2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
39 #endif /* CONFIG_NFSD_V4 */
44 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
48 * This is a cache of readahead params that help us choose the proper
49 * readahead strategy. Initially, we set all readahead parameters to 0
50 * and let the VFS handle things.
51 * If you increase the number of cached files very much, you'll need to
52 * add a hash table here.
55 struct raparms
*p_next
;
60 struct file_ra_state p_ra
;
61 unsigned int p_hindex
;
64 struct raparm_hbucket
{
65 struct raparms
*pb_head
;
67 } ____cacheline_aligned_in_smp
;
69 #define RAPARM_HASH_BITS 4
70 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
71 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
72 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
75 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
77 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
78 * or nfs_ok having possibly changed *dpp and *expp
81 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
82 struct svc_export
**expp
)
84 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
85 struct dentry
*dentry
= *dpp
;
86 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
87 .dentry
= dget(dentry
)};
90 err
= follow_down(&path
);
94 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
104 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
109 if (nfsd_v4client(rqstp
) ||
110 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
119 path
.dentry
= dentry
;
129 static void follow_to_parent(struct path
*path
)
133 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
135 dp
= dget_parent(path
->dentry
);
140 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
142 struct svc_export
*exp2
;
143 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
144 .dentry
= dget(dparent
)};
146 follow_to_parent(&path
);
148 exp2
= rqst_exp_parent(rqstp
, &path
);
149 if (PTR_ERR(exp2
) == -ENOENT
) {
150 *dentryp
= dget(dparent
);
151 } else if (IS_ERR(exp2
)) {
153 return PTR_ERR(exp2
);
155 *dentryp
= dget(path
.dentry
);
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
169 if (d_mountpoint(dentry
))
171 if (nfsd4_is_junction(dentry
))
173 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
175 return dentry
->d_inode
!= NULL
;
179 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
180 const char *name
, unsigned int len
,
181 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
183 struct svc_export
*exp
;
184 struct dentry
*dparent
;
185 struct dentry
*dentry
;
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
190 dparent
= fhp
->fh_dentry
;
191 exp
= fhp
->fh_export
;
194 /* Lookup the name, but don't follow links */
195 if (isdotent(name
, len
)) {
197 dentry
= dget(dparent
);
198 else if (dparent
!= exp
->ex_path
.dentry
)
199 dentry
= dget_parent(dparent
);
200 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
201 dentry
= dget(dparent
); /* .. == . just like at / */
203 /* checking mountpoint crossing is very different when stepping up */
204 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
210 dentry
= lookup_one_len(name
, dparent
, len
);
211 host_err
= PTR_ERR(dentry
);
215 * check if we have crossed a mount point ...
217 if (nfsd_mountpoint(dentry
, exp
)) {
218 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
224 *dentry_ret
= dentry
;
230 return nfserrno(host_err
);
234 * Look up one component of a pathname.
235 * N.B. After this call _both_ fhp and resfh need an fh_put
237 * If the lookup would cross a mountpoint, and the mounted filesystem
238 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
239 * accepted as it stands and the mounted directory is
240 * returned. Otherwise the covered directory is returned.
241 * NOTE: this mountpoint crossing is not supported properly by all
242 * clients and is explicitly disallowed for NFSv3
243 * NeilBrown <neilb@cse.unsw.edu.au>
246 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
247 unsigned int len
, struct svc_fh
*resfh
)
249 struct svc_export
*exp
;
250 struct dentry
*dentry
;
253 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
256 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
259 err
= check_nfsd_access(exp
, rqstp
);
263 * Note: we compose the file handle now, but as the
264 * dentry may be negative, it may need to be updated.
266 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
267 if (!err
&& !dentry
->d_inode
)
275 static int nfsd_break_lease(struct inode
*inode
)
277 if (!S_ISREG(inode
->i_mode
))
279 return break_lease(inode
, O_WRONLY
| O_NONBLOCK
);
283 * Commit metadata changes to stable storage.
286 commit_metadata(struct svc_fh
*fhp
)
288 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
289 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
291 if (!EX_ISSYNC(fhp
->fh_export
))
294 if (export_ops
->commit_metadata
)
295 return export_ops
->commit_metadata(inode
);
296 return sync_inode_metadata(inode
, 1);
300 * Go over the attributes and take care of the small differences between
301 * NFS semantics and what Linux expects.
304 nfsd_sanitize_attrs(struct inode
*inode
, struct iattr
*iap
)
307 * NFSv2 does not differentiate between "set-[ac]time-to-now"
308 * which only requires access, and "set-[ac]time-to-X" which
309 * requires ownership.
310 * So if it looks like it might be "set both to the same time which
311 * is close to now", and if inode_change_ok fails, then we
312 * convert to "set to now" instead of "set to explicit time"
314 * We only call inode_change_ok as the last test as technically
315 * it is not an interface that we should be using.
317 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
318 #define MAX_TOUCH_TIME_ERROR (30*60)
319 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
320 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
324 * Now just make sure time is in the right ballpark.
325 * Solaris, at least, doesn't seem to care what the time
326 * request is. We require it be within 30 minutes of now.
328 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
331 if (delta
< MAX_TOUCH_TIME_ERROR
&&
332 inode_change_ok(inode
, iap
) != 0) {
334 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
335 * This will cause notify_change to set these times
338 iap
->ia_valid
&= ~BOTH_TIME_SET
;
342 /* sanitize the mode change */
343 if (iap
->ia_valid
& ATTR_MODE
) {
344 iap
->ia_mode
&= S_IALLUGO
;
345 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
348 /* Revoke setuid/setgid on chown */
349 if (!S_ISDIR(inode
->i_mode
) &&
350 (((iap
->ia_valid
& ATTR_UID
) && !uid_eq(iap
->ia_uid
, inode
->i_uid
)) ||
351 ((iap
->ia_valid
& ATTR_GID
) && !gid_eq(iap
->ia_gid
, inode
->i_gid
)))) {
352 iap
->ia_valid
|= ATTR_KILL_PRIV
;
353 if (iap
->ia_valid
& ATTR_MODE
) {
354 /* we're setting mode too, just clear the s*id bits */
355 iap
->ia_mode
&= ~S_ISUID
;
356 if (iap
->ia_mode
& S_IXGRP
)
357 iap
->ia_mode
&= ~S_ISGID
;
359 /* set ATTR_KILL_* bits and let VFS handle it */
360 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
366 nfsd_get_write_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
369 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
372 if (iap
->ia_size
< inode
->i_size
) {
375 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
376 NFSD_MAY_TRUNC
| NFSD_MAY_OWNER_OVERRIDE
);
381 host_err
= get_write_access(inode
);
385 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
387 goto out_put_write_access
;
390 out_put_write_access
:
391 put_write_access(inode
);
393 return nfserrno(host_err
);
397 * Set various file attributes. After this call fhp needs an fh_put.
400 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
401 int check_guard
, time_t guardtime
)
403 struct dentry
*dentry
;
405 int accmode
= NFSD_MAY_SATTR
;
409 bool get_write_count
;
412 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
413 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
414 if (iap
->ia_valid
& ATTR_SIZE
)
417 /* Callers that do fh_verify should do the fh_want_write: */
418 get_write_count
= !fhp
->fh_dentry
;
421 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
424 if (get_write_count
) {
425 host_err
= fh_want_write(fhp
);
427 return nfserrno(host_err
);
430 dentry
= fhp
->fh_dentry
;
431 inode
= dentry
->d_inode
;
433 /* Ignore any mode updates on symlinks */
434 if (S_ISLNK(inode
->i_mode
))
435 iap
->ia_valid
&= ~ATTR_MODE
;
440 nfsd_sanitize_attrs(inode
, iap
);
443 * The size case is special, it changes the file in addition to the
446 if (iap
->ia_valid
& ATTR_SIZE
) {
447 err
= nfsd_get_write_access(rqstp
, fhp
, iap
);
453 iap
->ia_valid
|= ATTR_CTIME
;
455 if (check_guard
&& guardtime
!= inode
->i_ctime
.tv_sec
) {
456 err
= nfserr_notsync
;
457 goto out_put_write_access
;
460 host_err
= nfsd_break_lease(inode
);
462 goto out_put_write_access_nfserror
;
465 host_err
= notify_change(dentry
, iap
);
468 out_put_write_access_nfserror
:
469 err
= nfserrno(host_err
);
470 out_put_write_access
:
472 put_write_access(inode
);
474 commit_metadata(fhp
);
479 #if defined(CONFIG_NFSD_V2_ACL) || \
480 defined(CONFIG_NFSD_V3_ACL) || \
481 defined(CONFIG_NFSD_V4)
482 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
487 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
491 *buf
= kmalloc(buflen
, GFP_KERNEL
);
495 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
502 #if defined(CONFIG_NFSD_V4)
504 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
512 return vfs_setxattr(dentry
, key
, NULL
, 0, 0);
514 buflen
= posix_acl_xattr_size(pacl
->a_count
);
515 buf
= kmalloc(buflen
, GFP_KERNEL
);
520 len
= posix_acl_to_xattr(&init_user_ns
, pacl
, buf
, buflen
);
526 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
533 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
534 struct nfs4_acl
*acl
)
538 struct dentry
*dentry
;
540 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
541 unsigned int flags
= 0;
544 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_SATTR
);
548 dentry
= fhp
->fh_dentry
;
549 inode
= dentry
->d_inode
;
550 if (S_ISDIR(inode
->i_mode
))
551 flags
= NFS4_ACL_DIR
;
553 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
554 if (host_error
== -EINVAL
) {
555 return nfserr_attrnotsupp
;
556 } else if (host_error
< 0)
559 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
563 if (S_ISDIR(inode
->i_mode
))
564 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
567 posix_acl_release(pacl
);
568 posix_acl_release(dpacl
);
570 if (host_error
== -EOPNOTSUPP
)
571 return nfserr_attrnotsupp
;
573 return nfserrno(host_error
);
576 static struct posix_acl
*
577 _get_posix_acl(struct dentry
*dentry
, char *key
)
580 struct posix_acl
*pacl
= NULL
;
583 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
587 return ERR_PTR(buflen
);
589 pacl
= posix_acl_from_xattr(&init_user_ns
, buf
, buflen
);
595 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
597 struct inode
*inode
= dentry
->d_inode
;
599 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
600 unsigned int flags
= 0;
602 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
603 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
604 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
606 error
= PTR_ERR(pacl
);
611 if (S_ISDIR(inode
->i_mode
)) {
612 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
613 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
615 else if (IS_ERR(dpacl
)) {
616 error
= PTR_ERR(dpacl
);
620 flags
= NFS4_ACL_DIR
;
623 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
625 error
= PTR_ERR(*acl
);
629 posix_acl_release(pacl
);
630 posix_acl_release(dpacl
);
635 * NFS junction information is stored in an extended attribute.
