splice: fix deadlock in splicing to file
[linux-2.6/mini2440.git] / fs / nfsd / nfsfh.c
blob9f1ca17293d3e0a54a42234643f250aa22c6e062
1 /*
2 * linux/fs/nfsd/nfsfh.c
4 * NFS server file handle treatment.
6 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
7 * Portions Copyright (C) 1999 G. Allen Morris III <gam3@acm.org>
8 * Extensive rewrite by Neil Brown <neilb@cse.unsw.edu.au> Southern-Spring 1999
9 * ... and again Southern-Winter 2001 to support export_operations
12 #include <linux/slab.h>
13 #include <linux/fs.h>
14 #include <linux/unistd.h>
15 #include <linux/string.h>
16 #include <linux/stat.h>
17 #include <linux/dcache.h>
18 #include <linux/exportfs.h>
19 #include <linux/mount.h>
21 #include <linux/sunrpc/clnt.h>
22 #include <linux/sunrpc/svc.h>
23 #include <linux/sunrpc/svcauth_gss.h>
24 #include <linux/nfsd/nfsd.h>
25 #include "auth.h"
27 #define NFSDDBG_FACILITY NFSDDBG_FH
30 static int nfsd_nr_verified;
31 static int nfsd_nr_put;
34 * our acceptability function.
35 * if NOSUBTREECHECK, accept anything
36 * if not, require that we can walk up to exp->ex_dentry
37 * doing some checks on the 'x' bits
39 static int nfsd_acceptable(void *expv, struct dentry *dentry)
41 struct svc_export *exp = expv;
42 int rv;
43 struct dentry *tdentry;
44 struct dentry *parent;
46 if (exp->ex_flags & NFSEXP_NOSUBTREECHECK)
47 return 1;
49 tdentry = dget(dentry);
50 while (tdentry != exp->ex_path.dentry && !IS_ROOT(tdentry)) {
51 /* make sure parents give x permission to user */
52 int err;
53 parent = dget_parent(tdentry);
54 err = inode_permission(parent->d_inode, MAY_EXEC);
55 if (err < 0) {
56 dput(parent);
57 break;
59 dput(tdentry);
60 tdentry = parent;
62 if (tdentry != exp->ex_path.dentry)
63 dprintk("nfsd_acceptable failed at %p %s\n", tdentry, tdentry->d_name.name);
64 rv = (tdentry == exp->ex_path.dentry);
65 dput(tdentry);
66 return rv;
69 /* Type check. The correct error return for type mismatches does not seem to be
70 * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
71 * comment in the NFSv3 spec says this is incorrect (implementation notes for
72 * the write call).
74 static inline __be32
75 nfsd_mode_check(struct svc_rqst *rqstp, umode_t mode, int type)
77 /* Type can be negative when creating hardlinks - not to a dir */
78 if (type > 0 && (mode & S_IFMT) != type) {
79 if (rqstp->rq_vers == 4 && (mode & S_IFMT) == S_IFLNK)
80 return nfserr_symlink;
81 else if (type == S_IFDIR)
82 return nfserr_notdir;
83 else if ((mode & S_IFMT) == S_IFDIR)
84 return nfserr_isdir;
85 else
86 return nfserr_inval;
88 if (type < 0 && (mode & S_IFMT) == -type) {
89 if (rqstp->rq_vers == 4 && (mode & S_IFMT) == S_IFLNK)
90 return nfserr_symlink;
91 else if (type == -S_IFDIR)
92 return nfserr_isdir;
93 else
94 return nfserr_notdir;
96 return 0;
99 static __be32 nfsd_setuser_and_check_port(struct svc_rqst *rqstp,
100 struct svc_export *exp)
102 /* Check if the request originated from a secure port. */
103 if (!rqstp->rq_secure && EX_SECURE(exp)) {
104 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
105 dprintk(KERN_WARNING
106 "nfsd: request from insecure port %s!\n",
107 svc_print_addr(rqstp, buf, sizeof(buf)));
108 return nfserr_perm;
111 /* Set user creds for this exportpoint */
112 return nfserrno(nfsd_setuser(rqstp, exp));
116 * Use the given filehandle to look up the corresponding export and
117 * dentry. On success, the results are used to set fh_export and
118 * fh_dentry.
120 static __be32 nfsd_set_fh_dentry(struct svc_rqst *rqstp, struct svc_fh *fhp)
122 struct knfsd_fh *fh = &fhp->fh_handle;
123 struct fid *fid = NULL, sfid;
124 struct svc_export *exp;
125 struct dentry *dentry;
126 int fileid_type;
127 int data_left = fh->fh_size/4;
128 __be32 error;
130 error = nfserr_stale;
131 if (rqstp->rq_vers > 2)
132 error = nfserr_badhandle;
133 if (rqstp->rq_vers == 4 && fh->fh_size == 0)
134 return nfserr_nofilehandle;
136 if (fh->fh_version == 1) {
137 int len;
139 if (--data_left < 0)
140 return error;
141 if (fh->fh_auth_type != 0)
142 return error;
143 len = key_len(fh->fh_fsid_type) / 4;
144 if (len == 0)
145 return error;
146 if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
147 /* deprecated, convert to type 3 */
148 len = key_len(FSID_ENCODE_DEV)/4;
149 fh->fh_fsid_type = FSID_ENCODE_DEV;
150 fh->fh_fsid[0] = new_encode_dev(MKDEV(ntohl(fh->fh_fsid[0]), ntohl(fh->fh_fsid[1])));
151 fh->fh_fsid[1] = fh->fh_fsid[2];
153 data_left -= len;
154 if (data_left < 0)
155 return error;
156 exp = rqst_exp_find(rqstp, fh->fh_fsid_type, fh->fh_auth);
157 fid = (struct fid *)(fh->fh_auth + len);
158 } else {
159 __u32 tfh[2];
160 dev_t xdev;
161 ino_t xino;
163 if (fh->fh_size != NFS_FHSIZE)
164 return error;
165 /* assume old filehandle format */
166 xdev = old_decode_dev(fh->ofh_xdev);
167 xino = u32_to_ino_t(fh->ofh_xino);
168 mk_fsid(FSID_DEV, tfh, xdev, xino, 0, NULL);
169 exp = rqst_exp_find(rqstp, FSID_DEV, tfh);
172 error = nfserr_stale;
173 if (PTR_ERR(exp) == -ENOENT)
174 return error;
176 if (IS_ERR(exp))
177 return nfserrno(PTR_ERR(exp));
179 if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
180 /* Elevate privileges so that the lack of 'r' or 'x'
181 * permission on some parent directory will
182 * not stop exportfs_decode_fh from being able
183 * to reconnect a directory into the dentry cache.
184 * The same problem can affect "SUBTREECHECK" exports,
185 * but as nfsd_acceptable depends on correct
186 * access control settings being in effect, we cannot
187 * fix that case easily.
189 struct cred *new = prepare_creds();
190 if (!new)
191 return nfserrno(-ENOMEM);
192 new->cap_effective =
193 cap_raise_nfsd_set(new->cap_effective,
194 new->cap_permitted);
195 put_cred(override_creds(new));
196 put_cred(new);
197 } else {
198 error = nfsd_setuser_and_check_port(rqstp, exp);
199 if (error)
200 goto out;
204 * Look up the dentry using the NFS file handle.
206 error = nfserr_stale;
207 if (rqstp->rq_vers > 2)
208 error = nfserr_badhandle;
210 if (fh->fh_version != 1) {
211 sfid.i32.ino = fh->ofh_ino;
212 sfid.i32.gen = fh->ofh_generation;
213 sfid.i32.parent_ino = fh->ofh_dirino;
214 fid = &sfid;
215 data_left = 3;
216 if (fh->ofh_dirino == 0)
217 fileid_type = FILEID_INO32_GEN;
218 else
219 fileid_type = FILEID_INO32_GEN_PARENT;
220 } else
221 fileid_type = fh->fh_fileid_type;
223 if (fileid_type == FILEID_ROOT)
224 dentry = dget(exp->ex_path.dentry);
225 else {
226 dentry = exportfs_decode_fh(exp->ex_path.mnt, fid,
227 data_left, fileid_type,
228 nfsd_acceptable, exp);
230 if (dentry == NULL)
231 goto out;
232 if (IS_ERR(dentry)) {
233 if (PTR_ERR(dentry) != -EINVAL)
234 error = nfserrno(PTR_ERR(dentry));
235 goto out;
238 if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
239 error = nfsd_setuser_and_check_port(rqstp, exp);
240 if (error) {
241 dput(dentry);
242 goto out;
246 if (S_ISDIR(dentry->d_inode->i_mode) &&
247 (dentry->d_flags & DCACHE_DISCONNECTED)) {
248 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %s/%s\n",
249 dentry->d_parent->d_name.name, dentry->d_name.name);
252 fhp->fh_dentry = dentry;
253 fhp->fh_export = exp;
254 nfsd_nr_verified++;
255 return 0;
256 out:
257 exp_put(exp);
258 return error;
262 * fh_verify - filehandle lookup and access checking
263 * @rqstp: pointer to current rpc request
264 * @fhp: filehandle to be verified
265 * @type: expected type of object pointed to by filehandle
266 * @access: type of access needed to object
268 * Look up a dentry from the on-the-wire filehandle, check the client's
269 * access to the export, and set the current task's credentials.
271 * Regardless of success or failure of fh_verify(), fh_put() should be
272 * called on @fhp when the caller is finished with the filehandle.
274 * fh_verify() may be called multiple times on a given filehandle, for
275 * example, when processing an NFSv4 compound. The first call will look
276 * up a dentry using the on-the-wire filehandle. Subsequent calls will
277 * skip the lookup and just perform the other checks and possibly change
278 * the current task's credentials.
280 * @type specifies the type of object expected using one of the S_IF*
281 * constants defined in include/linux/stat.h. The caller may use zero
282 * to indicate that it doesn't care, or a negative integer to indicate
283 * that it expects something not of the given type.
285 * @access is formed from the NFSD_MAY_* constants defined in
286 * include/linux/nfsd/nfsd.h.
288 __be32
289 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, int type, int access)
291 struct svc_export *exp;
292 struct dentry *dentry;
293 __be32 error;
295 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp));
297 if (!fhp->fh_dentry) {
298 error = nfsd_set_fh_dentry(rqstp, fhp);
299 if (error)
300 goto out;
301 dentry = fhp->fh_dentry;
302 exp = fhp->fh_export;
303 } else {
305 * just rechecking permissions
306 * (e.g. nfsproc_create calls fh_verify, then nfsd_create
307 * does as well)
309 dprintk("nfsd: fh_verify - just checking\n");
310 dentry = fhp->fh_dentry;
311 exp = fhp->fh_export;
313 * Set user creds for this exportpoint; necessary even
314 * in the "just checking" case because this may be a
315 * filehandle that was created by fh_compose, and that
316 * is about to be used in another nfsv4 compound
317 * operation.
319 error = nfsd_setuser_and_check_port(rqstp, exp);
320 if (error)
321 goto out;
324 error = nfsd_mode_check(rqstp, dentry->d_inode->i_mode, type);
325 if (error)
326 goto out;
329 * pseudoflavor restrictions are not enforced on NLM,
330 * which clients virtually always use auth_sys for,
331 * even while using RPCSEC_GSS for NFS.
333 if (access & NFSD_MAY_LOCK)
334 goto skip_pseudoflavor_check;
336 * Clients may expect to be able to use auth_sys during mount,
337 * even if they use gss for everything else; see section 2.3.2
338 * of rfc 2623.
340 if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
341 && exp->ex_path.dentry == dentry)
342 goto skip_pseudoflavor_check;
344 error = check_nfsd_access(exp, rqstp);
345 if (error)
346 goto out;
348 skip_pseudoflavor_check:
349 /* Finally, check access permissions. */
350 error = nfsd_permission(rqstp, exp, dentry, access);
352 if (error) {
353 dprintk("fh_verify: %s/%s permission failure, "
354 "acc=%x, error=%d\n",
355 dentry->d_parent->d_name.name,
356 dentry->d_name.name,
357 access, ntohl(error));
359 out:
360 if (error == nfserr_stale)
361 nfsdstats.fh_stale++;
362 return error;
367 * Compose a file handle for an NFS reply.
369 * Note that when first composed, the dentry may not yet have
370 * an inode. In this case a call to fh_update should be made
371 * before the fh goes out on the wire ...
373 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
374 struct dentry *dentry)
376 if (dentry != exp->ex_path.dentry) {
377 struct fid *fid = (struct fid *)
378 (fhp->fh_handle.fh_auth + fhp->fh_handle.fh_size/4 - 1);
379 int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
380 int subtreecheck = !(exp->ex_flags & NFSEXP_NOSUBTREECHECK);
382 fhp->fh_handle.fh_fileid_type =
383 exportfs_encode_fh(dentry, fid, &maxsize, subtreecheck);
384 fhp->fh_handle.fh_size += maxsize * 4;
385 } else {
386 fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
391 * for composing old style file handles
393 static inline void _fh_update_old(struct dentry *dentry,
394 struct svc_export *exp,
395 struct knfsd_fh *fh)
397 fh->ofh_ino = ino_t_to_u32(dentry->d_inode->i_ino);
398 fh->ofh_generation = dentry->d_inode->i_generation;
399 if (S_ISDIR(dentry->d_inode->i_mode) ||
400 (exp->ex_flags & NFSEXP_NOSUBTREECHECK))
401 fh->ofh_dirino = 0;
404 __be32
405 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
406 struct svc_fh *ref_fh)
408 /* ref_fh is a reference file handle.
409 * if it is non-null and for the same filesystem, then we should compose
410 * a filehandle which is of the same version, where possible.
411 * Currently, that means that if ref_fh->fh_handle.fh_version == 0xca
412 * Then create a 32byte filehandle using nfs_fhbase_old
416 u8 version;
417 u8 fsid_type = 0;
418 struct inode * inode = dentry->d_inode;
419 struct dentry *parent = dentry->d_parent;
420 __u32 *datap;
421 dev_t ex_dev = exp->ex_path.dentry->d_inode->i_sb->s_dev;
422 int root_export = (exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root);
424 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %s/%s, ino=%ld)\n",
425 MAJOR(ex_dev), MINOR(ex_dev),
426 (long) exp->ex_path.dentry->d_inode->i_ino,
427 parent->d_name.name, dentry->d_name.name,
428 (inode ? inode->i_ino : 0));
430 /* Choose filehandle version and fsid type based on
431 * the reference filehandle (if it is in the same export)
432 * or the export options.
434 retry:
435 version = 1;
436 if (ref_fh && ref_fh->fh_export == exp) {
437 version = ref_fh->fh_handle.fh_version;
438 fsid_type = ref_fh->fh_handle.fh_fsid_type;
440 if (ref_fh == fhp)
441 fh_put(ref_fh);
442 ref_fh = NULL;
444 switch (version) {
445 case 0xca:
446 fsid_type = FSID_DEV;
447 break;
448 case 1:
449 break;
450 default:
451 goto retry;
454 /* Need to check that this type works for this
455 * export point. As the fsid -> filesystem mapping
456 * was guided by user-space, there is no guarantee
457 * that the filesystem actually supports that fsid
458 * type. If it doesn't we loop around again without
459 * ref_fh set.
461 switch(fsid_type) {
462 case FSID_DEV:
463 if (!old_valid_dev(ex_dev))
464 goto retry;
465 /* FALL THROUGH */
466 case FSID_MAJOR_MINOR:
467 case FSID_ENCODE_DEV:
468 if (!(exp->ex_path.dentry->d_inode->i_sb->s_type->fs_flags
469 & FS_REQUIRES_DEV))
470 goto retry;
471 break;
472 case FSID_NUM:
473 if (! (exp->ex_flags & NFSEXP_FSID))
474 goto retry;
475 break;
476 case FSID_UUID8:
477 case FSID_UUID16:
478 if (!root_export)
479 goto retry;
480 /* fall through */
481 case FSID_UUID4_INUM:
482 case FSID_UUID16_INUM:
483 if (exp->ex_uuid == NULL)
484 goto retry;
485 break;
487 } else if (exp->ex_flags & NFSEXP_FSID) {
488 fsid_type = FSID_NUM;
489 } else if (exp->ex_uuid) {
490 if (fhp->fh_maxsize >= 64) {
491 if (root_export)
492 fsid_type = FSID_UUID16;
493 else
494 fsid_type = FSID_UUID16_INUM;
495 } else {
496 if (root_export)
497 fsid_type = FSID_UUID8;
498 else
499 fsid_type = FSID_UUID4_INUM;
501 } else if (!old_valid_dev(ex_dev))
502 /* for newer device numbers, we must use a newer fsid format */
503 fsid_type = FSID_ENCODE_DEV;
504 else
505 fsid_type = FSID_DEV;
507 if (ref_fh == fhp)
508 fh_put(ref_fh);
510 if (fhp->fh_locked || fhp->fh_dentry) {
511 printk(KERN_ERR "fh_compose: fh %s/%s not initialized!\n",
512 parent->d_name.name, dentry->d_name.name);
514 if (fhp->fh_maxsize < NFS_FHSIZE)
515 printk(KERN_ERR "fh_compose: called with maxsize %d! %s/%s\n",
516 fhp->fh_maxsize,
517 parent->d_name.name, dentry->d_name.name);
519 fhp->fh_dentry = dget(dentry); /* our internal copy */
520 fhp->fh_export = exp;
521 cache_get(&exp->h);
523 if (version == 0xca) {
524 /* old style filehandle please */
525 memset(&fhp->fh_handle.fh_base, 0, NFS_FHSIZE);
526 fhp->fh_handle.fh_size = NFS_FHSIZE;
527 fhp->fh_handle.ofh_dcookie = 0xfeebbaca;
528 fhp->fh_handle.ofh_dev = old_encode_dev(ex_dev);
529 fhp->fh_handle.ofh_xdev = fhp->fh_handle.ofh_dev;
530 fhp->fh_handle.ofh_xino =
531 ino_t_to_u32(exp->ex_path.dentry->d_inode->i_ino);
532 fhp->fh_handle.ofh_dirino = ino_t_to_u32(parent_ino(dentry));
533 if (inode)
534 _fh_update_old(dentry, exp, &fhp->fh_handle);
535 } else {
536 int len;
537 fhp->fh_handle.fh_version = 1;
538 fhp->fh_handle.fh_auth_type = 0;
539 datap = fhp->fh_handle.fh_auth+0;
540 fhp->fh_handle.fh_fsid_type = fsid_type;
541 mk_fsid(fsid_type, datap, ex_dev,
542 exp->ex_path.dentry->d_inode->i_ino,
543 exp->ex_fsid, exp->ex_uuid);
545 len = key_len(fsid_type);
546 datap += len/4;
547 fhp->fh_handle.fh_size = 4 + len;
549 if (inode)
550 _fh_update(fhp, exp, dentry);
551 if (fhp->fh_handle.fh_fileid_type == 255)
552 return nfserr_opnotsupp;
555 nfsd_nr_verified++;
556 return 0;
560 * Update file handle information after changing a dentry.
561 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
563 __be32
564 fh_update(struct svc_fh *fhp)
566 struct dentry *dentry;
568 if (!fhp->fh_dentry)
569 goto out_bad;
571 dentry = fhp->fh_dentry;
572 if (!dentry->d_inode)
573 goto out_negative;
574 if (fhp->fh_handle.fh_version != 1) {
575 _fh_update_old(dentry, fhp->fh_export, &fhp->fh_handle);
576 } else {
577 if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
578 goto out;
580 _fh_update(fhp, fhp->fh_export, dentry);
581 if (fhp->fh_handle.fh_fileid_type == 255)
582 return nfserr_opnotsupp;
584 out:
585 return 0;
587 out_bad:
588 printk(KERN_ERR "fh_update: fh not verified!\n");
589 goto out;
590 out_negative:
591 printk(KERN_ERR "fh_update: %s/%s still negative!\n",
592 dentry->d_parent->d_name.name, dentry->d_name.name);
593 goto out;
597 * Release a file handle.
599 void
600 fh_put(struct svc_fh *fhp)
602 struct dentry * dentry = fhp->fh_dentry;
603 struct svc_export * exp = fhp->fh_export;
604 if (dentry) {
605 fh_unlock(fhp);
606 fhp->fh_dentry = NULL;
607 dput(dentry);
608 #ifdef CONFIG_NFSD_V3
609 fhp->fh_pre_saved = 0;
610 fhp->fh_post_saved = 0;
611 #endif
612 nfsd_nr_put++;
614 if (exp) {
615 cache_put(&exp->h, &svc_export_cache);
616 fhp->fh_export = NULL;
618 return;
622 * Shorthand for dprintk()'s
624 char * SVCFH_fmt(struct svc_fh *fhp)
626 struct knfsd_fh *fh = &fhp->fh_handle;
628 static char buf[80];
629 sprintf(buf, "%d: %08x %08x %08x %08x %08x %08x",
630 fh->fh_size,
631 fh->fh_base.fh_pad[0],
632 fh->fh_base.fh_pad[1],
633 fh->fh_base.fh_pad[2],
634 fh->fh_base.fh_pad[3],
635 fh->fh_base.fh_pad[4],
636 fh->fh_base.fh_pad[5]);
637 return buf;
640 enum fsid_source fsid_source(struct svc_fh *fhp)
642 if (fhp->fh_handle.fh_version != 1)
643 return FSIDSOURCE_DEV;
644 switch(fhp->fh_handle.fh_fsid_type) {
645 case FSID_DEV:
646 case FSID_ENCODE_DEV:
647 case FSID_MAJOR_MINOR:
648 if (fhp->fh_export->ex_path.dentry->d_inode->i_sb->s_type->fs_flags
649 & FS_REQUIRES_DEV)
650 return FSIDSOURCE_DEV;
651 break;
652 case FSID_NUM:
653 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
654 return FSIDSOURCE_FSID;
655 break;
656 default:
657 break;
659 /* either a UUID type filehandle, or the filehandle doesn't
660 * match the export.
662 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
663 return FSIDSOURCE_FSID;
664 if (fhp->fh_export->ex_uuid)
665 return FSIDSOURCE_UUID;
666 return FSIDSOURCE_DEV;