1 // SPDX-License-Identifier: GPL-2.0
3 * NFS server file handle treatment.
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(&nop_mnt_idmap,
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 dentry *dentry, umode_t requested)
67 umode_t mode = d_inode(dentry)->i_mode & S_IFMT;
69 if (requested == 0) /* the caller doesn't care */
71 if (mode == requested) {
72 if (mode == S_IFDIR && !d_can_lookup(dentry)) {
78 if (mode == S_IFLNK) {
79 if (requested == S_IFDIR)
80 return nfserr_symlink_not_dir;
81 return nfserr_symlink;
83 if (requested == S_IFDIR)
87 return nfserr_wrong_type;
90 static bool nfsd_originating_port_ok(struct svc_rqst *rqstp,
91 struct svc_cred *cred,
92 struct svc_export *exp)
94 if (nfsexp_flags(cred, exp) & NFSEXP_INSECURE_PORT)
96 /* We don't require gss requests to use low ports: */
97 if (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_cred *cred,
104 struct svc_export *exp)
106 /* Check if the request originated from a secure port. */
107 if (rqstp && !nfsd_originating_port_ok(rqstp, cred, exp)) {
108 RPC_IFDEBUG(char buf[RPC_MAX_ADDRBUFLEN]);
109 dprintk("nfsd: request from insecure port %s!\n",
110 svc_print_addr(rqstp, buf, sizeof(buf)));
114 /* Set user creds for this exportpoint */
115 return nfserrno(nfsd_setuser(cred, exp));
118 static inline __be32 check_pseudo_root(struct dentry *dentry,
119 struct svc_export *exp)
121 if (!(exp->ex_flags & NFSEXP_V4ROOT))
124 * We're exposing only the directories and symlinks that have to be
125 * traversed on the way to real exports:
127 if (unlikely(!d_is_dir(dentry) &&
128 !d_is_symlink(dentry)))
131 * A pseudoroot export gives permission to access only one
132 * single directory; the kernel has to make another upcall
133 * before granting access to anything else under it:
135 if (unlikely(dentry != exp->ex_path.dentry))
141 * Use the given filehandle to look up the corresponding export and
142 * dentry. On success, the results are used to set fh_export and
145 static __be32 nfsd_set_fh_dentry(struct svc_rqst *rqstp, struct net *net,
146 struct svc_cred *cred,
147 struct auth_domain *client,
148 struct auth_domain *gssclient,
151 struct knfsd_fh *fh = &fhp->fh_handle;
152 struct fid *fid = NULL;
153 struct svc_export *exp;
154 struct dentry *dentry;
156 int data_left = fh->fh_size/4;
160 error = nfserr_badhandle;
161 if (fh->fh_size == 0)
162 return nfserr_nofilehandle;
164 if (fh->fh_version != 1)
169 if (fh->fh_auth_type != 0)
171 len = key_len(fh->fh_fsid_type) / 4;
174 if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
175 /* deprecated, convert to type 3 */
176 len = key_len(FSID_ENCODE_DEV)/4;
177 fh->fh_fsid_type = FSID_ENCODE_DEV;
179 * struct knfsd_fh uses host-endian fields, which are
180 * sometimes used to hold net-endian values. This
181 * confuses sparse, so we must use __force here to
182 * keep it from complaining.
184 fh->fh_fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fh->fh_fsid[0]),
185 ntohl((__force __be32)fh->fh_fsid[1])));
186 fh->fh_fsid[1] = fh->fh_fsid[2];
191 exp = rqst_exp_find(rqstp ? &rqstp->rq_chandle : NULL,
192 net, client, gssclient,
193 fh->fh_fsid_type, fh->fh_fsid);
194 fid = (struct fid *)(fh->fh_fsid + len);
196 error = nfserr_stale;
198 trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
200 if (PTR_ERR(exp) == -ENOENT)
203 return nfserrno(PTR_ERR(exp));
206 if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
207 /* Elevate privileges so that the lack of 'r' or 'x'
208 * permission on some parent directory will
209 * not stop exportfs_decode_fh from being able
210 * to reconnect a directory into the dentry cache.
211 * The same problem can affect "SUBTREECHECK" exports,
212 * but as nfsd_acceptable depends on correct
213 * access control settings being in effect, we cannot
214 * fix that case easily.
216 struct cred *new = prepare_creds();
218 error = nfserrno(-ENOMEM);
222 cap_raise_nfsd_set(new->cap_effective,
224 put_cred(override_creds(new));
226 error = nfsd_setuser_and_check_port(rqstp, cred, exp);
232 * Look up the dentry using the NFS file handle.
234 error = nfserr_badhandle;
236 fileid_type = fh->fh_fileid_type;
238 if (fileid_type == FILEID_ROOT)
239 dentry = dget(exp->ex_path.dentry);
241 dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
242 data_left, fileid_type, 0,
243 nfsd_acceptable, exp);
244 if (IS_ERR_OR_NULL(dentry)) {
245 trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
246 dentry ? PTR_ERR(dentry) : -ESTALE);
247 switch (PTR_ERR(dentry)) {
252 dentry = ERR_PTR(-ESTALE);
258 if (IS_ERR(dentry)) {
259 if (PTR_ERR(dentry) != -EINVAL)
260 error = nfserrno(PTR_ERR(dentry));
264 if (d_is_dir(dentry) &&
265 (dentry->d_flags & DCACHE_DISCONNECTED)) {
266 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
270 fhp->fh_dentry = dentry;
271 fhp->fh_export = exp;
273 switch (fhp->fh_maxsize) {
275 if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
276 fhp->fh_no_atomic_attr = true;
277 fhp->fh_64bit_cookies = true;
280 if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
281 fhp->fh_no_wcc = true;
282 fhp->fh_64bit_cookies = true;
283 if (exp->ex_flags & NFSEXP_V4ROOT)
287 fhp->fh_no_wcc = true;
289 fhp->fh_use_wgather = true;
290 if (exp->ex_flags & NFSEXP_V4ROOT)
301 * __fh_verify - filehandle lookup and access checking
302 * @rqstp: RPC transaction context, or NULL
303 * @net: net namespace in which to perform the export lookup
304 * @cred: RPC user credential
305 * @client: RPC auth domain
306 * @gssclient: RPC GSS auth domain, or NULL
307 * @fhp: filehandle to be verified
308 * @type: expected type of object pointed to by filehandle
309 * @access: type of access needed to object
311 * See fh_verify() for further descriptions of @fhp, @type, and @access.
314 __fh_verify(struct svc_rqst *rqstp,
315 struct net *net, struct svc_cred *cred,
316 struct auth_domain *client,
317 struct auth_domain *gssclient,
318 struct svc_fh *fhp, umode_t type, int access)
320 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
321 struct svc_export *exp = NULL;
322 bool may_bypass_gss = false;
323 struct dentry *dentry;
326 if (!fhp->fh_dentry) {
327 error = nfsd_set_fh_dentry(rqstp, net, cred, client,
332 dentry = fhp->fh_dentry;
333 exp = fhp->fh_export;
335 trace_nfsd_fh_verify(rqstp, fhp, type, access);
338 * We still have to do all these permission checks, even when
339 * fh_dentry is already set:
340 * - fh_verify may be called multiple times with different
341 * "access" arguments (e.g. nfsd_proc_create calls
342 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
343 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
344 * - in the NFSv4 case, the filehandle may have been filled
345 * in by fh_compose, and given a dentry, but further
346 * compound operations performed with that filehandle
347 * still need permissions checks. In the worst case, a
348 * mountpoint crossing may have changed the export
349 * options, and we may now need to use a different uid
350 * (for example, if different id-squashing options are in
351 * effect on the new filesystem).
353 error = check_pseudo_root(dentry, exp);
357 error = nfsd_setuser_and_check_port(rqstp, cred, exp);
361 error = nfsd_mode_check(dentry, type);
365 if ((access & NFSD_MAY_NLM) && (exp->ex_flags & NFSEXP_NOAUTHNLM))
366 /* NLM is allowed to fully bypass authentication */
369 if (access & NFSD_MAY_BYPASS_GSS)
370 may_bypass_gss = true;
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 may_bypass_gss = true;
380 error = check_nfsd_access(exp, rqstp, may_bypass_gss);
383 /* During LOCALIO call to fh_verify will be called with a NULL rqstp */
385 svc_xprt_set_valid(rqstp->rq_xprt);
387 /* Finally, check access permissions. */
388 error = nfsd_permission(cred, exp, dentry, access);
390 trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
391 if (error == nfserr_stale)
392 nfsd_stats_fh_stale_inc(nn, exp);
397 * fh_verify_local - filehandle lookup and access checking
398 * @net: net namespace in which to perform the export lookup
399 * @cred: RPC user credential
400 * @client: RPC auth domain
401 * @fhp: filehandle to be verified
402 * @type: expected type of object pointed to by filehandle
403 * @access: type of access needed to object
405 * This API can be used by callers who do not have an RPC
406 * transaction context (ie are not running in an nfsd thread).
408 * See fh_verify() for further descriptions of @fhp, @type, and @access.
411 fh_verify_local(struct net *net, struct svc_cred *cred,
412 struct auth_domain *client, struct svc_fh *fhp,
413 umode_t type, int access)
415 return __fh_verify(NULL, net, cred, client, NULL,
420 * fh_verify - filehandle lookup and access checking
421 * @rqstp: pointer to current rpc request
422 * @fhp: filehandle to be verified
423 * @type: expected type of object pointed to by filehandle
424 * @access: type of access needed to object
426 * Look up a dentry from the on-the-wire filehandle, check the client's
427 * access to the export, and set the current task's credentials.
429 * Regardless of success or failure of fh_verify(), fh_put() should be
430 * called on @fhp when the caller is finished with the filehandle.
432 * fh_verify() may be called multiple times on a given filehandle, for
433 * example, when processing an NFSv4 compound. The first call will look
434 * up a dentry using the on-the-wire filehandle. Subsequent calls will
435 * skip the lookup and just perform the other checks and possibly change
436 * the current task's credentials.
438 * @type specifies the type of object expected using one of the S_IF*
439 * constants defined in include/linux/stat.h. The caller may use zero
440 * to indicate that it doesn't care, or a negative integer to indicate
441 * that it expects something not of the given type.
443 * @access is formed from the NFSD_MAY_* constants defined in
447 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
449 return __fh_verify(rqstp, SVC_NET(rqstp), &rqstp->rq_cred,
450 rqstp->rq_client, rqstp->rq_gssclient,
455 * Compose a file handle for an NFS reply.
457 * Note that when first composed, the dentry may not yet have
458 * an inode. In this case a call to fh_update should be made
459 * before the fh goes out on the wire ...
461 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
462 struct dentry *dentry)
464 if (dentry != exp->ex_path.dentry) {
465 struct fid *fid = (struct fid *)
466 (fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1);
467 int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
468 int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
469 EXPORT_FH_CONNECTABLE;
471 exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
473 fhp->fh_handle.fh_fileid_type =
474 fileid_type > 0 ? fileid_type : FILEID_INVALID;
475 fhp->fh_handle.fh_size += maxsize * 4;
477 fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
481 static bool is_root_export(struct svc_export *exp)
483 return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
486 static struct super_block *exp_sb(struct svc_export *exp)
488 return exp->ex_path.dentry->d_sb;
491 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
495 if (!old_valid_dev(exp_sb(exp)->s_dev))
498 case FSID_MAJOR_MINOR:
499 case FSID_ENCODE_DEV:
500 return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
502 return exp->ex_flags & NFSEXP_FSID;
505 if (!is_root_export(exp))
508 case FSID_UUID4_INUM:
509 case FSID_UUID16_INUM:
510 return exp->ex_uuid != NULL;
516 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
522 if (ref_fh && ref_fh->fh_export == exp) {
523 version = ref_fh->fh_handle.fh_version;
524 fsid_type = ref_fh->fh_handle.fh_fsid_type;
530 fsid_type = FSID_DEV;
539 * As the fsid -> filesystem mapping was guided by
540 * user-space, there is no guarantee that the filesystem
541 * actually supports that fsid type. If it doesn't we
542 * loop around again without ref_fh set.
544 if (!fsid_type_ok_for_exp(fsid_type, exp))
546 } else if (exp->ex_flags & NFSEXP_FSID) {
547 fsid_type = FSID_NUM;
548 } else if (exp->ex_uuid) {
549 if (fhp->fh_maxsize >= 64) {
550 if (is_root_export(exp))
551 fsid_type = FSID_UUID16;
553 fsid_type = FSID_UUID16_INUM;
555 if (is_root_export(exp))
556 fsid_type = FSID_UUID8;
558 fsid_type = FSID_UUID4_INUM;
560 } else if (!old_valid_dev(exp_sb(exp)->s_dev))
561 /* for newer device numbers, we must use a newer fsid format */
562 fsid_type = FSID_ENCODE_DEV;
564 fsid_type = FSID_DEV;
565 fhp->fh_handle.fh_version = version;
567 fhp->fh_handle.fh_fsid_type = fsid_type;
571 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
572 struct svc_fh *ref_fh)
574 /* ref_fh is a reference file handle.
575 * if it is non-null and for the same filesystem, then we should compose
576 * a filehandle which is of the same version, where possible.
579 struct inode * inode = d_inode(dentry);
580 dev_t ex_dev = exp_sb(exp)->s_dev;
582 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
583 MAJOR(ex_dev), MINOR(ex_dev),
584 (long) d_inode(exp->ex_path.dentry)->i_ino,
586 (inode ? inode->i_ino : 0));
588 /* Choose filehandle version and fsid type based on
589 * the reference filehandle (if it is in the same export)
590 * or the export options.
592 set_version_and_fsid_type(fhp, exp, ref_fh);
594 /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
595 fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
600 if (fhp->fh_dentry) {
601 printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
604 if (fhp->fh_maxsize < NFS_FHSIZE)
605 printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
609 fhp->fh_dentry = dget(dentry); /* our internal copy */
610 fhp->fh_export = exp_get(exp);
612 fhp->fh_handle.fh_size =
613 key_len(fhp->fh_handle.fh_fsid_type) + 4;
614 fhp->fh_handle.fh_auth_type = 0;
616 mk_fsid(fhp->fh_handle.fh_fsid_type,
617 fhp->fh_handle.fh_fsid,
619 d_inode(exp->ex_path.dentry)->i_ino,
620 exp->ex_fsid, exp->ex_uuid);
623 _fh_update(fhp, exp, dentry);
624 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
633 * Update file handle information after changing a dentry.
634 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
637 fh_update(struct svc_fh *fhp)
639 struct dentry *dentry;
644 dentry = fhp->fh_dentry;
645 if (d_really_is_negative(dentry))
647 if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
650 _fh_update(fhp, fhp->fh_export, dentry);
651 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
655 printk(KERN_ERR "fh_update: fh not verified!\n");
656 return nfserr_serverfault;
658 printk(KERN_ERR "fh_update: %pd2 still negative!\n",
660 return nfserr_serverfault;
664 * fh_fill_pre_attrs - Fill in pre-op attributes
665 * @fhp: file handle to be updated
668 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
670 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
674 if (fhp->fh_no_wcc || fhp->fh_pre_saved)
677 err = fh_getattr(fhp, &stat);
682 fhp->fh_pre_change = nfsd4_change_attribute(&stat);
684 fhp->fh_pre_mtime = stat.mtime;
685 fhp->fh_pre_ctime = stat.ctime;
686 fhp->fh_pre_size = stat.size;
687 fhp->fh_pre_saved = true;
692 * fh_fill_post_attrs - Fill in post-op attributes
693 * @fhp: file handle to be updated
696 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
698 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
704 if (fhp->fh_post_saved)
705 printk("nfsd: inode locked twice during operation.\n");
707 err = fh_getattr(fhp, &fhp->fh_post_attr);
711 fhp->fh_post_saved = true;
713 fhp->fh_post_change =
714 nfsd4_change_attribute(&fhp->fh_post_attr);
719 * fh_fill_both_attrs - Fill pre-op and post-op attributes
720 * @fhp: file handle to be updated
722 * This is used when the directory wasn't changed, but wcc attributes
725 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
729 err = fh_fill_post_attrs(fhp);
733 fhp->fh_pre_change = fhp->fh_post_change;
734 fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
735 fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
736 fhp->fh_pre_size = fhp->fh_post_attr.size;
737 fhp->fh_pre_saved = true;
742 * Release a file handle.
745 fh_put(struct svc_fh *fhp)
747 struct dentry * dentry = fhp->fh_dentry;
748 struct svc_export * exp = fhp->fh_export;
750 fhp->fh_dentry = NULL;
752 fh_clear_pre_post_attrs(fhp);
757 fhp->fh_export = NULL;
759 fhp->fh_no_wcc = false;
764 * Shorthand for dprintk()'s
766 char * SVCFH_fmt(struct svc_fh *fhp)
768 struct knfsd_fh *fh = &fhp->fh_handle;
769 static char buf[2+1+1+64*3+1];
771 if (fh->fh_size > 64)
773 sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
777 enum fsid_source fsid_source(const struct svc_fh *fhp)
779 if (fhp->fh_handle.fh_version != 1)
780 return FSIDSOURCE_DEV;
781 switch(fhp->fh_handle.fh_fsid_type) {
783 case FSID_ENCODE_DEV:
784 case FSID_MAJOR_MINOR:
785 if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
786 return FSIDSOURCE_DEV;
789 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
790 return FSIDSOURCE_FSID;
795 /* either a UUID type filehandle, or the filehandle doesn't
798 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
799 return FSIDSOURCE_FSID;
800 if (fhp->fh_export->ex_uuid)
801 return FSIDSOURCE_UUID;
802 return FSIDSOURCE_DEV;
806 * nfsd4_change_attribute - Generate an NFSv4 change_attribute value
807 * @stat: inode attributes
809 * Caller must fill in @stat before calling, typically by invoking
810 * vfs_getattr() with STATX_MODE, STATX_CTIME, and STATX_CHANGE_COOKIE.
811 * Returns an unsigned 64-bit changeid4 value (RFC 8881 Section 3.2).
813 * We could use i_version alone as the change attribute. However, i_version
814 * can go backwards on a regular file after an unclean shutdown. On its own
815 * that doesn't necessarily cause a problem, but if i_version goes backwards
816 * and then is incremented again it could reuse a value that was previously
817 * used before boot, and a client who queried the two values might incorrectly
818 * assume nothing changed.
820 * By using both ctime and the i_version counter we guarantee that as long as
821 * time doesn't go backwards we never reuse an old value. If the filesystem
822 * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
825 * We only need to do this for regular files as well. For directories, we
826 * assume that the new change attr is always logged to stable storage in some
827 * fashion before the results can be seen.
829 u64 nfsd4_change_attribute(const struct kstat *stat)
833 if (stat->result_mask & STATX_CHANGE_COOKIE) {
834 chattr = stat->change_cookie;
835 if (S_ISREG(stat->mode) &&
836 !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
837 chattr += (u64)stat->ctime.tv_sec << 30;
838 chattr += stat->ctime.tv_nsec;
841 chattr = time_to_chattr(&stat->ctime);