4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/nfs_idmap.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
42 #include <asm/system.h>
43 #include <asm/uaccess.h>
47 #include "delegation.h"
51 #include "dns_resolve.h"
54 #define NFSDBG_FACILITY NFSDBG_VFS
56 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
58 /* Default is to see 64-bit inode numbers */
59 static int enable_ino64
= NFS_64_BIT_INODE_NUMBERS_ENABLED
;
61 static void nfs_invalidate_inode(struct inode
*);
62 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
64 static struct kmem_cache
* nfs_inode_cachep
;
66 static inline unsigned long
67 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
69 return nfs_fileid_to_ino_t(fattr
->fileid
);
73 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
74 * @word: long word containing the bit lock
76 int nfs_wait_bit_killable(void *word
)
78 if (fatal_signal_pending(current
))
85 * nfs_compat_user_ino64 - returns the user-visible inode number
86 * @fileid: 64-bit fileid
88 * This function returns a 32-bit inode number if the boot parameter
89 * nfs.enable_ino64 is zero.
91 u64
nfs_compat_user_ino64(u64 fileid
)
102 if (sizeof(ino
) < sizeof(fileid
))
103 ino
^= fileid
>> (sizeof(fileid
)-sizeof(ino
)) * 8;
107 static void nfs_clear_inode(struct inode
*inode
)
110 * The following should never happen...
112 BUG_ON(nfs_have_writebacks(inode
));
113 BUG_ON(!list_empty(&NFS_I(inode
)->open_files
));
114 nfs_zap_acl_cache(inode
);
115 nfs_access_zap_cache(inode
);
116 nfs_fscache_release_inode_cookie(inode
);
119 void nfs_evict_inode(struct inode
*inode
)
121 truncate_inode_pages(&inode
->i_data
, 0);
122 end_writeback(inode
);
123 nfs_clear_inode(inode
);
127 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
129 int nfs_sync_mapping(struct address_space
*mapping
)
133 if (mapping
->nrpages
!= 0) {
134 unmap_mapping_range(mapping
, 0, 0, 0);
135 ret
= nfs_wb_all(mapping
->host
);
141 * Invalidate the local caches
143 static void nfs_zap_caches_locked(struct inode
*inode
)
145 struct nfs_inode
*nfsi
= NFS_I(inode
);
146 int mode
= inode
->i_mode
;
148 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
150 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
151 nfsi
->attrtimeo_timestamp
= jiffies
;
153 memset(NFS_COOKIEVERF(inode
), 0, sizeof(NFS_COOKIEVERF(inode
)));
154 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))
155 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
157 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
160 void nfs_zap_caches(struct inode
*inode
)
162 spin_lock(&inode
->i_lock
);
163 nfs_zap_caches_locked(inode
);
164 spin_unlock(&inode
->i_lock
);
167 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
169 if (mapping
->nrpages
!= 0) {
170 spin_lock(&inode
->i_lock
);
171 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
172 spin_unlock(&inode
->i_lock
);
176 void nfs_zap_acl_cache(struct inode
*inode
)
178 void (*clear_acl_cache
)(struct inode
*);
180 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
181 if (clear_acl_cache
!= NULL
)
182 clear_acl_cache(inode
);
183 spin_lock(&inode
->i_lock
);
184 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
185 spin_unlock(&inode
->i_lock
);
188 void nfs_invalidate_atime(struct inode
*inode
)
190 spin_lock(&inode
->i_lock
);
191 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
192 spin_unlock(&inode
->i_lock
);
196 * Invalidate, but do not unhash, the inode.
197 * NB: must be called with inode->i_lock held!
199 static void nfs_invalidate_inode(struct inode
*inode
)
201 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
202 nfs_zap_caches_locked(inode
);
205 struct nfs_find_desc
{
207 struct nfs_fattr
*fattr
;
211 * In NFSv3 we can have 64bit inode numbers. In order to support
212 * this, and re-exported directories (also seen in NFSv2)
213 * we are forced to allow 2 different inodes to have the same
217 nfs_find_actor(struct inode
*inode
, void *opaque
)
219 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
220 struct nfs_fh
*fh
= desc
->fh
;
221 struct nfs_fattr
*fattr
= desc
->fattr
;
223 if (NFS_FILEID(inode
) != fattr
->fileid
)
225 if (nfs_compare_fh(NFS_FH(inode
), fh
))
227 if (is_bad_inode(inode
) || NFS_STALE(inode
))
233 nfs_init_locked(struct inode
*inode
, void *opaque
)
235 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
236 struct nfs_fattr
*fattr
= desc
->fattr
;
238 set_nfs_fileid(inode
, fattr
->fileid
);
239 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
244 * This is our front-end to iget that looks up inodes by file handle
245 * instead of inode number.
248 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
250 struct nfs_find_desc desc
= {
254 struct inode
*inode
= ERR_PTR(-ENOENT
);
257 if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) == 0)
259 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) == 0)
262 hash
= nfs_fattr_to_ino_t(fattr
);
264 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
266 inode
= ERR_PTR(-ENOMEM
);
270 if (inode
->i_state
& I_NEW
) {
271 struct nfs_inode
*nfsi
= NFS_I(inode
);
272 unsigned long now
= jiffies
;
274 /* We set i_ino for the few things that still rely on it,
278 /* We can't support update_atime(), since the server will reset it */
279 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
280 inode
->i_mode
= fattr
->mode
;
281 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) == 0
282 && nfs_server_capable(inode
, NFS_CAP_MODE
))
283 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
284 | NFS_INO_INVALID_ACCESS
285 | NFS_INO_INVALID_ACL
;
286 /* Why so? Because we want revalidate for devices/FIFOs, and
287 * that's precisely what we have in nfs_file_inode_operations.
289 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
290 if (S_ISREG(inode
->i_mode
)) {
291 inode
->i_fop
= &nfs_file_operations
;
292 inode
->i_data
.a_ops
= &nfs_file_aops
;
293 inode
->i_data
.backing_dev_info
= &NFS_SB(sb
)->backing_dev_info
;
294 } else if (S_ISDIR(inode
->i_mode
)) {
295 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
296 inode
->i_fop
= &nfs_dir_operations
;
297 inode
->i_data
.a_ops
= &nfs_dir_aops
;
298 if (nfs_server_capable(inode
, NFS_CAP_READDIRPLUS
))
299 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
300 /* Deal with crossing mountpoints */
301 if ((fattr
->valid
& NFS_ATTR_FATTR_FSID
)
302 && !nfs_fsid_equal(&NFS_SB(sb
)->fsid
, &fattr
->fsid
)) {
303 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
304 inode
->i_op
= &nfs_referral_inode_operations
;
306 inode
->i_op
= &nfs_mountpoint_inode_operations
;
308 inode
->i_flags
|= S_AUTOMOUNT
;
310 } else if (S_ISLNK(inode
->i_mode
))
311 inode
->i_op
= &nfs_symlink_inode_operations
;
313 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
315 memset(&inode
->i_atime
, 0, sizeof(inode
->i_atime
));
316 memset(&inode
->i_mtime
, 0, sizeof(inode
->i_mtime
));
317 memset(&inode
->i_ctime
, 0, sizeof(inode
->i_ctime
));
318 nfsi
->change_attr
= 0;
324 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
326 nfsi
->read_cache_jiffies
= fattr
->time_start
;
327 nfsi
->attr_gencount
= fattr
->gencount
;
328 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
329 inode
->i_atime
= fattr
->atime
;
330 else if (nfs_server_capable(inode
, NFS_CAP_ATIME
))
331 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
332 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
333 inode
->i_mtime
= fattr
->mtime
;
334 else if (nfs_server_capable(inode
, NFS_CAP_MTIME
))
335 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
336 | NFS_INO_INVALID_DATA
;
337 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
338 inode
->i_ctime
= fattr
->ctime
;
339 else if (nfs_server_capable(inode
, NFS_CAP_CTIME
))
340 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
341 | NFS_INO_INVALID_ACCESS
342 | NFS_INO_INVALID_ACL
;
343 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
344 nfsi
->change_attr
= fattr
->change_attr
;
345 else if (nfs_server_capable(inode
, NFS_CAP_CHANGE_ATTR
))
346 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
347 | NFS_INO_INVALID_DATA
;
348 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
349 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
351 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
352 | NFS_INO_INVALID_DATA
353 | NFS_INO_REVAL_PAGECACHE
;
354 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
)
355 inode
->i_nlink
= fattr
->nlink
;
356 else if (nfs_server_capable(inode
, NFS_CAP_NLINK
))
357 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
358 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
)
359 inode
->i_uid
= fattr
->uid
;
360 else if (nfs_server_capable(inode
, NFS_CAP_OWNER
))
361 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
362 | NFS_INO_INVALID_ACCESS
363 | NFS_INO_INVALID_ACL
;
364 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
)
365 inode
->i_gid
= fattr
->gid
;
366 else if (nfs_server_capable(inode
, NFS_CAP_OWNER_GROUP
))
367 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
368 | NFS_INO_INVALID_ACCESS
369 | NFS_INO_INVALID_ACL
;
370 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
371 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
372 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
374 * report the blocks in 512byte units
376 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
378 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
379 nfsi
->attrtimeo_timestamp
= now
;
380 nfsi
->access_cache
= RB_ROOT
;
382 nfs_fscache_init_inode_cookie(inode
);
384 unlock_new_inode(inode
);
386 nfs_refresh_inode(inode
, fattr
);
387 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
389 (long long)NFS_FILEID(inode
),
390 atomic_read(&inode
->i_count
));
396 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
400 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
403 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
405 struct inode
*inode
= dentry
->d_inode
;
406 struct nfs_fattr
*fattr
;
409 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
411 /* skip mode change if it's just for clearing setuid/setgid */
412 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
413 attr
->ia_valid
&= ~ATTR_MODE
;
415 if (attr
->ia_valid
& ATTR_SIZE
) {
416 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
417 attr
->ia_valid
&= ~ATTR_SIZE
;
420 /* Optimization: if the end result is no change, don't RPC */
421 attr
->ia_valid
&= NFS_VALID_ATTRS
;
422 if ((attr
->ia_valid
& ~ATTR_FILE
) == 0)
425 /* Write all dirty data */
426 if (S_ISREG(inode
->i_mode
))
429 fattr
= nfs_alloc_fattr();
433 * Return any delegations if we're going to change ACLs
435 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
436 nfs_inode_return_delegation(inode
);
437 error
= NFS_PROTO(inode
)->setattr(dentry
, fattr
, attr
);
439 nfs_refresh_inode(inode
, fattr
);
440 nfs_free_fattr(fattr
);
446 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
447 * @inode: inode of the file used
448 * @offset: file offset to start truncating
450 * This is a copy of the common vmtruncate, but with the locking
451 * corrected to take into account the fact that NFS requires
452 * inode->i_size to be updated under the inode->i_lock.
454 static int nfs_vmtruncate(struct inode
* inode
, loff_t offset
)
459 err
= inode_newsize_ok(inode
, offset
);
463 spin_lock(&inode
->i_lock
);
464 oldsize
= inode
->i_size
;
465 i_size_write(inode
, offset
);
466 spin_unlock(&inode
->i_lock
);
468 truncate_pagecache(inode
, oldsize
, offset
);
474 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
475 * @inode: pointer to struct inode
476 * @attr: pointer to struct iattr
478 * Note: we do this in the *proc.c in order to ensure that
479 * it works for things like exclusive creates too.
481 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
483 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
484 spin_lock(&inode
->i_lock
);
485 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
486 int mode
= attr
->ia_mode
& S_IALLUGO
;
487 mode
|= inode
->i_mode
& ~S_IALLUGO
;
488 inode
->i_mode
= mode
;
490 if ((attr
->ia_valid
& ATTR_UID
) != 0)
491 inode
->i_uid
= attr
->ia_uid
;
492 if ((attr
->ia_valid
& ATTR_GID
) != 0)
493 inode
->i_gid
= attr
->ia_gid
;
494 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
495 spin_unlock(&inode
->i_lock
);
497 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
498 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
499 nfs_vmtruncate(inode
, attr
->ia_size
);
503 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
505 struct inode
*inode
= dentry
->d_inode
;
506 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
509 /* Flush out writes to the server in order to update c/mtime. */
510 if (S_ISREG(inode
->i_mode
)) {
511 err
= filemap_write_and_wait(inode
->i_mapping
);
517 * We may force a getattr if the user cares about atime.
519 * Note that we only have to check the vfsmount flags here:
520 * - NFS always sets S_NOATIME by so checking it would give a
522 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
523 * no point in checking those.
525 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
526 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
530 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
532 err
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
534 generic_fillattr(inode
, stat
);
535 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
541 static void nfs_init_lock_context(struct nfs_lock_context
*l_ctx
)
543 atomic_set(&l_ctx
->count
, 1);
544 l_ctx
->lockowner
= current
->files
;
545 l_ctx
->pid
= current
->tgid
;
546 INIT_LIST_HEAD(&l_ctx
->list
);
549 static struct nfs_lock_context
*__nfs_find_lock_context(struct nfs_open_context
*ctx
)
551 struct nfs_lock_context
*pos
;
553 list_for_each_entry(pos
, &ctx
->lock_context
.list
, list
) {
554 if (pos
->lockowner
!= current
->files
)
556 if (pos
->pid
!= current
->tgid
)
558 atomic_inc(&pos
->count
);
564 struct nfs_lock_context
*nfs_get_lock_context(struct nfs_open_context
*ctx
)
566 struct nfs_lock_context
*res
, *new = NULL
;
567 struct inode
*inode
= ctx
->path
.dentry
->d_inode
;
569 spin_lock(&inode
->i_lock
);
570 res
= __nfs_find_lock_context(ctx
);
572 spin_unlock(&inode
->i_lock
);
573 new = kmalloc(sizeof(*new), GFP_KERNEL
);
576 nfs_init_lock_context(new);
577 spin_lock(&inode
->i_lock
);
578 res
= __nfs_find_lock_context(ctx
);
580 list_add_tail(&new->list
, &ctx
->lock_context
.list
);
581 new->open_context
= ctx
;
586 spin_unlock(&inode
->i_lock
);
591 void nfs_put_lock_context(struct nfs_lock_context
*l_ctx
)
593 struct nfs_open_context
*ctx
= l_ctx
->open_context
;
594 struct inode
*inode
= ctx
->path
.dentry
->d_inode
;
596 if (!atomic_dec_and_lock(&l_ctx
->count
, &inode
->i_lock
))
598 list_del(&l_ctx
->list
);
599 spin_unlock(&inode
->i_lock
);
604 * nfs_close_context - Common close_context() routine NFSv2/v3
605 * @ctx: pointer to context
606 * @is_sync: is this a synchronous close
608 * always ensure that the attributes are up to date if we're mounted
609 * with close-to-open semantics
611 void nfs_close_context(struct nfs_open_context
*ctx
, int is_sync
)
614 struct nfs_server
*server
;
616 if (!(ctx
->mode
& FMODE_WRITE
))
620 inode
= ctx
->path
.dentry
->d_inode
;
621 if (!list_empty(&NFS_I(inode
)->open_files
))
623 server
= NFS_SERVER(inode
);
624 if (server
->flags
& NFS_MOUNT_NOCTO
)
626 nfs_revalidate_inode(server
, inode
);
629 struct nfs_open_context
*alloc_nfs_open_context(struct path
*path
, struct rpc_cred
*cred
, fmode_t f_mode
)
631 struct nfs_open_context
*ctx
;
633 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
636 path_get(&ctx
->path
);
637 ctx
->cred
= get_rpccred(cred
);
643 nfs_init_lock_context(&ctx
->lock_context
);
644 ctx
->lock_context
.open_context
= ctx
;
645 INIT_LIST_HEAD(&ctx
->list
);
650 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
653 atomic_inc(&ctx
->lock_context
.count
);
657 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int is_sync
)
659 struct inode
*inode
= ctx
->path
.dentry
->d_inode
;
661 if (!list_empty(&ctx
->list
)) {
662 if (!atomic_dec_and_lock(&ctx
->lock_context
.count
, &inode
->i_lock
))
664 list_del(&ctx
->list
);
665 spin_unlock(&inode
->i_lock
);
666 } else if (!atomic_dec_and_test(&ctx
->lock_context
.count
))
669 NFS_PROTO(inode
)->close_context(ctx
, is_sync
);
670 if (ctx
->cred
!= NULL
)
671 put_rpccred(ctx
->cred
);
672 path_put(&ctx
->path
);
676 void put_nfs_open_context(struct nfs_open_context
*ctx
)
678 __put_nfs_open_context(ctx
, 0);
682 * Ensure that mmap has a recent RPC credential for use when writing out
685 void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
687 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
688 struct nfs_inode
*nfsi
= NFS_I(inode
);
690 filp
->private_data
= get_nfs_open_context(ctx
);
691 spin_lock(&inode
->i_lock
);
692 list_add(&ctx
->list
, &nfsi
->open_files
);
693 spin_unlock(&inode
->i_lock
);
697 * Given an inode, search for an open context with the desired characteristics
699 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, fmode_t mode
)
701 struct nfs_inode
*nfsi
= NFS_I(inode
);
702 struct nfs_open_context
*pos
, *ctx
= NULL
;
704 spin_lock(&inode
->i_lock
);
705 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
706 if (cred
!= NULL
&& pos
->cred
!= cred
)
708 if ((pos
->mode
& (FMODE_READ
|FMODE_WRITE
)) != mode
)
710 ctx
= get_nfs_open_context(pos
);
713 spin_unlock(&inode
->i_lock
);
717 static void nfs_file_clear_open_context(struct file
*filp
)
719 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
720 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
723 filp
->private_data
= NULL
;
724 spin_lock(&inode
->i_lock
);
725 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
726 spin_unlock(&inode
->i_lock
);
727 __put_nfs_open_context(ctx
, filp
->f_flags
& O_DIRECT
? 0 : 1);
732 * These allocate and release file read/write context information.
734 int nfs_open(struct inode
*inode
, struct file
*filp
)
736 struct nfs_open_context
*ctx
;
737 struct rpc_cred
*cred
;
739 cred
= rpc_lookup_cred();
741 return PTR_ERR(cred
);
742 ctx
= alloc_nfs_open_context(&filp
->f_path
, cred
, filp
->f_mode
);
746 nfs_file_set_open_context(filp
, ctx
);
747 put_nfs_open_context(ctx
);
748 nfs_fscache_set_inode_cookie(inode
, filp
);
752 int nfs_release(struct inode
*inode
, struct file
*filp
)
754 nfs_file_clear_open_context(filp
);
759 * This function is called whenever some part of NFS notices that
760 * the cached attributes have to be refreshed.
763 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
765 int status
= -ESTALE
;
766 struct nfs_fattr
*fattr
= NULL
;
767 struct nfs_inode
*nfsi
= NFS_I(inode
);
769 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
770 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
772 if (is_bad_inode(inode
))
774 if (NFS_STALE(inode
))
778 fattr
= nfs_alloc_fattr();
782 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
783 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), fattr
);
785 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
787 (long long)NFS_FILEID(inode
), status
);
788 if (status
== -ESTALE
) {
789 nfs_zap_caches(inode
);
790 if (!S_ISDIR(inode
->i_mode
))
791 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
796 status
= nfs_refresh_inode(inode
, fattr
);
798 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
800 (long long)NFS_FILEID(inode
), status
);
804 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
805 nfs_zap_acl_cache(inode
);
807 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
809 (long long)NFS_FILEID(inode
));
812 nfs_free_fattr(fattr
);
816 int nfs_attribute_timeout(struct inode
*inode
)
818 struct nfs_inode
*nfsi
= NFS_I(inode
);
820 return !time_in_range_open(jiffies
, nfsi
->read_cache_jiffies
, nfsi
->read_cache_jiffies
+ nfsi
->attrtimeo
);
823 static int nfs_attribute_cache_expired(struct inode
*inode
)
825 if (nfs_have_delegated_attributes(inode
))
827 return nfs_attribute_timeout(inode
);
831 * nfs_revalidate_inode - Revalidate the inode attributes
832 * @server - pointer to nfs_server struct
833 * @inode - pointer to inode struct
835 * Updates inode attribute information by retrieving the data from the server.
837 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
839 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
840 && !nfs_attribute_cache_expired(inode
))
841 return NFS_STALE(inode
) ? -ESTALE
: 0;
842 return __nfs_revalidate_inode(server
, inode
);
845 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
847 struct nfs_inode
*nfsi
= NFS_I(inode
);
849 if (mapping
->nrpages
!= 0) {
850 int ret
= invalidate_inode_pages2(mapping
);
854 spin_lock(&inode
->i_lock
);
855 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
856 if (S_ISDIR(inode
->i_mode
))
857 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
858 spin_unlock(&inode
->i_lock
);
859 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
860 nfs_fscache_reset_inode_cookie(inode
);
861 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
862 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
867 * nfs_revalidate_mapping - Revalidate the pagecache
868 * @inode - pointer to host inode
869 * @mapping - pointer to mapping
871 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
873 struct nfs_inode
*nfsi
= NFS_I(inode
);
876 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
877 || nfs_attribute_cache_expired(inode
)
878 || NFS_STALE(inode
)) {
879 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
883 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
884 ret
= nfs_invalidate_mapping(inode
, mapping
);
889 static unsigned long nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
891 struct nfs_inode
*nfsi
= NFS_I(inode
);
892 unsigned long ret
= 0;
894 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
)
895 && (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
896 && nfsi
->change_attr
== fattr
->pre_change_attr
) {
897 nfsi
->change_attr
= fattr
->change_attr
;
898 if (S_ISDIR(inode
->i_mode
))
899 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
900 ret
|= NFS_INO_INVALID_ATTR
;
902 /* If we have atomic WCC data, we may update some attributes */
903 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
)
904 && (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
905 && timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
)) {
906 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
907 ret
|= NFS_INO_INVALID_ATTR
;
910 if ((fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
)
911 && (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
912 && timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
913 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
914 if (S_ISDIR(inode
->i_mode
))
915 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
916 ret
|= NFS_INO_INVALID_ATTR
;
918 if ((fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
)
919 && (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
920 && i_size_read(inode
) == nfs_size_to_loff_t(fattr
->pre_size
)
921 && nfsi
->npages
== 0) {
922 i_size_write(inode
, nfs_size_to_loff_t(fattr
->size
));
923 ret
|= NFS_INO_INVALID_ATTR
;
929 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
930 * @inode - pointer to inode
931 * @fattr - updated attributes
933 * Verifies the attribute cache. If we have just changed the attributes,
934 * so that fattr carries weak cache consistency data, then it may
935 * also update the ctime/mtime/change_attribute.
937 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
939 struct nfs_inode
*nfsi
= NFS_I(inode
);
940 loff_t cur_size
, new_isize
;
941 unsigned long invalid
= 0;
944 /* Has the inode gone and changed behind our back? */
945 if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) && nfsi
->fileid
!= fattr
->fileid
)
947 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
950 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
951 nfsi
->change_attr
!= fattr
->change_attr
)
952 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
954 /* Verify a few of the more important attributes */
955 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) && !timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
956 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
958 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
959 cur_size
= i_size_read(inode
);
960 new_isize
= nfs_size_to_loff_t(fattr
->size
);
961 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
962 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
965 /* Have any file permissions changed? */
966 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) && (inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
))
967 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
968 if ((fattr
->valid
& NFS_ATTR_FATTR_OWNER
) && inode
->i_uid
!= fattr
->uid
)
969 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
970 if ((fattr
->valid
& NFS_ATTR_FATTR_GROUP
) && inode
->i_gid
!= fattr
->gid
)
971 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
973 /* Has the link count changed? */
974 if ((fattr
->valid
& NFS_ATTR_FATTR_NLINK
) && inode
->i_nlink
!= fattr
->nlink
)
975 invalid
|= NFS_INO_INVALID_ATTR
;
977 if ((fattr
->valid
& NFS_ATTR_FATTR_ATIME
) && !timespec_equal(&inode
->i_atime
, &fattr
->atime
))
978 invalid
|= NFS_INO_INVALID_ATIME
;
981 nfsi
->cache_validity
|= invalid
;
983 nfsi
->read_cache_jiffies
= fattr
->time_start
;
987 static int nfs_ctime_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
989 if (!(fattr
->valid
& NFS_ATTR_FATTR_CTIME
))
991 return timespec_compare(&fattr
->ctime
, &inode
->i_ctime
) > 0;
994 static int nfs_size_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
996 if (!(fattr
->valid
& NFS_ATTR_FATTR_SIZE
))
998 return nfs_size_to_loff_t(fattr
->size
) > i_size_read(inode
);
1001 static atomic_long_t nfs_attr_generation_counter
;
1003 static unsigned long nfs_read_attr_generation_counter(void)
1005 return atomic_long_read(&nfs_attr_generation_counter
);
1008 unsigned long nfs_inc_attr_generation_counter(void)
1010 return atomic_long_inc_return(&nfs_attr_generation_counter
);
1013 void nfs_fattr_init(struct nfs_fattr
*fattr
)
1016 fattr
->time_start
= jiffies
;
1017 fattr
->gencount
= nfs_inc_attr_generation_counter();
1020 struct nfs_fattr
*nfs_alloc_fattr(void)
1022 struct nfs_fattr
*fattr
;
1024 fattr
= kmalloc(sizeof(*fattr
), GFP_NOFS
);
1026 nfs_fattr_init(fattr
);
1030 struct nfs_fh
*nfs_alloc_fhandle(void)
1034 fh
= kmalloc(sizeof(struct nfs_fh
), GFP_NOFS
);
1041 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1042 * @inode - pointer to inode
1043 * @fattr - attributes
1045 * Attempt to divine whether or not an RPC call reply carrying stale
1046 * attributes got scheduled after another call carrying updated ones.
1048 * To do so, the function first assumes that a more recent ctime means
1049 * that the attributes in fattr are newer, however it also attempt to
1050 * catch the case where ctime either didn't change, or went backwards
1051 * (if someone reset the clock on the server) by looking at whether
1052 * or not this RPC call was started after the inode was last updated.
1053 * Note also the check for wraparound of 'attr_gencount'
1055 * The function returns 'true' if it thinks the attributes in 'fattr' are
1056 * more recent than the ones cached in the inode.
1059 static int nfs_inode_attrs_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
1061 const struct nfs_inode
*nfsi
= NFS_I(inode
);
1063 return ((long)fattr
->gencount
- (long)nfsi
->attr_gencount
) > 0 ||
1064 nfs_ctime_need_update(inode
, fattr
) ||
1065 nfs_size_need_update(inode
, fattr
) ||
1066 ((long)nfsi
->attr_gencount
- (long)nfs_read_attr_generation_counter() > 0);
1069 static int nfs_refresh_inode_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1071 if (nfs_inode_attrs_need_update(inode
, fattr
))
1072 return nfs_update_inode(inode
, fattr
);
1073 return nfs_check_inode_attributes(inode
, fattr
);
1077 * nfs_refresh_inode - try to update the inode attribute cache
1078 * @inode - pointer to inode
1079 * @fattr - updated attributes
1081 * Check that an RPC call that returned attributes has not overlapped with
1082 * other recent updates of the inode metadata, then decide whether it is
1083 * safe to do a full update of the inode attributes, or whether just to
1084 * call nfs_check_inode_attributes.
1086 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1090 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1092 spin_lock(&inode
->i_lock
);
1093 status
= nfs_refresh_inode_locked(inode
, fattr
);
1094 spin_unlock(&inode
->i_lock
);
1099 static int nfs_post_op_update_inode_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1101 struct nfs_inode
*nfsi
= NFS_I(inode
);
1103 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
1104 if (S_ISDIR(inode
->i_mode
))
1105 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
1106 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1108 return nfs_refresh_inode_locked(inode
, fattr
);
1112 * nfs_post_op_update_inode - try to update the inode attribute cache
1113 * @inode - pointer to inode
1114 * @fattr - updated attributes
1116 * After an operation that has changed the inode metadata, mark the
1117 * attribute cache as being invalid, then try to update it.
1119 * NB: if the server didn't return any post op attributes, this
1120 * function will force the retrieval of attributes before the next
1121 * NFS request. Thus it should be used only for operations that
1122 * are expected to change one or more attributes, to avoid
1123 * unnecessary NFS requests and trips through nfs_update_inode().
1125 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1129 spin_lock(&inode
->i_lock
);
1130 status
= nfs_post_op_update_inode_locked(inode
, fattr
);
1131 spin_unlock(&inode
->i_lock
);
1136 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1137 * @inode - pointer to inode
1138 * @fattr - updated attributes
1140 * After an operation that has changed the inode metadata, mark the
1141 * attribute cache as being invalid, then try to update it. Fake up
1142 * weak cache consistency data, if none exist.
1144 * This function is mainly designed to be used by the ->write_done() functions.
1146 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
1150 spin_lock(&inode
->i_lock
);
1151 /* Don't do a WCC update if these attributes are already stale */
1152 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0 ||
1153 !nfs_inode_attrs_need_update(inode
, fattr
)) {
1154 fattr
->valid
&= ~(NFS_ATTR_FATTR_PRECHANGE
1155 | NFS_ATTR_FATTR_PRESIZE
1156 | NFS_ATTR_FATTR_PREMTIME
1157 | NFS_ATTR_FATTR_PRECTIME
);
1160 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1161 (fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
) == 0) {
1162 fattr
->pre_change_attr
= NFS_I(inode
)->change_attr
;
1163 fattr
->valid
|= NFS_ATTR_FATTR_PRECHANGE
;
1165 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) != 0 &&
1166 (fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
) == 0) {
1167 memcpy(&fattr
->pre_ctime
, &inode
->i_ctime
, sizeof(fattr
->pre_ctime
));
1168 fattr
->valid
|= NFS_ATTR_FATTR_PRECTIME
;
1170 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) != 0 &&
1171 (fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
) == 0) {
1172 memcpy(&fattr
->pre_mtime
, &inode
->i_mtime
, sizeof(fattr
->pre_mtime
));
1173 fattr
->valid
|= NFS_ATTR_FATTR_PREMTIME
;
1175 if ((fattr
->valid
& NFS_ATTR_FATTR_SIZE
) != 0 &&
1176 (fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
) == 0) {
1177 fattr
->pre_size
= i_size_read(inode
);
1178 fattr
->valid
|= NFS_ATTR_FATTR_PRESIZE
;
1181 status
= nfs_post_op_update_inode_locked(inode
, fattr
);
1182 spin_unlock(&inode
->i_lock
);
1187 * Many nfs protocol calls return the new file attributes after
1188 * an operation. Here we update the inode to reflect the state
1189 * of the server's inode.
1191 * This is a bit tricky because we have to make sure all dirty pages
1192 * have been sent off to the server before calling invalidate_inode_pages.
1193 * To make sure no other process adds more write requests while we try
1194 * our best to flush them, we make them sleep during the attribute refresh.
1196 * A very similar scenario holds for the dir cache.
1198 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1200 struct nfs_server
*server
;
1201 struct nfs_inode
*nfsi
= NFS_I(inode
);
1202 loff_t cur_isize
, new_isize
;
1203 unsigned long invalid
= 0;
1204 unsigned long now
= jiffies
;
1205 unsigned long save_cache_validity
;
1207 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1208 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1209 atomic_read(&inode
->i_count
), fattr
->valid
);
1211 if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) && nfsi
->fileid
!= fattr
->fileid
)
1215 * Make sure the inode's type hasn't changed.
1217 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
1220 server
= NFS_SERVER(inode
);
1221 /* Update the fsid? */
1222 if (S_ISDIR(inode
->i_mode
) && (fattr
->valid
& NFS_ATTR_FATTR_FSID
) &&
1223 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
1224 !IS_AUTOMOUNT(inode
))
1225 server
->fsid
= fattr
->fsid
;
1228 * Update the read time so we don't revalidate too often.
1230 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1232 save_cache_validity
= nfsi
->cache_validity
;
1233 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
1234 | NFS_INO_INVALID_ATIME
1235 | NFS_INO_REVAL_FORCED
1236 | NFS_INO_REVAL_PAGECACHE
);
1238 /* Do atomic weak cache consistency updates */
1239 invalid
|= nfs_wcc_update_inode(inode
, fattr
);
1241 /* More cache consistency checks */
1242 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) {
1243 if (nfsi
->change_attr
!= fattr
->change_attr
) {
1244 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1245 inode
->i_sb
->s_id
, inode
->i_ino
);
1246 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1247 if (S_ISDIR(inode
->i_mode
))
1248 nfs_force_lookup_revalidate(inode
);
1249 nfsi
->change_attr
= fattr
->change_attr
;
1251 } else if (server
->caps
& NFS_CAP_CHANGE_ATTR
)
1252 invalid
|= save_cache_validity
;
1254 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
) {
1255 /* NFSv2/v3: Check if the mtime agrees */
1256 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
1257 dprintk("NFS: mtime change on server for file %s/%ld\n",
1258 inode
->i_sb
->s_id
, inode
->i_ino
);
1259 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1260 if (S_ISDIR(inode
->i_mode
))
1261 nfs_force_lookup_revalidate(inode
);
1262 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1264 } else if (server
->caps
& NFS_CAP_MTIME
)
1265 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1266 | NFS_INO_INVALID_DATA
1267 | NFS_INO_REVAL_PAGECACHE
1268 | NFS_INO_REVAL_FORCED
);
1270 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
) {
1271 /* If ctime has changed we should definitely clear access+acl caches */
1272 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
)) {
1273 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1274 /* and probably clear data for a directory too as utimes can cause
1275 * havoc with our cache.
1277 if (S_ISDIR(inode
->i_mode
)) {
1278 invalid
|= NFS_INO_INVALID_DATA
;
1279 nfs_force_lookup_revalidate(inode
);
1281 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1283 } else if (server
->caps
& NFS_CAP_CTIME
)
1284 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1285 | NFS_INO_INVALID_ACCESS
1286 | NFS_INO_INVALID_ACL
1287 | NFS_INO_REVAL_FORCED
);
1289 /* Check if our cached file size is stale */
1290 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1291 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1292 cur_isize
= i_size_read(inode
);
1293 if (new_isize
!= cur_isize
) {
1294 /* Do we perhaps have any outstanding writes, or has
1295 * the file grown beyond our last write? */
1296 if (nfsi
->npages
== 0 || new_isize
> cur_isize
) {
1297 i_size_write(inode
, new_isize
);
1298 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1300 dprintk("NFS: isize change on server for file %s/%ld\n",
1301 inode
->i_sb
->s_id
, inode
->i_ino
);
1304 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1305 | NFS_INO_REVAL_PAGECACHE
1306 | NFS_INO_REVAL_FORCED
);
1309 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
1310 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1311 else if (server
->caps
& NFS_CAP_ATIME
)
1312 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATIME
1313 | NFS_INO_REVAL_FORCED
);
1315 if (fattr
->valid
& NFS_ATTR_FATTR_MODE
) {
1316 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)) {
1317 umode_t newmode
= inode
->i_mode
& S_IFMT
;
1318 newmode
|= fattr
->mode
& S_IALLUGO
;
1319 inode
->i_mode
= newmode
;
1320 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1322 } else if (server
->caps
& NFS_CAP_MODE
)
1323 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1324 | NFS_INO_INVALID_ACCESS
1325 | NFS_INO_INVALID_ACL
1326 | NFS_INO_REVAL_FORCED
);
1328 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
) {
1329 if (inode
->i_uid
!= fattr
->uid
) {
1330 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1331 inode
->i_uid
= fattr
->uid
;
1333 } else if (server
->caps
& NFS_CAP_OWNER
)
1334 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1335 | NFS_INO_INVALID_ACCESS
1336 | NFS_INO_INVALID_ACL
1337 | NFS_INO_REVAL_FORCED
);
1339 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
) {
1340 if (inode
->i_gid
!= fattr
->gid
) {
1341 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1342 inode
->i_gid
= fattr
->gid
;
1344 } else if (server
->caps
& NFS_CAP_OWNER_GROUP
)
1345 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1346 | NFS_INO_INVALID_ACCESS
1347 | NFS_INO_INVALID_ACL
1348 | NFS_INO_REVAL_FORCED
);
1350 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
) {
1351 if (inode
->i_nlink
!= fattr
->nlink
) {
1352 invalid
|= NFS_INO_INVALID_ATTR
;
1353 if (S_ISDIR(inode
->i_mode
))
1354 invalid
|= NFS_INO_INVALID_DATA
;
1355 inode
->i_nlink
= fattr
->nlink
;
1357 } else if (server
->caps
& NFS_CAP_NLINK
)
1358 invalid
|= save_cache_validity
& (NFS_INO_INVALID_ATTR
1359 | NFS_INO_REVAL_FORCED
);
1361 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
1363 * report the blocks in 512byte units
1365 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1367 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
1368 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1370 /* Update attrtimeo value if we're out of the unstable period */
1371 if (invalid
& NFS_INO_INVALID_ATTR
) {
1372 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1373 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1374 nfsi
->attrtimeo_timestamp
= now
;
1375 nfsi
->attr_gencount
= nfs_inc_attr_generation_counter();
1377 if (!time_in_range_open(now
, nfsi
->attrtimeo_timestamp
, nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
1378 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1379 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1380 nfsi
->attrtimeo_timestamp
= now
;
1383 invalid
&= ~NFS_INO_INVALID_ATTR
;
1384 /* Don't invalidate the data if we were to blame */
1385 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1386 || S_ISLNK(inode
->i_mode
)))
1387 invalid
&= ~NFS_INO_INVALID_DATA
;
1388 if (!nfs_have_delegation(inode
, FMODE_READ
) ||
1389 (save_cache_validity
& NFS_INO_REVAL_FORCED
))
1390 nfsi
->cache_validity
|= invalid
;
1395 * Big trouble! The inode has become a different object.
1397 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1398 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1401 * No need to worry about unhashing the dentry, as the
1402 * lookup validation will know that the inode is bad.
1403 * (But we fall through to invalidate the caches.)
1405 nfs_invalidate_inode(inode
);
1409 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1410 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1411 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1412 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1417 #ifdef CONFIG_NFS_V4
1420 * Clean out any remaining NFSv4 state that might be left over due
1421 * to open() calls that passed nfs_atomic_lookup, but failed to call
1424 void nfs4_evict_inode(struct inode
*inode
)
1426 pnfs_destroy_layout(NFS_I(inode
));
1427 truncate_inode_pages(&inode
->i_data
, 0);
1428 end_writeback(inode
);
1429 /* If we are holding a delegation, return it! */
1430 nfs_inode_return_delegation_noreclaim(inode
);
1431 /* First call standard NFS clear_inode() code */
1432 nfs_clear_inode(inode
);
1436 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1438 struct nfs_inode
*nfsi
;
1439 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1443 nfsi
->cache_validity
= 0UL;
1444 #ifdef CONFIG_NFS_V3_ACL
1445 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1446 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1448 #ifdef CONFIG_NFS_V4
1449 nfsi
->nfs4_acl
= NULL
;
1450 #endif /* CONFIG_NFS_V4 */
1451 return &nfsi
->vfs_inode
;
1454 static void nfs_i_callback(struct rcu_head
*head
)
1456 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
1457 INIT_LIST_HEAD(&inode
->i_dentry
);
1458 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1461 void nfs_destroy_inode(struct inode
*inode
)
1463 call_rcu(&inode
->i_rcu
, nfs_i_callback
);
1466 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1468 #ifdef CONFIG_NFS_V4
1469 INIT_LIST_HEAD(&nfsi
->open_states
);
1470 nfsi
->delegation
= NULL
;
1471 nfsi
->delegation_state
= 0;
1472 init_rwsem(&nfsi
->rwsem
);
1473 nfsi
->layout
= NULL
;
1477 static void init_once(void *foo
)
1479 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1481 inode_init_once(&nfsi
->vfs_inode
);
1482 INIT_LIST_HEAD(&nfsi
->open_files
);
1483 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1484 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1485 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1488 atomic_set(&nfsi
->silly_count
, 1);
1489 INIT_HLIST_HEAD(&nfsi
->silly_list
);
1490 init_waitqueue_head(&nfsi
->waitqueue
);
1491 nfs4_init_once(nfsi
);
1494 static int __init
nfs_init_inodecache(void)
1496 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1497 sizeof(struct nfs_inode
),
1498 0, (SLAB_RECLAIM_ACCOUNT
|
1501 if (nfs_inode_cachep
== NULL
)
1507 static void nfs_destroy_inodecache(void)
1509 kmem_cache_destroy(nfs_inode_cachep
);
1512 struct workqueue_struct
*nfsiod_workqueue
;
1515 * start up the nfsiod workqueue
1517 static int nfsiod_start(void)
1519 struct workqueue_struct
*wq
;
1520 dprintk("RPC: creating workqueue nfsiod\n");
1521 wq
= alloc_workqueue("nfsiod", WQ_RESCUER
, 0);
1524 nfsiod_workqueue
= wq
;
1529 * Destroy the nfsiod workqueue
1531 static void nfsiod_stop(void)
1533 struct workqueue_struct
*wq
;
1535 wq
= nfsiod_workqueue
;
1538 nfsiod_workqueue
= NULL
;
1539 destroy_workqueue(wq
);
1545 static int __init
init_nfs_fs(void)
1549 err
= nfs_idmap_init();
1553 err
= nfs_dns_resolver_init();
1557 err
= nfs_fscache_register();
1561 err
= nfsiod_start();
1565 err
= nfs_fs_proc_init();
1569 err
= nfs_init_nfspagecache();
1573 err
= nfs_init_inodecache();
1577 err
= nfs_init_readpagecache();
1581 err
= nfs_init_writepagecache();
1585 err
= nfs_init_directcache();
1589 #ifdef CONFIG_PROC_FS
1590 rpc_proc_register(&nfs_rpcstat
);
1592 if ((err
= register_nfs_fs()) != 0)
1596 #ifdef CONFIG_PROC_FS
1597 rpc_proc_unregister("nfs");
1599 nfs_destroy_directcache();
1601 nfs_destroy_writepagecache();
1603 nfs_destroy_readpagecache();
1605 nfs_destroy_inodecache();
1607 nfs_destroy_nfspagecache();
1613 nfs_fscache_unregister();
1615 nfs_dns_resolver_destroy();
1622 static void __exit
exit_nfs_fs(void)
1624 nfs_destroy_directcache();
1625 nfs_destroy_writepagecache();
1626 nfs_destroy_readpagecache();
1627 nfs_destroy_inodecache();
1628 nfs_destroy_nfspagecache();
1629 nfs_fscache_unregister();
1630 nfs_dns_resolver_destroy();
1632 #ifdef CONFIG_PROC_FS
1633 rpc_proc_unregister("nfs");
1635 nfs_cleanup_cb_ident_idr();
1636 unregister_nfs_fs();
1641 /* Not quite true; I just maintain it */
1642 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1643 MODULE_LICENSE("GPL");
1644 module_param(enable_ino64
, bool, 0644);
1646 module_init(init_nfs_fs
)
1647 module_exit(exit_nfs_fs
)