637 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
640 * nfsd4_is_junction - Test if an object could be an NFS junction
642 * @dentry: object to test
644 * Returns 1 if "dentry" appears to contain NFS junction information.
645 * Otherwise 0 is returned.
647 int nfsd4_is_junction(struct dentry
*dentry
)
649 struct inode
*inode
= dentry
->d_inode
;
653 if (inode
->i_mode
& S_IXUGO
)
655 if (!(inode
->i_mode
& S_ISVTX
))
657 if (vfs_getxattr(dentry
, NFSD_JUNCTION_XATTR_NAME
, NULL
, 0) <= 0)
661 #endif /* defined(CONFIG_NFSD_V4) */
663 #ifdef CONFIG_NFSD_V3
665 * Check server access rights to a file system object
671 static struct accessmap nfs3_regaccess
[] = {
672 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
673 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
674 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
675 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
680 static struct accessmap nfs3_diraccess
[] = {
681 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
682 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
683 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
684 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
685 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
690 static struct accessmap nfs3_anyaccess
[] = {
691 /* Some clients - Solaris 2.6 at least, make an access call
692 * to the server to check for access for things like /dev/null
693 * (which really, the server doesn't care about). So
694 * We provide simple access checking for them, looking
695 * mainly at mode bits, and we make sure to ignore read-only
698 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
699 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
700 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
701 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
707 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
709 struct accessmap
*map
;
710 struct svc_export
*export
;
711 struct dentry
*dentry
;
712 u32 query
, result
= 0, sresult
= 0;
715 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
719 export
= fhp
->fh_export
;
720 dentry
= fhp
->fh_dentry
;
722 if (S_ISREG(dentry
->d_inode
->i_mode
))
723 map
= nfs3_regaccess
;
724 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
725 map
= nfs3_diraccess
;
727 map
= nfs3_anyaccess
;
731 for (; map
->access
; map
++) {
732 if (map
->access
& query
) {
735 sresult
|= map
->access
;
737 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
740 result
|= map
->access
;
743 /* the following error codes just mean the access was not allowed,
744 * rather than an error occurred */
748 /* simply don't "or" in the access bit. */
758 *supported
= sresult
;
763 #endif /* CONFIG_NFSD_V3 */
765 static int nfsd_open_break_lease(struct inode
*inode
, int access
)
769 if (access
& NFSD_MAY_NOT_BREAK_LEASE
)
771 mode
= (access
& NFSD_MAY_WRITE
) ? O_WRONLY
: O_RDONLY
;
772 return break_lease(inode
, mode
| O_NONBLOCK
);
776 * Open an existing file or directory.
777 * The may_flags argument indicates the type of open (read/write/lock)
778 * and additional flags.
779 * N.B. After this call fhp needs an fh_put
782 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
,
783 int may_flags
, struct file
**filp
)
787 int flags
= O_RDONLY
|O_LARGEFILE
;
791 validate_process_creds();
794 * If we get here, then the client has already done an "open",
795 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
796 * in case a chmod has now revoked permission.
798 * Arguably we should also allow the owner override for
799 * directories, but we never have and it doesn't seem to have
800 * caused anyone a problem. If we were to change this, note
801 * also that our filldir callbacks would need a variant of
802 * lookup_one_len that doesn't check permissions.
805 may_flags
|= NFSD_MAY_OWNER_OVERRIDE
;
806 err
= fh_verify(rqstp
, fhp
, type
, may_flags
);
810 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
811 path
.dentry
= fhp
->fh_dentry
;
812 inode
= path
.dentry
->d_inode
;
814 /* Disallow write access to files with the append-only bit set
815 * or any access when mandatory locking enabled
818 if (IS_APPEND(inode
) && (may_flags
& NFSD_MAY_WRITE
))
821 * We must ignore files (but only files) which might have mandatory
822 * locks on them because there is no way to know if the accesser has
825 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
831 host_err
= nfsd_open_break_lease(inode
, may_flags
);
832 if (host_err
) /* NOMEM or WOULDBLOCK */
835 if (may_flags
& NFSD_MAY_WRITE
) {
836 if (may_flags
& NFSD_MAY_READ
)
837 flags
= O_RDWR
|O_LARGEFILE
;
839 flags
= O_WRONLY
|O_LARGEFILE
;
841 *filp
= dentry_open(&path
, flags
, current_cred());
843 host_err
= PTR_ERR(*filp
);
846 host_err
= ima_file_check(*filp
, may_flags
);
848 if (may_flags
& NFSD_MAY_64BIT_COOKIE
)
849 (*filp
)->f_mode
|= FMODE_64BITHASH
;
851 (*filp
)->f_mode
|= FMODE_32BITHASH
;
855 err
= nfserrno(host_err
);
857 validate_process_creds();
865 nfsd_close(struct file
*filp
)
871 * Obtain the readahead parameters for the file
872 * specified by (dev, ino).
875 static inline struct raparms
*
876 nfsd_get_raparms(dev_t dev
, ino_t ino
)
878 struct raparms
*ra
, **rap
, **frap
= NULL
;
881 struct raparm_hbucket
*rab
;
883 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
884 rab
= &raparm_hash
[hash
];
886 spin_lock(&rab
->pb_lock
);
887 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
888 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
891 if (ra
->p_count
== 0)
894 depth
= nfsdstats
.ra_size
;
896 spin_unlock(&rab
->pb_lock
);
906 if (rap
!= &rab
->pb_head
) {
908 ra
->p_next
= rab
->pb_head
;
912 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
913 spin_unlock(&rab
->pb_lock
);
918 * Grab and keep cached pages associated with a file in the svc_rqst
919 * so that they can be passed to the network sendmsg/sendpage routines
920 * directly. They will be released after the sending has completed.
923 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
924 struct splice_desc
*sd
)
926 struct svc_rqst
*rqstp
= sd
->u
.data
;
927 struct page
**pp
= rqstp
->rq_next_page
;
928 struct page
*page
= buf
->page
;
933 if (rqstp
->rq_res
.page_len
== 0) {
935 put_page(*rqstp
->rq_next_page
);
936 *(rqstp
->rq_next_page
++) = page
;
937 rqstp
->rq_res
.page_base
= buf
->offset
;
938 rqstp
->rq_res
.page_len
= size
;
939 } else if (page
!= pp
[-1]) {
941 if (*rqstp
->rq_next_page
)
942 put_page(*rqstp
->rq_next_page
);
943 *(rqstp
->rq_next_page
++) = page
;
944 rqstp
->rq_res
.page_len
+= size
;
946 rqstp
->rq_res
.page_len
+= size
;
951 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
952 struct splice_desc
*sd
)
954 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
958 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
959 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
967 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
968 struct splice_desc sd
= {
975 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
976 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
980 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
985 nfsdstats
.io_read
+= host_err
;
988 fsnotify_access(file
);
990 err
= nfserrno(host_err
);
994 static void kill_suid(struct dentry
*dentry
)
997 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
999 mutex_lock(&dentry
->d_inode
->i_mutex
);
1000 notify_change(dentry
, &ia
);
1001 mutex_unlock(&dentry
->d_inode
->i_mutex
);
1005 * Gathered writes: If another process is currently writing to the file,
1006 * there's a high chance this is another nfsd (triggered by a bulk write
1007 * from a client's biod). Rather than syncing the file with each write
1008 * request, we sleep for 10 msec.
1010 * I don't know if this roughly approximates C. Juszak's idea of
1011 * gathered writes, but it's a nice and simple solution (IMHO), and it
1014 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1015 * better tool (separate unstable writes and commits) for solving this
1018 static int wait_for_concurrent_writes(struct file
*file
)
1020 struct inode
*inode
= file_inode(file
);
1021 static ino_t last_ino
;
1022 static dev_t last_dev
;
1025 if (atomic_read(&inode
->i_writecount
) > 1
1026 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1027 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1029 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1032 if (inode
->i_state
& I_DIRTY
) {
1033 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1034 err
= vfs_fsync(file
, 0);
1036 last_ino
= inode
->i_ino
;
1037 last_dev
= inode
->i_sb
->s_dev
;
1042 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1043 loff_t offset
, struct kvec
*vec
, int vlen
,
1044 unsigned long *cnt
, int *stablep
)
1046 struct svc_export
*exp
;
1047 struct dentry
*dentry
;
1048 struct inode
*inode
;
1052 int stable
= *stablep
;
1054 loff_t pos
= offset
;
1056 dentry
= file
->f_path
.dentry
;
1057 inode
= dentry
->d_inode
;
1058 exp
= fhp
->fh_export
;
1060 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1062 if (!EX_ISSYNC(exp
))
1065 /* Write the data. */
1066 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1067 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &pos
);
1072 nfsdstats
.io_write
+= host_err
;
1073 fsnotify_modify(file
);
1075 /* clear setuid/setgid flag after write */
1076 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1081 host_err
= wait_for_concurrent_writes(file
);
1083 host_err
= vfs_fsync_range(file
, offset
, offset
+*cnt
, 0);
1087 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1091 err
= nfserrno(host_err
);
1096 * Read data from a file. count must contain the requested read count
1097 * on entry. On return, *count contains the number of bytes actually read.
1098 * N.B. After this call fhp needs an fh_put
1100 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1101 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
1104 struct inode
*inode
;
1108 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1112 inode
= file_inode(file
);
1114 /* Get readahead parameters */
1115 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
1117 if (ra
&& ra
->p_set
)
1118 file
->f_ra
= ra
->p_ra
;
1120 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1122 /* Write back readahead params */
1124 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
1125 spin_lock(&rab
->pb_lock
);
1126 ra
->p_ra
= file
->f_ra
;
1129 spin_unlock(&rab
->pb_lock
);
1136 /* As above, but use the provided file descriptor. */
1138 nfsd_read_file(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1139 loff_t offset
, struct kvec
*vec
, int vlen
,
1140 unsigned long *count
)
1145 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1146 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1149 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1150 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1151 err
= nfsd_read(rqstp
, fhp
, offset
, vec
, vlen
, count
);
1157 * Write data to a file.
1158 * The stable flag requests synchronous writes.
1159 * N.B. After this call fhp needs an fh_put
1162 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1163 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1169 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1170 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1173 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1176 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1181 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1189 #ifdef CONFIG_NFSD_V3
1191 * Commit all pending writes to stable storage.
1193 * Note: we only guarantee that data that lies within the range specified
1194 * by the 'offset' and 'count' parameters will be synced.
1196 * Unfortunately we cannot lock the file to make sure we return full WCC
1197 * data to the client, as locking happens lower down in the filesystem.
1200 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1201 loff_t offset
, unsigned long count
)
1204 loff_t end
= LLONG_MAX
;
1205 __be32 err
= nfserr_inval
;
1210 end
= offset
+ (loff_t
)count
- 1;
1215 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1216 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1219 if (EX_ISSYNC(fhp
->fh_export
)) {
1220 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1222 if (err2
!= -EINVAL
)
1223 err
= nfserrno(err2
);
1225 err
= nfserr_notsupp
;
1232 #endif /* CONFIG_NFSD_V3 */
1235 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1239 * Mode has already been set earlier in create:
1241 iap
->ia_valid
&= ~ATTR_MODE
;
1243 * Setting uid/gid works only for root. Irix appears to
1244 * send along the gid on create when it tries to implement
1245 * setgid directories via NFS:
1247 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID
))
1248 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1250 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1254 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1255 * setting size to 0 may fail for some specific file systems by the permission
1256 * checking which requires WRITE permission but the mode is 000.
1257 * we ignore the resizing(to 0) on the just new created file, since the size is
1258 * 0 after file created.
1260 * call this only after vfs_create() is called.
1263 nfsd_check_ignore_resizing(struct iattr
*iap
)
1265 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1266 iap
->ia_valid
&= ~ATTR_SIZE
;
1270 * Create a file (regular, directory, device, fifo); UNIX sockets
1271 * not yet implemented.
1272 * If the response fh has been verified, the parent directory should
1273 * already be locked. Note that the parent directory is left locked.
1275 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1278 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1279 char *fname
, int flen
, struct iattr
*iap
,
1280 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1282 struct dentry
*dentry
, *dchild
= NULL
;
1292 if (isdotent(fname
, flen
))
1295 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1299 dentry
= fhp
->fh_dentry
;
1300 dirp
= dentry
->d_inode
;
1302 err
= nfserr_notdir
;
1303 if (!dirp
->i_op
->lookup
)
1306 * Check whether the response file handle has been verified yet.
1307 * If it has, the parent directory should already be locked.
1309 if (!resfhp
->fh_dentry
) {
1310 host_err
= fh_want_write(fhp
);
1314 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1315 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1316 dchild
= lookup_one_len(fname
, dentry
, flen
);
1317 host_err
= PTR_ERR(dchild
);
1320 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1324 /* called from nfsd_proc_create */
1325 dchild
= dget(resfhp
->fh_dentry
);
1326 if (!fhp
->fh_locked
) {
1327 /* not actually possible */
1329 "nfsd_create: parent %s/%s not locked!\n",
1330 dentry
->d_parent
->d_name
.name
,
1331 dentry
->d_name
.name
);
1337 * Make sure the child dentry is still negative ...
1340 if (dchild
->d_inode
) {
1341 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1342 dentry
->d_name
.name
, dchild
->d_name
.name
);
1346 if (!(iap
->ia_valid
& ATTR_MODE
))
1348 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1351 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1352 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1358 * Get the dir op function pointer.
1364 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1366 nfsd_check_ignore_resizing(iap
);
1369 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1375 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1381 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1384 * nfsd_setattr already committed the child. Transactional filesystems
1385 * had a chance to commit changes for both parent and child
1386 * simultaneously making the following commit_metadata a noop.
1388 err2
= nfserrno(commit_metadata(fhp
));
1392 * Update the file handle to get the new inode info.
1395 err
= fh_update(resfhp
);
1397 if (dchild
&& !IS_ERR(dchild
))
1402 err
= nfserrno(host_err
);
1406 #ifdef CONFIG_NFSD_V3
1408 static inline int nfsd_create_is_exclusive(int createmode
)
1410 return createmode
== NFS3_CREATE_EXCLUSIVE
1411 || createmode
== NFS4_CREATE_EXCLUSIVE4_1
;
1415 * NFSv3 and NFSv4 version of nfsd_create
1418 do_nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1419 char *fname
, int flen
, struct iattr
*iap
,
1420 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1421 bool *truncp
, bool *created
)
1423 struct dentry
*dentry
, *dchild
= NULL
;
1427 __u32 v_mtime
=0, v_atime
=0;
1433 if (isdotent(fname
, flen
))
1435 if (!(iap
->ia_valid
& ATTR_MODE
))
1437 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
1441 dentry
= fhp
->fh_dentry
;
1442 dirp
= dentry
->d_inode
;
1444 /* Get all the sanity checks out of the way before
1445 * we lock the parent. */
1446 err
= nfserr_notdir
;
1447 if (!dirp
->i_op
->lookup
)
1450 host_err
= fh_want_write(fhp
);
1454 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1457 * Compose the response file handle.
1459 dchild
= lookup_one_len(fname
, dentry
, flen
);
1460 host_err
= PTR_ERR(dchild
);
1464 /* If file doesn't exist, check for permissions to create one */
1465 if (!dchild
->d_inode
) {
1466 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1471 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1475 if (nfsd_create_is_exclusive(createmode
)) {
1476 /* solaris7 gets confused (bugid 4218508) if these have
1477 * the high bit set, so just clear the high bits. If this is
1478 * ever changed to use different attrs for storing the
1479 * verifier, then do_open_lookup() will also need to be fixed
1482 v_mtime
= verifier
[0]&0x7fffffff;
1483 v_atime
= verifier
[1]&0x7fffffff;
1486 if (dchild
->d_inode
) {
1489 switch (createmode
) {
1490 case NFS3_CREATE_UNCHECKED
:
1491 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1494 /* in nfsv4, we need to treat this case a little
1495 * differently. we don't want to truncate the
1496 * file now; this would be wrong if the OPEN
1497 * fails for some other reason. furthermore,
1498 * if the size is nonzero, we should ignore it
1499 * according to spec!
1501 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1504 iap
->ia_valid
&= ATTR_SIZE
;
1508 case NFS3_CREATE_EXCLUSIVE
:
1509 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1510 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1511 && dchild
->d_inode
->i_size
== 0 ) {
1516 case NFS4_CREATE_EXCLUSIVE4_1
:
1517 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1518 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1519 && dchild
->d_inode
->i_size
== 0 ) {
1525 case NFS3_CREATE_GUARDED
:
1532 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1540 nfsd_check_ignore_resizing(iap
);
1542 if (nfsd_create_is_exclusive(createmode
)) {
1543 /* Cram the verifier into atime/mtime */
1544 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1545 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1546 /* XXX someone who knows this better please fix it for nsec */
1547 iap
->ia_mtime
.tv_sec
= v_mtime
;
1548 iap
->ia_atime
.tv_sec
= v_atime
;
1549 iap
->ia_mtime
.tv_nsec
= 0;
1550 iap
->ia_atime
.tv_nsec
= 0;
1554 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1557 * nfsd_setattr already committed the child (and possibly also the parent).
1560 err
= nfserrno(commit_metadata(fhp
));
1563 * Update the filehandle to get the new inode info.
1566 err
= fh_update(resfhp
);
1570 if (dchild
&& !IS_ERR(dchild
))
1576 err
= nfserrno(host_err
);
1579 #endif /* CONFIG_NFSD_V3 */
1582 * Read a symlink. On entry, *lenp must contain the maximum path length that
1583 * fits into the buffer. On return, it contains the true length.
1584 * N.B. After this call fhp needs an fh_put
1587 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1589 struct inode
*inode
;
1595 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1599 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
1600 path
.dentry
= fhp
->fh_dentry
;
1601 inode
= path
.dentry
->d_inode
;
1604 if (!inode
->i_op
->readlink
)
1608 /* N.B. Why does this call need a get_fs()??
1609 * Remove the set_fs and watch the fireworks:-) --okir
1612 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1613 host_err
= inode
->i_op
->readlink(path
.dentry
, (char __user
*)buf
, *lenp
);
1624 err
= nfserrno(host_err
);
1629 * Create a symlink and look up its inode
1630 * N.B. After this call _both_ fhp and resfhp need an fh_put
1633 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1634 char *fname
, int flen
,
1635 char *path
, int plen
,
1636 struct svc_fh
*resfhp
,
1639 struct dentry
*dentry
, *dnew
;
1647 if (isdotent(fname
, flen
))
1650 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1654 host_err
= fh_want_write(fhp
);
1659 dentry
= fhp
->fh_dentry
;
1660 dnew
= lookup_one_len(fname
, dentry
, flen
);
1661 host_err
= PTR_ERR(dnew
);
1665 if (unlikely(path
[plen
] != 0)) {
1666 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1667 if (path_alloced
== NULL
)
1670 strncpy(path_alloced
, path
, plen
);
1671 path_alloced
[plen
] = 0;
1672 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1673 kfree(path_alloced
);
1676 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1677 err
= nfserrno(host_err
);
1679 err
= nfserrno(commit_metadata(fhp
));
1684 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1686 if (err
==0) err
= cerr
;
1691 err
= nfserrno(host_err
);
1697 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1700 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1701 char *name
, int len
, struct svc_fh
*tfhp
)
1703 struct dentry
*ddir
, *dnew
, *dold
;
1708 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1711 err
= fh_verify(rqstp
, tfhp
, 0, NFSD_MAY_NOP
);
1715 if (S_ISDIR(tfhp
->fh_dentry
->d_inode
->i_mode
))
1721 if (isdotent(name
, len
))
1724 host_err
= fh_want_write(tfhp
);
1726 err
= nfserrno(host_err
);
1730 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1731 ddir
= ffhp
->fh_dentry
;
1732 dirp
= ddir
->d_inode
;
1734 dnew
= lookup_one_len(name
, ddir
, len
);
1735 host_err
= PTR_ERR(dnew
);
1739 dold
= tfhp
->fh_dentry
;
1744 host_err
= nfsd_break_lease(dold
->d_inode
);
1746 err
= nfserrno(host_err
);
1749 host_err
= vfs_link(dold
, dirp
, dnew
);
1751 err
= nfserrno(commit_metadata(ffhp
));
1753 err
= nfserrno(commit_metadata(tfhp
));
1755 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1758 err
= nfserrno(host_err
);
1764 fh_drop_write(tfhp
);
1769 err
= nfserrno(host_err
);
1775 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1778 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1779 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1781 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1782 struct inode
*fdir
, *tdir
;
1786 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1789 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1793 fdentry
= ffhp
->fh_dentry
;
1794 fdir
= fdentry
->d_inode
;
1796 tdentry
= tfhp
->fh_dentry
;
1797 tdir
= tdentry
->d_inode
;
1800 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1803 host_err
= fh_want_write(ffhp
);
1805 err
= nfserrno(host_err
);
1809 /* cannot use fh_lock as we need deadlock protective ordering
1810 * so do it by hand */
1811 trap
= lock_rename(tdentry
, fdentry
);
1812 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1816 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1817 host_err
= PTR_ERR(odentry
);
1818 if (IS_ERR(odentry
))
1822 if (!odentry
->d_inode
)
1825 if (odentry
== trap
)
1828 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1829 host_err
= PTR_ERR(ndentry
);
1830 if (IS_ERR(ndentry
))
1832 host_err
= -ENOTEMPTY
;
1833 if (ndentry
== trap
)
1837 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1839 if (ffhp
->fh_export
->ex_path
.dentry
!= tfhp
->fh_export
->ex_path
.dentry
)
1842 host_err
= nfsd_break_lease(odentry
->d_inode
);
1845 if (ndentry
->d_inode
) {
1846 host_err
= nfsd_break_lease(ndentry
->d_inode
);
1850 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1852 host_err
= commit_metadata(tfhp
);
1854 host_err
= commit_metadata(ffhp
);
1861 err
= nfserrno(host_err
);
1863 /* we cannot reply on fh_unlock on the two filehandles,
1864 * as that would do the wrong thing if the two directories
1865 * were the same, so again we do it by hand
1867 fill_post_wcc(ffhp
);
1868 fill_post_wcc(tfhp
);
1869 unlock_rename(tdentry
, fdentry
);
1870 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1871 fh_drop_write(ffhp
);
1878 * Unlink a file or directory
1879 * N.B. After this call fhp needs an fh_put
1882 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1883 char *fname
, int flen
)
1885 struct dentry
*dentry
, *rdentry
;
1891 if (!flen
|| isdotent(fname
, flen
))
1893 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1897 host_err
= fh_want_write(fhp
);
1901 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1902 dentry
= fhp
->fh_dentry
;
1903 dirp
= dentry
->d_inode
;
1905 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1906 host_err
= PTR_ERR(rdentry
);
1907 if (IS_ERR(rdentry
))
1910 if (!rdentry
->d_inode
) {
1917 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1919 host_err
= nfsd_break_lease(rdentry
->d_inode
);
1922 if (type
!= S_IFDIR
)
1923 host_err
= vfs_unlink(dirp
, rdentry
);
1925 host_err
= vfs_rmdir(dirp
, rdentry
);
1927 host_err
= commit_metadata(fhp
);
1932 err
= nfserrno(host_err
);
1938 * We do this buffering because we must not call back into the file
1939 * system's ->lookup() method from the filldir callback. That may well
1940 * deadlock a number of file systems.
1942 * This is based heavily on the implementation of same in XFS.
1944 struct buffered_dirent
{
1948 unsigned int d_type
;
1952 struct readdir_data
{
1958 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1959 loff_t offset
, u64 ino
, unsigned int d_type
)
1961 struct readdir_data
*buf
= __buf
;
1962 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1963 unsigned int reclen
;
1965 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1966 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1971 de
->namlen
= namlen
;
1972 de
->offset
= offset
;
1974 de
->d_type
= d_type
;
1975 memcpy(de
->name
, name
, namlen
);
1976 buf
->used
+= reclen
;
1981 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1982 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1984 struct readdir_data buf
;
1985 struct buffered_dirent
*de
;
1990 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1992 return nfserrno(-ENOMEM
);
1997 struct inode
*dir_inode
= file_inode(file
);
1998 unsigned int reclen
;
2000 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
2004 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
2017 * Various filldir functions may end up calling back into
2018 * lookup_one_len() and the file system's ->lookup() method.
2019 * These expect i_mutex to be held, as it would within readdir.
2021 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
2025 de
= (struct buffered_dirent
*)buf
.dirent
;
2027 offset
= de
->offset
;
2029 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
2030 de
->ino
, de
->d_type
))
2033 if (cdp
->err
!= nfs_ok
)
2036 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
2039 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
2041 mutex_unlock(&dir_inode
->i_mutex
);
2042 if (size
> 0) /* We bailed out early */
2045 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
2048 free_page((unsigned long)(buf
.dirent
));
2051 return nfserrno(host_err
);
2058 * Read entries from a directory.
2059 * The NFSv3/4 verifier we ignore for now.
2062 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
2063 struct readdir_cd
*cdp
, filldir_t func
)
2067 loff_t offset
= *offsetp
;
2068 int may_flags
= NFSD_MAY_READ
;
2070 /* NFSv2 only supports 32 bit cookies */
2071 if (rqstp
->rq_vers
> 2)
2072 may_flags
|= NFSD_MAY_64BIT_COOKIE
;
2074 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, may_flags
, &file
);
2078 offset
= vfs_llseek(file
, offset
, SEEK_SET
);
2080 err
= nfserrno((int)offset
);
2084 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2086 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2087 err
= nfs_ok
; /* can still be found in ->err */
2095 * Get file system stats
2096 * N.B. After this call fhp needs an fh_put
2099 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2103 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2105 struct path path
= {
2106 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
2107 .dentry
= fhp
->fh_dentry
,
2109 if (vfs_statfs(&path
, stat
))
2115 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2117 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2121 * Check for a user's access permissions to this inode.
2124 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2125 struct dentry
*dentry
, int acc
)
2127 struct inode
*inode
= dentry
->d_inode
;
2130 if ((acc
& NFSD_MAY_MASK
) == NFSD_MAY_NOP
)
2133 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2135 (acc
& NFSD_MAY_READ
)? " read" : "",
2136 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2137 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2138 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2139 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2140 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2141 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2143 IS_IMMUTABLE(inode
)? " immut" : "",
2144 IS_APPEND(inode
)? " append" : "",
2145 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2146 dprintk(" owner %d/%d user %d/%d\n",
2147 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2150 /* Normally we reject any write/sattr etc access on a read-only file
2151 * system. But if it is IRIX doing check on write-access for a
2152 * device special file, we ignore rofs.
2154 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2155 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2156 if (exp_rdonly(rqstp
, exp
) ||
2157 __mnt_is_readonly(exp
->ex_path
.mnt
))
2159 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2162 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2165 if (acc
& NFSD_MAY_LOCK
) {
2166 /* If we cannot rely on authentication in NLM requests,
2167 * just allow locks, otherwise require read permission, or
2170 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2173 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2176 * The file owner always gets access permission for accesses that
2177 * would normally be checked at open time. This is to make
2178 * file access work even when the client has done a fchmod(fd, 0).
2180 * However, `cp foo bar' should fail nevertheless when bar is
2181 * readonly. A sensible way to do this might be to reject all
2182 * attempts to truncate a read-only file, because a creat() call
2183 * always implies file truncation.
2184 * ... but this isn't really fair. A process may reasonably call
2185 * ftruncate on an open file descriptor on a file with perm 000.
2186 * We must trust the client to do permission checking - using "ACCESS"
2189 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2190 uid_eq(inode
->i_uid
, current_fsuid()))
2193 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2194 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2196 /* Allow read access to binaries even when mode 111 */
2197 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2198 (acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
) ||
2199 acc
== (NFSD_MAY_READ
| NFSD_MAY_READ_IF_EXEC
)))
2200 err
= inode_permission(inode
, MAY_EXEC
);
2202 return err
? nfserrno(err
) : 0;
2206 nfsd_racache_shutdown(void)
2208 struct raparms
*raparm
, *last_raparm
;
2211 dprintk("nfsd: freeing readahead buffers.\n");
2213 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2214 raparm
= raparm_hash
[i
].pb_head
;
2216 last_raparm
= raparm
;
2217 raparm
= raparm
->p_next
;
2220 raparm_hash
[i
].pb_head
= NULL
;
2224 * Initialize readahead param cache
2227 nfsd_racache_init(int cache_size
)
2232 struct raparms
**raparm
= NULL
;
2235 if (raparm_hash
[0].pb_head
)
2237 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2240 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2242 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2244 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2245 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2247 raparm
= &raparm_hash
[i
].pb_head
;
2248 for (j
= 0; j
< nperbucket
; j
++) {
2249 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2252 raparm
= &(*raparm
)->p_next
;
2257 nfsdstats
.ra_size
= cache_size
;
2261 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2262 nfsd_racache_shutdown();
2266 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2268 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2270 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2274 struct posix_acl
*acl
;
2276 if (!IS_POSIXACL(inode
))
2277 return ERR_PTR(-EOPNOTSUPP
);
2280 case ACL_TYPE_ACCESS
:
2281 name
= POSIX_ACL_XATTR_ACCESS
;
2283 case ACL_TYPE_DEFAULT
:
2284 name
= POSIX_ACL_XATTR_DEFAULT
;
2287 return ERR_PTR(-EOPNOTSUPP
);
2290 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2292 return ERR_PTR(size
);
2294 acl
= posix_acl_from_xattr(&init_user_ns
, value
, size
);
2300 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2302 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2308 if (!IS_POSIXACL(inode
) ||
2309 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2312 case ACL_TYPE_ACCESS
:
2313 name
= POSIX_ACL_XATTR_ACCESS
;
2315 case ACL_TYPE_DEFAULT
:
2316 name
= POSIX_ACL_XATTR_DEFAULT
;
2322 if (acl
&& acl
->a_count
) {
2323 size
= posix_acl_xattr_size(acl
->a_count
);
2324 value
= kmalloc(size
, GFP_KERNEL
);
2327 error
= posix_acl_to_xattr(&init_user_ns
, acl
, value
, size
);
2334 error
= fh_want_write(fhp
);
2338 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2340 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2343 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2344 if (error
== -ENODATA
)
2354 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */