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>
19 #define NFSDDBG_FACILITY NFSDDBG_FH
23 * our acceptability function.
24 * if NOSUBTREECHECK, accept anything
25 * if not, require that we can walk up to exp->ex_dentry
26 * doing some checks on the 'x' bits
28 static int nfsd_acceptable(void *expv
, struct dentry
*dentry
)
30 struct svc_export
*exp
= expv
;
32 struct dentry
*tdentry
;
33 struct dentry
*parent
;
35 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
)
38 tdentry
= dget(dentry
);
39 while (tdentry
!= exp
->ex_path
.dentry
&& !IS_ROOT(tdentry
)) {
40 /* make sure parents give x permission to user */
42 parent
= dget_parent(tdentry
);
43 err
= inode_permission(&init_user_ns
,
44 d_inode(parent
), MAY_EXEC
);
52 if (tdentry
!= exp
->ex_path
.dentry
)
53 dprintk("nfsd_acceptable failed at %p %pd\n", tdentry
, tdentry
);
54 rv
= (tdentry
== exp
->ex_path
.dentry
);
59 /* Type check. The correct error return for type mismatches does not seem to be
60 * generally agreed upon. SunOS seems to use EISDIR if file isn't S_IFREG; a
61 * comment in the NFSv3 spec says this is incorrect (implementation notes for
65 nfsd_mode_check(struct svc_rqst
*rqstp
, struct dentry
*dentry
,
68 umode_t mode
= d_inode(dentry
)->i_mode
& S_IFMT
;
70 if (requested
== 0) /* the caller doesn't care */
72 if (mode
== requested
) {
73 if (mode
== S_IFDIR
&& !d_can_lookup(dentry
)) {
80 * v4 has an error more specific than err_notdir which we should
81 * return in preference to err_notdir:
83 if (rqstp
->rq_vers
== 4 && mode
== S_IFLNK
)
84 return nfserr_symlink
;
85 if (requested
== S_IFDIR
)
92 static bool nfsd_originating_port_ok(struct svc_rqst
*rqstp
, int flags
)
94 if (flags
& NFSEXP_INSECURE_PORT
)
96 /* We don't require gss requests to use low ports: */
97 if (rqstp
->rq_cred
.cr_flavor
>= RPC_AUTH_GSS
)
99 return test_bit(RQ_SECURE
, &rqstp
->rq_flags
);
102 static __be32
nfsd_setuser_and_check_port(struct svc_rqst
*rqstp
,
103 struct svc_export
*exp
)
105 int flags
= nfsexp_flags(rqstp
, exp
);
107 /* Check if the request originated from a secure port. */
108 if (!nfsd_originating_port_ok(rqstp
, flags
)) {
109 RPC_IFDEBUG(char buf
[RPC_MAX_ADDRBUFLEN
]);
110 dprintk("nfsd: request from insecure port %s!\n",
111 svc_print_addr(rqstp
, buf
, sizeof(buf
)));
115 /* Set user creds for this exportpoint */
116 return nfserrno(nfsd_setuser(rqstp
, exp
));
119 static inline __be32
check_pseudo_root(struct svc_rqst
*rqstp
,
120 struct dentry
*dentry
, struct svc_export
*exp
)
122 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
125 * v2/v3 clients have no need for the V4ROOT export--they use
126 * the mount protocl instead; also, further V4ROOT checks may be
127 * in v4-specific code, in which case v2/v3 clients could bypass
130 if (!nfsd_v4client(rqstp
))
133 * We're exposing only the directories and symlinks that have to be
134 * traversed on the way to real exports:
136 if (unlikely(!d_is_dir(dentry
) &&
137 !d_is_symlink(dentry
)))
140 * A pseudoroot export gives permission to access only one
141 * single directory; the kernel has to make another upcall
142 * before granting access to anything else under it:
144 if (unlikely(dentry
!= exp
->ex_path
.dentry
))
150 * Use the given filehandle to look up the corresponding export and
151 * dentry. On success, the results are used to set fh_export and
154 static __be32
nfsd_set_fh_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
)
156 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
157 struct fid
*fid
= NULL
, sfid
;
158 struct svc_export
*exp
;
159 struct dentry
*dentry
;
161 int data_left
= fh
->fh_size
/4;
164 error
= nfserr_stale
;
165 if (rqstp
->rq_vers
> 2)
166 error
= nfserr_badhandle
;
167 if (rqstp
->rq_vers
== 4 && fh
->fh_size
== 0)
168 return nfserr_nofilehandle
;
170 if (fh
->fh_version
== 1) {
175 if (fh
->fh_auth_type
!= 0)
177 len
= key_len(fh
->fh_fsid_type
) / 4;
180 if (fh
->fh_fsid_type
== FSID_MAJOR_MINOR
) {
181 /* deprecated, convert to type 3 */
182 len
= key_len(FSID_ENCODE_DEV
)/4;
183 fh
->fh_fsid_type
= FSID_ENCODE_DEV
;
185 * struct knfsd_fh uses host-endian fields, which are
186 * sometimes used to hold net-endian values. This
187 * confuses sparse, so we must use __force here to
188 * keep it from complaining.
