2 * NFS server file handle treatment.
4 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
5 * Portions Copyright (C) 1999 G. Allen Morris III <gam3@acm.org>
6 * Extensive rewrite by Neil Brown <neilb@cse.unsw.edu.au> Southern-Spring 1999
7 * ... and again Southern-Winter 2001 to support export_operations
10 #include <linux/exportfs.h>
12 #include <linux/sunrpc/svcauth_gss.h>
17 #define NFSDDBG_FACILITY NFSDDBG_FH
21 * our acceptability function.
22 * if NOSUBTREECHECK, accept anything
23 * if not, require that we can walk up to exp->ex_dentry
24 * doing some checks on the 'x' bits
26 static int nfsd_acceptable(void *expv
, struct dentry
*dentry
)
28 struct svc_export
*exp
= expv
;
30 struct dentry
*tdentry
;
31 struct dentry
*parent
;
33 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
)
36 tdentry
= dget(dentry
);
37 while (tdentry
!= exp
->ex_path
.dentry
&& !IS_ROOT(tdentry
)) {
38 /* make sure parents give x permission to user */
40 parent
= dget_parent(tdentry
);
41 err
= inode_permission(parent
->d_inode
, MAY_EXEC
);
49 if (tdentry
!= exp
->ex_path
.dentry
)
50 dprintk("nfsd_acceptable failed at %p %s\n", tdentry
, tdentry
->d_name
.name
);
51 rv
= (tdentry
== exp
->ex_path
.dentry
);
56 /* Type check. The correct error return for type mismatches does not seem to be
57 * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
58 * comment in the NFSv3 spec says this is incorrect (implementation notes for
62 nfsd_mode_check(struct svc_rqst
*rqstp
, umode_t mode
, int type
)
64 /* Type can be negative when creating hardlinks - not to a dir */
65 if (type
> 0 && (mode
& S_IFMT
) != type
) {
66 if (rqstp
->rq_vers
== 4 && (mode
& S_IFMT
) == S_IFLNK
)
67 return nfserr_symlink
;
68 else if (type
== S_IFDIR
)
70 else if ((mode
& S_IFMT
) == S_IFDIR
)
75 if (type
< 0 && (mode
& S_IFMT
) == -type
) {
76 if (rqstp
->rq_vers
== 4 && (mode
& S_IFMT
) == S_IFLNK
)
77 return nfserr_symlink
;
78 else if (type
== -S_IFDIR
)
86 static __be32
nfsd_setuser_and_check_port(struct svc_rqst
*rqstp
,
87 struct svc_export
*exp
)
89 int flags
= nfsexp_flags(rqstp
, exp
);
91 /* Check if the request originated from a secure port. */
92 if (!rqstp
->rq_secure
&& (flags
& NFSEXP_INSECURE_PORT
)) {
93 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
95 "nfsd: request from insecure port %s!\n",
96 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
100 /* Set user creds for this exportpoint */
101 return nfserrno(nfsd_setuser(rqstp
, exp
));
104 static inline __be32
check_pseudo_root(struct svc_rqst
*rqstp
,
105 struct dentry
*dentry
, struct svc_export
*exp
)
107 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
110 * v2/v3 clients have no need for the V4ROOT export--they use
111 * the mount protocl instead; also, further V4ROOT checks may be
112 * in v4-specific code, in which case v2/v3 clients could bypass
115 if (!nfsd_v4client(rqstp
))
118 * We're exposing only the directories and symlinks that have to be
119 * traversed on the way to real exports:
121 if (unlikely(!S_ISDIR(dentry
->d_inode
->i_mode
) &&
122 !S_ISLNK(dentry
->d_inode
->i_mode
)))
125 * A pseudoroot export gives permission to access only one
126 * single directory; the kernel has to make another upcall
127 * before granting access to anything else under it:
129 if (unlikely(dentry
!= exp
->ex_path
.dentry
))
135 * Use the given filehandle to look up the corresponding export and
136 * dentry. On success, the results are used to set fh_export and
139 static __be32
nfsd_set_fh_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
)
141 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
142 struct fid
*fid
= NULL
, sfid
;
143 struct svc_export
*exp
;
144 struct dentry
*dentry
;
146 int data_left
= fh
->fh_size
/4;
149 error
= nfserr_stale
;
150 if (rqstp
->rq_vers
> 2)
151 error
= nfserr_badhandle
;
152 if (rqstp
->rq_vers
== 4 && fh
->fh_size
== 0)
153 return nfserr_nofilehandle
;
155 if (fh
->fh_version
== 1) {
160 if (fh
->fh_auth_type
!= 0)
162 len
= key_len(fh
->fh_fsid_type
) / 4;
165 if (fh
->fh_fsid_type
== FSID_MAJOR_MINOR
) {
166 /* deprecated, convert to type 3 */
167 len
= key_len(FSID_ENCODE_DEV
)/4;
168 fh
->fh_fsid_type
= FSID_ENCODE_DEV
;
169 fh
->fh_fsid
[0] = new_encode_dev(MKDEV(ntohl(fh
->fh_fsid
[0]), ntohl(fh
->fh_fsid
[1])));
170 fh
->fh_fsid
[1] = fh
->fh_fsid
[2];
175 exp
= rqst_exp_find(rqstp
, fh
->fh_fsid_type
, fh
->fh_auth
);
176 fid
= (struct fid
*)(fh
->fh_auth
+ len
);
182 if (fh
->fh_size
!= NFS_FHSIZE
)
184 /* assume old filehandle format */
185 xdev
= old_decode_dev(fh
->ofh_xdev
);
186 xino
= u32_to_ino_t(fh
->ofh_xino
);
187 mk_fsid(FSID_DEV
, tfh
, xdev
, xino
, 0, NULL
);
188 exp
= rqst_exp_find(rqstp
, FSID_DEV
, tfh
);
191 error
= nfserr_stale
;
192 if (PTR_ERR(exp
) == -ENOENT
)
196 return nfserrno(PTR_ERR(exp
));
198 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
) {
199 /* Elevate privileges so that the lack of 'r' or 'x'
200 * permission on some parent directory will
201 * not stop exportfs_decode_fh from being able
202 * to reconnect a directory into the dentry cache.
203 * The same problem can affect "SUBTREECHECK" exports,
204 * but as nfsd_acceptable depends on correct
205 * access control settings being in effect, we cannot
206 * fix that case easily.
208 struct cred
*new = prepare_creds();
210 return nfserrno(-ENOMEM
);
212 cap_raise_nfsd_set(new->cap_effective
,
214 put_cred(override_creds(new));
217 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
223 * Look up the dentry using the NFS file handle.
225 error
= nfserr_stale
;
226 if (rqstp
->rq_vers
> 2)
227 error
= nfserr_badhandle
;
229 if (fh
->fh_version
!= 1) {
230 sfid
.i32
.ino
= fh
->ofh_ino
;
231 sfid
.i32
.gen
= fh
->ofh_generation
;
232 sfid
.i32
.parent_ino
= fh
->ofh_dirino
;
235 if (fh
->ofh_dirino
== 0)
236 fileid_type
= FILEID_INO32_GEN
;
238 fileid_type
= FILEID_INO32_GEN_PARENT
;
240 fileid_type
= fh
->fh_fileid_type
;
242 if (fileid_type
== FILEID_ROOT
)
243 dentry
= dget(exp
->ex_path
.dentry
);
245 dentry
= exportfs_decode_fh(exp
->ex_path
.mnt
, fid
,
246 data_left
, fileid_type
,
247 nfsd_acceptable
, exp
);
251 if (IS_ERR(dentry
)) {
252 if (PTR_ERR(dentry
) != -EINVAL
)
253 error
= nfserrno(PTR_ERR(dentry
));
257 if (S_ISDIR(dentry
->d_inode
->i_mode
) &&
258 (dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
259 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %s/%s\n",
260 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
263 fhp
->fh_dentry
= dentry
;
264 fhp
->fh_export
= exp
;
272 * fh_verify - filehandle lookup and access checking
273 * @rqstp: pointer to current rpc request
274 * @fhp: filehandle to be verified
275 * @type: expected type of object pointed to by filehandle
276 * @access: type of access needed to object
278 * Look up a dentry from the on-the-wire filehandle, check the client's
279 * access to the export, and set the current task's credentials.
281 * Regardless of success or failure of fh_verify(), fh_put() should be
282 * called on @fhp when the caller is finished with the filehandle.
284 * fh_verify() may be called multiple times on a given filehandle, for
285 * example, when processing an NFSv4 compound. The first call will look
286 * up a dentry using the on-the-wire filehandle. Subsequent calls will
287 * skip the lookup and just perform the other checks and possibly change
288 * the current task's credentials.
290 * @type specifies the type of object expected using one of the S_IF*
291 * constants defined in include/linux/stat.h. The caller may use zero
292 * to indicate that it doesn't care, or a negative integer to indicate
293 * that it expects something not of the given type.
295 * @access is formed from the NFSD_MAY_* constants defined in
296 * include/linux/nfsd/nfsd.h.
299 fh_verify(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
, int access
)
301 struct svc_export
*exp
;
302 struct dentry
*dentry
;
305 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp
));
307 if (!fhp
->fh_dentry
) {
308 error
= nfsd_set_fh_dentry(rqstp
, fhp
);
312 dentry
= fhp
->fh_dentry
;
313 exp
= fhp
->fh_export
;
315 * We still have to do all these permission checks, even when
316 * fh_dentry is already set:
317 * - fh_verify may be called multiple times with different
318 * "access" arguments (e.g. nfsd_proc_create calls
319 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
320 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
321 * - in the NFSv4 case, the filehandle may have been filled
322 * in by fh_compose, and given a dentry, but further
323 * compound operations performed with that filehandle
324 * still need permissions checks. In the worst case, a
325 * mountpoint crossing may have changed the export
326 * options, and we may now need to use a different uid
327 * (for example, if different id-squashing options are in
328 * effect on the new filesystem).
330 error
= check_pseudo_root(rqstp
, dentry
, exp
);
334 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
338 error
= nfsd_mode_check(rqstp
, dentry
->d_inode
->i_mode
, type
);
343 * pseudoflavor restrictions are not enforced on NLM,
344 * which clients virtually always use auth_sys for,
345 * even while using RPCSEC_GSS for NFS.
347 if (access
& NFSD_MAY_LOCK
)
348 goto skip_pseudoflavor_check
;
350 * Clients may expect to be able to use auth_sys during mount,
351 * even if they use gss for everything else; see section 2.3.2
354 if (access
& NFSD_MAY_BYPASS_GSS_ON_ROOT
355 && exp
->ex_path
.dentry
== dentry
)
356 goto skip_pseudoflavor_check
;
358 error
= check_nfsd_access(exp
, rqstp
);
362 skip_pseudoflavor_check
:
363 /* Finally, check access permissions. */
364 error
= nfsd_permission(rqstp
, exp
, dentry
, access
);
367 dprintk("fh_verify: %s/%s permission failure, "
368 "acc=%x, error=%d\n",
369 dentry
->d_parent
->d_name
.name
,
371 access
, ntohl(error
));
374 if (error
== nfserr_stale
)
375 nfsdstats
.fh_stale
++;
381 * Compose a file handle for an NFS reply.
383 * Note that when first composed, the dentry may not yet have
384 * an inode. In this case a call to fh_update should be made
385 * before the fh goes out on the wire ...
387 static void _fh_update(struct svc_fh
*fhp
, struct svc_export
*exp
,
388 struct dentry
*dentry
)
390 if (dentry
!= exp
->ex_path
.dentry
) {
391 struct fid
*fid
= (struct fid
*)
392 (fhp
->fh_handle
.fh_auth
+ fhp
->fh_handle
.fh_size
/4 - 1);
393 int maxsize
= (fhp
->fh_maxsize
- fhp
->fh_handle
.fh_size
)/4;
394 int subtreecheck
= !(exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
);
396 fhp
->fh_handle
.fh_fileid_type
=
397 exportfs_encode_fh(dentry
, fid
, &maxsize
, subtreecheck
);
398 fhp
->fh_handle
.fh_size
+= maxsize
* 4;
400 fhp
->fh_handle
.fh_fileid_type
= FILEID_ROOT
;
405 * for composing old style file handles
407 static inline void _fh_update_old(struct dentry
*dentry
,
408 struct svc_export
*exp
,
411 fh
->ofh_ino
= ino_t_to_u32(dentry
->d_inode
->i_ino
);
412 fh
->ofh_generation
= dentry
->d_inode
->i_generation
;
413 if (S_ISDIR(dentry
->d_inode
->i_mode
) ||
414 (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
))
418 static bool is_root_export(struct svc_export
*exp
)
420 return exp
->ex_path
.dentry
== exp
->ex_path
.dentry
->d_sb
->s_root
;
423 static struct super_block
*exp_sb(struct svc_export
*exp
)
425 return exp
->ex_path
.dentry
->d_inode
->i_sb
;
428 static bool fsid_type_ok_for_exp(u8 fsid_type
, struct svc_export
*exp
)
432 if (!old_valid_dev(exp_sb(exp
)->s_dev
))
435 case FSID_MAJOR_MINOR
:
436 case FSID_ENCODE_DEV
:
437 return exp_sb(exp
)->s_type
->fs_flags
& FS_REQUIRES_DEV
;
439 return exp
->ex_flags
& NFSEXP_FSID
;
442 if (!is_root_export(exp
))
445 case FSID_UUID4_INUM
:
446 case FSID_UUID16_INUM
:
447 return exp
->ex_uuid
!= NULL
;
453 static void set_version_and_fsid_type(struct svc_fh
*fhp
, struct svc_export
*exp
, struct svc_fh
*ref_fh
)
459 if (ref_fh
&& ref_fh
->fh_export
== exp
) {
460 version
= ref_fh
->fh_handle
.fh_version
;
461 fsid_type
= ref_fh
->fh_handle
.fh_fsid_type
;
467 fsid_type
= FSID_DEV
;
476 * As the fsid -> filesystem mapping was guided by
477 * user-space, there is no guarantee that the filesystem
478 * actually supports that fsid type. If it doesn't we
479 * loop around again without ref_fh set.
481 if (!fsid_type_ok_for_exp(fsid_type
, exp
))
483 } else if (exp
->ex_flags
& NFSEXP_FSID
) {
484 fsid_type
= FSID_NUM
;
485 } else if (exp
->ex_uuid
) {
486 if (fhp
->fh_maxsize
>= 64) {
487 if (is_root_export(exp
))
488 fsid_type
= FSID_UUID16
;
490 fsid_type
= FSID_UUID16_INUM
;
492 if (is_root_export(exp
))
493 fsid_type
= FSID_UUID8
;
495 fsid_type
= FSID_UUID4_INUM
;
497 } else if (!old_valid_dev(exp_sb(exp
)->s_dev
))
498 /* for newer device numbers, we must use a newer fsid format */
499 fsid_type
= FSID_ENCODE_DEV
;
501 fsid_type
= FSID_DEV
;
502 fhp
->fh_handle
.fh_version
= version
;
504 fhp
->fh_handle
.fh_fsid_type
= fsid_type
;
508 fh_compose(struct svc_fh
*fhp
, struct svc_export
*exp
, struct dentry
*dentry
,
509 struct svc_fh
*ref_fh
)
511 /* ref_fh is a reference file handle.
512 * if it is non-null and for the same filesystem, then we should compose
513 * a filehandle which is of the same version, where possible.
514 * Currently, that means that if ref_fh->fh_handle.fh_version == 0xca
515 * Then create a 32byte filehandle using nfs_fhbase_old
519 struct inode
* inode
= dentry
->d_inode
;
520 struct dentry
*parent
= dentry
->d_parent
;
522 dev_t ex_dev
= exp_sb(exp
)->s_dev
;
524 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %s/%s, ino=%ld)\n",
525 MAJOR(ex_dev
), MINOR(ex_dev
),
526 (long) exp
->ex_path
.dentry
->d_inode
->i_ino
,
527 parent
->d_name
.name
, dentry
->d_name
.name
,
528 (inode
? inode
->i_ino
: 0));
530 /* Choose filehandle version and fsid type based on
531 * the reference filehandle (if it is in the same export)
532 * or the export options.
534 set_version_and_fsid_type(fhp
, exp
, ref_fh
);
539 if (fhp
->fh_locked
|| fhp
->fh_dentry
) {
540 printk(KERN_ERR
"fh_compose: fh %s/%s not initialized!\n",
541 parent
->d_name
.name
, dentry
->d_name
.name
);
543 if (fhp
->fh_maxsize
< NFS_FHSIZE
)
544 printk(KERN_ERR
"fh_compose: called with maxsize %d! %s/%s\n",
546 parent
->d_name
.name
, dentry
->d_name
.name
);
548 fhp
->fh_dentry
= dget(dentry
); /* our internal copy */
549 fhp
->fh_export
= exp
;
552 if (fhp
->fh_handle
.fh_version
== 0xca) {
553 /* old style filehandle please */
554 memset(&fhp
->fh_handle
.fh_base
, 0, NFS_FHSIZE
);
555 fhp
->fh_handle
.fh_size
= NFS_FHSIZE
;
556 fhp
->fh_handle
.ofh_dcookie
= 0xfeebbaca;
557 fhp
->fh_handle
.ofh_dev
= old_encode_dev(ex_dev
);
558 fhp
->fh_handle
.ofh_xdev
= fhp
->fh_handle
.ofh_dev
;
559 fhp
->fh_handle
.ofh_xino
=
560 ino_t_to_u32(exp
->ex_path
.dentry
->d_inode
->i_ino
);
561 fhp
->fh_handle
.ofh_dirino
= ino_t_to_u32(parent_ino(dentry
));
563 _fh_update_old(dentry
, exp
, &fhp
->fh_handle
);
566 fhp
->fh_handle
.fh_auth_type
= 0;
567 datap
= fhp
->fh_handle
.fh_auth
+0;
568 mk_fsid(fhp
->fh_handle
.fh_fsid_type
, datap
, ex_dev
,
569 exp
->ex_path
.dentry
->d_inode
->i_ino
,
570 exp
->ex_fsid
, exp
->ex_uuid
);
572 len
= key_len(fhp
->fh_handle
.fh_fsid_type
);
574 fhp
->fh_handle
.fh_size
= 4 + len
;
577 _fh_update(fhp
, exp
, dentry
);
578 if (fhp
->fh_handle
.fh_fileid_type
== 255) {
580 return nfserr_opnotsupp
;
588 * Update file handle information after changing a dentry.
589 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
592 fh_update(struct svc_fh
*fhp
)
594 struct dentry
*dentry
;
599 dentry
= fhp
->fh_dentry
;
600 if (!dentry
->d_inode
)
602 if (fhp
->fh_handle
.fh_version
!= 1) {
603 _fh_update_old(dentry
, fhp
->fh_export
, &fhp
->fh_handle
);
605 if (fhp
->fh_handle
.fh_fileid_type
!= FILEID_ROOT
)
608 _fh_update(fhp
, fhp
->fh_export
, dentry
);
609 if (fhp
->fh_handle
.fh_fileid_type
== 255)
610 return nfserr_opnotsupp
;
616 printk(KERN_ERR
"fh_update: fh not verified!\n");
619 printk(KERN_ERR
"fh_update: %s/%s still negative!\n",
620 dentry
->d_parent
->d_name
.name
, dentry
->d_name
.name
);
625 * Release a file handle.
628 fh_put(struct svc_fh
*fhp
)
630 struct dentry
* dentry
= fhp
->fh_dentry
;
631 struct svc_export
* exp
= fhp
->fh_export
;
634 fhp
->fh_dentry
= NULL
;
636 #ifdef CONFIG_NFSD_V3
637 fhp
->fh_pre_saved
= 0;
638 fhp
->fh_post_saved
= 0;
642 cache_put(&exp
->h
, &svc_export_cache
);
643 fhp
->fh_export
= NULL
;
649 * Shorthand for dprintk()'s
651 char * SVCFH_fmt(struct svc_fh
*fhp
)
653 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
656 sprintf(buf
, "%d: %08x %08x %08x %08x %08x %08x",
658 fh
->fh_base
.fh_pad
[0],
659 fh
->fh_base
.fh_pad
[1],
660 fh
->fh_base
.fh_pad
[2],
661 fh
->fh_base
.fh_pad
[3],
662 fh
->fh_base
.fh_pad
[4],
663 fh
->fh_base
.fh_pad
[5]);
667 enum fsid_source
fsid_source(struct svc_fh
*fhp
)
669 if (fhp
->fh_handle
.fh_version
!= 1)
670 return FSIDSOURCE_DEV
;
671 switch(fhp
->fh_handle
.fh_fsid_type
) {
673 case FSID_ENCODE_DEV
:
674 case FSID_MAJOR_MINOR
:
675 if (exp_sb(fhp
->fh_export
)->s_type
->fs_flags
& FS_REQUIRES_DEV
)
676 return FSIDSOURCE_DEV
;
679 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
680 return FSIDSOURCE_FSID
;
685 /* either a UUID type filehandle, or the filehandle doesn't
688 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
689 return FSIDSOURCE_FSID
;
690 if (fhp
->fh_export
->ex_uuid
)
691 return FSIDSOURCE_UUID
;
692 return FSIDSOURCE_DEV
;