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/fcntl.h>
20 #include <linux/namei.h>
21 #include <linux/delay.h>
22 #include <linux/fsnotify.h>
23 #include <linux/posix_acl_xattr.h>
24 #include <linux/xattr.h>
25 #include <linux/jhash.h>
26 #include <linux/ima.h>
27 #include <linux/slab.h>
28 #include <asm/uaccess.h>
29 #include <linux/exportfs.h>
30 #include <linux/writeback.h>
31 #include <linux/security.h>
35 #endif /* CONFIG_NFSD_V3 */
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
);
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 (nfsd4_is_junction(dentry
))
174 if (!(exp
->ex_flags
& NFSEXP_V4ROOT
))
176 return dentry
->d_inode
!= NULL
;
180 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
181 const char *name
, unsigned int len
,
182 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
184 struct svc_export
*exp
;
185 struct dentry
*dparent
;
186 struct dentry
*dentry
;
189 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
191 dparent
= fhp
->fh_dentry
;
192 exp
= fhp
->fh_export
;
195 /* Lookup the name, but don't follow links */
196 if (isdotent(name
, len
)) {
198 dentry
= dget(dparent
);
199 else if (dparent
!= exp
->ex_path
.dentry
)
200 dentry
= dget_parent(dparent
);
201 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
202 dentry
= dget(dparent
); /* .. == . just like at / */
204 /* checking mountpoint crossing is very different when stepping up */
205 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
211 * In the nfsd4_open() case, this may be held across
212 * subsequent open and delegation acquisition which may
213 * need to take the child's i_mutex:
215 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
216 dentry
= lookup_one_len(name
, dparent
, len
);
217 host_err
= PTR_ERR(dentry
);
221 * check if we have crossed a mount point ...
223 if (nfsd_mountpoint(dentry
, exp
)) {
224 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
230 *dentry_ret
= dentry
;
236 return nfserrno(host_err
);
240 * Look up one component of a pathname.
241 * N.B. After this call _both_ fhp and resfh need an fh_put
243 * If the lookup would cross a mountpoint, and the mounted filesystem
244 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
245 * accepted as it stands and the mounted directory is
246 * returned. Otherwise the covered directory is returned.
247 * NOTE: this mountpoint crossing is not supported properly by all
248 * clients and is explicitly disallowed for NFSv3
249 * NeilBrown <neilb@cse.unsw.edu.au>
252 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
253 unsigned int len
, struct svc_fh
*resfh
)
255 struct svc_export
*exp
;
256 struct dentry
*dentry
;
259 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
262 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
265 err
= check_nfsd_access(exp
, rqstp
);
269 * Note: we compose the file handle now, but as the
270 * dentry may be negative, it may need to be updated.
272 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
273 if (!err
&& !dentry
->d_inode
)
282 * Commit metadata changes to stable storage.
285 commit_metadata(struct svc_fh
*fhp
)
287 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
288 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
290 if (!EX_ISSYNC(fhp
->fh_export
))
293 if (export_ops
->commit_metadata
)
294 return export_ops
->commit_metadata(inode
);
295 return sync_inode_metadata(inode
, 1);
299 * Go over the attributes and take care of the small differences between
300 * NFS semantics and what Linux expects.
303 nfsd_sanitize_attrs(struct inode
*inode
, struct iattr
*iap
)
306 * NFSv2 does not differentiate between "set-[ac]time-to-now"
307 * which only requires access, and "set-[ac]time-to-X" which
308 * requires ownership.
309 * So if it looks like it might be "set both to the same time which
310 * is close to now", and if inode_change_ok fails, then we
311 * convert to "set to now" instead of "set to explicit time"
313 * We only call inode_change_ok as the last test as technically
314 * it is not an interface that we should be using.
316 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
317 #define MAX_TOUCH_TIME_ERROR (30*60)
318 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
319 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
323 * Now just make sure time is in the right ballpark.
324 * Solaris, at least, doesn't seem to care what the time
325 * request is. We require it be within 30 minutes of now.
327 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
330 if (delta
< MAX_TOUCH_TIME_ERROR
&&
331 inode_change_ok(inode
, iap
) != 0) {
333 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
334 * This will cause notify_change to set these times
337 iap
->ia_valid
&= ~BOTH_TIME_SET
;
341 /* sanitize the mode change */
342 if (iap
->ia_valid
& ATTR_MODE
) {
343 iap
->ia_mode
&= S_IALLUGO
;
344 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
347 /* Revoke setuid/setgid on chown */
348 if (!S_ISDIR(inode
->i_mode
) &&
349 ((iap
->ia_valid
& ATTR_UID
) || (iap
->ia_valid
& ATTR_GID
))) {
350 iap
->ia_valid
|= ATTR_KILL_PRIV
;
351 if (iap
->ia_valid
& ATTR_MODE
) {
352 /* we're setting mode too, just clear the s*id bits */
353 iap
->ia_mode
&= ~S_ISUID
;
354 if (iap
->ia_mode
& S_IXGRP
)
355 iap
->ia_mode
&= ~S_ISGID
;
357 /* set ATTR_KILL_* bits and let VFS handle it */
358 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
364 nfsd_get_write_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
367 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
370 if (iap
->ia_size
< inode
->i_size
) {
373 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
374 NFSD_MAY_TRUNC
| NFSD_MAY_OWNER_OVERRIDE
);
379 host_err
= get_write_access(inode
);
383 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
385 goto out_put_write_access
;
388 out_put_write_access
:
389 put_write_access(inode
);
391 return nfserrno(host_err
);
395 * Set various file attributes. After this call fhp needs an fh_put.
398 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
399 int check_guard
, time_t guardtime
)
401 struct dentry
*dentry
;
403 int accmode
= NFSD_MAY_SATTR
;
407 bool get_write_count
;
410 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
411 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
412 if (iap
->ia_valid
& ATTR_SIZE
)
415 /* Callers that do fh_verify should do the fh_want_write: */
416 get_write_count
= !fhp
->fh_dentry
;
419 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
422 if (get_write_count
) {
423 host_err
= fh_want_write(fhp
);
425 return nfserrno(host_err
);
428 dentry
= fhp
->fh_dentry
;
429 inode
= dentry
->d_inode
;
431 /* Ignore any mode updates on symlinks */
432 if (S_ISLNK(inode
->i_mode
))
433 iap
->ia_valid
&= ~ATTR_MODE
;
438 nfsd_sanitize_attrs(inode
, iap
);
441 * The size case is special, it changes the file in addition to the
444 if (iap
->ia_valid
& ATTR_SIZE
) {
445 err
= nfsd_get_write_access(rqstp
, fhp
, iap
);
451 iap
->ia_valid
|= ATTR_CTIME
;
453 if (check_guard
&& guardtime
!= inode
->i_ctime
.tv_sec
) {
454 err
= nfserr_notsync
;
455 goto out_put_write_access
;
459 host_err
= notify_change(dentry
, iap
, NULL
);
461 err
= nfserrno(host_err
);
463 out_put_write_access
:
465 put_write_access(inode
);
467 commit_metadata(fhp
);
472 #if defined(CONFIG_NFSD_V4)
474 * NFS junction information is stored in an extended attribute.
476 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
479 * nfsd4_is_junction - Test if an object could be an NFS junction
481 * @dentry: object to test
483 * Returns 1 if "dentry" appears to contain NFS junction information.
484 * Otherwise 0 is returned.
486 int nfsd4_is_junction(struct dentry
*dentry
)
488 struct inode
*inode
= dentry
->d_inode
;
492 if (inode
->i_mode
& S_IXUGO
)
494 if (!(inode
->i_mode
& S_ISVTX
))
496 if (vfs_getxattr(dentry
, NFSD_JUNCTION_XATTR_NAME
, NULL
, 0) <= 0)
500 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
501 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
502 struct xdr_netobj
*label
)
506 struct dentry
*dentry
;
508 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
512 dentry
= fhp
->fh_dentry
;
514 mutex_lock(&dentry
->d_inode
->i_mutex
);
515 host_error
= security_inode_setsecctx(dentry
, label
->data
, label
->len
);
516 mutex_unlock(&dentry
->d_inode
->i_mutex
);
517 return nfserrno(host_error
);
520 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
521 struct xdr_netobj
*label
)
523 return nfserr_notsupp
;
527 #endif /* defined(CONFIG_NFSD_V4) */
529 #ifdef CONFIG_NFSD_V3
531 * Check server access rights to a file system object
537 static struct accessmap nfs3_regaccess
[] = {
538 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
539 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
540 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
541 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
546 static struct accessmap nfs3_diraccess
[] = {
547 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
548 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
549 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
550 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
551 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
556 static struct accessmap nfs3_anyaccess
[] = {
557 /* Some clients - Solaris 2.6 at least, make an access call
558 * to the server to check for access for things like /dev/null
559 * (which really, the server doesn't care about). So
560 * We provide simple access checking for them, looking
561 * mainly at mode bits, and we make sure to ignore read-only
564 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
565 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
566 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
567 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
573 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
575 struct accessmap
*map
;
576 struct svc_export
*export
;
577 struct dentry
*dentry
;
578 u32 query
, result
= 0, sresult
= 0;
581 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
585 export
= fhp
->fh_export
;
586 dentry
= fhp
->fh_dentry
;
588 if (S_ISREG(dentry
->d_inode
->i_mode
))
589 map
= nfs3_regaccess
;
590 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
591 map
= nfs3_diraccess
;
593 map
= nfs3_anyaccess
;
597 for (; map
->access
; map
++) {
598 if (map
->access
& query
) {
601 sresult
|= map
->access
;
603 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
606 result
|= map
->access
;
609 /* the following error codes just mean the access was not allowed,
610 * rather than an error occurred */
614 /* simply don't "or" in the access bit. */
624 *supported
= sresult
;
629 #endif /* CONFIG_NFSD_V3 */
631 static int nfsd_open_break_lease(struct inode
*inode
, int access
)
635 if (access
& NFSD_MAY_NOT_BREAK_LEASE
)
637 mode
= (access
& NFSD_MAY_WRITE
) ? O_WRONLY
: O_RDONLY
;
638 return break_lease(inode
, mode
| O_NONBLOCK
);
642 * Open an existing file or directory.
643 * The may_flags argument indicates the type of open (read/write/lock)
644 * and additional flags.
645 * N.B. After this call fhp needs an fh_put
648 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
,
649 int may_flags
, struct file
**filp
)
653 int flags
= O_RDONLY
|O_LARGEFILE
;
657 validate_process_creds();
660 * If we get here, then the client has already done an "open",
661 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
662 * in case a chmod has now revoked permission.
664 * Arguably we should also allow the owner override for
665 * directories, but we never have and it doesn't seem to have
666 * caused anyone a problem. If we were to change this, note
667 * also that our filldir callbacks would need a variant of
668 * lookup_one_len that doesn't check permissions.
671 may_flags
|= NFSD_MAY_OWNER_OVERRIDE
;
672 err
= fh_verify(rqstp
, fhp
, type
, may_flags
);
676 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
677 path
.dentry
= fhp
->fh_dentry
;
678 inode
= path
.dentry
->d_inode
;
680 /* Disallow write access to files with the append-only bit set
681 * or any access when mandatory locking enabled
684 if (IS_APPEND(inode
) && (may_flags
& NFSD_MAY_WRITE
))
687 * We must ignore files (but only files) which might have mandatory
688 * locks on them because there is no way to know if the accesser has
691 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
697 host_err
= nfsd_open_break_lease(inode
, may_flags
);
698 if (host_err
) /* NOMEM or WOULDBLOCK */
701 if (may_flags
& NFSD_MAY_WRITE
) {
702 if (may_flags
& NFSD_MAY_READ
)
703 flags
= O_RDWR
|O_LARGEFILE
;
705 flags
= O_WRONLY
|O_LARGEFILE
;
707 *filp
= dentry_open(&path
, flags
, current_cred());
709 host_err
= PTR_ERR(*filp
);
712 host_err
= ima_file_check(*filp
, may_flags
);
714 if (may_flags
& NFSD_MAY_64BIT_COOKIE
)
715 (*filp
)->f_mode
|= FMODE_64BITHASH
;
717 (*filp
)->f_mode
|= FMODE_32BITHASH
;
721 err
= nfserrno(host_err
);
723 validate_process_creds();
731 nfsd_close(struct file
*filp
)
737 * Obtain the readahead parameters for the file
738 * specified by (dev, ino).
741 static inline struct raparms
*
742 nfsd_get_raparms(dev_t dev
, ino_t 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
);
784 * Grab and keep cached pages associated with a file in the svc_rqst
785 * so that they can be passed to the network sendmsg/sendpage routines
786 * directly. They will be released after the sending has completed.
789 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
790 struct splice_desc
*sd
)
792 struct svc_rqst
*rqstp
= sd
->u
.data
;
793 struct page
**pp
= rqstp
->rq_next_page
;
794 struct page
*page
= buf
->page
;
799 if (rqstp
->rq_res
.page_len
== 0) {
801 put_page(*rqstp
->rq_next_page
);
802 *(rqstp
->rq_next_page
++) = page
;
803 rqstp
->rq_res
.page_base
= buf
->offset
;
804 rqstp
->rq_res
.page_len
= size
;
805 } else if (page
!= pp
[-1]) {
807 if (*rqstp
->rq_next_page
)
808 put_page(*rqstp
->rq_next_page
);
809 *(rqstp
->rq_next_page
++) = page
;
810 rqstp
->rq_res
.page_len
+= size
;
812 rqstp
->rq_res
.page_len
+= size
;
817 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
818 struct splice_desc
*sd
)
820 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
824 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct file
*file
,
825 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
833 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
834 struct splice_desc sd
= {
841 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
842 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
846 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
851 nfsdstats
.io_read
+= host_err
;
854 fsnotify_access(file
);
856 err
= nfserrno(host_err
);
861 * Gathered writes: If another process is currently writing to the file,
862 * there's a high chance this is another nfsd (triggered by a bulk write
863 * from a client's biod). Rather than syncing the file with each write
864 * request, we sleep for 10 msec.
866 * I don't know if this roughly approximates C. Juszak's idea of
867 * gathered writes, but it's a nice and simple solution (IMHO), and it
870 * Note: we do this only in the NFSv2 case, since v3 and higher have a
871 * better tool (separate unstable writes and commits) for solving this
874 static int wait_for_concurrent_writes(struct file
*file
)
876 struct inode
*inode
= file_inode(file
);
877 static ino_t last_ino
;
878 static dev_t last_dev
;
881 if (atomic_read(&inode
->i_writecount
) > 1
882 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
883 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
885 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
888 if (inode
->i_state
& I_DIRTY
) {
889 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
890 err
= vfs_fsync(file
, 0);
892 last_ino
= inode
->i_ino
;
893 last_dev
= inode
->i_sb
->s_dev
;
898 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
899 loff_t offset
, struct kvec
*vec
, int vlen
,
900 unsigned long *cnt
, int *stablep
)
902 struct svc_export
*exp
;
903 struct dentry
*dentry
;
908 int stable
= *stablep
;
911 unsigned int pflags
= current
->flags
;
915 * We want less throttling in balance_dirty_pages()
916 * and shrink_inactive_list() so that nfs to
917 * localhost doesn't cause nfsd to lock up due to all
918 * the client's dirty pages or its congested queue.
920 current
->flags
|= PF_LESS_THROTTLE
;
922 dentry
= file
->f_path
.dentry
;
923 inode
= dentry
->d_inode
;
924 exp
= fhp
->fh_export
;
926 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
931 /* Write the data. */
932 oldfs
= get_fs(); set_fs(KERNEL_DS
);
933 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &pos
);
938 nfsdstats
.io_write
+= host_err
;
939 fsnotify_modify(file
);
943 host_err
= wait_for_concurrent_writes(file
);
945 host_err
= vfs_fsync_range(file
, offset
, offset
+*cnt
, 0);
949 dprintk("nfsd: write complete host_err=%d\n", host_err
);
953 err
= nfserrno(host_err
);
955 tsk_restore_flags(current
, pflags
, PF_LESS_THROTTLE
);
960 * Read data from a file. count must contain the requested read count
961 * on entry. On return, *count contains the number of bytes actually read.
962 * N.B. After this call fhp needs an fh_put
964 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
965 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
972 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
976 inode
= file_inode(file
);
978 /* Get readahead parameters */
979 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
982 file
->f_ra
= ra
->p_ra
;
984 err
= nfsd_vfs_read(rqstp
, file
, offset
, vec
, vlen
, count
);
986 /* Write back readahead params */
988 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
989 spin_lock(&rab
->pb_lock
);
990 ra
->p_ra
= file
->f_ra
;
993 spin_unlock(&rab
->pb_lock
);
1000 /* As above, but use the provided file descriptor. */
1002 nfsd_read_file(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1003 loff_t offset
, struct kvec
*vec
, int vlen
,
1004 unsigned long *count
)
1009 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1010 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1013 err
= nfsd_vfs_read(rqstp
, file
, offset
, vec
, vlen
, count
);
1014 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1015 err
= nfsd_read(rqstp
, fhp
, offset
, vec
, vlen
, count
);
1021 * Write data to a file.
1022 * The stable flag requests synchronous writes.
1023 * N.B. After this call fhp needs an fh_put
1026 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1027 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1033 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1034 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1037 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1040 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1045 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1053 #ifdef CONFIG_NFSD_V3
1055 * Commit all pending writes to stable storage.
1057 * Note: we only guarantee that data that lies within the range specified
1058 * by the 'offset' and 'count' parameters will be synced.
1060 * Unfortunately we cannot lock the file to make sure we return full WCC
1061 * data to the client, as locking happens lower down in the filesystem.
1064 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1065 loff_t offset
, unsigned long count
)
1068 loff_t end
= LLONG_MAX
;
1069 __be32 err
= nfserr_inval
;
1074 end
= offset
+ (loff_t
)count
- 1;
1079 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1080 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1083 if (EX_ISSYNC(fhp
->fh_export
)) {
1084 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1086 if (err2
!= -EINVAL
)
1087 err
= nfserrno(err2
);
1089 err
= nfserr_notsupp
;
1096 #endif /* CONFIG_NFSD_V3 */
1099 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1103 * Mode has already been set earlier in create:
1105 iap
->ia_valid
&= ~ATTR_MODE
;
1107 * Setting uid/gid works only for root. Irix appears to
1108 * send along the gid on create when it tries to implement
1109 * setgid directories via NFS:
1111 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID
))
1112 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1114 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1118 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1119 * setting size to 0 may fail for some specific file systems by the permission
1120 * checking which requires WRITE permission but the mode is 000.
1121 * we ignore the resizing(to 0) on the just new created file, since the size is
1122 * 0 after file created.
1124 * call this only after vfs_create() is called.
1127 nfsd_check_ignore_resizing(struct iattr
*iap
)
1129 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1130 iap
->ia_valid
&= ~ATTR_SIZE
;
1134 * Create a file (regular, directory, device, fifo); UNIX sockets
1135 * not yet implemented.
1136 * If the response fh has been verified, the parent directory should
1137 * already be locked. Note that the parent directory is left locked.
1139 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1142 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1143 char *fname
, int flen
, struct iattr
*iap
,
1144 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1146 struct dentry
*dentry
, *dchild
= NULL
;
1156 if (isdotent(fname
, flen
))
1159 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1163 dentry
= fhp
->fh_dentry
;
1164 dirp
= dentry
->d_inode
;
1166 err
= nfserr_notdir
;
1167 if (!dirp
->i_op
->lookup
)
1170 * Check whether the response file handle has been verified yet.
1171 * If it has, the parent directory should already be locked.
1173 if (!resfhp
->fh_dentry
) {
1174 host_err
= fh_want_write(fhp
);
1178 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1179 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1180 dchild
= lookup_one_len(fname
, dentry
, flen
);
1181 host_err
= PTR_ERR(dchild
);
1184 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1188 /* called from nfsd_proc_create */
1189 dchild
= dget(resfhp
->fh_dentry
);
1190 if (!fhp
->fh_locked
) {
1191 /* not actually possible */
1193 "nfsd_create: parent %pd2 not locked!\n",
1200 * Make sure the child dentry is still negative ...
1203 if (dchild
->d_inode
) {
1204 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1209 if (!(iap
->ia_valid
& ATTR_MODE
))
1211 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1214 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1215 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1221 * Get the dir op function pointer.
1227 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1229 nfsd_check_ignore_resizing(iap
);
1232 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1238 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1244 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1247 * nfsd_setattr already committed the child. Transactional filesystems
1248 * had a chance to commit changes for both parent and child
1249 * simultaneously making the following commit_metadata a noop.
1251 err2
= nfserrno(commit_metadata(fhp
));
1255 * Update the file handle to get the new inode info.
1258 err
= fh_update(resfhp
);
1260 if (dchild
&& !IS_ERR(dchild
))
1265 err
= nfserrno(host_err
);
1269 #ifdef CONFIG_NFSD_V3
1271 static inline int nfsd_create_is_exclusive(int createmode
)
1273 return createmode
== NFS3_CREATE_EXCLUSIVE
1274 || createmode
== NFS4_CREATE_EXCLUSIVE4_1
;
1278 * NFSv3 and NFSv4 version of nfsd_create
1281 do_nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1282 char *fname
, int flen
, struct iattr
*iap
,
1283 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1284 bool *truncp
, bool *created
)
1286 struct dentry
*dentry
, *dchild
= NULL
;
1290 __u32 v_mtime
=0, v_atime
=0;
1296 if (isdotent(fname
, flen
))
1298 if (!(iap
->ia_valid
& ATTR_MODE
))
1300 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
1304 dentry
= fhp
->fh_dentry
;
1305 dirp
= dentry
->d_inode
;
1307 /* Get all the sanity checks out of the way before
1308 * we lock the parent. */
1309 err
= nfserr_notdir
;
1310 if (!dirp
->i_op
->lookup
)
1313 host_err
= fh_want_write(fhp
);
1317 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1320 * Compose the response file handle.
1322 dchild
= lookup_one_len(fname
, dentry
, flen
);
1323 host_err
= PTR_ERR(dchild
);
1327 /* If file doesn't exist, check for permissions to create one */
1328 if (!dchild
->d_inode
) {
1329 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1334 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1338 if (nfsd_create_is_exclusive(createmode
)) {
1339 /* solaris7 gets confused (bugid 4218508) if these have
1340 * the high bit set, so just clear the high bits. If this is
1341 * ever changed to use different attrs for storing the
1342 * verifier, then do_open_lookup() will also need to be fixed
1345 v_mtime
= verifier
[0]&0x7fffffff;
1346 v_atime
= verifier
[1]&0x7fffffff;
1349 if (dchild
->d_inode
) {
1352 switch (createmode
) {
1353 case NFS3_CREATE_UNCHECKED
:
1354 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1357 /* in nfsv4, we need to treat this case a little
1358 * differently. we don't want to truncate the
1359 * file now; this would be wrong if the OPEN
1360 * fails for some other reason. furthermore,
1361 * if the size is nonzero, we should ignore it
1362 * according to spec!
1364 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1367 iap
->ia_valid
&= ATTR_SIZE
;
1371 case NFS3_CREATE_EXCLUSIVE
:
1372 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1373 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1374 && dchild
->d_inode
->i_size
== 0 ) {
1379 case NFS4_CREATE_EXCLUSIVE4_1
:
1380 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1381 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1382 && dchild
->d_inode
->i_size
== 0 ) {
1388 case NFS3_CREATE_GUARDED
:
1395 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1403 nfsd_check_ignore_resizing(iap
);
1405 if (nfsd_create_is_exclusive(createmode
)) {
1406 /* Cram the verifier into atime/mtime */
1407 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1408 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1409 /* XXX someone who knows this better please fix it for nsec */
1410 iap
->ia_mtime
.tv_sec
= v_mtime
;
1411 iap
->ia_atime
.tv_sec
= v_atime
;
1412 iap
->ia_mtime
.tv_nsec
= 0;
1413 iap
->ia_atime
.tv_nsec
= 0;
1417 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1420 * nfsd_setattr already committed the child (and possibly also the parent).
1423 err
= nfserrno(commit_metadata(fhp
));
1426 * Update the filehandle to get the new inode info.
1429 err
= fh_update(resfhp
);
1433 if (dchild
&& !IS_ERR(dchild
))
1439 err
= nfserrno(host_err
);
1442 #endif /* CONFIG_NFSD_V3 */
1445 * Read a symlink. On entry, *lenp must contain the maximum path length that
1446 * fits into the buffer. On return, it contains the true length.
1447 * N.B. After this call fhp needs an fh_put
1450 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1452 struct inode
*inode
;
1458 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1462 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
1463 path
.dentry
= fhp
->fh_dentry
;
1464 inode
= path
.dentry
->d_inode
;
1467 if (!inode
->i_op
->readlink
)
1471 /* N.B. Why does this call need a get_fs()??
1472 * Remove the set_fs and watch the fireworks:-) --okir
1475 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1476 host_err
= inode
->i_op
->readlink(path
.dentry
, (char __user
*)buf
, *lenp
);
1487 err
= nfserrno(host_err
);
1492 * Create a symlink and look up its inode
1493 * N.B. After this call _both_ fhp and resfhp need an fh_put
1496 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1497 char *fname
, int flen
,
1498 char *path
, int plen
,
1499 struct svc_fh
*resfhp
,
1502 struct dentry
*dentry
, *dnew
;
1510 if (isdotent(fname
, flen
))
1513 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1517 host_err
= fh_want_write(fhp
);
1522 dentry
= fhp
->fh_dentry
;
1523 dnew
= lookup_one_len(fname
, dentry
, flen
);
1524 host_err
= PTR_ERR(dnew
);
1528 if (unlikely(path
[plen
] != 0)) {
1529 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1530 if (path_alloced
== NULL
)
1533 strncpy(path_alloced
, path
, plen
);
1534 path_alloced
[plen
] = 0;
1535 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1536 kfree(path_alloced
);
1539 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1540 err
= nfserrno(host_err
);
1542 err
= nfserrno(commit_metadata(fhp
));
1547 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1549 if (err
==0) err
= cerr
;
1554 err
= nfserrno(host_err
);
1560 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1563 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1564 char *name
, int len
, struct svc_fh
*tfhp
)
1566 struct dentry
*ddir
, *dnew
, *dold
;
1571 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1574 err
= fh_verify(rqstp
, tfhp
, 0, NFSD_MAY_NOP
);
1578 if (S_ISDIR(tfhp
->fh_dentry
->d_inode
->i_mode
))
1584 if (isdotent(name
, len
))
1587 host_err
= fh_want_write(tfhp
);
1589 err
= nfserrno(host_err
);
1593 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1594 ddir
= ffhp
->fh_dentry
;
1595 dirp
= ddir
->d_inode
;
1597 dnew
= lookup_one_len(name
, ddir
, len
);
1598 host_err
= PTR_ERR(dnew
);
1602 dold
= tfhp
->fh_dentry
;
1607 host_err
= vfs_link(dold
, dirp
, dnew
, NULL
);
1609 err
= nfserrno(commit_metadata(ffhp
));
1611 err
= nfserrno(commit_metadata(tfhp
));
1613 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1616 err
= nfserrno(host_err
);
1622 fh_drop_write(tfhp
);
1627 err
= nfserrno(host_err
);
1633 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1636 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1637 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1639 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1640 struct inode
*fdir
, *tdir
;
1644 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1647 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1651 fdentry
= ffhp
->fh_dentry
;
1652 fdir
= fdentry
->d_inode
;
1654 tdentry
= tfhp
->fh_dentry
;
1655 tdir
= tdentry
->d_inode
;
1658 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1661 host_err
= fh_want_write(ffhp
);
1663 err
= nfserrno(host_err
);
1667 /* cannot use fh_lock as we need deadlock protective ordering
1668 * so do it by hand */
1669 trap
= lock_rename(tdentry
, fdentry
);
1670 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1674 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1675 host_err
= PTR_ERR(odentry
);
1676 if (IS_ERR(odentry
))
1680 if (!odentry
->d_inode
)
1683 if (odentry
== trap
)
1686 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1687 host_err
= PTR_ERR(ndentry
);
1688 if (IS_ERR(ndentry
))
1690 host_err
= -ENOTEMPTY
;
1691 if (ndentry
== trap
)
1695 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1697 if (ffhp
->fh_export
->ex_path
.dentry
!= tfhp
->fh_export
->ex_path
.dentry
)
1700 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
, NULL
, 0);
1702 host_err
= commit_metadata(tfhp
);
1704 host_err
= commit_metadata(ffhp
);
1711 err
= nfserrno(host_err
);
1713 * We cannot rely on fh_unlock on the two filehandles,
1714 * as that would do the wrong thing if the two directories
1715 * were the same, so again we do it by hand.
1717 fill_post_wcc(ffhp
);
1718 fill_post_wcc(tfhp
);
1719 unlock_rename(tdentry
, fdentry
);
1720 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1721 fh_drop_write(ffhp
);
1728 * Unlink a file or directory
1729 * N.B. After this call fhp needs an fh_put
1732 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1733 char *fname
, int flen
)
1735 struct dentry
*dentry
, *rdentry
;
1741 if (!flen
|| isdotent(fname
, flen
))
1743 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1747 host_err
= fh_want_write(fhp
);
1751 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1752 dentry
= fhp
->fh_dentry
;
1753 dirp
= dentry
->d_inode
;
1755 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1756 host_err
= PTR_ERR(rdentry
);
1757 if (IS_ERR(rdentry
))
1760 if (!rdentry
->d_inode
) {
1767 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1769 if (type
!= S_IFDIR
)
1770 host_err
= vfs_unlink(dirp
, rdentry
, NULL
);
1772 host_err
= vfs_rmdir(dirp
, rdentry
);
1774 host_err
= commit_metadata(fhp
);
1778 err
= nfserrno(host_err
);
1784 * We do this buffering because we must not call back into the file
1785 * system's ->lookup() method from the filldir callback. That may well
1786 * deadlock a number of file systems.
1788 * This is based heavily on the implementation of same in XFS.
1790 struct buffered_dirent
{
1794 unsigned int d_type
;
1798 struct readdir_data
{
1799 struct dir_context ctx
;
1805 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1806 loff_t offset
, u64 ino
, unsigned int d_type
)
1808 struct readdir_data
*buf
= __buf
;
1809 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1810 unsigned int reclen
;
1812 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1813 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1818 de
->namlen
= namlen
;
1819 de
->offset
= offset
;
1821 de
->d_type
= d_type
;
1822 memcpy(de
->name
, name
, namlen
);
1823 buf
->used
+= reclen
;
1828 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1829 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1831 struct buffered_dirent
*de
;
1835 struct readdir_data buf
= {
1836 .ctx
.actor
= nfsd_buffered_filldir
,
1837 .dirent
= (void *)__get_free_page(GFP_KERNEL
)
1841 return nfserrno(-ENOMEM
);
1846 struct inode
*dir_inode
= file_inode(file
);
1847 unsigned int reclen
;
1849 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1853 host_err
= iterate_dir(file
, &buf
.ctx
);
1866 * Various filldir functions may end up calling back into
1867 * lookup_one_len() and the file system's ->lookup() method.
1868 * These expect i_mutex to be held, as it would within readdir.
1870 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1874 de
= (struct buffered_dirent
*)buf
.dirent
;
1876 offset
= de
->offset
;
1878 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1879 de
->ino
, de
->d_type
))
1882 if (cdp
->err
!= nfs_ok
)
1885 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1888 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1890 mutex_unlock(&dir_inode
->i_mutex
);
1891 if (size
> 0) /* We bailed out early */
1894 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1897 free_page((unsigned long)(buf
.dirent
));
1900 return nfserrno(host_err
);
1907 * Read entries from a directory.
1908 * The NFSv3/4 verifier we ignore for now.
1911 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1912 struct readdir_cd
*cdp
, filldir_t func
)
1916 loff_t offset
= *offsetp
;
1917 int may_flags
= NFSD_MAY_READ
;
1919 /* NFSv2 only supports 32 bit cookies */
1920 if (rqstp
->rq_vers
> 2)
1921 may_flags
|= NFSD_MAY_64BIT_COOKIE
;
1923 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, may_flags
, &file
);
1927 offset
= vfs_llseek(file
, offset
, SEEK_SET
);
1929 err
= nfserrno((int)offset
);
1933 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1935 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1936 err
= nfs_ok
; /* can still be found in ->err */
1944 * Get file system stats
1945 * N.B. After this call fhp needs an fh_put
1948 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
1952 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
1954 struct path path
= {
1955 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
1956 .dentry
= fhp
->fh_dentry
,
1958 if (vfs_statfs(&path
, stat
))
1964 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
1966 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
1970 * Check for a user's access permissions to this inode.
1973 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
1974 struct dentry
*dentry
, int acc
)
1976 struct inode
*inode
= dentry
->d_inode
;
1979 if ((acc
& NFSD_MAY_MASK
) == NFSD_MAY_NOP
)
1982 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1984 (acc
& NFSD_MAY_READ
)? " read" : "",
1985 (acc
& NFSD_MAY_WRITE
)? " write" : "",
1986 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
1987 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
1988 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
1989 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
1990 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
1992 IS_IMMUTABLE(inode
)? " immut" : "",
1993 IS_APPEND(inode
)? " append" : "",
1994 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
1995 dprintk(" owner %d/%d user %d/%d\n",
1996 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
1999 /* Normally we reject any write/sattr etc access on a read-only file
2000 * system. But if it is IRIX doing check on write-access for a
2001 * device special file, we ignore rofs.
2003 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2004 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2005 if (exp_rdonly(rqstp
, exp
) ||
2006 __mnt_is_readonly(exp
->ex_path
.mnt
))
2008 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2011 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2014 if (acc
& NFSD_MAY_LOCK
) {
2015 /* If we cannot rely on authentication in NLM requests,
2016 * just allow locks, otherwise require read permission, or
2019 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2022 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2025 * The file owner always gets access permission for accesses that
2026 * would normally be checked at open time. This is to make
2027 * file access work even when the client has done a fchmod(fd, 0).
2029 * However, `cp foo bar' should fail nevertheless when bar is
2030 * readonly. A sensible way to do this might be to reject all
2031 * attempts to truncate a read-only file, because a creat() call
2032 * always implies file truncation.
2033 * ... but this isn't really fair. A process may reasonably call
2034 * ftruncate on an open file descriptor on a file with perm 000.
2035 * We must trust the client to do permission checking - using "ACCESS"
2038 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2039 uid_eq(inode
->i_uid
, current_fsuid()))
2042 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2043 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2045 /* Allow read access to binaries even when mode 111 */
2046 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2047 (acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
) ||
2048 acc
== (NFSD_MAY_READ
| NFSD_MAY_READ_IF_EXEC
)))
2049 err
= inode_permission(inode
, MAY_EXEC
);
2051 return err
? nfserrno(err
) : 0;
2055 nfsd_racache_shutdown(void)
2057 struct raparms
*raparm
, *last_raparm
;
2060 dprintk("nfsd: freeing readahead buffers.\n");
2062 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2063 raparm
= raparm_hash
[i
].pb_head
;
2065 last_raparm
= raparm
;
2066 raparm
= raparm
->p_next
;
2069 raparm_hash
[i
].pb_head
= NULL
;
2073 * Initialize readahead param cache
2076 nfsd_racache_init(int cache_size
)
2081 struct raparms
**raparm
= NULL
;
2084 if (raparm_hash
[0].pb_head
)
2086 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2089 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2091 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2093 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2094 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2096 raparm
= &raparm_hash
[i
].pb_head
;
2097 for (j
= 0; j
< nperbucket
; j
++) {
2098 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2101 raparm
= &(*raparm
)->p_next
;
2106 nfsdstats
.ra_size
= cache_size
;
2110 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2111 nfsd_racache_shutdown();