1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
34 #endif /* CONFIG_NFSD_V3 */
37 #include <linux/nfs4_acl.h>
38 #include <linux/nfsd_idmap.h>
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 while (d_mountpoint(path
.dentry
) && follow_down(&path
))
93 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
97 * We normally allow NFS clients to continue
98 * "underneath" a mountpoint that is not exported.
99 * The exception is V4ROOT, where no traversal is ever
100 * allowed without an explicit export of the new
103 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
108 if (nfsd_v4client(rqstp
) ||
109 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
110 /* successfully crossed mount point */
112 * This is subtle: path.dentry is *not* on path.mnt
113 * at this point. The only reason we are safe is that
114 * original mnt is pinned down by exp, so we should
115 * put path *before* putting exp
118 path
.dentry
= dentry
;
128 static void follow_to_parent(struct path
*path
)
132 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
134 dp
= dget_parent(path
->dentry
);
139 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
141 struct svc_export
*exp2
;
142 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
143 .dentry
= dget(dparent
)};
145 follow_to_parent(&path
);
147 exp2
= rqst_exp_parent(rqstp
, &path
);
148 if (PTR_ERR(exp2
) == -ENOENT
) {
149 *dentryp
= dget(dparent
);
150 } else if (IS_ERR(exp2
)) {
152 return PTR_ERR(exp2
);
154 *dentryp
= dget(path
.dentry
);
163 * For nfsd purposes, we treat V4ROOT exports as though there was an
164 * export at *every* directory.
166 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
168 if (d_mountpoint(dentry
))
170 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
172 return dentry
->d_inode
!= NULL
;
176 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
177 const char *name
, unsigned int len
,
178 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
180 struct svc_export
*exp
;
181 struct dentry
*dparent
;
182 struct dentry
*dentry
;
186 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
188 /* Obtain dentry and export. */
189 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
193 dparent
= fhp
->fh_dentry
;
194 exp
= fhp
->fh_export
;
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name
, len
)) {
200 dentry
= dget(dparent
);
201 else if (dparent
!= exp
->ex_path
.dentry
)
202 dentry
= dget_parent(dparent
);
203 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
204 dentry
= dget(dparent
); /* .. == . just like at / */
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
213 dentry
= lookup_one_len(name
, dparent
, len
);
214 host_err
= PTR_ERR(dentry
);
218 * check if we have crossed a mount point ...
220 if (nfsd_mountpoint(dentry
, exp
)) {
221 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
227 *dentry_ret
= dentry
;
233 return nfserrno(host_err
);
237 * Look up one component of a pathname.
238 * N.B. After this call _both_ fhp and resfh need an fh_put
240 * If the lookup would cross a mountpoint, and the mounted filesystem
241 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
242 * accepted as it stands and the mounted directory is
243 * returned. Otherwise the covered directory is returned.
244 * NOTE: this mountpoint crossing is not supported properly by all
245 * clients and is explicitly disallowed for NFSv3
246 * NeilBrown <neilb@cse.unsw.edu.au>
249 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
250 unsigned int len
, struct svc_fh
*resfh
)
252 struct svc_export
*exp
;
253 struct dentry
*dentry
;
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
)
276 * Commit metadata changes to stable storage.
279 commit_metadata(struct svc_fh
*fhp
)
281 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
282 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
285 if (!EX_ISSYNC(fhp
->fh_export
))
288 if (export_ops
->commit_metadata
) {
289 error
= export_ops
->commit_metadata(inode
);
291 struct writeback_control wbc
= {
292 .sync_mode
= WB_SYNC_ALL
,
293 .nr_to_write
= 0, /* metadata only */
296 error
= sync_inode(inode
, &wbc
);
303 * Set various file attributes.
304 * N.B. After this call fhp needs an fh_put
307 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
308 int check_guard
, time_t guardtime
)
310 struct dentry
*dentry
;
312 int accmode
= NFSD_MAY_SATTR
;
318 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
319 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
320 if (iap
->ia_valid
& ATTR_SIZE
)
324 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
328 dentry
= fhp
->fh_dentry
;
329 inode
= dentry
->d_inode
;
331 /* Ignore any mode updates on symlinks */
332 if (S_ISLNK(inode
->i_mode
))
333 iap
->ia_valid
&= ~ATTR_MODE
;
339 * NFSv2 does not differentiate between "set-[ac]time-to-now"
340 * which only requires access, and "set-[ac]time-to-X" which
341 * requires ownership.
342 * So if it looks like it might be "set both to the same time which
343 * is close to now", and if inode_change_ok fails, then we
344 * convert to "set to now" instead of "set to explicit time"
346 * We only call inode_change_ok as the last test as technically
347 * it is not an interface that we should be using. It is only
348 * valid if the filesystem does not define it's own i_op->setattr.
350 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
351 #define MAX_TOUCH_TIME_ERROR (30*60)
352 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
353 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
357 * Now just make sure time is in the right ballpark.
358 * Solaris, at least, doesn't seem to care what the time
359 * request is. We require it be within 30 minutes of now.
361 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
364 if (delta
< MAX_TOUCH_TIME_ERROR
&&
365 inode_change_ok(inode
, iap
) != 0) {
367 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
368 * This will cause notify_change to set these times
371 iap
->ia_valid
&= ~BOTH_TIME_SET
;
376 * The size case is special.
377 * It changes the file as well as the attributes.
379 if (iap
->ia_valid
& ATTR_SIZE
) {
380 if (iap
->ia_size
< inode
->i_size
) {
381 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
382 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
388 * If we are changing the size of the file, then
389 * we need to break all leases.
391 host_err
= break_lease(inode
, O_WRONLY
| O_NONBLOCK
);
392 if (host_err
== -EWOULDBLOCK
)
393 host_err
= -ETIMEDOUT
;
394 if (host_err
) /* ENOMEM or EWOULDBLOCK */
397 host_err
= get_write_access(inode
);
402 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
404 put_write_access(inode
);
409 /* sanitize the mode change */
410 if (iap
->ia_valid
& ATTR_MODE
) {
411 iap
->ia_mode
&= S_IALLUGO
;
412 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
415 /* Revoke setuid/setgid on chown */
416 if (!S_ISDIR(inode
->i_mode
) &&
417 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
418 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
419 iap
->ia_valid
|= ATTR_KILL_PRIV
;
420 if (iap
->ia_valid
& ATTR_MODE
) {
421 /* we're setting mode too, just clear the s*id bits */
422 iap
->ia_mode
&= ~S_ISUID
;
423 if (iap
->ia_mode
& S_IXGRP
)
424 iap
->ia_mode
&= ~S_ISGID
;
426 /* set ATTR_KILL_* bits and let VFS handle it */
427 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
431 /* Change the attributes. */
433 iap
->ia_valid
|= ATTR_CTIME
;
435 err
= nfserr_notsync
;
436 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
438 host_err
= notify_change(dentry
, iap
);
439 err
= nfserrno(host_err
);
443 put_write_access(inode
);
445 if (EX_ISSYNC(fhp
->fh_export
))
446 write_inode_now(inode
, 1);
451 err
= nfserrno(host_err
);
455 #if defined(CONFIG_NFSD_V2_ACL) || \
456 defined(CONFIG_NFSD_V3_ACL) || \
457 defined(CONFIG_NFSD_V4)
458 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
463 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
467 *buf
= kmalloc(buflen
, GFP_KERNEL
);
471 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
478 #if defined(CONFIG_NFSD_V4)
480 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
487 buflen
= posix_acl_xattr_size(pacl
->a_count
);
488 buf
= kmalloc(buflen
, GFP_KERNEL
);
493 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
499 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
506 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
507 struct nfs4_acl
*acl
)
511 struct dentry
*dentry
;
513 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
514 unsigned int flags
= 0;
517 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
521 dentry
= fhp
->fh_dentry
;
522 inode
= dentry
->d_inode
;
523 if (S_ISDIR(inode
->i_mode
))
524 flags
= NFS4_ACL_DIR
;
526 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
527 if (host_error
== -EINVAL
) {
528 return nfserr_attrnotsupp
;
529 } else if (host_error
< 0)
532 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
536 if (S_ISDIR(inode
->i_mode
))
537 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
540 posix_acl_release(pacl
);
541 posix_acl_release(dpacl
);
543 if (host_error
== -EOPNOTSUPP
)
544 return nfserr_attrnotsupp
;
546 return nfserrno(host_error
);
549 static struct posix_acl
*
550 _get_posix_acl(struct dentry
*dentry
, char *key
)
553 struct posix_acl
*pacl
= NULL
;
556 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
560 return ERR_PTR(buflen
);
562 pacl
= posix_acl_from_xattr(buf
, buflen
);
568 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
570 struct inode
*inode
= dentry
->d_inode
;
572 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
573 unsigned int flags
= 0;
575 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
576 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
577 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
579 error
= PTR_ERR(pacl
);
584 if (S_ISDIR(inode
->i_mode
)) {
585 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
586 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
588 else if (IS_ERR(dpacl
)) {
589 error
= PTR_ERR(dpacl
);
593 flags
= NFS4_ACL_DIR
;
596 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
598 error
= PTR_ERR(*acl
);
602 posix_acl_release(pacl
);
603 posix_acl_release(dpacl
