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 nfsd_v4client(struct svc_rqst
*rq
)
95 return rq
->rq_prog
== NFS_PROGRAM
&& rq
->rq_vers
== 4;
99 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
101 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
102 * or nfs_ok having possibly changed *dpp and *expp
105 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
106 struct svc_export
**expp
)
108 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
109 struct dentry
*dentry
= *dpp
;
110 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
111 .dentry
= dget(dentry
)};
114 while (d_mountpoint(path
.dentry
) && follow_down(&path
))
117 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
119 if (PTR_ERR(exp2
) != -ENOENT
)
124 if (nfsd_v4client(rqstp
) ||
125 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
126 /* successfully crossed mount point */
128 * This is subtle: path.dentry is *not* on path.mnt
129 * at this point. The only reason we are safe is that
130 * original mnt is pinned down by exp, so we should
131 * put path *before* putting exp
134 path
.dentry
= dentry
;
145 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
146 const char *name
, unsigned int len
,
147 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
149 struct svc_export
*exp
;
150 struct dentry
*dparent
;
151 struct dentry
*dentry
;
155 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
157 /* Obtain dentry and export. */
158 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
162 dparent
= fhp
->fh_dentry
;
163 exp
= fhp
->fh_export
;
166 /* Lookup the name, but don't follow links */
167 if (isdotent(name
, len
)) {
169 dentry
= dget(dparent
);
170 else if (dparent
!= exp
->ex_path
.dentry
)
171 dentry
= dget_parent(dparent
);
172 else if (!EX_NOHIDE(exp
))
173 dentry
= dget(dparent
); /* .. == . just like at / */
175 /* checking mountpoint crossing is very different when stepping up */
176 struct svc_export
*exp2
= NULL
;
178 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
179 .dentry
= dget(dparent
)};
181 while (path
.dentry
== path
.mnt
->mnt_root
&&
184 dp
= dget_parent(path
.dentry
);
188 exp2
= rqst_exp_parent(rqstp
, &path
);
189 if (PTR_ERR(exp2
) == -ENOENT
) {
190 dentry
= dget(dparent
);
191 } else if (IS_ERR(exp2
)) {
192 host_err
= PTR_ERR(exp2
);
196 dentry
= dget(path
.dentry
);
204 dentry
= lookup_one_len(name
, dparent
, len
);
205 host_err
= PTR_ERR(dentry
);
209 * check if we have crossed a mount point ...
211 if (d_mountpoint(dentry
)) {
212 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
218 *dentry_ret
= dentry
;
224 return nfserrno(host_err
);
228 * Look up one component of a pathname.
229 * N.B. After this call _both_ fhp and resfh need an fh_put
231 * If the lookup would cross a mountpoint, and the mounted filesystem
232 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
233 * accepted as it stands and the mounted directory is
234 * returned. Otherwise the covered directory is returned.
235 * NOTE: this mountpoint crossing is not supported properly by all
236 * clients and is explicitly disallowed for NFSv3
237 * NeilBrown <neilb@cse.unsw.edu.au>
240 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
241 unsigned int len
, struct svc_fh
*resfh
)
243 struct svc_export
*exp
;
244 struct dentry
*dentry
;
247 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
250 err
= check_nfsd_access(exp
, rqstp
);
254 * Note: we compose the file handle now, but as the
255 * dentry may be negative, it may need to be updated.
257 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
258 if (!err
&& !dentry
->d_inode
)
268 * Set various file attributes.
269 * N.B. After this call fhp needs an fh_put
272 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
273 int check_guard
, time_t guardtime
)
275 struct dentry
*dentry
;
277 int accmode
= NFSD_MAY_SATTR
;
283 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
284 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
285 if (iap
->ia_valid
& ATTR_SIZE
)
289 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
293 dentry
= fhp
->fh_dentry
;
294 inode
= dentry
->d_inode
;
296 /* Ignore any mode updates on symlinks */
297 if (S_ISLNK(inode
->i_mode
))
298 iap
->ia_valid
&= ~ATTR_MODE
;
304 * NFSv2 does not differentiate between "set-[ac]time-to-now"
305 * which only requires access, and "set-[ac]time-to-X" which
306 * requires ownership.
307 * So if it looks like it might be "set both to the same time which
308 * is close to now", and if inode_change_ok fails, then we
309 * convert to "set to now" instead of "set to explicit time"
311 * We only call inode_change_ok as the last test as technically
312 * it is not an interface that we should be using. It is only
313 * valid if the filesystem does not define it's own i_op->setattr.
315 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
316 #define MAX_TOUCH_TIME_ERROR (30*60)
317 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
318 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
322 * Now just make sure time is in the right ballpark.
323 * Solaris, at least, doesn't seem to care what the time
324 * request is. We require it be within 30 minutes of now.
326 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
329 if (delta
< MAX_TOUCH_TIME_ERROR
&&
330 inode_change_ok(inode
, iap
) != 0) {
332 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
333 * This will cause notify_change to set these times
336 iap
->ia_valid
&= ~BOTH_TIME_SET
;
341 * The size case is special.
342 * It changes the file as well as the attributes.
344 if (iap
->ia_valid
& ATTR_SIZE
) {
345 if (iap
->ia_size
< inode
->i_size
) {
346 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
347 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
353 * If we are changing the size of the file, then
354 * we need to break all leases.
356 host_err
= break_lease(inode
, FMODE_WRITE
| O_NONBLOCK
);
357 if (host_err
== -EWOULDBLOCK
)
358 host_err
= -ETIMEDOUT
;
359 if (host_err
) /* ENOMEM or EWOULDBLOCK */
362 host_err
= get_write_access(inode
);
367 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
369 put_write_access(inode
);
375 /* sanitize the mode change */
376 if (iap
->ia_valid
& ATTR_MODE
) {
377 iap
->ia_mode
&= S_IALLUGO
;
378 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
381 /* Revoke setuid/setgid on chown */
382 if (!S_ISDIR(inode
->i_mode
) &&
383 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
384 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
385 iap
->ia_valid
|= ATTR_KILL_PRIV
;
386 if (iap
->ia_valid
& ATTR_MODE
) {
387 /* we're setting mode too, just clear the s*id bits */
388 iap
->ia_mode
&= ~S_ISUID
;
389 if (iap
->ia_mode
& S_IXGRP
)
390 iap
->ia_mode
&= ~S_ISGID
;
392 /* set ATTR_KILL_* bits and let VFS handle it */
393 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
397 /* Change the attributes. */
399 iap
->ia_valid
|= ATTR_CTIME
;
401 err
= nfserr_notsync
;
402 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
404 host_err
= notify_change(dentry
, iap
);
405 err
= nfserrno(host_err
);
409 put_write_access(inode
);
411 if (EX_ISSYNC(fhp
->fh_export
))
412 write_inode_now(inode
, 1);
417 err
= nfserrno(host_err
);
421 #if defined(CONFIG_NFSD_V2_ACL) || \
422 defined(CONFIG_NFSD_V3_ACL) || \
423 defined(CONFIG_NFSD_V4)
424 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
429 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
433 *buf
= kmalloc(buflen
, GFP_KERNEL
);
437 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
444 #if defined(CONFIG_NFSD_V4)
446 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
453 buflen
= posix_acl_xattr_size(pacl
->a_count
);
454 buf
= kmalloc(buflen
, GFP_KERNEL
);
459 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
465 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
472 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
473 struct nfs4_acl
*acl
)
477 struct dentry
*dentry
;
479 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
480 unsigned int flags
= 0;
483 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
487 dentry
= fhp
->fh_dentry
;
488 inode
= dentry
->d_inode
;
489 if (S_ISDIR(inode
->i_mode
))
490 flags
= NFS4_ACL_DIR
;
492 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
493 if (host_error
== -EINVAL
) {
494 return nfserr_attrnotsupp
;
495 } else if (host_error
< 0)
498 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
502 if (S_ISDIR(inode
->i_mode
))
503 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
506 posix_acl_release(pacl
);
507 posix_acl_release(dpacl
);
509 if (host_error
== -EOPNOTSUPP
)
510 return nfserr_attrnotsupp
;
512 return nfserrno(host_error
);
515 static struct posix_acl
*
516 _get_posix_acl(struct dentry
*dentry
, char *key
)
519 struct posix_acl
*pacl
= NULL
;
522 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
526 return ERR_PTR(buflen
);
528 pacl
= posix_acl_from_xattr(buf
, buflen
);
534 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
536 struct inode
*inode
= dentry
->d_inode
;
538 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
539 unsigned int flags
= 0;
541 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
542 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
543 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
545 error
= PTR_ERR(pacl
);
550 if (S_ISDIR(inode
->i_mode
)) {
551 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
552 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
554 else if (IS_ERR(dpacl
)) {
555 error
= PTR_ERR(dpacl
);
559 flags
= NFS4_ACL_DIR
;
562 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
564 error
= PTR_ERR(*acl
);
568 posix_acl_release(pacl
);
569 posix_acl_release(dpacl
);
573 #endif /* defined(CONFIG_NFS_V4) */
575 #ifdef CONFIG_NFSD_V3
577 * Check server access rights to a file system object
583 static struct accessmap nfs3_regaccess
[] = {
584 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
585 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
586 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
587 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
592 static struct accessmap nfs3_diraccess
[] = {
593 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
594 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
595 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
596 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
597 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
602 static struct accessmap nfs3_anyaccess
[] = {
603 /* Some clients - Solaris 2.6 at least, make an access call
604 * to the server to check for access for things like /dev/null
605 * (which really, the server doesn't care about). So
606 * We provide simple access checking for them, looking
607 * mainly at mode bits, and we make sure to ignore read-only
610 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
611 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
612 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
613 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
619 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
621 struct accessmap
*map
;
622 struct svc_export
*export
;
623 struct dentry
*dentry
;
624 u32 query
, result
= 0, sresult
= 0;
627 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
631 export
= fhp
->fh_export
;
632 dentry
= fhp
->fh_dentry
;
634 if (S_ISREG(dentry
->d_inode
->i_mode
))
635 map
= nfs3_regaccess
;
636 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
637 map
= nfs3_diraccess
;
639 map
= nfs3_anyaccess
;
643 for (; map
->access
; map
++) {
644 if (map
->access
& query
) {
647 sresult
|= map
->access
;
649 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
652 result
|= map
->access
;
655 /* the following error codes just mean the access was not allowed,
656 * rather than an error occurred */
660 /* simply don't "or" in the access bit. */
670 *supported
= sresult
;
675 #endif /* CONFIG_NFSD_V3 */
680 * Open an existing file or directory.
681 * The access argument indicates the type of open (read/write/lock)
682 * N.B. After this call fhp needs an fh_put
685 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
686 int access
, struct file
**filp
)
688 struct dentry
*dentry
;
690 int flags
= O_RDONLY
|O_LARGEFILE
;
694 validate_process_creds();
697 * If we get here, then the client has already done an "open",
698 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
699 * in case a chmod has now revoked permission.
701 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
705 dentry
= fhp
->fh_dentry
;
706 inode
= dentry
->d_inode
;
708 /* Disallow write access to files with the append-only bit set
709 * or any access when mandatory locking enabled
712 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
715 * We must ignore files (but only files) which might have mandatory
716 * locks on them because there is no way to know if the accesser has
719 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
726 * Check to see if there are any leases on this file.
727 * This may block while leases are broken.
729 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? FMODE_WRITE
: 0));
730 if (host_err
== -EWOULDBLOCK
)
731 host_err
= -ETIMEDOUT
;
732 if (host_err
) /* NOMEM or WOULDBLOCK */
735 if (access
& NFSD_MAY_WRITE
) {
736 if (access
& NFSD_MAY_READ
)
737 flags
= O_RDWR
|O_LARGEFILE
;
739 flags
= O_WRONLY
|O_LARGEFILE
;
743 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
744 flags
, current_cred());
746 host_err
= PTR_ERR(*filp
);
748 ima_counts_get(*filp
);
750 err
= nfserrno(host_err
);
752 validate_process_creds();
760 nfsd_close(struct file
*filp
)
767 * As this calls fsync (not fdatasync) there is no need for a write_inode
770 static inline int nfsd_dosync(struct file
*filp
, struct dentry
*dp
,
771 const struct file_operations
*fop
)
773 struct inode
*inode
= dp
->d_inode
;
774 int (*fsync
) (struct file
*, struct dentry
*, int);
777 err
= filemap_write_and_wait(inode
->i_mapping
);
778 if (err
== 0 && fop
&& (fsync
= fop
->fsync
))
779 err
= fsync(filp
, dp
, 0);
784 nfsd_sync(struct file
*filp
)
787 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
788 dprintk("nfsd: sync file %s\n", filp
->f_path
.dentry
->d_name
.name
);
789 mutex_lock(&inode
->i_mutex
);
790 err
=nfsd_dosync(filp
, filp
->f_path
.dentry
, filp
->f_op
);
791 mutex_unlock(&inode
->i_mutex
);
797 nfsd_sync_dir(struct dentry
*dp
)
799 return nfsd_dosync(NULL
, dp
, dp
->d_inode
->i_fop
);
803 * Obtain the readahead parameters for the file
804 * specified by (dev, ino).
807 static inline struct raparms
*
808 nfsd_get_raparms(dev_t dev
, ino_t ino
)
810 struct raparms
*ra
, **rap
, **frap
= NULL
;
813 struct raparm_hbucket
*rab
;
815 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
816 rab
= &raparm_hash
[hash
];
818 spin_lock(&rab
->pb_lock
);
819 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
820 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
823 if (ra
->p_count
== 0)
826 depth
= nfsdstats
.ra_size
*11/10;
828 spin_unlock(&rab
->pb_lock
);
838 if (rap
!= &rab
->pb_head
) {
840 ra
->p_next
= rab
->pb_head
;
844 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
845 spin_unlock(&rab
->pb_lock
);
850 * Grab and keep cached pages associated with a file in the svc_rqst
851 * so that they can be passed to the network sendmsg/sendpage routines
852 * directly. They will be released after the sending has completed.
855 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
856 struct splice_desc
*sd
)
858 struct svc_rqst
*rqstp
= sd
->u
.data
;
859 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
860 struct page
*page
= buf
->page
;
864 ret
= buf
->ops
->confirm(pipe
, buf
);
870 if (rqstp
->rq_res
.page_len
== 0) {
875 rqstp
->rq_res
.page_base
= buf
->offset
;
876 rqstp
->rq_res
.page_len
= size
;
877 } else if (page
!= pp
[-1]) {
883 rqstp
->rq_res
.page_len
+= size
;
885 rqstp
->rq_res
.page_len
+= size
;
890 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
891 struct splice_desc
*sd
)
893 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
896 static inline int svc_msnfs(struct svc_fh
*ffhp
)
899 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
906 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
907 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
916 inode
= file
->f_path
.dentry
->d_inode
;
918 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
921 /* Get readahead parameters */
922 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
925 file
->f_ra
= ra
->p_ra
;
927 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
928 struct splice_desc sd
= {
935 rqstp
->rq_resused
= 1;
936 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
940 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
944 /* Write back readahead params */
946 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
947 spin_lock(&rab
->pb_lock
);
948 ra
->p_ra
= file
->f_ra
;
951 spin_unlock(&rab
->pb_lock
);
955 nfsdstats
.io_read
+= host_err
;
958 fsnotify_access(file
->f_path
.dentry
);
960 err
= nfserrno(host_err
);
965 static void kill_suid(struct dentry
*dentry
)
968 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
970 mutex_lock(&dentry
->d_inode
->i_mutex
);
971 notify_change(dentry
, &ia
);
972 mutex_unlock(&dentry
->d_inode
->i_mutex
);
976 * Gathered writes: If another process is currently writing to the file,
977 * there's a high chance this is another nfsd (triggered by a bulk write
978 * from a client's biod). Rather than syncing the file with each write
979 * request, we sleep for 10 msec.
981 * I don't know if this roughly approximates C. Juszak's idea of
982 * gathered writes, but it's a nice and simple solution (IMHO), and it
985 * Note: we do this only in the NFSv2 case, since v3 and higher have a
986 * better tool (separate unstable writes and commits) for solving this
989 static int wait_for_concurrent_writes(struct file
*file
)
991 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
992 static ino_t last_ino
;
993 static dev_t last_dev
;
996 if (atomic_read(&inode
->i_writecount
) > 1
997 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
998 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1000 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1003 if (inode
->i_state
& I_DIRTY
) {
1004 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1005 err
= nfsd_sync(file
);
1007 last_ino
= inode
->i_ino
;
1008 last_dev
= inode
->i_sb
->s_dev
;
1013 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1014 loff_t offset
, struct kvec
*vec
, int vlen
,
1015 unsigned long *cnt
, int *stablep
)
1017 struct svc_export
*exp
;
1018 struct dentry
*dentry
;
1019 struct inode
*inode
;
1023 int stable
= *stablep
;
1029 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1030 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
1034 dentry
= file
->f_path
.dentry
;
1035 inode
= dentry
->d_inode
;
1036 exp
= fhp
->fh_export
;
1039 * Request sync writes if
1040 * - the sync export option has been set, or
1041 * - the client requested O_SYNC behavior (NFSv3 feature).
1042 * - The file system doesn't support fsync().
1043 * When NFSv2 gathered writes have been configured for this volume,
1044 * flushing the data to disk is handled separately below.
1046 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1048 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1050 *stablep
= 2; /* FILE_SYNC */
1053 if (!EX_ISSYNC(exp
))
1055 if (stable
&& !use_wgather
) {
1056 spin_lock(&file
->f_lock
);
1057 file
->f_flags
|= O_SYNC
;
1058 spin_unlock(&file
->f_lock
);
1061 /* Write the data. */
1062 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1063 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1068 nfsdstats
.io_write
+= host_err
;
1069 fsnotify_modify(file
->f_path
.dentry
);
1071 /* clear setuid/setgid flag after write */
1072 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1075 if (stable
&& use_wgather
)
1076 host_err
= wait_for_concurrent_writes(file
);
1079 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1083 err
= nfserrno(host_err
);
1089 * Read data from a file. count must contain the requested read count
1090 * on entry. On return, *count contains the number of bytes actually read.
1091 * N.B. After this call fhp needs an fh_put
1094 nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1095 loff_t offset
, struct kvec
*vec
, int vlen
,
1096 unsigned long *count
)
1101 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1102 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1105 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1107 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1110 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1118 * Write data to a file.
1119 * The stable flag requests synchronous writes.
1120 * N.B. After this call fhp needs an fh_put
1123 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1124 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1130 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1131 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1134 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1137 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1142 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1150 #ifdef CONFIG_NFSD_V3
1152 * Commit all pending writes to stable storage.
1153 * Strictly speaking, we could sync just the indicated file region here,
1154 * but there's currently no way we can ask the VFS to do so.
1156 * Unfortunately we cannot lock the file to make sure we return full WCC
1157 * data to the client, as locking happens lower down in the filesystem.
1160 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1161 loff_t offset
, unsigned long count
)
1166 if ((u64
)count
> ~(u64
)offset
)
1167 return nfserr_inval
;
1169 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1172 if (EX_ISSYNC(fhp
->fh_export
)) {
1173 if (file
->f_op
&& file
->f_op
->fsync
) {
1174 err
= nfserrno(nfsd_sync(file
));
1176 err
= nfserr_notsupp
;
1183 #endif /* CONFIG_NFSD_V3 */
1186 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1190 * Mode has already been set earlier in create:
1192 iap
->ia_valid
&= ~ATTR_MODE
;
1194 * Setting uid/gid works only for root. Irix appears to
1195 * send along the gid on create when it tries to implement
1196 * setgid directories via NFS:
1198 if (current_fsuid() != 0)
1199 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1201 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1205 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1206 * setting size to 0 may fail for some specific file systems by the permission
1207 * checking which requires WRITE permission but the mode is 000.
1208 * we ignore the resizing(to 0) on the just new created file, since the size is
1209 * 0 after file created.
1211 * call this only after vfs_create() is called.
1214 nfsd_check_ignore_resizing(struct iattr
*iap
)
1216 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1217 iap
->ia_valid
&= ~ATTR_SIZE
;
1221 * Create a file (regular, directory, device, fifo); UNIX sockets
1222 * not yet implemented.
1223 * If the response fh has been verified, the parent directory should
1224 * already be locked. Note that the parent directory is left locked.
1226 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1229 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1230 char *fname
, int flen
, struct iattr
*iap
,
1231 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1233 struct dentry
*dentry
, *dchild
= NULL
;
1243 if (isdotent(fname
, flen
))
1246 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1250 dentry
= fhp
->fh_dentry
;
1251 dirp
= dentry
->d_inode
;
1253 err
= nfserr_notdir
;
1254 if (!dirp
->i_op
->lookup
)
1257 * Check whether the response file handle has been verified yet.
1258 * If it has, the parent directory should already be locked.
1260 if (!resfhp
->fh_dentry
) {
1261 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1262 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1263 dchild
= lookup_one_len(fname
, dentry
, flen
);
1264 host_err
= PTR_ERR(dchild
);
1267 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1271 /* called from nfsd_proc_create */
1272 dchild
= dget(resfhp
->fh_dentry
);
1273 if (!fhp
->fh_locked
) {
1274 /* not actually possible */
1276 "nfsd_create: parent %s/%s not locked!\n",
1277 dentry
->d_parent
->d_name
.name
,
1278 dentry
->d_name
.name
);
1284 * Make sure the child dentry is still negative ...
1287 if (dchild
->d_inode
) {
1288 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1289 dentry
->d_name
.name
, dchild
->d_name
.name
);
1293 if (!(iap
->ia_valid
& ATTR_MODE
))
1295 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1298 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1299 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1304 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1309 * Get the dir op function pointer.
1314 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1316 nfsd_check_ignore_resizing(iap
);
1319 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1325 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1329 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1333 if (EX_ISSYNC(fhp
->fh_export
)) {
1334 err
= nfserrno(nfsd_sync_dir(dentry
));
1335 write_inode_now(dchild
->d_inode
, 1);
1338 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1341 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1343 * Update the file handle to get the new inode info.
1346 err
= fh_update(resfhp
);
1348 if (dchild
&& !IS_ERR(dchild
))
1353 err
= nfserrno(host_err
);
1357 #ifdef CONFIG_NFSD_V3
1359 * NFSv3 version of nfsd_create
1362 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1363 char *fname
, int flen
, struct iattr
*iap
,
1364 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1365 int *truncp
, int *created
)
1367 struct dentry
*dentry
, *dchild
= NULL
;
1372 __u32 v_mtime
=0, v_atime
=0;
1378 if (isdotent(fname
, flen
))
1380 if (!(iap
->ia_valid
& ATTR_MODE
))
1382 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1386 dentry
= fhp
->fh_dentry
;
1387 dirp
= dentry
->d_inode
;
1389 /* Get all the sanity checks out of the way before
1390 * we lock the parent. */
1391 err
= nfserr_notdir
;
1392 if (!dirp
->i_op
->lookup
)
1394 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1397 * Compose the response file handle.
1399 dchild
= lookup_one_len(fname
, dentry
, flen
);
1400 host_err
= PTR_ERR(dchild
);
1404 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1408 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1409 /* solaris7 gets confused (bugid 4218508) if these have
1410 * the high bit set, so just clear the high bits. If this is
1411 * ever changed to use different attrs for storing the
1412 * verifier, then do_open_lookup() will also need to be fixed
1415 v_mtime
= verifier
[0]&0x7fffffff;
1416 v_atime
= verifier
[1]&0x7fffffff;
1419 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1422 if (dchild
->d_inode
) {
1425 switch (createmode
) {
1426 case NFS3_CREATE_UNCHECKED
:
1427 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1430 /* in nfsv4, we need to treat this case a little
1431 * differently. we don't want to truncate the
1432 * file now; this would be wrong if the OPEN
1433 * fails for some other reason. furthermore,
1434 * if the size is nonzero, we should ignore it
1435 * according to spec!
1437 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1440 iap
->ia_valid
&= ATTR_SIZE
;
1444 case NFS3_CREATE_EXCLUSIVE
:
1445 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1446 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1447 && dchild
->d_inode
->i_size
== 0 )
1450 case NFS3_CREATE_GUARDED
:
1453 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1457 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1459 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1465 if (EX_ISSYNC(fhp
->fh_export
)) {
1466 err
= nfserrno(nfsd_sync_dir(dentry
));
1467 /* setattr will sync the child (or not) */
1470 nfsd_check_ignore_resizing(iap
);
1472 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1473 /* Cram the verifier into atime/mtime */
1474 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1475 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1476 /* XXX someone who knows this better please fix it for nsec */
1477 iap
->ia_mtime
.tv_sec
= v_mtime
;
1478 iap
->ia_atime
.tv_sec
= v_atime
;
1479 iap
->ia_mtime
.tv_nsec
= 0;
1480 iap
->ia_atime
.tv_nsec
= 0;
1484 err2
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1488 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1490 * Update the filehandle to get the new inode info.
1493 err
= fh_update(resfhp
);
1497 if (dchild
&& !IS_ERR(dchild
))
1502 err
= nfserrno(host_err
);
1505 #endif /* CONFIG_NFSD_V3 */
1508 * Read a symlink. On entry, *lenp must contain the maximum path length that
1509 * fits into the buffer. On return, it contains the true length.
1510 * N.B. After this call fhp needs an fh_put
1513 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1515 struct dentry
*dentry
;
1516 struct inode
*inode
;
1521 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1525 dentry
= fhp
->fh_dentry
;
1526 inode
= dentry
->d_inode
;
1529 if (!inode
->i_op
->readlink
)
1532 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1533 /* N.B. Why does this call need a get_fs()??
1534 * Remove the set_fs and watch the fireworks:-) --okir
1537 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1538 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1549 err
= nfserrno(host_err
);
1554 * Create a symlink and look up its inode
1555 * N.B. After this call _both_ fhp and resfhp need an fh_put
1558 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1559 char *fname
, int flen
,
1560 char *path
, int plen
,
1561 struct svc_fh
*resfhp
,
1564 struct dentry
*dentry
, *dnew
;
1572 if (isdotent(fname
, flen
))
1575 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1579 dentry
= fhp
->fh_dentry
;
1580 dnew
= lookup_one_len(fname
, dentry
, flen
);
1581 host_err
= PTR_ERR(dnew
);
1585 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1589 if (unlikely(path
[plen
] != 0)) {
1590 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1591 if (path_alloced
== NULL
)
1594 strncpy(path_alloced
, path
, plen
);
1595 path_alloced
[plen
] = 0;
1596 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1597 kfree(path_alloced
);
1600 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1603 if (EX_ISSYNC(fhp
->fh_export
))
1604 host_err
= nfsd_sync_dir(dentry
);
1606 err
= nfserrno(host_err
);
1609 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1611 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1613 if (err
==0) err
= cerr
;
1618 err
= nfserrno(host_err
);
1624 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1627 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1628 char *name
, int len
, struct svc_fh
*tfhp
)
1630 struct dentry
*ddir
, *dnew
, *dold
;
1631 struct inode
*dirp
, *dest
;
1635 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1638 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1646 if (isdotent(name
, len
))
1649 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1650 ddir
= ffhp
->fh_dentry
;
1651 dirp
= ddir
->d_inode
;
1653 dnew
= lookup_one_len(name
, ddir
, len
);
1654 host_err
= PTR_ERR(dnew
);
1658 dold
= tfhp
->fh_dentry
;
1659 dest
= dold
->d_inode
;
1661 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1663 err
= nfserrno(host_err
);
1666 host_err
= vfs_link(dold
, dirp
, dnew
);
1668 if (EX_ISSYNC(ffhp
->fh_export
)) {
1669 err
= nfserrno(nfsd_sync_dir(ddir
));
1670 write_inode_now(dest
, 1);
1674 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1677 err
= nfserrno(host_err
);
1679 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1688 err
= nfserrno(host_err
);
1694 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1697 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1698 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1700 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1701 struct inode
*fdir
, *tdir
;
1705 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1708 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1712 fdentry
= ffhp
->fh_dentry
;
1713 fdir
= fdentry
->d_inode
;
1715 tdentry
= tfhp
->fh_dentry
;
1716 tdir
= tdentry
->d_inode
;
1718 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1719 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1723 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1726 /* cannot use fh_lock as we need deadlock protective ordering
1727 * so do it by hand */
1728 trap
= lock_rename(tdentry
, fdentry
);
1729 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1733 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1734 host_err
= PTR_ERR(odentry
);
1735 if (IS_ERR(odentry
))
1739 if (!odentry
->d_inode
)
1742 if (odentry
== trap
)
1745 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1746 host_err
= PTR_ERR(ndentry
);
1747 if (IS_ERR(ndentry
))
1749 host_err
= -ENOTEMPTY
;
1750 if (ndentry
== trap
)
1753 if (svc_msnfs(ffhp
) &&
1754 ((atomic_read(&odentry
->d_count
) > 1)
1755 || (atomic_read(&ndentry
->d_count
) > 1))) {
1761 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1763 host_err
= mnt_want_write(ffhp
->fh_export
->ex_path
.mnt
);
1767 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1768 if (!host_err
&& EX_ISSYNC(tfhp
->fh_export
)) {
1769 host_err
= nfsd_sync_dir(tdentry
);
1771 host_err
= nfsd_sync_dir(fdentry
);
1774 mnt_drop_write(ffhp
->fh_export
->ex_path
.mnt
);
1781 err
= nfserrno(host_err
);
1783 /* we cannot reply on fh_unlock on the two filehandles,
1784 * as that would do the wrong thing if the two directories
1785 * were the same, so again we do it by hand
1787 fill_post_wcc(ffhp
);
1788 fill_post_wcc(tfhp
);
1789 unlock_rename(tdentry
, fdentry
);
1790 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1797 * Unlink a file or directory
1798 * N.B. After this call fhp needs an fh_put
1801 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1802 char *fname
, int flen
)
1804 struct dentry
*dentry
, *rdentry
;
1810 if (!flen
|| isdotent(fname
, flen
))
1812 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1816 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1817 dentry
= fhp
->fh_dentry
;
1818 dirp
= dentry
->d_inode
;
1820 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1821 host_err
= PTR_ERR(rdentry
);
1822 if (IS_ERR(rdentry
))
1825 if (!rdentry
->d_inode
) {
1832 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1834 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1838 if (type
!= S_IFDIR
) { /* It's UNLINK */
1840 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
1841 (atomic_read(&rdentry
->d_count
) > 1)) {
1845 host_err
= vfs_unlink(dirp
, rdentry
);
1846 } else { /* It's RMDIR */
1847 host_err
= vfs_rmdir(dirp
, rdentry
);
1854 if (EX_ISSYNC(fhp
->fh_export
))
1855 host_err
= nfsd_sync_dir(dentry
);
1858 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1860 err
= nfserrno(host_err
);
1866 * We do this buffering because we must not call back into the file
1867 * system's ->lookup() method from the filldir callback. That may well
1868 * deadlock a number of file systems.
1870 * This is based heavily on the implementation of same in XFS.
1872 struct buffered_dirent
{
1876 unsigned int d_type
;
1880 struct readdir_data
{
1886 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1887 loff_t offset
, u64 ino
, unsigned int d_type
)
1889 struct readdir_data
*buf
= __buf
;
1890 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1891 unsigned int reclen
;
1893 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1894 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1899 de
->namlen
= namlen
;
1900 de
->offset
= offset
;
1902 de
->d_type
= d_type
;
1903 memcpy(de
->name
, name
, namlen
);
1904 buf
->used
+= reclen
;
1909 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1910 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1912 struct readdir_data buf
;
1913 struct buffered_dirent
*de
;
1918 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1920 return nfserrno(-ENOMEM
);
1925 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1926 unsigned int reclen
;
1928 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1932 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1945 * Various filldir functions may end up calling back into
1946 * lookup_one_len() and the file system's ->lookup() method.
1947 * These expect i_mutex to be held, as it would within readdir.
1949 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1953 de
= (struct buffered_dirent
*)buf
.dirent
;
1955 offset
= de
->offset
;
1957 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1958 de
->ino
, de
->d_type
))
1961 if (cdp
->err
!= nfs_ok
)
1964 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1967 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1969 mutex_unlock(&dir_inode
->i_mutex
);
1970 if (size
> 0) /* We bailed out early */
1973 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1976 free_page((unsigned long)(buf
.dirent
));
1979 return nfserrno(host_err
);
1986 * Read entries from a directory.
1987 * The NFSv3/4 verifier we ignore for now.
1990 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1991 struct readdir_cd
*cdp
, filldir_t func
)
1995 loff_t offset
= *offsetp
;
1997 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
2001 offset
= vfs_llseek(file
, offset
, 0);
2003 err
= nfserrno((int)offset
);
2007 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2009 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2010 err
= nfs_ok
; /* can still be found in ->err */
2018 * Get file system stats
2019 * N.B. After this call fhp needs an fh_put
2022 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2024 __be32 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2025 if (!err
&& vfs_statfs(fhp
->fh_dentry
,stat
))
2030 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2032 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2036 * Check for a user's access permissions to this inode.
2039 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2040 struct dentry
*dentry
, int acc
)
2042 struct inode
*inode
= dentry
->d_inode
;
2046 if (acc
== NFSD_MAY_NOP
)
2049 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2051 (acc
& NFSD_MAY_READ
)? " read" : "",
2052 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2053 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2054 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2055 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2056 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2057 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2059 IS_IMMUTABLE(inode
)? " immut" : "",
2060 IS_APPEND(inode
)? " append" : "",
2061 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2062 dprintk(" owner %d/%d user %d/%d\n",
2063 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2066 /* Normally we reject any write/sattr etc access on a read-only file
2067 * system. But if it is IRIX doing check on write-access for a
2068 * device special file, we ignore rofs.
2070 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2071 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2072 if (exp_rdonly(rqstp
, exp
) ||
2073 __mnt_is_readonly(exp
->ex_path
.mnt
))
2075 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2078 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2081 if (acc
& NFSD_MAY_LOCK
) {
2082 /* If we cannot rely on authentication in NLM requests,
2083 * just allow locks, otherwise require read permission, or
2086 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2089 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2092 * The file owner always gets access permission for accesses that
2093 * would normally be checked at open time. This is to make
2094 * file access work even when the client has done a fchmod(fd, 0).
2096 * However, `cp foo bar' should fail nevertheless when bar is
2097 * readonly. A sensible way to do this might be to reject all
2098 * attempts to truncate a read-only file, because a creat() call
2099 * always implies file truncation.
2100 * ... but this isn't really fair. A process may reasonably call
2101 * ftruncate on an open file descriptor on a file with perm 000.
2102 * We must trust the client to do permission checking - using "ACCESS"
2105 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2106 inode
->i_uid
== current_fsuid())
2109 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2110 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2112 /* Allow read access to binaries even when mode 111 */
2113 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2114 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2115 err
= inode_permission(inode
, MAY_EXEC
);
2119 /* Do integrity (permission) checking now, but defer incrementing
2120 * IMA counts to the actual file open.
2122 path
.mnt
= exp
->ex_path
.mnt
;
2123 path
.dentry
= dentry
;
2124 err
= ima_path_check(&path
, acc
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
),
2127 return err
? nfserrno(err
) : 0;
2131 nfsd_racache_shutdown(void)
2133 struct raparms
*raparm
, *last_raparm
;
2136 dprintk("nfsd: freeing readahead buffers.\n");
2138 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2139 raparm
= raparm_hash
[i
].pb_head
;
2141 last_raparm
= raparm
;
2142 raparm
= raparm
->p_next
;
2145 raparm_hash
[i
].pb_head
= NULL
;
2149 * Initialize readahead param cache
2152 nfsd_racache_init(int cache_size
)
2157 struct raparms
**raparm
= NULL
;
2160 if (raparm_hash
[0].pb_head
)
2162 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2165 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2167 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2169 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2170 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2172 raparm
= &raparm_hash
[i
].pb_head
;
2173 for (j
= 0; j
< nperbucket
; j
++) {
2174 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2177 raparm
= &(*raparm
)->p_next
;
2182 nfsdstats
.ra_size
= cache_size
;
2186 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2187 nfsd_racache_shutdown();
2191 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2193 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2195 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2199 struct posix_acl
*acl
;
2201 if (!IS_POSIXACL(inode
))
2202 return ERR_PTR(-EOPNOTSUPP
);
2205 case ACL_TYPE_ACCESS
:
2206 name
= POSIX_ACL_XATTR_ACCESS
;
2208 case ACL_TYPE_DEFAULT
:
2209 name
= POSIX_ACL_XATTR_DEFAULT
;
2212 return ERR_PTR(-EOPNOTSUPP
);
2215 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2217 return ERR_PTR(size
);
2219 acl
= posix_acl_from_xattr(value
, size
);
2225 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2227 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2233 if (!IS_POSIXACL(inode
) ||
2234 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2237 case ACL_TYPE_ACCESS
:
2238 name
= POSIX_ACL_XATTR_ACCESS
;
2240 case ACL_TYPE_DEFAULT
:
2241 name
= POSIX_ACL_XATTR_DEFAULT
;
2247 if (acl
&& acl
->a_count
) {
2248 size
= posix_acl_xattr_size(acl
->a_count
);
2249 value
= kmalloc(size
, GFP_KERNEL
);
2252 error
= posix_acl_to_xattr(acl
, value
, size
);
2259 error
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
2263 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2265 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2268 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2269 if (error
== -ENODATA
)
2273 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
2279 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */