2 * File operations used by nfsd. Some of these have been ripped from
3 * other parts of the kernel because they weren't exported, others
4 * are partial duplicates with added or changed functionality.
6 * Note that several functions dget() the dentry upon which they want
7 * to act, most notably those that create directory entries. Response
8 * dentry's are dput()'d if necessary in the release callback.
9 * So if you notice code paths that apparently fail to dput() the
10 * dentry, don't worry--they have been taken care of.
12 * Copyright (C) 1995-1999 Olaf Kirch <okir@monad.swb.de>
13 * Zerocpy NFS support (C) 2002 Hirokazu Takahashi <taka@valinux.co.jp>
17 #include <linux/file.h>
18 #include <linux/splice.h>
19 #include <linux/falloc.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>
32 #include <linux/security.h>
36 #endif /* CONFIG_NFSD_V3 */
39 #include "../internal.h"
42 #endif /* CONFIG_NFSD_V4 */
48 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
52 * This is a cache of readahead params that help us choose the proper
53 * readahead strategy. Initially, we set all readahead parameters to 0
54 * and let the VFS handle things.
55 * If you increase the number of cached files very much, you'll need to
56 * add a hash table here.
59 struct raparms
*p_next
;
64 struct file_ra_state p_ra
;
65 unsigned int p_hindex
;
68 struct raparm_hbucket
{
69 struct raparms
*pb_head
;
71 } ____cacheline_aligned_in_smp
;
73 #define RAPARM_HASH_BITS 4
74 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
75 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
76 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
79 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
81 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
82 * or nfs_ok having possibly changed *dpp and *expp
85 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
86 struct svc_export
**expp
)
88 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
89 struct dentry
*dentry
= *dpp
;
90 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
91 .dentry
= dget(dentry
)};
94 err
= follow_down(&path
);
98 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
102 * We normally allow NFS clients to continue
103 * "underneath" a mountpoint that is not exported.
104 * The exception is V4ROOT, where no traversal is ever
105 * allowed without an explicit export of the new
108 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
113 if (nfsd_v4client(rqstp
) ||
114 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
115 /* successfully crossed mount point */
117 * This is subtle: path.dentry is *not* on path.mnt
118 * at this point. The only reason we are safe is that
119 * original mnt is pinned down by exp, so we should
120 * put path *before* putting exp
123 path
.dentry
= dentry
;
133 static void follow_to_parent(struct path
*path
)
137 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
139 dp
= dget_parent(path
->dentry
);
144 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
146 struct svc_export
*exp2
;
147 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
148 .dentry
= dget(dparent
)};
150 follow_to_parent(&path
);
152 exp2
= rqst_exp_parent(rqstp
, &path
);
153 if (PTR_ERR(exp2
) == -ENOENT
) {
154 *dentryp
= dget(dparent
);
155 } else if (IS_ERR(exp2
)) {
157 return PTR_ERR(exp2
);
159 *dentryp
= dget(path
.dentry
);
168 * For nfsd purposes, we treat V4ROOT exports as though there was an
169 * export at *every* directory.
171 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
173 if (d_mountpoint(dentry
))
175 if (nfsd4_is_junction(dentry
))
177 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
179 return d_inode(dentry
) != NULL
;
183 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
184 const char *name
, unsigned int len
,
185 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
187 struct svc_export
*exp
;
188 struct dentry
*dparent
;
189 struct dentry
*dentry
;
192 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
194 dparent
= fhp
->fh_dentry
;
195 exp
= exp_get(fhp
->fh_export
);
197 /* Lookup the name, but don't follow links */
198 if (isdotent(name
, len
)) {
200 dentry
= dget(dparent
);
201 else if (dparent
!= exp
->ex_path
.dentry
)
202 dentry
= dget_parent(dparent
);
203 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
204 dentry
= dget(dparent
); /* .. == . just like at / */
206 /* checking mountpoint crossing is very different when stepping up */
207 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
213 * In the nfsd4_open() case, this may be held across
214 * subsequent open and delegation acquisition which may
215 * need to take the child's i_mutex:
217 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
218 dentry
= lookup_one_len(name
, dparent
, len
);
219 host_err
= PTR_ERR(dentry
);
222 if (nfsd_mountpoint(dentry
, exp
)) {
224 * We don't need the i_mutex after all. It's
225 * still possible we could open this (regular
226 * files can be mountpoints too), but the
227 * i_mutex is just there to prevent renames of
228 * something that we might be about to delegate,
229 * and a mountpoint won't be renamed:
232 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
238 *dentry_ret
= dentry
;
244 return nfserrno(host_err
);
248 * Look up one component of a pathname.
249 * N.B. After this call _both_ fhp and resfh need an fh_put
251 * If the lookup would cross a mountpoint, and the mounted filesystem
252 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
253 * accepted as it stands and the mounted directory is
254 * returned. Otherwise the covered directory is returned.
255 * NOTE: this mountpoint crossing is not supported properly by all
256 * clients and is explicitly disallowed for NFSv3
257 * NeilBrown <neilb@cse.unsw.edu.au>
260 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
261 unsigned int len
, struct svc_fh
*resfh
)
263 struct svc_export
*exp
;
264 struct dentry
*dentry
;
267 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
270 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
273 err
= check_nfsd_access(exp
, rqstp
);
277 * Note: we compose the file handle now, but as the
278 * dentry may be negative, it may need to be updated.
280 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
281 if (!err
&& d_really_is_negative(dentry
))
290 * Commit metadata changes to stable storage.
293 commit_metadata(struct svc_fh
*fhp
)
295 struct inode
*inode
= d_inode(fhp
->fh_dentry
);
296 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
298 if (!EX_ISSYNC(fhp
->fh_export
))
301 if (export_ops
->commit_metadata
)
302 return export_ops
->commit_metadata(inode
);
303 return sync_inode_metadata(inode
, 1);
307 * Go over the attributes and take care of the small differences between
308 * NFS semantics and what Linux expects.
311 nfsd_sanitize_attrs(struct inode
*inode
, struct iattr
*iap
)
313 /* sanitize the mode change */
314 if (iap
->ia_valid
& ATTR_MODE
) {
315 iap
->ia_mode
&= S_IALLUGO
;
316 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
319 /* Revoke setuid/setgid on chown */
320 if (!S_ISDIR(inode
->i_mode
) &&
321 ((iap
->ia_valid
& ATTR_UID
) || (iap
->ia_valid
& ATTR_GID
))) {
322 iap
->ia_valid
|= ATTR_KILL_PRIV
;
323 if (iap
->ia_valid
& ATTR_MODE
) {
324 /* we're setting mode too, just clear the s*id bits */
325 iap
->ia_mode
&= ~S_ISUID
;
326 if (iap
->ia_mode
& S_IXGRP
)
327 iap
->ia_mode
&= ~S_ISGID
;
329 /* set ATTR_KILL_* bits and let VFS handle it */
330 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
336 nfsd_get_write_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
339 struct inode
*inode
= d_inode(fhp
->fh_dentry
);
342 if (iap
->ia_size
< inode
->i_size
) {
345 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
346 NFSD_MAY_TRUNC
| NFSD_MAY_OWNER_OVERRIDE
);
351 host_err
= get_write_access(inode
);
355 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
357 goto out_put_write_access
;
360 out_put_write_access
:
361 put_write_access(inode
);
363 return nfserrno(host_err
);
367 * Set various file attributes. After this call fhp needs an fh_put.
370 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
371 int check_guard
, time_t guardtime
)
373 struct dentry
*dentry
;
375 int accmode
= NFSD_MAY_SATTR
;
379 bool get_write_count
;
382 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
383 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
384 if (iap
->ia_valid
& ATTR_SIZE
)
387 /* Callers that do fh_verify should do the fh_want_write: */
388 get_write_count
= !fhp
->fh_dentry
;
391 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
394 if (get_write_count
) {
395 host_err
= fh_want_write(fhp
);
397 return nfserrno(host_err
);
400 dentry
= fhp
->fh_dentry
;
401 inode
= d_inode(dentry
);
403 /* Ignore any mode updates on symlinks */
404 if (S_ISLNK(inode
->i_mode
))
405 iap
->ia_valid
&= ~ATTR_MODE
;
410 nfsd_sanitize_attrs(inode
, iap
);
413 * The size case is special, it changes the file in addition to the
416 if (iap
->ia_valid
& ATTR_SIZE
) {
417 err
= nfsd_get_write_access(rqstp
, fhp
, iap
);
423 * RFC5661, Section 18.30.4:
424 * Changing the size of a file with SETATTR indirectly
425 * changes the time_modify and change attributes.
427 * (and similar for the older RFCs)
429 if (iap
->ia_size
!= i_size_read(inode
))
430 iap
->ia_valid
|= ATTR_MTIME
;
433 iap
->ia_valid
|= ATTR_CTIME
;
435 if (check_guard
&& guardtime
!= inode
->i_ctime
.tv_sec
) {
436 err
= nfserr_notsync
;
437 goto out_put_write_access
;
441 host_err
= notify_change(dentry
, iap
, NULL
);
443 err
= nfserrno(host_err
);
445 out_put_write_access
:
447 put_write_access(inode
);
449 err
= nfserrno(commit_metadata(fhp
));
454 #if defined(CONFIG_NFSD_V4)
456 * NFS junction information is stored in an extended attribute.
458 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
461 * nfsd4_is_junction - Test if an object could be an NFS junction
463 * @dentry: object to test
465 * Returns 1 if "dentry" appears to contain NFS junction information.
466 * Otherwise 0 is returned.
468 int nfsd4_is_junction(struct dentry
*dentry
)
470 struct inode
*inode
= d_inode(dentry
);
474 if (inode
->i_mode
& S_IXUGO
)
476 if (!(inode
->i_mode
& S_ISVTX
))
478 if (vfs_getxattr(dentry
, NFSD_JUNCTION_XATTR_NAME
, NULL
, 0) <= 0)
482 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
483 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
484 struct xdr_netobj
*label
)
488 struct dentry
*dentry
;
490 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
494 dentry
= fhp
->fh_dentry
;
496 inode_lock(d_inode(dentry
));
497 host_error
= security_inode_setsecctx(dentry
, label
->data
, label
->len
);
498 inode_unlock(d_inode(dentry
));
499 return nfserrno(host_error
);
502 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
503 struct xdr_netobj
*label
)
505 return nfserr_notsupp
;
509 __be32
nfsd4_clone_file_range(struct file
*src
, u64 src_pos
, struct file
*dst
,
510 u64 dst_pos
, u64 count
)
512 return nfserrno(vfs_clone_file_range(src
, src_pos
, dst
, dst_pos
,
516 __be32
nfsd4_vfs_fallocate(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
517 struct file
*file
, loff_t offset
, loff_t len
,
522 if (!S_ISREG(file_inode(file
)->i_mode
))
525 error
= vfs_fallocate(file
, flags
, offset
, len
);
527 error
= commit_metadata(fhp
);
529 return nfserrno(error
);
531 #endif /* defined(CONFIG_NFSD_V4) */
533 #ifdef CONFIG_NFSD_V3
535 * Check server access rights to a file system object
541 static struct accessmap nfs3_regaccess
[] = {
542 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
543 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
544 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
545 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
550 static struct accessmap nfs3_diraccess
[] = {
551 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
552 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
553 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
554 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
555 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
560 static struct accessmap nfs3_anyaccess
[] = {
561 /* Some clients - Solaris 2.6 at least, make an access call
562 * to the server to check for access for things like /dev/null
563 * (which really, the server doesn't care about). So
564 * We provide simple access checking for them, looking
565 * mainly at mode bits, and we make sure to ignore read-only
568 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
569 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
570 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
571 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
577 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
579 struct accessmap
*map
;
580 struct svc_export
*export
;
581 struct dentry
*dentry
;
582 u32 query
, result
= 0, sresult
= 0;
585 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
589 export
= fhp
->fh_export
;
590 dentry
= fhp
->fh_dentry
;
592 if (d_is_reg(dentry
))
593 map
= nfs3_regaccess
;
594 else if (d_is_dir(dentry
))
595 map
= nfs3_diraccess
;
597 map
= nfs3_anyaccess
;
601 for (; map
->access
; map
++) {
602 if (map
->access
& query
) {
605 sresult
|= map
->access
;
607 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
610 result
|= map
->access
;
613 /* the following error codes just mean the access was not allowed,
614 * rather than an error occurred */
618 /* simply don't "or" in the access bit. */
628 *supported
= sresult
;
633 #endif /* CONFIG_NFSD_V3 */
635 static int nfsd_open_break_lease(struct inode
*inode
, int access
)
639 if (access
& NFSD_MAY_NOT_BREAK_LEASE
)
641 mode
= (access
& NFSD_MAY_WRITE
) ? O_WRONLY
: O_RDONLY
;
642 return break_lease(inode
, mode
| O_NONBLOCK
);
646 * Open an existing file or directory.
647 * The may_flags argument indicates the type of open (read/write/lock)
648 * and additional flags.
649 * N.B. After this call fhp needs an fh_put
652 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
,
653 int may_flags
, struct file
**filp
)
658 int flags
= O_RDONLY
|O_LARGEFILE
;
662 validate_process_creds();
665 * If we get here, then the client has already done an "open",
666 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
667 * in case a chmod has now revoked permission.
669 * Arguably we should also allow the owner override for
670 * directories, but we never have and it doesn't seem to have
671 * caused anyone a problem. If we were to change this, note
672 * also that our filldir callbacks would need a variant of
673 * lookup_one_len that doesn't check permissions.
676 may_flags
|= NFSD_MAY_OWNER_OVERRIDE
;
677 err
= fh_verify(rqstp
, fhp
, type
, may_flags
);
681 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
682 path
.dentry
= fhp
->fh_dentry
;
683 inode
= d_inode(path
.dentry
);
685 /* Disallow write access to files with the append-only bit set
686 * or any access when mandatory locking enabled
689 if (IS_APPEND(inode
) && (may_flags
& NFSD_MAY_WRITE
))
692 * We must ignore files (but only files) which might have mandatory
693 * locks on them because there is no way to know if the accesser has
696 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
702 host_err
= nfsd_open_break_lease(inode
, may_flags
);
703 if (host_err
) /* NOMEM or WOULDBLOCK */
706 if (may_flags
& NFSD_MAY_WRITE
) {
707 if (may_flags
& NFSD_MAY_READ
)
708 flags
= O_RDWR
|O_LARGEFILE
;
710 flags
= O_WRONLY
|O_LARGEFILE
;
713 file
= dentry_open(&path
, flags
, current_cred());
715 host_err
= PTR_ERR(file
);
719 host_err
= ima_file_check(file
, may_flags
, 0);
725 if (may_flags
& NFSD_MAY_64BIT_COOKIE
)
726 file
->f_mode
|= FMODE_64BITHASH
;
728 file
->f_mode
|= FMODE_32BITHASH
;
732 err
= nfserrno(host_err
);
734 validate_process_creds();
739 nfsd_init_raparms(struct file
*file
)
741 struct inode
*inode
= file_inode(file
);
742 dev_t dev
= inode
->i_sb
->s_dev
;
743 ino_t ino
= inode
->i_ino
;
744 struct raparms
*ra
, **rap
, **frap
= NULL
;
747 struct raparm_hbucket
*rab
;
749 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
750 rab
= &raparm_hash
[hash
];
752 spin_lock(&rab
->pb_lock
);
753 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
754 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
757 if (ra
->p_count
== 0)
760 depth
= nfsdstats
.ra_size
;
762 spin_unlock(&rab
->pb_lock
);
772 if (rap
!= &rab
->pb_head
) {
774 ra
->p_next
= rab
->pb_head
;
778 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
779 spin_unlock(&rab
->pb_lock
);
782 file
->f_ra
= ra
->p_ra
;
786 void nfsd_put_raparams(struct file
*file
, struct raparms
*ra
)
788 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
790 spin_lock(&rab
->pb_lock
);
791 ra
->p_ra
= file
->f_ra
;
794 spin_unlock(&rab
->pb_lock
);
798 * Grab and keep cached pages associated with a file in the svc_rqst
799 * so that they can be passed to the network sendmsg/sendpage routines
800 * directly. They will be released after the sending has completed.
803 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
804 struct splice_desc
*sd
)
806 struct svc_rqst
*rqstp
= sd
->u
.data
;
807 struct page
**pp
= rqstp
->rq_next_page
;
808 struct page
*page
= buf
->page
;
813 if (rqstp
->rq_res
.page_len
== 0) {
815 put_page(*rqstp
->rq_next_page
);
816 *(rqstp
->rq_next_page
++) = page
;
817 rqstp
->rq_res
.page_base
= buf
->offset
;
818 rqstp
->rq_res
.page_len
= size
;
819 } else if (page
!= pp
[-1]) {
821 if (*rqstp
->rq_next_page
)
822 put_page(*rqstp
->rq_next_page
);
823 *(rqstp
->rq_next_page
++) = page
;
824 rqstp
->rq_res
.page_len
+= size
;
826 rqstp
->rq_res
.page_len
+= size
;
831 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
832 struct splice_desc
*sd
)
834 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
838 nfsd_finish_read(struct file
*file
, unsigned long *count
, int host_err
)
841 nfsdstats
.io_read
+= host_err
;
843 fsnotify_access(file
);
846 return nfserrno(host_err
);
849 __be32
nfsd_splice_read(struct svc_rqst
*rqstp
,
850 struct file
*file
, loff_t offset
, unsigned long *count
)
852 struct splice_desc sd
= {
860 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
861 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
862 return nfsd_finish_read(file
, count
, host_err
);
865 __be32
nfsd_readv(struct file
*file
, loff_t offset
, struct kvec
*vec
, int vlen
,
866 unsigned long *count
)
873 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
, 0);
875 return nfsd_finish_read(file
, count
, host_err
);
879 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct file
*file
,
880 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
882 if (file
->f_op
->splice_read
&& test_bit(RQ_SPLICE_OK
, &rqstp
->rq_flags
))
883 return nfsd_splice_read(rqstp
, file
, offset
, count
);
885 return nfsd_readv(file
, offset
, vec
, vlen
, count
);
889 * Gathered writes: If another process is currently writing to the file,
890 * there's a high chance this is another nfsd (triggered by a bulk write
891 * from a client's biod). Rather than syncing the file with each write
892 * request, we sleep for 10 msec.
894 * I don't know if this roughly approximates C. Juszak's idea of
895 * gathered writes, but it's a nice and simple solution (IMHO), and it
898 * Note: we do this only in the NFSv2 case, since v3 and higher have a
899 * better tool (separate unstable writes and commits) for solving this
902 static int wait_for_concurrent_writes(struct file
*file
)
904 struct inode
*inode
= file_inode(file
);
905 static ino_t last_ino
;
906 static dev_t last_dev
;
909 if (atomic_read(&inode
->i_writecount
) > 1
910 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
911 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
913 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
916 if (inode
->i_state
& I_DIRTY
) {
917 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
918 err
= vfs_fsync(file
, 0);
920 last_ino
= inode
->i_ino
;
921 last_dev
= inode
->i_sb
->s_dev
;
926 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
927 loff_t offset
, struct kvec
*vec
, int vlen
,
928 unsigned long *cnt
, int *stablep
)
930 struct svc_export
*exp
;
935 int stable
= *stablep
;
938 unsigned int pflags
= current
->flags
;
941 if (test_bit(RQ_LOCAL
, &rqstp
->rq_flags
))
943 * We want less throttling in balance_dirty_pages()
944 * and shrink_inactive_list() so that nfs to
945 * localhost doesn't cause nfsd to lock up due to all
946 * the client's dirty pages or its congested queue.
948 current
->flags
|= PF_LESS_THROTTLE
;
950 inode
= file_inode(file
);
951 exp
= fhp
->fh_export
;
953 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
958 if (stable
&& !use_wgather
)
961 /* Write the data. */
962 oldfs
= get_fs(); set_fs(KERNEL_DS
);
963 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &pos
, flags
);
968 nfsdstats
.io_write
+= host_err
;
969 fsnotify_modify(file
);
971 if (stable
&& use_wgather
)
972 host_err
= wait_for_concurrent_writes(file
);
975 dprintk("nfsd: write complete host_err=%d\n", host_err
);
979 err
= nfserrno(host_err
);
980 if (test_bit(RQ_LOCAL
, &rqstp
->rq_flags
))
981 tsk_restore_flags(current
, pflags
, PF_LESS_THROTTLE
);
986 * Read data from a file. count must contain the requested read count
987 * on entry. On return, *count contains the number of bytes actually read.
988 * N.B. After this call fhp needs an fh_put
990 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
991 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
997 trace_read_start(rqstp
, fhp
, offset
, vlen
);
998 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1002 ra
= nfsd_init_raparms(file
);
1004 trace_read_opened(rqstp
, fhp
, offset
, vlen
);
1005 err
= nfsd_vfs_read(rqstp
, file
, offset
, vec
, vlen
, count
);
1006 trace_read_io_done(rqstp
, fhp
, offset
, vlen
);
1009 nfsd_put_raparams(file
, ra
);
1012 trace_read_done(rqstp
, fhp
, offset
, vlen
);
1018 * Write data to a file.
1019 * The stable flag requests synchronous writes.
1020 * N.B. After this call fhp needs an fh_put
1023 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1024 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1029 trace_write_start(rqstp
, fhp
, offset
, vlen
);
1032 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1033 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1036 trace_write_opened(rqstp
, fhp
, offset
, vlen
);
1037 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1039 trace_write_io_done(rqstp
, fhp
, offset
, vlen
);
1041 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1045 trace_write_opened(rqstp
, fhp
, offset
, vlen
);
1047 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1049 trace_write_io_done(rqstp
, fhp
, offset
, vlen
);
1053 trace_write_done(rqstp
, fhp
, offset
, vlen
);
1057 #ifdef CONFIG_NFSD_V3
1059 * Commit all pending writes to stable storage.
1061 * Note: we only guarantee that data that lies within the range specified
1062 * by the 'offset' and 'count' parameters will be synced.
1064 * Unfortunately we cannot lock the file to make sure we return full WCC
1065 * data to the client, as locking happens lower down in the filesystem.
1068 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1069 loff_t offset
, unsigned long count
)
1072 loff_t end
= LLONG_MAX
;
1073 __be32 err
= nfserr_inval
;
1078 end
= offset
+ (loff_t
)count
- 1;
1083 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1084 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1087 if (EX_ISSYNC(fhp
->fh_export
)) {
1088 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1090 if (err2
!= -EINVAL
)
1091 err
= nfserrno(err2
);
1093 err
= nfserr_notsupp
;
1100 #endif /* CONFIG_NFSD_V3 */
1103 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1107 * Mode has already been set earlier in create:
1109 iap
->ia_valid
&= ~ATTR_MODE
;
1111 * Setting uid/gid works only for root. Irix appears to
1112 * send along the gid on create when it tries to implement
1113 * setgid directories via NFS:
1115 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID
))
1116 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1118 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1119 /* Callers expect file metadata to be committed here */
1120 return nfserrno(commit_metadata(resfhp
));
1123 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1124 * setting size to 0 may fail for some specific file systems by the permission
1125 * checking which requires WRITE permission but the mode is 000.
1126 * we ignore the resizing(to 0) on the just new created file, since the size is
1127 * 0 after file created.
1129 * call this only after vfs_create() is called.
1132 nfsd_check_ignore_resizing(struct iattr
*iap
)
1134 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1135 iap
->ia_valid
&= ~ATTR_SIZE
;
1138 /* The parent directory should already be locked: */
1140 nfsd_create_locked(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1141 char *fname
, int flen
, struct iattr
*iap
,
1142 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1144 struct dentry
*dentry
, *dchild
= NULL
;
1150 dentry
= fhp
->fh_dentry
;
1151 dirp
= d_inode(dentry
);
1153 dchild
= dget(resfhp
->fh_dentry
);
1154 if (!fhp
->fh_locked
) {
1155 WARN_ONCE(1, "nfsd_create: parent %pd2 not locked!\n",
1161 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
, NFSD_MAY_CREATE
);
1165 if (!(iap
->ia_valid
& ATTR_MODE
))
1167 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1173 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1175 nfsd_check_ignore_resizing(iap
);
1178 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1184 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1187 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1194 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1197 * nfsd_create_setattr already committed the child. Transactional
1198 * filesystems had a chance to commit changes for both parent and
1199 * child simultaneously making the following commit_metadata a
1202 err2
= nfserrno(commit_metadata(fhp
));
1206 * Update the file handle to get the new inode info.
1209 err
= fh_update(resfhp
);
1211 if (dchild
&& !IS_ERR(dchild
))
1216 err
= nfserrno(host_err
);
1221 * Create a filesystem object (regular, directory, special).
1222 * Note that the parent directory is left locked.
1224 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1227 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1228 char *fname
, int flen
, struct iattr
*iap
,
1229 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1231 struct dentry
*dentry
, *dchild
= NULL
;
1236 if (isdotent(fname
, flen
))
1237 return nfserr_exist
;
1239 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_NOP
);
1243 dentry
= fhp
->fh_dentry
;
1244 dirp
= d_inode(dentry
);
1246 host_err
= fh_want_write(fhp
);
1248 return nfserrno(host_err
);
1250 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1251 dchild
= lookup_one_len(fname
, dentry
, flen
);
1252 host_err
= PTR_ERR(dchild
);
1254 return nfserrno(host_err
);
1255 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1260 return nfsd_create_locked(rqstp
, fhp
, fname
, flen
, iap
, type
,
1264 #ifdef CONFIG_NFSD_V3
1267 * NFSv3 and NFSv4 version of nfsd_create
1270 do_nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1271 char *fname
, int flen
, struct iattr
*iap
,
1272 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1273 bool *truncp
, bool *created
)
1275 struct dentry
*dentry
, *dchild
= NULL
;
1279 __u32 v_mtime
=0, v_atime
=0;
1285 if (isdotent(fname
, flen
))
1287 if (!(iap
->ia_valid
& ATTR_MODE
))
1289 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
1293 dentry
= fhp
->fh_dentry
;
1294 dirp
= d_inode(dentry
);
1296 host_err
= fh_want_write(fhp
);
1300 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1303 * Compose the response file handle.
1305 dchild
= lookup_one_len(fname
, dentry
, flen
);
1306 host_err
= PTR_ERR(dchild
);
1310 /* If file doesn't exist, check for permissions to create one */
1311 if (d_really_is_negative(dchild
)) {
1312 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1317 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1321 if (nfsd_create_is_exclusive(createmode
)) {
1322 /* solaris7 gets confused (bugid 4218508) if these have
1323 * the high bit set, so just clear the high bits. If this is
1324 * ever changed to use different attrs for storing the
1325 * verifier, then do_open_lookup() will also need to be fixed
1328 v_mtime
= verifier
[0]&0x7fffffff;
1329 v_atime
= verifier
[1]&0x7fffffff;
1332 if (d_really_is_positive(dchild
)) {
1335 switch (createmode
) {
1336 case NFS3_CREATE_UNCHECKED
:
1337 if (! d_is_reg(dchild
))
1340 /* in nfsv4, we need to treat this case a little
1341 * differently. we don't want to truncate the
1342 * file now; this would be wrong if the OPEN
1343 * fails for some other reason. furthermore,
1344 * if the size is nonzero, we should ignore it
1345 * according to spec!
1347 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1350 iap
->ia_valid
&= ATTR_SIZE
;
1354 case NFS3_CREATE_EXCLUSIVE
:
1355 if ( d_inode(dchild
)->i_mtime
.tv_sec
== v_mtime
1356 && d_inode(dchild
)->i_atime
.tv_sec
== v_atime
1357 && d_inode(dchild
)->i_size
== 0 ) {
1362 case NFS4_CREATE_EXCLUSIVE4_1
:
1363 if ( d_inode(dchild
)->i_mtime
.tv_sec
== v_mtime
1364 && d_inode(dchild
)->i_atime
.tv_sec
== v_atime
1365 && d_inode(dchild
)->i_size
== 0 ) {
1371 case NFS3_CREATE_GUARDED
:
1378 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1386 nfsd_check_ignore_resizing(iap
);
1388 if (nfsd_create_is_exclusive(createmode
)) {
1389 /* Cram the verifier into atime/mtime */
1390 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1391 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1392 /* XXX someone who knows this better please fix it for nsec */
1393 iap
->ia_mtime
.tv_sec
= v_mtime
;
1394 iap
->ia_atime
.tv_sec
= v_atime
;
1395 iap
->ia_mtime
.tv_nsec
= 0;
1396 iap
->ia_atime
.tv_nsec
= 0;
1400 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1403 * nfsd_create_setattr already committed the child
1404 * (and possibly also the parent).
1407 err
= nfserrno(commit_metadata(fhp
));
1410 * Update the filehandle to get the new inode info.
1413 err
= fh_update(resfhp
);
1417 if (dchild
&& !IS_ERR(dchild
))
1423 err
= nfserrno(host_err
);
1426 #endif /* CONFIG_NFSD_V3 */
1429 * Read a symlink. On entry, *lenp must contain the maximum path length that
1430 * fits into the buffer. On return, it contains the true length.
1431 * N.B. After this call fhp needs an fh_put
1434 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1436 struct inode
*inode
;
1442 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1446 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
1447 path
.dentry
= fhp
->fh_dentry
;
1448 inode
= d_inode(path
.dentry
);
1451 if (!inode
->i_op
->readlink
)
1455 /* N.B. Why does this call need a get_fs()??
1456 * Remove the set_fs and watch the fireworks:-) --okir
1459 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1460 host_err
= inode
->i_op
->readlink(path
.dentry
, (char __user
*)buf
, *lenp
);
1471 err
= nfserrno(host_err
);
1476 * Create a symlink and look up its inode
1477 * N.B. After this call _both_ fhp and resfhp need an fh_put
1480 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1481 char *fname
, int flen
,
1483 struct svc_fh
*resfhp
)
1485 struct dentry
*dentry
, *dnew
;
1490 if (!flen
|| path
[0] == '\0')
1493 if (isdotent(fname
, flen
))
1496 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1500 host_err
= fh_want_write(fhp
);
1505 dentry
= fhp
->fh_dentry
;
1506 dnew
= lookup_one_len(fname
, dentry
, flen
);
1507 host_err
= PTR_ERR(dnew
);
1511 host_err
= vfs_symlink(d_inode(dentry
), dnew
, path
);
1512 err
= nfserrno(host_err
);
1514 err
= nfserrno(commit_metadata(fhp
));
1519 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1521 if (err
==0) err
= cerr
;
1526 err
= nfserrno(host_err
);
1532 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1535 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1536 char *name
, int len
, struct svc_fh
*tfhp
)
1538 struct dentry
*ddir
, *dnew
, *dold
;
1543 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1546 err
= fh_verify(rqstp
, tfhp
, 0, NFSD_MAY_NOP
);
1550 if (d_is_dir(tfhp
->fh_dentry
))
1556 if (isdotent(name
, len
))
1559 host_err
= fh_want_write(tfhp
);
1561 err
= nfserrno(host_err
);
1565 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1566 ddir
= ffhp
->fh_dentry
;
1567 dirp
= d_inode(ddir
);
1569 dnew
= lookup_one_len(name
, ddir
, len
);
1570 host_err
= PTR_ERR(dnew
);
1574 dold
= tfhp
->fh_dentry
;
1577 if (d_really_is_negative(dold
))
1579 host_err
= vfs_link(dold
, dirp
, dnew
, NULL
);
1581 err
= nfserrno(commit_metadata(ffhp
));
1583 err
= nfserrno(commit_metadata(tfhp
));
1585 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1588 err
= nfserrno(host_err
);
1594 fh_drop_write(tfhp
);
1599 err
= nfserrno(host_err
);
1605 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1608 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1609 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1611 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1612 struct inode
*fdir
, *tdir
;
1616 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1619 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1623 fdentry
= ffhp
->fh_dentry
;
1624 fdir
= d_inode(fdentry
);
1626 tdentry
= tfhp
->fh_dentry
;
1627 tdir
= d_inode(tdentry
);
1630 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1633 host_err
= fh_want_write(ffhp
);
1635 err
= nfserrno(host_err
);
1639 /* cannot use fh_lock as we need deadlock protective ordering
1640 * so do it by hand */
1641 trap
= lock_rename(tdentry
, fdentry
);
1642 ffhp
->fh_locked
= tfhp
->fh_locked
= true;
1646 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1647 host_err
= PTR_ERR(odentry
);
1648 if (IS_ERR(odentry
))
1652 if (d_really_is_negative(odentry
))
1655 if (odentry
== trap
)
1658 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1659 host_err
= PTR_ERR(ndentry
);
1660 if (IS_ERR(ndentry
))
1662 host_err
= -ENOTEMPTY
;
1663 if (ndentry
== trap
)
1667 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1669 if (ffhp
->fh_export
->ex_path
.dentry
!= tfhp
->fh_export
->ex_path
.dentry
)
1672 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
, NULL
, 0);
1674 host_err
= commit_metadata(tfhp
);
1676 host_err
= commit_metadata(ffhp
);
1683 err
= nfserrno(host_err
);
1685 * We cannot rely on fh_unlock on the two filehandles,
1686 * as that would do the wrong thing if the two directories
1687 * were the same, so again we do it by hand.
1689 fill_post_wcc(ffhp
);
1690 fill_post_wcc(tfhp
);
1691 unlock_rename(tdentry
, fdentry
);
1692 ffhp
->fh_locked
= tfhp
->fh_locked
= false;
1693 fh_drop_write(ffhp
);
1700 * Unlink a file or directory
1701 * N.B. After this call fhp needs an fh_put
1704 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1705 char *fname
, int flen
)
1707 struct dentry
*dentry
, *rdentry
;
1713 if (!flen
|| isdotent(fname
, flen
))
1715 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1719 host_err
= fh_want_write(fhp
);
1723 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1724 dentry
= fhp
->fh_dentry
;
1725 dirp
= d_inode(dentry
);
1727 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1728 host_err
= PTR_ERR(rdentry
);
1729 if (IS_ERR(rdentry
))
1732 if (d_really_is_negative(rdentry
)) {
1739 type
= d_inode(rdentry
)->i_mode
& S_IFMT
;
1741 if (type
!= S_IFDIR
)
1742 host_err
= vfs_unlink(dirp
, rdentry
, NULL
);
1744 host_err
= vfs_rmdir(dirp
, rdentry
);
1746 host_err
= commit_metadata(fhp
);
1750 err
= nfserrno(host_err
);
1756 * We do this buffering because we must not call back into the file
1757 * system's ->lookup() method from the filldir callback. That may well
1758 * deadlock a number of file systems.
1760 * This is based heavily on the implementation of same in XFS.
1762 struct buffered_dirent
{
1766 unsigned int d_type
;
1770 struct readdir_data
{
1771 struct dir_context ctx
;
1777 static int nfsd_buffered_filldir(struct dir_context
*ctx
, const char *name
,
1778 int namlen
, loff_t offset
, u64 ino
,
1779 unsigned int d_type
)
1781 struct readdir_data
*buf
=
1782 container_of(ctx
, struct readdir_data
, ctx
);
1783 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1784 unsigned int reclen
;
1786 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1787 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1792 de
->namlen
= namlen
;
1793 de
->offset
= offset
;
1795 de
->d_type
= d_type
;
1796 memcpy(de
->name
, name
, namlen
);
1797 buf
->used
+= reclen
;
1802 static __be32
nfsd_buffered_readdir(struct file
*file
, nfsd_filldir_t func
,
1803 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1805 struct buffered_dirent
*de
;
1809 struct readdir_data buf
= {
1810 .ctx
.actor
= nfsd_buffered_filldir
,
1811 .dirent
= (void *)__get_free_page(GFP_KERNEL
)
1815 return nfserrno(-ENOMEM
);
1820 unsigned int reclen
;
1822 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1826 host_err
= iterate_dir(file
, &buf
.ctx
);
1838 de
= (struct buffered_dirent
*)buf
.dirent
;
1840 offset
= de
->offset
;
1842 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1843 de
->ino
, de
->d_type
))
1846 if (cdp
->err
!= nfs_ok
)
1849 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1852 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1854 if (size
> 0) /* We bailed out early */
1857 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1860 free_page((unsigned long)(buf
.dirent
));
1863 return nfserrno(host_err
);
1870 * Read entries from a directory.
1871 * The NFSv3/4 verifier we ignore for now.
1874 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1875 struct readdir_cd
*cdp
, nfsd_filldir_t func
)
1879 loff_t offset
= *offsetp
;
1880 int may_flags
= NFSD_MAY_READ
;
1882 /* NFSv2 only supports 32 bit cookies */
1883 if (rqstp
->rq_vers
> 2)
1884 may_flags
|= NFSD_MAY_64BIT_COOKIE
;
1886 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, may_flags
, &file
);
1890 offset
= vfs_llseek(file
, offset
, SEEK_SET
);
1892 err
= nfserrno((int)offset
);
1896 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1898 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1899 err
= nfs_ok
; /* can still be found in ->err */
1907 * Get file system stats
1908 * N.B. After this call fhp needs an fh_put
1911 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
1915 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
1917 struct path path
= {
1918 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
1919 .dentry
= fhp
->fh_dentry
,
1921 if (vfs_statfs(&path
, stat
))
1927 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
1929 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
1933 * Check for a user's access permissions to this inode.
1936 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
1937 struct dentry
*dentry
, int acc
)
1939 struct inode
*inode
= d_inode(dentry
);
1942 if ((acc
& NFSD_MAY_MASK
) == NFSD_MAY_NOP
)
1945 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1947 (acc
& NFSD_MAY_READ
)? " read" : "",
1948 (acc
& NFSD_MAY_WRITE
)? " write" : "",
1949 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
1950 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
1951 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
1952 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
1953 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
1955 IS_IMMUTABLE(inode
)? " immut" : "",
1956 IS_APPEND(inode
)? " append" : "",
1957 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
1958 dprintk(" owner %d/%d user %d/%d\n",
1959 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
1962 /* Normally we reject any write/sattr etc access on a read-only file
1963 * system. But if it is IRIX doing check on write-access for a
1964 * device special file, we ignore rofs.
1966 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
1967 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
1968 if (exp_rdonly(rqstp
, exp
) ||
1969 __mnt_is_readonly(exp
->ex_path
.mnt
))
1971 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
1974 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
1977 if (acc
& NFSD_MAY_LOCK
) {
1978 /* If we cannot rely on authentication in NLM requests,
1979 * just allow locks, otherwise require read permission, or
1982 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
1985 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
1988 * The file owner always gets access permission for accesses that
1989 * would normally be checked at open time. This is to make
1990 * file access work even when the client has done a fchmod(fd, 0).
1992 * However, `cp foo bar' should fail nevertheless when bar is
1993 * readonly. A sensible way to do this might be to reject all
1994 * attempts to truncate a read-only file, because a creat() call
1995 * always implies file truncation.
1996 * ... but this isn't really fair. A process may reasonably call
1997 * ftruncate on an open file descriptor on a file with perm 000.
1998 * We must trust the client to do permission checking - using "ACCESS"
2001 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2002 uid_eq(inode
->i_uid
, current_fsuid()))
2005 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2006 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2008 /* Allow read access to binaries even when mode 111 */
2009 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2010 (acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
) ||
2011 acc
== (NFSD_MAY_READ
| NFSD_MAY_READ_IF_EXEC
)))
2012 err
= inode_permission(inode
, MAY_EXEC
);
2014 return err
? nfserrno(err
) : 0;
2018 nfsd_racache_shutdown(void)
2020 struct raparms
*raparm
, *last_raparm
;
2023 dprintk("nfsd: freeing readahead buffers.\n");
2025 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2026 raparm
= raparm_hash
[i
].pb_head
;
2028 last_raparm
= raparm
;
2029 raparm
= raparm
->p_next
;
2032 raparm_hash
[i
].pb_head
= NULL
;
2036 * Initialize readahead param cache
2039 nfsd_racache_init(int cache_size
)
2044 struct raparms
**raparm
= NULL
;
2047 if (raparm_hash
[0].pb_head
)
2049 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2050 nperbucket
= max(2, nperbucket
);
2051 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2053 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2055 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2056 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2058 raparm
= &raparm_hash
[i
].pb_head
;
2059 for (j
= 0; j
< nperbucket
; j
++) {
2060 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2063 raparm
= &(*raparm
)->p_next
;
2068 nfsdstats
.ra_size
= cache_size
;
2072 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2073 nfsd_racache_shutdown();