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>
58 #include <linux/ima.h>
60 #include <asm/uaccess.h>
62 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
66 * This is a cache of readahead params that help us choose the proper
67 * readahead strategy. Initially, we set all readahead parameters to 0
68 * and let the VFS handle things.
69 * If you increase the number of cached files very much, you'll need to
70 * add a hash table here.
73 struct raparms
*p_next
;
78 struct file_ra_state p_ra
;
79 unsigned int p_hindex
;
82 struct raparm_hbucket
{
83 struct raparms
*pb_head
;
85 } ____cacheline_aligned_in_smp
;
87 #define RAPARM_HASH_BITS 4
88 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
89 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
90 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
93 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
95 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
96 * or nfs_ok having possibly changed *dpp and *expp
99 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
100 struct svc_export
**expp
)
102 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
103 struct dentry
*dentry
= *dpp
;
104 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
105 .dentry
= dget(dentry
)};
108 while (follow_down(&path
.mnt
, &path
.dentry
) &&
109 d_mountpoint(path
.dentry
))
112 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
114 if (PTR_ERR(exp2
) != -ENOENT
)
119 if ((exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
120 /* successfully crossed mount point */
122 * This is subtle: path.dentry is *not* on path.mnt
123 * at this point. The only reason we are safe is that
124 * original mnt is pinned down by exp, so we should
125 * put path *before* putting exp
128 path
.dentry
= dentry
;
139 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
140 const char *name
, unsigned int len
,
141 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
143 struct svc_export
*exp
;
144 struct dentry
*dparent
;
145 struct dentry
*dentry
;
149 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
151 /* Obtain dentry and export. */
152 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
156 dparent
= fhp
->fh_dentry
;
157 exp
= fhp
->fh_export
;
160 /* Lookup the name, but don't follow links */
161 if (isdotent(name
, len
)) {
163 dentry
= dget(dparent
);
164 else if (dparent
!= exp
->ex_path
.dentry
)
165 dentry
= dget_parent(dparent
);
166 else if (!EX_NOHIDE(exp
))
167 dentry
= dget(dparent
); /* .. == . just like at / */
169 /* checking mountpoint crossing is very different when stepping up */
170 struct svc_export
*exp2
= NULL
;
172 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
173 .dentry
= dget(dparent
)};
175 while (path
.dentry
== path
.mnt
->mnt_root
&&
176 follow_up(&path
.mnt
, &path
.dentry
))
178 dp
= dget_parent(path
.dentry
);
182 exp2
= rqst_exp_parent(rqstp
, &path
);
183 if (PTR_ERR(exp2
) == -ENOENT
) {
184 dentry
= dget(dparent
);
185 } else if (IS_ERR(exp2
)) {
186 host_err
= PTR_ERR(exp2
);
190 dentry
= dget(path
.dentry
);
198 dentry
= lookup_one_len(name
, dparent
, len
);
199 host_err
= PTR_ERR(dentry
);
203 * check if we have crossed a mount point ...
205 if (d_mountpoint(dentry
)) {
206 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
212 *dentry_ret
= dentry
;
218 return nfserrno(host_err
);
222 * Look up one component of a pathname.
223 * N.B. After this call _both_ fhp and resfh need an fh_put
225 * If the lookup would cross a mountpoint, and the mounted filesystem
226 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
227 * accepted as it stands and the mounted directory is
228 * returned. Otherwise the covered directory is returned.
229 * NOTE: this mountpoint crossing is not supported properly by all
230 * clients and is explicitly disallowed for NFSv3
231 * NeilBrown <neilb@cse.unsw.edu.au>
234 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
235 unsigned int len
, struct svc_fh
*resfh
)
237 struct svc_export
*exp
;
238 struct dentry
*dentry
;
241 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
244 err
= check_nfsd_access(exp
, rqstp
);
248 * Note: we compose the file handle now, but as the
249 * dentry may be negative, it may need to be updated.
251 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
252 if (!err
&& !dentry
->d_inode
)
262 * Set various file attributes.
263 * N.B. After this call fhp needs an fh_put
266 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
267 int check_guard
, time_t guardtime
)
269 struct dentry
*dentry
;
271 int accmode
= NFSD_MAY_SATTR
;
277 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
278 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
279 if (iap
->ia_valid
& ATTR_SIZE
)
283 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
287 dentry
= fhp
->fh_dentry
;
288 inode
= dentry
->d_inode
;
290 /* Ignore any mode updates on symlinks */
291 if (S_ISLNK(inode
->i_mode
))
292 iap
->ia_valid
&= ~ATTR_MODE
;
298 * NFSv2 does not differentiate between "set-[ac]time-to-now"
299 * which only requires access, and "set-[ac]time-to-X" which
300 * requires ownership.
301 * So if it looks like it might be "set both to the same time which
302 * is close to now", and if inode_change_ok fails, then we
303 * convert to "set to now" instead of "set to explicit time"
305 * We only call inode_change_ok as the last test as technically
306 * it is not an interface that we should be using. It is only
307 * valid if the filesystem does not define it's own i_op->setattr.
309 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
310 #define MAX_TOUCH_TIME_ERROR (30*60)
311 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
312 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
316 * Now just make sure time is in the right ballpark.
317 * Solaris, at least, doesn't seem to care what the time
318 * request is. We require it be within 30 minutes of now.
320 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
323 if (delta
< MAX_TOUCH_TIME_ERROR
&&
324 inode_change_ok(inode
, iap
) != 0) {
326 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
327 * This will cause notify_change to set these times
330 iap
->ia_valid
&= ~BOTH_TIME_SET
;
335 * The size case is special.
336 * It changes the file as well as the attributes.
338 if (iap
->ia_valid
& ATTR_SIZE
) {
339 if (iap
->ia_size
< inode
->i_size
) {
340 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
341 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
347 * If we are changing the size of the file, then
348 * we need to break all leases.
350 host_err
= break_lease(inode
, FMODE_WRITE
| O_NONBLOCK
);
351 if (host_err
== -EWOULDBLOCK
)
352 host_err
= -ETIMEDOUT
;
353 if (host_err
) /* ENOMEM or EWOULDBLOCK */
356 host_err
= get_write_access(inode
);
361 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
363 put_write_access(inode
);
369 /* sanitize the mode change */
370 if (iap
->ia_valid
& ATTR_MODE
) {
371 iap
->ia_mode
&= S_IALLUGO
;
372 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
375 /* Revoke setuid/setgid on chown */
376 if (!S_ISDIR(inode
->i_mode
) &&
377 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
378 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
379 iap
->ia_valid
|= ATTR_KILL_PRIV
;
380 if (iap
->ia_valid
& ATTR_MODE
) {
381 /* we're setting mode too, just clear the s*id bits */
382 iap
->ia_mode
&= ~S_ISUID
;
383 if (iap
->ia_mode
& S_IXGRP
)
384 iap
->ia_mode
&= ~S_ISGID
;
386 /* set ATTR_KILL_* bits and let VFS handle it */
387 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
391 /* Change the attributes. */
393 iap
->ia_valid
|= ATTR_CTIME
;
395 err
= nfserr_notsync
;
396 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
398 host_err
= notify_change(dentry
, iap
);
399 err
= nfserrno(host_err
);
403 put_write_access(inode
);
405 if (EX_ISSYNC(fhp
->fh_export
))
406 write_inode_now(inode
, 1);
411 err
= nfserrno(host_err
);
415 #if defined(CONFIG_NFSD_V2_ACL) || \
416 defined(CONFIG_NFSD_V3_ACL) || \
417 defined(CONFIG_NFSD_V4)
418 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
423 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
427 *buf
= kmalloc(buflen
, GFP_KERNEL
);
431 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
438 #if defined(CONFIG_NFSD_V4)
440 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
447 buflen
= posix_acl_xattr_size(pacl
->a_count
);
448 buf
= kmalloc(buflen
, GFP_KERNEL
);
453 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
459 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
466 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
467 struct nfs4_acl
*acl
)
471 struct dentry
*dentry
;
473 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
474 unsigned int flags
= 0;
477 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
481 dentry
= fhp
->fh_dentry
;
482 inode
= dentry
->d_inode
;
483 if (S_ISDIR(inode
->i_mode
))
484 flags
= NFS4_ACL_DIR
;
486 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
487 if (host_error
== -EINVAL
) {
488 return nfserr_attrnotsupp
;
489 } else if (host_error
< 0)
492 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
496 if (S_ISDIR(inode
->i_mode
))
497 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
500 posix_acl_release(pacl
);
501 posix_acl_release(dpacl
);
503 if (host_error
== -EOPNOTSUPP
)
504 return nfserr_attrnotsupp
;
506 return nfserrno(host_error
);
509 static struct posix_acl
*
510 _get_posix_acl(struct dentry
*dentry
, char *key
)
513 struct posix_acl
*pacl
= NULL
;
516 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
520 return ERR_PTR(buflen
);
522 pacl
= posix_acl_from_xattr(buf
, buflen
);
528 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
530 struct inode
*inode
= dentry
->d_inode
;
532 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
533 unsigned int flags
= 0;
535 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
536 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
537 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
539 error
= PTR_ERR(pacl
);
544 if (S_ISDIR(inode
->i_mode
)) {
545 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
546 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
548 else if (IS_ERR(dpacl
)) {
549 error
= PTR_ERR(dpacl
);
553 flags
= NFS4_ACL_DIR
;
556 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
558 error
= PTR_ERR(*acl
);
562 posix_acl_release(pacl
);
563 posix_acl_release(dpacl
);
567 #endif /* defined(CONFIG_NFS_V4) */
569 #ifdef CONFIG_NFSD_V3
571 * Check server access rights to a file system object
577 static struct accessmap nfs3_regaccess
[] = {
578 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
579 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
580 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
581 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
586 static struct accessmap nfs3_diraccess
[] = {
587 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
588 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
589 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
590 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
591 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
596 static struct accessmap nfs3_anyaccess
[] = {
597 /* Some clients - Solaris 2.6 at least, make an access call
598 * to the server to check for access for things like /dev/null
599 * (which really, the server doesn't care about). So
600 * We provide simple access checking for them, looking
601 * mainly at mode bits, and we make sure to ignore read-only
604 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
605 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
606 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
607 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
613 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
615 struct accessmap
*map
;
616 struct svc_export
*export
;
617 struct dentry
*dentry
;
618 u32 query
, result
= 0, sresult
= 0;
621 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
625 export
= fhp
->fh_export
;
626 dentry
= fhp
->fh_dentry
;
628 if (S_ISREG(dentry
->d_inode
->i_mode
))
629 map
= nfs3_regaccess
;
630 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
631 map
= nfs3_diraccess
;
633 map
= nfs3_anyaccess
;
637 for (; map
->access
; map
++) {
638 if (map
->access
& query
) {
641 sresult
|= map
->access
;
643 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
646 result
|= map
->access
;
649 /* the following error codes just mean the access was not allowed,
650 * rather than an error occurred */
654 /* simply don't "or" in the access bit. */
664 *supported
= sresult
;
669 #endif /* CONFIG_NFSD_V3 */
674 * Open an existing file or directory.
675 * The access argument indicates the type of open (read/write/lock)
676 * N.B. After this call fhp needs an fh_put
679 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
680 int access
, struct file
**filp
)
682 const struct cred
*cred
= current_cred();
683 struct dentry
*dentry
;
685 int flags
= O_RDONLY
|O_LARGEFILE
;
690 * If we get here, then the client has already done an "open",
691 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
692 * in case a chmod has now revoked permission.
694 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
698 dentry
= fhp
->fh_dentry
;
699 inode
= dentry
->d_inode
;
701 /* Disallow write access to files with the append-only bit set
702 * or any access when mandatory locking enabled
705 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
708 * We must ignore files (but only files) which might have mandatory
709 * locks on them because there is no way to know if the accesser has
712 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
719 * Check to see if there are any leases on this file.
720 * This may block while leases are broken.
722 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
723 if (host_err
== -EWOULDBLOCK
)
724 host_err
= -ETIMEDOUT
;
725 if (host_err
) /* NOMEM or WOULDBLOCK */
728 if (access
& NFSD_MAY_WRITE
) {
729 if (access
& NFSD_MAY_READ
)
730 flags
= O_RDWR
|O_LARGEFILE
;
732 flags
= O_WRONLY
|O_LARGEFILE
;
736 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
739 host_err
= PTR_ERR(*filp
);
741 ima_counts_get(*filp
);
743 err
= nfserrno(host_err
);
752 nfsd_close(struct file
*filp
)
759 * As this calls fsync (not fdatasync) there is no need for a write_inode
762 static inline int nfsd_dosync(struct file
*filp
, struct dentry
*dp
,
763 const struct file_operations
*fop
)
765 struct inode
*inode
= dp
->d_inode
;
766 int (*fsync
) (struct file
*, struct dentry
*, int);
769 err
= filemap_fdatawrite(inode
->i_mapping
);
770 if (err
== 0 && fop
&& (fsync
= fop
->fsync
))
771 err
= fsync(filp
, dp
, 0);
773 err
= filemap_fdatawait(inode
->i_mapping
);
779 nfsd_sync(struct file
*filp
)
782 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
783 dprintk("nfsd: sync file %s\n", filp
->f_path
.dentry
->d_name
.name
);
784 mutex_lock(&inode
->i_mutex
);
785 err
=nfsd_dosync(filp
, filp
->f_path
.dentry
, filp
->f_op
);
786 mutex_unlock(&inode
->i_mutex
);
792 nfsd_sync_dir(struct dentry
*dp
)
794 return nfsd_dosync(NULL
, dp
, dp
->d_inode
->i_fop
);
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 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
972 loff_t offset
, struct kvec
*vec
, int vlen
,
973 unsigned long *cnt
, int *stablep
)
975 struct svc_export
*exp
;
976 struct dentry
*dentry
;
981 int stable
= *stablep
;
986 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
987 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
991 dentry
= file
->f_path
.dentry
;
992 inode
= dentry
->d_inode
;
993 exp
= fhp
->fh_export
;
996 * Request sync writes if
997 * - the sync export option has been set, or
998 * - the client requested O_SYNC behavior (NFSv3 feature).
999 * - The file system doesn't support fsync().
1000 * When gathered writes have been configured for this volume,
1001 * flushing the data to disk is handled separately below.
1004 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1006 *stablep
= 2; /* FILE_SYNC */
1009 if (!EX_ISSYNC(exp
))
1011 if (stable
&& !EX_WGATHER(exp
)) {
1012 spin_lock(&file
->f_lock
);
1013 file
->f_flags
|= O_SYNC
;
1014 spin_unlock(&file
->f_lock
);
1017 /* Write the data. */
1018 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1019 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1021 if (host_err
>= 0) {
1023 nfsdstats
.io_write
+= host_err
;
1024 fsnotify_modify(file
->f_path
.dentry
);
1027 /* clear setuid/setgid flag after write */
1028 if (host_err
>= 0 && (inode
->i_mode
& (S_ISUID
| S_ISGID
)))
1031 if (host_err
>= 0 && stable
) {
1032 static ino_t last_ino
;
1033 static dev_t last_dev
;
1036 * Gathered writes: If another process is currently
1037 * writing to the file, there's a high chance
1038 * this is another nfsd (triggered by a bulk write
1039 * from a client's biod). Rather than syncing the
1040 * file with each write request, we sleep for 10 msec.
1042 * I don't know if this roughly approximates
1043 * C. Juszak's idea of gathered writes, but it's a
1044 * nice and simple solution (IMHO), and it seems to
1047 if (EX_WGATHER(exp
)) {
1048 if (atomic_read(&inode
->i_writecount
) > 1
1049 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1050 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1052 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1055 if (inode
->i_state
& I_DIRTY
) {
1056 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1057 host_err
=nfsd_sync(file
);
1060 wake_up(&inode
->i_wait
);
1063 last_ino
= inode
->i_ino
;
1064 last_dev
= inode
->i_sb
->s_dev
;
1067 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1071 err
= nfserrno(host_err
);
1077 * Read data from a file. count must contain the requested read count
1078 * on entry. On return, *count contains the number of bytes actually read.
1079 * N.B. After this call fhp needs an fh_put
1082 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1083 loff_t offset
, struct kvec
*vec
, int vlen
,
1084 unsigned long *count
)
1089 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1090 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1093 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1095 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1098 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1106 * Write data to a file.
1107 * The stable flag requests synchronous writes.
1108 * N.B. After this call fhp needs an fh_put
1111 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1112 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1118 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1119 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1122 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1125 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1130 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1138 #ifdef CONFIG_NFSD_V3
1140 * Commit all pending writes to stable storage.
1141 * Strictly speaking, we could sync just the indicated file region here,
1142 * but there's currently no way we can ask the VFS to do so.
1144 * Unfortunately we cannot lock the file to make sure we return full WCC
1145 * data to the client, as locking happens lower down in the filesystem.
1148 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1149 loff_t offset
, unsigned long count
)
1154 if ((u64
)count
> ~(u64
)offset
)
1155 return nfserr_inval
;
1157 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1160 if (EX_ISSYNC(fhp
->fh_export
)) {
1161 if (file
->f_op
&& file
->f_op
->fsync
) {
1162 err
= nfserrno(nfsd_sync(file
));
1164 err
= nfserr_notsupp
;
1171 #endif /* CONFIG_NFSD_V3 */
1174 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1178 * Mode has already been set earlier in create:
1180 iap
->ia_valid
&= ~ATTR_MODE
;
1182 * Setting uid/gid works only for root. Irix appears to
1183 * send along the gid on create when it tries to implement
1184 * setgid directories via NFS:
1186 if (current_fsuid() != 0)
1187 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1189 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1193 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1194 * setting size to 0 may fail for some specific file systems by the permission
1195 * checking which requires WRITE permission but the mode is 000.
1196 * we ignore the resizing(to 0) on the just new created file, since the size is
1197 * 0 after file created.
1199 * call this only after vfs_create() is called.
1202 nfsd_check_ignore_resizing(struct iattr
*iap
)
1204 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1205 iap
->ia_valid
&= ~ATTR_SIZE
;
1209 * Create a file (regular, directory, device, fifo); UNIX sockets
1210 * not yet implemented.
1211 * If the response fh has been verified, the parent directory should
1212 * already be locked. Note that the parent directory is left locked.
1214 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1217 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1218 char *fname
, int flen
, struct iattr
*iap
,
1219 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1221 struct dentry
*dentry
, *dchild
= NULL
;
1231 if (isdotent(fname
, flen
))
1234 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1238 dentry
= fhp
->fh_dentry
;
1239 dirp
= dentry
->d_inode
;
1241 err
= nfserr_notdir
;
1242 if (!dirp
->i_op
->lookup
)
1245 * Check whether the response file handle has been verified yet.
1246 * If it has, the parent directory should already be locked.
1248 if (!resfhp
->fh_dentry
) {
1249 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1250 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1251 dchild
= lookup_one_len(fname
, dentry
, flen
);
1252 host_err
= PTR_ERR(dchild
);
1255 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1259 /* called from nfsd_proc_create */
1260 dchild
= dget(resfhp
->fh_dentry
);
1261 if (!fhp
->fh_locked
) {
1262 /* not actually possible */
1264 "nfsd_create: parent %s/%s not locked!\n",
1265 dentry
->d_parent
->d_name
.name
,
1266 dentry
->d_name
.name
);
1272 * Make sure the child dentry is still negative ...
1275 if (dchild
->d_inode
) {
1276 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1277 dentry
->d_name
.name
, dchild
->d_name
.name
);
1281 if (!(iap
->ia_valid
& ATTR_MODE
))
1283 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1286 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1287 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1292 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1297 * Get the dir op function pointer.
1302 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1304 nfsd_check_ignore_resizing(iap
);
1307 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1313 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1317 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1321 if (EX_ISSYNC(fhp
->fh_export
)) {
1322 err
= nfserrno(nfsd_sync_dir(dentry
));
1323 write_inode_now(dchild
->d_inode
, 1);
1326 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1329 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1331 * Update the file handle to get the new inode info.
1334 err
= fh_update(resfhp
);
1336 if (dchild
&& !IS_ERR(dchild
))
1341 err
= nfserrno(host_err
);
1345 #ifdef CONFIG_NFSD_V3
1347 * NFSv3 version of nfsd_create
1350 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1351 char *fname
, int flen
, struct iattr
*iap
,
1352 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1353 int *truncp
, int *created
)
1355 struct dentry
*dentry
, *dchild
= NULL
;
1360 __u32 v_mtime
=0, v_atime
=0;
1366 if (isdotent(fname
, flen
))
1368 if (!(iap
->ia_valid
& ATTR_MODE
))
1370 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1374 dentry
= fhp
->fh_dentry
;
1375 dirp
= dentry
->d_inode
;
1377 /* Get all the sanity checks out of the way before
1378 * we lock the parent. */
1379 err
= nfserr_notdir
;
1380 if (!dirp
->i_op
->lookup
)
1382 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1385 * Compose the response file handle.
1387 dchild
= lookup_one_len(fname
, dentry
, flen
);
1388 host_err
= PTR_ERR(dchild
);
1392 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1396 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1397 /* solaris7 gets confused (bugid 4218508) if these have
1398 * the high bit set, so just clear the high bits. If this is
1399 * ever changed to use different attrs for storing the
1400 * verifier, then do_open_lookup() will also need to be fixed
1403 v_mtime
= verifier
[0]&0x7fffffff;
1404 v_atime
= verifier
[1]&0x7fffffff;
1407 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1410 if (dchild
->d_inode
) {
1413 switch (createmode
) {
1414 case NFS3_CREATE_UNCHECKED
:
1415 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1418 /* in nfsv4, we need to treat this case a little
1419 * differently. we don't want to truncate the
1420 * file now; this would be wrong if the OPEN
1421 * fails for some other reason. furthermore,
1422 * if the size is nonzero, we should ignore it
1423 * according to spec!
1425 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1428 iap
->ia_valid
&= ATTR_SIZE
;
1432 case NFS3_CREATE_EXCLUSIVE
:
1433 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1434 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1435 && dchild
->d_inode
->i_size
== 0 )
1438 case NFS3_CREATE_GUARDED
:
1441 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1445 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1447 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1453 if (EX_ISSYNC(fhp
->fh_export
)) {
1454 err
= nfserrno(nfsd_sync_dir(dentry
));
1455 /* setattr will sync the child (or not) */
1458 nfsd_check_ignore_resizing(iap
);
1460 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1461 /* Cram the verifier into atime/mtime */
1462 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1463 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1464 /* XXX someone who knows this better please fix it for nsec */
1465 iap
->ia_mtime
.tv_sec
= v_mtime
;
1466 iap
->ia_atime
.tv_sec
= v_atime
;
1467 iap
->ia_mtime
.tv_nsec
= 0;
1468 iap
->ia_atime
.tv_nsec
= 0;
1472 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1476 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1478 * Update the filehandle to get the new inode info.
1481 err
= fh_update(resfhp
);
1485 if (dchild
&& !IS_ERR(dchild
))
1490 err
= nfserrno(host_err
);
1493 #endif /* CONFIG_NFSD_V3 */
1496 * Read a symlink. On entry, *lenp must contain the maximum path length that
1497 * fits into the buffer. On return, it contains the true length.
1498 * N.B. After this call fhp needs an fh_put
1501 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1503 struct dentry
*dentry
;
1504 struct inode
*inode
;
1509 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1513 dentry
= fhp
->fh_dentry
;
1514 inode
= dentry
->d_inode
;
1517 if (!inode
->i_op
->readlink
)
1520 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1521 /* N.B. Why does this call need a get_fs()??
1522 * Remove the set_fs and watch the fireworks:-) --okir
1525 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1526 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1537 err
= nfserrno(host_err
);
1542 * Create a symlink and look up its inode
1543 * N.B. After this call _both_ fhp and resfhp need an fh_put
1546 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1547 char *fname
, int flen
,
1548 char *path
, int plen
,
1549 struct svc_fh
*resfhp
,
1552 struct dentry
*dentry
, *dnew
;
1560 if (isdotent(fname
, flen
))
1563 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1567 dentry
= fhp
->fh_dentry
;
1568 dnew
= lookup_one_len(fname
, dentry
, flen
);
1569 host_err
= PTR_ERR(dnew
);
1573 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1577 if (unlikely(path
[plen
] != 0)) {
1578 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1579 if (path_alloced
== NULL
)
1582 strncpy(path_alloced
, path
, plen
);
1583 path_alloced
[plen
] = 0;
1584 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1585 kfree(path_alloced
);
1588 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1591 if (EX_ISSYNC(fhp
->fh_export
))
1592 host_err
= nfsd_sync_dir(dentry
);
1594 err
= nfserrno(host_err
);
1597 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1599 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1601 if (err
==0) err
= cerr
;
1606 err
= nfserrno(host_err
);
1612 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1615 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1616 char *name
, int len
, struct svc_fh
*tfhp
)
1618 struct dentry
*ddir
, *dnew
, *dold
;
1619 struct inode
*dirp
, *dest
;
1623 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1626 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1634 if (isdotent(name
, len
))
1637 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1638 ddir
= ffhp
->fh_dentry
;
1639 dirp
= ddir
->d_inode
;
1641 dnew
= lookup_one_len(name
, ddir
, len
);
1642 host_err
= PTR_ERR(dnew
);
1646 dold
= tfhp
->fh_dentry
;
1647 dest
= dold
->d_inode
;
1649 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1651 err
= nfserrno(host_err
);
1654 host_err
= vfs_link(dold
, dirp
, dnew
);
1656 if (EX_ISSYNC(ffhp
->fh_export
)) {
1657 err
= nfserrno(nfsd_sync_dir(ddir
));
1658 write_inode_now(dest
, 1);
1662 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1665 err
= nfserrno(host_err
);
1667 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1676 err
= nfserrno(host_err
);
1682 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1685 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1686 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1688 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1689 struct inode
*fdir
, *tdir
;
1693 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1696 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1700 fdentry
= ffhp
->fh_dentry
;
1701 fdir
= fdentry
->d_inode
;
1703 tdentry
= tfhp
->fh_dentry
;
1704 tdir
= tdentry
->d_inode
;
1706 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1707 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1711 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1714 /* cannot use fh_lock as we need deadlock protective ordering
1715 * so do it by hand */
1716 trap
= lock_rename(tdentry
, fdentry
);
1717 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1721 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1722 host_err
= PTR_ERR(odentry
);
1723 if (IS_ERR(odentry
))
1727 if (!odentry
->d_inode
)
1730 if (odentry
== trap
)
1733 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1734 host_err
= PTR_ERR(ndentry
);
1735 if (IS_ERR(ndentry
))
1737 host_err
= -ENOTEMPTY
;
1738 if (ndentry
== trap
)
1741 if (svc_msnfs(ffhp
) &&
1742 ((atomic_read(&odentry
->d_count
) > 1)
1743 || (atomic_read(&ndentry
->d_count
) > 1))) {
1749 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1751 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1755 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1756 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1757 host_err
= nfsd_sync_dir(tdentry
);
1759 host_err
= nfsd_sync_dir(fdentry
);
1762 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1769 err
= nfserrno(host_err
);
1771 /* we cannot reply on fh_unlock on the two filehandles,
1772 * as that would do the wrong thing if the two directories
1773 * were the same, so again we do it by hand
1775 fill_post_wcc(ffhp
);
1776 fill_post_wcc(tfhp
);
1777 unlock_rename(tdentry
, fdentry
);
1778 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1785 * Unlink a file or directory
1786 * N.B. After this call fhp needs an fh_put
1789 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1790 char *fname
, int flen
)
1792 struct dentry
*dentry
, *rdentry
;
1798 if (!flen
|| isdotent(fname
, flen
))
1800 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1804 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1805 dentry
= fhp
->fh_dentry
;
1806 dirp
= dentry
->d_inode
;
1808 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1809 host_err
= PTR_ERR(rdentry
);
1810 if (IS_ERR(rdentry
))
1813 if (!rdentry
->d_inode
) {
1820 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1822 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1826 if (type
!= S_IFDIR
) { /* It's UNLINK */
1828 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1829 (atomic_read(&rdentry
->d_count
) > 1)) {
1833 host_err
= vfs_unlink(dirp
, rdentry
);
1834 } else { /* It's RMDIR */
1835 host_err
= vfs_rmdir(dirp
, rdentry
);
1842 if (EX_ISSYNC(fhp
->fh_export
))
1843 host_err
= nfsd_sync_dir(dentry
);
1846 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1848 err
= nfserrno(host_err
);
1854 * We do this buffering because we must not call back into the file
1855 * system's ->lookup() method from the filldir callback. That may well
1856 * deadlock a number of file systems.
1858 * This is based heavily on the implementation of same in XFS.
1860 struct buffered_dirent
{
1864 unsigned int d_type
;
1868 struct readdir_data
{
1874 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1875 loff_t offset
, u64 ino
, unsigned int d_type
)
1877 struct readdir_data
*buf
= __buf
;
1878 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1879 unsigned int reclen
;
1881 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1882 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1887 de
->namlen
= namlen
;
1888 de
->offset
= offset
;
1890 de
->d_type
= d_type
;
1891 memcpy(de
->name
, name
, namlen
);
1892 buf
->used
+= reclen
;
1897 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1898 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1900 struct readdir_data buf
;
1901 struct buffered_dirent
*de
;
1906 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1908 return nfserrno(-ENOMEM
);
1913 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1914 unsigned int reclen
;
1916 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1920 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1933 * Various filldir functions may end up calling back into
1934 * lookup_one_len() and the file system's ->lookup() method.
1935 * These expect i_mutex to be held, as it would within readdir.
1937 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1941 de
= (struct buffered_dirent
*)buf
.dirent
;
1943 offset
= de
->offset
;
1945 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1946 de
->ino
, de
->d_type
))
1949 if (cdp
->err
!= nfs_ok
)
1952 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1955 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1957 mutex_unlock(&dir_inode
->i_mutex
);
1958 if (size
> 0) /* We bailed out early */
1961 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1964 free_page((unsigned long)(buf
.dirent
));
1967 return nfserrno(host_err
);
1974 * Read entries from a directory.
1975 * The NFSv3/4 verifier we ignore for now.
1978 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1979 struct readdir_cd
*cdp
, filldir_t func
)
1983 loff_t offset
= *offsetp
;
1985 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1989 offset
= vfs_llseek(file
, offset
, 0);
1991 err
= nfserrno((int)offset
);
1995 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1997 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1998 err
= nfs_ok
; /* can still be found in ->err */
2006 * Get file system stats
2007 * N.B. After this call fhp needs an fh_put
2010 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2012 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2013 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
2018 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2020 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2024 * Check for a user's access permissions to this inode.
2027 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2028 struct dentry
*dentry
, int acc
)
2030 struct inode
*inode
= dentry
->d_inode
;
2034 if (acc
== NFSD_MAY_NOP
)
2037 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2039 (acc
& NFSD_MAY_READ
)? " read" : "",
2040 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2041 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2042 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2043 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2044 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2045 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2047 IS_IMMUTABLE(inode
)? " immut" : "",
2048 IS_APPEND(inode
)? " append" : "",
2049 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2050 dprintk(" owner %d/%d user %d/%d\n",
2051 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2054 /* Normally we reject any write/sattr etc access on a read-only file
2055 * system. But if it is IRIX doing check on write-access for a
2056 * device special file, we ignore rofs.
2058 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2059 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2060 if (exp_rdonly(rqstp
, exp
) ||
2061 __mnt_is_readonly(exp
->ex_path
.mnt
))
2063 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2066 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2069 if (acc
& NFSD_MAY_LOCK
) {
2070 /* If we cannot rely on authentication in NLM requests,
2071 * just allow locks, otherwise require read permission, or
2074 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2077 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2080 * The file owner always gets access permission for accesses that
2081 * would normally be checked at open time. This is to make
2082 * file access work even when the client has done a fchmod(fd, 0).
2084 * However, `cp foo bar' should fail nevertheless when bar is
2085 * readonly. A sensible way to do this might be to reject all
2086 * attempts to truncate a read-only file, because a creat() call
2087 * always implies file truncation.
2088 * ... but this isn't really fair. A process may reasonably call
2089 * ftruncate on an open file descriptor on a file with perm 000.
2090 * We must trust the client to do permission checking - using "ACCESS"
2093 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2094 inode
->i_uid
== current_fsuid())
2097 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2098 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2100 /* Allow read access to binaries even when mode 111 */
2101 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2102 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2103 err
= inode_permission(inode
, MAY_EXEC
);
2107 /* Do integrity (permission) checking now, but defer incrementing
2108 * IMA counts to the actual file open.
2110 path
.mnt
= exp
->ex_path
.mnt
;
2111 path
.dentry
= dentry
;
2112 err
= ima_path_check(&path
, acc
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
),
2115 return err
? nfserrno(err
) : 0;
2119 nfsd_racache_shutdown(void)
2121 struct raparms
*raparm
, *last_raparm
;
2124 dprintk("nfsd: freeing readahead buffers.\n");
2126 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2127 raparm
= raparm_hash
[i
].pb_head
;
2129 last_raparm
= raparm
;
2130 raparm
= raparm
->p_next
;
2133 raparm_hash
[i
].pb_head
= NULL
;
2137 * Initialize readahead param cache
2140 nfsd_racache_init(int cache_size
)
2145 struct raparms
**raparm
= NULL
;
2148 if (raparm_hash
[0].pb_head
)
2150 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2153 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2155 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2157 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2158 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2160 raparm
= &raparm_hash
[i
].pb_head
;
2161 for (j
= 0; j
< nperbucket
; j
++) {
2162 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2165 raparm
= &(*raparm
)->p_next
;
2170 nfsdstats
.ra_size
= cache_size
;
2174 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2175 nfsd_racache_shutdown();
2179 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2181 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2183 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2187 struct posix_acl
*acl
;
2189 if (!IS_POSIXACL(inode
))
2190 return ERR_PTR(-EOPNOTSUPP
);
2193 case ACL_TYPE_ACCESS
:
2194 name
= POSIX_ACL_XATTR_ACCESS
;
2196 case ACL_TYPE_DEFAULT
:
2197 name
= POSIX_ACL_XATTR_DEFAULT
;
2200 return ERR_PTR(-EOPNOTSUPP
);
2203 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2205 return ERR_PTR(size
);
2207 acl
= posix_acl_from_xattr(value
, size
);
2213 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2215 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2221 if (!IS_POSIXACL(inode
) ||
2222 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2225 case ACL_TYPE_ACCESS
:
2226 name
= POSIX_ACL_XATTR_ACCESS
;
2228 case ACL_TYPE_DEFAULT
:
2229 name
= POSIX_ACL_XATTR_DEFAULT
;
2235 if (acl
&& acl
->a_count
) {
2236 size
= posix_acl_xattr_size(acl
->a_count
);
2237 value
= kmalloc(size
, GFP_KERNEL
);
2240 error
= posix_acl_to_xattr(acl
, value
, size
);
2247 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2251 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2253 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2256 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2257 if (error
== -ENODATA
)
2261 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2267 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */