4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, 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/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/inet.h>
39 #include <linux/nfs_xdr.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
50 #define NFSDBG_FACILITY NFSDBG_VFS
52 static void nfs_invalidate_inode(struct inode
*);
53 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
55 static void nfs_zap_acl_cache(struct inode
*);
57 static struct kmem_cache
* nfs_inode_cachep
;
59 static inline unsigned long
60 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
62 return nfs_fileid_to_ino_t(fattr
->fileid
);
65 int nfs_write_inode(struct inode
*inode
, int sync
)
70 ret
= filemap_fdatawait(inode
->i_mapping
);
72 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
74 ret
= nfs_commit_inode(inode
, 0);
77 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
81 void nfs_clear_inode(struct inode
*inode
)
84 * The following should never happen...
86 BUG_ON(nfs_have_writebacks(inode
));
87 BUG_ON(!list_empty(&NFS_I(inode
)->open_files
));
88 BUG_ON(atomic_read(&NFS_I(inode
)->data_updates
) != 0);
89 nfs_zap_acl_cache(inode
);
90 nfs_access_zap_cache(inode
);
94 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
96 int nfs_sync_mapping(struct address_space
*mapping
)
100 if (mapping
->nrpages
== 0)
102 unmap_mapping_range(mapping
, 0, 0, 0);
103 ret
= filemap_write_and_wait(mapping
);
106 ret
= nfs_wb_all(mapping
->host
);
112 * Invalidate the local caches
114 static void nfs_zap_caches_locked(struct inode
*inode
)
116 struct nfs_inode
*nfsi
= NFS_I(inode
);
117 int mode
= inode
->i_mode
;
119 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
121 NFS_ATTRTIMEO(inode
) = NFS_MINATTRTIMEO(inode
);
122 NFS_ATTRTIMEO_UPDATE(inode
) = jiffies
;
124 memset(NFS_COOKIEVERF(inode
), 0, sizeof(NFS_COOKIEVERF(inode
)));
125 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))
126 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
128 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
131 void nfs_zap_caches(struct inode
*inode
)
133 spin_lock(&inode
->i_lock
);
134 nfs_zap_caches_locked(inode
);
135 spin_unlock(&inode
->i_lock
);
138 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
140 if (mapping
->nrpages
!= 0) {
141 spin_lock(&inode
->i_lock
);
142 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
143 spin_unlock(&inode
->i_lock
);
147 static void nfs_zap_acl_cache(struct inode
*inode
)
149 void (*clear_acl_cache
)(struct inode
*);
151 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
152 if (clear_acl_cache
!= NULL
)
153 clear_acl_cache(inode
);
154 spin_lock(&inode
->i_lock
);
155 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
156 spin_unlock(&inode
->i_lock
);
160 * Invalidate, but do not unhash, the inode.
161 * NB: must be called with inode->i_lock held!
163 static void nfs_invalidate_inode(struct inode
*inode
)
165 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
166 nfs_zap_caches_locked(inode
);
169 struct nfs_find_desc
{
171 struct nfs_fattr
*fattr
;
175 * In NFSv3 we can have 64bit inode numbers. In order to support
176 * this, and re-exported directories (also seen in NFSv2)
177 * we are forced to allow 2 different inodes to have the same
181 nfs_find_actor(struct inode
*inode
, void *opaque
)
183 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
184 struct nfs_fh
*fh
= desc
->fh
;
185 struct nfs_fattr
*fattr
= desc
->fattr
;
187 if (NFS_FILEID(inode
) != fattr
->fileid
)
189 if (nfs_compare_fh(NFS_FH(inode
), fh
))
191 if (is_bad_inode(inode
) || NFS_STALE(inode
))
197 nfs_init_locked(struct inode
*inode
, void *opaque
)
199 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
200 struct nfs_fattr
*fattr
= desc
->fattr
;
202 NFS_FILEID(inode
) = fattr
->fileid
;
203 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
207 /* Don't use READDIRPLUS on directories that we believe are too large */
208 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
211 * This is our front-end to iget that looks up inodes by file handle
212 * instead of inode number.
215 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
217 struct nfs_find_desc desc
= {
221 struct inode
*inode
= ERR_PTR(-ENOENT
);
224 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
228 printk("NFS: Buggy server - nlink == 0!\n");
232 hash
= nfs_fattr_to_ino_t(fattr
);
234 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
236 inode
= ERR_PTR(-ENOMEM
);
240 if (inode
->i_state
& I_NEW
) {
241 struct nfs_inode
*nfsi
= NFS_I(inode
);
242 unsigned long now
= jiffies
;
244 /* We set i_ino for the few things that still rely on it,
248 /* We can't support update_atime(), since the server will reset it */
249 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
250 inode
->i_mode
= fattr
->mode
;
251 /* Why so? Because we want revalidate for devices/FIFOs, and
252 * that's precisely what we have in nfs_file_inode_operations.
254 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
255 if (S_ISREG(inode
->i_mode
)) {
256 inode
->i_fop
= &nfs_file_operations
;
257 inode
->i_data
.a_ops
= &nfs_file_aops
;
258 inode
->i_data
.backing_dev_info
= &NFS_SB(sb
)->backing_dev_info
;
259 } else if (S_ISDIR(inode
->i_mode
)) {
260 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
261 inode
->i_fop
= &nfs_dir_operations
;
262 if (nfs_server_capable(inode
, NFS_CAP_READDIRPLUS
)
263 && fattr
->size
<= NFS_LIMIT_READDIRPLUS
)
264 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
265 /* Deal with crossing mountpoints */
266 if (!nfs_fsid_equal(&NFS_SB(sb
)->fsid
, &fattr
->fsid
)) {
267 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
268 inode
->i_op
= &nfs_referral_inode_operations
;
270 inode
->i_op
= &nfs_mountpoint_inode_operations
;
273 } else if (S_ISLNK(inode
->i_mode
))
274 inode
->i_op
= &nfs_symlink_inode_operations
;
276 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
278 nfsi
->read_cache_jiffies
= fattr
->time_start
;
279 nfsi
->last_updated
= now
;
280 nfsi
->cache_change_attribute
= now
;
281 inode
->i_atime
= fattr
->atime
;
282 inode
->i_mtime
= fattr
->mtime
;
283 inode
->i_ctime
= fattr
->ctime
;
284 if (fattr
->valid
& NFS_ATTR_FATTR_V4
)
285 nfsi
->change_attr
= fattr
->change_attr
;
286 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
287 inode
->i_nlink
= fattr
->nlink
;
288 inode
->i_uid
= fattr
->uid
;
289 inode
->i_gid
= fattr
->gid
;
290 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
292 * report the blocks in 512byte units
294 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
296 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
298 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
299 nfsi
->attrtimeo_timestamp
= now
;
300 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
301 nfsi
->access_cache
= RB_ROOT
;
303 unlock_new_inode(inode
);
305 nfs_refresh_inode(inode
, fattr
);
306 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
308 (long long)NFS_FILEID(inode
),
309 atomic_read(&inode
->i_count
));
315 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
319 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
322 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
324 struct inode
*inode
= dentry
->d_inode
;
325 struct nfs_fattr fattr
;
328 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
330 if (attr
->ia_valid
& ATTR_SIZE
) {
331 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
332 attr
->ia_valid
&= ~ATTR_SIZE
;
335 /* Optimization: if the end result is no change, don't RPC */
336 attr
->ia_valid
&= NFS_VALID_ATTRS
;
337 if (attr
->ia_valid
== 0)
341 nfs_begin_data_update(inode
);
342 /* Write all dirty data */
343 if (S_ISREG(inode
->i_mode
)) {
344 filemap_write_and_wait(inode
->i_mapping
);
348 * Return any delegations if we're going to change ACLs
350 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
351 nfs_inode_return_delegation(inode
);
352 error
= NFS_PROTO(inode
)->setattr(dentry
, &fattr
, attr
);
354 nfs_refresh_inode(inode
, &fattr
);
355 nfs_end_data_update(inode
);
361 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
362 * @inode: pointer to struct inode
363 * @attr: pointer to struct iattr
365 * Note: we do this in the *proc.c in order to ensure that
366 * it works for things like exclusive creates too.
368 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
370 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
371 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
372 int mode
= attr
->ia_mode
& S_IALLUGO
;
373 mode
|= inode
->i_mode
& ~S_IALLUGO
;
374 inode
->i_mode
= mode
;
376 if ((attr
->ia_valid
& ATTR_UID
) != 0)
377 inode
->i_uid
= attr
->ia_uid
;
378 if ((attr
->ia_valid
& ATTR_GID
) != 0)
379 inode
->i_gid
= attr
->ia_gid
;
380 spin_lock(&inode
->i_lock
);
381 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
382 spin_unlock(&inode
->i_lock
);
384 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
385 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
386 inode
->i_size
= attr
->ia_size
;
387 vmtruncate(inode
, attr
->ia_size
);
391 static int nfs_wait_schedule(void *word
)
393 if (signal_pending(current
))
400 * Wait for the inode to get unlocked.
402 static int nfs_wait_on_inode(struct inode
*inode
)
404 struct rpc_clnt
*clnt
= NFS_CLIENT(inode
);
405 struct nfs_inode
*nfsi
= NFS_I(inode
);
409 rpc_clnt_sigmask(clnt
, &oldmask
);
410 error
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_REVALIDATING
,
411 nfs_wait_schedule
, TASK_INTERRUPTIBLE
);
412 rpc_clnt_sigunmask(clnt
, &oldmask
);
417 static void nfs_wake_up_inode(struct inode
*inode
)
419 struct nfs_inode
*nfsi
= NFS_I(inode
);
421 clear_bit(NFS_INO_REVALIDATING
, &nfsi
->flags
);
422 smp_mb__after_clear_bit();
423 wake_up_bit(&nfsi
->flags
, NFS_INO_REVALIDATING
);
426 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
428 struct inode
*inode
= dentry
->d_inode
;
429 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
432 /* Flush out writes to the server in order to update c/mtime */
433 if (S_ISREG(inode
->i_mode
))
434 nfs_sync_mapping_range(inode
->i_mapping
, 0, 0, FLUSH_NOCOMMIT
);
437 * We may force a getattr if the user cares about atime.
439 * Note that we only have to check the vfsmount flags here:
440 * - NFS always sets S_NOATIME by so checking it would give a
442 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
443 * no point in checking those.
445 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
446 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
450 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
452 err
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
454 generic_fillattr(inode
, stat
);
458 static struct nfs_open_context
*alloc_nfs_open_context(struct vfsmount
*mnt
, struct dentry
*dentry
, struct rpc_cred
*cred
)
460 struct nfs_open_context
*ctx
;
462 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
464 ctx
->path
.dentry
= dget(dentry
);
465 ctx
->path
.mnt
= mntget(mnt
);
466 ctx
->cred
= get_rpccred(cred
);
468 ctx
->lockowner
= current
->files
;
471 kref_init(&ctx
->kref
);
476 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
479 kref_get(&ctx
->kref
);
483 static void nfs_free_open_context(struct kref
*kref
)
485 struct nfs_open_context
*ctx
= container_of(kref
,
486 struct nfs_open_context
, kref
);
488 if (!list_empty(&ctx
->list
)) {
489 struct inode
*inode
= ctx
->path
.dentry
->d_inode
;
490 spin_lock(&inode
->i_lock
);
491 list_del(&ctx
->list
);
492 spin_unlock(&inode
->i_lock
);
494 if (ctx
->state
!= NULL
)
495 nfs4_close_state(&ctx
->path
, ctx
->state
, ctx
->mode
);
496 if (ctx
->cred
!= NULL
)
497 put_rpccred(ctx
->cred
);
498 dput(ctx
->path
.dentry
);
499 mntput(ctx
->path
.mnt
);
503 void put_nfs_open_context(struct nfs_open_context
*ctx
)
505 kref_put(&ctx
->kref
, nfs_free_open_context
);
509 * Ensure that mmap has a recent RPC credential for use when writing out
512 static void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
514 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
515 struct nfs_inode
*nfsi
= NFS_I(inode
);
517 filp
->private_data
= get_nfs_open_context(ctx
);
518 spin_lock(&inode
->i_lock
);
519 list_add(&ctx
->list
, &nfsi
->open_files
);
520 spin_unlock(&inode
->i_lock
);
524 * Given an inode, search for an open context with the desired characteristics
526 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, int mode
)
528 struct nfs_inode
*nfsi
= NFS_I(inode
);
529 struct nfs_open_context
*pos
, *ctx
= NULL
;
531 spin_lock(&inode
->i_lock
);
532 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
533 if (cred
!= NULL
&& pos
->cred
!= cred
)
535 if ((pos
->mode
& mode
) == mode
) {
536 ctx
= get_nfs_open_context(pos
);
540 spin_unlock(&inode
->i_lock
);
544 static void nfs_file_clear_open_context(struct file
*filp
)
546 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
547 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)filp
->private_data
;
550 filp
->private_data
= NULL
;
551 spin_lock(&inode
->i_lock
);
552 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
553 spin_unlock(&inode
->i_lock
);
554 put_nfs_open_context(ctx
);
559 * These allocate and release file read/write context information.
561 int nfs_open(struct inode
*inode
, struct file
*filp
)
563 struct nfs_open_context
*ctx
;
564 struct rpc_cred
*cred
;
566 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
568 return PTR_ERR(cred
);
569 ctx
= alloc_nfs_open_context(filp
->f_path
.mnt
, filp
->f_path
.dentry
, cred
);
573 ctx
->mode
= filp
->f_mode
;
574 nfs_file_set_open_context(filp
, ctx
);
575 put_nfs_open_context(ctx
);
579 int nfs_release(struct inode
*inode
, struct file
*filp
)
581 nfs_file_clear_open_context(filp
);
586 * This function is called whenever some part of NFS notices that
587 * the cached attributes have to be refreshed.
590 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
592 int status
= -ESTALE
;
593 struct nfs_fattr fattr
;
594 struct nfs_inode
*nfsi
= NFS_I(inode
);
596 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
597 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
599 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
601 if (is_bad_inode(inode
))
603 if (NFS_STALE(inode
))
606 status
= nfs_wait_on_inode(inode
);
609 if (NFS_STALE(inode
)) {
611 /* Do we trust the cached ESTALE? */
612 if (NFS_ATTRTIMEO(inode
) != 0) {
613 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
)) {
620 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), &fattr
);
622 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
624 (long long)NFS_FILEID(inode
), status
);
625 if (status
== -ESTALE
) {
626 nfs_zap_caches(inode
);
627 if (!S_ISDIR(inode
->i_mode
))
628 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
633 spin_lock(&inode
->i_lock
);
634 status
= nfs_update_inode(inode
, &fattr
);
636 spin_unlock(&inode
->i_lock
);
637 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
639 (long long)NFS_FILEID(inode
), status
);
642 spin_unlock(&inode
->i_lock
);
644 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
645 nfs_zap_acl_cache(inode
);
647 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
649 (long long)NFS_FILEID(inode
));
652 nfs_wake_up_inode(inode
);
659 int nfs_attribute_timeout(struct inode
*inode
)
661 struct nfs_inode
*nfsi
= NFS_I(inode
);
663 if (nfs_have_delegation(inode
, FMODE_READ
))
665 return time_after(jiffies
, nfsi
->read_cache_jiffies
+nfsi
->attrtimeo
);
669 * nfs_revalidate_inode - Revalidate the inode attributes
670 * @server - pointer to nfs_server struct
671 * @inode - pointer to inode struct
673 * Updates inode attribute information by retrieving the data from the server.
675 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
677 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
678 && !nfs_attribute_timeout(inode
))
679 return NFS_STALE(inode
) ? -ESTALE
: 0;
680 return __nfs_revalidate_inode(server
, inode
);
683 static int nfs_invalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
685 struct nfs_inode
*nfsi
= NFS_I(inode
);
687 if (mapping
->nrpages
!= 0) {
688 int ret
= invalidate_inode_pages2(mapping
);
692 spin_lock(&inode
->i_lock
);
693 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
694 if (S_ISDIR(inode
->i_mode
)) {
695 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
696 /* This ensures we revalidate child dentries */
697 nfsi
->cache_change_attribute
= jiffies
;
699 spin_unlock(&inode
->i_lock
);
700 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
701 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
702 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
706 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
710 mutex_lock(&inode
->i_mutex
);
711 if (NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_DATA
) {
712 ret
= nfs_sync_mapping(mapping
);
714 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
716 mutex_unlock(&inode
->i_mutex
);
721 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
722 * @inode - pointer to host inode
723 * @mapping - pointer to mapping
725 int nfs_revalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
727 struct nfs_inode
*nfsi
= NFS_I(inode
);
730 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
731 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
732 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
736 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
737 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
743 * nfs_revalidate_mapping - Revalidate the pagecache
744 * @inode - pointer to host inode
745 * @mapping - pointer to mapping
747 * This version of the function will take the inode->i_mutex and attempt to
748 * flush out all dirty data if it needs to invalidate the page cache.
750 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
752 struct nfs_inode
*nfsi
= NFS_I(inode
);
755 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
756 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
757 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
761 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
762 ret
= nfs_invalidate_mapping(inode
, mapping
);
768 * nfs_begin_data_update
769 * @inode - pointer to inode
770 * Declare that a set of operations will update file data on the server
772 void nfs_begin_data_update(struct inode
*inode
)
774 atomic_inc(&NFS_I(inode
)->data_updates
);
778 * nfs_end_data_update
779 * @inode - pointer to inode
780 * Declare end of the operations that will update file data
781 * This will mark the inode as immediately needing revalidation
782 * of its attribute cache.
784 void nfs_end_data_update(struct inode
*inode
)
786 struct nfs_inode
*nfsi
= NFS_I(inode
);
788 /* Directories: invalidate page cache */
789 if (S_ISDIR(inode
->i_mode
)) {
790 spin_lock(&inode
->i_lock
);
791 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
792 spin_unlock(&inode
->i_lock
);
794 nfsi
->cache_change_attribute
= jiffies
;
795 atomic_dec(&nfsi
->data_updates
);
798 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
800 struct nfs_inode
*nfsi
= NFS_I(inode
);
801 unsigned long now
= jiffies
;
803 /* If we have atomic WCC data, we may update some attributes */
804 if ((fattr
->valid
& NFS_ATTR_WCC
) != 0) {
805 if (timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
)) {
806 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
807 nfsi
->cache_change_attribute
= now
;
809 if (timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
810 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
811 nfsi
->cache_change_attribute
= now
;
813 if (inode
->i_size
== fattr
->pre_size
&& nfsi
->npages
== 0) {
814 inode
->i_size
= fattr
->size
;
815 nfsi
->cache_change_attribute
= now
;
821 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
822 * @inode - pointer to inode
823 * @fattr - updated attributes
825 * Verifies the attribute cache. If we have just changed the attributes,
826 * so that fattr carries weak cache consistency data, then it may
827 * also update the ctime/mtime/change_attribute.
829 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
831 struct nfs_inode
*nfsi
= NFS_I(inode
);
832 loff_t cur_size
, new_isize
;
836 /* Has the inode gone and changed behind our back? */
837 if (nfsi
->fileid
!= fattr
->fileid
838 || (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
842 /* Are we in the process of updating data on the server? */
843 data_unstable
= nfs_caches_unstable(inode
);
845 /* Do atomic weak cache consistency updates */
846 nfs_wcc_update_inode(inode
, fattr
);
848 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
849 nfsi
->change_attr
!= fattr
->change_attr
)
850 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
852 /* Verify a few of the more important attributes */
853 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
854 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
856 cur_size
= i_size_read(inode
);
857 new_isize
= nfs_size_to_loff_t(fattr
->size
);
858 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
859 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
861 /* Have any file permissions changed? */
862 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)
863 || inode
->i_uid
!= fattr
->uid
864 || inode
->i_gid
!= fattr
->gid
)
865 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
867 /* Has the link count changed? */
868 if (inode
->i_nlink
!= fattr
->nlink
)
869 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
871 if (!timespec_equal(&inode
->i_atime
, &fattr
->atime
))
872 nfsi
->cache_validity
|= NFS_INO_INVALID_ATIME
;
874 nfsi
->read_cache_jiffies
= fattr
->time_start
;
879 * nfs_refresh_inode - try to update the inode attribute cache
880 * @inode - pointer to inode
881 * @fattr - updated attributes
883 * Check that an RPC call that returned attributes has not overlapped with
884 * other recent updates of the inode metadata, then decide whether it is
885 * safe to do a full update of the inode attributes, or whether just to
886 * call nfs_check_inode_attributes.
888 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
890 struct nfs_inode
*nfsi
= NFS_I(inode
);
893 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
895 spin_lock(&inode
->i_lock
);
896 if (time_after(fattr
->time_start
, nfsi
->last_updated
))
897 status
= nfs_update_inode(inode
, fattr
);
899 status
= nfs_check_inode_attributes(inode
, fattr
);
901 spin_unlock(&inode
->i_lock
);
906 * nfs_post_op_update_inode - try to update the inode attribute cache
907 * @inode - pointer to inode
908 * @fattr - updated attributes
910 * After an operation that has changed the inode metadata, mark the
911 * attribute cache as being invalid, then try to update it.
913 * NB: if the server didn't return any post op attributes, this
914 * function will force the retrieval of attributes before the next
915 * NFS request. Thus it should be used only for operations that
916 * are expected to change one or more attributes, to avoid
917 * unnecessary NFS requests and trips through nfs_update_inode().
919 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
921 struct nfs_inode
*nfsi
= NFS_I(inode
);
924 spin_lock(&inode
->i_lock
);
925 if (unlikely((fattr
->valid
& NFS_ATTR_FATTR
) == 0)) {
926 nfsi
->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
929 status
= nfs_update_inode(inode
, fattr
);
931 spin_unlock(&inode
->i_lock
);
936 * Many nfs protocol calls return the new file attributes after
937 * an operation. Here we update the inode to reflect the state
938 * of the server's inode.
940 * This is a bit tricky because we have to make sure all dirty pages
941 * have been sent off to the server before calling invalidate_inode_pages.
942 * To make sure no other process adds more write requests while we try
943 * our best to flush them, we make them sleep during the attribute refresh.
945 * A very similar scenario holds for the dir cache.
947 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
949 struct nfs_server
*server
;
950 struct nfs_inode
*nfsi
= NFS_I(inode
);
951 loff_t cur_isize
, new_isize
;
952 unsigned int invalid
= 0;
953 unsigned long now
= jiffies
;
956 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
957 __FUNCTION__
, inode
->i_sb
->s_id
, inode
->i_ino
,
958 atomic_read(&inode
->i_count
), fattr
->valid
);
960 if (nfsi
->fileid
!= fattr
->fileid
)
964 * Make sure the inode's type hasn't changed.
966 if ((inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
969 server
= NFS_SERVER(inode
);
970 /* Update the fsid? */
971 if (S_ISDIR(inode
->i_mode
)
972 && !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
973 server
->fsid
= fattr
->fsid
;
976 * Update the read time so we don't revalidate too often.
978 nfsi
->read_cache_jiffies
= fattr
->time_start
;
979 nfsi
->last_updated
= now
;
981 /* Fix a wraparound issue with nfsi->cache_change_attribute */
982 if (time_before(now
, nfsi
->cache_change_attribute
))
983 nfsi
->cache_change_attribute
= now
- 600*HZ
;
985 /* Are we racing with known updates of the metadata on the server? */
986 data_stable
= nfs_verify_change_attribute(inode
, fattr
->time_start
);
988 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
|NFS_INO_INVALID_ATIME
);
990 /* Do atomic weak cache consistency updates */
991 nfs_wcc_update_inode(inode
, fattr
);
993 /* Check if our cached file size is stale */
994 new_isize
= nfs_size_to_loff_t(fattr
->size
);
995 cur_isize
= i_size_read(inode
);
996 if (new_isize
!= cur_isize
) {
997 /* Do we perhaps have any outstanding writes? */
998 if (nfsi
->npages
== 0) {
999 /* No, but did we race with nfs_end_data_update()? */
1001 inode
->i_size
= new_isize
;
1002 invalid
|= NFS_INO_INVALID_DATA
;
1004 invalid
|= NFS_INO_INVALID_ATTR
;
1005 } else if (new_isize
> cur_isize
) {
1006 inode
->i_size
= new_isize
;
1007 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1009 nfsi
->cache_change_attribute
= now
;
1010 dprintk("NFS: isize change on server for file %s/%ld\n",
1011 inode
->i_sb
->s_id
, inode
->i_ino
);
1014 /* Check if the mtime agrees */
1015 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
1016 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1017 dprintk("NFS: mtime change on server for file %s/%ld\n",
1018 inode
->i_sb
->s_id
, inode
->i_ino
);
1019 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1020 nfsi
->cache_change_attribute
= now
;
1023 /* If ctime has changed we should definitely clear access+acl caches */
1024 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
)) {
1025 invalid
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1026 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1027 nfsi
->cache_change_attribute
= now
;
1029 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1031 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
) ||
1032 inode
->i_uid
!= fattr
->uid
||
1033 inode
->i_gid
!= fattr
->gid
)
1034 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1036 inode
->i_mode
= fattr
->mode
;
1037 inode
->i_nlink
= fattr
->nlink
;
1038 inode
->i_uid
= fattr
->uid
;
1039 inode
->i_gid
= fattr
->gid
;
1041 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
1043 * report the blocks in 512byte units
1045 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1047 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1050 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
1051 nfsi
->change_attr
!= fattr
->change_attr
) {
1052 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1053 inode
->i_sb
->s_id
, inode
->i_ino
);
1054 nfsi
->change_attr
= fattr
->change_attr
;
1055 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1056 nfsi
->cache_change_attribute
= now
;
1059 /* Update attrtimeo value if we're out of the unstable period */
1060 if (invalid
& NFS_INO_INVALID_ATTR
) {
1061 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1062 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1063 nfsi
->attrtimeo_timestamp
= now
;
1064 } else if (time_after(now
, nfsi
->attrtimeo_timestamp
+nfsi
->attrtimeo
)) {
1065 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1066 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1067 nfsi
->attrtimeo_timestamp
= now
;
1069 /* Don't invalidate the data if we were to blame */
1070 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1071 || S_ISLNK(inode
->i_mode
)))
1072 invalid
&= ~NFS_INO_INVALID_DATA
;
1074 invalid
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
|NFS_INO_REVAL_PAGECACHE
);
1075 if (!nfs_have_delegation(inode
, FMODE_READ
) ||
1076 (nfsi
->cache_validity
& NFS_INO_REVAL_FORCED
))
1077 nfsi
->cache_validity
|= invalid
;
1078 nfsi
->cache_validity
&= ~NFS_INO_REVAL_FORCED
;
1083 * Big trouble! The inode has become a different object.
