1 #define MSNFS /* HACK HACK */
5 * File operations used by nfsd. Some of these have been ripped from
6 * other parts of the kernel because they weren't exported, others
7 * are partial duplicates with added or changed functionality.
9 * Note that several functions dget() the dentry upon which they want
10 * to act, most notably those that create directory entries. Response
11 * dentry's are dput()'d if necessary in the release callback.
12 * So if you notice code paths that apparently fail to dput() the
13 * dentry, don't worry--they have been taken care of.
15 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
16 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
19 #include <linux/string.h>
20 #include <linux/time.h>
21 #include <linux/errno.h>
23 #include <linux/file.h>
24 #include <linux/mount.h>
25 #include <linux/major.h>
26 #include <linux/splice.h>
27 #include <linux/proc_fs.h>
28 #include <linux/stat.h>
29 #include <linux/fcntl.h>
30 #include <linux/net.h>
31 #include <linux/unistd.h>
32 #include <linux/slab.h>
33 #include <linux/pagemap.h>
35 #include <linux/module.h>
36 #include <linux/namei.h>
37 #include <linux/vfs.h>
38 #include <linux/delay.h>
39 #include <linux/sunrpc/svc.h>
40 #include <linux/nfsd/nfsd.h>
42 #include <linux/nfs3.h>
43 #include <linux/nfsd/xdr3.h>
44 #endif /* CONFIG_NFSD_V3 */
45 #include <linux/nfsd/nfsfh.h>
46 #include <linux/quotaops.h>
47 #include <linux/fsnotify.h>
48 #include <linux/posix_acl.h>
49 #include <linux/posix_acl_xattr.h>
50 #include <linux/xattr.h>
52 #include <linux/nfs4.h>
53 #include <linux/nfs4_acl.h>
54 #include <linux/nfsd_idmap.h>
55 #include <linux/security.h>
56 #endif /* CONFIG_NFSD_V4 */
57 #include <linux/jhash.h>
59 #include <asm/uaccess.h>
61 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
65 * This is a cache of readahead params that help us choose the proper
66 * readahead strategy. Initially, we set all readahead parameters to 0
67 * and let the VFS handle things.
68 * If you increase the number of cached files very much, you'll need to
69 * add a hash table here.
72 struct raparms
*p_next
;
77 struct file_ra_state p_ra
;
78 unsigned int p_hindex
;
81 struct raparm_hbucket
{
82 struct raparms
*pb_head
;
84 } ____cacheline_aligned_in_smp
;
86 #define RAPARM_HASH_BITS 4
87 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
88 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
89 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
92 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
94 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
95 * or nfs_ok having possibly changed *dpp and *expp
98 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
99 struct svc_export
**expp
)
101 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
102 struct dentry
*dentry
= *dpp
;
103 struct vfsmount
*mnt
= mntget(exp
->ex_path
.mnt
);
104 struct dentry
*mounts
= dget(dentry
);
107 while (follow_down(&mnt
,&mounts
)&&d_mountpoint(mounts
));
109 exp2
= rqst_exp_get_by_name(rqstp
, mnt
, mounts
);
111 if (PTR_ERR(exp2
) != -ENOENT
)
117 if ((exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
118 /* successfully crossed mount point */
120 * This is subtle: dentry is *not* under mnt at this point.
121 * The only reason we are safe is that original mnt is pinned
122 * down by exp, so we should dput before putting exp.
138 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
139 const char *name
, unsigned int len
,
140 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
142 struct svc_export
*exp
;
143 struct dentry
*dparent
;
144 struct dentry
*dentry
;
148 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
150 /* Obtain dentry and export. */
151 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
155 dparent
= fhp
->fh_dentry
;
156 exp
= fhp
->fh_export
;
159 /* Lookup the name, but don't follow links */
160 if (isdotent(name
, len
)) {
162 dentry
= dget(dparent
);
163 else if (dparent
!= exp
->ex_path
.dentry
)
164 dentry
= dget_parent(dparent
);
165 else if (!EX_NOHIDE(exp
))
166 dentry
= dget(dparent
); /* .. == . just like at / */
168 /* checking mountpoint crossing is very different when stepping up */
169 struct svc_export
*exp2
= NULL
;
171 struct vfsmount
*mnt
= mntget(exp
->ex_path
.mnt
);
172 dentry
= dget(dparent
);
173 while(dentry
== mnt
->mnt_root
&& follow_up(&mnt
, &dentry
))
175 dp
= dget_parent(dentry
);
179 exp2
= rqst_exp_parent(rqstp
, mnt
, dentry
);
180 if (PTR_ERR(exp2
) == -ENOENT
) {
182 dentry
= dget(dparent
);
183 } else if (IS_ERR(exp2
)) {
184 host_err
= PTR_ERR(exp2
);
196 dentry
= lookup_one_len(name
, dparent
, len
);
197 host_err
= PTR_ERR(dentry
);
201 * check if we have crossed a mount point ...
203 if (d_mountpoint(dentry
)) {
204 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
210 *dentry_ret
= dentry
;
216 return nfserrno(host_err
);
220 * Look up one component of a pathname.
221 * N.B. After this call _both_ fhp and resfh need an fh_put
223 * If the lookup would cross a mountpoint, and the mounted filesystem
224 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
225 * accepted as it stands and the mounted directory is
226 * returned. Otherwise the covered directory is returned.
227 * NOTE: this mountpoint crossing is not supported properly by all
228 * clients and is explicitly disallowed for NFSv3
229 * NeilBrown <neilb@cse.unsw.edu.au>
232 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
233 unsigned int len
, struct svc_fh
*resfh
)
235 struct svc_export
*exp
;
236 struct dentry
*dentry
;
239 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
242 err
= check_nfsd_access(exp
, rqstp
);
246 * Note: we compose the file handle now, but as the
247 * dentry may be negative, it may need to be updated.
249 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
250 if (!err
&& !dentry
->d_inode
)
260 * Set various file attributes.
261 * N.B. After this call fhp needs an fh_put
264 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
265 int check_guard
, time_t guardtime
)
267 struct dentry
*dentry
;
269 int accmode
= NFSD_MAY_SATTR
;
275 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
276 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
277 if (iap
->ia_valid
& ATTR_SIZE
)
281 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
285 dentry
= fhp
->fh_dentry
;
286 inode
= dentry
->d_inode
;
288 /* Ignore any mode updates on symlinks */
289 if (S_ISLNK(inode
->i_mode
))
290 iap
->ia_valid
&= ~ATTR_MODE
;
296 * NFSv2 does not differentiate between "set-[ac]time-to-now"
297 * which only requires access, and "set-[ac]time-to-X" which
298 * requires ownership.
299 * So if it looks like it might be "set both to the same time which
300 * is close to now", and if inode_change_ok fails, then we
301 * convert to "set to now" instead of "set to explicit time"
303 * We only call inode_change_ok as the last test as technically
304 * it is not an interface that we should be using. It is only
305 * valid if the filesystem does not define it's own i_op->setattr.
307 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
308 #define MAX_TOUCH_TIME_ERROR (30*60)
309 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
310 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
314 * Now just make sure time is in the right ballpark.
315 * Solaris, at least, doesn't seem to care what the time
316 * request is. We require it be within 30 minutes of now.
318 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
321 if (delta
< MAX_TOUCH_TIME_ERROR
&&
322 inode_change_ok(inode
, iap
) != 0) {
324 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
325 * This will cause notify_change to set these times
328 iap
->ia_valid
&= ~BOTH_TIME_SET
;
333 * The size case is special.
334 * It changes the file as well as the attributes.
336 if (iap
->ia_valid
& ATTR_SIZE
) {
337 if (iap
->ia_size
< inode
->i_size
) {
338 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
339 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
345 * If we are changing the size of the file, then
346 * we need to break all leases.
348 host_err
= break_lease(inode
, FMODE_WRITE
| O_NONBLOCK
);
349 if (host_err
== -EWOULDBLOCK
)
350 host_err
= -ETIMEDOUT
;
351 if (host_err
) /* ENOMEM or EWOULDBLOCK */
354 host_err
= get_write_access(inode
);
359 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
361 put_write_access(inode
);
367 /* sanitize the mode change */
368 if (iap
->ia_valid
& ATTR_MODE
) {
369 iap
->ia_mode
&= S_IALLUGO
;
370 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
373 /* Revoke setuid/setgid on chown */
374 if (!S_ISDIR(inode
->i_mode
) &&
375 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
376 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
377 iap
->ia_valid
|= ATTR_KILL_PRIV
;
378 if (iap
->ia_valid
& ATTR_MODE
) {
379 /* we're setting mode too, just clear the s*id bits */
380 iap
->ia_mode
&= ~S_ISUID
;
381 if (iap
->ia_mode
& S_IXGRP
)
382 iap
->ia_mode
&= ~S_ISGID
;
384 /* set ATTR_KILL_* bits and let VFS handle it */
385 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
389 /* Change the attributes. */
391 iap
->ia_valid
|= ATTR_CTIME
;
393 err
= nfserr_notsync
;
394 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
396 host_err
= notify_change(dentry
, iap
);
397 err
= nfserrno(host_err
);
401 put_write_access(inode
);
403 if (EX_ISSYNC(fhp
->fh_export
))
404 write_inode_now(inode
, 1);
409 err
= nfserrno(host_err
);
413 #if defined(CONFIG_NFSD_V2_ACL) || \
414 defined(CONFIG_NFSD_V3_ACL) || \
415 defined(CONFIG_NFSD_V4)
416 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
421 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
425 *buf
= kmalloc(buflen
, GFP_KERNEL
);
429 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
436 #if defined(CONFIG_NFSD_V4)
438 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
445 buflen
= posix_acl_xattr_size(pacl
->a_count
);
446 buf
= kmalloc(buflen
, GFP_KERNEL
);
451 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
457 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
464 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
465 struct nfs4_acl
*acl
)
469 struct dentry
*dentry
;
471 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
472 unsigned int flags
= 0;
475 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
479 dentry
= fhp
->fh_dentry
;
480 inode
= dentry
->d_inode
;
481 if (S_ISDIR(inode
->i_mode
))
482 flags
= NFS4_ACL_DIR
;
484 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
485 if (host_error
== -EINVAL
) {
486 return nfserr_attrnotsupp
;
487 } else if (host_error
< 0)
490 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
494 if (S_ISDIR(inode
->i_mode
))
495 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
498 posix_acl_release(pacl
);
499 posix_acl_release(dpacl
);
501 if (host_error
== -EOPNOTSUPP
)
502 return nfserr_attrnotsupp
;
504 return nfserrno(host_error
);
507 static struct posix_acl
*
508 _get_posix_acl(struct dentry
*dentry
, char *key
)
511 struct posix_acl
*pacl
= NULL
;
514 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
518 return ERR_PTR(buflen
);
520 pacl
= posix_acl_from_xattr(buf
, buflen
);
526 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
528 struct inode
*inode
= dentry
->d_inode
;
530 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
531 unsigned int flags
= 0;
533 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
534 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
535 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
537 error
= PTR_ERR(pacl
);
542 if (S_ISDIR(inode
->i_mode
)) {
543 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
544 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
546 else if (IS_ERR(dpacl
)) {
547 error
= PTR_ERR(dpacl
);
551 flags
= NFS4_ACL_DIR
;
554 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
556 error
= PTR_ERR(*acl
);
560 posix_acl_release(pacl
);
561 posix_acl_release(dpacl
);
565 #endif /* defined(CONFIG_NFS_V4) */
567 #ifdef CONFIG_NFSD_V3
569 * Check server access rights to a file system object
575 static struct accessmap nfs3_regaccess
[] = {
576 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
577 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
578 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
579 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
584 static struct accessmap nfs3_diraccess
[] = {
585 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
586 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
587 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
588 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
589 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
594 static struct accessmap nfs3_anyaccess
[] = {
595 /* Some clients - Solaris 2.6 at least, make an access call
596 * to the server to check for access for things like /dev/null
597 * (which really, the server doesn't care about). So
598 * We provide simple access checking for them, looking
599 * mainly at mode bits, and we make sure to ignore read-only
602 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
603 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
604 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
605 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
611 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
613 struct accessmap
*map
;
614 struct svc_export
*export
;
615 struct dentry
*dentry
;
616 u32 query
, result
= 0, sresult
= 0;
619 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
623 export
= fhp
->fh_export
;
624 dentry
= fhp
->fh_dentry
;
626 if (S_ISREG(dentry
->d_inode
->i_mode
))
627 map
= nfs3_regaccess
;
628 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
629 map
= nfs3_diraccess
;
631 map
= nfs3_anyaccess
;
635 for (; map
->access
; map
++) {
636 if (map
->access
& query
) {
639 sresult
|= map
->access
;
641 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
644 result
|= map
->access
;
647 /* the following error codes just mean the access was not allowed,
648 * rather than an error occurred */
652 /* simply don't "or" in the access bit. */
662 *supported
= sresult
;
667 #endif /* CONFIG_NFSD_V3 */
672 * Open an existing file or directory.
673 * The access argument indicates the type of open (read/write/lock)
674 * N.B. After this call fhp needs an fh_put
677 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
678 int access
, struct file
**filp
)
680 const struct cred
*cred
= current_cred();
681 struct dentry
*dentry
;
683 int flags
= O_RDONLY
|O_LARGEFILE
;
688 * If we get here, then the client has already done an "open",
689 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
690 * in case a chmod has now revoked permission.
692 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
696 dentry
= fhp
->fh_dentry
;
697 inode
= dentry
->d_inode
;
699 /* Disallow write access to files with the append-only bit set
700 * or any access when mandatory locking enabled
703 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
706 * We must ignore files (but only files) which might have mandatory
707 * locks on them because there is no way to know if the accesser has
710 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
717 * Check to see if there are any leases on this file.
718 * This may block while leases are broken.
720 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
721 if (host_err
== -EWOULDBLOCK
)
722 host_err
= -ETIMEDOUT
;
723 if (host_err
) /* NOMEM or WOULDBLOCK */
726 if (access
& NFSD_MAY_WRITE
) {
727 if (access
& NFSD_MAY_READ
)
728 flags
= O_RDWR
|O_LARGEFILE
;
730 flags
= O_WRONLY
|O_LARGEFILE
;
734 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
737 host_err
= PTR_ERR(*filp
);
739 err
= nfserrno(host_err
);
748 nfsd_close(struct file
*filp
)
755 * As this calls fsync (not fdatasync) there is no need for a write_inode
758 static inline int nfsd_dosync(struct file
*filp
, struct dentry
*dp
,
759 const struct file_operations
*fop
)
761 struct inode
*inode
= dp
->d_inode
;
762 int (*fsync
) (struct file
*, struct dentry
*, int);
765 err
= filemap_fdatawrite(inode
->i_mapping
);
766 if (err
== 0 && fop
&& (fsync
= fop
->fsync
))
767 err
= fsync(filp
, dp
, 0);
769 err
= filemap_fdatawait(inode
->i_mapping
);
775 nfsd_sync(struct file
*filp
)
778 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
779 dprintk("nfsd: sync file %s\n", filp
->f_path
.dentry
->d_name
.name
);
780 mutex_lock(&inode
->i_mutex
);
781 err
=nfsd_dosync(filp
, filp
->f_path
.dentry
, filp
->f_op
);
782 mutex_unlock(&inode
->i_mutex
);
788 nfsd_sync_dir(struct dentry
*dp
)
790 return nfsd_dosync(NULL
, dp
, dp
->d_inode
->i_fop
);
794 * Obtain the readahead parameters for the file
795 * specified by (dev, ino).
798 static inline struct raparms
*
799 nfsd_get_raparms(dev_t dev
, ino_t ino
)
801 struct raparms
*ra
, **rap
, **frap
= NULL
;
804 struct raparm_hbucket
*rab
;
806 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
807 rab
= &raparm_hash
[hash
];
809 spin_lock(&rab
->pb_lock
);
810 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
811 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
814 if (ra
->p_count
== 0)
817 depth
= nfsdstats
.ra_size
*11/10;
819 spin_unlock(&rab
->pb_lock
);
829 if (rap
!= &rab
->pb_head
) {
831 ra
->p_next
= rab
->pb_head
;
835 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
836 spin_unlock(&rab
->pb_lock
);
841 * Grab and keep cached pages associated with a file in the svc_rqst
842 * so that they can be passed to the network sendmsg/sendpage routines
843 * directly. They will be released after the sending has completed.
846 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
847 struct splice_desc
*sd
)
849 struct svc_rqst
*rqstp
= sd
->u
.data
;
850 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
851 struct page
*page
= buf
->page
;
855 ret
= buf
->ops
->confirm(pipe
, buf
);
861 if (rqstp
->rq_res
.page_len
== 0) {
866 rqstp
->rq_res
.page_base
= buf
->offset
;
867 rqstp
->rq_res
.page_len
= size
;
868 } else if (page
!= pp
[-1]) {
874 rqstp
->rq_res
.page_len
+= size
;
876 rqstp
->rq_res
.page_len
+= size
;
881 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
882 struct splice_desc
*sd
)
884 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
887 static inline int svc_msnfs(struct svc_fh
*ffhp
)
890 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
897 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
898 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
907 inode
= file
->f_path
.dentry
->d_inode
;
909 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
912 /* Get readahead parameters */
913 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
916 file
->f_ra
= ra
->p_ra
;
918 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
919 struct splice_desc sd
= {
926 rqstp
->rq_resused
= 1;
927 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
931 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
935 /* Write back readahead params */
937 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
938 spin_lock(&rab
->pb_lock
);
939 ra
->p_ra
= file
->f_ra
;
942 spin_unlock(&rab
->pb_lock
);
946 nfsdstats
.io_read
+= host_err
;
949 fsnotify_access(file
->f_path
.dentry
);
951 err
= nfserrno(host_err
);
956 static void kill_suid(struct dentry
*dentry
)
959 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
961 mutex_lock(&dentry
->d_inode
->i_mutex
);
962 notify_change(dentry
, &ia
);
963 mutex_unlock(&dentry
->d_inode
->i_mutex
);
967 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
968 loff_t offset
, struct kvec
*vec
, int vlen
,
969 unsigned long *cnt
, int *stablep
)
971 struct svc_export
*exp
;
972 struct dentry
*dentry
;
977 int stable
= *stablep
;
982 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
983 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
987 dentry
= file
->f_path
.dentry
;
988 inode
= dentry
->d_inode
;
989 exp
= fhp
->fh_export
;
992 * Request sync writes if
993 * - the sync export option has been set, or
994 * - the client requested O_SYNC behavior (NFSv3 feature).
995 * - The file system doesn't support fsync().
996 * When gathered writes have been configured for this volume,
997 * flushing the data to disk is handled separately below.
1000 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1002 *stablep
= 2; /* FILE_SYNC */
1005 if (!EX_ISSYNC(exp
))
1007 if (stable
&& !EX_WGATHER(exp
)) {
1008 spin_lock(&file
->f_lock
);
1009 file
->f_flags
|= O_SYNC
;
1010 spin_unlock(&file
->f_lock
);
1013 /* Write the data. */
1014 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1015 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1017 if (host_err
>= 0) {
1019 nfsdstats
.io_write
+= host_err
;
1020 fsnotify_modify(file
->f_path
.dentry
);
1023 /* clear setuid/setgid flag after write */
1024 if (host_err
>= 0 && (inode
->i_mode
& (S_ISUID
| S_ISGID
)))
1027 if (host_err
>= 0 && stable
) {
1028 static ino_t last_ino
;
1029 static dev_t last_dev
;
1032 * Gathered writes: If another process is currently
1033 * writing to the file, there's a high chance
1034 * this is another nfsd (triggered by a bulk write
1035 * from a client's biod). Rather than syncing the
1036 * file with each write request, we sleep for 10 msec.
1038 * I don't know if this roughly approximates
1039 * C. Juszak's idea of gathered writes, but it's a
1040 * nice and simple solution (IMHO), and it seems to
1043 if (EX_WGATHER(exp
)) {
1044 if (atomic_read(&inode
->i_writecount
) > 1
1045 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1046 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1048 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1051 if (inode
->i_state
& I_DIRTY
) {
1052 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1053 host_err
=nfsd_sync(file
);
1056 wake_up(&inode
->i_wait
);
1059 last_ino
= inode
->i_ino
;
1060 last_dev
= inode
->i_sb
->s_dev
;
1063 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1067 err
= nfserrno(host_err
);
1073 * Read data from a file. count must contain the requested read count
1074 * on entry. On return, *count contains the number of bytes actually read.
1075 * N.B. After this call fhp needs an fh_put
1078 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1079 loff_t offset
, struct kvec
*vec
, int vlen
,
1080 unsigned long *count
)
1085 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1086 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1089 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1091 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1094 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1102 * Write data to a file.
1103 * The stable flag requests synchronous writes.
1104 * N.B. After this call fhp needs an fh_put
1107 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1108 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1114 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1115 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1118 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1121 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1126 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1134 #ifdef CONFIG_NFSD_V3
1136 * Commit all pending writes to stable storage.
1137 * Strictly speaking, we could sync just the indicated file region here,
1138 * but there's currently no way we can ask the VFS to do so.
1140 * Unfortunately we cannot lock the file to make sure we return full WCC
1141 * data to the client, as locking happens lower down in the filesystem.
1144 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1145 loff_t offset
, unsigned long count
)
1150 if ((u64
)count
> ~(u64
)offset
)
1151 return nfserr_inval
;
1153 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1156 if (EX_ISSYNC(fhp
->fh_export
)) {
1157 if (file
->f_op
&& file
->f_op
->fsync
) {
1158 err
= nfserrno(nfsd_sync(file
));
1160 err
= nfserr_notsupp
;
1167 #endif /* CONFIG_NFSD_V3 */
1170 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1174 * Mode has already been set earlier in create:
1176 iap
->ia_valid
&= ~ATTR_MODE
;
1178 * Setting uid/gid works only for root. Irix appears to
1179 * send along the gid on create when it tries to implement
1180 * setgid directories via NFS:
1182 if (current_fsuid() != 0)
1183 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1185 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1189 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1190 * setting size to 0 may fail for some specific file systems by the permission
1191 * checking which requires WRITE permission but the mode is 000.
1192 * we ignore the resizing(to 0) on the just new created file, since the size is
1193 * 0 after file created.
1195 * call this only after vfs_create() is called.
1198 nfsd_check_ignore_resizing(struct iattr
*iap
)
1200 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1201 iap
->ia_valid
&= ~ATTR_SIZE
;
1205 * Create a file (regular, directory, device, fifo); UNIX sockets
1206 * not yet implemented.
1207 * If the response fh has been verified, the parent directory should
1208 * already be locked. Note that the parent directory is left locked.
1210 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1213 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1214 char *fname
, int flen
, struct iattr
*iap
,
1215 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1217 struct dentry
*dentry
, *dchild
= NULL
;
1227 if (isdotent(fname
, flen
))
1230 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1234 dentry
= fhp
->fh_dentry
;
1235 dirp
= dentry
->d_inode
;
1237 err
= nfserr_notdir
;
1238 if (!dirp
->i_op
->lookup
)
1241 * Check whether the response file handle has been verified yet.
1242 * If it has, the parent directory should already be locked.
1244 if (!resfhp
->fh_dentry
) {
1245 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1246 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1247 dchild
= lookup_one_len(fname
, dentry
, flen
);
1248 host_err
= PTR_ERR(dchild
);
1251 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1255 /* called from nfsd_proc_create */
1256 dchild
= dget(resfhp
->fh_dentry
);
1257 if (!fhp
->fh_locked
) {
1258 /* not actually possible */
1260 "nfsd_create: parent %s/%s not locked!\n",
1261 dentry
->d_parent
->d_name
.name
,
1262 dentry
->d_name
.name
);
1268 * Make sure the child dentry is still negative ...
1271 if (dchild
->d_inode
) {
1272 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1273 dentry
->d_name
.name
, dchild
->d_name
.name
);
1277 if (!(iap
->ia_valid
& ATTR_MODE
))
1279 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1282 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1283 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1288 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1293 * Get the dir op function pointer.
1298 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1300 nfsd_check_ignore_resizing(iap
);
1303 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1309 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1313 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1317 if (EX_ISSYNC(fhp
->fh_export
)) {
1318 err
= nfserrno(nfsd_sync_dir(dentry
));
1319 write_inode_now(dchild
->d_inode
, 1);
1322 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1325 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1327 * Update the file handle to get the new inode info.
1330 err
= fh_update(resfhp
);
1332 if (dchild
&& !IS_ERR(dchild
))
1337 err
= nfserrno(host_err
);
1341 #ifdef CONFIG_NFSD_V3
1343 * NFSv3 version of nfsd_create
1346 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1347 char *fname
, int flen
, struct iattr
*iap
,
1348 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1349 int *truncp
, int *created
)
1351 struct dentry
*dentry
, *dchild
= NULL
;
1356 __u32 v_mtime
=0, v_atime
=0;
1362 if (isdotent(fname
, flen
))
1364 if (!(iap
->ia_valid
& ATTR_MODE
))
1366 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1370 dentry
= fhp
->fh_dentry
;
1371 dirp
= dentry
->d_inode
;
1373 /* Get all the sanity checks out of the way before
1374 * we lock the parent. */
1375 err
= nfserr_notdir
;
1376 if (!dirp
->i_op
->lookup
)
1378 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1381 * Compose the response file handle.
1383 dchild
= lookup_one_len(fname
, dentry
, flen
);
1384 host_err
= PTR_ERR(dchild
);
1388 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1392 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1393 /* solaris7 gets confused (bugid 4218508) if these have
1394 * the high bit set, so just clear the high bits. If this is
1395 * ever changed to use different attrs for storing the
1396 * verifier, then do_open_lookup() will also need to be fixed
1399 v_mtime
= verifier
[0]&0x7fffffff;
1400 v_atime
= verifier
[1]&0x7fffffff;
1403 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1406 if (dchild
->d_inode
) {
1409 switch (createmode
) {
1410 case NFS3_CREATE_UNCHECKED
:
1411 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1414 /* in nfsv4, we need to treat this case a little
1415 * differently. we don't want to truncate the
1416 * file now; this would be wrong if the OPEN
1417 * fails for some other reason. furthermore,
1418 * if the size is nonzero, we should ignore it
1419 * according to spec!
1421 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1424 iap
->ia_valid
&= ATTR_SIZE
;
1428 case NFS3_CREATE_EXCLUSIVE
:
1429 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1430 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1431 && dchild
->d_inode
->i_size
== 0 )
1434 case NFS3_CREATE_GUARDED
:
1437 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1441 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1443 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1449 if (EX_ISSYNC(fhp
->fh_export
)) {
1450 err
= nfserrno(nfsd_sync_dir(dentry
));
1451 /* setattr will sync the child (or not) */
1454 nfsd_check_ignore_resizing(iap
);
1456 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1457 /* Cram the verifier into atime/mtime */
1458 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1459 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1460 /* XXX someone who knows this better please fix it for nsec */
1461 iap
->ia_mtime
.tv_sec
= v_mtime
;
1462 iap
->ia_atime
.tv_sec
= v_atime
;
1463 iap
->ia_mtime
.tv_nsec
= 0;
1464 iap
->ia_atime
.tv_nsec
= 0;
1468 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1472 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1474 * Update the filehandle to get the new inode info.
1477 err
= fh_update(resfhp
);
1481 if (dchild
&& !IS_ERR(dchild
))
1486 err
= nfserrno(host_err
);
1489 #endif /* CONFIG_NFSD_V3 */
1492 * Read a symlink. On entry, *lenp must contain the maximum path length that
1493 * fits into the buffer. On return, it contains the true length.
1494 * N.B. After this call fhp needs an fh_put
1497 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1499 struct dentry
*dentry
;
1500 struct inode
*inode
;
1505 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1509 dentry
= fhp
->fh_dentry
;
1510 inode
= dentry
->d_inode
;
1513 if (!inode
->i_op
->readlink
)
1516 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1517 /* N.B. Why does this call need a get_fs()??
1518 * Remove the set_fs and watch the fireworks:-) --okir
1521 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1522 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1533 err
= nfserrno(host_err
);
1538 * Create a symlink and look up its inode
1539 * N.B. After this call _both_ fhp and resfhp need an fh_put
1542 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1543 char *fname
, int flen
,
1544 char *path
, int plen
,
1545 struct svc_fh
*resfhp
,
1548 struct dentry
*dentry
, *dnew
;
1556 if (isdotent(fname
, flen
))
1559 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1563 dentry
= fhp
->fh_dentry
;
1564 dnew
= lookup_one_len(fname
, dentry
, flen
);
1565 host_err
= PTR_ERR(dnew
);
1569 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1573 if (unlikely(path
[plen
] != 0)) {
1574 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1575 if (path_alloced
== NULL
)
1578 strncpy(path_alloced
, path
, plen
);
1579 path_alloced
[plen
] = 0;
1580 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1581 kfree(path_alloced
);
1584 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1587 if (EX_ISSYNC(fhp
->fh_export
))
1588 host_err
= nfsd_sync_dir(dentry
);
1590 err
= nfserrno(host_err
);
1593 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1595 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1597 if (err
==0) err
= cerr
;
1602 err
= nfserrno(host_err
);
1608 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1611 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1612 char *name
, int len
, struct svc_fh
*tfhp
)
1614 struct dentry
*ddir
, *dnew
, *dold
;
1615 struct inode
*dirp
, *dest
;
1619 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1622 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1630 if (isdotent(name
, len
))
1633 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1634 ddir
= ffhp
->fh_dentry
;
1635 dirp
= ddir
->d_inode
;
1637 dnew
= lookup_one_len(name
, ddir
, len
);
1638 host_err
= PTR_ERR(dnew
);
1642 dold
= tfhp
->fh_dentry
;
1643 dest
= dold
->d_inode
;
1645 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1647 err
= nfserrno(host_err
);
1650 host_err
= vfs_link(dold
, dirp
, dnew
);
1652 if (EX_ISSYNC(ffhp
->fh_export
)) {
1653 err
= nfserrno(nfsd_sync_dir(ddir
));
1654 write_inode_now(dest
, 1);
1658 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1661 err
= nfserrno(host_err
);
1663 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1672 err
= nfserrno(host_err
);
1678 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1681 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1682 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1684 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1685 struct inode
*fdir
, *tdir
;
1689 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1692 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1696 fdentry
= ffhp
->fh_dentry
;
1697 fdir
= fdentry
->d_inode
;
1699 tdentry
= tfhp
->fh_dentry
;
1700 tdir
= tdentry
->d_inode
;
1702 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1703 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1707 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1710 /* cannot use fh_lock as we need deadlock protective ordering
1711 * so do it by hand */
1712 trap
= lock_rename(tdentry
, fdentry
);
1713 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1717 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1718 host_err
= PTR_ERR(odentry
);
1719 if (IS_ERR(odentry
))
1723 if (!odentry
->d_inode
)
1726 if (odentry
== trap
)
1729 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1730 host_err
= PTR_ERR(ndentry
);
1731 if (IS_ERR(ndentry
))
1733 host_err
= -ENOTEMPTY
;
1734 if (ndentry
== trap
)
1737 if (svc_msnfs(ffhp
) &&
1738 ((atomic_read(&odentry
->d_count
) > 1)
1739 || (atomic_read(&ndentry
->d_count
) > 1))) {
1745 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1747 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1751 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1752 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1753 host_err
= nfsd_sync_dir(tdentry
);
1755 host_err
= nfsd_sync_dir(fdentry
);
1758 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1765 err
= nfserrno(host_err
);
1767 /* we cannot reply on fh_unlock on the two filehandles,
1768 * as that would do the wrong thing if the two directories
1769 * were the same, so again we do it by hand
1771 fill_post_wcc(ffhp
);
1772 fill_post_wcc(tfhp
);
1773 unlock_rename(tdentry
, fdentry
);
1774 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1781 * Unlink a file or directory
1782 * N.B. After this call fhp needs an fh_put
1785 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1786 char *fname
, int flen
)
1788 struct dentry
*dentry
, *rdentry
;
1794 if (!flen
|| isdotent(fname
, flen
))
1796 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1800 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1801 dentry
= fhp
->fh_dentry
;
1802 dirp
= dentry
->d_inode
;
1804 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1805 host_err
= PTR_ERR(rdentry
);
1806 if (IS_ERR(rdentry
))
1809 if (!rdentry
->d_inode
) {
1816 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1818 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1822 if (type
!= S_IFDIR
) { /* It's UNLINK */
1824 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1825 (atomic_read(&rdentry
->d_count
) > 1)) {
1829 host_err
= vfs_unlink(dirp
, rdentry
);
1830 } else { /* It's RMDIR */
1831 host_err
= vfs_rmdir(dirp
, rdentry
);
1838 if (EX_ISSYNC(fhp
->fh_export
))
1839 host_err
= nfsd_sync_dir(dentry
);
1842 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1844 err
= nfserrno(host_err
);
1850 * We do this buffering because we must not call back into the file
1851 * system's ->lookup() method from the filldir callback. That may well
1852 * deadlock a number of file systems.
1854 * This is based heavily on the implementation of same in XFS.
1856 struct buffered_dirent
{
1860 unsigned int d_type
;
1864 struct readdir_data
{
1870 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1871 loff_t offset
, u64 ino
, unsigned int d_type
)
1873 struct readdir_data
*buf
= __buf
;
1874 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1875 unsigned int reclen
;
1877 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1878 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1883 de
->namlen
= namlen
;
1884 de
->offset
= offset
;
1886 de
->d_type
= d_type
;
1887 memcpy(de
->name
, name
, namlen
);
1888 buf
->used
+= reclen
;
1893 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1894 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1896 struct readdir_data buf
;
1897 struct buffered_dirent
*de
;
1902 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1904 return nfserrno(-ENOMEM
);
1909 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1910 unsigned int reclen
;
1912 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1916 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1929 * Various filldir functions may end up calling back into
1930 * lookup_one_len() and the file system's ->lookup() method.
1931 * These expect i_mutex to be held, as it would within readdir.
1933 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1937 de
= (struct buffered_dirent
*)buf
.dirent
;
1939 offset
= de
->offset
;
1941 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1942 de
->ino
, de
->d_type
))
1945 if (cdp
->err
!= nfs_ok
)
1948 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1951 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1953 mutex_unlock(&dir_inode
->i_mutex
);
1954 if (size
> 0) /* We bailed out early */
1957 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1960 free_page((unsigned long)(buf
.dirent
));
1963 return nfserrno(host_err
);
1970 * Read entries from a directory.
1971 * The NFSv3/4 verifier we ignore for now.
1974 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1975 struct readdir_cd
*cdp
, filldir_t func
)
1979 loff_t offset
= *offsetp
;
1981 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1985 offset
= vfs_llseek(file
, offset
, 0);
1987 err
= nfserrno((int)offset
);
1991 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1993 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1994 err
= nfs_ok
; /* can still be found in ->err */
2002 * Get file system stats
2003 * N.B. After this call fhp needs an fh_put
2006 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2008 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2009 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
2014 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2016 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2020 * Check for a user's access permissions to this inode.
2023 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2024 struct dentry
*dentry
, int acc
)
2026 struct inode
*inode
= dentry
->d_inode
;
2029 if (acc
== NFSD_MAY_NOP
)
2032 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2034 (acc
& NFSD_MAY_READ
)? " read" : "",
2035 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2036 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2037 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2038 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2039 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2040 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2042 IS_IMMUTABLE(inode
)? " immut" : "",
2043 IS_APPEND(inode
)? " append" : "",
2044 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2045 dprintk(" owner %d/%d user %d/%d\n",
2046 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2049 /* Normally we reject any write/sattr etc access on a read-only file
2050 * system. But if it is IRIX doing check on write-access for a
2051 * device special file, we ignore rofs.
2053 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2054 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2055 if (exp_rdonly(rqstp
, exp
) ||
2056 __mnt_is_readonly(exp
->ex_path
.mnt
))
2058 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2061 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2064 if (acc
& NFSD_MAY_LOCK
) {
2065 /* If we cannot rely on authentication in NLM requests,
2066 * just allow locks, otherwise require read permission, or
2069 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2072 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2075 * The file owner always gets access permission for accesses that
2076 * would normally be checked at open time. This is to make
2077 * file access work even when the client has done a fchmod(fd, 0).
2079 * However, `cp foo bar' should fail nevertheless when bar is
2080 * readonly. A sensible way to do this might be to reject all
2081 * attempts to truncate a read-only file, because a creat() call
2082 * always implies file truncation.
2083 * ... but this isn't really fair. A process may reasonably call
2084 * ftruncate on an open file descriptor on a file with perm 000.
2085 * We must trust the client to do permission checking - using "ACCESS"
2088 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2089 inode
->i_uid
== current_fsuid())
2092 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2093 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2095 /* Allow read access to binaries even when mode 111 */
2096 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2097 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2098 err
= inode_permission(inode
, MAY_EXEC
);
2100 return err
? nfserrno(err
) : 0;
2104 nfsd_racache_shutdown(void)
2106 struct raparms
*raparm
, *last_raparm
;
2109 dprintk("nfsd: freeing readahead buffers.\n");
2111 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2112 raparm
= raparm_hash
[i
].pb_head
;
2114 last_raparm
= raparm
;
2115 raparm
= raparm
->p_next
;
2118 raparm_hash
[i
].pb_head
= NULL
;
2122 * Initialize readahead param cache
2125 nfsd_racache_init(int cache_size
)
2130 struct raparms
**raparm
= NULL
;
2133 if (raparm_hash
[0].pb_head
)
2135 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2138 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2140 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2142 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2143 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2145 raparm
= &raparm_hash
[i
].pb_head
;
2146 for (j
= 0; j
< nperbucket
; j
++) {
2147 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2150 raparm
= &(*raparm
)->p_next
;
2155 nfsdstats
.ra_size
= cache_size
;
2159 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2160 nfsd_racache_shutdown();
2164 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2166 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2168 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2172 struct posix_acl
*acl
;
2174 if (!IS_POSIXACL(inode
))
2175 return ERR_PTR(-EOPNOTSUPP
);
2178 case ACL_TYPE_ACCESS
:
2179 name
= POSIX_ACL_XATTR_ACCESS
;
2181 case ACL_TYPE_DEFAULT
:
2182 name
= POSIX_ACL_XATTR_DEFAULT
;
2185 return ERR_PTR(-EOPNOTSUPP
);
2188 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2190 return ERR_PTR(size
);
2192 acl
= posix_acl_from_xattr(value
, size
);
2198 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2200 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2206 if (!IS_POSIXACL(inode
) ||
2207 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2210 case ACL_TYPE_ACCESS
:
2211 name
= POSIX_ACL_XATTR_ACCESS
;
2213 case ACL_TYPE_DEFAULT
:
2214 name
= POSIX_ACL_XATTR_DEFAULT
;
2220 if (acl
&& acl
->a_count
) {
2221 size
= posix_acl_xattr_size(acl
->a_count
);
2222 value
= kmalloc(size
, GFP_KERNEL
);
2225 error
= posix_acl_to_xattr(acl
, value
, size
);
2232 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2236 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2238 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2241 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2242 if (error
== -ENODATA
)
2246 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2252 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */