1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
35 #endif /* CONFIG_NFSD_V3 */
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms
*p_next
;
61 struct file_ra_state p_ra
;
62 unsigned int p_hindex
;
65 struct raparm_hbucket
{
66 struct raparms
*pb_head
;
68 } ____cacheline_aligned_in_smp
;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
83 struct svc_export
**expp
)
85 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
86 struct dentry
*dentry
= *dpp
;
87 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
88 .dentry
= dget(dentry
)};
91 while (d_mountpoint(path
.dentry
) && follow_down(&path
))
94 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
98 * We normally allow NFS clients to continue
99 * "underneath" a mountpoint that is not exported.
100 * The exception is V4ROOT, where no traversal is ever
101 * allowed without an explicit export of the new
104 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
109 if (nfsd_v4client(rqstp
) ||
110 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
111 /* successfully crossed mount point */
113 * This is subtle: path.dentry is *not* on path.mnt
114 * at this point. The only reason we are safe is that
115 * original mnt is pinned down by exp, so we should
116 * put path *before* putting exp
119 path
.dentry
= dentry
;
129 static void follow_to_parent(struct path
*path
)
133 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
135 dp
= dget_parent(path
->dentry
);
140 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
142 struct svc_export
*exp2
;
143 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
144 .dentry
= dget(dparent
)};
146 follow_to_parent(&path
);
148 exp2
= rqst_exp_parent(rqstp
, &path
);
149 if (PTR_ERR(exp2
) == -ENOENT
) {
150 *dentryp
= dget(dparent
);
151 } else if (IS_ERR(exp2
)) {
153 return PTR_ERR(exp2
);
155 *dentryp
= dget(path
.dentry
);
164 * For nfsd purposes, we treat V4ROOT exports as though there was an
165 * export at *every* directory.
167 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
169 if (d_mountpoint(dentry
))
171 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
173 return dentry
->d_inode
!= NULL
;
177 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
178 const char *name
, unsigned int len
,
179 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
181 struct svc_export
*exp
;
182 struct dentry
*dparent
;
183 struct dentry
*dentry
;
187 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
189 /* Obtain dentry and export. */
190 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
194 dparent
= fhp
->fh_dentry
;
195 exp
= fhp
->fh_export
;
198 /* Lookup the name, but don't follow links */
199 if (isdotent(name
, len
)) {
201 dentry
= dget(dparent
);
202 else if (dparent
!= exp
->ex_path
.dentry
)
203 dentry
= dget_parent(dparent
);
204 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
205 dentry
= dget(dparent
); /* .. == . just like at / */
207 /* checking mountpoint crossing is very different when stepping up */
208 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
214 dentry
= lookup_one_len(name
, dparent
, len
);
215 host_err
= PTR_ERR(dentry
);
219 * check if we have crossed a mount point ...
221 if (nfsd_mountpoint(dentry
, exp
)) {
222 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
228 *dentry_ret
= dentry
;
234 return nfserrno(host_err
);
238 * Look up one component of a pathname.
239 * N.B. After this call _both_ fhp and resfh need an fh_put
241 * If the lookup would cross a mountpoint, and the mounted filesystem
242 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
243 * accepted as it stands and the mounted directory is
244 * returned. Otherwise the covered directory is returned.
245 * NOTE: this mountpoint crossing is not supported properly by all
246 * clients and is explicitly disallowed for NFSv3
247 * NeilBrown <neilb@cse.unsw.edu.au>
250 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
251 unsigned int len
, struct svc_fh
*resfh
)
253 struct svc_export
*exp
;
254 struct dentry
*dentry
;
257 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
260 err
= check_nfsd_access(exp
, rqstp
);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
268 if (!err
&& !dentry
->d_inode
)
277 * Commit metadata changes to stable storage.
280 commit_metadata(struct svc_fh
*fhp
)
282 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
283 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
285 if (!EX_ISSYNC(fhp
->fh_export
))
288 if (export_ops
->commit_metadata
)
289 return export_ops
->commit_metadata(inode
);
290 return sync_inode_metadata(inode
, 1);
294 * Set various file attributes.
295 * N.B. After this call fhp needs an fh_put
298 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
299 int check_guard
, time_t guardtime
)
301 struct dentry
*dentry
;
303 int accmode
= NFSD_MAY_SATTR
;
309 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
310 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
311 if (iap
->ia_valid
& ATTR_SIZE
)
315 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
319 dentry
= fhp
->fh_dentry
;
320 inode
= dentry
->d_inode
;
322 /* Ignore any mode updates on symlinks */
323 if (S_ISLNK(inode
->i_mode
))
324 iap
->ia_valid
&= ~ATTR_MODE
;
330 * NFSv2 does not differentiate between "set-[ac]time-to-now"
331 * which only requires access, and "set-[ac]time-to-X" which
332 * requires ownership.
