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
)
415 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
419 *buf
= kmalloc(buflen
, GFP_KERNEL
);
423 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
430 #if defined(CONFIG_NFSD_V4)
432 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
439 buflen
= posix_acl_xattr_size(pacl
->a_count
);
440 buf
= kmalloc(buflen
, GFP_KERNEL
);
445 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
451 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
458 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
459 struct nfs4_acl
*acl
)
463 struct dentry
*dentry
;
465 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
466 unsigned int flags
= 0;
469 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
473 dentry
= fhp
->fh_dentry
;
474 inode
= dentry
->d_inode
;
475 if (S_ISDIR(inode
->i_mode
))
476 flags
= NFS4_ACL_DIR
;
478 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
479 if (host_error
== -EINVAL
) {
480 return nfserr_attrnotsupp
;
481 } else if (host_error
< 0)
484 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
488 if (S_ISDIR(inode
->i_mode
))
489 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
492 posix_acl_release(pacl
);
493 posix_acl_release(dpacl
);
495 if (host_error
== -EOPNOTSUPP
)
496 return nfserr_attrnotsupp
;
498 return nfserrno(host_error
);
501 static struct posix_acl
*
502 _get_posix_acl(struct dentry
*dentry
, char *key
)
505 struct posix_acl
*pacl
= NULL
;
508 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
512 return ERR_PTR(buflen
);
514 pacl
= posix_acl_from_xattr(buf
, buflen
);
520 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
522 struct inode
*inode
= dentry
->d_inode
;
524 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
525 unsigned int flags
= 0;
527 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
528 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
529 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
531 error
= PTR_ERR(pacl
);
536 if (S_ISDIR(inode
->i_mode
)) {
537 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
538 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
540 else if (IS_ERR(dpacl
)) {
541 error
= PTR_ERR(dpacl
);
545 flags
= NFS4_ACL_DIR
;
548 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
550 error
= PTR_ERR(*acl
);
554 posix_acl_release(pacl
);
555 posix_acl_release(dpacl
);
559 #endif /* defined(CONFIG_NFS_V4) */
561 #ifdef CONFIG_NFSD_V3
563 * Check server access rights to a file system object
569 static struct accessmap nfs3_regaccess
[] = {
570 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
571 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
572 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
573 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
578 static struct accessmap nfs3_diraccess
[] = {
579 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
580 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
581 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
582 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
583 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
588 static struct accessmap nfs3_anyaccess
[] = {
589 /* Some clients - Solaris 2.6 at least, make an access call
590 * to the server to check for access for things like /dev/null
591 * (which really, the server doesn't care about). So
592 * We provide simple access checking for them, looking
593 * mainly at mode bits, and we make sure to ignore read-only
596 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
597 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
598 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
599 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
605 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
607 struct accessmap
*map
;
608 struct svc_export
*export
;
609 struct dentry
*dentry
;
610 u32 query
, result
= 0, sresult
= 0;
613 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
617 export
= fhp
->fh_export
;
618 dentry
= fhp
->fh_dentry
;
620 if (S_ISREG(dentry
->d_inode
->i_mode
))
621 map
= nfs3_regaccess
;
622 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
623 map
= nfs3_diraccess
;
625 map
= nfs3_anyaccess
;
629 for (; map
->access
; map
++) {
630 if (map
->access
& query
) {
633 sresult
|= map
->access
;
635 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
638 result
|= map
->access
;
641 /* the following error codes just mean the access was not allowed,
642 * rather than an error occurred */
646 /* simply don't "or" in the access bit. */
656 *supported
= sresult
;
661 #endif /* CONFIG_NFSD_V3 */
666 * Open an existing file or directory.
667 * The access argument indicates the type of open (read/write/lock)
668 * N.B. After this call fhp needs an fh_put
671 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
672 int access
, struct file
**filp
)
674 const struct cred
*cred
= current_cred();
675 struct dentry
*dentry
;
677 int flags
= O_RDONLY
|O_LARGEFILE
;
682 * If we get here, then the client has already done an "open",
683 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
684 * in case a chmod has now revoked permission.
686 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
690 dentry
= fhp
->fh_dentry
;
691 inode
= dentry
->d_inode
;
693 /* Disallow write access to files with the append-only bit set
694 * or any access when mandatory locking enabled
697 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
700 * We must ignore files (but only files) which might have mandatory
701 * locks on them because there is no way to know if the accesser has
704 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
711 * Check to see if there are any leases on this file.
712 * This may block while leases are broken.
714 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
715 if (host_err
== -EWOULDBLOCK
)
716 host_err
= -ETIMEDOUT
;
717 if (host_err
) /* NOMEM or WOULDBLOCK */
720 if (access
& NFSD_MAY_WRITE
) {
721 if (access
& NFSD_MAY_READ
)
722 flags
= O_RDWR
|O_LARGEFILE
;
724 flags
= O_WRONLY
|O_LARGEFILE
;
728 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
731 host_err
= PTR_ERR(*filp
);
733 err
= nfserrno(host_err
);
742 nfsd_close(struct file
*filp
)
748 nfsd_sync(struct file
*filp
)
750 return vfs_fsync(filp
, filp
->f_path
.dentry
, 0);
754 nfsd_sync_dir(struct dentry
*dentry
)
756 return vfs_fsync(NULL
, dentry
, 0);
760 * Obtain the readahead parameters for the file
761 * specified by (dev, ino).
764 static inline struct raparms
*
765 nfsd_get_raparms(dev_t dev
, ino_t ino
)
767 struct raparms
*ra
, **rap
, **frap
= NULL
;
770 struct raparm_hbucket
*rab
;
772 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
773 rab
= &raparm_hash
[hash
];
775 spin_lock(&rab
->pb_lock
);
776 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
777 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
780 if (ra
->p_count
== 0)
783 depth
= nfsdstats
.ra_size
*11/10;
785 spin_unlock(&rab
->pb_lock
);
795 if (rap
!= &rab
->pb_head
) {
797 ra
->p_next
= rab
->pb_head
;
801 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
802 spin_unlock(&rab
->pb_lock
);
807 * Grab and keep cached pages associated with a file in the svc_rqst
808 * so that they can be passed to the network sendmsg/sendpage routines
809 * directly. They will be released after the sending has completed.
812 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
813 struct splice_desc
*sd
)
815 struct svc_rqst
*rqstp
= sd
->u
.data
;
816 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
817 struct page
*page
= buf
->page
;
821 ret
= buf
->ops
->confirm(pipe
, buf
);
827 if (rqstp
->rq_res
.page_len
== 0) {
832 rqstp
->rq_res
.page_base
= buf
->offset
;
833 rqstp
->rq_res
.page_len
= size
;
834 } else if (page
!= pp
[-1]) {
840 rqstp
->rq_res
.page_len
+= size
;
842 rqstp
->rq_res
.page_len
+= size
;
847 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
848 struct splice_desc
*sd
)
850 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
853 static inline int svc_msnfs(struct svc_fh
*ffhp
)
856 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
863 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
864 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
873 inode
= file
->f_path
.dentry
->d_inode
;
875 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
878 /* Get readahead parameters */
879 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
882 file
->f_ra
= ra
->p_ra
;
884 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
885 struct splice_desc sd
= {
892 rqstp
->rq_resused
= 1;
893 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
897 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
901 /* Write back readahead params */
903 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
904 spin_lock(&rab
->pb_lock
);
905 ra
->p_ra
= file
->f_ra
;
908 spin_unlock(&rab
->pb_lock
);
912 nfsdstats
.io_read
+= host_err
;
915 fsnotify_access(file
->f_path
.dentry
);
917 err
= nfserrno(host_err
);
922 static void kill_suid(struct dentry
*dentry
)
925 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
927 mutex_lock(&dentry
->d_inode
->i_mutex
);
928 notify_change(dentry
, &ia
);
929 mutex_unlock(&dentry
->d_inode
->i_mutex
);
933 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
934 loff_t offset
, struct kvec
*vec
, int vlen
,
935 unsigned long cnt
, int *stablep
)
937 struct svc_export
*exp
;
938 struct dentry
*dentry
;
943 int stable
= *stablep
;
948 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
949 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, cnt
)))
953 dentry
= file
->f_path
.dentry
;
954 inode
= dentry
->d_inode
;
955 exp
= fhp
->fh_export
;
958 * Request sync writes if
959 * - the sync export option has been set, or
960 * - the client requested O_SYNC behavior (NFSv3 feature).
961 * - The file system doesn't support fsync().
962 * When gathered writes have been configured for this volume,
963 * flushing the data to disk is handled separately below.
966 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
968 *stablep
= 2; /* FILE_SYNC */
973 if (stable
&& !EX_WGATHER(exp
))
974 file
->f_flags
|= O_SYNC
;
976 /* Write the data. */
977 oldfs
= get_fs(); set_fs(KERNEL_DS
);
978 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
981 nfsdstats
.io_write
+= cnt
;
982 fsnotify_modify(file
->f_path
.dentry
);
985 /* clear setuid/setgid flag after write */
986 if (host_err
>= 0 && (inode
->i_mode
& (S_ISUID
| S_ISGID
)))
989 if (host_err
>= 0 && stable
) {
990 static ino_t last_ino
;
991 static dev_t last_dev
;
994 * Gathered writes: If another process is currently
995 * writing to the file, there's a high chance
996 * this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the
998 * file with each write request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates
1001 * C. Juszak's idea of gathered writes, but it's a
1002 * nice and simple solution (IMHO), and it seems to
1005 if (EX_WGATHER(exp
)) {
1006 if (atomic_read(&inode
->i_writecount
) > 1
1007 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1008 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1010 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1013 if (inode
->i_state
& I_DIRTY
) {
1014 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1015 host_err
=nfsd_sync(file
);
1018 wake_up(&inode
->i_wait
);
1021 last_ino
= inode
->i_ino
;
1022 last_dev
= inode
->i_sb
->s_dev
;
1025 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1029 err
= nfserrno(host_err
);
1035 * Read data from a file. count must contain the requested read count
1036 * on entry. On return, *count contains the number of bytes actually read.
1037 * N.B. After this call fhp needs an fh_put
1040 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1041 loff_t offset
, struct kvec
*vec
, int vlen
,
1042 unsigned long *count
)
1047 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1048 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1051 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1053 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1056 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1064 * Write data to a file.
1065 * The stable flag requests synchronous writes.
1066 * N.B. After this call fhp needs an fh_put
1069 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1070 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long cnt
,
1076 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1077 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1080 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1083 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1088 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1096 #ifdef CONFIG_NFSD_V3
1098 * Commit all pending writes to stable storage.
1099 * Strictly speaking, we could sync just the indicated file region here,
1100 * but there's currently no way we can ask the VFS to do so.
1102 * Unfortunately we cannot lock the file to make sure we return full WCC
1103 * data to the client, as locking happens lower down in the filesystem.
1106 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1107 loff_t offset
, unsigned long count
)
1112 if ((u64
)count
> ~(u64
)offset
)
1113 return nfserr_inval
;
1115 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1118 if (EX_ISSYNC(fhp
->fh_export
)) {
1119 if (file
->f_op
&& file
->f_op
->fsync
) {
1120 err
= nfserrno(nfsd_sync(file
));
1122 err
= nfserr_notsupp
;
1129 #endif /* CONFIG_NFSD_V3 */
1132 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1136 * Mode has already been set earlier in create:
1138 iap
->ia_valid
&= ~ATTR_MODE
;
1140 * Setting uid/gid works only for root. Irix appears to
1141 * send along the gid on create when it tries to implement
1142 * setgid directories via NFS:
1144 if (current_fsuid() != 0)
1145 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1147 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1152 * Create a file (regular, directory, device, fifo); UNIX sockets
1153 * not yet implemented.
1154 * If the response fh has been verified, the parent directory should
1155 * already be locked. Note that the parent directory is left locked.
1157 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1160 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1161 char *fname
, int flen
, struct iattr
*iap
,
1162 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1164 struct dentry
*dentry
, *dchild
= NULL
;
1174 if (isdotent(fname
, flen
))
1177 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1181 dentry
= fhp
->fh_dentry
;
1182 dirp
= dentry
->d_inode
;
1184 err
= nfserr_notdir
;
1185 if (!dirp
->i_op
->lookup
)
1188 * Check whether the response file handle has been verified yet.
1189 * If it has, the parent directory should already be locked.
1191 if (!resfhp
->fh_dentry
) {
1192 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1193 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1194 dchild
= lookup_one_len(fname
, dentry
, flen
);
1195 host_err
= PTR_ERR(dchild
);
1198 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1202 /* called from nfsd_proc_create */
1203 dchild
= dget(resfhp
->fh_dentry
);
1204 if (!fhp
->fh_locked
) {
1205 /* not actually possible */
1207 "nfsd_create: parent %s/%s not locked!\n",
1208 dentry
->d_parent
->d_name
.name
,
1209 dentry
->d_name
.name
);
1215 * Make sure the child dentry is still negative ...
1218 if (dchild
->d_inode
) {
1219 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1220 dentry
->d_name
.name
, dchild
->d_name
.name
);
1224 if (!(iap
->ia_valid
& ATTR_MODE
))
1226 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1229 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1230 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1235 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1240 * Get the dir op function pointer.
1245 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1248 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1254 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1258 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1262 if (EX_ISSYNC(fhp
->fh_export
)) {
1263 err
= nfserrno(nfsd_sync_dir(dentry
));
1264 write_inode_now(dchild
->d_inode
, 1);
1267 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1270 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1272 * Update the file handle to get the new inode info.
1275 err
= fh_update(resfhp
);
1277 if (dchild
&& !IS_ERR(dchild
))
1282 err
= nfserrno(host_err
);
1286 #ifdef CONFIG_NFSD_V3
1288 * NFSv3 version of nfsd_create
1291 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1292 char *fname
, int flen
, struct iattr
*iap
,
1293 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1294 int *truncp
, int *created
)
1296 struct dentry
*dentry
, *dchild
= NULL
;
1301 __u32 v_mtime
=0, v_atime
=0;
1307 if (isdotent(fname
, flen
))
1309 if (!(iap
->ia_valid
& ATTR_MODE
))
1311 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1315 dentry
= fhp
->fh_dentry
;
1316 dirp
= dentry
->d_inode
;
1318 /* Get all the sanity checks out of the way before
1319 * we lock the parent. */
1320 err
= nfserr_notdir
;
1321 if (!dirp
->i_op
->lookup
)
1323 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1326 * Compose the response file handle.
1328 dchild
= lookup_one_len(fname
, dentry
, flen
);
1329 host_err
= PTR_ERR(dchild
);
1333 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1337 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1338 /* solaris7 gets confused (bugid 4218508) if these have
1339 * the high bit set, so just clear the high bits. If this is
1340 * ever changed to use different attrs for storing the
1341 * verifier, then do_open_lookup() will also need to be fixed
1344 v_mtime
= verifier
[0]&0x7fffffff;
1345 v_atime
= verifier
[1]&0x7fffffff;
1348 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1351 if (dchild
->d_inode
) {
1354 switch (createmode
) {
1355 case NFS3_CREATE_UNCHECKED
:
1356 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1359 /* in nfsv4, we need to treat this case a little
1360 * differently. we don't want to truncate the
1361 * file now; this would be wrong if the OPEN
1362 * fails for some other reason. furthermore,
1363 * if the size is nonzero, we should ignore it
1364 * according to spec!
1366 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1369 iap
->ia_valid
&= ATTR_SIZE
;
1373 case NFS3_CREATE_EXCLUSIVE
:
1374 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1375 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1376 && dchild
->d_inode
->i_size
== 0 )
1379 case NFS3_CREATE_GUARDED
:
1382 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1386 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1388 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1394 if (EX_ISSYNC(fhp
->fh_export
)) {
1395 err
= nfserrno(nfsd_sync_dir(dentry
));
1396 /* setattr will sync the child (or not) */
1399 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1400 /* Cram the verifier into atime/mtime */
1401 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1402 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1403 /* XXX someone who knows this better please fix it for nsec */
1404 iap
->ia_mtime
.tv_sec
= v_mtime
;
1405 iap
->ia_atime
.tv_sec
= v_atime
;
1406 iap
->ia_mtime
.tv_nsec
= 0;
1407 iap
->ia_atime
.tv_nsec
= 0;
1411 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1415 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1417 * Update the filehandle to get the new inode info.
1420 err
= fh_update(resfhp
);
1424 if (dchild
&& !IS_ERR(dchild
))
1429 err
= nfserrno(host_err
);
1432 #endif /* CONFIG_NFSD_V3 */
1435 * Read a symlink. On entry, *lenp must contain the maximum path length that
1436 * fits into the buffer. On return, it contains the true length.
1437 * N.B. After this call fhp needs an fh_put
1440 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1442 struct dentry
*dentry
;
1443 struct inode
*inode
;
1448 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1452 dentry
= fhp
->fh_dentry
;
1453 inode
= dentry
->d_inode
;
1456 if (!inode
->i_op
->readlink
)
1459 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1460 /* N.B. Why does this call need a get_fs()??
1461 * Remove the set_fs and watch the fireworks:-) --okir
1464 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1465 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1476 err
= nfserrno(host_err
);
1481 * Create a symlink and look up its inode
1482 * N.B. After this call _both_ fhp and resfhp need an fh_put
1485 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1486 char *fname
, int flen
,
1487 char *path
, int plen
,
1488 struct svc_fh
*resfhp
,
1491 struct dentry
*dentry
, *dnew
;
1499 if (isdotent(fname
, flen
))
1502 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1506 dentry
= fhp
->fh_dentry
;
1507 dnew
= lookup_one_len(fname
, dentry
, flen
);
1508 host_err
= PTR_ERR(dnew
);
1512 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1516 if (unlikely(path
[plen
] != 0)) {
1517 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1518 if (path_alloced
== NULL
)
1521 strncpy(path_alloced
, path
, plen
);
1522 path_alloced
[plen
] = 0;
1523 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1524 kfree(path_alloced
);
1527 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1530 if (EX_ISSYNC(fhp
->fh_export
))
1531 host_err
= nfsd_sync_dir(dentry
);
1533 err
= nfserrno(host_err
);
1536 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1538 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1540 if (err
==0) err
= cerr
;
1545 err
= nfserrno(host_err
);
1551 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1554 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1555 char *name
, int len
, struct svc_fh
*tfhp
)
1557 struct dentry
*ddir
, *dnew
, *dold
;
1558 struct inode
*dirp
, *dest
;
1562 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1565 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1573 if (isdotent(name
, len
))
1576 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1577 ddir
= ffhp
->fh_dentry
;
1578 dirp
= ddir
->d_inode
;
1580 dnew
= lookup_one_len(name
, ddir
, len
);
1581 host_err
= PTR_ERR(dnew
);
1585 dold
= tfhp
->fh_dentry
;
1586 dest
= dold
->d_inode
;
1588 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1590 err
= nfserrno(host_err
);
1593 host_err
= vfs_link(dold
, dirp
, dnew
);
1595 if (EX_ISSYNC(ffhp
->fh_export
)) {
1596 err
= nfserrno(nfsd_sync_dir(ddir
));
1597 write_inode_now(dest
, 1);
1601 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1604 err
= nfserrno(host_err
);
1606 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1615 err
= nfserrno(host_err
);
1621 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1624 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1625 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1627 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1628 struct inode
*fdir
, *tdir
;
1632 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1635 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1639 fdentry
= ffhp
->fh_dentry
;
1640 fdir
= fdentry
->d_inode
;
1642 tdentry
= tfhp
->fh_dentry
;
1643 tdir
= tdentry
->d_inode
;
1645 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1646 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1650 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1653 /* cannot use fh_lock as we need deadlock protective ordering
1654 * so do it by hand */
1655 trap
= lock_rename(tdentry
, fdentry
);
1656 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1660 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1661 host_err
= PTR_ERR(odentry
);
1662 if (IS_ERR(odentry
))
1666 if (!odentry
->d_inode
)
1669 if (odentry
== trap
)
1672 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1673 host_err
= PTR_ERR(ndentry
);
1674 if (IS_ERR(ndentry
))
1676 host_err
= -ENOTEMPTY
;
1677 if (ndentry
== trap
)
1680 if (svc_msnfs(ffhp
) &&
1681 ((atomic_read(&odentry
->d_count
) > 1)
1682 || (atomic_read(&ndentry
->d_count
) > 1))) {
1688 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1690 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1694 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1695 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1696 host_err
= nfsd_sync_dir(tdentry
);
1698 host_err
= nfsd_sync_dir(fdentry
);
1701 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1708 err
= nfserrno(host_err
);
1710 /* we cannot reply on fh_unlock on the two filehandles,
1711 * as that would do the wrong thing if the two directories
1712 * were the same, so again we do it by hand
1714 fill_post_wcc(ffhp
);
1715 fill_post_wcc(tfhp
);
1716 unlock_rename(tdentry
, fdentry
);
1717 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1724 * Unlink a file or directory
1725 * N.B. After this call fhp needs an fh_put
1728 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1729 char *fname
, int flen
)
1731 struct dentry
*dentry
, *rdentry
;
1737 if (!flen
|| isdotent(fname
, flen
))
1739 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1743 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1744 dentry
= fhp
->fh_dentry
;
1745 dirp
= dentry
->d_inode
;
1747 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1748 host_err
= PTR_ERR(rdentry
);
1749 if (IS_ERR(rdentry
))
1752 if (!rdentry
->d_inode
) {
1759 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1761 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1765 if (type
!= S_IFDIR
) { /* It's UNLINK */
1767 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1768 (atomic_read(&rdentry
->d_count
) > 1)) {
1772 host_err
= vfs_unlink(dirp
, rdentry
);
1773 } else { /* It's RMDIR */
1774 host_err
= vfs_rmdir(dirp
, rdentry
);
1781 if (EX_ISSYNC(fhp
->fh_export
))
1782 host_err
= nfsd_sync_dir(dentry
);
1785 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1787 err
= nfserrno(host_err
);
1793 * We do this buffering because we must not call back into the file
1794 * system's ->lookup() method from the filldir callback. That may well
1795 * deadlock a number of file systems.
1797 * This is based heavily on the implementation of same in XFS.
1799 struct buffered_dirent
{
1803 unsigned int d_type
;
1807 struct readdir_data
{
1813 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1814 loff_t offset
, u64 ino
, unsigned int d_type
)
1816 struct readdir_data
*buf
= __buf
;
1817 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1818 unsigned int reclen
;
1820 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1821 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1826 de
->namlen
= namlen
;
1827 de
->offset
= offset
;
1829 de
->d_type
= d_type
;
1830 memcpy(de
->name
, name
, namlen
);
1831 buf
->used
+= reclen
;
1836 static int nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1837 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1839 struct readdir_data buf
;
1840 struct buffered_dirent
*de
;
1845 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1852 unsigned int reclen
;
1854 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1858 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1870 de
= (struct buffered_dirent
*)buf
.dirent
;
1872 offset
= de
->offset
;
1874 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1875 de
->ino
, de
->d_type
))
1878 if (cdp
->err
!= nfs_ok
)
1881 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1884 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1886 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1890 free_page((unsigned long)(buf
.dirent
));
1893 return nfserrno(host_err
);
1900 * Read entries from a directory.
1901 * The NFSv3/4 verifier we ignore for now.
1904 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1905 struct readdir_cd
*cdp
, filldir_t func
)
1909 loff_t offset
= *offsetp
;
1911 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1915 offset
= vfs_llseek(file
, offset
, 0);
1917 err
= nfserrno((int)offset
);
1921 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1923 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1924 err
= nfs_ok
; /* can still be found in ->err */
1932 * Get file system stats
1933 * N.B. After this call fhp needs an fh_put
1936 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
1938 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
1939 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
1944 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
1946 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
1950 * Check for a user's access permissions to this inode.
1953 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
1954 struct dentry
*dentry
, int acc
)
1956 struct inode
*inode
= dentry
->d_inode
;
1959 if (acc
== NFSD_MAY_NOP
)
1962 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1964 (acc
& NFSD_MAY_READ
)? " read" : "",
1965 (acc
& NFSD_MAY_WRITE
)? " write" : "",
1966 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
1967 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
1968 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
1969 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
1970 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
1972 IS_IMMUTABLE(inode
)? " immut" : "",
1973 IS_APPEND(inode
)? " append" : "",
1974 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
1975 dprintk(" owner %d/%d user %d/%d\n",
1976 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
1979 /* Normally we reject any write/sattr etc access on a read-only file
1980 * system. But if it is IRIX doing check on write-access for a
1981 * device special file, we ignore rofs.
1983 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
1984 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
1985 if (exp_rdonly(rqstp
, exp
) ||
1986 __mnt_is_readonly(exp
->ex_path
.mnt
))
1988 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
1991 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
1994 if (acc
& NFSD_MAY_LOCK
) {
1995 /* If we cannot rely on authentication in NLM requests,
1996 * just allow locks, otherwise require read permission, or
1999 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2002 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2005 * The file owner always gets access permission for accesses that
2006 * would normally be checked at open time. This is to make
2007 * file access work even when the client has done a fchmod(fd, 0).
2009 * However, `cp foo bar' should fail nevertheless when bar is
2010 * readonly. A sensible way to do this might be to reject all
2011 * attempts to truncate a read-only file, because a creat() call
2012 * always implies file truncation.
2013 * ... but this isn't really fair. A process may reasonably call
2014 * ftruncate on an open file descriptor on a file with perm 000.
2015 * We must trust the client to do permission checking - using "ACCESS"
2018 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2019 inode
->i_uid
== current_fsuid())
2022 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2023 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2025 /* Allow read access to binaries even when mode 111 */
2026 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2027 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2028 err
= inode_permission(inode
, MAY_EXEC
);
2030 return err
? nfserrno(err
) : 0;
2034 nfsd_racache_shutdown(void)
2036 struct raparms
*raparm
, *last_raparm
;
2039 dprintk("nfsd: freeing readahead buffers.\n");
2041 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2042 raparm
= raparm_hash
[i
].pb_head
;
2044 last_raparm
= raparm
;
2045 raparm
= raparm
->p_next
;
2048 raparm_hash
[i
].pb_head
= NULL
;
2052 * Initialize readahead param cache
2055 nfsd_racache_init(int cache_size
)
2060 struct raparms
**raparm
= NULL
;
2063 if (raparm_hash
[0].pb_head
)
2065 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2068 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2070 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2072 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2073 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2075 raparm
= &raparm_hash
[i
].pb_head
;
2076 for (j
= 0; j
< nperbucket
; j
++) {
2077 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2080 raparm
= &(*raparm
)->p_next
;
2085 nfsdstats
.ra_size
= cache_size
;
2089 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2090 nfsd_racache_shutdown();
2094 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2096 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2098 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2102 struct posix_acl
*acl
;
2104 if (!IS_POSIXACL(inode
))
2105 return ERR_PTR(-EOPNOTSUPP
);
2108 case ACL_TYPE_ACCESS
:
2109 name
= POSIX_ACL_XATTR_ACCESS
;
2111 case ACL_TYPE_DEFAULT
:
2112 name
= POSIX_ACL_XATTR_DEFAULT
;
2115 return ERR_PTR(-EOPNOTSUPP
);
2118 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2120 return ERR_PTR(size
);
2122 acl
= posix_acl_from_xattr(value
, size
);
2128 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2130 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2136 if (!IS_POSIXACL(inode
) ||
2137 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2140 case ACL_TYPE_ACCESS
:
2141 name
= POSIX_ACL_XATTR_ACCESS
;
2143 case ACL_TYPE_DEFAULT
:
2144 name
= POSIX_ACL_XATTR_DEFAULT
;
2150 if (acl
&& acl
->a_count
) {
2151 size
= posix_acl_xattr_size(acl
->a_count
);
2152 value
= kmalloc(size
, GFP_KERNEL
);
2155 error
= posix_acl_to_xattr(acl
, value
, size
);
2162 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2166 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2168 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2171 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2172 if (error
== -ENODATA
)
2176 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2182 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */