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 dentry
= lookup_one_len(name
, dparent
, len
);
212 host_err
= PTR_ERR(dentry
);
216 * check if we have crossed a mount point ...
218 if (nfsd_mountpoint(dentry
, exp
)) {
219 if ((host_err
= nfsd_cross_mnt(rqstp
, &dentry
, &exp
))) {
225 *dentry_ret
= dentry
;
231 return nfserrno(host_err
);
235 * Look up one component of a pathname.
236 * N.B. After this call _both_ fhp and resfh need an fh_put
238 * If the lookup would cross a mountpoint, and the mounted filesystem
239 * is exported to the client with NFSEXP_NOHIDE, then the lookup is
240 * accepted as it stands and the mounted directory is
241 * returned. Otherwise the covered directory is returned.
242 * NOTE: this mountpoint crossing is not supported properly by all
243 * clients and is explicitly disallowed for NFSv3
244 * NeilBrown <neilb@cse.unsw.edu.au>
247 nfsd_lookup(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, const char *name
,
248 unsigned int len
, struct svc_fh
*resfh
)
250 struct svc_export
*exp
;
251 struct dentry
*dentry
;
254 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
257 err
= nfsd_lookup_dentry(rqstp
, fhp
, name
, len
, &exp
, &dentry
);
260 err
= check_nfsd_access(exp
, rqstp
);
264 * Note: we compose the file handle now, but as the
265 * dentry may be negative, it may need to be updated.
267 err
= fh_compose(resfh
, exp
, dentry
, fhp
);
268 if (!err
&& !dentry
->d_inode
)
276 static int nfsd_break_lease(struct inode
*inode
)
278 if (!S_ISREG(inode
->i_mode
))
280 return break_lease(inode
, O_WRONLY
| O_NONBLOCK
);
284 * Commit metadata changes to stable storage.
287 commit_metadata(struct svc_fh
*fhp
)
289 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
290 const struct export_operations
*export_ops
= inode
->i_sb
->s_export_op
;
292 if (!EX_ISSYNC(fhp
->fh_export
))
295 if (export_ops
->commit_metadata
)
296 return export_ops
->commit_metadata(inode
);
297 return sync_inode_metadata(inode
, 1);
301 * Set various file attributes.
302 * N.B. After this call fhp needs an fh_put
305 nfsd_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct iattr
*iap
,
306 int check_guard
, time_t guardtime
)
308 struct dentry
*dentry
;
310 int accmode
= NFSD_MAY_SATTR
;
316 if (iap
->ia_valid
& (ATTR_ATIME
| ATTR_MTIME
| ATTR_SIZE
))
317 accmode
|= NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
;
318 if (iap
->ia_valid
& ATTR_SIZE
)
322 err
= fh_verify(rqstp
, fhp
, ftype
, accmode
);
326 dentry
= fhp
->fh_dentry
;
327 inode
= dentry
->d_inode
;
329 /* Ignore any mode updates on symlinks */
330 if (S_ISLNK(inode
->i_mode
))
331 iap
->ia_valid
&= ~ATTR_MODE
;
337 * NFSv2 does not differentiate between "set-[ac]time-to-now"
338 * which only requires access, and "set-[ac]time-to-X" which
339 * requires ownership.
340 * So if it looks like it might be "set both to the same time which
341 * is close to now", and if inode_change_ok fails, then we
342 * convert to "set to now" instead of "set to explicit time"
344 * We only call inode_change_ok as the last test as technically
345 * it is not an interface that we should be using. It is only
346 * valid if the filesystem does not define it's own i_op->setattr.
348 #define BOTH_TIME_SET (ATTR_ATIME_SET | ATTR_MTIME_SET)
349 #define MAX_TOUCH_TIME_ERROR (30*60)
350 if ((iap
->ia_valid
& BOTH_TIME_SET
) == BOTH_TIME_SET
&&
351 iap
->ia_mtime
.tv_sec
== iap
->ia_atime
.tv_sec
) {
355 * Now just make sure time is in the right ballpark.
356 * Solaris, at least, doesn't seem to care what the time
357 * request is. We require it be within 30 minutes of now.
359 time_t delta
= iap
->ia_atime
.tv_sec
- get_seconds();
362 if (delta
< MAX_TOUCH_TIME_ERROR
&&
363 inode_change_ok(inode
, iap
) != 0) {
365 * Turn off ATTR_[AM]TIME_SET but leave ATTR_[AM]TIME.
366 * This will cause notify_change to set these times
369 iap
->ia_valid
&= ~BOTH_TIME_SET
;
374 * The size case is special.
375 * It changes the file as well as the attributes.
377 if (iap
->ia_valid
& ATTR_SIZE
) {
378 if (iap
->ia_size
< inode
->i_size
) {
379 err
= nfsd_permission(rqstp
, fhp
->fh_export
, dentry
,
380 NFSD_MAY_TRUNC
|NFSD_MAY_OWNER_OVERRIDE
);
385 host_err
= get_write_access(inode
);
390 host_err
= locks_verify_truncate(inode
, NULL
, iap
->ia_size
);
392 put_write_access(inode
);
397 /* sanitize the mode change */
398 if (iap
->ia_valid
& ATTR_MODE
) {
399 iap
->ia_mode
&= S_IALLUGO
;
400 iap
->ia_mode
|= (inode
->i_mode
& ~S_IALLUGO
);
403 /* Revoke setuid/setgid on chown */
404 if (!S_ISDIR(inode
->i_mode
) &&
405 (((iap
->ia_valid
& ATTR_UID
) && !uid_eq(iap
->ia_uid
, inode
->i_uid
)) ||
406 ((iap
->ia_valid
& ATTR_GID
) && !gid_eq(iap
->ia_gid
, inode
->i_gid
)))) {
407 iap
->ia_valid
|= ATTR_KILL_PRIV
;
408 if (iap
->ia_valid
& ATTR_MODE
) {
409 /* we're setting mode too, just clear the s*id bits */
410 iap
->ia_mode
&= ~S_ISUID
;
411 if (iap
->ia_mode
& S_IXGRP
)
412 iap
->ia_mode
&= ~S_ISGID
;
414 /* set ATTR_KILL_* bits and let VFS handle it */
415 iap
->ia_valid
|= (ATTR_KILL_SUID
| ATTR_KILL_SGID
);
419 /* Change the attributes. */
421 iap
->ia_valid
|= ATTR_CTIME
;
423 err
= nfserr_notsync
;
424 if (!check_guard
|| guardtime
== inode
->i_ctime
.tv_sec
) {
425 host_err
= nfsd_break_lease(inode
);
430 host_err
= notify_change(dentry
, iap
);
431 err
= nfserrno(host_err
);
435 put_write_access(inode
);
437 commit_metadata(fhp
);
442 err
= nfserrno(host_err
);
446 #if defined(CONFIG_NFSD_V2_ACL) || \
447 defined(CONFIG_NFSD_V3_ACL) || \
448 defined(CONFIG_NFSD_V4)
449 static ssize_t
nfsd_getxattr(struct dentry
*dentry
, char *key
, void **buf
)
454 buflen
= vfs_getxattr(dentry
, key
, NULL
, 0);
458 *buf
= kmalloc(buflen
, GFP_KERNEL
);
462 ret
= vfs_getxattr(dentry
, key
, *buf
, buflen
);
469 #if defined(CONFIG_NFSD_V4)
471 set_nfsv4_acl_one(struct dentry
*dentry
, struct posix_acl
*pacl
, char *key
)
478 buflen
= posix_acl_xattr_size(pacl
->a_count
);
479 buf
= kmalloc(buflen
, GFP_KERNEL
);
484 len
= posix_acl_to_xattr(&init_user_ns
, pacl
, buf
, buflen
);
490 error
= vfs_setxattr(dentry
, key
, buf
, len
, 0);
497 nfsd4_set_nfs4_acl(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
498 struct nfs4_acl
*acl
)
502 struct dentry
*dentry
;
504 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
505 unsigned int flags
= 0;
508 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_SATTR
);
512 dentry
= fhp
->fh_dentry
;
513 inode
= dentry
->d_inode
;
514 if (S_ISDIR(inode
->i_mode
))
515 flags
= NFS4_ACL_DIR
;
517 host_error
= nfs4_acl_nfsv4_to_posix(acl
, &pacl
, &dpacl
, flags
);
518 if (host_error
== -EINVAL
) {
519 return nfserr_attrnotsupp
;
520 } else if (host_error
< 0)
523 host_error
= set_nfsv4_acl_one(dentry
, pacl
, POSIX_ACL_XATTR_ACCESS
);
527 if (S_ISDIR(inode
->i_mode
))
528 host_error
= set_nfsv4_acl_one(dentry
, dpacl
, POSIX_ACL_XATTR_DEFAULT
);
531 posix_acl_release(pacl
);
532 posix_acl_release(dpacl
);
534 if (host_error
== -EOPNOTSUPP
)
535 return nfserr_attrnotsupp
;
537 return nfserrno(host_error
);
540 static struct posix_acl
*
541 _get_posix_acl(struct dentry
*dentry
, char *key
)
544 struct posix_acl
*pacl
= NULL
;
547 buflen
= nfsd_getxattr(dentry
, key
, &buf
);
551 return ERR_PTR(buflen
);
553 pacl
= posix_acl_from_xattr(&init_user_ns
, buf
, buflen
);
559 nfsd4_get_nfs4_acl(struct svc_rqst
*rqstp
, struct dentry
*dentry
, struct nfs4_acl
**acl
)
561 struct inode
*inode
= dentry
->d_inode
;
563 struct posix_acl
*pacl
= NULL
, *dpacl
= NULL
;
564 unsigned int flags
= 0;
566 pacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_ACCESS
);
567 if (IS_ERR(pacl
) && PTR_ERR(pacl
) == -ENODATA
)
568 pacl
= posix_acl_from_mode(inode
->i_mode
, GFP_KERNEL
);
570 error
= PTR_ERR(pacl
);
575 if (S_ISDIR(inode
->i_mode
)) {
576 dpacl
= _get_posix_acl(dentry
, POSIX_ACL_XATTR_DEFAULT
);
577 if (IS_ERR(dpacl
) && PTR_ERR(dpacl
) == -ENODATA
)
579 else if (IS_ERR(dpacl
)) {
580 error
= PTR_ERR(dpacl
);
584 flags
= NFS4_ACL_DIR
;
587 *acl
= nfs4_acl_posix_to_nfsv4(pacl
, dpacl
, flags
);
589 error
= PTR_ERR(*acl
);
593 posix_acl_release(pacl
);
594 posix_acl_release(dpacl
);
599 * NFS junction information is stored in an extended attribute.
601 #define NFSD_JUNCTION_XATTR_NAME XATTR_TRUSTED_PREFIX "junction.nfs"
604 * nfsd4_is_junction - Test if an object could be an NFS junction
606 * @dentry: object to test
608 * Returns 1 if "dentry" appears to contain NFS junction information.
609 * Otherwise 0 is returned.
611 int nfsd4_is_junction(struct dentry
*dentry
)
613 struct inode
*inode
= dentry
->d_inode
;
617 if (inode
->i_mode
& S_IXUGO
)
619 if (!(inode
->i_mode
& S_ISVTX
))
621 if (vfs_getxattr(dentry
, NFSD_JUNCTION_XATTR_NAME
, NULL
, 0) <= 0)
625 #ifdef CONFIG_NFSD_V4_SECURITY_LABEL
626 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
627 struct xdr_netobj
*label
)
631 struct dentry
*dentry
;
633 error
= fh_verify(rqstp
, fhp
, 0 /* S_IFREG */, NFSD_MAY_SATTR
);
637 dentry
= fhp
->fh_dentry
;
639 mutex_lock(&dentry
->d_inode
->i_mutex
);
640 host_error
= security_inode_setsecctx(dentry
, label
->data
, label
->len
);
641 mutex_unlock(&dentry
->d_inode
->i_mutex
);
642 return nfserrno(host_error
);
645 __be32
nfsd4_set_nfs4_label(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
646 struct xdr_netobj
*label
)
648 return nfserr_notsupp
;
652 #endif /* defined(CONFIG_NFSD_V4) */
654 #ifdef CONFIG_NFSD_V3
656 * Check server access rights to a file system object
662 static struct accessmap nfs3_regaccess
[] = {
663 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
664 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
665 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
666 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
},
671 static struct accessmap nfs3_diraccess
[] = {
672 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
673 { NFS3_ACCESS_LOOKUP
, NFSD_MAY_EXEC
},
674 { NFS3_ACCESS_MODIFY
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
|NFSD_MAY_TRUNC
},
675 { NFS3_ACCESS_EXTEND
, NFSD_MAY_EXEC
|NFSD_MAY_WRITE
},
676 { NFS3_ACCESS_DELETE
, NFSD_MAY_REMOVE
},
681 static struct accessmap nfs3_anyaccess
[] = {
682 /* Some clients - Solaris 2.6 at least, make an access call
683 * to the server to check for access for things like /dev/null
684 * (which really, the server doesn't care about). So
685 * We provide simple access checking for them, looking
686 * mainly at mode bits, and we make sure to ignore read-only
689 { NFS3_ACCESS_READ
, NFSD_MAY_READ
},
690 { NFS3_ACCESS_EXECUTE
, NFSD_MAY_EXEC
},
691 { NFS3_ACCESS_MODIFY
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
692 { NFS3_ACCESS_EXTEND
, NFSD_MAY_WRITE
|NFSD_MAY_LOCAL_ACCESS
},
698 nfsd_access(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, u32
*access
, u32
*supported
)
700 struct accessmap
*map
;
701 struct svc_export
*export
;
702 struct dentry
*dentry
;
703 u32 query
, result
= 0, sresult
= 0;
706 error
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
);
710 export
= fhp
->fh_export
;
711 dentry
= fhp
->fh_dentry
;
713 if (S_ISREG(dentry
->d_inode
->i_mode
))
714 map
= nfs3_regaccess
;
715 else if (S_ISDIR(dentry
->d_inode
->i_mode
))
716 map
= nfs3_diraccess
;
718 map
= nfs3_anyaccess
;
722 for (; map
->access
; map
++) {
723 if (map
->access
& query
) {
726 sresult
|= map
->access
;
728 err2
= nfsd_permission(rqstp
, export
, dentry
, map
->how
);
731 result
|= map
->access
;
734 /* the following error codes just mean the access was not allowed,
735 * rather than an error occurred */
739 /* simply don't "or" in the access bit. */
749 *supported
= sresult
;
754 #endif /* CONFIG_NFSD_V3 */
756 static int nfsd_open_break_lease(struct inode
*inode
, int access
)
760 if (access
& NFSD_MAY_NOT_BREAK_LEASE
)
762 mode
= (access
& NFSD_MAY_WRITE
) ? O_WRONLY
: O_RDONLY
;
763 return break_lease(inode
, mode
| O_NONBLOCK
);
767 * Open an existing file or directory.
768 * The may_flags argument indicates the type of open (read/write/lock)
769 * and additional flags.
770 * N.B. After this call fhp needs an fh_put
773 nfsd_open(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, umode_t type
,
774 int may_flags
, struct file
**filp
)
778 int flags
= O_RDONLY
|O_LARGEFILE
;
782 validate_process_creds();
785 * If we get here, then the client has already done an "open",
786 * and (hopefully) checked permission - so allow OWNER_OVERRIDE
787 * in case a chmod has now revoked permission.
789 * Arguably we should also allow the owner override for
790 * directories, but we never have and it doesn't seem to have
791 * caused anyone a problem. If we were to change this, note
792 * also that our filldir callbacks would need a variant of
793 * lookup_one_len that doesn't check permissions.
796 may_flags
|= NFSD_MAY_OWNER_OVERRIDE
;
797 err
= fh_verify(rqstp
, fhp
, type
, may_flags
);
801 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
802 path
.dentry
= fhp
->fh_dentry
;
803 inode
= path
.dentry
->d_inode
;
805 /* Disallow write access to files with the append-only bit set
806 * or any access when mandatory locking enabled
809 if (IS_APPEND(inode
) && (may_flags
& NFSD_MAY_WRITE
))
812 * We must ignore files (but only files) which might have mandatory
813 * locks on them because there is no way to know if the accesser has
816 if (S_ISREG((inode
)->i_mode
) && mandatory_lock(inode
))
822 host_err
= nfsd_open_break_lease(inode
, may_flags
);
823 if (host_err
) /* NOMEM or WOULDBLOCK */
826 if (may_flags
& NFSD_MAY_WRITE
) {
827 if (may_flags
& NFSD_MAY_READ
)
828 flags
= O_RDWR
|O_LARGEFILE
;
830 flags
= O_WRONLY
|O_LARGEFILE
;
832 *filp
= dentry_open(&path
, flags
, current_cred());
834 host_err
= PTR_ERR(*filp
);
836 host_err
= ima_file_check(*filp
, may_flags
);
838 if (may_flags
& NFSD_MAY_64BIT_COOKIE
)
839 (*filp
)->f_mode
|= FMODE_64BITHASH
;
841 (*filp
)->f_mode
|= FMODE_32BITHASH
;
845 err
= nfserrno(host_err
);
847 validate_process_creds();
855 nfsd_close(struct file
*filp
)
861 * Obtain the readahead parameters for the file
862 * specified by (dev, ino).
865 static inline struct raparms
*
866 nfsd_get_raparms(dev_t dev
, ino_t ino
)
868 struct raparms
*ra
, **rap
, **frap
= NULL
;
871 struct raparm_hbucket
*rab
;
873 hash
= jhash_2words(dev
, ino
, 0xfeedbeef) & RAPARM_HASH_MASK
;
874 rab
= &raparm_hash
[hash
];
876 spin_lock(&rab
->pb_lock
);
877 for (rap
= &rab
->pb_head
; (ra
= *rap
); rap
= &ra
->p_next
) {
878 if (ra
->p_ino
== ino
&& ra
->p_dev
== dev
)
881 if (ra
->p_count
== 0)
884 depth
= nfsdstats
.ra_size
;
886 spin_unlock(&rab
->pb_lock
);
896 if (rap
!= &rab
->pb_head
) {
898 ra
->p_next
= rab
->pb_head
;
902 nfsdstats
.ra_depth
[depth
*10/nfsdstats
.ra_size
]++;
903 spin_unlock(&rab
->pb_lock
);
908 * Grab and keep cached pages associated with a file in the svc_rqst
909 * so that they can be passed to the network sendmsg/sendpage routines
910 * directly. They will be released after the sending has completed.
913 nfsd_splice_actor(struct pipe_inode_info
*pipe
, struct pipe_buffer
*buf
,
914 struct splice_desc
*sd
)
916 struct svc_rqst
*rqstp
= sd
->u
.data
;
917 struct page
**pp
= rqstp
->rq_next_page
;
918 struct page
*page
= buf
->page
;
923 if (rqstp
->rq_res
.page_len
== 0) {
925 put_page(*rqstp
->rq_next_page
);
926 *(rqstp
->rq_next_page
++) = page
;
927 rqstp
->rq_res
.page_base
= buf
->offset
;
928 rqstp
->rq_res
.page_len
= size
;
929 } else if (page
!= pp
[-1]) {
931 if (*rqstp
->rq_next_page
)
932 put_page(*rqstp
->rq_next_page
);
933 *(rqstp
->rq_next_page
++) = page
;
934 rqstp
->rq_res
.page_len
+= size
;
936 rqstp
->rq_res
.page_len
+= size
;
941 static int nfsd_direct_splice_actor(struct pipe_inode_info
*pipe
,
942 struct splice_desc
*sd
)
944 return __splice_from_pipe(pipe
, sd
, nfsd_splice_actor
);
948 nfsd_vfs_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
949 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
957 if (file
->f_op
->splice_read
&& rqstp
->rq_splice_ok
) {
958 struct splice_desc sd
= {
965 rqstp
->rq_next_page
= rqstp
->rq_respages
+ 1;
966 host_err
= splice_direct_to_actor(file
, &sd
, nfsd_direct_splice_actor
);
970 host_err
= vfs_readv(file
, (struct iovec __user
*)vec
, vlen
, &offset
);
975 nfsdstats
.io_read
+= host_err
;
978 fsnotify_access(file
);
980 err
= nfserrno(host_err
);
984 static void kill_suid(struct dentry
*dentry
)
987 ia
.ia_valid
= ATTR_KILL_SUID
| ATTR_KILL_SGID
| ATTR_KILL_PRIV
;
989 mutex_lock(&dentry
->d_inode
->i_mutex
);
990 notify_change(dentry
, &ia
);
991 mutex_unlock(&dentry
->d_inode
->i_mutex
);
995 * Gathered writes: If another process is currently writing to the file,
996 * there's a high chance this is another nfsd (triggered by a bulk write
997 * from a client's biod). Rather than syncing the file with each write
998 * request, we sleep for 10 msec.
1000 * I don't know if this roughly approximates C. Juszak's idea of
1001 * gathered writes, but it's a nice and simple solution (IMHO), and it
1004 * Note: we do this only in the NFSv2 case, since v3 and higher have a
1005 * better tool (separate unstable writes and commits) for solving this
1008 static int wait_for_concurrent_writes(struct file
*file
)
1010 struct inode
*inode
= file_inode(file
);
1011 static ino_t last_ino
;
1012 static dev_t last_dev
;
1015 if (atomic_read(&inode
->i_writecount
) > 1
1016 || (last_ino
== inode
->i_ino
&& last_dev
== inode
->i_sb
->s_dev
)) {
1017 dprintk("nfsd: write defer %d\n", task_pid_nr(current
));
1019 dprintk("nfsd: write resume %d\n", task_pid_nr(current
));
1022 if (inode
->i_state
& I_DIRTY
) {
1023 dprintk("nfsd: write sync %d\n", task_pid_nr(current
));
1024 err
= vfs_fsync(file
, 0);
1026 last_ino
= inode
->i_ino
;
1027 last_dev
= inode
->i_sb
->s_dev
;
1032 nfsd_vfs_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1033 loff_t offset
, struct kvec
*vec
, int vlen
,
1034 unsigned long *cnt
, int *stablep
)
1036 struct svc_export
*exp
;
1037 struct dentry
*dentry
;
1038 struct inode
*inode
;
1042 int stable
= *stablep
;
1044 loff_t pos
= offset
;
1046 dentry
= file
->f_path
.dentry
;
1047 inode
= dentry
->d_inode
;
1048 exp
= fhp
->fh_export
;
1050 use_wgather
= (rqstp
->rq_vers
== 2) && EX_WGATHER(exp
);
1052 if (!EX_ISSYNC(exp
))
1055 /* Write the data. */
1056 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1057 host_err
= vfs_writev(file
, (struct iovec __user
*)vec
, vlen
, &pos
);
1062 nfsdstats
.io_write
+= host_err
;
1063 fsnotify_modify(file
);
1065 /* clear setuid/setgid flag after write */
1066 if (inode
->i_mode
& (S_ISUID
| S_ISGID
))
1071 host_err
= wait_for_concurrent_writes(file
);
1073 host_err
= vfs_fsync_range(file
, offset
, offset
+*cnt
, 0);
1077 dprintk("nfsd: write complete host_err=%d\n", host_err
);
1081 err
= nfserrno(host_err
);
1086 * Read data from a file. count must contain the requested read count
1087 * on entry. On return, *count contains the number of bytes actually read.
1088 * N.B. After this call fhp needs an fh_put
1090 __be32
nfsd_read(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1091 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *count
)
1094 struct inode
*inode
;
1098 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_READ
, &file
);
1102 inode
= file_inode(file
);
1104 /* Get readahead parameters */
1105 ra
= nfsd_get_raparms(inode
->i_sb
->s_dev
, inode
->i_ino
);
1107 if (ra
&& ra
->p_set
)
1108 file
->f_ra
= ra
->p_ra
;
1110 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1112 /* Write back readahead params */
1114 struct raparm_hbucket
*rab
= &raparm_hash
[ra
->p_hindex
];
1115 spin_lock(&rab
->pb_lock
);
1116 ra
->p_ra
= file
->f_ra
;
1119 spin_unlock(&rab
->pb_lock
);
1126 /* As above, but use the provided file descriptor. */
1128 nfsd_read_file(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1129 loff_t offset
, struct kvec
*vec
, int vlen
,
1130 unsigned long *count
)
1135 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1136 NFSD_MAY_READ
|NFSD_MAY_OWNER_OVERRIDE
);
1139 err
= nfsd_vfs_read(rqstp
, fhp
, file
, offset
, vec
, vlen
, count
);
1140 } else /* Note file may still be NULL in NFSv4 special stateid case: */
1141 err
= nfsd_read(rqstp
, fhp
, offset
, vec
, vlen
, count
);
1147 * Write data to a file.
1148 * The stable flag requests synchronous writes.
1149 * N.B. After this call fhp needs an fh_put
1152 nfsd_write(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct file
*file
,
1153 loff_t offset
, struct kvec
*vec
, int vlen
, unsigned long *cnt
,
1159 err
= nfsd_permission(rqstp
, fhp
->fh_export
, fhp
->fh_dentry
,
1160 NFSD_MAY_WRITE
|NFSD_MAY_OWNER_OVERRIDE
);
1163 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
, cnt
,
1166 err
= nfsd_open(rqstp
, fhp
, S_IFREG
, NFSD_MAY_WRITE
, &file
);
1171 err
= nfsd_vfs_write(rqstp
, fhp
, file
, offset
, vec
, vlen
,
1179 #ifdef CONFIG_NFSD_V3
1181 * Commit all pending writes to stable storage.
1183 * Note: we only guarantee that data that lies within the range specified
1184 * by the 'offset' and 'count' parameters will be synced.
1186 * Unfortunately we cannot lock the file to make sure we return full WCC
1187 * data to the client, as locking happens lower down in the filesystem.
1190 nfsd_commit(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1191 loff_t offset
, unsigned long count
)
1194 loff_t end
= LLONG_MAX
;
1195 __be32 err
= nfserr_inval
;
1200 end
= offset
+ (loff_t
)count
- 1;
1205 err
= nfsd_open(rqstp
, fhp
, S_IFREG
,
1206 NFSD_MAY_WRITE
|NFSD_MAY_NOT_BREAK_LEASE
, &file
);
1209 if (EX_ISSYNC(fhp
->fh_export
)) {
1210 int err2
= vfs_fsync_range(file
, offset
, end
, 0);
1212 if (err2
!= -EINVAL
)
1213 err
= nfserrno(err2
);
1215 err
= nfserr_notsupp
;
1222 #endif /* CONFIG_NFSD_V3 */
1225 nfsd_create_setattr(struct svc_rqst
*rqstp
, struct svc_fh
*resfhp
,
1229 * Mode has already been set earlier in create:
1231 iap
->ia_valid
&= ~ATTR_MODE
;
1233 * Setting uid/gid works only for root. Irix appears to
1234 * send along the gid on create when it tries to implement
1235 * setgid directories via NFS:
1237 if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID
))
1238 iap
->ia_valid
&= ~(ATTR_UID
|ATTR_GID
);
1240 return nfsd_setattr(rqstp
, resfhp
, iap
, 0, (time_t)0);
1244 /* HPUX client sometimes creates a file in mode 000, and sets size to 0.
1245 * setting size to 0 may fail for some specific file systems by the permission
1246 * checking which requires WRITE permission but the mode is 000.
1247 * we ignore the resizing(to 0) on the just new created file, since the size is
1248 * 0 after file created.
1250 * call this only after vfs_create() is called.
1253 nfsd_check_ignore_resizing(struct iattr
*iap
)
1255 if ((iap
->ia_valid
& ATTR_SIZE
) && (iap
->ia_size
== 0))
1256 iap
->ia_valid
&= ~ATTR_SIZE
;
1260 * Create a file (regular, directory, device, fifo); UNIX sockets
1261 * not yet implemented.
1262 * If the response fh has been verified, the parent directory should
1263 * already be locked. Note that the parent directory is left locked.
1265 * N.B. Every call to nfsd_create needs an fh_put for _both_ fhp and resfhp
1268 nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1269 char *fname
, int flen
, struct iattr
*iap
,
1270 int type
, dev_t rdev
, struct svc_fh
*resfhp
)
1272 struct dentry
*dentry
, *dchild
= NULL
;
1282 if (isdotent(fname
, flen
))
1285 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1289 dentry
= fhp
->fh_dentry
;
1290 dirp
= dentry
->d_inode
;
1292 err
= nfserr_notdir
;
1293 if (!dirp
->i_op
->lookup
)
1296 * Check whether the response file handle has been verified yet.
1297 * If it has, the parent directory should already be locked.
1299 if (!resfhp
->fh_dentry
) {
1300 host_err
= fh_want_write(fhp
);
1304 /* called from nfsd_proc_mkdir, or possibly nfsd3_proc_create */
1305 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1306 dchild
= lookup_one_len(fname
, dentry
, flen
);
1307 host_err
= PTR_ERR(dchild
);
1310 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1314 /* called from nfsd_proc_create */
1315 dchild
= dget(resfhp
->fh_dentry
);
1316 if (!fhp
->fh_locked
) {
1317 /* not actually possible */
1319 "nfsd_create: parent %s/%s not locked!\n",
1320 dentry
->d_parent
->d_name
.name
,
1321 dentry
->d_name
.name
);
1327 * Make sure the child dentry is still negative ...
1330 if (dchild
->d_inode
) {
1331 dprintk("nfsd_create: dentry %s/%s not negative!\n",
1332 dentry
->d_name
.name
, dchild
->d_name
.name
);
1336 if (!(iap
->ia_valid
& ATTR_MODE
))
1338 iap
->ia_mode
= (iap
->ia_mode
& S_IALLUGO
) | type
;
1341 if (!S_ISREG(type
) && !S_ISDIR(type
) && !special_file(type
)) {
1342 printk(KERN_WARNING
"nfsd: bad file type %o in nfsd_create\n",
1348 * Get the dir op function pointer.
1354 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1356 nfsd_check_ignore_resizing(iap
);
1359 host_err
= vfs_mkdir(dirp
, dchild
, iap
->ia_mode
);
1365 host_err
= vfs_mknod(dirp
, dchild
, iap
->ia_mode
, rdev
);
1371 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1374 * nfsd_setattr already committed the child. Transactional filesystems
1375 * had a chance to commit changes for both parent and child
1376 * simultaneously making the following commit_metadata a noop.
1378 err2
= nfserrno(commit_metadata(fhp
));
1382 * Update the file handle to get the new inode info.
1385 err
= fh_update(resfhp
);
1387 if (dchild
&& !IS_ERR(dchild
))
1392 err
= nfserrno(host_err
);
1396 #ifdef CONFIG_NFSD_V3
1398 static inline int nfsd_create_is_exclusive(int createmode
)
1400 return createmode
== NFS3_CREATE_EXCLUSIVE
1401 || createmode
== NFS4_CREATE_EXCLUSIVE4_1
;
1405 * NFSv3 and NFSv4 version of nfsd_create
1408 do_nfsd_create(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1409 char *fname
, int flen
, struct iattr
*iap
,
1410 struct svc_fh
*resfhp
, int createmode
, u32
*verifier
,
1411 bool *truncp
, bool *created
)
1413 struct dentry
*dentry
, *dchild
= NULL
;
1417 __u32 v_mtime
=0, v_atime
=0;
1423 if (isdotent(fname
, flen
))
1425 if (!(iap
->ia_valid
& ATTR_MODE
))
1427 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_EXEC
);
1431 dentry
= fhp
->fh_dentry
;
1432 dirp
= dentry
->d_inode
;
1434 /* Get all the sanity checks out of the way before
1435 * we lock the parent. */
1436 err
= nfserr_notdir
;
1437 if (!dirp
->i_op
->lookup
)
1440 host_err
= fh_want_write(fhp
);
1444 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1447 * Compose the response file handle.
1449 dchild
= lookup_one_len(fname
, dentry
, flen
);
1450 host_err
= PTR_ERR(dchild
);
1454 /* If file doesn't exist, check for permissions to create one */
1455 if (!dchild
->d_inode
) {
1456 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1461 err
= fh_compose(resfhp
, fhp
->fh_export
, dchild
, fhp
);
1465 if (nfsd_create_is_exclusive(createmode
)) {
1466 /* solaris7 gets confused (bugid 4218508) if these have
1467 * the high bit set, so just clear the high bits. If this is
1468 * ever changed to use different attrs for storing the
1469 * verifier, then do_open_lookup() will also need to be fixed
1472 v_mtime
= verifier
[0]&0x7fffffff;
1473 v_atime
= verifier
[1]&0x7fffffff;
1476 if (dchild
->d_inode
) {
1479 switch (createmode
) {
1480 case NFS3_CREATE_UNCHECKED
:
1481 if (! S_ISREG(dchild
->d_inode
->i_mode
))
1484 /* in nfsv4, we need to treat this case a little
1485 * differently. we don't want to truncate the
1486 * file now; this would be wrong if the OPEN
1487 * fails for some other reason. furthermore,
1488 * if the size is nonzero, we should ignore it
1489 * according to spec!
1491 *truncp
= (iap
->ia_valid
& ATTR_SIZE
) && !iap
->ia_size
;
1494 iap
->ia_valid
&= ATTR_SIZE
;
1498 case NFS3_CREATE_EXCLUSIVE
:
1499 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1500 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1501 && dchild
->d_inode
->i_size
== 0 ) {
1506 case NFS4_CREATE_EXCLUSIVE4_1
:
1507 if ( dchild
->d_inode
->i_mtime
.tv_sec
== v_mtime
1508 && dchild
->d_inode
->i_atime
.tv_sec
== v_atime
1509 && dchild
->d_inode
->i_size
== 0 ) {
1515 case NFS3_CREATE_GUARDED
:
1522 host_err
= vfs_create(dirp
, dchild
, iap
->ia_mode
, true);
1530 nfsd_check_ignore_resizing(iap
);
1532 if (nfsd_create_is_exclusive(createmode
)) {
1533 /* Cram the verifier into atime/mtime */
1534 iap
->ia_valid
= ATTR_MTIME
|ATTR_ATIME
1535 | ATTR_MTIME_SET
|ATTR_ATIME_SET
;
1536 /* XXX someone who knows this better please fix it for nsec */
1537 iap
->ia_mtime
.tv_sec
= v_mtime
;
1538 iap
->ia_atime
.tv_sec
= v_atime
;
1539 iap
->ia_mtime
.tv_nsec
= 0;
1540 iap
->ia_atime
.tv_nsec
= 0;
1544 err
= nfsd_create_setattr(rqstp
, resfhp
, iap
);
1547 * nfsd_setattr already committed the child (and possibly also the parent).
1550 err
= nfserrno(commit_metadata(fhp
));
1553 * Update the filehandle to get the new inode info.
1556 err
= fh_update(resfhp
);
1560 if (dchild
&& !IS_ERR(dchild
))
1566 err
= nfserrno(host_err
);
1569 #endif /* CONFIG_NFSD_V3 */
1572 * Read a symlink. On entry, *lenp must contain the maximum path length that
1573 * fits into the buffer. On return, it contains the true length.
1574 * N.B. After this call fhp needs an fh_put
1577 nfsd_readlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, char *buf
, int *lenp
)
1579 struct inode
*inode
;
1585 err
= fh_verify(rqstp
, fhp
, S_IFLNK
, NFSD_MAY_NOP
);
1589 path
.mnt
= fhp
->fh_export
->ex_path
.mnt
;
1590 path
.dentry
= fhp
->fh_dentry
;
1591 inode
= path
.dentry
->d_inode
;
1594 if (!inode
->i_op
->readlink
)
1598 /* N.B. Why does this call need a get_fs()??
1599 * Remove the set_fs and watch the fireworks:-) --okir
1602 oldfs
= get_fs(); set_fs(KERNEL_DS
);
1603 host_err
= inode
->i_op
->readlink(path
.dentry
, (char __user
*)buf
, *lenp
);
1614 err
= nfserrno(host_err
);
1619 * Create a symlink and look up its inode
1620 * N.B. After this call _both_ fhp and resfhp need an fh_put
1623 nfsd_symlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
,
1624 char *fname
, int flen
,
1625 char *path
, int plen
,
1626 struct svc_fh
*resfhp
,
1629 struct dentry
*dentry
, *dnew
;
1637 if (isdotent(fname
, flen
))
1640 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1644 host_err
= fh_want_write(fhp
);
1649 dentry
= fhp
->fh_dentry
;
1650 dnew
= lookup_one_len(fname
, dentry
, flen
);
1651 host_err
= PTR_ERR(dnew
);
1655 if (unlikely(path
[plen
] != 0)) {
1656 char *path_alloced
= kmalloc(plen
+1, GFP_KERNEL
);
1657 if (path_alloced
== NULL
)
1660 strncpy(path_alloced
, path
, plen
);
1661 path_alloced
[plen
] = 0;
1662 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path_alloced
);
1663 kfree(path_alloced
);
1666 host_err
= vfs_symlink(dentry
->d_inode
, dnew
, path
);
1667 err
= nfserrno(host_err
);
1669 err
= nfserrno(commit_metadata(fhp
));
1674 cerr
= fh_compose(resfhp
, fhp
->fh_export
, dnew
, fhp
);
1676 if (err
==0) err
= cerr
;
1681 err
= nfserrno(host_err
);
1687 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1690 nfsd_link(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
,
1691 char *name
, int len
, struct svc_fh
*tfhp
)
1693 struct dentry
*ddir
, *dnew
, *dold
;
1698 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1701 err
= fh_verify(rqstp
, tfhp
, 0, NFSD_MAY_NOP
);
1705 if (S_ISDIR(tfhp
->fh_dentry
->d_inode
->i_mode
))
1711 if (isdotent(name
, len
))
1714 host_err
= fh_want_write(tfhp
);
1716 err
= nfserrno(host_err
);
1720 fh_lock_nested(ffhp
, I_MUTEX_PARENT
);
1721 ddir
= ffhp
->fh_dentry
;
1722 dirp
= ddir
->d_inode
;
1724 dnew
= lookup_one_len(name
, ddir
, len
);
1725 host_err
= PTR_ERR(dnew
);
1729 dold
= tfhp
->fh_dentry
;
1734 host_err
= nfsd_break_lease(dold
->d_inode
);
1736 err
= nfserrno(host_err
);
1739 host_err
= vfs_link(dold
, dirp
, dnew
);
1741 err
= nfserrno(commit_metadata(ffhp
));
1743 err
= nfserrno(commit_metadata(tfhp
));
1745 if (host_err
== -EXDEV
&& rqstp
->rq_vers
== 2)
1748 err
= nfserrno(host_err
);
1754 fh_drop_write(tfhp
);
1759 err
= nfserrno(host_err
);
1765 * N.B. After this call _both_ ffhp and tfhp need an fh_put
1768 nfsd_rename(struct svc_rqst
*rqstp
, struct svc_fh
*ffhp
, char *fname
, int flen
,
1769 struct svc_fh
*tfhp
, char *tname
, int tlen
)
1771 struct dentry
*fdentry
, *tdentry
, *odentry
, *ndentry
, *trap
;
1772 struct inode
*fdir
, *tdir
;
1776 err
= fh_verify(rqstp
, ffhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1779 err
= fh_verify(rqstp
, tfhp
, S_IFDIR
, NFSD_MAY_CREATE
);
1783 fdentry
= ffhp
->fh_dentry
;
1784 fdir
= fdentry
->d_inode
;
1786 tdentry
= tfhp
->fh_dentry
;
1787 tdir
= tdentry
->d_inode
;
1790 if (!flen
|| isdotent(fname
, flen
) || !tlen
|| isdotent(tname
, tlen
))
1793 host_err
= fh_want_write(ffhp
);
1795 err
= nfserrno(host_err
);
1799 /* cannot use fh_lock as we need deadlock protective ordering
1800 * so do it by hand */
1801 trap
= lock_rename(tdentry
, fdentry
);
1802 ffhp
->fh_locked
= tfhp
->fh_locked
= 1;
1806 odentry
= lookup_one_len(fname
, fdentry
, flen
);
1807 host_err
= PTR_ERR(odentry
);
1808 if (IS_ERR(odentry
))
1812 if (!odentry
->d_inode
)
1815 if (odentry
== trap
)
1818 ndentry
= lookup_one_len(tname
, tdentry
, tlen
);
1819 host_err
= PTR_ERR(ndentry
);
1820 if (IS_ERR(ndentry
))
1822 host_err
= -ENOTEMPTY
;
1823 if (ndentry
== trap
)
1827 if (ffhp
->fh_export
->ex_path
.mnt
!= tfhp
->fh_export
->ex_path
.mnt
)
1829 if (ffhp
->fh_export
->ex_path
.dentry
!= tfhp
->fh_export
->ex_path
.dentry
)
1832 host_err
= nfsd_break_lease(odentry
->d_inode
);
1835 if (ndentry
->d_inode
) {
1836 host_err
= nfsd_break_lease(ndentry
->d_inode
);
1840 host_err
= vfs_rename(fdir
, odentry
, tdir
, ndentry
);
1842 host_err
= commit_metadata(tfhp
);
1844 host_err
= commit_metadata(ffhp
);
1851 err
= nfserrno(host_err
);
1853 /* we cannot reply on fh_unlock on the two filehandles,
1854 * as that would do the wrong thing if the two directories
1855 * were the same, so again we do it by hand
1857 fill_post_wcc(ffhp
);
1858 fill_post_wcc(tfhp
);
1859 unlock_rename(tdentry
, fdentry
);
1860 ffhp
->fh_locked
= tfhp
->fh_locked
= 0;
1861 fh_drop_write(ffhp
);
1868 * Unlink a file or directory
1869 * N.B. After this call fhp needs an fh_put
1872 nfsd_unlink(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, int type
,
1873 char *fname
, int flen
)
1875 struct dentry
*dentry
, *rdentry
;
1881 if (!flen
|| isdotent(fname
, flen
))
1883 err
= fh_verify(rqstp
, fhp
, S_IFDIR
, NFSD_MAY_REMOVE
);
1887 host_err
= fh_want_write(fhp
);
1891 fh_lock_nested(fhp
, I_MUTEX_PARENT
);
1892 dentry
= fhp
->fh_dentry
;
1893 dirp
= dentry
->d_inode
;
1895 rdentry
= lookup_one_len(fname
, dentry
, flen
);
1896 host_err
= PTR_ERR(rdentry
);
1897 if (IS_ERR(rdentry
))
1900 if (!rdentry
->d_inode
) {
1907 type
= rdentry
->d_inode
->i_mode
& S_IFMT
;
1909 host_err
= nfsd_break_lease(rdentry
->d_inode
);
1912 if (type
!= S_IFDIR
)
1913 host_err
= vfs_unlink(dirp
, rdentry
);
1915 host_err
= vfs_rmdir(dirp
, rdentry
);
1917 host_err
= commit_metadata(fhp
);
1922 err
= nfserrno(host_err
);
1928 * We do this buffering because we must not call back into the file
1929 * system's ->lookup() method from the filldir callback. That may well
1930 * deadlock a number of file systems.
1932 * This is based heavily on the implementation of same in XFS.
1934 struct buffered_dirent
{
1938 unsigned int d_type
;
1942 struct readdir_data
{
1943 struct dir_context ctx
;
1949 static int nfsd_buffered_filldir(void *__buf
, const char *name
, int namlen
,
1950 loff_t offset
, u64 ino
, unsigned int d_type
)
1952 struct readdir_data
*buf
= __buf
;
1953 struct buffered_dirent
*de
= (void *)(buf
->dirent
+ buf
->used
);
1954 unsigned int reclen
;
1956 reclen
= ALIGN(sizeof(struct buffered_dirent
) + namlen
, sizeof(u64
));
1957 if (buf
->used
+ reclen
> PAGE_SIZE
) {
1962 de
->namlen
= namlen
;
1963 de
->offset
= offset
;
1965 de
->d_type
= d_type
;
1966 memcpy(de
->name
, name
, namlen
);
1967 buf
->used
+= reclen
;
1972 static __be32
nfsd_buffered_readdir(struct file
*file
, filldir_t func
,
1973 struct readdir_cd
*cdp
, loff_t
*offsetp
)
1975 struct buffered_dirent
*de
;
1979 struct readdir_data buf
= {
1980 .ctx
.actor
= nfsd_buffered_filldir
,
1981 .dirent
= (void *)__get_free_page(GFP_KERNEL
)
1985 return nfserrno(-ENOMEM
);
1990 struct inode
*dir_inode
= file_inode(file
);
1991 unsigned int reclen
;
1993 cdp
->err
= nfserr_eof
; /* will be cleared on successful read */
1997 host_err
= iterate_dir(file
, &buf
.ctx
);
2010 * Various filldir functions may end up calling back into
2011 * lookup_one_len() and the file system's ->lookup() method.
2012 * These expect i_mutex to be held, as it would within readdir.
2014 host_err
= mutex_lock_killable(&dir_inode
->i_mutex
);
2018 de
= (struct buffered_dirent
*)buf
.dirent
;
2020 offset
= de
->offset
;
2022 if (func(cdp
, de
->name
, de
->namlen
, de
->offset
,
2023 de
->ino
, de
->d_type
))
2026 if (cdp
->err
!= nfs_ok
)
2029 reclen
= ALIGN(sizeof(*de
) + de
->namlen
,
2032 de
= (struct buffered_dirent
*)((char *)de
+ reclen
);
2034 mutex_unlock(&dir_inode
->i_mutex
);
2035 if (size
> 0) /* We bailed out early */
2038 offset
= vfs_llseek(file
, 0, SEEK_CUR
);
2041 free_page((unsigned long)(buf
.dirent
));
2044 return nfserrno(host_err
);
2051 * Read entries from a directory.
2052 * The NFSv3/4 verifier we ignore for now.
2055 nfsd_readdir(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, loff_t
*offsetp
,
2056 struct readdir_cd
*cdp
, filldir_t func
)
2060 loff_t offset
= *offsetp
;
2061 int may_flags
= NFSD_MAY_READ
;
2063 /* NFSv2 only supports 32 bit cookies */
2064 if (rqstp
->rq_vers
> 2)
2065 may_flags
|= NFSD_MAY_64BIT_COOKIE
;
2067 err
= nfsd_open(rqstp
, fhp
, S_IFDIR
, may_flags
, &file
);
2071 offset
= vfs_llseek(file
, offset
, SEEK_SET
);
2073 err
= nfserrno((int)offset
);
2077 err
= nfsd_buffered_readdir(file
, func
, cdp
, offsetp
);
2079 if (err
== nfserr_eof
|| err
== nfserr_toosmall
)
2080 err
= nfs_ok
; /* can still be found in ->err */
2088 * Get file system stats
2089 * N.B. After this call fhp needs an fh_put
2092 nfsd_statfs(struct svc_rqst
*rqstp
, struct svc_fh
*fhp
, struct kstatfs
*stat
, int access
)
2096 err
= fh_verify(rqstp
, fhp
, 0, NFSD_MAY_NOP
| access
);
2098 struct path path
= {
2099 .mnt
= fhp
->fh_export
->ex_path
.mnt
,
2100 .dentry
= fhp
->fh_dentry
,
2102 if (vfs_statfs(&path
, stat
))
2108 static int exp_rdonly(struct svc_rqst
*rqstp
, struct svc_export
*exp
)
2110 return nfsexp_flags(rqstp
, exp
) & NFSEXP_READONLY
;
2114 * Check for a user's access permissions to this inode.
2117 nfsd_permission(struct svc_rqst
*rqstp
, struct svc_export
*exp
,
2118 struct dentry
*dentry
, int acc
)
2120 struct inode
*inode
= dentry
->d_inode
;
2123 if ((acc
& NFSD_MAY_MASK
) == NFSD_MAY_NOP
)
2126 dprintk("nfsd: permission 0x%x%s%s%s%s%s%s%s mode 0%o%s%s%s\n",
2128 (acc
& NFSD_MAY_READ
)? " read" : "",
2129 (acc
& NFSD_MAY_WRITE
)? " write" : "",
2130 (acc
& NFSD_MAY_EXEC
)? " exec" : "",
2131 (acc
& NFSD_MAY_SATTR
)? " sattr" : "",
2132 (acc
& NFSD_MAY_TRUNC
)? " trunc" : "",
2133 (acc
& NFSD_MAY_LOCK
)? " lock" : "",
2134 (acc
& NFSD_MAY_OWNER_OVERRIDE
)? " owneroverride" : "",
2136 IS_IMMUTABLE(inode
)? " immut" : "",
2137 IS_APPEND(inode
)? " append" : "",
2138 __mnt_is_readonly(exp
->ex_path
.mnt
)? " ro" : "");
2139 dprintk(" owner %d/%d user %d/%d\n",
2140 inode
->i_uid
, inode
->i_gid
, current_fsuid(), current_fsgid());
2143 /* Normally we reject any write/sattr etc access on a read-only file
2144 * system. But if it is IRIX doing check on write-access for a
2145 * device special file, we ignore rofs.
2147 if (!(acc
& NFSD_MAY_LOCAL_ACCESS
))
2148 if (acc
& (NFSD_MAY_WRITE
| NFSD_MAY_SATTR
| NFSD_MAY_TRUNC
)) {
2149 if (exp_rdonly(rqstp
, exp
) ||
2150 __mnt_is_readonly(exp
->ex_path
.mnt
))
2152 if (/* (acc & NFSD_MAY_WRITE) && */ IS_IMMUTABLE(inode
))
2155 if ((acc
& NFSD_MAY_TRUNC
) && IS_APPEND(inode
))
2158 if (acc
& NFSD_MAY_LOCK
) {
2159 /* If we cannot rely on authentication in NLM requests,
2160 * just allow locks, otherwise require read permission, or
2163 if (exp
->ex_flags
& NFSEXP_NOAUTHNLM
)
2166 acc
= NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
;
2169 * The file owner always gets access permission for accesses that
2170 * would normally be checked at open time. This is to make
2171 * file access work even when the client has done a fchmod(fd, 0).
2173 * However, `cp foo bar' should fail nevertheless when bar is
2174 * readonly. A sensible way to do this might be to reject all
2175 * attempts to truncate a read-only file, because a creat() call
2176 * always implies file truncation.
2177 * ... but this isn't really fair. A process may reasonably call
2178 * ftruncate on an open file descriptor on a file with perm 000.
2179 * We must trust the client to do permission checking - using "ACCESS"
2182 if ((acc
& NFSD_MAY_OWNER_OVERRIDE
) &&
2183 uid_eq(inode
->i_uid
, current_fsuid()))
2186 /* This assumes NFSD_MAY_{READ,WRITE,EXEC} == MAY_{READ,WRITE,EXEC} */
2187 err
= inode_permission(inode
, acc
& (MAY_READ
|MAY_WRITE
|MAY_EXEC
));
2189 /* Allow read access to binaries even when mode 111 */
2190 if (err
== -EACCES
&& S_ISREG(inode
->i_mode
) &&
2191 (acc
== (NFSD_MAY_READ
| NFSD_MAY_OWNER_OVERRIDE
) ||
2192 acc
== (NFSD_MAY_READ
| NFSD_MAY_READ_IF_EXEC
)))
2193 err
= inode_permission(inode
, MAY_EXEC
);
2195 return err
? nfserrno(err
) : 0;
2199 nfsd_racache_shutdown(void)
2201 struct raparms
*raparm
, *last_raparm
;
2204 dprintk("nfsd: freeing readahead buffers.\n");
2206 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2207 raparm
= raparm_hash
[i
].pb_head
;
2209 last_raparm
= raparm
;
2210 raparm
= raparm
->p_next
;
2213 raparm_hash
[i
].pb_head
= NULL
;
2217 * Initialize readahead param cache
2220 nfsd_racache_init(int cache_size
)
2225 struct raparms
**raparm
= NULL
;
2228 if (raparm_hash
[0].pb_head
)
2230 nperbucket
= DIV_ROUND_UP(cache_size
, RAPARM_HASH_SIZE
);
2233 cache_size
= nperbucket
* RAPARM_HASH_SIZE
;
2235 dprintk("nfsd: allocating %d readahead buffers.\n", cache_size
);
2237 for (i
= 0; i
< RAPARM_HASH_SIZE
; i
++) {
2238 spin_lock_init(&raparm_hash
[i
].pb_lock
);
2240 raparm
= &raparm_hash
[i
].pb_head
;
2241 for (j
= 0; j
< nperbucket
; j
++) {
2242 *raparm
= kzalloc(sizeof(struct raparms
), GFP_KERNEL
);
2245 raparm
= &(*raparm
)->p_next
;
2250 nfsdstats
.ra_size
= cache_size
;
2254 dprintk("nfsd: kmalloc failed, freeing readahead buffers\n");
2255 nfsd_racache_shutdown();
2259 #if defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL)
2261 nfsd_get_posix_acl(struct svc_fh
*fhp
, int type
)
2263 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2267 struct posix_acl
*acl
;
2269 if (!IS_POSIXACL(inode
))
2270 return ERR_PTR(-EOPNOTSUPP
);
2273 case ACL_TYPE_ACCESS
:
2274 name
= POSIX_ACL_XATTR_ACCESS
;
2276 case ACL_TYPE_DEFAULT
:
2277 name
= POSIX_ACL_XATTR_DEFAULT
;
2280 return ERR_PTR(-EOPNOTSUPP
);
2283 size
= nfsd_getxattr(fhp
->fh_dentry
, name
, &value
);
2285 return ERR_PTR(size
);
2287 acl
= posix_acl_from_xattr(&init_user_ns
, value
, size
);
2293 nfsd_set_posix_acl(struct svc_fh
*fhp
, int type
, struct posix_acl
*acl
)
2295 struct inode
*inode
= fhp
->fh_dentry
->d_inode
;
2301 if (!IS_POSIXACL(inode
) ||
2302 !inode
->i_op
->setxattr
|| !inode
->i_op
->removexattr
)
2305 case ACL_TYPE_ACCESS
:
2306 name
= POSIX_ACL_XATTR_ACCESS
;
2308 case ACL_TYPE_DEFAULT
:
2309 name
= POSIX_ACL_XATTR_DEFAULT
;
2315 if (acl
&& acl
->a_count
) {
2316 size
= posix_acl_xattr_size(acl
->a_count
);
2317 value
= kmalloc(size
, GFP_KERNEL
);
2320 error
= posix_acl_to_xattr(&init_user_ns
, acl
, value
, size
);
2327 error
= fh_want_write(fhp
);
2331 error
= vfs_setxattr(fhp
->fh_dentry
, name
, value
, size
, 0);
2333 if (!S_ISDIR(inode
->i_mode
) && type
== ACL_TYPE_DEFAULT
)
2336 error
= vfs_removexattr(fhp
->fh_dentry
, name
);
2337 if (error
== -ENODATA
)
2347 #endif /* defined(CONFIG_NFSD_V2_ACL) || defined(CONFIG_NFSD_V3_ACL) */