190 fh
->fh_fsid
[0] = new_encode_dev(MKDEV(ntohl((__force __be32
)fh
->fh_fsid
[0]),
191 ntohl((__force __be32
)fh
->fh_fsid
[1])));
192 fh
->fh_fsid
[1] = fh
->fh_fsid
[2];
197 exp
= rqst_exp_find(rqstp
, fh
->fh_fsid_type
, fh
->fh_fsid
);
198 fid
= (struct fid
*)(fh
->fh_fsid
+ len
);
204 if (fh
->fh_size
!= NFS_FHSIZE
)
206 /* assume old filehandle format */
207 xdev
= old_decode_dev(fh
->ofh_xdev
);
208 xino
= u32_to_ino_t(fh
->ofh_xino
);
209 mk_fsid(FSID_DEV
, tfh
, xdev
, xino
, 0, NULL
);
210 exp
= rqst_exp_find(rqstp
, FSID_DEV
, tfh
);
213 error
= nfserr_stale
;
215 trace_nfsd_set_fh_dentry_badexport(rqstp
, fhp
, PTR_ERR(exp
));
217 if (PTR_ERR(exp
) == -ENOENT
)
220 return nfserrno(PTR_ERR(exp
));
223 if (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
) {
224 /* Elevate privileges so that the lack of 'r' or 'x'
225 * permission on some parent directory will
226 * not stop exportfs_decode_fh from being able
227 * to reconnect a directory into the dentry cache.
228 * The same problem can affect "SUBTREECHECK" exports,
229 * but as nfsd_acceptable depends on correct
230 * access control settings being in effect, we cannot
231 * fix that case easily.
233 struct cred
*new = prepare_creds();
235 error
= nfserrno(-ENOMEM
);
239 cap_raise_nfsd_set(new->cap_effective
,
241 put_cred(override_creds(new));
244 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
250 * Look up the dentry using the NFS file handle.
252 error
= nfserr_stale
;
253 if (rqstp
->rq_vers
> 2)
254 error
= nfserr_badhandle
;
256 if (fh
->fh_version
!= 1) {
257 sfid
.i32
.ino
= fh
->ofh_ino
;
258 sfid
.i32
.gen
= fh
->ofh_generation
;
259 sfid
.i32
.parent_ino
= fh
->ofh_dirino
;
262 if (fh
->ofh_dirino
== 0)
263 fileid_type
= FILEID_INO32_GEN
;
265 fileid_type
= FILEID_INO32_GEN_PARENT
;
267 fileid_type
= fh
->fh_fileid_type
;
269 if (fileid_type
== FILEID_ROOT
)
270 dentry
= dget(exp
->ex_path
.dentry
);
272 dentry
= exportfs_decode_fh_raw(exp
->ex_path
.mnt
, fid
,
273 data_left
, fileid_type
,
274 nfsd_acceptable
, exp
);
275 if (IS_ERR_OR_NULL(dentry
)) {
276 trace_nfsd_set_fh_dentry_badhandle(rqstp
, fhp
,
277 dentry
? PTR_ERR(dentry
) : -ESTALE
);
278 switch (PTR_ERR(dentry
)) {
283 dentry
= ERR_PTR(-ESTALE
);
289 if (IS_ERR(dentry
)) {
290 if (PTR_ERR(dentry
) != -EINVAL
)
291 error
= nfserrno(PTR_ERR(dentry
));
295 if (d_is_dir(dentry
) &&
296 (dentry
->d_flags
& DCACHE_DISCONNECTED
)) {
297 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
301 fhp
->fh_dentry
= dentry
;
302 fhp
->fh_export
= exp
;
304 switch (rqstp
->rq_vers
) {
306 if (dentry
->d_sb
->s_export_op
->flags
& EXPORT_OP_NOATOMIC_ATTR
)
307 fhp
->fh_no_atomic_attr
= true;
310 if (dentry
->d_sb
->s_export_op
->flags
& EXPORT_OP_NOWCC
)
311 fhp
->fh_no_wcc
= true;
314 fhp
->fh_no_wcc
= true;
324 * fh_verify - filehandle lookup and access checking
325 * @rqstp: pointer to current rpc request
326 * @fhp: filehandle to be verified
327 * @type: expected type of object pointed to by filehandle
328 * @access: type of access needed to object
330 * Look up a dentry from the on-the-wire filehandle, check the client's
331 * access to the export, and set the current task's credentials.
333 * Regardless of success or failure of fh_verify(), fh_put() should be
334 * called on @fhp when the caller is finished with the filehandle.
336 * fh_verify() may be called multiple times on a given filehandle, for
337 * example, when processing an NFSv4 compound. The first call will look
338 * up a dentry using the on-the-wire filehandle. Subsequent calls will
339 * skip the lookup and just perform the other checks and possibly change
340 * the current task's credentials.
342 * @type specifies the type of object expected using one of the S_IF*
343 * constants defined in include/linux/stat.h. The caller may use zero
344 * to indicate that it doesn't care, or a negative integer to indicate
345 * that it expects something not of the given type.
347 * @access is formed from the NFSD_MAY_* constants defined in
351 fh_verify(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
, int access
)
353 struct svc_export
*exp
= NULL
;
354 struct dentry
*dentry
;
357 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp
));
359 if (!fhp
->fh_dentry
) {
360 error
= nfsd_set_fh_dentry(rqstp
, fhp
);
364 dentry
= fhp
->fh_dentry
;
365 exp
= fhp
->fh_export
;
367 * We still have to do all these permission checks, even when
368 * fh_dentry is already set:
369 * - fh_verify may be called multiple times with different
370 * "access" arguments (e.g. nfsd_proc_create calls
371 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
372 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
373 * - in the NFSv4 case, the filehandle may have been filled
374 * in by fh_compose, and given a dentry, but further
375 * compound operations performed with that filehandle
376 * still need permissions checks. In the worst case, a
377 * mountpoint crossing may have changed the export
378 * options, and we may now need to use a different uid
379 * (for example, if different id-squashing options are in
380 * effect on the new filesystem).
382 error
= check_pseudo_root(rqstp
, dentry
, exp
);
386 error
= nfsd_setuser_and_check_port(rqstp
, exp
);
390 error
= nfsd_mode_check(rqstp
, dentry
, type
);
395 * pseudoflavor restrictions are not enforced on NLM,
396 * which clients virtually always use auth_sys for,
397 * even while using RPCSEC_GSS for NFS.
399 if (access
& NFSD_MAY_LOCK
|| access
& NFSD_MAY_BYPASS_GSS
)
400 goto skip_pseudoflavor_check
;
402 * Clients may expect to be able to use auth_sys during mount,
403 * even if they use gss for everything else; see section 2.3.2
406 if (access
& NFSD_MAY_BYPASS_GSS_ON_ROOT
407 && exp
->ex_path
.dentry
== dentry
)
408 goto skip_pseudoflavor_check
;
410 error
= check_nfsd_access(exp
, rqstp
);
414 skip_pseudoflavor_check
:
415 /* Finally, check access permissions. */
416 error
= nfsd_permission(rqstp
, exp
, dentry
, access
);
419 dprintk("fh_verify: %pd2 permission failure, "
420 "acc=%x, error=%d\n",
422 access
, ntohl(error
));
425 if (error
== nfserr_stale
)
426 nfsd_stats_fh_stale_inc(exp
);
432 * Compose a file handle for an NFS reply.
434 * Note that when first composed, the dentry may not yet have
435 * an inode. In this case a call to fh_update should be made
436 * before the fh goes out on the wire ...
438 static void _fh_update(struct svc_fh
*fhp
, struct svc_export
*exp
,
439 struct dentry
*dentry
)
441 if (dentry
!= exp
->ex_path
.dentry
) {
442 struct fid
*fid
= (struct fid
*)
443 (fhp
->fh_handle
.fh_fsid
+ fhp
->fh_handle
.fh_size
/4 - 1);
444 int maxsize
= (fhp
->fh_maxsize
- fhp
->fh_handle
.fh_size
)/4;
445 int subtreecheck
= !(exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
);
447 fhp
->fh_handle
.fh_fileid_type
=
448 exportfs_encode_fh(dentry
, fid
, &maxsize
, subtreecheck
);
449 fhp
->fh_handle
.fh_size
+= maxsize
* 4;
451 fhp
->fh_handle
.fh_fileid_type
= FILEID_ROOT
;
456 * for composing old style file handles
458 static inline void _fh_update_old(struct dentry
*dentry
,
459 struct svc_export
*exp
,
462 fh
->ofh_ino
= ino_t_to_u32(d_inode(dentry
)->i_ino
);
463 fh
->ofh_generation
= d_inode(dentry
)->i_generation
;
464 if (d_is_dir(dentry
) ||
465 (exp
->ex_flags
& NFSEXP_NOSUBTREECHECK
))
469 static bool is_root_export(struct svc_export
*exp
)
471 return exp
->ex_path
.dentry
== exp
->ex_path
.dentry
->d_sb
->s_root
;
474 static struct super_block
*exp_sb(struct svc_export
*exp
)
476 return exp
->ex_path
.dentry
->d_sb
;
479 static bool fsid_type_ok_for_exp(u8 fsid_type
, struct svc_export
*exp
)
483 if (!old_valid_dev(exp_sb(exp
)->s_dev
))
486 case FSID_MAJOR_MINOR
:
487 case FSID_ENCODE_DEV
:
488 return exp_sb(exp
)->s_type
->fs_flags
& FS_REQUIRES_DEV
;
490 return exp
->ex_flags
& NFSEXP_FSID
;
493 if (!is_root_export(exp
))
496 case FSID_UUID4_INUM
:
497 case FSID_UUID16_INUM
:
498 return exp
->ex_uuid
!= NULL
;
504 static void set_version_and_fsid_type(struct svc_fh
*fhp
, struct svc_export
*exp
, struct svc_fh
*ref_fh
)
510 if (ref_fh
&& ref_fh
->fh_export
== exp
) {
511 version
= ref_fh
->fh_handle
.fh_version
;
512 fsid_type
= ref_fh
->fh_handle
.fh_fsid_type
;
518 fsid_type
= FSID_DEV
;
527 * As the fsid -> filesystem mapping was guided by
528 * user-space, there is no guarantee that the filesystem
529 * actually supports that fsid type. If it doesn't we
530 * loop around again without ref_fh set.
532 if (!fsid_type_ok_for_exp(fsid_type
, exp
))
534 } else if (exp
->ex_flags
& NFSEXP_FSID
) {
535 fsid_type
= FSID_NUM
;
536 } else if (exp
->ex_uuid
) {
537 if (fhp
->fh_maxsize
>= 64) {
538 if (is_root_export(exp
))
539 fsid_type
= FSID_UUID16
;
541 fsid_type
= FSID_UUID16_INUM
;
543 if (is_root_export(exp
))
544 fsid_type
= FSID_UUID8
;
546 fsid_type
= FSID_UUID4_INUM
;
548 } else if (!old_valid_dev(exp_sb(exp
)->s_dev
))
549 /* for newer device numbers, we must use a newer fsid format */
550 fsid_type
= FSID_ENCODE_DEV
;
552 fsid_type
= FSID_DEV
;
553 fhp
->fh_handle
.fh_version
= version
;
555 fhp
->fh_handle
.fh_fsid_type
= fsid_type
;
559 fh_compose(struct svc_fh
*fhp
, struct svc_export
*exp
, struct dentry
*dentry
,
560 struct svc_fh
*ref_fh
)
562 /* ref_fh is a reference file handle.
563 * if it is non-null and for the same filesystem, then we should compose
564 * a filehandle which is of the same version, where possible.
565 * Currently, that means that if ref_fh->fh_handle.fh_version == 0xca
566 * Then create a 32byte filehandle using nfs_fhbase_old
570 struct inode
* inode
= d_inode(dentry
);
571 dev_t ex_dev
= exp_sb(exp
)->s_dev
;
573 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
574 MAJOR(ex_dev
), MINOR(ex_dev
),
575 (long) d_inode(exp
->ex_path
.dentry
)->i_ino
,
577 (inode
? inode
->i_ino
: 0));
579 /* Choose filehandle version and fsid type based on
580 * the reference filehandle (if it is in the same export)
581 * or the export options.
583 set_version_and_fsid_type(fhp
, exp
, ref_fh
);
585 /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
586 fhp
->fh_no_wcc
= ref_fh
? ref_fh
->fh_no_wcc
: false;
591 if (fhp
->fh_locked
|| fhp
->fh_dentry
) {
592 printk(KERN_ERR
"fh_compose: fh %pd2 not initialized!\n",
595 if (fhp
->fh_maxsize
< NFS_FHSIZE
)
596 printk(KERN_ERR
"fh_compose: called with maxsize %d! %pd2\n",
600 fhp
->fh_dentry
= dget(dentry
); /* our internal copy */
601 fhp
->fh_export
= exp_get(exp
);
603 if (fhp
->fh_handle
.fh_version
== 0xca) {
604 /* old style filehandle please */
605 memset(&fhp
->fh_handle
.fh_base
, 0, NFS_FHSIZE
);
606 fhp
->fh_handle
.fh_size
= NFS_FHSIZE
;
607 fhp
->fh_handle
.ofh_dcookie
= 0xfeebbaca;
608 fhp
->fh_handle
.ofh_dev
= old_encode_dev(ex_dev
);
609 fhp
->fh_handle
.ofh_xdev
= fhp
->fh_handle
.ofh_dev
;
610 fhp
->fh_handle
.ofh_xino
=
611 ino_t_to_u32(d_inode(exp
->ex_path
.dentry
)->i_ino
);
612 fhp
->fh_handle
.ofh_dirino
= ino_t_to_u32(parent_ino(dentry
));
614 _fh_update_old(dentry
, exp
, &fhp
->fh_handle
);
616 fhp
->fh_handle
.fh_size
=
617 key_len(fhp
->fh_handle
.fh_fsid_type
) + 4;
618 fhp
->fh_handle
.fh_auth_type
= 0;
620 mk_fsid(fhp
->fh_handle
.fh_fsid_type
,
621 fhp
->fh_handle
.fh_fsid
,
623 d_inode(exp
->ex_path
.dentry
)->i_ino
,
624 exp
->ex_fsid
, exp
->ex_uuid
);
627 _fh_update(fhp
, exp
, dentry
);
628 if (fhp
->fh_handle
.fh_fileid_type
== FILEID_INVALID
) {
630 return nfserr_opnotsupp
;
638 * Update file handle information after changing a dentry.
639 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
642 fh_update(struct svc_fh
*fhp
)
644 struct dentry
*dentry
;
649 dentry
= fhp
->fh_dentry
;
650 if (d_really_is_negative(dentry
))
652 if (fhp
->fh_handle
.fh_version
!= 1) {
653 _fh_update_old(dentry
, fhp
->fh_export
, &fhp
->fh_handle
);
655 if (fhp
->fh_handle
.fh_fileid_type
!= FILEID_ROOT
)
658 _fh_update(fhp
, fhp
->fh_export
, dentry
);
659 if (fhp
->fh_handle
.fh_fileid_type
== FILEID_INVALID
)
660 return nfserr_opnotsupp
;
664 printk(KERN_ERR
"fh_update: fh not verified!\n");
665 return nfserr_serverfault
;
667 printk(KERN_ERR
"fh_update: %pd2 still negative!\n",
669 return nfserr_serverfault
;
673 * Release a file handle.
676 fh_put(struct svc_fh
*fhp
)
678 struct dentry
* dentry
= fhp
->fh_dentry
;
679 struct svc_export
* exp
= fhp
->fh_export
;
682 fhp
->fh_dentry
= NULL
;
689 fhp
->fh_export
= NULL
;
691 fhp
->fh_no_wcc
= false;
696 * Shorthand for dprintk()'s
698 char * SVCFH_fmt(struct svc_fh
*fhp
)
700 struct knfsd_fh
*fh
= &fhp
->fh_handle
;
703 sprintf(buf
, "%d: %08x %08x %08x %08x %08x %08x",
705 fh
->fh_base
.fh_pad
[0],
706 fh
->fh_base
.fh_pad
[1],
707 fh
->fh_base
.fh_pad
[2],
708 fh
->fh_base
.fh_pad
[3],
709 fh
->fh_base
.fh_pad
[4],
710 fh
->fh_base
.fh_pad
[5]);
714 enum fsid_source
fsid_source(const struct svc_fh
*fhp
)
716 if (fhp
->fh_handle
.fh_version
!= 1)
717 return FSIDSOURCE_DEV
;
718 switch(fhp
->fh_handle
.fh_fsid_type
) {
720 case FSID_ENCODE_DEV
:
721 case FSID_MAJOR_MINOR
:
722 if (exp_sb(fhp
->fh_export
)->s_type
->fs_flags
& FS_REQUIRES_DEV
)
723 return FSIDSOURCE_DEV
;
726 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
727 return FSIDSOURCE_FSID
;
732 /* either a UUID type filehandle, or the filehandle doesn't
735 if (fhp
->fh_export
->ex_flags
& NFSEXP_FSID
)
736 return FSIDSOURCE_FSID
;
737 if (fhp
->fh_export
->ex_uuid
)
738 return FSIDSOURCE_UUID
;
739 return FSIDSOURCE_DEV
;