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 (d_mountpoint(path
.dentry
) && follow_down(&path
))
111 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
113 if (PTR_ERR(exp2
) != -ENOENT
)
118 if ((exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
119 /* successfully crossed mount point */
121 * This is subtle: path.dentry is *not* on path.mnt
122 * at this point. The only reason we are safe is that
123 * original mnt is pinned down by exp, so we should
124 * put path *before* putting exp
127 path
.dentry
= dentry
;
138 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
139 const char *name
, unsigned int len
,
140 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
142 struct svc_export
*exp
;
143 struct dentry
*dparent
;
144 struct dentry
*dentry
;
148 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
150 /* Obtain dentry and export. */
151 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
155 dparent
= fhp
->fh_dentry
;
156 exp
= fhp
->fh_export
;
159 /* Lookup the name, but don't follow links */
160 if (isdotent(name
, len
)) {
162 dentry
= dget(dparent
);
163 else if (dparent
!= exp
->ex_path
.dentry
)
164 dentry
= dget_parent(dparent
);
165 else if (!EX_NOHIDE(exp
))
166 dentry
= dget(dparent
); /* .. == . just like at / */
168 /* checking mountpoint crossing is very different when stepping up */
169 struct svc_export
*exp2
= NULL
;
171 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
172 .dentry
= dget(dparent
)};
174 while (path
.dentry
== path
.mnt
->mnt_root
&&
177 dp
= dget_parent(path
.dentry
);
181 exp2
= rqst_exp_parent(rqstp
, &path
);
182 if (PTR_ERR(exp2
) == -ENOENT
) {
183 dentry
= dget(dparent
);
184 } else if (IS_ERR(exp2
)) {
185 host_err
= PTR_ERR(exp2
);
189 dentry
= dget(path
.dentry
);
197 dentry
= lookup_one_len(name
, dparent
, len
);
198 host_err
= PTR_ERR(dentry
);
202 * check if we have crossed a mount point ...
204 if (d_mountpoint(dentry
)) {
205 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
211 *dentry_ret
= dentry
;
217 return nfserrno(host_err
);
221 * Look up one component of a pathname.
222 * N.B. After this call _both_ fhp and resfh need an fh_put
224 * If the lookup would cross a mountpoint, and the mounted filesystem
225 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
226 * accepted as it stands and the mounted directory is
227 * returned. Otherwise the covered directory is returned.
228 * NOTE: this mountpoint crossing is not supported properly by all
229 * clients and is explicitly disallowed for NFSv3
230 * NeilBrown <neilb@cse.unsw.edu.au>
233 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
234 unsigned int len
, struct svc_fh
*resfh
)
236 struct svc_export
*exp
;
237 struct dentry
*dentry
;
240 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
243 err
= check_nfsd_access(exp
, rqstp
);
247 * Note: we compose the file handle now, but as the
248 * dentry may be negative, it may need to be updated.
250 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
251 if (!err
&& !dentry
->d_inode
)
261 * Set various file attributes.
262 * N.B. After this call fhp needs an fh_put
265 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
266 int check_guard
, time_t guardtime
)
268 struct dentry
*dentry
;
270 int accmode
= NFSD_MAY_SATTR
;
276 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
277 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
278 if (iap
->ia_valid
& ATTR_SIZE
)
282 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
286 dentry
= fhp
->fh_dentry
;
287 inode
= dentry
->d_inode
;
289 /* Ignore any mode updates on symlinks */
290 if (S_ISLNK(inode
->i_mode
))
291 iap
->ia_valid
&= ~ATTR_MODE
;
297 * NFSv2 does not differentiate between "set-[ac]time-to-now"
298 * which only requires access, and "set-[ac]time-to-X" which
299 * requires ownership.
300 * So if it looks like it might be "set both to the same time which
301 * is close to now", and if inode_change_ok fails, then we
302 * convert to "set to now" instead of "set to explicit time"
304 * We only call inode_change_ok as the last test as technically
305 * it is not an interface that we should be using. It is only
306 * valid if the filesystem does not define it's own i_op->setattr.
308 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
309 #define MAX_TOUCH_TIME_ERROR (30*60)
310 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
311 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
315 * Now just make sure time is in the right ballpark.
316 * Solaris, at least, doesn't seem to care what the time
317 * request is. We require it be within 30 minutes of now.
319 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
322 if (delta
< MAX_TOUCH_TIME_ERROR
&&
323 inode_change_ok(inode
, iap
) != 0) {
325 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
326 * This will cause notify_change to set these times
329 iap
->ia_valid
&= ~BOTH_TIME_SET
;
334 * The size case is special.
335 * It changes the file as well as the attributes.
337 if (iap
->ia_valid
& ATTR_SIZE
) {
338 if (iap
->ia_size
< inode
->i_size
) {
339 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
340 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
346 * If we are changing the size of the file, then
347 * we need to break all leases.
349 host_err
= break_lease(inode
, FMODE_WRITE
| O_NONBLOCK
);
350 if (host_err
== -EWOULDBLOCK
)
351 host_err
= -ETIMEDOUT
;
352 if (host_err
) /* ENOMEM or EWOULDBLOCK */
355 host_err
= get_write_access(inode
);
360 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
362 put_write_access(inode
);
368 /* sanitize the mode change */
369 if (iap
->ia_valid
& ATTR_MODE
) {
370 iap
->ia_mode
&= S_IALLUGO
;
371 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
374 /* Revoke setuid/setgid on chown */
375 if (!S_ISDIR(inode
->i_mode
) &&
376 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
377 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
378 iap
->ia_valid
|= ATTR_KILL_PRIV
;
379 if (iap
->ia_valid
& ATTR_MODE
) {
380 /* we're setting mode too, just clear the s*id bits */
381 iap
->ia_mode
&= ~S_ISUID
;
382 if (iap
->ia_mode
& S_IXGRP
)
383 iap
->ia_mode
&= ~S_ISGID
;
385 /* set ATTR_KILL_* bits and let VFS handle it */
386 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
390 /* Change the attributes. */
392 iap
->ia_valid
|= ATTR_CTIME
;
394 err
= nfserr_notsync
;
395 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
397 host_err
= notify_change(dentry
, iap
);
398 err
= nfserrno(host_err
);
402 put_write_access(inode
);
404 if (EX_ISSYNC(fhp
->fh_export
))
405 write_inode_now(inode
, 1);
410 err
= nfserrno(host_err
);
414 #if defined(CONFIG_NFSD_V2_ACL) || \
415 defined(CONFIG_NFSD_V3_ACL) || \
416 defined(CONFIG_NFSD_V4)
417 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
422 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
426 *buf
= kmalloc(buflen
, GFP_KERNEL
);
430 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
437 #if defined(CONFIG_NFSD_V4)
439 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
446 buflen
= posix_acl_xattr_size(pacl
->a_count
);
447 buf
= kmalloc(buflen
, GFP_KERNEL
);
452 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
458 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
465 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
466 struct nfs4_acl
*acl
)
470 struct dentry
*dentry
;
472 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
473 unsigned int flags
= 0;
476 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
480 dentry
= fhp
->fh_dentry
;
481 inode
= dentry
->d_inode
;
482 if (S_ISDIR(inode
->i_mode
))
483 flags
= NFS4_ACL_DIR
;
485 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
486 if (host_error
== -EINVAL
) {
487 return nfserr_attrnotsupp
;
488 } else if (host_error
< 0)
491 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
495 if (S_ISDIR(inode
->i_mode
))
496 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
499 posix_acl_release(pacl
);
500 posix_acl_release(dpacl
);
502 if (host_error
== -EOPNOTSUPP
)
503 return nfserr_attrnotsupp
;
505 return nfserrno(host_error
);
508 static struct posix_acl
*
509 _get_posix_acl(struct dentry
*dentry
, char *key
)
512 struct posix_acl
*pacl
= NULL
;
515 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
519 return ERR_PTR(buflen
);
521 pacl
= posix_acl_from_xattr(buf
, buflen
);
527 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
529 struct inode
*inode
= dentry
->d_inode
;
531 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
532 unsigned int flags
= 0;
534 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
535 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
536 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
538 error
= PTR_ERR(pacl
);
543 if (S_ISDIR(inode
->i_mode
)) {
544 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
545 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
547 else if (IS_ERR(dpacl
)) {
548 error
= PTR_ERR(dpacl
);
552 flags
= NFS4_ACL_DIR
;
555 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
557 error
= PTR_ERR(*acl
);
561 posix_acl_release(pacl
);
562 posix_acl_release(dpacl
);
566 #endif /* defined(CONFIG_NFS_V4) */
568 #ifdef CONFIG_NFSD_V3
570 * Check server access rights to a file system object
576 static struct accessmap nfs3_regaccess
[] = {
577 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
578 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
579 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
580 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
585 static struct accessmap nfs3_diraccess
[] = {
586 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
587 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
588 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
589 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
590 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
595 static struct accessmap nfs3_anyaccess
[] = {
596 /* Some clients - Solaris 2.6 at least, make an access call
597 * to the server to check for access for things like /dev/null
598 * (which really, the server doesn't care about). So
599 * We provide simple access checking for them, looking
600 * mainly at mode bits, and we make sure to ignore read-only
603 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
604 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
605 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
606 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
612 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
614 struct accessmap
*map
;
615 struct svc_export
*export
;
616 struct dentry
*dentry
;
617 u32 query
, result
= 0, sresult
= 0;
620 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
624 export
= fhp
->fh_export
;
625 dentry
= fhp
->fh_dentry
;
627 if (S_ISREG(dentry
->d_inode
->i_mode
))
628 map
= nfs3_regaccess
;
629 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
630 map
= nfs3_diraccess
;
632 map
= nfs3_anyaccess
;
636 for (; map
->access
; map
++) {
637 if (map
->access
& query
) {
640 sresult
|= map
->access
;
642 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
645 result
|= map
->access
;
648 /* the following error codes just mean the access was not allowed,
649 * rather than an error occurred */
653 /* simply don't "or" in the access bit. */
663 *supported
= sresult
;
668 #endif /* CONFIG_NFSD_V3 */
673 * Open an existing file or directory.
674 * The access argument indicates the type of open (read/write/lock)
675 * N.B. After this call fhp needs an fh_put
678 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
679 int access
, struct file
**filp
)
681 const struct cred
*cred
= current_cred();
682 struct dentry
*dentry
;
684 int flags
= O_RDONLY
|O_LARGEFILE
;
689 * If we get here, then the client has already done an "open",
690 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
691 * in case a chmod has now revoked permission.
693 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
697 dentry
= fhp
->fh_dentry
;
698 inode
= dentry
->d_inode
;
700 /* Disallow write access to files with the append-only bit set
701 * or any access when mandatory locking enabled
704 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
707 * We must ignore files (but only files) which might have mandatory
708 * locks on them because there is no way to know if the accesser has
711 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
718 * Check to see if there are any leases on this file.
719 * This may block while leases are broken.
721 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
722 if (host_err
== -EWOULDBLOCK
)
723 host_err
= -ETIMEDOUT
;
724 if (host_err
) /* NOMEM or WOULDBLOCK */
727 if (access
& NFSD_MAY_WRITE
) {
728 if (access
& NFSD_MAY_READ
)
729 flags
= O_RDWR
|O_LARGEFILE
;
731 flags
= O_WRONLY
|O_LARGEFILE
;
735 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
738 host_err
= PTR_ERR(*filp
);
740 ima_counts_get(*filp
);
742 err
= nfserrno(host_err
);
751 nfsd_close(struct file
*filp
)
758 * As this calls fsync (not fdatasync) there is no need for a write_inode
761 static inline int nfsd_dosync(struct file
*filp
, struct dentry
*dp
,
762 const struct file_operations
*fop
)
764 struct inode
*inode
= dp
->d_inode
;
765 int (*fsync
) (struct file
*, struct dentry
*, int);
768 err
= filemap_fdatawrite(inode
->i_mapping
);
769 if (err
== 0 && fop
&& (fsync
= fop
->fsync
))
770 err
= fsync(filp
, dp
, 0);
772 err
= filemap_fdatawait(inode
->i_mapping
);
778 nfsd_sync(struct file
*filp
)
781 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
782 dprintk("nfsd: sync file %s\n", filp
->f_path
.dentry
->d_name
.name
);
783 mutex_lock(&inode
->i_mutex
);
784 err
=nfsd_dosync(filp
, filp
->f_path
.dentry
, filp
->f_op
);
785 mutex_unlock(&inode
->i_mutex
);
791 nfsd_sync_dir(struct dentry
*dp
)
793 return nfsd_dosync(NULL
, dp
, dp
->d_inode
->i_fop
);
797 * Obtain the readahead parameters for the file
798 * specified by (dev, ino).
801 static inline struct raparms
*
802 nfsd_get_raparms(dev_t dev
, ino_t ino
)
804 struct raparms
*ra
, **rap
, **frap
= NULL
;
807 struct raparm_hbucket
*rab
;
809 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
810 rab
= &raparm_hash
[hash
];
812 spin_lock(&rab
->pb_lock
);
813 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
814 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
817 if (ra
->p_count
== 0)
820 depth
= nfsdstats
.ra_size
*11/10;
822 spin_unlock(&rab
->pb_lock
);
832 if (rap
!= &rab
->pb_head
) {
834 ra
->p_next
= rab
->pb_head
;
838 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
839 spin_unlock(&rab
->pb_lock
);
844 * Grab and keep cached pages associated with a file in the svc_rqst
845 * so that they can be passed to the network sendmsg/sendpage routines
846 * directly. They will be released after the sending has completed.
849 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
850 struct splice_desc
*sd
)
852 struct svc_rqst
*rqstp
= sd
->u
.data
;
853 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
854 struct page
*page
= buf
->page
;
858 ret
= buf
->ops
->confirm(pipe
, buf
);
864 if (rqstp
->rq_res
.page_len
== 0) {
869 rqstp
->rq_res
.page_base
= buf
->offset
;
870 rqstp
->rq_res
.page_len
= size
;
871 } else if (page
!= pp
[-1]) {
877 rqstp
->rq_res
.page_len
+= size
;
879 rqstp
->rq_res
.page_len
+= size
;
884 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
885 struct splice_desc
*sd
)
887 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
890 static inline int svc_msnfs(struct svc_fh
*ffhp
)
893 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
900 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
901 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
910 inode
= file
->f_path
.dentry
->d_inode
;
912 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
915 /* Get readahead parameters */
916 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
919 file
->f_ra
= ra
->p_ra
;
921 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
922 struct splice_desc sd
= {
929 rqstp
->rq_resused
= 1;
930 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
934 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
938 /* Write back readahead params */
940 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
941 spin_lock(&rab
->pb_lock
);
942 ra
->p_ra
= file
->f_ra
;
945 spin_unlock(&rab
->pb_lock
);
949 nfsdstats
.io_read
+= host_err
;
952 fsnotify_access(file
->f_path
.dentry
);
954 err
= nfserrno(host_err
);
959 static void kill_suid(struct dentry
*dentry
)
962 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
964 mutex_lock(&dentry
->d_inode
->i_mutex
);
965 notify_change(dentry
, &ia
);
966 mutex_unlock(&dentry
->d_inode
->i_mutex
);
970 * Gathered writes: If another process is currently writing to the file,
971 * there's a high chance this is another nfsd (triggered by a bulk write
972 * from a client's biod). Rather than syncing the file with each write
973 * request, we sleep for 10 msec.
975 * I don't know if this roughly approximates C. Juszak's idea of
976 * gathered writes, but it's a nice and simple solution (IMHO), and it
979 * Note: we do this only in the NFSv2 case, since v3 and higher have a
980 * better tool (separate unstable writes and commits) for solving this
983 static int wait_for_concurrent_writes(struct file
*file
)
985 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
986 static ino_t last_ino
;
987 static dev_t last_dev
;
990 if (atomic_read(&inode
->i_writecount
) > 1
991 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
992 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
994 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
997 if (inode
->i_state
& I_DIRTY
) {
998 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
999 err
= nfsd_sync(file
);
1001 last_ino
= inode
->i_ino
;
1002 last_dev
= inode
->i_sb
->s_dev
;
1007 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1008 loff_t offset
, struct kvec
*vec
, int vlen
,
1009 unsigned long *cnt
, int *stablep
)
1011 struct svc_export
*exp
;
1012 struct dentry
*dentry
;
1013 struct inode
*inode
;
1017 int stable
= *stablep
;
1023 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1024 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
1028 dentry
= file
->f_path
.dentry
;
1029 inode
= dentry
->d_inode
;
1030 exp
= fhp
->fh_export
;
1033 * Request sync writes if
1034 * - the sync export option has been set, or
1035 * - the client requested O_SYNC behavior (NFSv3 feature).
1036 * - The file system doesn't support fsync().
1037 * When NFSv2 gathered writes have been configured for this volume,
1038 * flushing the data to disk is handled separately below.
1040 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1042 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1044 *stablep
= 2; /* FILE_SYNC */
1047 if (!EX_ISSYNC(exp
))
1049 if (stable
&& !use_wgather
) {
1050 spin_lock(&file
->f_lock
);
1051 file
->f_flags
|= O_SYNC
;
1052 spin_unlock(&file
->f_lock
);
1055 /* Write the data. */
1056 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1057 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1062 nfsdstats
.io_write
+= host_err
;
1063 fsnotify_modify(file
->f_path
.dentry
);
1065 /* clear setuid/setgid flag after write */
1066 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1069 if (stable
&& use_wgather
)
1070 host_err
= wait_for_concurrent_writes(file
);
1073 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1077 err
= nfserrno(host_err
);
1083 * Read data from a file. count must contain the requested read count
1084 * on entry. On return, *count contains the number of bytes actually read.
1085 * N.B. After this call fhp needs an fh_put
1088 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1089 loff_t offset
, struct kvec
*vec
, int vlen
,
1090 unsigned long *count
)
1095 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1096 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1099 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1101 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1104 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1112 * Write data to a file.
1113 * The stable flag requests synchronous writes.
1114 * N.B. After this call fhp needs an fh_put
1117 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1118 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1124 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1125 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1128 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1131 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1136 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1144 #ifdef CONFIG_NFSD_V3
1146 * Commit all pending writes to stable storage.
1147 * Strictly speaking, we could sync just the indicated file region here,
1148 * but there's currently no way we can ask the VFS to do so.
1150 * Unfortunately we cannot lock the file to make sure we return full WCC
1151 * data to the client, as locking happens lower down in the filesystem.
1154 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1155 loff_t offset
, unsigned long count
)
1160 if ((u64
)count
> ~(u64
)offset
)
1161 return nfserr_inval
;
1163 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1166 if (EX_ISSYNC(fhp
->fh_export
)) {
1167 if (file
->f_op
&& file
->f_op
->fsync
) {
1168 err
= nfserrno(nfsd_sync(file
));
1170 err
= nfserr_notsupp
;
1177 #endif /* CONFIG_NFSD_V3 */
1180 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1184 * Mode has already been set earlier in create:
1186 iap
->ia_valid
&= ~ATTR_MODE
;
1188 * Setting uid/gid works only for root. Irix appears to
1189 * send along the gid on create when it tries to implement
1190 * setgid directories via NFS:
1192 if (current_fsuid() != 0)
1193 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1195 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1199 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1200 * setting size to 0 may fail for some specific file systems by the permission
1201 * checking which requires WRITE permission but the mode is 000.
1202 * we ignore the resizing(to 0) on the just new created file, since the size is
1203 * 0 after file created.
1205 * call this only after vfs_create() is called.
1208 nfsd_check_ignore_resizing(struct iattr
*iap
)
1210 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1211 iap
->ia_valid
&= ~ATTR_SIZE
;
1215 * Create a file (regular, directory, device, fifo); UNIX sockets
1216 * not yet implemented.
1217 * If the response fh has been verified, the parent directory should
1218 * already be locked. Note that the parent directory is left locked.
1220 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1223 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1224 char *fname
, int flen
, struct iattr
*iap
,
1225 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1227 struct dentry
*dentry
, *dchild
= NULL
;
1237 if (isdotent(fname
, flen
))
1240 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1244 dentry
= fhp
->fh_dentry
;
1245 dirp
= dentry
->d_inode
;
1247 err
= nfserr_notdir
;
1248 if (!dirp
->i_op
->lookup
)
1251 * Check whether the response file handle has been verified yet.
1252 * If it has, the parent directory should already be locked.
1254 if (!resfhp
->fh_dentry
) {
1255 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1256 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1257 dchild
= lookup_one_len(fname
, dentry
, flen
);
1258 host_err
= PTR_ERR(dchild
);
1261 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1265 /* called from nfsd_proc_create */
1266 dchild
= dget(resfhp
->fh_dentry
);
1267 if (!fhp
->fh_locked
) {
1268 /* not actually possible */
1270 "nfsd_create: parent %s/%s not locked!\n",
1271 dentry
->d_parent
->d_name
.name
,
1272 dentry
->d_name
.name
);
1278 * Make sure the child dentry is still negative ...
1281 if (dchild
->d_inode
) {
1282 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1283 dentry
->d_name
.name
, dchild
->d_name
.name
);
1287 if (!(iap
->ia_valid
& ATTR_MODE
))
1289 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1292 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1293 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1298 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1303 * Get the dir op function pointer.
1308 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1310 nfsd_check_ignore_resizing(iap
);
1313 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1319 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1323 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1327 if (EX_ISSYNC(fhp
->fh_export
)) {
1328 err
= nfserrno(nfsd_sync_dir(dentry
));
1329 write_inode_now(dchild
->d_inode
, 1);
1332 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1335 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1337 * Update the file handle to get the new inode info.
1340 err
= fh_update(resfhp
);
1342 if (dchild
&& !IS_ERR(dchild
))
1347 err
= nfserrno(host_err
);
1351 #ifdef CONFIG_NFSD_V3
1353 * NFSv3 version of nfsd_create
1356 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1357 char *fname
, int flen
, struct iattr
*iap
,
1358 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1359 int *truncp
, int *created
)
1361 struct dentry
*dentry
, *dchild
= NULL
;
1366 __u32 v_mtime
=0, v_atime
=0;
1372 if (isdotent(fname
, flen
))
1374 if (!(iap
->ia_valid
& ATTR_MODE
))
1376 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1380 dentry
= fhp
->fh_dentry
;
1381 dirp
= dentry
->d_inode
;
1383 /* Get all the sanity checks out of the way before
1384 * we lock the parent. */
1385 err
= nfserr_notdir
;
1386 if (!dirp
->i_op
->lookup
)
1388 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1391 * Compose the response file handle.
1393 dchild
= lookup_one_len(fname
, dentry
, flen
);
1394 host_err
= PTR_ERR(dchild
);
1398 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1402 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1403 /* solaris7 gets confused (bugid 4218508) if these have
1404 * the high bit set, so just clear the high bits. If this is
1405 * ever changed to use different attrs for storing the
1406 * verifier, then do_open_lookup() will also need to be fixed
1409 v_mtime
= verifier
[0]&0x7fffffff;
1410 v_atime
= verifier
[1]&0x7fffffff;
1413 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1416 if (dchild
->d_inode
) {
1419 switch (createmode
) {
1420 case NFS3_CREATE_UNCHECKED
:
1421 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1424 /* in nfsv4, we need to treat this case a little
1425 * differently. we don't want to truncate the
1426 * file now; this would be wrong if the OPEN
1427 * fails for some other reason. furthermore,
1428 * if the size is nonzero, we should ignore it
1429 * according to spec!
1431 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1434 iap
->ia_valid
&= ATTR_SIZE
;
1438 case NFS3_CREATE_EXCLUSIVE
:
1439 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1440 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1441 && dchild
->d_inode
->i_size
== 0 )
1444 case NFS3_CREATE_GUARDED
:
1447 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1451 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1453 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1459 if (EX_ISSYNC(fhp
->fh_export
)) {
1460 err
= nfserrno(nfsd_sync_dir(dentry
));
1461 /* setattr will sync the child (or not) */
1464 nfsd_check_ignore_resizing(iap
);
1466 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1467 /* Cram the verifier into atime/mtime */
1468 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1469 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1470 /* XXX someone who knows this better please fix it for nsec */
1471 iap
->ia_mtime
.tv_sec
= v_mtime
;
1472 iap
->ia_atime
.tv_sec
= v_atime
;
1473 iap
->ia_mtime
.tv_nsec
= 0;
1474 iap
->ia_atime
.tv_nsec
= 0;
1478 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1482 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1484 * Update the filehandle to get the new inode info.
1487 err
= fh_update(resfhp
);
1491 if (dchild
&& !IS_ERR(dchild
))
1496 err
= nfserrno(host_err
);
1499 #endif /* CONFIG_NFSD_V3 */
1502 * Read a symlink. On entry, *lenp must contain the maximum path length that
1503 * fits into the buffer. On return, it contains the true length.
1504 * N.B. After this call fhp needs an fh_put
1507 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1509 struct dentry
*dentry
;
1510 struct inode
*inode
;
1515 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1519 dentry
= fhp
->fh_dentry
;
1520 inode
= dentry
->d_inode
;
1523 if (!inode
->i_op
->readlink
)
1526 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1527 /* N.B. Why does this call need a get_fs()??
1528 * Remove the set_fs and watch the fireworks:-) --okir
1531 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1532 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1543 err
= nfserrno(host_err
);
1548 * Create a symlink and look up its inode
1549 * N.B. After this call _both_ fhp and resfhp need an fh_put
1552 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1553 char *fname
, int flen
,
1554 char *path
, int plen
,
1555 struct svc_fh
*resfhp
,
1558 struct dentry
*dentry
, *dnew
;
1566 if (isdotent(fname
, flen
))
1569 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1573 dentry
= fhp
->fh_dentry
;
1574 dnew
= lookup_one_len(fname
, dentry
, flen
);
1575 host_err
= PTR_ERR(dnew
);
1579 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1583 if (unlikely(path
[plen
] != 0)) {
1584 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1585 if (path_alloced
== NULL
)
1588 strncpy(path_alloced
, path
, plen
);
1589 path_alloced
[plen
] = 0;
1590 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1591 kfree(path_alloced
);
1594 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1597 if (EX_ISSYNC(fhp
->fh_export
))
1598 host_err
= nfsd_sync_dir(dentry
);
1600 err
= nfserrno(host_err
);
1603 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1605 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1607 if (err
==0) err
= cerr
;
1612 err
= nfserrno(host_err
);
1618 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1621 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1622 char *name
, int len
, struct svc_fh
*tfhp
)
1624 struct dentry
*ddir
, *dnew
, *dold
;
1625 struct inode
*dirp
, *dest
;
1629 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1632 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1640 if (isdotent(name
, len
))
1643 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1644 ddir
= ffhp
->fh_dentry
;
1645 dirp
= ddir
->d_inode
;
1647 dnew
= lookup_one_len(name
, ddir
, len
);
1648 host_err
= PTR_ERR(dnew
);
1652 dold
= tfhp
->fh_dentry
;
1653 dest
= dold
->d_inode
;
1655 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1657 err
= nfserrno(host_err
);
1660 host_err
= vfs_link(dold
, dirp
, dnew
);
1662 if (EX_ISSYNC(ffhp
->fh_export
)) {
1663 err
= nfserrno(nfsd_sync_dir(ddir
));
1664 write_inode_now(dest
, 1);
1668 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1671 err
= nfserrno(host_err
);
1673 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1682 err
= nfserrno(host_err
);
1688 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1691 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1692 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1694 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1695 struct inode
*fdir
, *tdir
;
1699 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1702 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1706 fdentry
= ffhp
->fh_dentry
;
1707 fdir
= fdentry
->d_inode
;
1709 tdentry
= tfhp
->fh_dentry
;
1710 tdir
= tdentry
->d_inode
;
1712 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1713 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1717 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1720 /* cannot use fh_lock as we need deadlock protective ordering
1721 * so do it by hand */
1722 trap
= lock_rename(tdentry
, fdentry
);
1723 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1727 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1728 host_err
= PTR_ERR(odentry
);
1729 if (IS_ERR(odentry
))
1733 if (!odentry
->d_inode
)
1736 if (odentry
== trap
)
1739 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1740 host_err
= PTR_ERR(ndentry
);
1741 if (IS_ERR(ndentry
))
1743 host_err
= -ENOTEMPTY
;
1744 if (ndentry
== trap
)
1747 if (svc_msnfs(ffhp
) &&
1748 ((atomic_read(&odentry
->d_count
) > 1)
1749 || (atomic_read(&ndentry
->d_count
) > 1))) {
1755 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1757 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1761 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1762 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1763 host_err
= nfsd_sync_dir(tdentry
);
1765 host_err
= nfsd_sync_dir(fdentry
);
1768 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1775 err
= nfserrno(host_err
);
1777 /* we cannot reply on fh_unlock on the two filehandles,
1778 * as that would do the wrong thing if the two directories
1779 * were the same, so again we do it by hand
1781 fill_post_wcc(ffhp
);
1782 fill_post_wcc(tfhp
);
1783 unlock_rename(tdentry
, fdentry
);
1784 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1791 * Unlink a file or directory
1792 * N.B. After this call fhp needs an fh_put
1795 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1796 char *fname
, int flen
)
1798 struct dentry
*dentry
, *rdentry
;
1804 if (!flen
|| isdotent(fname
, flen
))
1806 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1810 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1811 dentry
= fhp
->fh_dentry
;
1812 dirp
= dentry
->d_inode
;
1814 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1815 host_err
= PTR_ERR(rdentry
);
1816 if (IS_ERR(rdentry
))
1819 if (!rdentry
->d_inode
) {
1826 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1828 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1832 if (type
!= S_IFDIR
) { /* It's UNLINK */
1834 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1835 (atomic_read(&rdentry
->d_count
) > 1)) {
1839 host_err
= vfs_unlink(dirp
, rdentry
);
1840 } else { /* It's RMDIR */
1841 host_err
= vfs_rmdir(dirp
, rdentry
);
1848 if (EX_ISSYNC(fhp
->fh_export
))
1849 host_err
= nfsd_sync_dir(dentry
);
1852 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1854 err
= nfserrno(host_err
);
1860 * We do this buffering because we must not call back into the file
1861 * system's ->lookup() method from the filldir callback. That may well
1862 * deadlock a number of file systems.
1864 * This is based heavily on the implementation of same in XFS.
1866 struct buffered_dirent
{
1870 unsigned int d_type
;
1874 struct readdir_data
{
1880 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1881 loff_t offset
, u64 ino
, unsigned int d_type
)
1883 struct readdir_data
*buf
= __buf
;
1884 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1885 unsigned int reclen
;
1887 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1888 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1893 de
->namlen
= namlen
;
1894 de
->offset
= offset
;
1896 de
->d_type
= d_type
;
1897 memcpy(de
->name
, name
, namlen
);
1898 buf
->used
+= reclen
;
1903 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1904 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1906 struct readdir_data buf
;
1907 struct buffered_dirent
*de
;
1912 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1914 return nfserrno(-ENOMEM
);
1919 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1920 unsigned int reclen
;
1922 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1926 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1939 * Various filldir functions may end up calling back into
1940 * lookup_one_len() and the file system's ->lookup() method.
1941 * These expect i_mutex to be held, as it would within readdir.
1943 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1947 de
= (struct buffered_dirent
*)buf
.dirent
;
1949 offset
= de
->offset
;
1951 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1952 de
->ino
, de
->d_type
))
1955 if (cdp
->err
!= nfs_ok
)
1958 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1961 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1963 mutex_unlock(&dir_inode
->i_mutex
);
1964 if (size
> 0) /* We bailed out early */
1967 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1970 free_page((unsigned long)(buf
.dirent
));
1973 return nfserrno(host_err
);
1980 * Read entries from a directory.
1981 * The NFSv3/4 verifier we ignore for now.
1984 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1985 struct readdir_cd
*cdp
, filldir_t func
)
1989 loff_t offset
= *offsetp
;
1991 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1995 offset
= vfs_llseek(file
, offset
, 0);
1997 err
= nfserrno((int)offset
);
2001 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2003 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2004 err
= nfs_ok
; /* can still be found in ->err */
2012 * Get file system stats
2013 * N.B. After this call fhp needs an fh_put
2016 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2018 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2019 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
2024 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2026 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2030 * Check for a user's access permissions to this inode.
2033 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2034 struct dentry
*dentry
, int acc
)
2036 struct inode
*inode
= dentry
->d_inode
;
2040 if (acc
== NFSD_MAY_NOP
)
2043 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2045 (acc
& NFSD_MAY_READ
)? " read" : "",
2046 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2047 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2048 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2049 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2050 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2051 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2053 IS_IMMUTABLE(inode
)? " immut" : "",
2054 IS_APPEND(inode
)? " append" : "",
2055 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2056 dprintk(" owner %d/%d user %d/%d\n",
2057 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2060 /* Normally we reject any write/sattr etc access on a read-only file
2061 * system. But if it is IRIX doing check on write-access for a
2062 * device special file, we ignore rofs.
2064 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2065 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2066 if (exp_rdonly(rqstp
, exp
) ||
2067 __mnt_is_readonly(exp
->ex_path
.mnt
))
2069 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2072 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2075 if (acc
& NFSD_MAY_LOCK
) {
2076 /* If we cannot rely on authentication in NLM requests,
2077 * just allow locks, otherwise require read permission, or
2080 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2083 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2086 * The file owner always gets access permission for accesses that
2087 * would normally be checked at open time. This is to make
2088 * file access work even when the client has done a fchmod(fd, 0).
2090 * However, `cp foo bar' should fail nevertheless when bar is
2091 * readonly. A sensible way to do this might be to reject all
2092 * attempts to truncate a read-only file, because a creat() call
2093 * always implies file truncation.
2094 * ... but this isn't really fair. A process may reasonably call
2095 * ftruncate on an open file descriptor on a file with perm 000.
2096 * We must trust the client to do permission checking - using "ACCESS"
2099 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2100 inode
->i_uid
== current_fsuid())
2103 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2104 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2106 /* Allow read access to binaries even when mode 111 */
2107 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2108 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2109 err
= inode_permission(inode
, MAY_EXEC
);
2113 /* Do integrity (permission) checking now, but defer incrementing
2114 * IMA counts to the actual file open.
2116 path
.mnt
= exp
->ex_path
.mnt
;
2117 path
.dentry
= dentry
;
2118 err
= ima_path_check(&path
, acc
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
),
2121 return err
? nfserrno(err
) : 0;
2125 nfsd_racache_shutdown(void)
2127 struct raparms
*raparm
, *last_raparm
;
2130 dprintk("nfsd: freeing readahead buffers.\n");
2132 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2133 raparm
= raparm_hash
[i
].pb_head
;
2135 last_raparm
= raparm
;
2136 raparm
= raparm
->p_next
;
2139 raparm_hash
[i
].pb_head
= NULL
;
2143 * Initialize readahead param cache
2146 nfsd_racache_init(int cache_size
)
2151 struct raparms
**raparm
= NULL
;
2154 if (raparm_hash
[0].pb_head
)
2156 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2159 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2161 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2163 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2164 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2166 raparm
= &raparm_hash
[i
].pb_head
;
2167 for (j
= 0; j
< nperbucket
; j
++) {
2168 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2171 raparm
= &(*raparm
)->p_next
;
2176 nfsdstats
.ra_size
= cache_size
;
2180 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2181 nfsd_racache_shutdown();
2185 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2187 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2189 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2193 struct posix_acl
*acl
;
2195 if (!IS_POSIXACL(inode
))
2196 return ERR_PTR(-EOPNOTSUPP
);
2199 case ACL_TYPE_ACCESS
:
2200 name
= POSIX_ACL_XATTR_ACCESS
;
2202 case ACL_TYPE_DEFAULT
:
2203 name
= POSIX_ACL_XATTR_DEFAULT
;
2206 return ERR_PTR(-EOPNOTSUPP
);
2209 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2211 return ERR_PTR(size
);
2213 acl
= posix_acl_from_xattr(value
, size
);
2219 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2221 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2227 if (!IS_POSIXACL(inode
) ||
2228 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2231 case ACL_TYPE_ACCESS
:
2232 name
= POSIX_ACL_XATTR_ACCESS
;
2234 case ACL_TYPE_DEFAULT
:
2235 name
= POSIX_ACL_XATTR_DEFAULT
;
2241 if (acl
&& acl
->a_count
) {
2242 size
= posix_acl_xattr_size(acl
->a_count
);
2243 value
= kmalloc(size
, GFP_KERNEL
);
2246 error
= posix_acl_to_xattr(acl
, value
, size
);
2253 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2257 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2259 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2262 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2263 if (error
== -ENODATA
)
2267 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2273 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */