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 */
133 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
134 const char *name
, unsigned int len
,
135 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
137 struct svc_export
*exp
;
138 struct dentry
*dparent
;
139 struct dentry
*dentry
;
143 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
145 /* Obtain dentry and export. */
146 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
150 dparent
= fhp
->fh_dentry
;
151 exp
= fhp
->fh_export
;
154 /* Lookup the name, but don't follow links */
155 if (isdotent(name
, len
)) {
157 dentry
= dget(dparent
);
158 else if (dparent
!= exp
->ex_path
.dentry
)
159 dentry
= dget_parent(dparent
);
160 else if (!EX_NOHIDE(exp
))
161 dentry
= dget(dparent
); /* .. == . just like at / */
163 /* checking mountpoint crossing is very different when stepping up */
164 struct svc_export
*exp2
= NULL
;
166 struct vfsmount
*mnt
= mntget(exp
->ex_path
.mnt
);
167 dentry
= dget(dparent
);
168 while(dentry
== mnt
->mnt_root
&& follow_up(&mnt
, &dentry
))
170 dp
= dget_parent(dentry
);
174 exp2
= rqst_exp_parent(rqstp
, mnt
, dentry
);
175 if (PTR_ERR(exp2
) == -ENOENT
) {
177 dentry
= dget(dparent
);
178 } else if (IS_ERR(exp2
)) {
179 host_err
= PTR_ERR(exp2
);
191 dentry
= lookup_one_len(name
, dparent
, len
);
192 host_err
= PTR_ERR(dentry
);
196 * check if we have crossed a mount point ...
198 if (d_mountpoint(dentry
)) {
199 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
205 *dentry_ret
= dentry
;
211 return nfserrno(host_err
);
215 * Look up one component of a pathname.
216 * N.B. After this call _both_ fhp and resfh need an fh_put
218 * If the lookup would cross a mountpoint, and the mounted filesystem
219 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
220 * accepted as it stands and the mounted directory is
221 * returned. Otherwise the covered directory is returned.
222 * NOTE: this mountpoint crossing is not supported properly by all
223 * clients and is explicitly disallowed for NFSv3
224 * NeilBrown <neilb@cse.unsw.edu.au>
227 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
228 unsigned int len
, struct svc_fh
*resfh
)
230 struct svc_export
*exp
;
231 struct dentry
*dentry
;
234 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
237 err
= check_nfsd_access(exp
, rqstp
);
241 * Note: we compose the file handle now, but as the
242 * dentry may be negative, it may need to be updated.
244 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
245 if (!err
&& !dentry
->d_inode
)
255 * Set various file attributes.
256 * N.B. After this call fhp needs an fh_put
259 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
260 int check_guard
, time_t guardtime
)
262 struct dentry
*dentry
;
264 int accmode
= NFSD_MAY_SATTR
;
270 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
271 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
272 if (iap
->ia_valid
& ATTR_SIZE
)
276 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
280 dentry
= fhp
->fh_dentry
;
281 inode
= dentry
->d_inode
;
283 /* Ignore any mode updates on symlinks */
284 if (S_ISLNK(inode
->i_mode
))
285 iap
->ia_valid
&= ~ATTR_MODE
;
291 * NFSv2 does not differentiate between "set-[ac]time-to-now"
292 * which only requires access, and "set-[ac]time-to-X" which
293 * requires ownership.
294 * So if it looks like it might be "set both to the same time which
295 * is close to now", and if inode_change_ok fails, then we
296 * convert to "set to now" instead of "set to explicit time"
298 * We only call inode_change_ok as the last test as technically
299 * it is not an interface that we should be using. It is only
300 * valid if the filesystem does not define it's own i_op->setattr.
302 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
303 #define MAX_TOUCH_TIME_ERROR (30*60)
304 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
305 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
309 * Now just make sure time is in the right ballpark.
310 * Solaris, at least, doesn't seem to care what the time
311 * request is. We require it be within 30 minutes of now.
313 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
316 if (delta
< MAX_TOUCH_TIME_ERROR
&&
317 inode_change_ok(inode
, iap
) != 0) {
319 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
320 * This will cause notify_change to set these times
323 iap
->ia_valid
&= ~BOTH_TIME_SET
;
328 * The size case is special.
329 * It changes the file as well as the attributes.
331 if (iap
->ia_valid
& ATTR_SIZE
) {
332 if (iap
->ia_size
< inode
->i_size
) {
333 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
334 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
340 * If we are changing the size of the file, then
341 * we need to break all leases.
343 host_err
= break_lease(inode
, FMODE_WRITE
| O_NONBLOCK
);
344 if (host_err
== -EWOULDBLOCK
)
345 host_err
= -ETIMEDOUT
;
346 if (host_err
) /* ENOMEM or EWOULDBLOCK */
349 host_err
= get_write_access(inode
);
354 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
356 put_write_access(inode
);
362 /* sanitize the mode change */
363 if (iap
->ia_valid
& ATTR_MODE
) {
364 iap
->ia_mode
&= S_IALLUGO
;
365 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
368 /* Revoke setuid/setgid on chown */
369 if (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
370 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
)) {
371 iap
->ia_valid
|= ATTR_KILL_PRIV
;
372 if (iap
->ia_valid
& ATTR_MODE
) {
373 /* we're setting mode too, just clear the s*id bits */
374 iap
->ia_mode
&= ~S_ISUID
;
375 if (iap
->ia_mode
& S_IXGRP
)
376 iap
->ia_mode
&= ~S_ISGID
;
378 /* set ATTR_KILL_* bits and let VFS handle it */
379 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
383 /* Change the attributes. */
385 iap
->ia_valid
|= ATTR_CTIME
;
387 err
= nfserr_notsync
;
388 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
390 host_err
= notify_change(dentry
, iap
);
391 err
= nfserrno(host_err
);
395 put_write_access(inode
);
397 if (EX_ISSYNC(fhp
->fh_export
))
398 write_inode_now(inode
, 1);
403 err
= nfserrno(host_err
);
407 #if defined(CONFIG_NFSD_V2_ACL) || \
408 defined(CONFIG_NFSD_V3_ACL) || \
409 defined(CONFIG_NFSD_V4)
410 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
414 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
418 *buf
= kmalloc(buflen
, GFP_KERNEL
);
422 return vfs_getxattr(dentry
, key
, *buf
, buflen
);
426 #if defined(CONFIG_NFSD_V4)
428 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
435 buflen
= posix_acl_xattr_size(pacl
->a_count
);
436 buf
= kmalloc(buflen
, GFP_KERNEL
);
441 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
447 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
454 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
455 struct nfs4_acl
*acl
)
459 struct dentry
*dentry
;
461 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
462 unsigned int flags
= 0;
465 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
469 dentry
= fhp
->fh_dentry
;
470 inode
= dentry
->d_inode
;
471 if (S_ISDIR(inode
->i_mode
))
472 flags
= NFS4_ACL_DIR
;
474 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
475 if (host_error
== -EINVAL
) {
476 return nfserr_attrnotsupp
;
477 } else if (host_error
< 0)
480 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
484 if (S_ISDIR(inode
->i_mode
))
485 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
488 posix_acl_release(pacl
);
489 posix_acl_release(dpacl
);
491 if (host_error
== -EOPNOTSUPP
)
492 return nfserr_attrnotsupp
;
494 return nfserrno(host_error
);
497 static struct posix_acl
*
498 _get_posix_acl(struct dentry
*dentry
, char *key
)
501 struct posix_acl
*pacl
= NULL
;
504 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
508 return ERR_PTR(buflen
);
510 pacl
= posix_acl_from_xattr(buf
, buflen
);
516 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
518 struct inode
*inode
= dentry
->d_inode
;
520 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
521 unsigned int flags
= 0;
523 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
524 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
525 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
527 error
= PTR_ERR(pacl
);
532 if (S_ISDIR(inode
->i_mode
)) {
533 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
534 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
536 else if (IS_ERR(dpacl
)) {
537 error
= PTR_ERR(dpacl
);
541 flags
= NFS4_ACL_DIR
;
544 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
546 error
= PTR_ERR(*acl
);
550 posix_acl_release(pacl
);
551 posix_acl_release(dpacl
);
555 #endif /* defined(CONFIG_NFS_V4) */
557 #ifdef CONFIG_NFSD_V3
559 * Check server access rights to a file system object
565 static struct accessmap nfs3_regaccess
[] = {
566 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
567 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
568 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
569 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
574 static struct accessmap nfs3_diraccess
[] = {
575 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
576 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
577 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
578 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
579 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
584 static struct accessmap nfs3_anyaccess
[] = {
585 /* Some clients - Solaris 2.6 at least, make an access call
586 * to the server to check for access for things like /dev/null
587 * (which really, the server doesn't care about). So
588 * We provide simple access checking for them, looking
589 * mainly at mode bits, and we make sure to ignore read-only
592 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
593 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
594 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
595 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
601 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
603 struct accessmap
*map
;
604 struct svc_export
*export
;
605 struct dentry
*dentry
;
606 u32 query
, result
= 0, sresult
= 0;
609 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
613 export
= fhp
->fh_export
;
614 dentry
= fhp
->fh_dentry
;
616 if (S_ISREG(dentry
->d_inode
->i_mode
))
617 map
= nfs3_regaccess
;
618 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
619 map
= nfs3_diraccess
;
621 map
= nfs3_anyaccess
;
625 for (; map
->access
; map
++) {
626 if (map
->access
& query
) {
629 sresult
|= map
->access
;
631 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
634 result
|= map
->access
;
637 /* the following error codes just mean the access was not allowed,
638 * rather than an error occurred */
642 /* simply don't "or" in the access bit. */
652 *supported
= sresult
;
657 #endif /* CONFIG_NFSD_V3 */
662 * Open an existing file or directory.
663 * The access argument indicates the type of open (read/write/lock)
664 * N.B. After this call fhp needs an fh_put
667 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
668 int access
, struct file
**filp
)
670 struct dentry
*dentry
;
672 int flags
= O_RDONLY
|O_LARGEFILE
;
677 * If we get here, then the client has already done an "open",
678 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
679 * in case a chmod has now revoked permission.
681 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
685 dentry
= fhp
->fh_dentry
;
686 inode
= dentry
->d_inode
;
688 /* Disallow write access to files with the append-only bit set
689 * or any access when mandatory locking enabled
692 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
695 * We must ignore files (but only files) which might have mandatory
696 * locks on them because there is no way to know if the accesser has
699 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
706 * Check to see if there are any leases on this file.
707 * This may block while leases are broken.
709 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
710 if (host_err
== -EWOULDBLOCK
)
711 host_err
= -ETIMEDOUT
;
712 if (host_err
) /* NOMEM or WOULDBLOCK */
715 if (access
& NFSD_MAY_WRITE
) {
716 if (access
& NFSD_MAY_READ
)
717 flags
= O_RDWR
|O_LARGEFILE
;
719 flags
= O_WRONLY
|O_LARGEFILE
;
723 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
726 host_err
= PTR_ERR(*filp
);
728 err
= nfserrno(host_err
);
737 nfsd_close(struct file
*filp
)
744 * As this calls fsync (not fdatasync) there is no need for a write_inode
747 static inline int nfsd_dosync(struct file
*filp
, struct dentry
*dp
,
748 const struct file_operations
*fop
)
750 struct inode
*inode
= dp
->d_inode
;
751 int (*fsync
) (struct file
*, struct dentry
*, int);
754 err
= filemap_fdatawrite(inode
->i_mapping
);
755 if (err
== 0 && fop
&& (fsync
= fop
->fsync
))
756 err
= fsync(filp
, dp
, 0);
758 err
= filemap_fdatawait(inode
->i_mapping
);
765 nfsd_sync(struct file
*filp
)
768 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
769 dprintk("nfsd: sync file %s\n", filp
->f_path
.dentry
->d_name
.name
);
770 mutex_lock(&inode
->i_mutex
);
771 err
=nfsd_dosync(filp
, filp
->f_path
.dentry
, filp
->f_op
);
772 mutex_unlock(&inode
->i_mutex
);
778 nfsd_sync_dir(struct dentry
*dp
)
780 return nfsd_dosync(NULL
, dp
, dp
->d_inode
->i_fop
);
784 * Obtain the readahead parameters for the file
785 * specified by (dev, ino).
788 static inline struct raparms
*
789 nfsd_get_raparms(dev_t dev
, ino_t ino
)
791 struct raparms
*ra
, **rap
, **frap
= NULL
;
794 struct raparm_hbucket
*rab
;
796 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
797 rab
= &raparm_hash
[hash
];
799 spin_lock(&rab
->pb_lock
);
800 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
801 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
804 if (ra
->p_count
== 0)
807 depth
= nfsdstats
.ra_size
*11/10;
809 spin_unlock(&rab
->pb_lock
);
819 if (rap
!= &rab
->pb_head
) {
821 ra
->p_next
= rab
->pb_head
;
825 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
826 spin_unlock(&rab
->pb_lock
);
831 * Grab and keep cached pages associated with a file in the svc_rqst
832 * so that they can be passed to the network sendmsg/sendpage routines
833 * directly. They will be released after the sending has completed.
836 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
837 struct splice_desc
*sd
)
839 struct svc_rqst
*rqstp
= sd
->u
.data
;
840 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
841 struct page
*page
= buf
->page
;
845 ret
= buf
->ops
->confirm(pipe
, buf
);
851 if (rqstp
->rq_res
.page_len
== 0) {
856 rqstp
->rq_res
.page_base
= buf
->offset
;
857 rqstp
->rq_res
.page_len
= size
;
858 } else if (page
!= pp
[-1]) {
864 rqstp
->rq_res
.page_len
+= size
;
866 rqstp
->rq_res
.page_len
+= size
;
871 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
872 struct splice_desc
*sd
)
874 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
877 static inline int svc_msnfs(struct svc_fh
*ffhp
)
880 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
887 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
888 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
897 inode
= file
->f_path
.dentry
->d_inode
;
899 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
902 /* Get readahead parameters */
903 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
906 file
->f_ra
= ra
->p_ra
;
908 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
909 struct splice_desc sd
= {
916 rqstp
->rq_resused
= 1;
917 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
921 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
925 /* Write back readahead params */
927 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
928 spin_lock(&rab
->pb_lock
);
929 ra
->p_ra
= file
->f_ra
;
932 spin_unlock(&rab
->pb_lock
);
936 nfsdstats
.io_read
+= host_err
;
939 fsnotify_access(file
->f_path
.dentry
);
941 err
= nfserrno(host_err
);
946 static void kill_suid(struct dentry
*dentry
)
949 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
951 mutex_lock(&dentry
->d_inode
->i_mutex
);
952 notify_change(dentry
, &ia
);
953 mutex_unlock(&dentry
->d_inode
->i_mutex
);
957 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
958 loff_t offset
, struct kvec
*vec
, int vlen
,
959 unsigned long cnt
, int *stablep
)
961 struct svc_export
*exp
;
962 struct dentry
*dentry
;
967 int stable
= *stablep
;
972 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
973 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, cnt
)))
977 dentry
= file
->f_path
.dentry
;
978 inode
= dentry
->d_inode
;
979 exp
= fhp
->fh_export
;
982 * Request sync writes if
983 * - the sync export option has been set, or
984 * - the client requested O_SYNC behavior (NFSv3 feature).
985 * - The file system doesn't support fsync().
986 * When gathered writes have been configured for this volume,
987 * flushing the data to disk is handled separately below.
990 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
992 *stablep
= 2; /* FILE_SYNC */
997 if (stable
&& !EX_WGATHER(exp
))
998 file
->f_flags
|= O_SYNC
;
1000 /* Write the data. */
1001 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1002 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1004 if (host_err
>= 0) {
1005 nfsdstats
.io_write
+= cnt
;
1006 fsnotify_modify(file
->f_path
.dentry
);
1009 /* clear setuid/setgid flag after write */
1010 if (host_err
>= 0 && (inode
->i_mode
& (S_ISUID
| S_ISGID
)))
1013 if (host_err
>= 0 && stable
) {
1014 static ino_t last_ino
;
1015 static dev_t last_dev
;
1018 * Gathered writes: If another process is currently
1019 * writing to the file, there's a high chance
1020 * this is another nfsd (triggered by a bulk write
1021 * from a client's biod). Rather than syncing the
1022 * file with each write request, we sleep for 10 msec.
1024 * I don't know if this roughly approximates
1025 * C. Juszak's idea of gathered writes, but it's a
1026 * nice and simple solution (IMHO), and it seems to
1029 if (EX_WGATHER(exp
)) {
1030 if (atomic_read(&inode
->i_writecount
) > 1
1031 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1032 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1034 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1037 if (inode
->i_state
& I_DIRTY
) {
1038 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1039 host_err
=nfsd_sync(file
);
1042 wake_up(&inode
->i_wait
);
1045 last_ino
= inode
->i_ino
;
1046 last_dev
= inode
->i_sb
->s_dev
;
1049 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1053 err
= nfserrno(host_err
);
1059 * Read data from a file. count must contain the requested read count
1060 * on entry. On return, *count contains the number of bytes actually read.
1061 * N.B. After this call fhp needs an fh_put
1064 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1065 loff_t offset
, struct kvec
*vec
, int vlen
,
1066 unsigned long *count
)
1071 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1072 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1075 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1077 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1080 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1088 * Write data to a file.
1089 * The stable flag requests synchronous writes.
1090 * N.B. After this call fhp needs an fh_put
1093 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1094 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long cnt
,
1100 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1101 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1104 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1107 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1112 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1120 #ifdef CONFIG_NFSD_V3
1122 * Commit all pending writes to stable storage.
1123 * Strictly speaking, we could sync just the indicated file region here,
1124 * but there's currently no way we can ask the VFS to do so.
1126 * Unfortunately we cannot lock the file to make sure we return full WCC
1127 * data to the client, as locking happens lower down in the filesystem.
1130 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1131 loff_t offset
, unsigned long count
)
1136 if ((u64
)count
> ~(u64
)offset
)
1137 return nfserr_inval
;
1139 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1142 if (EX_ISSYNC(fhp
->fh_export
)) {
1143 if (file
->f_op
&& file
->f_op
->fsync
) {
1144 err
= nfserrno(nfsd_sync(file
));
1146 err
= nfserr_notsupp
;
1153 #endif /* CONFIG_NFSD_V3 */
1156 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1160 * Mode has already been set earlier in create:
1162 iap
->ia_valid
&= ~ATTR_MODE
;
1164 * Setting uid/gid works only for root. Irix appears to
1165 * send along the gid on create when it tries to implement
1166 * setgid directories via NFS:
1168 if (current
->fsuid
!= 0)
1169 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1171 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1176 * Create a file (regular, directory, device, fifo); UNIX sockets
1177 * not yet implemented.
1178 * If the response fh has been verified, the parent directory should
1179 * already be locked. Note that the parent directory is left locked.
1181 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1184 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1185 char *fname
, int flen
, struct iattr
*iap
,
1186 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1188 struct dentry
*dentry
, *dchild
= NULL
;
1198 if (isdotent(fname
, flen
))
1201 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1205 dentry
= fhp
->fh_dentry
;
1206 dirp
= dentry
->d_inode
;
1208 err
= nfserr_notdir
;
1209 if(!dirp
->i_op
|| !dirp
->i_op
->lookup
)
1212 * Check whether the response file handle has been verified yet.
1213 * If it has, the parent directory should already be locked.
1215 if (!resfhp
->fh_dentry
) {
1216 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1217 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1218 dchild
= lookup_one_len(fname
, dentry
, flen
);
1219 host_err
= PTR_ERR(dchild
);
1222 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1226 /* called from nfsd_proc_create */
1227 dchild
= dget(resfhp
->fh_dentry
);
1228 if (!fhp
->fh_locked
) {
1229 /* not actually possible */
1231 "nfsd_create: parent %s/%s not locked!\n",
1232 dentry
->d_parent
->d_name
.name
,
1233 dentry
->d_name
.name
);
1239 * Make sure the child dentry is still negative ...
1242 if (dchild
->d_inode
) {
1243 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1244 dentry
->d_name
.name
, dchild
->d_name
.name
);
1248 if (!(iap
->ia_valid
& ATTR_MODE
))
1250 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1253 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1254 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1259 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1264 * Get the dir op function pointer.
1269 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1272 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1278 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1282 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1286 if (EX_ISSYNC(fhp
->fh_export
)) {
1287 err
= nfserrno(nfsd_sync_dir(dentry
));
1288 write_inode_now(dchild
->d_inode
, 1);
1291 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1294 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1296 * Update the file handle to get the new inode info.
1299 err
= fh_update(resfhp
);
1301 if (dchild
&& !IS_ERR(dchild
))
1306 err
= nfserrno(host_err
);
1310 #ifdef CONFIG_NFSD_V3
1312 * NFSv3 version of nfsd_create
1315 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1316 char *fname
, int flen
, struct iattr
*iap
,
1317 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1318 int *truncp
, int *created
)
1320 struct dentry
*dentry
, *dchild
= NULL
;
1325 __u32 v_mtime
=0, v_atime
=0;
1331 if (isdotent(fname
, flen
))
1333 if (!(iap
->ia_valid
& ATTR_MODE
))
1335 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1339 dentry
= fhp
->fh_dentry
;
1340 dirp
= dentry
->d_inode
;
1342 /* Get all the sanity checks out of the way before
1343 * we lock the parent. */
1344 err
= nfserr_notdir
;
1345 if(!dirp
->i_op
|| !dirp
->i_op
->lookup
)
1347 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1350 * Compose the response file handle.
1352 dchild
= lookup_one_len(fname
, dentry
, flen
);
1353 host_err
= PTR_ERR(dchild
);
1357 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1361 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1362 /* solaris7 gets confused (bugid 4218508) if these have
1363 * the high bit set, so just clear the high bits. If this is
1364 * ever changed to use different attrs for storing the
1365 * verifier, then do_open_lookup() will also need to be fixed
1368 v_mtime
= verifier
[0]&0x7fffffff;
1369 v_atime
= verifier
[1]&0x7fffffff;
1372 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1375 if (dchild
->d_inode
) {
1378 switch (createmode
) {
1379 case NFS3_CREATE_UNCHECKED
:
1380 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1383 /* in nfsv4, we need to treat this case a little
1384 * differently. we don't want to truncate the
1385 * file now; this would be wrong if the OPEN
1386 * fails for some other reason. furthermore,
1387 * if the size is nonzero, we should ignore it
1388 * according to spec!
1390 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1393 iap
->ia_valid
&= ATTR_SIZE
;
1397 case NFS3_CREATE_EXCLUSIVE
:
1398 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1399 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1400 && dchild
->d_inode
->i_size
== 0 )
1403 case NFS3_CREATE_GUARDED
:
1406 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1410 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1412 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1418 if (EX_ISSYNC(fhp
->fh_export
)) {
1419 err
= nfserrno(nfsd_sync_dir(dentry
));
1420 /* setattr will sync the child (or not) */
1423 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1424 /* Cram the verifier into atime/mtime */
1425 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1426 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1427 /* XXX someone who knows this better please fix it for nsec */
1428 iap
->ia_mtime
.tv_sec
= v_mtime
;
1429 iap
->ia_atime
.tv_sec
= v_atime
;
1430 iap
->ia_mtime
.tv_nsec
= 0;
1431 iap
->ia_atime
.tv_nsec
= 0;
1435 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1439 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1441 * Update the filehandle to get the new inode info.
1444 err
= fh_update(resfhp
);
1448 if (dchild
&& !IS_ERR(dchild
))
1453 err
= nfserrno(host_err
);
1456 #endif /* CONFIG_NFSD_V3 */
1459 * Read a symlink. On entry, *lenp must contain the maximum path length that
1460 * fits into the buffer. On return, it contains the true length.
1461 * N.B. After this call fhp needs an fh_put
1464 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1466 struct dentry
*dentry
;
1467 struct inode
*inode
;
1472 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1476 dentry
= fhp
->fh_dentry
;
1477 inode
= dentry
->d_inode
;
1480 if (!inode
->i_op
|| !inode
->i_op
->readlink
)
1483 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1484 /* N.B. Why does this call need a get_fs()??
1485 * Remove the set_fs and watch the fireworks:-) --okir
1488 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1489 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1500 err
= nfserrno(host_err
);
1505 * Create a symlink and look up its inode
1506 * N.B. After this call _both_ fhp and resfhp need an fh_put
1509 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1510 char *fname
, int flen
,
1511 char *path
, int plen
,
1512 struct svc_fh
*resfhp
,
1515 struct dentry
*dentry
, *dnew
;
1523 if (isdotent(fname
, flen
))
1526 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1530 dentry
= fhp
->fh_dentry
;
1531 dnew
= lookup_one_len(fname
, dentry
, flen
);
1532 host_err
= PTR_ERR(dnew
);
1536 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1540 if (unlikely(path
[plen
] != 0)) {
1541 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1542 if (path_alloced
== NULL
)
1545 strncpy(path_alloced
, path
, plen
);
1546 path_alloced
[plen
] = 0;
1547 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1548 kfree(path_alloced
);
1551 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1554 if (EX_ISSYNC(fhp
->fh_export
))
1555 host_err
= nfsd_sync_dir(dentry
);
1557 err
= nfserrno(host_err
);
1560 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1562 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1564 if (err
==0) err
= cerr
;
1569 err
= nfserrno(host_err
);
1575 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1578 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1579 char *name
, int len
, struct svc_fh
*tfhp
)
1581 struct dentry
*ddir
, *dnew
, *dold
;
1582 struct inode
*dirp
, *dest
;
1586 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1589 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1597 if (isdotent(name
, len
))
1600 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1601 ddir
= ffhp
->fh_dentry
;
1602 dirp
= ddir
->d_inode
;
1604 dnew
= lookup_one_len(name
, ddir
, len
);
1605 host_err
= PTR_ERR(dnew
);
1609 dold
= tfhp
->fh_dentry
;
1610 dest
= dold
->d_inode
;
1612 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1614 err
= nfserrno(host_err
);
1617 host_err
= vfs_link(dold
, dirp
, dnew
);
1619 if (EX_ISSYNC(ffhp
->fh_export
)) {
1620 err
= nfserrno(nfsd_sync_dir(ddir
));
1621 write_inode_now(dest
, 1);
1625 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1628 err
= nfserrno(host_err
);
1630 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1639 err
= nfserrno(host_err
);
1645 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1648 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1649 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1651 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1652 struct inode
*fdir
, *tdir
;
1656 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1659 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1663 fdentry
= ffhp
->fh_dentry
;
1664 fdir
= fdentry
->d_inode
;
1666 tdentry
= tfhp
->fh_dentry
;
1667 tdir
= tdentry
->d_inode
;
1669 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1670 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1674 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1677 /* cannot use fh_lock as we need deadlock protective ordering
1678 * so do it by hand */
1679 trap
= lock_rename(tdentry
, fdentry
);
1680 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1684 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1685 host_err
= PTR_ERR(odentry
);
1686 if (IS_ERR(odentry
))
1690 if (!odentry
->d_inode
)
1693 if (odentry
== trap
)
1696 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1697 host_err
= PTR_ERR(ndentry
);
1698 if (IS_ERR(ndentry
))
1700 host_err
= -ENOTEMPTY
;
1701 if (ndentry
== trap
)
1704 if (svc_msnfs(ffhp
) &&
1705 ((atomic_read(&odentry
->d_count
) > 1)
1706 || (atomic_read(&ndentry
->d_count
) > 1))) {
1712 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1714 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1718 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1719 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1720 host_err
= nfsd_sync_dir(tdentry
);
1722 host_err
= nfsd_sync_dir(fdentry
);
1725 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1732 err
= nfserrno(host_err
);
1734 /* we cannot reply on fh_unlock on the two filehandles,
1735 * as that would do the wrong thing if the two directories
1736 * were the same, so again we do it by hand
1738 fill_post_wcc(ffhp
);
1739 fill_post_wcc(tfhp
);
1740 unlock_rename(tdentry
, fdentry
);
1741 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1748 * Unlink a file or directory
1749 * N.B. After this call fhp needs an fh_put
1752 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1753 char *fname
, int flen
)
1755 struct dentry
*dentry
, *rdentry
;
1761 if (!flen
|| isdotent(fname
, flen
))
1763 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1767 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1768 dentry
= fhp
->fh_dentry
;
1769 dirp
= dentry
->d_inode
;
1771 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1772 host_err
= PTR_ERR(rdentry
);
1773 if (IS_ERR(rdentry
))
1776 if (!rdentry
->d_inode
) {
1783 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1785 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1789 if (type
!= S_IFDIR
) { /* It's UNLINK */
1791 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1792 (atomic_read(&rdentry
->d_count
) > 1)) {
1796 host_err
= vfs_unlink(dirp
, rdentry
);
1797 } else { /* It's RMDIR */
1798 host_err
= vfs_rmdir(dirp
, rdentry
);
1805 if (EX_ISSYNC(fhp
->fh_export
))
1806 host_err
= nfsd_sync_dir(dentry
);
1809 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1811 err
= nfserrno(host_err
);
1817 * Read entries from a directory.
1818 * The NFSv3/4 verifier we ignore for now.
1821 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1822 struct readdir_cd
*cdp
, filldir_t func
)
1827 loff_t offset
= *offsetp
;
1829 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1833 offset
= vfs_llseek(file
, offset
, 0);
1835 err
= nfserrno((int)offset
);
1840 * Read the directory entries. This silly loop is necessary because
1841 * readdir() is not guaranteed to fill up the entire buffer, but
1842 * may choose to do less.
1846 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1847 host_err
= vfs_readdir(file
, func
, cdp
);
1848 } while (host_err
>=0 && cdp
->err
== nfs_ok
);
1850 err
= nfserrno(host_err
);
1853 *offsetp
= vfs_llseek(file
, 0, 1);
1855 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1856 err
= nfs_ok
; /* can still be found in ->err */
1864 * Get file system stats
1865 * N.B. After this call fhp needs an fh_put
1868 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
1870 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
1871 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
1876 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
1878 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
1882 * Check for a user's access permissions to this inode.
1885 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
1886 struct dentry
*dentry
, int acc
)
1888 struct inode
*inode
= dentry
->d_inode
;
1891 if (acc
== NFSD_MAY_NOP
)
1894 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1896 (acc
& NFSD_MAY_READ
)? " read" : "",
1897 (acc
& NFSD_MAY_WRITE
)? " write" : "",
1898 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
1899 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
1900 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
1901 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
1902 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
1904 IS_IMMUTABLE(inode
)? " immut" : "",
1905 IS_APPEND(inode
)? " append" : "",
1906 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
1907 dprintk(" owner %d/%d user %d/%d\n",
1908 inode
->i_uid
, inode
->i_gid
, current
->fsuid
, current
->fsgid
);
1911 /* Normally we reject any write/sattr etc access on a read-only file
1912 * system. But if it is IRIX doing check on write-access for a
1913 * device special file, we ignore rofs.
1915 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
1916 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
1917 if (exp_rdonly(rqstp
, exp
) ||
1918 __mnt_is_readonly(exp
->ex_path
.mnt
))
1920 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
1923 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
1926 if (acc
& NFSD_MAY_LOCK
) {
1927 /* If we cannot rely on authentication in NLM requests,
1928 * just allow locks, otherwise require read permission, or
1931 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
1934 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
1937 * The file owner always gets access permission for accesses that
1938 * would normally be checked at open time. This is to make
1939 * file access work even when the client has done a fchmod(fd, 0).
1941 * However, `cp foo bar' should fail nevertheless when bar is
1942 * readonly. A sensible way to do this might be to reject all
1943 * attempts to truncate a read-only file, because a creat() call
1944 * always implies file truncation.
1945 * ... but this isn't really fair. A process may reasonably call
1946 * ftruncate on an open file descriptor on a file with perm 000.
1947 * We must trust the client to do permission checking - using "ACCESS"
1950 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
1951 inode
->i_uid
== current
->fsuid
)
1954 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
1955 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
1957 /* Allow read access to binaries even when mode 111 */
1958 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
1959 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
1960 err
= inode_permission(inode
, MAY_EXEC
);
1962 return err
? nfserrno(err
) : 0;
1966 nfsd_racache_shutdown(void)
1968 struct raparms
*raparm
, *last_raparm
;
1971 dprintk("nfsd: freeing readahead buffers.\n");
1973 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
1974 raparm
= raparm_hash
[i
].pb_head
;
1976 last_raparm
= raparm
;
1977 raparm
= raparm
->p_next
;
1980 raparm_hash
[i
].pb_head
= NULL
;
1984 * Initialize readahead param cache
1987 nfsd_racache_init(int cache_size
)
1992 struct raparms
**raparm
= NULL
;
1995 if (raparm_hash
[0].pb_head
)
1997 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2000 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2002 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2004 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2005 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2007 raparm
= &raparm_hash
[i
].pb_head
;
2008 for (j
= 0; j
< nperbucket
; j
++) {
2009 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2012 raparm
= &(*raparm
)->p_next
;
2017 nfsdstats
.ra_size
= cache_size
;
2021 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2022 nfsd_racache_shutdown();
2026 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2028 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2030 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2034 struct posix_acl
*acl
;
2036 if (!IS_POSIXACL(inode
))
2037 return ERR_PTR(-EOPNOTSUPP
);
2040 case ACL_TYPE_ACCESS
:
2041 name
= POSIX_ACL_XATTR_ACCESS
;
2043 case ACL_TYPE_DEFAULT
:
2044 name
= POSIX_ACL_XATTR_DEFAULT
;
2047 return ERR_PTR(-EOPNOTSUPP
);
2050 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2052 return ERR_PTR(size
);
2054 acl
= posix_acl_from_xattr(value
, size
);
2060 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2062 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2068 if (!IS_POSIXACL(inode
) || !inode
->i_op
||
2069 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2072 case ACL_TYPE_ACCESS
:
2073 name
= POSIX_ACL_XATTR_ACCESS
;
2075 case ACL_TYPE_DEFAULT
:
2076 name
= POSIX_ACL_XATTR_DEFAULT
;
2082 if (acl
&& acl
->a_count
) {
2083 size
= posix_acl_xattr_size(acl
->a_count
);
2084 value
= kmalloc(size
, GFP_KERNEL
);
2087 error
= posix_acl_to_xattr(acl
, value
, size
);
2094 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2098 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2100 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2103 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2104 if (error
== -ENODATA
)
2108 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2114 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */