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 */
41 #endif /* CONFIG_NFSD_V4 */
46 #define NFSDDBG_FACILITY NFSDDBG_FILEOP
50 * This is a cache of readahead params that help us choose the proper
51 * readahead strategy. Initially, we set all readahead parameters to 0
52 * and let the VFS handle things.
53 * If you increase the number of cached files very much, you'll need to
54 * add a hash table here.
57 struct raparms
*p_next
;
62 struct file_ra_state p_ra
;
63 unsigned int p_hindex
;
66 struct raparm_hbucket
{
67 struct raparms
*pb_head
;
69 } ____cacheline_aligned_in_smp
;
71 #define RAPARM_HASH_BITS 4
72 #define RAPARM_HASH_SIZE (1<<RAPARM_HASH_BITS)
73 #define RAPARM_HASH_MASK (RAPARM_HASH_SIZE-1)
74 static struct raparm_hbucket raparm_hash
[RAPARM_HASH_SIZE
];
77 * Called from nfsd_lookup and encode_dirent. Check if we have crossed
79 * Returns -EAGAIN or -ETIMEDOUT leaving *dpp and *expp unchanged,
80 * or nfs_ok having possibly changed *dpp and *expp
83 nfsd_cross_mnt(struct svc_rqst
*rqstp
, struct dentry
**dpp
,
84 struct svc_export
**expp
)
86 struct svc_export
*exp
= *expp
, *exp2
= NULL
;
87 struct dentry
*dentry
= *dpp
;
88 struct path path
= {.mnt
= mntget(exp
->ex_path
.mnt
),
89 .dentry
= dget(dentry
)};
92 err
= follow_down(&path
);
95 if (path
.mnt
== exp
->ex_path
.mnt
&& path
.dentry
== dentry
&&
96 nfsd_mountpoint(dentry
, exp
) == 2) {
97 /* This is only a mountpoint in some other namespace */
102 exp2
= rqst_exp_get_by_name(rqstp
, &path
);
106 * We normally allow NFS clients to continue
107 * "underneath" a mountpoint that is not exported.
108 * The exception is V4ROOT, where no traversal is ever
109 * allowed without an explicit export of the new
112 if (err
== -ENOENT
&& !(exp
->ex_flags
& NFSEXP_V4ROOT
))
117 if (nfsd_v4client(rqstp
) ||
118 (exp
->ex_flags
& NFSEXP_CROSSMOUNT
) || EX_NOHIDE(exp2
)) {
119 /* successfully crossed mount point */
121 * This is subtle: path.dentry is *not* on path.mnt
122 * at this point. The only reason we are safe is that
123 * original mnt is pinned down by exp, so we should
124 * put path *before* putting exp
127 path
.dentry
= dentry
;
137 static void follow_to_parent(struct path
*path
)
141 while (path
->dentry
== path
->mnt
->mnt_root
&& follow_up(path
))
143 dp
= dget_parent(path
->dentry
);
148 static int nfsd_lookup_parent(struct svc_rqst
*rqstp
, struct dentry
*dparent
, struct svc_export
**exp
, struct dentry
**dentryp
)
150 struct svc_export
*exp2
;
151 struct path path
= {.mnt
= mntget((*exp
)->ex_path
.mnt
),
152 .dentry
= dget(dparent
)};
154 follow_to_parent(&path
);
156 exp2
= rqst_exp_parent(rqstp
, &path
);
157 if (PTR_ERR(exp2
) == -ENOENT
) {
158 *dentryp
= dget(dparent
);
159 } else if (IS_ERR(exp2
)) {
161 return PTR_ERR(exp2
);
163 *dentryp
= dget(path
.dentry
);
172 * For nfsd purposes, we treat V4ROOT exports as though there was an
173 * export at *every* directory.
175 * '1' if this dentry *must* be an export point,
176 * '2' if it might be, if there is really a mount here, and
177 * '0' if there is no chance of an export point here.
179 int nfsd_mountpoint(struct dentry
*dentry
, struct svc_export
*exp
)
181 if (!d_inode(dentry
))
183 if (exp
->ex_flags
& NFSEXP_V4ROOT
)
185 if (nfsd4_is_junction(dentry
))
187 if (d_mountpoint(dentry
))
189 * Might only be a mountpoint in a different namespace,
190 * but we need to check.
197 nfsd_lookup_dentry(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
198 const char *name
, unsigned int len
,
199 struct svc_export
**exp_ret
, struct dentry
**dentry_ret
)
201 struct svc_export
*exp
;
202 struct dentry
*dparent
;
203 struct dentry
*dentry
;
206 dprintk("nfsd: nfsd_lookup(fh %s, %.*s)\n", SVCFH_fmt(fhp
), len
,name
);
208 dparent
= fhp
->fh_dentry
;
209 exp
= exp_get(fhp
->fh_export
);
211 /* Lookup the name, but don't follow links */
212 if (isdotent(name
, len
)) {
214 dentry
= dget(dparent
);
215 else if (dparent
!= exp
->ex_path
.dentry
)
216 dentry
= dget_parent(dparent
);
217 else if (!EX_NOHIDE(exp
) && !nfsd_v4client(rqstp
))
218 dentry
= dget(dparent
); /* .. == . just like at / */
220 /* checking mountpoint crossing is very different when stepping up */
221 host_err
= nfsd_lookup_parent(rqstp
, dparent
, &exp
, &dentry
);
227 * In the nfsd4_open() case, this may be held across
228 * subsequent open and delegation acquisition which may
229 * need to take the child's i_mutex:
231 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
232 dentry
= lookup_one_len(name
, dparent
, len
);
233 host_err
= PTR_ERR(dentry
);
237 * check if we have crossed a mount point ...
239 if (nfsd_mountpoint(dentry
, exp
)) {
240 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
246 *dentry_ret
= dentry
;
252 return nfserrno(host_err
);
256 * Look up one component of a pathname.
257 * N.B. After this call _both_ fhp and resfh need an fh_put
259 * If the lookup would cross a mountpoint, and the mounted filesystem
260 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
261 * accepted as it stands and the mounted directory is
262 * returned. Otherwise the covered directory is returned.
263 * NOTE: this mountpoint crossing is not supported properly by all
264 * clients and is explicitly disallowed for NFSv3
265 * NeilBrown <neilb@cse.unsw.edu.au>
268 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
269 unsigned int len
, struct svc_fh
*resfh
)
271 struct svc_export
*exp
;
272 struct dentry
*dentry
;
275 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
278 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
281 err
= check_nfsd_access(exp
, rqstp
);
285 * Note: we compose the file handle now, but as the
286 * dentry may be negative, it may need to be updated.
288 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
289 if (!err
&& d_really_is_negative(dentry
))
298 * Commit metadata changes to stable storage.
301 commit_metadata(struct svc_fh
*fhp
)
303 struct inode
*inode
= d_inode(fhp
->fh_dentry
);
304 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
306 if (!EX_ISSYNC(fhp
->fh_export
))
309 if (export_ops
->commit_metadata
)
310 return export_ops
->commit_metadata(inode
);
311 return sync_inode_metadata(inode
, 1);
315 * Go over the attributes and take care of the small differences between
316 * NFS semantics and what Linux expects.
319 nfsd_sanitize_attrs(struct inode
*inode
, struct iattr
*iap
)
321 /* sanitize the mode change */
322 if (iap
->ia_valid
& ATTR_MODE
) {
323 iap
->ia_mode
&= S_IALLUGO
;
324 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
327 /* Revoke setuid/setgid on chown */
328 if (!S_ISDIR(inode
->i_mode
) &&
329 ((iap
->ia_valid
& ATTR_UID
) || (iap
->ia_valid
& ATTR_GID
))) {
330 iap
->ia_valid
|= ATTR_KILL_PRIV
;
331 if (iap
->ia_valid
& ATTR_MODE
) {
332 /* we're setting mode too, just clear the s*id bits */
333 iap
->ia_mode
&= ~S_ISUID
;
334 if (iap
->ia_mode
& S_IXGRP
)
335 iap
->ia_mode
&= ~S_ISGID
;
337 /* set ATTR_KILL_* bits and let VFS handle it */
338 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
344 nfsd_get_write_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
347 struct inode
*inode
= d_inode(fhp
->fh_dentry
);
350 if (iap
->ia_size
< inode
->i_size
) {
353 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
354 NFSD_MAY_TRUNC
| NFSD_MAY_OWNER_OVERRIDE
);
359 host_err
= get_write_access(inode
);
363 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
365 goto out_put_write_access
;
368 out_put_write_access
:
369 put_write_access(inode
);
371 return nfserrno(host_err
);
375 * Set various file attributes. After this call fhp needs an fh_put.
378 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
379 int check_guard
, time_t guardtime
)
381 struct dentry
*dentry
;
383 int accmode
= NFSD_MAY_SATTR
;
387 bool get_write_count
;
388 bool size_change
= (iap
->ia_valid
& ATTR_SIZE
);
390 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
391 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
392 if (iap
->ia_valid
& ATTR_SIZE
)
395 /* Callers that do fh_verify should do the fh_want_write: */
396 get_write_count
= !fhp
->fh_dentry
;
399 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
402 if (get_write_count
) {
403 host_err
= fh_want_write(fhp
);
408 dentry
= fhp
->fh_dentry
;
409 inode
= d_inode(dentry
);
411 /* Ignore any mode updates on symlinks */
412 if (S_ISLNK(inode
->i_mode
))
413 iap
->ia_valid
&= ~ATTR_MODE
;
418 nfsd_sanitize_attrs(inode
, iap
);
420 if (check_guard
&& guardtime
!= inode
->i_ctime
.tv_sec
)
421 return nfserr_notsync
;
424 * The size case is special, it changes the file in addition to the
425 * attributes, and file systems don't expect it to be mixed with
426 * "random" attribute changes. We thus split out the size change
427 * into a separate call to ->setattr, and do the rest as a separate
431 err
= nfsd_get_write_access(rqstp
, fhp
, iap
);
439 * RFC5661, Section 18.30.4:
440 * Changing the size of a file with SETATTR indirectly
441 * changes the time_modify and change attributes.
443 * (and similar for the older RFCs)
445 struct iattr size_attr
= {
446 .ia_valid
= ATTR_SIZE
| ATTR_CTIME
| ATTR_MTIME
,
447 .ia_size
= iap
->ia_size
,
450 host_err
= notify_change(dentry
, &size_attr
, NULL
);
453 iap
->ia_valid
&= ~ATTR_SIZE
;
456 * Avoid the additional setattr call below if the only other
457 * attribute that the client sends is the mtime, as we update
458 * it as part of the size change above.
460 if ((iap
->ia_valid
& ~ATTR_MTIME
) == 0)
464 iap
->ia_valid
|= ATTR_CTIME
;
465 host_err
= notify_change(dentry
, iap
, NULL
);
470 put_write_access(inode
);
473 host_err
= commit_metadata(fhp
);
474 return nfserrno(host_err
);
477 #if defined(CONFIG_NFSD_V4)
479 * NFS junction information is stored in an extended attribute.
481 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
484 * nfsd4_is_junction - Test if an object could be an NFS junction
486 * @dentry: object to test
488 * Returns 1 if "dentry" appears to contain NFS junction information.
489 * Otherwise 0 is returned.
491 int nfsd4_is_junction(struct dentry
*dentry
)
493 struct inode
*inode
= d_inode(dentry
);
497 if (inode
->i_mode
& S_IXUGO
)
499 if (!(inode
->i_mode
& S_ISVTX
))
501 if (vfs_getxattr(dentry
, NFSD_JUNCTION_XATTR_NAME
, NULL
, 0) <= 0)
505 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
506 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
507 struct xdr_netobj
*label
)
511 struct dentry
*dentry
;
513 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
517 dentry
= fhp
->fh_dentry
;
519 mutex_lock(&d_inode(dentry
)->i_mutex
);
520 host_error
= security_inode_setsecctx(dentry
, label
->data
, label
->len
);
521 mutex_unlock(&d_inode(dentry
)->i_mutex
);
522 return nfserrno(host_error
);
525 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
526 struct xdr_netobj
*label
)
528 return nfserr_notsupp
;
532 __be32
nfsd4_vfs_fallocate(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
533 struct file
*file
, loff_t offset
, loff_t len
,
538 if (!S_ISREG(file_inode(file
)->i_mode
))
541 error
= vfs_fallocate(file
, flags
, offset
, len
);
543 error
= commit_metadata(fhp
);
545 return nfserrno(error
);
547 #endif /* defined(CONFIG_NFSD_V4) */
549 #ifdef CONFIG_NFSD_V3
551 * Check server access rights to a file system object
557 static struct accessmap nfs3_regaccess
[] = {
558 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
559 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
560 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
561 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
566 static struct accessmap nfs3_diraccess
[] = {
567 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
568 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
569 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
570 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
571 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
576 static struct accessmap nfs3_anyaccess
[] = {
577 /* Some clients - Solaris 2.6 at least, make an access call
578 * to the server to check for access for things like /dev/null
579 * (which really, the server doesn't care about). So
580 * We provide simple access checking for them, looking
581 * mainly at mode bits, and we make sure to ignore read-only
584 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
585 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
586 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
587 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
593 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
595 struct accessmap
*map
;
596 struct svc_export
*export
;
597 struct dentry
*dentry
;
598 u32 query
, result
= 0, sresult
= 0;
601 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
605 export
= fhp
->fh_export
;
606 dentry
= fhp
->fh_dentry
;
608 if (d_is_reg(dentry
))
609 map
= nfs3_regaccess
;
610 else if (d_is_dir(dentry
))
611 map
= nfs3_diraccess
;
613 map
= nfs3_anyaccess
;
617 for (; map
->access
; map
++) {
618 if (map
->access
& query
) {
621 sresult
|= map
->access
;
623 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
626 result
|= map
->access
;
629 /* the following error codes just mean the access was not allowed,
630 * rather than an error occurred */
634 /* simply don't "or" in the access bit. */
644 *supported
= sresult
;
649 #endif /* CONFIG_NFSD_V3 */
651 static int nfsd_open_break_lease(struct inode
*inode
, int access
)
655 if (access
& NFSD_MAY_NOT_BREAK_LEASE
)
657 mode
= (access
& NFSD_MAY_WRITE
) ? O_WRONLY
: O_RDONLY
;
658 return break_lease(inode
, mode
| O_NONBLOCK
);
662 * Open an existing file or directory.
663 * The may_flags argument indicates the type of open (read/write/lock)
664 * and additional flags.
665 * N.B. After this call fhp needs an fh_put
668 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
,
669 int may_flags
, struct file
**filp
)
674 int flags
= O_RDONLY
|O_LARGEFILE
;
678 validate_process_creds();
681 * If we get here, then the client has already done an "open",
682 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
683 * in case a chmod has now revoked permission.
685 * Arguably we should also allow the owner override for
686 * directories, but we never have and it doesn't seem to have
687 * caused anyone a problem. If we were to change this, note
688 * also that our filldir callbacks would need a variant of
689 * lookup_one_len that doesn't check permissions.
692 may_flags
|= NFSD_MAY_OWNER_OVERRIDE
;
693 err
= fh_verify(rqstp
, fhp
, type
, may_flags
);
697 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
698 path
.dentry
= fhp
->fh_dentry
;
699 inode
= d_inode(path
.dentry
);
701 /* Disallow write access to files with the append-only bit set
702 * or any access when mandatory locking enabled
705 if (IS_APPEND(inode
) && (may_flags
& NFSD_MAY_WRITE
))
708 * We must ignore files (but only files) which might have mandatory
709 * locks on them because there is no way to know if the accesser has
712 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
718 host_err
= nfsd_open_break_lease(inode
, may_flags
);
719 if (host_err
) /* NOMEM or WOULDBLOCK */
722 if (may_flags
& NFSD_MAY_WRITE
) {
723 if (may_flags
& NFSD_MAY_READ
)
724 flags
= O_RDWR
|O_LARGEFILE
;
726 flags
= O_WRONLY
|O_LARGEFILE
;
729 file
= dentry_open(&path
, flags
, current_cred());
731 host_err
= PTR_ERR(file
);
735 host_err
= ima_file_check(file
, may_flags
, 0);
741 if (may_flags
& NFSD_MAY_64BIT_COOKIE
)
742 file
->f_mode
|= FMODE_64BITHASH
;
744 file
->f_mode
|= FMODE_32BITHASH
;
748 err
= nfserrno(host_err
);
750 validate_process_creds();
755 nfsd_init_raparms(struct file
*file
)
757 struct inode
*inode
= file_inode(file
);
758 dev_t dev
= inode
->i_sb
->s_dev
;
759 ino_t ino
= inode
->i_ino
;
760 struct raparms
*ra
, **rap
, **frap
= NULL
;
763 struct raparm_hbucket
*rab
;
765 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
766 rab
= &raparm_hash
[hash
];
768 spin_lock(&rab
->pb_lock
);
769 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
770 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
773 if (ra
->p_count
== 0)
776 depth
= nfsdstats
.ra_size
;
778 spin_unlock(&rab
->pb_lock
);
788 if (rap
!= &rab
->pb_head
) {
790 ra
->p_next
= rab
->pb_head
;
794 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
795 spin_unlock(&rab
->pb_lock
);
798 file
->f_ra
= ra
->p_ra
;
802 void nfsd_put_raparams(struct file
*file
, struct raparms
*ra
)
804 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
806 spin_lock(&rab
->pb_lock
);
807 ra
->p_ra
= file
->f_ra
;
810 spin_unlock(&rab
->pb_lock
);
814 * Grab and keep cached pages associated with a file in the svc_rqst
815 * so that they can be passed to the network sendmsg/sendpage routines
816 * directly. They will be released after the sending has completed.
819 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
820 struct splice_desc
*sd
)
822 struct svc_rqst
*rqstp
= sd
->u
.data
;
823 struct page
**pp
= rqstp
->rq_next_page
;
824 struct page
*page
= buf
->page
;
829 if (rqstp
->rq_res
.page_len
== 0) {
831 put_page(*rqstp
->rq_next_page
);
832 *(rqstp
->rq_next_page
++) = page
;
833 rqstp
->rq_res
.page_base
= buf
->offset
;
834 rqstp
->rq_res
.page_len
= size
;
835 } else if (page
!= pp
[-1]) {
837 if (*rqstp
->rq_next_page
)
838 put_page(*rqstp
->rq_next_page
);
839 *(rqstp
->rq_next_page
++) = page
;
840 rqstp
->rq_res
.page_len
+= size
;
842 rqstp
->rq_res
.page_len
+= size
;
847 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
848 struct splice_desc
*sd
)
850 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
854 nfsd_finish_read(struct file
*file
, unsigned long *count
, int host_err
)
857 nfsdstats
.io_read
+= host_err
;
859 fsnotify_access(file
);
862 return nfserrno(host_err
);
865 __be32
nfsd_splice_read(struct svc_rqst
*rqstp
,
866 struct file
*file
, loff_t offset
, unsigned long *count
)
868 struct splice_desc sd
= {
876 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
877 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
878 return nfsd_finish_read(file
, count
, host_err
);
881 __be32
nfsd_readv(struct file
*file
, loff_t offset
, struct kvec
*vec
, int vlen
,
882 unsigned long *count
)
889 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
891 return nfsd_finish_read(file
, count
, host_err
);
895 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct file
*file
,
896 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
898 if (file
->f_op
->splice_read
&& test_bit(RQ_SPLICE_OK
, &rqstp
->rq_flags
))
899 return nfsd_splice_read(rqstp
, file
, offset
, count
);
901 return nfsd_readv(file
, offset
, vec
, vlen
, count
);
905 * Gathered writes: If another process is currently writing to the file,
906 * there's a high chance this is another nfsd (triggered by a bulk write
907 * from a client's biod). Rather than syncing the file with each write
908 * request, we sleep for 10 msec.
910 * I don't know if this roughly approximates C. Juszak's idea of
911 * gathered writes, but it's a nice and simple solution (IMHO), and it
914 * Note: we do this only in the NFSv2 case, since v3 and higher have a
915 * better tool (separate unstable writes and commits) for solving this
918 static int wait_for_concurrent_writes(struct file
*file
)
920 struct inode
*inode
= file_inode(file
);
921 static ino_t last_ino
;
922 static dev_t last_dev
;
925 if (atomic_read(&inode
->i_writecount
) > 1
926 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
927 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
929 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
932 if (inode
->i_state
& I_DIRTY
) {
933 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
934 err
= vfs_fsync(file
, 0);
936 last_ino
= inode
->i_ino
;
937 last_dev
= inode
->i_sb
->s_dev
;
942 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
943 loff_t offset
, struct kvec
*vec
, int vlen
,
944 unsigned long *cnt
, int *stablep
)
946 struct svc_export
*exp
;
951 int stable
= *stablep
;
954 loff_t end
= LLONG_MAX
;
955 unsigned int pflags
= current
->flags
;
957 if (test_bit(RQ_LOCAL
, &rqstp
->rq_flags
))
959 * We want less throttling in balance_dirty_pages()
960 * and shrink_inactive_list() so that nfs to
961 * localhost doesn't cause nfsd to lock up due to all
962 * the client's dirty pages or its congested queue.
964 current
->flags
|= PF_LESS_THROTTLE
;
966 inode
= file_inode(file
);
967 exp
= fhp
->fh_export
;
969 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
974 /* Write the data. */
975 oldfs
= get_fs(); set_fs(KERNEL_DS
);
976 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &pos
);
981 nfsdstats
.io_write
+= host_err
;
982 fsnotify_modify(file
);
986 host_err
= wait_for_concurrent_writes(file
);
989 end
= offset
+ *cnt
- 1;
990 host_err
= vfs_fsync_range(file
, offset
, end
, 0);
995 dprintk("nfsd: write complete host_err=%d\n", host_err
);
999 err
= nfserrno(host_err
);
1000 if (test_bit(RQ_LOCAL
, &rqstp
->rq_flags
))
1001 tsk_restore_flags(current
, pflags
, PF_LESS_THROTTLE
);
1006 * Read data from a file. count must contain the requested read count
1007 * on entry. On return, *count contains the number of bytes actually read.
1008 * N.B. After this call fhp needs an fh_put
1010 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1011 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
1017 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1021 ra
= nfsd_init_raparms(file
);
1022 err
= nfsd_vfs_read(rqstp
, file
, offset
, vec
, vlen
, count
);
1024 nfsd_put_raparams(file
, ra
);
1031 * Write data to a file.
1032 * The stable flag requests synchronous writes.
1033 * N.B. After this call fhp needs an fh_put
1036 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1037 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1043 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1044 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1047 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1050 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1055 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1063 #ifdef CONFIG_NFSD_V3
1065 * Commit all pending writes to stable storage.
1067 * Note: we only guarantee that data that lies within the range specified
1068 * by the 'offset' and 'count' parameters will be synced.
1070 * Unfortunately we cannot lock the file to make sure we return full WCC
1071 * data to the client, as locking happens lower down in the filesystem.
1074 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1075 loff_t offset
, unsigned long count
)
1078 loff_t end
= LLONG_MAX
;
1079 __be32 err
= nfserr_inval
;
1084 end
= offset
+ (loff_t
)count
- 1;
1089 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1090 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1093 if (EX_ISSYNC(fhp
->fh_export
)) {
1094 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1096 if (err2
!= -EINVAL
)
1097 err
= nfserrno(err2
);
1099 err
= nfserr_notsupp
;
1106 #endif /* CONFIG_NFSD_V3 */
1109 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1113 * Mode has already been set earlier in create:
1115 iap
->ia_valid
&= ~ATTR_MODE
;
1117 * Setting uid/gid works only for root. Irix appears to
1118 * send along the gid on create when it tries to implement
1119 * setgid directories via NFS:
1121 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID
))
1122 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1124 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1125 /* Callers expect file metadata to be committed here */
1126 return nfserrno(commit_metadata(resfhp
));
1129 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1130 * setting size to 0 may fail for some specific file systems by the permission
1131 * checking which requires WRITE permission but the mode is 000.
1132 * we ignore the resizing(to 0) on the just new created file, since the size is
1133 * 0 after file created.
1135 * call this only after vfs_create() is called.
1138 nfsd_check_ignore_resizing(struct iattr
*iap
)
1140 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1141 iap
->ia_valid
&= ~ATTR_SIZE
;
1145 * Create a file (regular, directory, device, fifo); UNIX sockets
1146 * not yet implemented.
1147 * If the response fh has been verified, the parent directory should
1148 * already be locked. Note that the parent directory is left locked.
1150 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1153 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1154 char *fname
, int flen
, struct iattr
*iap
,
1155 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1157 struct dentry
*dentry
, *dchild
= NULL
;
1167 if (isdotent(fname
, flen
))
1170 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1174 dentry
= fhp
->fh_dentry
;
1175 dirp
= d_inode(dentry
);
1177 err
= nfserr_notdir
;
1178 if (!dirp
->i_op
->lookup
)
1181 * Check whether the response file handle has been verified yet.
1182 * If it has, the parent directory should already be locked.
1184 if (!resfhp
->fh_dentry
) {
1185 host_err
= fh_want_write(fhp
);
1189 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1190 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1191 dchild
= lookup_one_len(fname
, dentry
, flen
);
1192 host_err
= PTR_ERR(dchild
);
1195 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1199 /* called from nfsd_proc_create */
1200 dchild
= dget(resfhp
->fh_dentry
);
1201 if (!fhp
->fh_locked
) {
1202 /* not actually possible */
1204 "nfsd_create: parent %pd2 not locked!\n",
1211 * Make sure the child dentry is still negative ...
1214 if (d_really_is_positive(dchild
)) {
1215 dprintk("nfsd_create: dentry %pd/%pd not negative!\n",
1220 if (!(iap
->ia_valid
& ATTR_MODE
))
1222 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1225 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1226 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1232 * Get the dir op function pointer.
1238 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1240 nfsd_check_ignore_resizing(iap
);
1243 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1249 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1255 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1258 * nfsd_create_setattr already committed the child. Transactional
1259 * filesystems had a chance to commit changes for both parent and
1260 * child * simultaneously making the following commit_metadata a
1263 err2
= nfserrno(commit_metadata(fhp
));
1267 * Update the file handle to get the new inode info.
1270 err
= fh_update(resfhp
);
1272 if (dchild
&& !IS_ERR(dchild
))
1277 err
= nfserrno(host_err
);
1281 #ifdef CONFIG_NFSD_V3
1284 * NFSv3 and NFSv4 version of nfsd_create
1287 do_nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1288 char *fname
, int flen
, struct iattr
*iap
,
1289 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1290 bool *truncp
, bool *created
)
1292 struct dentry
*dentry
, *dchild
= NULL
;
1296 __u32 v_mtime
=0, v_atime
=0;
1302 if (isdotent(fname
, flen
))
1304 if (!(iap
->ia_valid
& ATTR_MODE
))
1306 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
1310 dentry
= fhp
->fh_dentry
;
1311 dirp
= d_inode(dentry
);
1313 /* Get all the sanity checks out of the way before
1314 * we lock the parent. */
1315 err
= nfserr_notdir
;
1316 if (!dirp
->i_op
->lookup
)
1319 host_err
= fh_want_write(fhp
);
1323 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1326 * Compose the response file handle.
1328 dchild
= lookup_one_len(fname
, dentry
, flen
);
1329 host_err
= PTR_ERR(dchild
);
1333 /* If file doesn't exist, check for permissions to create one */
1334 if (d_really_is_negative(dchild
)) {
1335 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1340 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1344 if (nfsd_create_is_exclusive(createmode
)) {
1345 /* solaris7 gets confused (bugid 4218508) if these have
1346 * the high bit set, so just clear the high bits. If this is
1347 * ever changed to use different attrs for storing the
1348 * verifier, then do_open_lookup() will also need to be fixed
1351 v_mtime
= verifier
[0]&0x7fffffff;
1352 v_atime
= verifier
[1]&0x7fffffff;
1355 if (d_really_is_positive(dchild
)) {
1358 switch (createmode
) {
1359 case NFS3_CREATE_UNCHECKED
:
1360 if (! d_is_reg(dchild
))
1363 /* in nfsv4, we need to treat this case a little
1364 * differently. we don't want to truncate the
1365 * file now; this would be wrong if the OPEN
1366 * fails for some other reason. furthermore,
1367 * if the size is nonzero, we should ignore it
1368 * according to spec!
1370 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1373 iap
->ia_valid
&= ATTR_SIZE
;
1377 case NFS3_CREATE_EXCLUSIVE
:
1378 if ( d_inode(dchild
)->i_mtime
.tv_sec
== v_mtime
1379 && d_inode(dchild
)->i_atime
.tv_sec
== v_atime
1380 && d_inode(dchild
)->i_size
== 0 ) {
1385 case NFS4_CREATE_EXCLUSIVE4_1
:
1386 if ( d_inode(dchild
)->i_mtime
.tv_sec
== v_mtime
1387 && d_inode(dchild
)->i_atime
.tv_sec
== v_atime
1388 && d_inode(dchild
)->i_size
== 0 ) {
1394 case NFS3_CREATE_GUARDED
:
1401 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1409 nfsd_check_ignore_resizing(iap
);
1411 if (nfsd_create_is_exclusive(createmode
)) {
1412 /* Cram the verifier into atime/mtime */
1413 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1414 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1415 /* XXX someone who knows this better please fix it for nsec */
1416 iap
->ia_mtime
.tv_sec
= v_mtime
;
1417 iap
->ia_atime
.tv_sec
= v_atime
;
1418 iap
->ia_mtime
.tv_nsec
= 0;
1419 iap
->ia_atime
.tv_nsec
= 0;
1423 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1426 * nfsd_create_setattr already committed the child
1427 * (and possibly also the parent).
1430 err
= nfserrno(commit_metadata(fhp
));
1433 * Update the filehandle to get the new inode info.
1436 err
= fh_update(resfhp
);
1440 if (dchild
&& !IS_ERR(dchild
))
1446 err
= nfserrno(host_err
);
1449 #endif /* CONFIG_NFSD_V3 */
1452 * Read a symlink. On entry, *lenp must contain the maximum path length that
1453 * fits into the buffer. On return, it contains the true length.
1454 * N.B. After this call fhp needs an fh_put
1457 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1459 struct inode
*inode
;
1465 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1469 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
1470 path
.dentry
= fhp
->fh_dentry
;
1471 inode
= d_inode(path
.dentry
);
1474 if (!inode
->i_op
->readlink
)
1478 /* N.B. Why does this call need a get_fs()??
1479 * Remove the set_fs and watch the fireworks:-) --okir
1482 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1483 host_err
= inode
->i_op
->readlink(path
.dentry
, (char __user
*)buf
, *lenp
);
1494 err
= nfserrno(host_err
);
1499 * Create a symlink and look up its inode
1500 * N.B. After this call _both_ fhp and resfhp need an fh_put
1503 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1504 char *fname
, int flen
,
1506 struct svc_fh
*resfhp
)
1508 struct dentry
*dentry
, *dnew
;
1513 if (!flen
|| path
[0] == '\0')
1516 if (isdotent(fname
, flen
))
1519 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1523 host_err
= fh_want_write(fhp
);
1528 dentry
= fhp
->fh_dentry
;
1529 dnew
= lookup_one_len(fname
, dentry
, flen
);
1530 host_err
= PTR_ERR(dnew
);
1534 host_err
= vfs_symlink(d_inode(dentry
), dnew
, path
);
1535 err
= nfserrno(host_err
);
1537 err
= nfserrno(commit_metadata(fhp
));
1542 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1544 if (err
==0) err
= cerr
;
1549 err
= nfserrno(host_err
);
1555 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1558 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1559 char *name
, int len
, struct svc_fh
*tfhp
)
1561 struct dentry
*ddir
, *dnew
, *dold
;
1566 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1569 err
= fh_verify(rqstp
, tfhp
, 0, NFSD_MAY_NOP
);
1573 if (d_is_dir(tfhp
->fh_dentry
))
1579 if (isdotent(name
, len
))
1582 host_err
= fh_want_write(tfhp
);
1584 err
= nfserrno(host_err
);
1588 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1589 ddir
= ffhp
->fh_dentry
;
1590 dirp
= d_inode(ddir
);
1592 dnew
= lookup_one_len(name
, ddir
, len
);
1593 host_err
= PTR_ERR(dnew
);
1597 dold
= tfhp
->fh_dentry
;
1600 if (d_really_is_negative(dold
))
1602 host_err
= vfs_link(dold
, dirp
, dnew
, NULL
);
1604 err
= nfserrno(commit_metadata(ffhp
));
1606 err
= nfserrno(commit_metadata(tfhp
));
1608 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1611 err
= nfserrno(host_err
);
1617 fh_drop_write(tfhp
);
1622 err
= nfserrno(host_err
);
1628 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1631 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1632 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1634 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1635 struct inode
*fdir
, *tdir
;
1639 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1642 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1646 fdentry
= ffhp
->fh_dentry
;
1647 fdir
= d_inode(fdentry
);
1649 tdentry
= tfhp
->fh_dentry
;
1650 tdir
= d_inode(tdentry
);
1653 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1656 host_err
= fh_want_write(ffhp
);
1658 err
= nfserrno(host_err
);
1662 /* cannot use fh_lock as we need deadlock protective ordering
1663 * so do it by hand */
1664 trap
= lock_rename(tdentry
, fdentry
);
1665 ffhp
->fh_locked
= tfhp
->fh_locked
= true;
1669 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1670 host_err
= PTR_ERR(odentry
);
1671 if (IS_ERR(odentry
))
1675 if (d_really_is_negative(odentry
))
1678 if (odentry
== trap
)
1681 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1682 host_err
= PTR_ERR(ndentry
);
1683 if (IS_ERR(ndentry
))
1685 host_err
= -ENOTEMPTY
;
1686 if (ndentry
== trap
)
1690 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1692 if (ffhp
->fh_export
->ex_path
.dentry
!= tfhp
->fh_export
->ex_path
.dentry
)
1695 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
, NULL
, 0);
1697 host_err
= commit_metadata(tfhp
);
1699 host_err
= commit_metadata(ffhp
);
1706 err
= nfserrno(host_err
);
1708 * We cannot rely on fh_unlock on the two filehandles,
1709 * as that would do the wrong thing if the two directories
1710 * were the same, so again we do it by hand.
1712 fill_post_wcc(ffhp
);
1713 fill_post_wcc(tfhp
);
1714 unlock_rename(tdentry
, fdentry
);
1715 ffhp
->fh_locked
= tfhp
->fh_locked
= false;
1716 fh_drop_write(ffhp
);
1723 * Unlink a file or directory
1724 * N.B. After this call fhp needs an fh_put
1727 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1728 char *fname
, int flen
)
1730 struct dentry
*dentry
, *rdentry
;
1736 if (!flen
|| isdotent(fname
, flen
))
1738 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1742 host_err
= fh_want_write(fhp
);
1746 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1747 dentry
= fhp
->fh_dentry
;
1748 dirp
= d_inode(dentry
);
1750 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1751 host_err
= PTR_ERR(rdentry
);
1752 if (IS_ERR(rdentry
))
1755 if (d_really_is_negative(rdentry
)) {
1762 type
= d_inode(rdentry
)->i_mode
& S_IFMT
;
1764 if (type
!= S_IFDIR
)
1765 host_err
= vfs_unlink(dirp
, rdentry
, NULL
);
1767 host_err
= vfs_rmdir(dirp
, rdentry
);
1769 host_err
= commit_metadata(fhp
);
1773 err
= nfserrno(host_err
);
1779 * We do this buffering because we must not call back into the file
1780 * system's ->lookup() method from the filldir callback. That may well
1781 * deadlock a number of file systems.
1783 * This is based heavily on the implementation of same in XFS.
1785 struct buffered_dirent
{
1789 unsigned int d_type
;
1793 struct readdir_data
{
1794 struct dir_context ctx
;
1800 static int nfsd_buffered_filldir(struct dir_context
*ctx
, const char *name
,
1801 int namlen
, loff_t offset
, u64 ino
,
1802 unsigned int d_type
)
1804 struct readdir_data
*buf
=
1805 container_of(ctx
, struct readdir_data
, ctx
);
1806 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1807 unsigned int reclen
;
1809 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1810 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1815 de
->namlen
= namlen
;
1816 de
->offset
= offset
;
1818 de
->d_type
= d_type
;
1819 memcpy(de
->name
, name
, namlen
);
1820 buf
->used
+= reclen
;
1825 static __be32
nfsd_buffered_readdir(struct file
*file
, nfsd_filldir_t func
,
1826 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1828 struct buffered_dirent
*de
;
1832 struct readdir_data buf
= {
1833 .ctx
.actor
= nfsd_buffered_filldir
,
1834 .dirent
= (void *)__get_free_page(GFP_KERNEL
)
1838 return nfserrno(-ENOMEM
);
1843 struct inode
*dir_inode
= file_inode(file
);
1844 unsigned int reclen
;
1846 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1850 host_err
= iterate_dir(file
, &buf
.ctx
);
1863 * Various filldir functions may end up calling back into
1864 * lookup_one_len() and the file system's ->lookup() method.
1865 * These expect i_mutex to be held, as it would within readdir.
1867 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
1871 de
= (struct buffered_dirent
*)buf
.dirent
;
1873 offset
= de
->offset
;
1875 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
1876 de
->ino
, de
->d_type
))
1879 if (cdp
->err
!= nfs_ok
)
1882 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
1885 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
1887 mutex_unlock(&dir_inode
->i_mutex
);
1888 if (size
> 0) /* We bailed out early */
1891 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
1894 free_page((unsigned long)(buf
.dirent
));
1897 return nfserrno(host_err
);
1904 * Read entries from a directory.
1905 * The NFSv3/4 verifier we ignore for now.
1908 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
1909 struct readdir_cd
*cdp
, nfsd_filldir_t func
)
1913 loff_t offset
= *offsetp
;
1914 int may_flags
= NFSD_MAY_READ
;
1916 /* NFSv2 only supports 32 bit cookies */
1917 if (rqstp
->rq_vers
> 2)
1918 may_flags
|= NFSD_MAY_64BIT_COOKIE
;
1920 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, may_flags
, &file
);
1924 offset
= vfs_llseek(file
, offset
, SEEK_SET
);
1926 err
= nfserrno((int)offset
);
1930 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
1932 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
1933 err
= nfs_ok
; /* can still be found in ->err */
1941 * Get file system stats
1942 * N.B. After this call fhp needs an fh_put
1945 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
1949 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
1951 struct path path
= {
1952 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
1953 .dentry
= fhp
->fh_dentry
,
1955 if (vfs_statfs(&path
, stat
))
1961 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
1963 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
1967 * Check for a user's access permissions to this inode.
1970 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
1971 struct dentry
*dentry
, int acc
)
1973 struct inode
*inode
= d_inode(dentry
);
1976 if ((acc
& NFSD_MAY_MASK
) == NFSD_MAY_NOP
)
1979 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
1981 (acc
& NFSD_MAY_READ
)? " read" : "",
1982 (acc
& NFSD_MAY_WRITE
)? " write" : "",
1983 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
1984 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
1985 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
1986 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
1987 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
1989 IS_IMMUTABLE(inode
)? " immut" : "",
1990 IS_APPEND(inode
)? " append" : "",
1991 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
1992 dprintk(" owner %d/%d user %d/%d\n",
1993 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
1996 /* Normally we reject any write/sattr etc access on a read-only file
1997 * system. But if it is IRIX doing check on write-access for a
1998 * device special file, we ignore rofs.
2000 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2001 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2002 if (exp_rdonly(rqstp
, exp
) ||
2003 __mnt_is_readonly(exp
->ex_path
.mnt
))
2005 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2008 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2011 if (acc
& NFSD_MAY_LOCK
) {
2012 /* If we cannot rely on authentication in NLM requests,
2013 * just allow locks, otherwise require read permission, or
2016 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2019 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2022 * The file owner always gets access permission for accesses that
2023 * would normally be checked at open time. This is to make
2024 * file access work even when the client has done a fchmod(fd, 0).
2026 * However, `cp foo bar' should fail nevertheless when bar is
2027 * readonly. A sensible way to do this might be to reject all
2028 * attempts to truncate a read-only file, because a creat() call
2029 * always implies file truncation.
2030 * ... but this isn't really fair. A process may reasonably call
2031 * ftruncate on an open file descriptor on a file with perm 000.
2032 * We must trust the client to do permission checking - using "ACCESS"
2035 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2036 uid_eq(inode
->i_uid
, current_fsuid()))
2039 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2040 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2042 /* Allow read access to binaries even when mode 111 */
2043 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2044 (acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
) ||
2045 acc
== (NFSD_MAY_READ
| NFSD_MAY_READ_IF_EXEC
)))
2046 err
= inode_permission(inode
, MAY_EXEC
);
2048 return err
? nfserrno(err
) : 0;
2052 nfsd_racache_shutdown(void)
2054 struct raparms
*raparm
, *last_raparm
;
2057 dprintk("nfsd: freeing readahead buffers.\n");
2059 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2060 raparm
= raparm_hash
[i
].pb_head
;
2062 last_raparm
= raparm
;
2063 raparm
= raparm
->p_next
;
2066 raparm_hash
[i
].pb_head
= NULL
;
2070 * Initialize readahead param cache
2073 nfsd_racache_init(int cache_size
)
2078 struct raparms
**raparm
= NULL
;
2081 if (raparm_hash
[0].pb_head
)
2083 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2084 nperbucket
= max(2, nperbucket
);
2085 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2087 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2089 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2090 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2092 raparm
= &raparm_hash
[i
].pb_head
;
2093 for (j
= 0; j
< nperbucket
; j
++) {
2094 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2097 raparm
= &(*raparm
)->p_next
;
2102 nfsdstats
.ra_size
= cache_size
;
2106 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2107 nfsd_racache_shutdown();