1 #define MSNFS /* HACK HACK */
3 * File operations used by nfsd. Some of these have been ripped from
4 * other parts of the kernel because they weren't exported, others
5 * are partial duplicates with added or changed functionality.
7 * Note that several functions dget() the dentry upon which they want
8 * to act, most notably those that create directory entries. Response
9 * dentry's are dput()'d if necessary in the release callback.
10 * So if you notice code paths that apparently fail to dput() the
11 * dentry, don't worry--they have been taken care of.
13 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
14 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
18 #include <linux/file.h>
19 #include <linux/splice.h>
20 #include <linux/fcntl.h>
21 #include <linux/namei.h>
22 #include <linux/delay.h>
23 #include <linux/fsnotify.h>
24 #include <linux/posix_acl_xattr.h>
25 #include <linux/xattr.h>
26 #include <linux/jhash.h>
27 #include <linux/ima.h>
28 #include <linux/slab.h>
29 #include <asm/uaccess.h>
30 #include <linux/exportfs.h>
31 #include <linux/writeback.h>
35 #endif /* CONFIG_NFSD_V3 */
38 #include <linux/nfs4_acl.h>
39 #include <linux/nfsd_idmap.h>
40 #endif /* CONFIG_NFSD_V4 */
45 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
49 * This is a cache of readahead params that help us choose the proper
50 * readahead strategy. Initially, we set all readahead parameters to 0
51 * and let the VFS handle things.
52 * If you increase the number of cached files very much, you'll need to
53 * add a hash table here.
56 struct raparms
*p_next
;
61 struct file_ra_state p_ra
;
62 unsigned int p_hindex
;
65 struct raparm_hbucket
{
66 struct raparms
*pb_head
;
68 } ____cacheline_aligned_in_smp
;
70 #define RAPARM_HASH_BITS 4
71 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
72 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
73 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
76 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
78 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
79 * or nfs_ok having possibly changed *dpp and *expp
82 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
83 struct svc_export
**expp
)
85 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
86 struct dentry
*dentry
= *dpp
;
87 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
88 .dentry
= dget(dentry
)};
91 err
= follow_down(&path
, false);
95 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
99 * We normally allow NFS clients to continue
100 * "underneath" a mountpoint that is not exported.
101 * The exception is V4ROOT, where no traversal is ever
102 * allowed without an explicit export of the new
105 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
110 if (nfsd_v4client(rqstp
) ||
111 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
112 /* successfully crossed mount point */
114 * This is subtle: path.dentry is *not* on path.mnt
115 * at this point. The only reason we are safe is that
116 * original mnt is pinned down by exp, so we should
117 * put path *before* putting exp
120 path
.dentry
= dentry
;
130 static void follow_to_parent(struct path
*path
)
134 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
136 dp
= dget_parent(path
->dentry
);
141 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
143 struct svc_export
*exp2
;
144 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
145 .dentry
= dget(dparent
)};
147 follow_to_parent(&path
);
149 exp2
= rqst_exp_parent(rqstp
, &path
);
150 if (PTR_ERR(exp2
) == -ENOENT
) {
151 *dentryp
= dget(dparent
);
152 } else if (IS_ERR(exp2
)) {
154 return PTR_ERR(exp2
);
156 *dentryp
= dget(path
.dentry
);
165 * For nfsd purposes, we treat V4ROOT exports as though there was an
166 * export at *every* directory.
168 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
170 if (d_mountpoint(dentry
))
172 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
174 return dentry
->d_inode
!= NULL
;
178 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
179 const char *name
, unsigned int len
,
180 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
182 struct svc_export
*exp
;
183 struct dentry
*dparent
;
184 struct dentry
*dentry
;
188 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
190 /* Obtain dentry and export. */
191 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
195 dparent
= fhp
->fh_dentry
;
196 exp
= fhp
->fh_export
;
199 /* Lookup the name, but don't follow links */
200 if (isdotent(name
, len
)) {
202 dentry
= dget(dparent
);
203 else if (dparent
!= exp
->ex_path
.dentry
)
204 dentry
= dget_parent(dparent
);
205 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
206 dentry
= dget(dparent
); /* .. == . just like at / */
208 /* checking mountpoint crossing is very different when stepping up */
209 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
215 dentry
= lookup_one_len(name
, dparent
, len
);
216 host_err
= PTR_ERR(dentry
);
220 * check if we have crossed a mount point ...
222 if (nfsd_mountpoint(dentry
, exp
)) {
223 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
229 *dentry_ret
= dentry
;
235 return nfserrno(host_err
);
239 * Look up one component of a pathname.
240 * N.B. After this call _both_ fhp and resfh need an fh_put
242 * If the lookup would cross a mountpoint, and the mounted filesystem
243 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
244 * accepted as it stands and the mounted directory is
245 * returned. Otherwise the covered directory is returned.
246 * NOTE: this mountpoint crossing is not supported properly by all
247 * clients and is explicitly disallowed for NFSv3
248 * NeilBrown <neilb@cse.unsw.edu.au>
251 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
252 unsigned int len
, struct svc_fh
*resfh
)
254 struct svc_export
*exp
;
255 struct dentry
*dentry
;
258 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
261 err
= check_nfsd_access(exp
, rqstp
);
265 * Note: we compose the file handle now, but as the
266 * dentry may be negative, it may need to be updated.
268 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
269 if (!err
&& !dentry
->d_inode
)
278 * Commit metadata changes to stable storage.
281 commit_metadata(struct svc_fh
*fhp
)
283 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
284 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
286 if (!EX_ISSYNC(fhp
->fh_export
))
289 if (export_ops
->commit_metadata
)
290 return export_ops
->commit_metadata(inode
);
291 return sync_inode_metadata(inode
, 1);
295 * Set various file attributes.
296 * N.B. After this call fhp needs an fh_put
299 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
300 int check_guard
, time_t guardtime
)
302 struct dentry
*dentry
;
304 int accmode
= NFSD_MAY_SATTR
;
310 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
311 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
312 if (iap
->ia_valid
& ATTR_SIZE
)
316 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
320 dentry
= fhp
->fh_dentry
;
321 inode
= dentry
->d_inode
;
323 /* Ignore any mode updates on symlinks */
324 if (S_ISLNK(inode
->i_mode
))
325 iap
->ia_valid
&= ~ATTR_MODE
;
331 * NFSv2 does not differentiate between "set-[ac]time-to-now"
332 * which only requires access, and "set-[ac]time-to-X" which
333 * requires ownership.
334 * So if it looks like it might be "set both to the same time which
335 * is close to now", and if inode_change_ok fails, then we
336 * convert to "set to now" instead of "set to explicit time"
338 * We only call inode_change_ok as the last test as technically
339 * it is not an interface that we should be using. It is only
340 * valid if the filesystem does not define it's own i_op->setattr.
342 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
343 #define MAX_TOUCH_TIME_ERROR (30*60)
344 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
345 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
349 * Now just make sure time is in the right ballpark.
350 * Solaris, at least, doesn't seem to care what the time
351 * request is. We require it be within 30 minutes of now.
353 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
356 if (delta
< MAX_TOUCH_TIME_ERROR
&&
357 inode_change_ok(inode
, iap
) != 0) {
359 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
360 * This will cause notify_change to set these times
363 iap
->ia_valid
&= ~BOTH_TIME_SET
;
368 * The size case is special.
369 * It changes the file as well as the attributes.
371 if (iap
->ia_valid
& ATTR_SIZE
) {
372 if (iap
->ia_size
< inode
->i_size
) {
373 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
374 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
380 * If we are changing the size of the file, then
381 * we need to break all leases.
383 host_err
= break_lease(inode
, O_WRONLY
| O_NONBLOCK
);
384 if (host_err
== -EWOULDBLOCK
)
385 host_err
= -ETIMEDOUT
;
386 if (host_err
) /* ENOMEM or EWOULDBLOCK */
389 host_err
= get_write_access(inode
);
394 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
396 put_write_access(inode
);
401 /* sanitize the mode change */
402 if (iap
->ia_valid
& ATTR_MODE
) {
403 iap
->ia_mode
&= S_IALLUGO
;
404 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
407 /* Revoke setuid/setgid on chown */
408 if (!S_ISDIR(inode
->i_mode
) &&
409 (((iap
->ia_valid
& ATTR_UID
) && iap
->ia_uid
!= inode
->i_uid
) ||
410 ((iap
->ia_valid
& ATTR_GID
) && iap
->ia_gid
!= inode
->i_gid
))) {
411 iap
->ia_valid
|= ATTR_KILL_PRIV
;
412 if (iap
->ia_valid
& ATTR_MODE
) {
413 /* we're setting mode too, just clear the s*id bits */
414 iap
->ia_mode
&= ~S_ISUID
;
415 if (iap
->ia_mode
& S_IXGRP
)
416 iap
->ia_mode
&= ~S_ISGID
;
418 /* set ATTR_KILL_* bits and let VFS handle it */
419 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
423 /* Change the attributes. */
425 iap
->ia_valid
|= ATTR_CTIME
;
427 err
= nfserr_notsync
;
428 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
430 host_err
= notify_change(dentry
, iap
);
431 err
= nfserrno(host_err
);
435 put_write_access(inode
);
437 commit_metadata(fhp
);
442 err
= nfserrno(host_err
);
446 #if defined(CONFIG_NFSD_V2_ACL) || \
447 defined(CONFIG_NFSD_V3_ACL) || \
448 defined(CONFIG_NFSD_V4)
449 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
454 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
458 *buf
= kmalloc(buflen
, GFP_KERNEL
);
462 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
469 #if defined(CONFIG_NFSD_V4)
471 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
478 buflen
= posix_acl_xattr_size(pacl
->a_count
);
479 buf
= kmalloc(buflen
, GFP_KERNEL
);
484 len
= posix_acl_to_xattr(pacl
, buf
, buflen
);
490 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
497 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
498 struct nfs4_acl
*acl
)
502 struct dentry
*dentry
;
504 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
505 unsigned int flags
= 0;
508 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
512 dentry
= fhp
->fh_dentry
;
513 inode
= dentry
->d_inode
;
514 if (S_ISDIR(inode
->i_mode
))
515 flags
= NFS4_ACL_DIR
;
517 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
518 if (host_error
== -EINVAL
) {
519 return nfserr_attrnotsupp
;
520 } else if (host_error
< 0)
523 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
527 if (S_ISDIR(inode
->i_mode
))
528 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
531 posix_acl_release(pacl
);
532 posix_acl_release(dpacl
);
534 if (host_error
== -EOPNOTSUPP
)
535 return nfserr_attrnotsupp
;
537 return nfserrno(host_error
);
540 static struct posix_acl
*
541 _get_posix_acl(struct dentry
*dentry
, char *key
)
544 struct posix_acl
*pacl
= NULL
;
547 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
551 return ERR_PTR(buflen
);
553 pacl
= posix_acl_from_xattr(buf
, buflen
);
559 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
561 struct inode
*inode
= dentry
->d_inode
;
563 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
564 unsigned int flags
= 0;
566 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
567 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
568 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
570 error
= PTR_ERR(pacl
);
575 if (S_ISDIR(inode
->i_mode
)) {
576 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
577 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
579 else if (IS_ERR(dpacl
)) {
580 error
= PTR_ERR(dpacl
);
584 flags
= NFS4_ACL_DIR
;
587 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
589 error
= PTR_ERR(*acl
);
593 posix_acl_release(pacl
);
594 posix_acl_release(dpacl
);
598 #endif /* defined(CONFIG_NFSD_V4) */
600 #ifdef CONFIG_NFSD_V3
602 * Check server access rights to a file system object
608 static struct accessmap nfs3_regaccess
[] = {
609 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
610 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
611 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
612 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
617 static struct accessmap nfs3_diraccess
[] = {
618 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
619 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
620 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
621 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
622 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
627 static struct accessmap nfs3_anyaccess
[] = {
628 /* Some clients - Solaris 2.6 at least, make an access call
629 * to the server to check for access for things like /dev/null
630 * (which really, the server doesn't care about). So
631 * We provide simple access checking for them, looking
632 * mainly at mode bits, and we make sure to ignore read-only
635 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
636 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
637 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
638 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
644 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
646 struct accessmap
*map
;
647 struct svc_export
*export
;
648 struct dentry
*dentry
;
649 u32 query
, result
= 0, sresult
= 0;
652 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
656 export
= fhp
->fh_export
;
657 dentry
= fhp
->fh_dentry
;
659 if (S_ISREG(dentry
->d_inode
->i_mode
))
660 map
= nfs3_regaccess
;
661 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
662 map
= nfs3_diraccess
;
664 map
= nfs3_anyaccess
;
668 for (; map
->access
; map
++) {
669 if (map
->access
& query
) {
672 sresult
|= map
->access
;
674 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
677 result
|= map
->access
;
680 /* the following error codes just mean the access was not allowed,
681 * rather than an error occurred */
685 /* simply don't "or" in the access bit. */
695 *supported
= sresult
;
700 #endif /* CONFIG_NFSD_V3 */
705 * Open an existing file or directory.
706 * The access argument indicates the type of open (read/write/lock)
707 * N.B. After this call fhp needs an fh_put
710 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
711 int access
, struct file
**filp
)
713 struct dentry
*dentry
;
715 int flags
= O_RDONLY
|O_LARGEFILE
;
719 validate_process_creds();
722 * If we get here, then the client has already done an "open",
723 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
724 * in case a chmod has now revoked permission.
726 err
= fh_verify(rqstp
, fhp
, type
, access
| NFSD_MAY_OWNER_OVERRIDE
);
730 dentry
= fhp
->fh_dentry
;
731 inode
= dentry
->d_inode
;
733 /* Disallow write access to files with the append-only bit set
734 * or any access when mandatory locking enabled
737 if (IS_APPEND(inode
) && (access
& NFSD_MAY_WRITE
))
740 * We must ignore files (but only files) which might have mandatory
741 * locks on them because there is no way to know if the accesser has
744 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
751 * Check to see if there are any leases on this file.
752 * This may block while leases are broken.
754 if (!(access
& NFSD_MAY_NOT_BREAK_LEASE
))
755 host_err
= break_lease(inode
, O_NONBLOCK
| ((access
& NFSD_MAY_WRITE
) ? O_WRONLY
: 0));
756 if (host_err
== -EWOULDBLOCK
)
757 host_err
= -ETIMEDOUT
;
758 if (host_err
) /* NOMEM or WOULDBLOCK */
761 if (access
& NFSD_MAY_WRITE
) {
762 if (access
& NFSD_MAY_READ
)
763 flags
= O_RDWR
|O_LARGEFILE
;
765 flags
= O_WRONLY
|O_LARGEFILE
;
767 *filp
= dentry_open(dget(dentry
), mntget(fhp
->fh_export
->ex_path
.mnt
),
768 flags
, current_cred());
770 host_err
= PTR_ERR(*filp
);
772 host_err
= ima_file_check(*filp
, access
);
774 err
= nfserrno(host_err
);
776 validate_process_creds();
784 nfsd_close(struct file
*filp
)
790 * Obtain the readahead parameters for the file
791 * specified by (dev, ino).
794 static inline struct raparms
*
795 nfsd_get_raparms(dev_t dev
, ino_t ino
)
797 struct raparms
*ra
, **rap
, **frap
= NULL
;
800 struct raparm_hbucket
*rab
;
802 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
803 rab
= &raparm_hash
[hash
];
805 spin_lock(&rab
->pb_lock
);
806 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
807 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
810 if (ra
->p_count
== 0)
813 depth
= nfsdstats
.ra_size
*11/10;
815 spin_unlock(&rab
->pb_lock
);
825 if (rap
!= &rab
->pb_head
) {
827 ra
->p_next
= rab
->pb_head
;
831 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
832 spin_unlock(&rab
->pb_lock
);
837 * Grab and keep cached pages associated with a file in the svc_rqst
838 * so that they can be passed to the network sendmsg/sendpage routines
839 * directly. They will be released after the sending has completed.
842 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
843 struct splice_desc
*sd
)
845 struct svc_rqst
*rqstp
= sd
->u
.data
;
846 struct page
**pp
= rqstp
->rq_respages
+ rqstp
->rq_resused
;
847 struct page
*page
= buf
->page
;
852 if (rqstp
->rq_res
.page_len
== 0) {
857 rqstp
->rq_res
.page_base
= buf
->offset
;
858 rqstp
->rq_res
.page_len
= size
;
859 } else if (page
!= pp
[-1]) {
865 rqstp
->rq_res
.page_len
+= size
;
867 rqstp
->rq_res
.page_len
+= size
;
872 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
873 struct splice_desc
*sd
)
875 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
878 static inline int svc_msnfs(struct svc_fh
*ffhp
)
881 return (ffhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
);
888 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
889 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
897 inode
= file
->f_path
.dentry
->d_inode
;
899 if (svc_msnfs(fhp
) && !lock_may_read(inode
, offset
, *count
))
902 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
903 struct splice_desc sd
= {
910 rqstp
->rq_resused
= 1;
911 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
915 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
920 nfsdstats
.io_read
+= host_err
;
923 fsnotify_access(file
);
925 err
= nfserrno(host_err
);
930 static void kill_suid(struct dentry
*dentry
)
933 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
935 mutex_lock(&dentry
->d_inode
->i_mutex
);
936 notify_change(dentry
, &ia
);
937 mutex_unlock(&dentry
->d_inode
->i_mutex
);
941 * Gathered writes: If another process is currently writing to the file,
942 * there's a high chance this is another nfsd (triggered by a bulk write
943 * from a client's biod). Rather than syncing the file with each write
944 * request, we sleep for 10 msec.
946 * I don't know if this roughly approximates C. Juszak's idea of
947 * gathered writes, but it's a nice and simple solution (IMHO), and it
950 * Note: we do this only in the NFSv2 case, since v3 and higher have a
951 * better tool (separate unstable writes and commits) for solving this
954 static int wait_for_concurrent_writes(struct file
*file
)
956 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
957 static ino_t last_ino
;
958 static dev_t last_dev
;
961 if (atomic_read(&inode
->i_writecount
) > 1
962 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
963 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
965 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
968 if (inode
->i_state
& I_DIRTY
) {
969 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
970 err
= vfs_fsync(file
, 0);
972 last_ino
= inode
->i_ino
;
973 last_dev
= inode
->i_sb
->s_dev
;
978 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
979 loff_t offset
, struct kvec
*vec
, int vlen
,
980 unsigned long *cnt
, int *stablep
)
982 struct svc_export
*exp
;
983 struct dentry
*dentry
;
988 int stable
= *stablep
;
994 if ((fhp
->fh_export
->ex_flags
& NFSEXP_MSNFS
) &&
995 (!lock_may_write(file
->f_path
.dentry
->d_inode
, offset
, *cnt
)))
999 dentry
= file
->f_path
.dentry
;
1000 inode
= dentry
->d_inode
;
1001 exp
= fhp
->fh_export
;
1004 * Request sync writes if
1005 * - the sync export option has been set, or
1006 * - the client requested O_SYNC behavior (NFSv3 feature).
1007 * - The file system doesn't support fsync().
1008 * When NFSv2 gathered writes have been configured for this volume,
1009 * flushing the data to disk is handled separately below.
1011 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1013 if (!file
->f_op
->fsync
) {/* COMMIT3 cannot work */
1015 *stablep
= 2; /* FILE_SYNC */
1018 if (!EX_ISSYNC(exp
))
1020 if (stable
&& !use_wgather
) {
1021 spin_lock(&file
->f_lock
);
1022 file
->f_flags
|= O_SYNC
;
1023 spin_unlock(&file
->f_lock
);
1026 /* Write the data. */
1027 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1028 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
1033 nfsdstats
.io_write
+= host_err
;
1034 fsnotify_modify(file
);
1036 /* clear setuid/setgid flag after write */
1037 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1040 if (stable
&& use_wgather
)
1041 host_err
= wait_for_concurrent_writes(file
);
1044 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1048 err
= nfserrno(host_err
);
1054 * Read data from a file. count must contain the requested read count
1055 * on entry. On return, *count contains the number of bytes actually read.
1056 * N.B. After this call fhp needs an fh_put
1058 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1059 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
1062 struct inode
*inode
;
1066 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1070 inode
= file
->f_path
.dentry
->d_inode
;
1072 /* Get readahead parameters */
1073 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
1075 if (ra
&& ra
->p_set
)
1076 file
->f_ra
= ra
->p_ra
;
1078 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1080 /* Write back readahead params */
1082 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
1083 spin_lock(&rab
->pb_lock
);
1084 ra
->p_ra
= file
->f_ra
;
1087 spin_unlock(&rab
->pb_lock
);
1094 /* As above, but use the provided file descriptor. */
1096 nfsd_read_file(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1097 loff_t offset
, struct kvec
*vec
, int vlen
,
1098 unsigned long *count
)
1103 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1104 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1107 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1108 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1109 err
= nfsd_read(rqstp
, fhp
, offset
, vec
, vlen
, count
);
1115 * Write data to a file.
1116 * The stable flag requests synchronous writes.
1117 * N.B. After this call fhp needs an fh_put
1120 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1121 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1127 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1128 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1131 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1134 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1139 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1147 #ifdef CONFIG_NFSD_V3
1149 * Commit all pending writes to stable storage.
1151 * Note: we only guarantee that data that lies within the range specified
1152 * by the 'offset' and 'count' parameters will be synced.
1154 * Unfortunately we cannot lock the file to make sure we return full WCC
1155 * data to the client, as locking happens lower down in the filesystem.
1158 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1159 loff_t offset
, unsigned long count
)
1162 loff_t end
= LLONG_MAX
;
1163 __be32 err
= nfserr_inval
;
1168 end
= offset
+ (loff_t
)count
- 1;
1173 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1174 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1177 if (EX_ISSYNC(fhp
->fh_export
)) {
1178 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1180 if (err2
!= -EINVAL
)
1181 err
= nfserrno(err2
);
1183 err
= nfserr_notsupp
;
1190 #endif /* CONFIG_NFSD_V3 */
1193 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1197 * Mode has already been set earlier in create:
1199 iap
->ia_valid
&= ~ATTR_MODE
;
1201 * Setting uid/gid works only for root. Irix appears to
1202 * send along the gid on create when it tries to implement
1203 * setgid directories via NFS:
1205 if (current_fsuid() != 0)
1206 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1208 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1212 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1213 * setting size to 0 may fail for some specific file systems by the permission
1214 * checking which requires WRITE permission but the mode is 000.
1215 * we ignore the resizing(to 0) on the just new created file, since the size is
1216 * 0 after file created.
1218 * call this only after vfs_create() is called.
1221 nfsd_check_ignore_resizing(struct iattr
*iap
)
1223 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1224 iap
->ia_valid
&= ~ATTR_SIZE
;
1228 * Create a file (regular, directory, device, fifo); UNIX sockets
1229 * not yet implemented.
1230 * If the response fh has been verified, the parent directory should
1231 * already be locked. Note that the parent directory is left locked.
1233 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1236 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1237 char *fname
, int flen
, struct iattr
*iap
,
1238 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1240 struct dentry
*dentry
, *dchild
= NULL
;
1250 if (isdotent(fname
, flen
))
1253 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1257 dentry
= fhp
->fh_dentry
;
1258 dirp
= dentry
->d_inode
;
1260 err
= nfserr_notdir
;
1261 if (!dirp
->i_op
->lookup
)
1264 * Check whether the response file handle has been verified yet.
1265 * If it has, the parent directory should already be locked.
1267 if (!resfhp
->fh_dentry
) {
1268 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1269 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1270 dchild
= lookup_one_len(fname
, dentry
, flen
);
1271 host_err
= PTR_ERR(dchild
);
1274 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1278 /* called from nfsd_proc_create */
1279 dchild
= dget(resfhp
->fh_dentry
);
1280 if (!fhp
->fh_locked
) {
1281 /* not actually possible */
1283 "nfsd_create: parent %s/%s not locked!\n",
1284 dentry
->d_parent
->d_name
.name
,
1285 dentry
->d_name
.name
);
1291 * Make sure the child dentry is still negative ...
1294 if (dchild
->d_inode
) {
1295 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1296 dentry
->d_name
.name
, dchild
->d_name
.name
);
1300 if (!(iap
->ia_valid
& ATTR_MODE
))
1302 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1305 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1306 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1311 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1316 * Get the dir op function pointer.
1321 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1323 nfsd_check_ignore_resizing(iap
);
1326 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1332 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1336 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1340 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1343 * nfsd_setattr already committed the child. Transactional filesystems
1344 * had a chance to commit changes for both parent and child
1345 * simultaneously making the following commit_metadata a noop.
1347 err2
= nfserrno(commit_metadata(fhp
));
1350 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1352 * Update the file handle to get the new inode info.
1355 err
= fh_update(resfhp
);
1357 if (dchild
&& !IS_ERR(dchild
))
1362 err
= nfserrno(host_err
);
1366 #ifdef CONFIG_NFSD_V3
1368 * NFSv3 version of nfsd_create
1371 nfsd_create_v3(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1372 char *fname
, int flen
, struct iattr
*iap
,
1373 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1374 int *truncp
, int *created
)
1376 struct dentry
*dentry
, *dchild
= NULL
;
1380 __u32 v_mtime
=0, v_atime
=0;
1386 if (isdotent(fname
, flen
))
1388 if (!(iap
->ia_valid
& ATTR_MODE
))
1390 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1394 dentry
= fhp
->fh_dentry
;
1395 dirp
= dentry
->d_inode
;
1397 /* Get all the sanity checks out of the way before
1398 * we lock the parent. */
1399 err
= nfserr_notdir
;
1400 if (!dirp
->i_op
->lookup
)
1402 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1405 * Compose the response file handle.
1407 dchild
= lookup_one_len(fname
, dentry
, flen
);
1408 host_err
= PTR_ERR(dchild
);
1412 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1416 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1417 /* solaris7 gets confused (bugid 4218508) if these have
1418 * the high bit set, so just clear the high bits. If this is
1419 * ever changed to use different attrs for storing the
1420 * verifier, then do_open_lookup() will also need to be fixed
1423 v_mtime
= verifier
[0]&0x7fffffff;
1424 v_atime
= verifier
[1]&0x7fffffff;
1427 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1430 if (dchild
->d_inode
) {
1433 switch (createmode
) {
1434 case NFS3_CREATE_UNCHECKED
:
1435 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1438 /* in nfsv4, we need to treat this case a little
1439 * differently. we don't want to truncate the
1440 * file now; this would be wrong if the OPEN
1441 * fails for some other reason. furthermore,
1442 * if the size is nonzero, we should ignore it
1443 * according to spec!
1445 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1448 iap
->ia_valid
&= ATTR_SIZE
;
1452 case NFS3_CREATE_EXCLUSIVE
:
1453 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1454 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1455 && dchild
->d_inode
->i_size
== 0 )
1458 case NFS3_CREATE_GUARDED
:
1461 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1465 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, NULL
);
1467 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1473 nfsd_check_ignore_resizing(iap
);
1475 if (createmode
== NFS3_CREATE_EXCLUSIVE
) {
1476 /* Cram the verifier into atime/mtime */
1477 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1478 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1479 /* XXX someone who knows this better please fix it for nsec */
1480 iap
->ia_mtime
.tv_sec
= v_mtime
;
1481 iap
->ia_atime
.tv_sec
= v_atime
;
1482 iap
->ia_mtime
.tv_nsec
= 0;
1483 iap
->ia_atime
.tv_nsec
= 0;
1487 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1490 * nfsd_setattr already committed the child (and possibly also the parent).
1493 err
= nfserrno(commit_metadata(fhp
));
1495 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1497 * Update the filehandle to get the new inode info.
1500 err
= fh_update(resfhp
);
1504 if (dchild
&& !IS_ERR(dchild
))
1509 err
= nfserrno(host_err
);
1512 #endif /* CONFIG_NFSD_V3 */
1515 * Read a symlink. On entry, *lenp must contain the maximum path length that
1516 * fits into the buffer. On return, it contains the true length.
1517 * N.B. After this call fhp needs an fh_put
1520 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1522 struct dentry
*dentry
;
1523 struct inode
*inode
;
1528 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1532 dentry
= fhp
->fh_dentry
;
1533 inode
= dentry
->d_inode
;
1536 if (!inode
->i_op
->readlink
)
1539 touch_atime(fhp
->fh_export
->ex_path
.mnt
, dentry
);
1540 /* N.B. Why does this call need a get_fs()??
1541 * Remove the set_fs and watch the fireworks:-) --okir
1544 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1545 host_err
= inode
->i_op
->readlink(dentry
, buf
, *lenp
);
1556 err
= nfserrno(host_err
);
1561 * Create a symlink and look up its inode
1562 * N.B. After this call _both_ fhp and resfhp need an fh_put
1565 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1566 char *fname
, int flen
,
1567 char *path
, int plen
,
1568 struct svc_fh
*resfhp
,
1571 struct dentry
*dentry
, *dnew
;
1579 if (isdotent(fname
, flen
))
1582 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1586 dentry
= fhp
->fh_dentry
;
1587 dnew
= lookup_one_len(fname
, dentry
, flen
);
1588 host_err
= PTR_ERR(dnew
);
1592 host_err
= mnt_want_write(fhp
->fh_export
->ex_path
.mnt
);
1596 if (unlikely(path
[plen
] != 0)) {
1597 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1598 if (path_alloced
== NULL
)
1601 strncpy(path_alloced
, path
, plen
);
1602 path_alloced
[plen
] = 0;
1603 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1604 kfree(path_alloced
);
1607 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1608 err
= nfserrno(host_err
);
1610 err
= nfserrno(commit_metadata(fhp
));
1613 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1615 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1617 if (err
==0) err
= cerr
;
1622 err
= nfserrno(host_err
);
1628 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1631 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1632 char *name
, int len
, struct svc_fh
*tfhp
)
1634 struct dentry
*ddir
, *dnew
, *dold
;
1639 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1642 err
= fh_verify(rqstp
, tfhp
, -S_IFDIR
, NFSD_MAY_NOP
);
1650 if (isdotent(name
, len
))
1653 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1654 ddir
= ffhp
->fh_dentry
;
1655 dirp
= ddir
->d_inode
;
1657 dnew
= lookup_one_len(name
, ddir
, len
);
1658 host_err
= PTR_ERR(dnew
);
1662 dold
= tfhp
->fh_dentry
;
1664 host_err
= mnt_want_write(tfhp
->fh_export
->ex_path
.mnt
);
1666 err
= nfserrno(host_err
);
1669 host_err
= vfs_link(dold
, dirp
, dnew
);
1671 err
= nfserrno(commit_metadata(ffhp
));
1673 err
= nfserrno(commit_metadata(tfhp
));
1675 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1678 err
= nfserrno(host_err
);
1680 mnt_drop_write(tfhp
->fh_export
->ex_path
.mnt
);
1689 err
= nfserrno(host_err
);
1695 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1698 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1699 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1701 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1702 struct inode
*fdir
, *tdir
;
1706 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1709 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1713 fdentry
= ffhp
->fh_dentry
;
1714 fdir
= fdentry
->d_inode
;
1716 tdentry
= tfhp
->fh_dentry
;
1717 tdir
= tdentry
->d_inode
;
1719 err
= (rqstp
->rq_vers
== 2) ? nfserr_acces
: nfserr_xdev
;
1720 if (ffhp
->fh_export
!= tfhp
->fh_export
)
1724 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1727 /* cannot use fh_lock as we need deadlock protective ordering
1728 * so do it by hand */
1729 trap
= lock_rename(tdentry
, fdentry
);
1730 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1734 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1735 host_err
= PTR_ERR(odentry
);
1736 if (IS_ERR(odentry
))
1740 if (!odentry
->d_inode
)
1743 if (odentry
== trap
)
1746 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1747 host_err
= PTR_ERR(ndentry
);
1748 if (IS_ERR(ndentry
))
1750 host_err
= -ENOTEMPTY
;
1751 if (ndentry
== trap
)
1754 if (svc_msnfs(ffhp
) &&
1755 ((odentry
->d_count
> 1) || (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
);
1769 host_err
= commit_metadata(tfhp
);
1771 host_err
= commit_metadata(ffhp
);
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 (rdentry
->d_count
> 1)) {
1845 host_err
= vfs_unlink(dirp
, rdentry
);
1846 } else { /* It's RMDIR */
1847 host_err
= vfs_rmdir(dirp
, rdentry
);
1853 host_err
= commit_metadata(fhp
);
1855 mnt_drop_write(fhp
->fh_export
->ex_path
.mnt
);
1857 err
= nfserrno(host_err
);
1863 * We do this buffering because we must not call back into the file
1864 * system's ->lookup() method from the filldir callback. That may well
1865 * deadlock a number of file systems.
1867 * This is based heavily on the implementation of same in XFS.
1869 struct buffered_dirent
{
1873 unsigned int d_type
;
1877 struct readdir_data
{
1883 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1884 loff_t offset
, u64 ino
, unsigned int d_type
)
1886 struct readdir_data
*buf
= __buf
;
1887 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1888 unsigned int reclen
;
1890 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1891 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1896 de
->namlen
= namlen
;
1897 de
->offset
= offset
;
1899 de
->d_type
= d_type
;
1900 memcpy(de
->name
, name
, namlen
);
1901 buf
->used
+= reclen
;
1906 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1907 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1909 struct readdir_data buf
;
1910 struct buffered_dirent
*de
;
1915 buf
.dirent
= (void *)__get_free_page(GFP_KERNEL
);
1917 return nfserrno(-ENOMEM
);
1922 struct inode
*dir_inode
= file
->f_path
.dentry
->d_inode
;
1923 unsigned int reclen
;
1925 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1929 host_err
= vfs_readdir(file
, nfsd_buffered_filldir
, &buf
);
1942 * Various filldir functions may end up calling back into
1943 * lookup_one_len() and the file system's ->lookup() method.
1944 * These expect i_mutex to be held, as it would within readdir.
1946 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1950 de
= (struct buffered_dirent
*)buf
.dirent
;
1952 offset
= de
->offset
;
1954 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1955 de
->ino
, de
->d_type
))
1958 if (cdp
->err
!= nfs_ok
)
1961 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1964 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1966 mutex_unlock(&dir_inode
->i_mutex
);
1967 if (size
> 0) /* We bailed out early */
1970 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1973 free_page((unsigned long)(buf
.dirent
));
1976 return nfserrno(host_err
);
1983 * Read entries from a directory.
1984 * The NFSv3/4 verifier we ignore for now.
1987 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1988 struct readdir_cd
*cdp
, filldir_t func
)
1992 loff_t offset
= *offsetp
;
1994 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_READ
, &file
);
1998 offset
= vfs_llseek(file
, offset
, 0);
2000 err
= nfserrno((int)offset
);
2004 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2006 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2007 err
= nfs_ok
; /* can still be found in ->err */
2015 * Get file system stats
2016 * N.B. After this call fhp needs an fh_put
2019 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2023 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2025 struct path path
= {
2026 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
2027 .dentry
= fhp
->fh_dentry
,
2029 if (vfs_statfs(&path
, stat
))
2035 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2037 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2041 * Check for a user's access permissions to this inode.
2044 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2045 struct dentry
*dentry
, int acc
)
2047 struct inode
*inode
= dentry
->d_inode
;
2050 if (acc
== NFSD_MAY_NOP
)
2053 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2055 (acc
& NFSD_MAY_READ
)? " read" : "",
2056 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2057 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2058 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2059 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2060 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2061 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2063 IS_IMMUTABLE(inode
)? " immut" : "",
2064 IS_APPEND(inode
)? " append" : "",
2065 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2066 dprintk(" owner %d/%d user %d/%d\n",
2067 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2070 /* Normally we reject any write/sattr etc access on a read-only file
2071 * system. But if it is IRIX doing check on write-access for a
2072 * device special file, we ignore rofs.
2074 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2075 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2076 if (exp_rdonly(rqstp
, exp
) ||
2077 __mnt_is_readonly(exp
->ex_path
.mnt
))
2079 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2082 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2085 if (acc
& NFSD_MAY_LOCK
) {
2086 /* If we cannot rely on authentication in NLM requests,
2087 * just allow locks, otherwise require read permission, or
2090 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2093 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2096 * The file owner always gets access permission for accesses that
2097 * would normally be checked at open time. This is to make
2098 * file access work even when the client has done a fchmod(fd, 0).
2100 * However, `cp foo bar' should fail nevertheless when bar is
2101 * readonly. A sensible way to do this might be to reject all
2102 * attempts to truncate a read-only file, because a creat() call
2103 * always implies file truncation.
2104 * ... but this isn't really fair. A process may reasonably call
2105 * ftruncate on an open file descriptor on a file with perm 000.
2106 * We must trust the client to do permission checking - using "ACCESS"
2109 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2110 inode
->i_uid
== current_fsuid())
2113 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2114 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2116 /* Allow read access to binaries even when mode 111 */
2117 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2118 acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
))
2119 err
= inode_permission(inode
, 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) */