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 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;
158 struct svc_export *exp;
159 struct dentry *dentry;
161 int data_left = fh->fh_size/4;
165 error = nfserr_stale;
166 if (rqstp->rq_vers > 2)
167 error = nfserr_badhandle;
168 if (rqstp->rq_vers == 4 && fh->fh_size == 0)
169 return nfserr_nofilehandle;
171 if (fh->fh_version != 1)
176 if (fh->fh_auth_type != 0)
178 len = key_len(fh->fh_fsid_type) / 4;
181 if (fh->fh_fsid_type == FSID_MAJOR_MINOR) {
182 /* deprecated, convert to type 3 */
183 len = key_len(FSID_ENCODE_DEV)/4;
184 fh->fh_fsid_type = FSID_ENCODE_DEV;
186 * struct knfsd_fh uses host-endian fields, which are
187 * sometimes used to hold net-endian values. This
188 * confuses sparse, so we must use __force here to
189 * keep it from complaining.
191 fh->fh_fsid[0] = new_encode_dev(MKDEV(ntohl((__force __be32)fh->fh_fsid[0]),
192 ntohl((__force __be32)fh->fh_fsid[1])));
193 fh->fh_fsid[1] = fh->fh_fsid[2];
198 exp = rqst_exp_find(rqstp, fh->fh_fsid_type, fh->fh_fsid);
199 fid = (struct fid *)(fh->fh_fsid + len);
201 error = nfserr_stale;
203 trace_nfsd_set_fh_dentry_badexport(rqstp, fhp, PTR_ERR(exp));
205 if (PTR_ERR(exp) == -ENOENT)
208 return nfserrno(PTR_ERR(exp));
211 if (exp->ex_flags & NFSEXP_NOSUBTREECHECK) {
212 /* Elevate privileges so that the lack of 'r' or 'x'
213 * permission on some parent directory will
214 * not stop exportfs_decode_fh from being able
215 * to reconnect a directory into the dentry cache.
216 * The same problem can affect "SUBTREECHECK" exports,
217 * but as nfsd_acceptable depends on correct
218 * access control settings being in effect, we cannot
219 * fix that case easily.
221 struct cred *new = prepare_creds();
223 error = nfserrno(-ENOMEM);
227 cap_raise_nfsd_set(new->cap_effective,
229 put_cred(override_creds(new));
232 error = nfsd_setuser_and_check_port(rqstp, exp);
238 * Look up the dentry using the NFS file handle.
240 error = nfserr_stale;
241 if (rqstp->rq_vers > 2)
242 error = nfserr_badhandle;
244 fileid_type = fh->fh_fileid_type;
246 if (fileid_type == FILEID_ROOT)
247 dentry = dget(exp->ex_path.dentry);
249 dentry = exportfs_decode_fh_raw(exp->ex_path.mnt, fid,
250 data_left, fileid_type,
251 nfsd_acceptable, exp);
252 if (IS_ERR_OR_NULL(dentry)) {
253 trace_nfsd_set_fh_dentry_badhandle(rqstp, fhp,
254 dentry ? PTR_ERR(dentry) : -ESTALE);
255 switch (PTR_ERR(dentry)) {
260 dentry = ERR_PTR(-ESTALE);
266 if (IS_ERR(dentry)) {
267 if (PTR_ERR(dentry) != -EINVAL)
268 error = nfserrno(PTR_ERR(dentry));
272 if (d_is_dir(dentry) &&
273 (dentry->d_flags & DCACHE_DISCONNECTED)) {
274 printk("nfsd: find_fh_dentry returned a DISCONNECTED directory: %pd2\n",
278 fhp->fh_dentry = dentry;
279 fhp->fh_export = exp;
281 switch (rqstp->rq_vers) {
283 if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOATOMIC_ATTR)
284 fhp->fh_no_atomic_attr = true;
287 if (dentry->d_sb->s_export_op->flags & EXPORT_OP_NOWCC)
288 fhp->fh_no_wcc = true;
291 fhp->fh_no_wcc = true;
301 * fh_verify - filehandle lookup and access checking
302 * @rqstp: pointer to current rpc request
303 * @fhp: filehandle to be verified
304 * @type: expected type of object pointed to by filehandle
305 * @access: type of access needed to object
307 * Look up a dentry from the on-the-wire filehandle, check the client's
308 * access to the export, and set the current task's credentials.
310 * Regardless of success or failure of fh_verify(), fh_put() should be
311 * called on @fhp when the caller is finished with the filehandle.
313 * fh_verify() may be called multiple times on a given filehandle, for
314 * example, when processing an NFSv4 compound. The first call will look
315 * up a dentry using the on-the-wire filehandle. Subsequent calls will
316 * skip the lookup and just perform the other checks and possibly change
317 * the current task's credentials.
319 * @type specifies the type of object expected using one of the S_IF*
320 * constants defined in include/linux/stat.h. The caller may use zero
321 * to indicate that it doesn't care, or a negative integer to indicate
322 * that it expects something not of the given type.
324 * @access is formed from the NFSD_MAY_* constants defined in
328 fh_verify(struct svc_rqst *rqstp, struct svc_fh *fhp, umode_t type, int access)
330 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
331 struct svc_export *exp = NULL;
332 struct dentry *dentry;
335 if (!fhp->fh_dentry) {
336 error = nfsd_set_fh_dentry(rqstp, fhp);
340 dentry = fhp->fh_dentry;
341 exp = fhp->fh_export;
343 trace_nfsd_fh_verify(rqstp, fhp, type, access);
346 * We still have to do all these permission checks, even when
347 * fh_dentry is already set:
348 * - fh_verify may be called multiple times with different
349 * "access" arguments (e.g. nfsd_proc_create calls
350 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
351 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
352 * - in the NFSv4 case, the filehandle may have been filled
353 * in by fh_compose, and given a dentry, but further
354 * compound operations performed with that filehandle
355 * still need permissions checks. In the worst case, a
356 * mountpoint crossing may have changed the export
357 * options, and we may now need to use a different uid
358 * (for example, if different id-squashing options are in
359 * effect on the new filesystem).
361 error = check_pseudo_root(rqstp, dentry, exp);
365 error = nfsd_setuser_and_check_port(rqstp, exp);
369 error = nfsd_mode_check(rqstp, dentry, type);
374 * pseudoflavor restrictions are not enforced on NLM,
375 * which clients virtually always use auth_sys for,
376 * even while using RPCSEC_GSS for NFS.
378 if (access & NFSD_MAY_LOCK || access & NFSD_MAY_BYPASS_GSS)
379 goto skip_pseudoflavor_check;
381 * Clients may expect to be able to use auth_sys during mount,
382 * even if they use gss for everything else; see section 2.3.2
385 if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
386 && exp->ex_path.dentry == dentry)
387 goto skip_pseudoflavor_check;
389 error = check_nfsd_access(exp, rqstp);
393 skip_pseudoflavor_check:
394 /* Finally, check access permissions. */
395 error = nfsd_permission(rqstp, exp, dentry, access);
397 trace_nfsd_fh_verify_err(rqstp, fhp, type, access, error);
398 if (error == nfserr_stale)
399 nfsd_stats_fh_stale_inc(nn, exp);
405 * Compose a file handle for an NFS reply.
407 * Note that when first composed, the dentry may not yet have
408 * an inode. In this case a call to fh_update should be made
409 * before the fh goes out on the wire ...
411 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
412 struct dentry *dentry)
414 if (dentry != exp->ex_path.dentry) {
415 struct fid *fid = (struct fid *)
416 (fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1);
417 int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
418 int fh_flags = (exp->ex_flags & NFSEXP_NOSUBTREECHECK) ? 0 :
419 EXPORT_FH_CONNECTABLE;
421 exportfs_encode_fh(dentry, fid, &maxsize, fh_flags);
423 fhp->fh_handle.fh_fileid_type =
424 fileid_type > 0 ? fileid_type : FILEID_INVALID;
425 fhp->fh_handle.fh_size += maxsize * 4;
427 fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
431 static bool is_root_export(struct svc_export *exp)
433 return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
436 static struct super_block *exp_sb(struct svc_export *exp)
438 return exp->ex_path.dentry->d_sb;
441 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
445 if (!old_valid_dev(exp_sb(exp)->s_dev))
448 case FSID_MAJOR_MINOR:
449 case FSID_ENCODE_DEV:
450 return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
452 return exp->ex_flags & NFSEXP_FSID;
455 if (!is_root_export(exp))
458 case FSID_UUID4_INUM:
459 case FSID_UUID16_INUM:
460 return exp->ex_uuid != NULL;
466 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
472 if (ref_fh && ref_fh->fh_export == exp) {
473 version = ref_fh->fh_handle.fh_version;
474 fsid_type = ref_fh->fh_handle.fh_fsid_type;
480 fsid_type = FSID_DEV;
489 * As the fsid -> filesystem mapping was guided by
490 * user-space, there is no guarantee that the filesystem
491 * actually supports that fsid type. If it doesn't we
492 * loop around again without ref_fh set.
494 if (!fsid_type_ok_for_exp(fsid_type, exp))
496 } else if (exp->ex_flags & NFSEXP_FSID) {
497 fsid_type = FSID_NUM;
498 } else if (exp->ex_uuid) {
499 if (fhp->fh_maxsize >= 64) {
500 if (is_root_export(exp))
501 fsid_type = FSID_UUID16;
503 fsid_type = FSID_UUID16_INUM;
505 if (is_root_export(exp))
506 fsid_type = FSID_UUID8;
508 fsid_type = FSID_UUID4_INUM;
510 } else if (!old_valid_dev(exp_sb(exp)->s_dev))
511 /* for newer device numbers, we must use a newer fsid format */
512 fsid_type = FSID_ENCODE_DEV;
514 fsid_type = FSID_DEV;
515 fhp->fh_handle.fh_version = version;
517 fhp->fh_handle.fh_fsid_type = fsid_type;
521 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
522 struct svc_fh *ref_fh)
524 /* ref_fh is a reference file handle.
525 * if it is non-null and for the same filesystem, then we should compose
526 * a filehandle which is of the same version, where possible.
529 struct inode * inode = d_inode(dentry);
530 dev_t ex_dev = exp_sb(exp)->s_dev;
532 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
533 MAJOR(ex_dev), MINOR(ex_dev),
534 (long) d_inode(exp->ex_path.dentry)->i_ino,
536 (inode ? inode->i_ino : 0));
538 /* Choose filehandle version and fsid type based on
539 * the reference filehandle (if it is in the same export)
540 * or the export options.
542 set_version_and_fsid_type(fhp, exp, ref_fh);
544 /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
545 fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
550 if (fhp->fh_dentry) {
551 printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
554 if (fhp->fh_maxsize < NFS_FHSIZE)
555 printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
559 fhp->fh_dentry = dget(dentry); /* our internal copy */
560 fhp->fh_export = exp_get(exp);
562 fhp->fh_handle.fh_size =
563 key_len(fhp->fh_handle.fh_fsid_type) + 4;
564 fhp->fh_handle.fh_auth_type = 0;
566 mk_fsid(fhp->fh_handle.fh_fsid_type,
567 fhp->fh_handle.fh_fsid,
569 d_inode(exp->ex_path.dentry)->i_ino,
570 exp->ex_fsid, exp->ex_uuid);
573 _fh_update(fhp, exp, dentry);
574 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
576 return nfserr_opnotsupp;
583 * Update file handle information after changing a dentry.
584 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
587 fh_update(struct svc_fh *fhp)
589 struct dentry *dentry;
594 dentry = fhp->fh_dentry;
595 if (d_really_is_negative(dentry))
597 if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
600 _fh_update(fhp, fhp->fh_export, dentry);
601 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
602 return nfserr_opnotsupp;
605 printk(KERN_ERR "fh_update: fh not verified!\n");
606 return nfserr_serverfault;
608 printk(KERN_ERR "fh_update: %pd2 still negative!\n",
610 return nfserr_serverfault;
614 * fh_fill_pre_attrs - Fill in pre-op attributes
615 * @fhp: file handle to be updated
618 __be32 __must_check fh_fill_pre_attrs(struct svc_fh *fhp)
620 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
625 if (fhp->fh_no_wcc || fhp->fh_pre_saved)
628 inode = d_inode(fhp->fh_dentry);
629 err = fh_getattr(fhp, &stat);
634 fhp->fh_pre_change = nfsd4_change_attribute(&stat, inode);
636 fhp->fh_pre_mtime = stat.mtime;
637 fhp->fh_pre_ctime = stat.ctime;
638 fhp->fh_pre_size = stat.size;
639 fhp->fh_pre_saved = true;
644 * fh_fill_post_attrs - Fill in post-op attributes
645 * @fhp: file handle to be updated
648 __be32 fh_fill_post_attrs(struct svc_fh *fhp)
650 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
651 struct inode *inode = d_inode(fhp->fh_dentry);
657 if (fhp->fh_post_saved)
658 printk("nfsd: inode locked twice during operation.\n");
660 err = fh_getattr(fhp, &fhp->fh_post_attr);
664 fhp->fh_post_saved = true;
666 fhp->fh_post_change =
667 nfsd4_change_attribute(&fhp->fh_post_attr, inode);
672 * fh_fill_both_attrs - Fill pre-op and post-op attributes
673 * @fhp: file handle to be updated
675 * This is used when the directory wasn't changed, but wcc attributes
678 __be32 __must_check fh_fill_both_attrs(struct svc_fh *fhp)
682 err = fh_fill_post_attrs(fhp);
686 fhp->fh_pre_change = fhp->fh_post_change;
687 fhp->fh_pre_mtime = fhp->fh_post_attr.mtime;
688 fhp->fh_pre_ctime = fhp->fh_post_attr.ctime;
689 fhp->fh_pre_size = fhp->fh_post_attr.size;
690 fhp->fh_pre_saved = true;
695 * Release a file handle.
698 fh_put(struct svc_fh *fhp)
700 struct dentry * dentry = fhp->fh_dentry;
701 struct svc_export * exp = fhp->fh_export;
703 fhp->fh_dentry = NULL;
705 fh_clear_pre_post_attrs(fhp);
710 fhp->fh_export = NULL;
712 fhp->fh_no_wcc = false;
717 * Shorthand for dprintk()'s
719 char * SVCFH_fmt(struct svc_fh *fhp)
721 struct knfsd_fh *fh = &fhp->fh_handle;
722 static char buf[2+1+1+64*3+1];
724 if (fh->fh_size < 0 || fh->fh_size> 64)
726 sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
730 enum fsid_source fsid_source(const struct svc_fh *fhp)
732 if (fhp->fh_handle.fh_version != 1)
733 return FSIDSOURCE_DEV;
734 switch(fhp->fh_handle.fh_fsid_type) {
736 case FSID_ENCODE_DEV:
737 case FSID_MAJOR_MINOR:
738 if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
739 return FSIDSOURCE_DEV;
742 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
743 return FSIDSOURCE_FSID;
748 /* either a UUID type filehandle, or the filehandle doesn't
751 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
752 return FSIDSOURCE_FSID;
753 if (fhp->fh_export->ex_uuid)
754 return FSIDSOURCE_UUID;
755 return FSIDSOURCE_DEV;
759 * We could use i_version alone as the change attribute. However, i_version
760 * can go backwards on a regular file after an unclean shutdown. On its own
761 * that doesn't necessarily cause a problem, but if i_version goes backwards
762 * and then is incremented again it could reuse a value that was previously
763 * used before boot, and a client who queried the two values might incorrectly
764 * assume nothing changed.
766 * By using both ctime and the i_version counter we guarantee that as long as
767 * time doesn't go backwards we never reuse an old value. If the filesystem
768 * advertises STATX_ATTR_CHANGE_MONOTONIC, then this mitigation is not
771 * We only need to do this for regular files as well. For directories, we
772 * assume that the new change attr is always logged to stable storage in some
773 * fashion before the results can be seen.
775 u64 nfsd4_change_attribute(const struct kstat *stat, const struct inode *inode)
779 if (stat->result_mask & STATX_CHANGE_COOKIE) {
780 chattr = stat->change_cookie;
781 if (S_ISREG(inode->i_mode) &&
782 !(stat->attributes & STATX_ATTR_CHANGE_MONOTONIC)) {
783 chattr += (u64)stat->ctime.tv_sec << 30;
784 chattr += stat->ctime.tv_nsec;
787 chattr = time_to_chattr(&stat->ctime);