333 * So if it looks like it might be "set both to the same time which
334 * is close to now", and if inode_change_ok fails, then we
335 * convert to "set to now" instead of "set to explicit time"
337 * We only call inode_change_ok as the last test as technically
338 * it is not an interface that we should be using. It is only
339 * valid if the filesystem does not define it's own i_op->setattr.
341 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
342 #define MAX_TOUCH_TIME_ERROR (30*60)
343 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
344 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
348 * Now just make sure time is in the right ballpark.
349 * Solaris, at least, doesn't seem to care what the time
350 * request is. We require it be within 30 minutes of now.
352 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
355 if (delta
< MAX_TOUCH_TIME_ERROR
&&
356 inode_change_ok(inode
, iap
) != 0) {
358 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
359 * This will cause notify_change to set these times
362 iap
->ia_valid
&= ~BOTH_TIME_SET
;
367 * The size case is special.
368 * It changes the file as well as the attributes.
370 if (iap
->ia_valid
& ATTR_SIZE
) {
371 if (iap
->ia_size
< inode
->i_size
) {
372 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
373 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
379 * If we are changing the size of the file, then
380 * we need to break all leases.
382 host_err
= break_lease(inode
, O_WRONLY
| O_NONBLOCK
);
383 if (host_err
== -EWOULDBLOCK
)
384 host_err
= -ETIMEDOUT
;
385 if (host_err
) /* ENOMEM or EWOULDBLOCK */
388 host_err
= get_write_access(inode
);
393 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
395 put_write_access(inode
);
400 /* sanitize the mode change */
401 if (iap
->ia_valid
& ATTR_MODE
) {
402 iap
->ia_mode
&= S_IALLUGO
;
403 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
406 /* Revoke setuid/setgid on chown */
407 if (!S_ISDIR(inode
->i_mode
) &&
408 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
409 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
410 iap
->ia_valid
|= ATTR_KILL_PRIV
;
411 if (iap
->ia_valid
& ATTR_MODE
) {
412 /* we're setting mode too, just clear the s*id bits */
413 iap
->ia_mode
&= ~S_ISUID
;
414 if (iap
->ia_mode
& S_IXGRP
)
415 iap
->ia_mode
&= ~S_ISGID
;
417 /* set ATTR_KILL_* bits and let VFS handle it */
418 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
422 /* Change the attributes. */
424 iap
->ia_valid
|= ATTR_CTIME
;
426 err
= nfserr_notsync
;
427 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
429 host_err
= notify_change(dentry
, iap
);
430 err
= nfserrno(host_err
);
434 put_write_access(inode
);
436 commit_metadata(fhp
);
441 err
= nfserrno(host_err
);
445 #if defined(CONFIG_NFSD_V2_ACL) || \
446 defined(CONFIG_NFSD_V3_ACL) || \
447 defined(CONFIG_NFSD_V4)
448 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
453 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
457 *buf
= kmalloc(buflen
, GFP_KERNEL
);
461 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
468 #if defined(CONFIG_NFSD_V4)
470 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
477 buflen
= posix_acl_xattr_size(pacl
->a_count
);
478 buf
= kmalloc(buflen
, GFP_KERNEL
);
483 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
489 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
496 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
497 struct nfs4_acl
*acl
)
501 struct dentry
*dentry
;
503 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
504 unsigned int flags
= 0;
507 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
511 dentry
= fhp
->fh_dentry
;
512 inode
= dentry
->d_inode
;
513 if (S_ISDIR(inode
->i_mode
))
514 flags
= NFS4_ACL_DIR
;
516 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
517 if (host_error
== -EINVAL
) {
518 return nfserr_attrnotsupp
;
519 } else if (host_error
< 0)
522 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
526 if (S_ISDIR(inode
->i_mode
))
527 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
530 posix_acl_release(pacl
);
531 posix_acl_release(dpacl
);
533 if (host_error
== -EOPNOTSUPP
)
534 return nfserr_attrnotsupp
;
536 return nfserrno(host_error
);
539 static struct posix_acl
*
540 _get_posix_acl(struct dentry
*dentry
, char *key
)
543 struct posix_acl
*pacl
= NULL
;
546 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
550 return ERR_PTR(buflen
);
552 pacl
= posix_acl_from_xattr(buf
, buflen
);
558 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
560 struct inode
*inode
= dentry
->d_inode
;
562 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
563 unsigned int flags
= 0;
565 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
566 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
567 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
569 error
= PTR_ERR(pacl
);
574 if (S_ISDIR(inode
->i_mode
)) {
575 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
576 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
578 else if (IS_ERR(dpacl
)) {
579 error
= PTR_ERR(dpacl
);
583 flags
= NFS4_ACL_DIR
;
586 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
588 error
= PTR_ERR(*acl
);
592 posix_acl_release(pacl
);
593 posix_acl_release(dpacl
);
597 #endif /* defined(CONFIG_NFSD_V4) */
599 #ifdef CONFIG_NFSD_V3
601 * Check server access rights to a file system object
607 static struct accessmap nfs3_regaccess
[] = {
608 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
609 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
610 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
611 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
616 static struct accessmap nfs3_diraccess
[] = {
617 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
618 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
619 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
620 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
621 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
626 static struct accessmap nfs3_anyaccess
[] = {
627 /* Some clients - Solaris 2.6 at least, make an access call
628 * to the server to check for access for things like /dev/null
629 * (which really, the server doesn't care about). So
630 * We provide simple access checking for them, looking
631 * mainly at mode bits, and we make sure to ignore read-only
634 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
635 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
636 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
637 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
643 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
645 struct accessmap
*map
;
646 struct svc_export
*export
;
647 struct dentry
*dentry
;
648 u32 query
, result
= 0, sresult
= 0;
651 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
655 export
= fhp
->fh_export
;
656 dentry
= fhp
->fh_dentry
;
658 if (S_ISREG(dentry
->d_inode
->i_mode
))
659 map
= nfs3_regaccess
;
660 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
661 map
= nfs3_diraccess
;
663 map
= nfs3_anyaccess
;
667 for (; map
->access
; map
++) {
668 if (map
->access
& query
) {
671 sresult
|= map
->access
;
673 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
676 result
|= map
->access
;
679 /* the following error codes just mean the access was not allowed,
680 * rather than an error occurred */
684 /* simply don't "or" in the access bit. */
694 *supported
= sresult
;
699 #endif /* CONFIG_NFSD_V3 */
704 * Open an existing file or directory.
705 * The access argument indicates the type of open (read/write/lock)
706 * N.B. After this call fhp needs an fh_put
709 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
710 int access
, struct file
**filp
)
712 struct dentry
*dentry
;
714 int flags
= O_RDONLY
|O_LARGEFILE
;
718 validate_process_creds();
721 * If we get here, then the client has already done an "open",
722 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
723 * in case a chmod has now revoked permission.
725 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
729 dentry
= fhp
->fh_dentry
;
730 inode
= dentry
->d_inode
;
732 /* Disallow write access to files with the append-only bit set
733 * or any access when mandatory locking enabled
736 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
739 * We must ignore files (but only files) which might have mandatory
740 * locks on them because there is no way to know if the accesser has
743 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
750 * Check to see if there are any leases on this file.
751 * This may block while leases are broken.
753 if (!(access
& NFSD_MAY_NOT_BREAK_LEASE
))
754 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? O_WRONLY
: 0));
755 if (host_err
== -EWOULDBLOCK
)
756 host_err
= -ETIMEDOUT
;
757 if (host_err
) /* NOMEM or WOULDBLOCK */
760 if (access
& NFSD_MAY_WRITE
) {
761 if (access
& NFSD_MAY_READ
)
762 flags
= O_RDWR
|O_LARGEFILE
;
764 flags
= O_WRONLY
|O_LARGEFILE
;
766 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
767 flags
, current_cred());
769 host_err
= PTR_ERR(*filp
);
771 host_err
= ima_file_check(*filp
, access
);
773 err
= nfserrno(host_err
);
775 validate_process_creds();
783 nfsd_close(struct file
*filp
)
789 * Obtain the readahead parameters for the file
790 * specified by (dev, ino).
793 static inline struct raparms
*
794 nfsd_get_raparms(dev_t dev
, ino_t ino
)
796 struct raparms
*ra
, **rap
, **frap
= NULL
;
799 struct raparm_hbucket
*rab
;
801 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
802 rab
= &raparm_hash
[hash
];
804 spin_lock(&rab
->pb_lock
);
805 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
806 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
809 if (ra
->p_count
== 0)
812 depth
= nfsdstats
.ra_size
*11/10;
814 spin_unlock(&rab
->pb_lock
);
824 if (rap
!= &rab
->pb_head
) {
826 ra
->p_next
= rab
->pb_head
;
830 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
831 spin_unlock(&rab
->pb_lock
);
836 * Grab and keep cached pages associated with a file in the svc_rqst
837 * so that they can be passed to the network sendmsg/sendpage routines
838 * directly. They will be released after the sending has completed.
841 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
842 struct splice_desc
*sd
)
844 struct svc_rqst
*rqstp
= sd
->u
.data
;
845 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
846 struct page
*page
= buf
->page
;
850 ret
= buf
->ops
->confirm(pipe
, buf
);
856 if (rqstp
->rq_res
.page_len
== 0) {
861 rqstp
->rq_res
.page_base
= buf
->offset
;
862 rqstp
->rq_res
.page_len
= size
;
863 } else if (page
!= pp
[-1]) {
869 rqstp
->rq_res
.page_len
+= size
;
871 rqstp
->rq_res
.page_len
+= size
;
876 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
877 struct splice_desc
*sd
)
879 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
882 static inline int svc_msnfs(struct svc_fh
*ffhp
)
885 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
892 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
893 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
901 inode
= file
->f_path
.dentry
->d_inode
;
903 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
906 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
907 struct splice_desc sd
= {
914 rqstp
->rq_resused
= 1;
915 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
919 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
924 nfsdstats
.io_read
+= host_err
;
927 fsnotify_access(file
);
929 err
= nfserrno(host_err
);
934 static void kill_suid(struct dentry
*dentry
)
937 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
939 mutex_lock(&dentry
->d_inode
->i_mutex
);
940 notify_change(dentry
, &ia
);
941 mutex_unlock(&dentry
->d_inode
->i_mutex
);
945 * Gathered writes: If another process is currently writing to the file,
946 * there's a high chance this is another nfsd (triggered by a bulk write
947 * from a client's biod). Rather than syncing the file with each write
948 * request, we sleep for 10 msec.
950 * I don't know if this roughly approximates C. Juszak's idea of
951 * gathered writes, but it's a nice and simple solution (IMHO), and it
954 * Note: we do this only in the NFSv2 case, since v3 and higher have a
955 * better tool (separate unstable writes and commits) for solving this
958 static int wait_for_concurrent_writes(struct file
*file
)
960 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
961 static ino_t last_ino
;
962 static dev_t last_dev
;
965 if (atomic_read(&inode
->i_writecount
) > 1
966 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
967 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
969 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
972 if (inode
->i_state
& I_DIRTY
) {
973 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
974 err
= vfs_fsync(file
, 0);
976 last_ino
= inode
->i_ino
;
977 last_dev
= inode
->i_sb
->s_dev
;
982 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
983 loff_t offset
, struct kvec
*vec
, int vlen
,
984 unsigned long *cnt
, int *stablep
)
986 struct svc_export
*exp
;
987 struct dentry
*dentry
;
992 int stable
= *stablep
;
998 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
999 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
1003 dentry
= file
->f_path
.dentry
;
1004 inode
= dentry
->d_inode
;
1005 exp
= fhp
->fh_export
;
1008 * Request sync writes if
1009 * - the sync export option has been set, or
1010 * - the client requested O_SYNC behavior (NFSv3 feature).
1011 * - The file system doesn't support fsync().
1012 * When NFSv2 gathered writes have been configured for this volume,
1013 * flushing the data to disk is handled separately below.
1015 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1017 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1019 *stablep
= 2; /* FILE_SYNC */
1022 if (!EX_ISSYNC(exp
))
1024 if (stable
&& !use_wgather
) {
1025 spin_lock(&file
->f_lock
);
1026 file
->f_flags
|= O_SYNC
;
1027 spin_unlock(&file
->f_lock
);
1030 /* Write the data. */
1031 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1032 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1037 nfsdstats
.io_write
+= host_err
;
1038 fsnotify_modify(file
);
1040 /* clear setuid/setgid flag after write */
1041 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1044 if (stable
&& use_wgather
)
1045 host_err
= wait_for_concurrent_writes(file
);
1048 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1052 err
= nfserrno(host_err
);
1058 * Read data from a file. count must contain the requested read count
1059 * on entry. On return, *count contains the number of bytes actually read.
1060 * N.B. After this call fhp needs an fh_put
1062 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1063 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
1066 struct inode
*inode
;
1070 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1074 inode
= file
->f_path
.dentry
->d_inode
;
1076 /* Get readahead parameters */
1077 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
1079 if (ra
&& ra
->p_set
)
1080 file
->f_ra
= ra
->p_ra
;
1082 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1084 /* Write back readahead params */
1086 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
1087 spin_lock(&rab
->pb_lock
);
1088 ra
->p_ra
= file
->f_ra
;
1091 spin_unlock(&rab
->pb_lock
);
1098 /* As above, but use the provided file descriptor. */
1100 nfsd_read_file(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1101 loff_t offset
, struct kvec
*vec
, int vlen
,
1102 unsigned long *count
)
1107 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1108 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1111 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1112 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1113 err
= nfsd_read(rqstp
, fhp
, offset
, vec
, vlen
, count
);
1119 * Write data to a file.
1120 * The stable flag requests synchronous writes.
1121 * N.B. After this call fhp needs an fh_put
1124 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1125 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1131 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1132 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1135 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1138 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1143 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1151 #ifdef CONFIG_NFSD_V3
1153 * Commit all pending writes to stable storage.
1155 * Note: we only guarantee that data that lies within the range specified
1156 * by the 'offset' and 'count' parameters will be synced.
1158 * Unfortunately we cannot lock the file to make sure we return full WCC
1159 * data to the client, as locking happens lower down in the filesystem.
1162 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1163 loff_t offset
, unsigned long count
)
1166 loff_t end
= LLONG_MAX
;
1167 __be32 err
= nfserr_inval
;
1172 end
= offset
+ (loff_t
)count
- 1;
1177 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1178 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1181 if (EX_ISSYNC(fhp
->fh_export
)) {
1182 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1184 if (err2
!= -EINVAL
)
1185 err
= nfserrno(err2
);
1187 err
= nfserr_notsupp
;
1194 #endif /* CONFIG_NFSD_V3 */
1197 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1201 * Mode has already been set earlier in create:
1203 iap
->ia_valid
&= ~ATTR_MODE
;
1205 * Setting uid/gid works only for root. Irix appears to
1206 * send along the gid on create when it tries to implement
1207 * setgid directories via NFS:
1209 if (current_fsuid() != 0)
1210 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1212 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1216 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1217 * setting size to 0 may fail for some specific file systems by the permission
1218 * checking which requires WRITE permission but the mode is 000.
1219 * we ignore the resizing(to 0) on the just new created file, since the size is
1220 * 0 after file created.
1222 * call this only after vfs_create() is called.
1225 nfsd_check_ignore_resizing(struct iattr
*iap
)
1227 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1228 iap
->ia_valid
&= ~ATTR_SIZE
;
1232 * Create a file (regular, directory, device, fifo); UNIX sockets
1233 * not yet implemented.
1234 * If the response fh has been verified, the parent directory should
1235 * already be locked. Note that the parent directory is left locked.
1237 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1240 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1241 char *fname
, int flen
, struct iattr
*iap
,
1242 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1244 struct dentry
*dentry
, *dchild
= NULL
;
1254 if (isdotent(fname
, flen
))
1257 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1261 dentry
= fhp
->fh_dentry
;
1262 dirp
= dentry
->d_inode
;
1264 err
= nfserr_notdir
;
1265 if (!dirp
->i_op
->lookup
)
1268 * Check whether the response file handle has been verified yet.
1269 * If it has, the parent directory should already be locked.
1271 if (!resfhp
->fh_dentry
) {
1272 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1273 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1274 dchild
= lookup_one_len(fname
, dentry
, flen
);
1275 host_err
= PTR_ERR(dchild
);
1278 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1282 /* called from nfsd_proc_create */
1283 dchild
= dget(resfhp
->fh_dentry
);
1284 if (!fhp
->fh_locked
) {
1285 /* not actually possible */
1287 "nfsd_create: parent %s/%s not locked!\n",
1288 dentry
->d_parent
->d_name
.name
,
1289 dentry
->d_name
.name
);
1295 * Make sure the child dentry is still negative ...
1298 if (dchild
->d_inode
) {
1299 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1300 dentry
->d_name
.name
, dchild
->d_name
.name
);
1304 if (!(iap
->ia_valid
& ATTR_MODE
))
1306 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1309 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1310 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1315 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1320 * Get the dir op function pointer.
1325 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1327 nfsd_check_ignore_resizing(iap
);
1330 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1336 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1340 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1344 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1347 * nfsd_setattr already committed the child. Transactional filesystems
1348 * had a chance to commit changes for both parent and child
1349 * simultaneously making the following commit_metadata a noop.
1351 err2
= nfserrno(commit_metadata(fhp
));
1354 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1356 * Update the file handle to get the new inode info.
1359 err
= fh_update(resfhp
);
1361 if (dchild
&& !IS_ERR(dchild
))
1366 err
= nfserrno(host_err
);
1370 #ifdef CONFIG_NFSD_V3
1372 * NFSv3 version of nfsd_create
1375 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1376 char *fname
, int flen
, struct iattr
*iap
,
1377 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1378 int *truncp
, int *created
)
1380 struct dentry
*dentry
, *dchild
= NULL
;
1384 __u32 v_mtime
=0, v_atime
=0;
1390 if (isdotent(fname
, flen
))
1392 if (!(iap
->ia_valid
& ATTR_MODE
))
1394 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1398 dentry
= fhp
->fh_dentry
;
1399 dirp
= dentry
->d_inode
;
1401 /* Get all the sanity checks out of the way before
1402 * we lock the parent. */
1403 err
= nfserr_notdir
;
1404 if (!dirp
->i_op
->lookup
)
1406 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1409 * Compose the response file handle.
1411 dchild
= lookup_one_len(fname
, dentry
, flen
);
1412 host_err
= PTR_ERR(dchild
);
1416 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1420 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1421 /* solaris7 gets confused (bugid 4218508) if these have
1422 * the high bit set, so just clear the high bits. If this is
1423 * ever changed to use different attrs for storing the
1424 * verifier, then do_open_lookup() will also need to be fixed
1427 v_mtime
= verifier
[0]&0x7fffffff;
1428 v_atime
= verifier
[1]&0x7fffffff;
1431 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1434 if (dchild
->d_inode
) {
1437 switch (createmode
) {
1438 case NFS3_CREATE_UNCHECKED
:
1439 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1442 /* in nfsv4, we need to treat this case a little
1443 * differently. we don't want to truncate the
1444 * file now; this would be wrong if the OPEN
1445 * fails for some other reason. furthermore,
1446 * if the size is nonzero, we should ignore it
1447 * according to spec!
1449 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1452 iap
->ia_valid
&= ATTR_SIZE
;
1456 case NFS3_CREATE_EXCLUSIVE
:
1457 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1458 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1459 && dchild
->d_inode
->i_size
== 0 )
1462 case NFS3_CREATE_GUARDED
:
1465 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1469 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1471 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1477 nfsd_check_ignore_resizing(iap
);
1479 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1480 /* Cram the verifier into atime/mtime */
1481 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1482 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1483 /* XXX someone who knows this better please fix it for nsec */
1484 iap
->ia_mtime
.tv_sec
= v_mtime
;
1485 iap
->ia_atime
.tv_sec
= v_atime
;
1486 iap
->ia_mtime
.tv_nsec
= 0;
1487 iap
->ia_atime
.tv_nsec
= 0;
1491 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1494 * nfsd_setattr already committed the child (and possibly also the parent).
1497 err
= nfserrno(commit_metadata(fhp
));
1499 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1501 * Update the filehandle to get the new inode info.
1504 err
= fh_update(resfhp
);
1508 if (dchild
&& !IS_ERR(dchild
))
1513 err
= nfserrno(host_err
);
1516 #endif /* CONFIG_NFSD_V3 */
1519 * Read a symlink. On entry, *lenp must contain the maximum path length that
1520 * fits into the buffer. On return, it contains the true length.
1521 * N.B. After this call fhp needs an fh_put
1524 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1526 struct dentry
*dentry
;
1527 struct inode
*inode
;
1532 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1536 dentry
= fhp
->fh_dentry
;
1537 inode
= dentry
->d_inode
;
1540 if (!inode
->i_op
->readlink
)
1543 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1544 /* N.B. Why does this call need a get_fs()??
1545 * Remove the set_fs and watch the fireworks:-) --okir
1548 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1549 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1560 err
= nfserrno(host_err
);
1565 * Create a symlink and look up its inode
1566 * N.B. After this call _both_ fhp and resfhp need an fh_put
1569 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1570 char *fname
, int flen
,
1571 char *path
, int plen
,
1572 struct svc_fh
*resfhp
,
1575 struct dentry
*dentry
, *dnew
;
1583 if (isdotent(fname
, flen
))
1586 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1590 dentry
= fhp
->fh_dentry
;
1591 dnew
= lookup_one_len(fname
, dentry
, flen
);
1592 host_err
= PTR_ERR(dnew
);
1596 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1600 if (unlikely(path
[plen
] != 0)) {
1601 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1602 if (path_alloced
== NULL
)
1605 strncpy(path_alloced
, path
, plen
);
1606 path_alloced
[plen
] = 0;
1607 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1608 kfree(path_alloced
);
1611 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1612 err
= nfserrno(host_err
);
1614 err
= nfserrno(commit_metadata(fhp
));
1617 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1619 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1621 if (err
==0) err
= cerr
;
1626 err
= nfserrno(host_err
);
1632 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1635 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1636 char *name
, int len
, struct svc_fh
*tfhp
)
1638 struct dentry
*ddir
, *dnew
, *dold
;
1643 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1646 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1654 if (isdotent(name
, len
))
1657 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1658 ddir
= ffhp
->fh_dentry
;
1659 dirp
= ddir
->d_inode
;
1661 dnew
= lookup_one_len(name
, ddir
, len
);
1662 host_err
= PTR_ERR(dnew
);
1666 dold
= tfhp
->fh_dentry
;
1668 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1670 err
= nfserrno(host_err
);
1673 host_err
= vfs_link(dold
, dirp
, dnew
);
1675 err
= nfserrno(commit_metadata(ffhp
));
1677 err
= nfserrno(commit_metadata(tfhp
));
1679 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1682 err
= nfserrno(host_err
);
1684 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1693 err
= nfserrno(host_err
);
1699 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1702 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1703 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1705 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1706 struct inode
*fdir
, *tdir
;
1710 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1713 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1717 fdentry
= ffhp
->fh_dentry
;
1718 fdir
= fdentry
->d_inode
;
1720 tdentry
= tfhp
->fh_dentry
;
1721 tdir
= tdentry
->d_inode
;
1723 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1724 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1728 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1731 /* cannot use fh_lock as we need deadlock protective ordering
1732 * so do it by hand */
1733 trap
= lock_rename(tdentry
, fdentry
);
1734 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1738 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1739 host_err
= PTR_ERR(odentry
);
1740 if (IS_ERR(odentry
))
1744 if (!odentry
->d_inode
)
1747 if (odentry
== trap
)
1750 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1751 host_err
= PTR_ERR(ndentry
);
1752 if (IS_ERR(ndentry
))
1754 host_err
= -ENOTEMPTY
;
1755 if (ndentry
== trap
)
1758 if (svc_msnfs(ffhp
) &&
1759 ((atomic_read(&odentry
->d_count
) > 1)
1760 || (atomic_read(&ndentry
->d_count
) > 1))) {
1766 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1768 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1772 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1774 host_err
= commit_metadata(tfhp
);
1776 host_err
= commit_metadata(ffhp
);
1779 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1786 err
= nfserrno(host_err
);
1788 /* we cannot reply on fh_unlock on the two filehandles,
1789 * as that would do the wrong thing if the two directories
1790 * were the same, so again we do it by hand
1792 fill_post_wcc(ffhp
);
1793 fill_post_wcc(tfhp
);
1794 unlock_rename(tdentry
, fdentry
);
1795 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1802 * Unlink a file or directory
1803 * N.B. After this call fhp needs an fh_put
1806 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1807 char *fname
, int flen
)
1809 struct dentry
*dentry
, *rdentry
;
1815 if (!flen
|| isdotent(fname
, flen
))
1817 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1821 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1822 dentry
= fhp
->fh_dentry
;
1823 dirp
= dentry
->d_inode
;
1825 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1826 host_err
= PTR_ERR(rdentry
);
1827 if (IS_ERR(rdentry
))
1830 if (!rdentry
->d_inode
) {
1837 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1839 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1843 if (type
!= S_IFDIR
) { /* It's UNLINK */
1845 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1846 (atomic_read(&rdentry
->d_count
) > 1)) {
1850 host_err
= vfs_unlink(dirp
, rdentry
);
1851 } else { /* It's RMDIR */
1852 host_err
= vfs_rmdir(dirp
, rdentry
);
1858 host_err
= commit_metadata(fhp
);
1860 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1862 err
= nfserrno(host_err
);
1868 * We do this buffering because we must not call back into the file
1869 * system's ->lookup() method from the filldir callback. That may well
1870 * deadlock a number of file systems.
1872 * This is based heavily on the implementation of same in XFS.
1874 struct buffered_dirent
{
1878 unsigned int d_type
;
1882 struct readdir_data
{
1888 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1889 loff_t offset
, u64 ino
, unsigned int d_type
)
1891 struct readdir_data
*buf
= __buf
;
1892 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1893 unsigned int reclen
;
1895 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1896 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1901 de
->namlen
= namlen
;
1902 de
->offset
= offset
;
1904 de
->d_type
= d_type
;
1905 memcpy(de
->name
, name
, namlen
);
1906 buf
->used
+= reclen
;
1911 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1912 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1914 struct readdir_data buf
;
1915 struct buffered_dirent
*de
;
1920 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1922 return nfserrno(-ENOMEM
);
1927 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1928 unsigned int reclen
;
1930 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1934 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1947 * Various filldir functions may end up calling back into
1948 * lookup_one_len() and the file system's ->lookup() method.
1949 * These expect i_mutex to be held, as it would within readdir.
1951 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1955 de
= (struct buffered_dirent
*)buf
.dirent
;
1957 offset
= de
->offset
;
1959 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1960 de
->ino
, de
->d_type
))
1963 if (cdp
->err
!= nfs_ok
)
1966 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1969 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1971 mutex_unlock(&dir_inode
->i_mutex
);
1972 if (size
> 0) /* We bailed out early */
1975 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1978 free_page((unsigned long)(buf
.dirent
));
1981 return nfserrno(host_err
);
1988 * Read entries from a directory.
1989 * The NFSv3/4 verifier we ignore for now.
1992 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1993 struct readdir_cd
*cdp
, filldir_t func
)
1997 loff_t offset
= *offsetp
;
1999 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
2003 offset
= vfs_llseek(file
, offset
, 0);
2005 err
= nfserrno((int)offset
);
2009 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2011 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2012 err
= nfs_ok
; /* can still be found in ->err */
2020 * Get file system stats
2021 * N.B. After this call fhp needs an fh_put
2024 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2028 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2030 struct path path
= {
2031 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
2032 .dentry
= fhp
->fh_dentry
,
2034 if (vfs_statfs(&path
, stat
))
2040 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2042 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2046 * Check for a user's access permissions to this inode.
2049 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2050 struct dentry
*dentry
, int acc
)
2052 struct inode
*inode
= dentry
->d_inode
;
2055 if (acc
== NFSD_MAY_NOP
)
2058 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2060 (acc
& NFSD_MAY_READ
)? " read" : "",
2061 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2062 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2063 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2064 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2065 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2066 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2068 IS_IMMUTABLE(inode
)? " immut" : "",
2069 IS_APPEND(inode
)? " append" : "",
2070 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2071 dprintk(" owner %d/%d user %d/%d\n",
2072 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2075 /* Normally we reject any write/sattr etc access on a read-only file
2076 * system. But if it is IRIX doing check on write-access for a
2077 * device special file, we ignore rofs.
2079 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2080 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2081 if (exp_rdonly(rqstp
, exp
) ||
2082 __mnt_is_readonly(exp
->ex_path
.mnt
))
2084 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2087 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2090 if (acc
& NFSD_MAY_LOCK
) {
2091 /* If we cannot rely on authentication in NLM requests,
2092 * just allow locks, otherwise require read permission, or
2095 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2098 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2101 * The file owner always gets access permission for accesses that
2102 * would normally be checked at open time. This is to make
2103 * file access work even when the client has done a fchmod(fd, 0).
2105 * However, `cp foo bar' should fail nevertheless when bar is
2106 * readonly. A sensible way to do this might be to reject all
2107 * attempts to truncate a read-only file, because a creat() call
2108 * always implies file truncation.
2109 * ... but this isn't really fair. A process may reasonably call
2110 * ftruncate on an open file descriptor on a file with perm 000.
2111 * We must trust the client to do permission checking - using "ACCESS"
2114 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2115 inode
->i_uid
== current_fsuid())
2118 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2119 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2121 /* Allow read access to binaries even when mode 111 */
2122 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2123 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2124 err
= inode_permission(inode
, MAY_EXEC
);
2126 return err
? nfserrno(err
) : 0;
2130 nfsd_racache_shutdown(void)
2132 struct raparms
*raparm
, *last_raparm
;
2135 dprintk("nfsd: freeing readahead buffers.\n");
2137 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2138 raparm
= raparm_hash
[i
].pb_head
;
2140 last_raparm
= raparm
;
2141 raparm
= raparm
->p_next
;
2144 raparm_hash
[i
].pb_head
= NULL
;
2148 * Initialize readahead param cache
2151 nfsd_racache_init(int cache_size
)
2156 struct raparms
**raparm
= NULL
;
2159 if (raparm_hash
[0].pb_head
)
2161 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2164 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2166 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2168 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2169 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2171 raparm
= &raparm_hash
[i
].pb_head
;
2172 for (j
= 0; j
< nperbucket
; j
++) {
2173 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2176 raparm
= &(*raparm
)->p_next
;
2181 nfsdstats
.ra_size
= cache_size
;
2185 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2186 nfsd_racache_shutdown();
2190 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2192 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2194 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2198 struct posix_acl
*acl
;
2200 if (!IS_POSIXACL(inode
))
2201 return ERR_PTR(-EOPNOTSUPP
);
2204 case ACL_TYPE_ACCESS
:
2205 name
= POSIX_ACL_XATTR_ACCESS
;
2207 case ACL_TYPE_DEFAULT
:
2208 name
= POSIX_ACL_XATTR_DEFAULT
;
2211 return ERR_PTR(-EOPNOTSUPP
);
2214 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2216 return ERR_PTR(size
);
2218 acl
= posix_acl_from_xattr(value
, size
);
2224 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2226 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2232 if (!IS_POSIXACL(inode
) ||
2233 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2236 case ACL_TYPE_ACCESS
:
2237 name
= POSIX_ACL_XATTR_ACCESS
;
2239 case ACL_TYPE_DEFAULT
:
2240 name
= POSIX_ACL_XATTR_DEFAULT
;
2246 if (acl
&& acl
->a_count
) {
2247 size
= posix_acl_xattr_size(acl
->a_count
);
2248 value
= kmalloc(size
, GFP_KERNEL
);
2251 error
= posix_acl_to_xattr(acl
, value
, size
);
2258 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2262 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2264 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2267 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2268 if (error
== -ENODATA
)
2272 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2278 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */