1 // SPDX-License-Identifier: GPL-2.0
3 * NFS server file handle treatment.
5 * Copyright (C) 1995, 1996 Olaf Kirch <okir@monad.swb.de>
6 * Portions Copyright (C) 1999 G. Allen Morris III <gam3@acm.org>
7 * Extensive rewrite by Neil Brown <neilb@cse.unsw.edu.au> Southern-Spring 1999
8 * ... and again Southern-Winter 2001 to support export_operations
11 #include <linux/exportfs.h>
13 #include <linux/sunrpc/svcauth_gss.h>
18 #define NFSDDBG_FACILITY NFSDDBG_FH
22 * our acceptability function.
23 * if NOSUBTREECHECK, accept anything
24 * if not, require that we can walk up to exp->ex_dentry
25 * doing some checks on the 'x' bits
27 static int nfsd_acceptable(void *expv
, struct dentry
*dentry
)
29 struct svc_export
*exp
= expv
;
31 struct dentry
*tdentry
;
32 struct dentry
*parent
;
34 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
)
37 tdentry
= dget(dentry
);
38 while (tdentry
!= exp
->ex_path
.dentry
&& !IS_ROOT(tdentry
)) {
39 /* make sure parents give x permission to user */
41 parent
= dget_parent(tdentry
);
42 err
= inode_permission(d_inode(parent
), MAY_EXEC
);
50 if (tdentry
!= exp
->ex_path
.dentry
)
51 dprintk("nfsd_acceptable failed at %p %pd\n", tdentry
, tdentry
);
52 rv
= (tdentry
== exp
->ex_path
.dentry
);
57 /* Type check. The correct error return for type mismatches does not seem to be
58 * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
59 * comment in the NFSv3 spec says this is incorrect (implementation notes for
63 nfsd_mode_check(struct svc_rqst
*rqstp
, struct dentry
*dentry
,
66 umode_t mode
= d_inode(dentry
)->i_mode
& S_IFMT
;
68 if (requested
== 0) /* the caller doesn't care */
70 if (mode
== requested
) {
71 if (mode
== S_IFDIR
&& !d_can_lookup(dentry
)) {
78 * v4 has an error more specific than err_notdir which we should
79 * return in preference to err_notdir:
81 if (rqstp
->rq_vers
== 4 && mode
== S_IFLNK
)
82 return nfserr_symlink
;
83 if (requested
== S_IFDIR
)
90 static bool nfsd_originating_port_ok(struct svc_rqst
*rqstp
, int flags
)
92 if (flags
& NFSEXP_INSECURE_PORT
)
94 /* We don't require gss requests to use low ports: */
95 if (rqstp
->rq_cred
.cr_flavor
>= RPC_AUTH_GSS
)
97 return test_bit(RQ_SECURE
, &rqstp
->rq_flags
);
100 static __be32
nfsd_setuser_and_check_port(struct svc_rqst
*rqstp
,
101 struct svc_export
*exp
)
103 int flags
= nfsexp_flags(rqstp
, exp
);
105 /* Check if the request originated from a secure port. */
106 if (!nfsd_originating_port_ok(rqstp
, flags
)) {
107 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
108 dprintk("nfsd: request from insecure port %s!\n",
109 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
113 /* Set user creds for this exportpoint */
114 return nfserrno(nfsd_setuser(rqstp
, exp
));
117 static inline __be32
check_pseudo_root(struct svc_rqst
*rqstp
,
118 struct dentry
*dentry
, struct svc_export
*exp
)
120 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
123 * v2/v3 clients have no need for the V4ROOT export--they use
124 * the mount protocl instead; also, further V4ROOT checks may be
125 * in v4-specific code, in which case v2/v3 clients could bypass
128 if (!nfsd_v4client(rqstp
))
131 * We're exposing only the directories and symlinks that have to be
132 * traversed on the way to real exports:
134 if (unlikely(!d_is_dir(dentry
) &&
135 !d_is_symlink(dentry
)))
138 * A pseudoroot export gives permission to access only one
139 * single directory; the kernel has to make another upcall
140 * before granting access to anything else under it:
142 if (unlikely(dentry
!= exp
->ex_path
.dentry
))
148 * Use the given filehandle to look up the corresponding export and
149 * dentry. On success, the results are used to set fh_export and
152 static __be32
nfsd_set_fh_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
)
154 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
155 struct fid
*fid
= NULL
, sfid
;
156 struct svc_export
*exp
;
157 struct dentry
*dentry
;
159 int data_left
= fh
->fh_size
/4;
162 error
= nfserr_stale
;
163 if (rqstp
->rq_vers
> 2)
164 error
= nfserr_badhandle
;
165 if (rqstp
->rq_vers
== 4 && fh
->fh_size
== 0)
166 return nfserr_nofilehandle
;
168 if (fh
->fh_version
== 1) {
173 if (fh
->fh_auth_type
!= 0)
175 len
= key_len(fh
->fh_fsid_type
) / 4;
178 if (fh
->fh_fsid_type
== FSID_MAJOR_MINOR
) {
179 /* deprecated, convert to type 3 */
180 len
= key_len(FSID_ENCODE_DEV
)/4;
181 fh
->fh_fsid_type
= FSID_ENCODE_DEV
;
183 * struct knfsd_fh uses host-endian fields, which are
184 * sometimes used to hold net-endian values. This
185 * confuses sparse, so we must use __force here to
186 * keep it from complaining.
188 fh
->fh_fsid
[0] = new_encode_dev(MKDEV(ntohl((__force __be32
)fh
->fh_fsid
[0]),
189 ntohl((__force __be32
)fh
->fh_fsid
[1])));
190 fh
->fh_fsid
[1] = fh
->fh_fsid
[2];
195 exp
= rqst_exp_find(rqstp
, fh
->fh_fsid_type
, fh
->fh_fsid
);
196 fid
= (struct fid
*)(fh
->fh_fsid
+ len
);
202 if (fh
->fh_size
!= NFS_FHSIZE
)
204 /* assume old filehandle format */
205 xdev
= old_decode_dev(fh
->ofh_xdev
);
206 xino
= u32_to_ino_t(fh
->ofh_xino
);
207 mk_fsid(FSID_DEV
, tfh
, xdev
, xino
, 0, NULL
);
208 exp
= rqst_exp_find(rqstp
, FSID_DEV
, tfh
);
211 error
= nfserr_stale
;
212 if (PTR_ERR(exp
) == -ENOENT
)
216 return nfserrno(PTR_ERR(exp
));
218 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
) {
219 /* Elevate privileges so that the lack of 'r' or 'x'
220 * permission on some parent directory will
221 * not stop exportfs_decode_fh from being able
222 * to reconnect a directory into the dentry cache.
223 * The same problem can affect "SUBTREECHECK" exports,
224 * but as nfsd_acceptable depends on correct
225 * access control settings being in effect, we cannot
226 * fix that case easily.
228 struct cred
*new = prepare_creds();
230 error
= nfserrno(-ENOMEM
);
234 cap_raise_nfsd_set(new->cap_effective
,
236 put_cred(override_creds(new));
239 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
245 * Look up the dentry using the NFS file handle.
247 error
= nfserr_stale
;
248 if (rqstp
->rq_vers
> 2)
249 error
= nfserr_badhandle
;
251 if (fh
->fh_version
!= 1) {
252 sfid
.i32
.ino
= fh
->ofh_ino
;
253 sfid
.i32
.gen
= fh
->ofh_generation
;
254 sfid
.i32
.parent_ino
= fh
->ofh_dirino
;
257 if (fh
->ofh_dirino
== 0)
258 fileid_type
= FILEID_INO32_GEN
;
260 fileid_type
= FILEID_INO32_GEN_PARENT
;
262 fileid_type
= fh
->fh_fileid_type
;
264 if (fileid_type
== FILEID_ROOT
)
265 dentry
= dget(exp
->ex_path
.dentry
);
267 dentry
= exportfs_decode_fh(exp
->ex_path
.mnt
, fid
,
268 data_left
, fileid_type
,
269 nfsd_acceptable
, exp
);
273 if (IS_ERR(dentry
)) {
274 if (PTR_ERR(dentry
) != -EINVAL
)
275 error
= nfserrno(PTR_ERR(dentry
));
279 if (d_is_dir(dentry
) &&
280 (dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
281 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
285 fhp
->fh_dentry
= dentry
;
286 fhp
->fh_export
= exp
;
294 * fh_verify - filehandle lookup and access checking
295 * @rqstp: pointer to current rpc request
296 * @fhp: filehandle to be verified
297 * @type: expected type of object pointed to by filehandle
298 * @access: type of access needed to object
300 * Look up a dentry from the on-the-wire filehandle, check the client's
301 * access to the export, and set the current task's credentials.
303 * Regardless of success or failure of fh_verify(), fh_put() should be
304 * called on @fhp when the caller is finished with the filehandle.
306 * fh_verify() may be called multiple times on a given filehandle, for
307 * example, when processing an NFSv4 compound. The first call will look
308 * up a dentry using the on-the-wire filehandle. Subsequent calls will
309 * skip the lookup and just perform the other checks and possibly change
310 * the current task's credentials.
312 * @type specifies the type of object expected using one of the S_IF*
313 * constants defined in include/linux/stat.h. The caller may use zero
314 * to indicate that it doesn't care, or a negative integer to indicate
315 * that it expects something not of the given type.
317 * @access is formed from the NFSD_MAY_* constants defined in
321 fh_verify(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
, int access
)
323 struct svc_export
*exp
;
324 struct dentry
*dentry
;
327 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp
));
329 if (!fhp
->fh_dentry
) {
330 error
= nfsd_set_fh_dentry(rqstp
, fhp
);
334 dentry
= fhp
->fh_dentry
;
335 exp
= fhp
->fh_export
;
337 * We still have to do all these permission checks, even when
338 * fh_dentry is already set:
339 * - fh_verify may be called multiple times with different
340 * "access" arguments (e.g. nfsd_proc_create calls
341 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
342 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
343 * - in the NFSv4 case, the filehandle may have been filled
344 * in by fh_compose, and given a dentry, but further
345 * compound operations performed with that filehandle
346 * still need permissions checks. In the worst case, a
347 * mountpoint crossing may have changed the export
348 * options, and we may now need to use a different uid
349 * (for example, if different id-squashing options are in
350 * effect on the new filesystem).
352 error
= check_pseudo_root(rqstp
, dentry
, exp
);
356 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
360 error
= nfsd_mode_check(rqstp
, dentry
, type
);
365 * pseudoflavor restrictions are not enforced on NLM,
366 * which clients virtually always use auth_sys for,
367 * even while using RPCSEC_GSS for NFS.
369 if (access
& NFSD_MAY_LOCK
|| access
& NFSD_MAY_BYPASS_GSS
)
370 goto skip_pseudoflavor_check
;
372 * Clients may expect to be able to use auth_sys during mount,
373 * even if they use gss for everything else; see section 2.3.2
376 if (access
& NFSD_MAY_BYPASS_GSS_ON_ROOT
377 && exp
->ex_path
.dentry
== dentry
)
378 goto skip_pseudoflavor_check
;
380 error
= check_nfsd_access(exp
, rqstp
);
384 skip_pseudoflavor_check
:
385 /* Finally, check access permissions. */
386 error
= nfsd_permission(rqstp
, exp
, dentry
, access
);
389 dprintk("fh_verify: %pd2 permission failure, "
390 "acc=%x, error=%d\n",
392 access
, ntohl(error
));
395 if (error
== nfserr_stale
)
396 nfsdstats
.fh_stale
++;
402 * Compose a file handle for an NFS reply.
404 * Note that when first composed, the dentry may not yet have
405 * an inode. In this case a call to fh_update should be made
406 * before the fh goes out on the wire ...
408 static void _fh_update(struct svc_fh
*fhp
, struct svc_export
*exp
,
409 struct dentry
*dentry
)
411 if (dentry
!= exp
->ex_path
.dentry
) {
412 struct fid
*fid
= (struct fid
*)
413 (fhp
->fh_handle
.fh_fsid
+ fhp
->fh_handle
.fh_size
/4 - 1);
414 int maxsize
= (fhp
->fh_maxsize
- fhp
->fh_handle
.fh_size
)/4;
415 int subtreecheck
= !(exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
);
417 fhp
->fh_handle
.fh_fileid_type
=
418 exportfs_encode_fh(dentry
, fid
, &maxsize
, subtreecheck
);
419 fhp
->fh_handle
.fh_size
+= maxsize
* 4;
421 fhp
->fh_handle
.fh_fileid_type
= FILEID_ROOT
;
426 * for composing old style file handles
428 static inline void _fh_update_old(struct dentry
*dentry
,
429 struct svc_export
*exp
,
432 fh
->ofh_ino
= ino_t_to_u32(d_inode(dentry
)->i_ino
);
433 fh
->ofh_generation
= d_inode(dentry
)->i_generation
;
434 if (d_is_dir(dentry
) ||
435 (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
))
439 static bool is_root_export(struct svc_export
*exp
)
441 return exp
->ex_path
.dentry
== exp
->ex_path
.dentry
->d_sb
->s_root
;
444 static struct super_block
*exp_sb(struct svc_export
*exp
)
446 return exp
->ex_path
.dentry
->d_sb
;
449 static bool fsid_type_ok_for_exp(u8 fsid_type
, struct svc_export
*exp
)
453 if (!old_valid_dev(exp_sb(exp
)->s_dev
))
456 case FSID_MAJOR_MINOR
:
457 case FSID_ENCODE_DEV
:
458 return exp_sb(exp
)->s_type
->fs_flags
& FS_REQUIRES_DEV
;
460 return exp
->ex_flags
& NFSEXP_FSID
;
463 if (!is_root_export(exp
))
466 case FSID_UUID4_INUM
:
467 case FSID_UUID16_INUM
:
468 return exp
->ex_uuid
!= NULL
;
474 static void set_version_and_fsid_type(struct svc_fh
*fhp
, struct svc_export
*exp
, struct svc_fh
*ref_fh
)
480 if (ref_fh
&& ref_fh
->fh_export
== exp
) {
481 version
= ref_fh
->fh_handle
.fh_version
;
482 fsid_type
= ref_fh
->fh_handle
.fh_fsid_type
;
488 fsid_type
= FSID_DEV
;
497 * As the fsid -> filesystem mapping was guided by
498 * user-space, there is no guarantee that the filesystem
499 * actually supports that fsid type. If it doesn't we
500 * loop around again without ref_fh set.
502 if (!fsid_type_ok_for_exp(fsid_type
, exp
))
504 } else if (exp
->ex_flags
& NFSEXP_FSID
) {
505 fsid_type
= FSID_NUM
;
506 } else if (exp
->ex_uuid
) {
507 if (fhp
->fh_maxsize
>= 64) {
508 if (is_root_export(exp
))
509 fsid_type
= FSID_UUID16
;
511 fsid_type
= FSID_UUID16_INUM
;
513 if (is_root_export(exp
))
514 fsid_type
= FSID_UUID8
;
516 fsid_type
= FSID_UUID4_INUM
;
518 } else if (!old_valid_dev(exp_sb(exp
)->s_dev
))
519 /* for newer device numbers, we must use a newer fsid format */
520 fsid_type
= FSID_ENCODE_DEV
;
522 fsid_type
= FSID_DEV
;
523 fhp
->fh_handle
.fh_version
= version
;
525 fhp
->fh_handle
.fh_fsid_type
= fsid_type
;
529 fh_compose(struct svc_fh
*fhp
, struct svc_export
*exp
, struct dentry
*dentry
,
530 struct svc_fh
*ref_fh
)
532 /* ref_fh is a reference file handle.
533 * if it is non-null and for the same filesystem, then we should compose
534 * a filehandle which is of the same version, where possible.
535 * Currently, that means that if ref_fh->fh_handle.fh_version == 0xca
536 * Then create a 32byte filehandle using nfs_fhbase_old
540 struct inode
* inode
= d_inode(dentry
);
541 dev_t ex_dev
= exp_sb(exp
)->s_dev
;
543 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
544 MAJOR(ex_dev
), MINOR(ex_dev
),
545 (long) d_inode(exp
->ex_path
.dentry
)->i_ino
,
547 (inode
? inode
->i_ino
: 0));
549 /* Choose filehandle version and fsid type based on
550 * the reference filehandle (if it is in the same export)
551 * or the export options.
553 set_version_and_fsid_type(fhp
, exp
, ref_fh
);
558 if (fhp
->fh_locked
|| fhp
->fh_dentry
) {
559 printk(KERN_ERR
"fh_compose: fh %pd2 not initialized!\n",
562 if (fhp
->fh_maxsize
< NFS_FHSIZE
)
563 printk(KERN_ERR
"fh_compose: called with maxsize %d! %pd2\n",
567 fhp
->fh_dentry
= dget(dentry
); /* our internal copy */
568 fhp
->fh_export
= exp_get(exp
);
570 if (fhp
->fh_handle
.fh_version
== 0xca) {
571 /* old style filehandle please */
572 memset(&fhp
->fh_handle
.fh_base
, 0, NFS_FHSIZE
);
573 fhp
->fh_handle
.fh_size
= NFS_FHSIZE
;
574 fhp
->fh_handle
.ofh_dcookie
= 0xfeebbaca;
575 fhp
->fh_handle
.ofh_dev
= old_encode_dev(ex_dev
);
576 fhp
->fh_handle
.ofh_xdev
= fhp
->fh_handle
.ofh_dev
;
577 fhp
->fh_handle
.ofh_xino
=
578 ino_t_to_u32(d_inode(exp
->ex_path
.dentry
)->i_ino
);
579 fhp
->fh_handle
.ofh_dirino
= ino_t_to_u32(parent_ino(dentry
));
581 _fh_update_old(dentry
, exp
, &fhp
->fh_handle
);
583 fhp
->fh_handle
.fh_size
=
584 key_len(fhp
->fh_handle
.fh_fsid_type
) + 4;
585 fhp
->fh_handle
.fh_auth_type
= 0;
587 mk_fsid(fhp
->fh_handle
.fh_fsid_type
,
588 fhp
->fh_handle
.fh_fsid
,
590 d_inode(exp
->ex_path
.dentry
)->i_ino
,
591 exp
->ex_fsid
, exp
->ex_uuid
);
594 _fh_update(fhp
, exp
, dentry
);
595 if (fhp
->fh_handle
.fh_fileid_type
== FILEID_INVALID
) {
597 return nfserr_opnotsupp
;
605 * Update file handle information after changing a dentry.
606 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
609 fh_update(struct svc_fh
*fhp
)
611 struct dentry
*dentry
;
616 dentry
= fhp
->fh_dentry
;
617 if (d_really_is_negative(dentry
))
619 if (fhp
->fh_handle
.fh_version
!= 1) {
620 _fh_update_old(dentry
, fhp
->fh_export
, &fhp
->fh_handle
);
622 if (fhp
->fh_handle
.fh_fileid_type
!= FILEID_ROOT
)
625 _fh_update(fhp
, fhp
->fh_export
, dentry
);
626 if (fhp
->fh_handle
.fh_fileid_type
== FILEID_INVALID
)
627 return nfserr_opnotsupp
;
631 printk(KERN_ERR
"fh_update: fh not verified!\n");
632 return nfserr_serverfault
;
634 printk(KERN_ERR
"fh_update: %pd2 still negative!\n",
636 return nfserr_serverfault
;
640 * Release a file handle.
643 fh_put(struct svc_fh
*fhp
)
645 struct dentry
* dentry
= fhp
->fh_dentry
;
646 struct svc_export
* exp
= fhp
->fh_export
;
649 fhp
->fh_dentry
= NULL
;
656 fhp
->fh_export
= NULL
;
662 * Shorthand for dprintk()'s
664 char * SVCFH_fmt(struct svc_fh
*fhp
)
666 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
669 sprintf(buf
, "%d: %08x %08x %08x %08x %08x %08x",
671 fh
->fh_base
.fh_pad
[0],
672 fh
->fh_base
.fh_pad
[1],
673 fh
->fh_base
.fh_pad
[2],
674 fh
->fh_base
.fh_pad
[3],
675 fh
->fh_base
.fh_pad
[4],
676 fh
->fh_base
.fh_pad
[5]);
680 enum fsid_source
fsid_source(struct svc_fh
*fhp
)
682 if (fhp
->fh_handle
.fh_version
!= 1)
683 return FSIDSOURCE_DEV
;
684 switch(fhp
->fh_handle
.fh_fsid_type
) {
686 case FSID_ENCODE_DEV
:
687 case FSID_MAJOR_MINOR
:
688 if (exp_sb(fhp
->fh_export
)->s_type
->fs_flags
& FS_REQUIRES_DEV
)
689 return FSIDSOURCE_DEV
;
692 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
693 return FSIDSOURCE_FSID
;
698 /* either a UUID type filehandle, or the filehandle doesn't
701 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
702 return FSIDSOURCE_FSID
;
703 if (fhp
->fh_export
->ex_uuid
)
704 return FSIDSOURCE_UUID
;
705 return FSIDSOURCE_DEV
;