);
607 #endif /* defined(CONFIG_NFS_V4) */
609 #ifdef CONFIG_NFSD_V3
611 * Check server access rights to a file system object
617 static struct accessmap nfs3_regaccess
[] = {
618 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
619 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
620 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
621 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
626 static struct accessmap nfs3_diraccess
[] = {
627 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
628 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
629 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
630 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
631 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
636 static struct accessmap nfs3_anyaccess
[] = {
637 /* Some clients - Solaris 2.6 at least, make an access call
638 * to the server to check for access for things like /dev/null
639 * (which really, the server doesn't care about). So
640 * We provide simple access checking for them, looking
641 * mainly at mode bits, and we make sure to ignore read-only
644 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
645 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
646 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
647 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
653 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
655 struct accessmap
*map
;
656 struct svc_export
*export
;
657 struct dentry
*dentry
;
658 u32 query
, result
= 0, sresult
= 0;
661 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
665 export
= fhp
->fh_export
;
666 dentry
= fhp
->fh_dentry
;
668 if (S_ISREG(dentry
->d_inode
->i_mode
))
669 map
= nfs3_regaccess
;
670 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
671 map
= nfs3_diraccess
;
673 map
= nfs3_anyaccess
;
677 for (; map
->access
; map
++) {
678 if (map
->access
& query
) {
681 sresult
|= map
->access
;
683 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
686 result
|= map
->access
;
689 /* the following error codes just mean the access was not allowed,
690 * rather than an error occurred */
694 /* simply don't "or" in the access bit. */
704 *supported
= sresult
;
709 #endif /* CONFIG_NFSD_V3 */
714 * Open an existing file or directory.
715 * The access argument indicates the type of open (read/write/lock)
716 * N.B. After this call fhp needs an fh_put
719 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
720 int access
, struct file
**filp
)
722 struct dentry
*dentry
;
724 int flags
= O_RDONLY
|O_LARGEFILE
;
728 validate_process_creds();
731 * If we get here, then the client has already done an "open",
732 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
733 * in case a chmod has now revoked permission.
735 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
739 dentry
= fhp
->fh_dentry
;
740 inode
= dentry
->d_inode
;
742 /* Disallow write access to files with the append-only bit set
743 * or any access when mandatory locking enabled
746 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
749 * We must ignore files (but only files) which might have mandatory
750 * locks on them because there is no way to know if the accesser has
753 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
760 * Check to see if there are any leases on this file.
761 * This may block while leases are broken.
763 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? O_WRONLY
: 0));
764 if (host_err
== -EWOULDBLOCK
)
765 host_err
= -ETIMEDOUT
;
766 if (host_err
) /* NOMEM or WOULDBLOCK */
769 if (access
& NFSD_MAY_WRITE
) {
770 if (access
& NFSD_MAY_READ
)
771 flags
= O_RDWR
|O_LARGEFILE
;
773 flags
= O_WRONLY
|O_LARGEFILE
;
775 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
776 flags
, current_cred());
778 host_err
= PTR_ERR(*filp
);
780 host_err
= ima_file_check(*filp
, access
);
782 err
= nfserrno(host_err
);
784 validate_process_creds();
792 nfsd_close(struct file
*filp
)
798 * Obtain the readahead parameters for the file
799 * specified by (dev, ino).
802 static inline struct raparms
*
803 nfsd_get_raparms(dev_t dev
, ino_t ino
)
805 struct raparms
*ra
, **rap
, **frap
= NULL
;
808 struct raparm_hbucket
*rab
;
810 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
811 rab
= &raparm_hash
[hash
];
813 spin_lock(&rab
->pb_lock
);
814 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
815 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
818 if (ra
->p_count
== 0)
821 depth
= nfsdstats
.ra_size
*11/10;
823 spin_unlock(&rab
->pb_lock
);
833 if (rap
!= &rab
->pb_head
) {
835 ra
->p_next
= rab
->pb_head
;
839 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
840 spin_unlock(&rab
->pb_lock
);
845 * Grab and keep cached pages associated with a file in the svc_rqst
846 * so that they can be passed to the network sendmsg/sendpage routines
847 * directly. They will be released after the sending has completed.
850 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
851 struct splice_desc
*sd
)
853 struct svc_rqst
*rqstp
= sd
->u
.data
;
854 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
855 struct page
*page
= buf
->page
;
859 ret
= buf
->ops
->confirm(pipe
, buf
);
865 if (rqstp
->rq_res
.page_len
== 0) {
870 rqstp
->rq_res
.page_base
= buf
->offset
;
871 rqstp
->rq_res
.page_len
= size
;
872 } else if (page
!= pp
[-1]) {
878 rqstp
->rq_res
.page_len
+= size
;
880 rqstp
->rq_res
.page_len
+= size
;
885 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
886 struct splice_desc
*sd
)
888 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
891 static inline int svc_msnfs(struct svc_fh
*ffhp
)
894 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
901 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
902 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
911 inode
= file
->f_path
.dentry
->d_inode
;
913 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
916 /* Get readahead parameters */
917 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
920 file
->f_ra
= ra
->p_ra
;
922 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
923 struct splice_desc sd
= {
930 rqstp
->rq_resused
= 1;
931 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
935 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
939 /* Write back readahead params */
941 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
942 spin_lock(&rab
->pb_lock
);
943 ra
->p_ra
= file
->f_ra
;
946 spin_unlock(&rab
->pb_lock
);
950 nfsdstats
.io_read
+= host_err
;
953 fsnotify_access(file
->f_path
.dentry
);
955 err
= nfserrno(host_err
);
960 static void kill_suid(struct dentry
*dentry
)
963 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
965 mutex_lock(&dentry
->d_inode
->i_mutex
);
966 notify_change(dentry
, &ia
);
967 mutex_unlock(&dentry
->d_inode
->i_mutex
);
971 * Gathered writes: If another process is currently writing to the file,
972 * there's a high chance this is another nfsd (triggered by a bulk write
973 * from a client's biod). Rather than syncing the file with each write
974 * request, we sleep for 10 msec.
976 * I don't know if this roughly approximates C. Juszak's idea of
977 * gathered writes, but it's a nice and simple solution (IMHO), and it
980 * Note: we do this only in the NFSv2 case, since v3 and higher have a
981 * better tool (separate unstable writes and commits) for solving this
984 static int wait_for_concurrent_writes(struct file
*file
)
986 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
987 static ino_t last_ino
;
988 static dev_t last_dev
;
991 if (atomic_read(&inode
->i_writecount
) > 1
992 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
993 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
995 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
998 if (inode
->i_state
& I_DIRTY
) {
999 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1000 err
= vfs_fsync(file
, file
->f_path
.dentry
, 0);
1002 last_ino
= inode
->i_ino
;
1003 last_dev
= inode
->i_sb
->s_dev
;
1008 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1009 loff_t offset
, struct kvec
*vec
, int vlen
,
1010 unsigned long *cnt
, int *stablep
)
1012 struct svc_export
*exp
;
1013 struct dentry
*dentry
;
1014 struct inode
*inode
;
1018 int stable
= *stablep
;
1024 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1025 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
1029 dentry
= file
->f_path
.dentry
;
1030 inode
= dentry
->d_inode
;
1031 exp
= fhp
->fh_export
;
1034 * Request sync writes if
1035 * - the sync export option has been set, or
1036 * - the client requested O_SYNC behavior (NFSv3 feature).
1037 * - The file system doesn't support fsync().
1038 * When NFSv2 gathered writes have been configured for this volume,
1039 * flushing the data to disk is handled separately below.
1041 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1043 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1045 *stablep
= 2; /* FILE_SYNC */
1048 if (!EX_ISSYNC(exp
))
1050 if (stable
&& !use_wgather
) {
1051 spin_lock(&file
->f_lock
);
1052 file
->f_flags
|= O_SYNC
;
1053 spin_unlock(&file
->f_lock
);
1056 /* Write the data. */
1057 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1058 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1063 nfsdstats
.io_write
+= host_err
;
1064 fsnotify_modify(file
->f_path
.dentry
);
1066 /* clear setuid/setgid flag after write */
1067 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1070 if (stable
&& use_wgather
)
1071 host_err
= wait_for_concurrent_writes(file
);
1074 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1078 err
= nfserrno(host_err
);
1084 * Read data from a file. count must contain the requested read count
1085 * on entry. On return, *count contains the number of bytes actually read.
1086 * N.B. After this call fhp needs an fh_put
1089 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1090 loff_t offset
, struct kvec
*vec
, int vlen
,
1091 unsigned long *count
)
1096 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1097 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1100 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1102 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1105 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1113 * Write data to a file.
1114 * The stable flag requests synchronous writes.
1115 * N.B. After this call fhp needs an fh_put
1118 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1119 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1125 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1126 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1129 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1132 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1137 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1145 #ifdef CONFIG_NFSD_V3
1147 * Commit all pending writes to stable storage.
1149 * Note: we only guarantee that data that lies within the range specified
1150 * by the 'offset' and 'count' parameters will be synced.
1152 * Unfortunately we cannot lock the file to make sure we return full WCC
1153 * data to the client, as locking happens lower down in the filesystem.
1156 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1157 loff_t offset
, unsigned long count
)
1160 loff_t end
= LLONG_MAX
;
1161 __be32 err
= nfserr_inval
;
1166 end
= offset
+ (loff_t
)count
- 1;
1171 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1174 if (EX_ISSYNC(fhp
->fh_export
)) {
1175 int err2
= vfs_fsync_range(file
, file
->f_path
.dentry
,
1178 if (err2
!= -EINVAL
)
1179 err
= nfserrno(err2
);
1181 err
= nfserr_notsupp
;
1188 #endif /* CONFIG_NFSD_V3 */
1191 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1195 * Mode has already been set earlier in create:
1197 iap
->ia_valid
&= ~ATTR_MODE
;
1199 * Setting uid/gid works only for root. Irix appears to
1200 * send along the gid on create when it tries to implement
1201 * setgid directories via NFS:
1203 if (current_fsuid() != 0)
1204 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1206 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1210 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1211 * setting size to 0 may fail for some specific file systems by the permission
1212 * checking which requires WRITE permission but the mode is 000.
1213 * we ignore the resizing(to 0) on the just new created file, since the size is
1214 * 0 after file created.
1216 * call this only after vfs_create() is called.
1219 nfsd_check_ignore_resizing(struct iattr
*iap
)
1221 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1222 iap
->ia_valid
&= ~ATTR_SIZE
;
1226 * Create a file (regular, directory, device, fifo); UNIX sockets
1227 * not yet implemented.
1228 * If the response fh has been verified, the parent directory should
1229 * already be locked. Note that the parent directory is left locked.
1231 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1234 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1235 char *fname
, int flen
, struct iattr
*iap
,
1236 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1238 struct dentry
*dentry
, *dchild
= NULL
;
1248 if (isdotent(fname
, flen
))
1251 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1255 dentry
= fhp
->fh_dentry
;
1256 dirp
= dentry
->d_inode
;
1258 err
= nfserr_notdir
;
1259 if (!dirp
->i_op
->lookup
)
1262 * Check whether the response file handle has been verified yet.
1263 * If it has, the parent directory should already be locked.
1265 if (!resfhp
->fh_dentry
) {
1266 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1267 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1268 dchild
= lookup_one_len(fname
, dentry
, flen
);
1269 host_err
= PTR_ERR(dchild
);
1272 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1276 /* called from nfsd_proc_create */
1277 dchild
= dget(resfhp
->fh_dentry
);
1278 if (!fhp
->fh_locked
) {
1279 /* not actually possible */
1281 "nfsd_create: parent %s/%s not locked!\n",
1282 dentry
->d_parent
->d_name
.name
,
1283 dentry
->d_name
.name
);
1289 * Make sure the child dentry is still negative ...
1292 if (dchild
->d_inode
) {
1293 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1294 dentry
->d_name
.name
, dchild
->d_name
.name
);
1298 if (!(iap
->ia_valid
& ATTR_MODE
))
1300 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1303 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1304 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1309 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1314 * Get the dir op function pointer.
1319 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1321 nfsd_check_ignore_resizing(iap
);
1324 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1330 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1334 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1338 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1341 * nfsd_setattr already committed the child. Transactional filesystems
1342 * had a chance to commit changes for both parent and child
1343 * simultaneously making the following commit_metadata a noop.
1345 err2
= nfserrno(commit_metadata(fhp
));
1348 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1350 * Update the file handle to get the new inode info.
1353 err
= fh_update(resfhp
);
1355 if (dchild
&& !IS_ERR(dchild
))
1360 err
= nfserrno(host_err
);
1364 #ifdef CONFIG_NFSD_V3
1366 * NFSv3 version of nfsd_create
1369 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1370 char *fname
, int flen
, struct iattr
*iap
,
1371 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1372 int *truncp
, int *created
)
1374 struct dentry
*dentry
, *dchild
= NULL
;
1378 __u32 v_mtime
=0, v_atime
=0;
1384 if (isdotent(fname
, flen
))
1386 if (!(iap
->ia_valid
& ATTR_MODE
))
1388 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1392 dentry
= fhp
->fh_dentry
;
1393 dirp
= dentry
->d_inode
;
1395 /* Get all the sanity checks out of the way before
1396 * we lock the parent. */
1397 err
= nfserr_notdir
;
1398 if (!dirp
->i_op
->lookup
)
1400 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1403 * Compose the response file handle.
1405 dchild
= lookup_one_len(fname
, dentry
, flen
);
1406 host_err
= PTR_ERR(dchild
);
1410 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1414 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1415 /* solaris7 gets confused (bugid 4218508) if these have
1416 * the high bit set, so just clear the high bits. If this is
1417 * ever changed to use different attrs for storing the
1418 * verifier, then do_open_lookup() will also need to be fixed
1421 v_mtime
= verifier
[0]&0x7fffffff;
1422 v_atime
= verifier
[1]&0x7fffffff;
1425 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1428 if (dchild
->d_inode
) {
1431 switch (createmode
) {
1432 case NFS3_CREATE_UNCHECKED
:
1433 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1436 /* in nfsv4, we need to treat this case a little
1437 * differently. we don't want to truncate the
1438 * file now; this would be wrong if the OPEN
1439 * fails for some other reason. furthermore,
1440 * if the size is nonzero, we should ignore it
1441 * according to spec!
1443 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1446 iap
->ia_valid
&= ATTR_SIZE
;
1450 case NFS3_CREATE_EXCLUSIVE
:
1451 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1452 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1453 && dchild
->d_inode
->i_size
== 0 )
1456 case NFS3_CREATE_GUARDED
:
1459 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1463 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1465 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1471 nfsd_check_ignore_resizing(iap
);
1473 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1474 /* Cram the verifier into atime/mtime */
1475 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1476 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1477 /* XXX someone who knows this better please fix it for nsec */
1478 iap
->ia_mtime
.tv_sec
= v_mtime
;
1479 iap
->ia_atime
.tv_sec
= v_atime
;
1480 iap
->ia_mtime
.tv_nsec
= 0;
1481 iap
->ia_atime
.tv_nsec
= 0;
1485 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1488 * nfsd_setattr already committed the child (and possibly also the parent).
1491 err
= nfserrno(commit_metadata(fhp
));
1493 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1495 * Update the filehandle to get the new inode info.
1498 err
= fh_update(resfhp
);
1502 if (dchild
&& !IS_ERR(dchild
))
1507 err
= nfserrno(host_err
);
1510 #endif /* CONFIG_NFSD_V3 */
1513 * Read a symlink. On entry, *lenp must contain the maximum path length that
1514 * fits into the buffer. On return, it contains the true length.
1515 * N.B. After this call fhp needs an fh_put
1518 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1520 struct dentry
*dentry
;
1521 struct inode
*inode
;
1526 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1530 dentry
= fhp
->fh_dentry
;
1531 inode
= dentry
->d_inode
;
1534 if (!inode
->i_op
->readlink
)
1537 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1538 /* N.B. Why does this call need a get_fs()??
1539 * Remove the set_fs and watch the fireworks:-) --okir
1542 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1543 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1554 err
= nfserrno(host_err
);
1559 * Create a symlink and look up its inode
1560 * N.B. After this call _both_ fhp and resfhp need an fh_put
1563 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1564 char *fname
, int flen
,
1565 char *path
, int plen
,
1566 struct svc_fh
*resfhp
,
1569 struct dentry
*dentry
, *dnew
;
1577 if (isdotent(fname
, flen
))
1580 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1584 dentry
= fhp
->fh_dentry
;
1585 dnew
= lookup_one_len(fname
, dentry
, flen
);
1586 host_err
= PTR_ERR(dnew
);
1590 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1594 if (unlikely(path
[plen
] != 0)) {
1595 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1596 if (path_alloced
== NULL
)
1599 strncpy(path_alloced
, path
, plen
);
1600 path_alloced
[plen
] = 0;
1601 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1602 kfree(path_alloced
);
1605 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1606 err
= nfserrno(host_err
);
1608 err
= nfserrno(commit_metadata(fhp
));
1611 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1613 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1615 if (err
==0) err
= cerr
;
1620 err
= nfserrno(host_err
);
1626 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1629 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1630 char *name
, int len
, struct svc_fh
*tfhp
)
1632 struct dentry
*ddir
, *dnew
, *dold
;
1633 struct inode
*dirp
, *dest
;
1637 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1640 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1648 if (isdotent(name
, len
))
1651 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1652 ddir
= ffhp
->fh_dentry
;
1653 dirp
= ddir
->d_inode
;
1655 dnew
= lookup_one_len(name
, ddir
, len
);
1656 host_err
= PTR_ERR(dnew
);
1660 dold
= tfhp
->fh_dentry
;
1661 dest
= dold
->d_inode
;
1663 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1665 err
= nfserrno(host_err
);
1668 host_err
= vfs_link(dold
, dirp
, dnew
);
1670 err
= nfserrno(commit_metadata(ffhp
));
1672 err
= nfserrno(commit_metadata(tfhp
));
1674 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1677 err
= nfserrno(host_err
);
1679 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1688 err
= nfserrno(host_err
);
1694 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1697 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1698 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1700 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1701 struct inode
*fdir
, *tdir
;
1705 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1708 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1712 fdentry
= ffhp
->fh_dentry
;
1713 fdir
= fdentry
->d_inode
;
1715 tdentry
= tfhp
->fh_dentry
;
1716 tdir
= tdentry
->d_inode
;
1718 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1719 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1723 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1726 /* cannot use fh_lock as we need deadlock protective ordering
1727 * so do it by hand */
1728 trap
= lock_rename(tdentry
, fdentry
);
1729 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1733 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1734 host_err
= PTR_ERR(odentry
);
1735 if (IS_ERR(odentry
))
1739 if (!odentry
->d_inode
)
1742 if (odentry
== trap
)
1745 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1746 host_err
= PTR_ERR(ndentry
);
1747 if (IS_ERR(ndentry
))
1749 host_err
= -ENOTEMPTY
;
1750 if (ndentry
== trap
)
1753 if (svc_msnfs(ffhp
) &&
1754 ((atomic_read(&odentry
->d_count
) > 1)
1755 || (atomic_read(&ndentry
->d_count
) > 1))) {
1761 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1763 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1767 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1769 host_err
= commit_metadata(tfhp
);
1771 host_err
= commit_metadata(ffhp
);
1774 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1781 err
= nfserrno(host_err
);
1783 /* we cannot reply on fh_unlock on the two filehandles,
1784 * as that would do the wrong thing if the two directories
1785 * were the same, so again we do it by hand
1787 fill_post_wcc(ffhp
);
1788 fill_post_wcc(tfhp
);
1789 unlock_rename(tdentry
, fdentry
);
1790 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1797 * Unlink a file or directory
1798 * N.B. After this call fhp needs an fh_put
1801 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1802 char *fname
, int flen
)
1804 struct dentry
*dentry
, *rdentry
;
1810 if (!flen
|| isdotent(fname
, flen
))
1812 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1816 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1817 dentry
= fhp
->fh_dentry
;
1818 dirp
= dentry
->d_inode
;
1820 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1821 host_err
= PTR_ERR(rdentry
);
1822 if (IS_ERR(rdentry
))
1825 if (!rdentry
->d_inode
) {
1832 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1834 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1838 if (type
!= S_IFDIR
) { /* It's UNLINK */
1840 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1841 (atomic_read(&rdentry
->d_count
) > 1)) {
1845 host_err
= vfs_unlink(dirp
, rdentry
);
1846 } else { /* It's RMDIR */
1847 host_err
= vfs_rmdir(dirp
, rdentry
);
1853 host_err
= commit_metadata(fhp
);
1855 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1857 err
= nfserrno(host_err
);
1863 * We do this buffering because we must not call back into the file
1864 * system's ->lookup() method from the filldir callback. That may well
1865 * deadlock a number of file systems.
1867 * This is based heavily on the implementation of same in XFS.
1869 struct buffered_dirent
{
1873 unsigned int d_type
;
1877 struct readdir_data
{
1883 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1884 loff_t offset
, u64 ino
, unsigned int d_type
)
1886 struct readdir_data
*buf
= __buf
;
1887 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1888 unsigned int reclen
;
1890 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1891 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1896 de
->namlen
= namlen
;
1897 de
->offset
= offset
;
1899 de
->d_type
= d_type
;
1900 memcpy(de
->name
, name
, namlen
);
1901 buf
->used
+= reclen
;
1906 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1907 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1909 struct readdir_data buf
;
1910 struct buffered_dirent
*de
;
1915 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1917 return nfserrno(-ENOMEM
);
1922 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1923 unsigned int reclen
;
1925 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1929 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1942 * Various filldir functions may end up calling back into
1943 * lookup_one_len() and the file system's ->lookup() method.
1944 * These expect i_mutex to be held, as it would within readdir.
1946 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1950 de
= (struct buffered_dirent
*)buf
.dirent
;
1952 offset
= de
->offset
;
1954 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1955 de
->ino
, de
->d_type
))
1958 if (cdp
->err
!= nfs_ok
)
1961 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1964 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1966 mutex_unlock(&dir_inode
->i_mutex
);
1967 if (size
> 0) /* We bailed out early */
1970 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1973 free_page((unsigned long)(buf
.dirent
));
1976 return nfserrno(host_err
);
1983 * Read entries from a directory.
1984 * The NFSv3/4 verifier we ignore for now.
1987 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1988 struct readdir_cd
*cdp
, filldir_t func
)
1992 loff_t offset
= *offsetp
;
1994 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1998 offset
= vfs_llseek(file
, offset
, 0);
2000 err
= nfserrno((int)offset
);
2004 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2006 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2007 err
= nfs_ok
; /* can still be found in ->err */
2015 * Get file system stats
2016 * N.B. After this call fhp needs an fh_put
2019 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2021 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2022 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
2027 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2029 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2033 * Check for a user's access permissions to this inode.
2036 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2037 struct dentry
*dentry
, int acc
)
2039 struct inode
*inode
= dentry
->d_inode
;
2043 if (acc
== NFSD_MAY_NOP
)
2046 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2048 (acc
& NFSD_MAY_READ
)? " read" : "",
2049 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2050 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2051 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2052 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2053 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2054 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2056 IS_IMMUTABLE(inode
)? " immut" : "",
2057 IS_APPEND(inode
)? " append" : "",
2058 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2059 dprintk(" owner %d/%d user %d/%d\n",
2060 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2063 /* Normally we reject any write/sattr etc access on a read-only file
2064 * system. But if it is IRIX doing check on write-access for a
2065 * device special file, we ignore rofs.
2067 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2068 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2069 if (exp_rdonly(rqstp
, exp
) ||
2070 __mnt_is_readonly(exp
->ex_path
.mnt
))
2072 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2075 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2078 if (acc
& NFSD_MAY_LOCK
) {
2079 /* If we cannot rely on authentication in NLM requests,
2080 * just allow locks, otherwise require read permission, or
2083 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2086 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2089 * The file owner always gets access permission for accesses that
2090 * would normally be checked at open time. This is to make
2091 * file access work even when the client has done a fchmod(fd, 0).
2093 * However, `cp foo bar' should fail nevertheless when bar is
2094 * readonly. A sensible way to do this might be to reject all
2095 * attempts to truncate a read-only file, because a creat() call
2096 * always implies file truncation.
2097 * ... but this isn't really fair. A process may reasonably call
2098 * ftruncate on an open file descriptor on a file with perm 000.
2099 * We must trust the client to do permission checking - using "ACCESS"
2102 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2103 inode
->i_uid
== current_fsuid())
2106 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2107 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2109 /* Allow read access to binaries even when mode 111 */
2110 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2111 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2112 err
= inode_permission(inode
, MAY_EXEC
);
2116 /* Do integrity (permission) checking now, but defer incrementing
2117 * IMA counts to the actual file open.
2119 path
.mnt
= exp
->ex_path
.mnt
;
2120 path
.dentry
= dentry
;
2122 return err
? nfserrno(err
) : 0;
2126 nfsd_racache_shutdown(void)
2128 struct raparms
*raparm
, *last_raparm
;
2131 dprintk("nfsd: freeing readahead buffers.\n");
2133 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2134 raparm
= raparm_hash
[i
].pb_head
;
2136 last_raparm
= raparm
;
2137 raparm
= raparm
->p_next
;
2140 raparm_hash
[i
].pb_head
= NULL
;
2144 * Initialize readahead param cache
2147 nfsd_racache_init(int cache_size
)
2152 struct raparms
**raparm
= NULL
;
2155 if (raparm_hash
[0].pb_head
)
2157 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2160 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2162 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2164 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2165 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2167 raparm
= &raparm_hash
[i
].pb_head
;
2168 for (j
= 0; j
< nperbucket
; j
++) {
2169 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2172 raparm
= &(*raparm
)->p_next
;
2177 nfsdstats
.ra_size
= cache_size
;
2181 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2182 nfsd_racache_shutdown();
2186 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2188 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2190 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2194 struct posix_acl
*acl
;
2196 if (!IS_POSIXACL(inode
))
2197 return ERR_PTR(-EOPNOTSUPP
);
2200 case ACL_TYPE_ACCESS
:
2201 name
= POSIX_ACL_XATTR_ACCESS
;
2203 case ACL_TYPE_DEFAULT
:
2204 name
= POSIX_ACL_XATTR_DEFAULT
;
2207 return ERR_PTR(-EOPNOTSUPP
);
2210 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2212 return ERR_PTR(size
);
2214 acl
= posix_acl_from_xattr(value
, size
);
2220 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2222 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2228 if (!IS_POSIXACL(inode
) ||
2229 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2232 case ACL_TYPE_ACCESS
:
2233 name
= POSIX_ACL_XATTR_ACCESS
;
2235 case ACL_TYPE_DEFAULT
:
2236 name
= POSIX_ACL_XATTR_DEFAULT
;
2242 if (acl
&& acl
->a_count
) {
2243 size
= posix_acl_xattr_size(acl
->a_count
);
2244 value
= kmalloc(size
, GFP_KERNEL
);
2247 error
= posix_acl_to_xattr(acl
, value
, size
);
2254 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2258 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2260 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2263 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2264 if (error
== -ENODATA
)
2268 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2274 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */