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(&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;
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 svc_export *exp = NULL;
331 struct dentry *dentry;
334 dprintk("nfsd: fh_verify(%s)\n", SVCFH_fmt(fhp));
336 if (!fhp->fh_dentry) {
337 error = nfsd_set_fh_dentry(rqstp, fhp);
341 dentry = fhp->fh_dentry;
342 exp = fhp->fh_export;
344 * We still have to do all these permission checks, even when
345 * fh_dentry is already set:
346 * - fh_verify may be called multiple times with different
347 * "access" arguments (e.g. nfsd_proc_create calls
348 * fh_verify(...,NFSD_MAY_EXEC) first, then later (in
349 * nfsd_create) calls fh_verify(...,NFSD_MAY_CREATE).
350 * - in the NFSv4 case, the filehandle may have been filled
351 * in by fh_compose, and given a dentry, but further
352 * compound operations performed with that filehandle
353 * still need permissions checks. In the worst case, a
354 * mountpoint crossing may have changed the export
355 * options, and we may now need to use a different uid
356 * (for example, if different id-squashing options are in
357 * effect on the new filesystem).
359 error = check_pseudo_root(rqstp, dentry, exp);
363 error = nfsd_setuser_and_check_port(rqstp, exp);
367 error = nfsd_mode_check(rqstp, dentry, type);
372 * pseudoflavor restrictions are not enforced on NLM,
373 * which clients virtually always use auth_sys for,
374 * even while using RPCSEC_GSS for NFS.
376 if (access & NFSD_MAY_LOCK || access & NFSD_MAY_BYPASS_GSS)
377 goto skip_pseudoflavor_check;
379 * Clients may expect to be able to use auth_sys during mount,
380 * even if they use gss for everything else; see section 2.3.2
383 if (access & NFSD_MAY_BYPASS_GSS_ON_ROOT
384 && exp->ex_path.dentry == dentry)
385 goto skip_pseudoflavor_check;
387 error = check_nfsd_access(exp, rqstp);
391 skip_pseudoflavor_check:
392 /* Finally, check access permissions. */
393 error = nfsd_permission(rqstp, exp, dentry, access);
396 dprintk("fh_verify: %pd2 permission failure, "
397 "acc=%x, error=%d\n",
399 access, ntohl(error));
402 if (error == nfserr_stale)
403 nfsd_stats_fh_stale_inc(exp);
409 * Compose a file handle for an NFS reply.
411 * Note that when first composed, the dentry may not yet have
412 * an inode. In this case a call to fh_update should be made
413 * before the fh goes out on the wire ...
415 static void _fh_update(struct svc_fh *fhp, struct svc_export *exp,
416 struct dentry *dentry)
418 if (dentry != exp->ex_path.dentry) {
419 struct fid *fid = (struct fid *)
420 (fhp->fh_handle.fh_fsid + fhp->fh_handle.fh_size/4 - 1);
421 int maxsize = (fhp->fh_maxsize - fhp->fh_handle.fh_size)/4;
422 int subtreecheck = !(exp->ex_flags & NFSEXP_NOSUBTREECHECK);
424 fhp->fh_handle.fh_fileid_type =
425 exportfs_encode_fh(dentry, fid, &maxsize, subtreecheck);
426 fhp->fh_handle.fh_size += maxsize * 4;
428 fhp->fh_handle.fh_fileid_type = FILEID_ROOT;
432 static bool is_root_export(struct svc_export *exp)
434 return exp->ex_path.dentry == exp->ex_path.dentry->d_sb->s_root;
437 static struct super_block *exp_sb(struct svc_export *exp)
439 return exp->ex_path.dentry->d_sb;
442 static bool fsid_type_ok_for_exp(u8 fsid_type, struct svc_export *exp)
446 if (!old_valid_dev(exp_sb(exp)->s_dev))
449 case FSID_MAJOR_MINOR:
450 case FSID_ENCODE_DEV:
451 return exp_sb(exp)->s_type->fs_flags & FS_REQUIRES_DEV;
453 return exp->ex_flags & NFSEXP_FSID;
456 if (!is_root_export(exp))
459 case FSID_UUID4_INUM:
460 case FSID_UUID16_INUM:
461 return exp->ex_uuid != NULL;
467 static void set_version_and_fsid_type(struct svc_fh *fhp, struct svc_export *exp, struct svc_fh *ref_fh)
473 if (ref_fh && ref_fh->fh_export == exp) {
474 version = ref_fh->fh_handle.fh_version;
475 fsid_type = ref_fh->fh_handle.fh_fsid_type;
481 fsid_type = FSID_DEV;
490 * As the fsid -> filesystem mapping was guided by
491 * user-space, there is no guarantee that the filesystem
492 * actually supports that fsid type. If it doesn't we
493 * loop around again without ref_fh set.
495 if (!fsid_type_ok_for_exp(fsid_type, exp))
497 } else if (exp->ex_flags & NFSEXP_FSID) {
498 fsid_type = FSID_NUM;
499 } else if (exp->ex_uuid) {
500 if (fhp->fh_maxsize >= 64) {
501 if (is_root_export(exp))
502 fsid_type = FSID_UUID16;
504 fsid_type = FSID_UUID16_INUM;
506 if (is_root_export(exp))
507 fsid_type = FSID_UUID8;
509 fsid_type = FSID_UUID4_INUM;
511 } else if (!old_valid_dev(exp_sb(exp)->s_dev))
512 /* for newer device numbers, we must use a newer fsid format */
513 fsid_type = FSID_ENCODE_DEV;
515 fsid_type = FSID_DEV;
516 fhp->fh_handle.fh_version = version;
518 fhp->fh_handle.fh_fsid_type = fsid_type;
522 fh_compose(struct svc_fh *fhp, struct svc_export *exp, struct dentry *dentry,
523 struct svc_fh *ref_fh)
525 /* ref_fh is a reference file handle.
526 * if it is non-null and for the same filesystem, then we should compose
527 * a filehandle which is of the same version, where possible.
530 struct inode * inode = d_inode(dentry);
531 dev_t ex_dev = exp_sb(exp)->s_dev;
533 dprintk("nfsd: fh_compose(exp %02x:%02x/%ld %pd2, ino=%ld)\n",
534 MAJOR(ex_dev), MINOR(ex_dev),
535 (long) d_inode(exp->ex_path.dentry)->i_ino,
537 (inode ? inode->i_ino : 0));
539 /* Choose filehandle version and fsid type based on
540 * the reference filehandle (if it is in the same export)
541 * or the export options.
543 set_version_and_fsid_type(fhp, exp, ref_fh);
545 /* If we have a ref_fh, then copy the fh_no_wcc setting from it. */
546 fhp->fh_no_wcc = ref_fh ? ref_fh->fh_no_wcc : false;
551 if (fhp->fh_locked || fhp->fh_dentry) {
552 printk(KERN_ERR "fh_compose: fh %pd2 not initialized!\n",
555 if (fhp->fh_maxsize < NFS_FHSIZE)
556 printk(KERN_ERR "fh_compose: called with maxsize %d! %pd2\n",
560 fhp->fh_dentry = dget(dentry); /* our internal copy */
561 fhp->fh_export = exp_get(exp);
563 fhp->fh_handle.fh_size =
564 key_len(fhp->fh_handle.fh_fsid_type) + 4;
565 fhp->fh_handle.fh_auth_type = 0;
567 mk_fsid(fhp->fh_handle.fh_fsid_type,
568 fhp->fh_handle.fh_fsid,
570 d_inode(exp->ex_path.dentry)->i_ino,
571 exp->ex_fsid, exp->ex_uuid);
574 _fh_update(fhp, exp, dentry);
575 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID) {
577 return nfserr_opnotsupp;
584 * Update file handle information after changing a dentry.
585 * This is only called by nfsd_create, nfsd_create_v3 and nfsd_proc_create
588 fh_update(struct svc_fh *fhp)
590 struct dentry *dentry;
595 dentry = fhp->fh_dentry;
596 if (d_really_is_negative(dentry))
598 if (fhp->fh_handle.fh_fileid_type != FILEID_ROOT)
601 _fh_update(fhp, fhp->fh_export, dentry);
602 if (fhp->fh_handle.fh_fileid_type == FILEID_INVALID)
603 return nfserr_opnotsupp;
606 printk(KERN_ERR "fh_update: fh not verified!\n");
607 return nfserr_serverfault;
609 printk(KERN_ERR "fh_update: %pd2 still negative!\n",
611 return nfserr_serverfault;
614 #ifdef CONFIG_NFSD_V3
617 * fh_fill_pre_attrs - Fill in pre-op attributes
618 * @fhp: file handle to be updated
621 void fh_fill_pre_attrs(struct svc_fh *fhp)
623 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
628 if (fhp->fh_no_wcc || fhp->fh_pre_saved)
631 inode = d_inode(fhp->fh_dentry);
632 err = fh_getattr(fhp, &stat);
634 /* Grab the times from inode anyway */
635 stat.mtime = inode->i_mtime;
636 stat.ctime = inode->i_ctime;
637 stat.size = inode->i_size;
640 fhp->fh_pre_change = nfsd4_change_attribute(&stat, inode);
642 fhp->fh_pre_mtime = stat.mtime;
643 fhp->fh_pre_ctime = stat.ctime;
644 fhp->fh_pre_size = stat.size;
645 fhp->fh_pre_saved = true;
649 * fh_fill_post_attrs - Fill in post-op attributes
650 * @fhp: file handle to be updated
653 void fh_fill_post_attrs(struct svc_fh *fhp)
655 bool v4 = (fhp->fh_maxsize == NFS4_FHSIZE);
656 struct inode *inode = d_inode(fhp->fh_dentry);
662 if (fhp->fh_post_saved)
663 printk("nfsd: inode locked twice during operation.\n");
665 err = fh_getattr(fhp, &fhp->fh_post_attr);
667 fhp->fh_post_saved = false;
668 fhp->fh_post_attr.ctime = inode->i_ctime;
670 fhp->fh_post_saved = true;
672 fhp->fh_post_change =
673 nfsd4_change_attribute(&fhp->fh_post_attr, inode);
676 #endif /* CONFIG_NFSD_V3 */
679 * Release a file handle.
682 fh_put(struct svc_fh *fhp)
684 struct dentry * dentry = fhp->fh_dentry;
685 struct svc_export * exp = fhp->fh_export;
688 fhp->fh_dentry = NULL;
690 fh_clear_pre_post_attrs(fhp);
695 fhp->fh_export = NULL;
697 fhp->fh_no_wcc = false;
702 * Shorthand for dprintk()'s
704 char * SVCFH_fmt(struct svc_fh *fhp)
706 struct knfsd_fh *fh = &fhp->fh_handle;
707 static char buf[2+1+1+64*3+1];
709 if (fh->fh_size < 0 || fh->fh_size> 64)
711 sprintf(buf, "%d: %*ph", fh->fh_size, fh->fh_size, fh->fh_raw);
715 enum fsid_source fsid_source(const struct svc_fh *fhp)
717 if (fhp->fh_handle.fh_version != 1)
718 return FSIDSOURCE_DEV;
719 switch(fhp->fh_handle.fh_fsid_type) {
721 case FSID_ENCODE_DEV:
722 case FSID_MAJOR_MINOR:
723 if (exp_sb(fhp->fh_export)->s_type->fs_flags & FS_REQUIRES_DEV)
724 return FSIDSOURCE_DEV;
727 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
728 return FSIDSOURCE_FSID;
733 /* either a UUID type filehandle, or the filehandle doesn't
736 if (fhp->fh_export->ex_flags & NFSEXP_FSID)
737 return FSIDSOURCE_FSID;
738 if (fhp->fh_export->ex_uuid)
739 return FSIDSOURCE_UUID;
740 return FSIDSOURCE_DEV;