mrq)
984263bc
MD
3881 struct nfsrv_descript *nfsd;
3882 struct nfssvc_sock *slp;