1085 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1086 __FUNCTION__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1089 * No need to worry about unhashing the dentry, as the
1090 * lookup validation will know that the inode is bad.
1091 * (But we fall through to invalidate the caches.)
1093 nfs_invalidate_inode(inode
);
1097 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1098 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1099 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1100 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1105 #ifdef CONFIG_NFS_V4
1108 * Clean out any remaining NFSv4 state that might be left over due
1109 * to open() calls that passed nfs_atomic_lookup, but failed to call
1112 void nfs4_clear_inode(struct inode
*inode
)
1114 /* If we are holding a delegation, return it! */
1115 nfs_inode_return_delegation(inode
);
1116 /* First call standard NFS clear_inode() code */
1117 nfs_clear_inode(inode
);
1121 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1123 struct nfs_inode
*nfsi
;
1124 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1128 nfsi
->cache_validity
= 0UL;
1129 #ifdef CONFIG_NFS_V3_ACL
1130 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1131 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1133 #ifdef CONFIG_NFS_V4
1134 nfsi
->nfs4_acl
= NULL
;
1135 #endif /* CONFIG_NFS_V4 */
1136 return &nfsi
->vfs_inode
;
1139 void nfs_destroy_inode(struct inode
*inode
)
1141 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1144 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1146 #ifdef CONFIG_NFS_V4
1147 INIT_LIST_HEAD(&nfsi
->open_states
);
1148 nfsi
->delegation
= NULL
;
1149 nfsi
->delegation_state
= 0;
1150 init_rwsem(&nfsi
->rwsem
);
1154 static void init_once(void * foo
, struct kmem_cache
* cachep
, unsigned long flags
)
1156 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1158 inode_init_once(&nfsi
->vfs_inode
);
1159 INIT_LIST_HEAD(&nfsi
->open_files
);
1160 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1161 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1162 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1163 atomic_set(&nfsi
->data_updates
, 0);
1166 nfs4_init_once(nfsi
);
1169 static int __init
nfs_init_inodecache(void)
1171 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1172 sizeof(struct nfs_inode
),
1173 0, (SLAB_RECLAIM_ACCOUNT
|
1176 if (nfs_inode_cachep
== NULL
)
1182 static void nfs_destroy_inodecache(void)
1184 kmem_cache_destroy(nfs_inode_cachep
);
1190 static int __init
init_nfs_fs(void)
1194 err
= nfs_fs_proc_init();
1198 err
= nfs_init_nfspagecache();
1202 err
= nfs_init_inodecache();
1206 err
= nfs_init_readpagecache();
1210 err
= nfs_init_writepagecache();
1214 err
= nfs_init_directcache();
1218 #ifdef CONFIG_PROC_FS
1219 rpc_proc_register(&nfs_rpcstat
);
1221 if ((err
= register_nfs_fs()) != 0)
1225 #ifdef CONFIG_PROC_FS
1226 rpc_proc_unregister("nfs");
1228 nfs_destroy_directcache();
1230 nfs_destroy_writepagecache();
1232 nfs_destroy_readpagecache();
1234 nfs_destroy_inodecache();
1236 nfs_destroy_nfspagecache();
1243 static void __exit
exit_nfs_fs(void)
1245 nfs_destroy_directcache();
1246 nfs_destroy_writepagecache();
1247 nfs_destroy_readpagecache();
1248 nfs_destroy_inodecache();
1249 nfs_destroy_nfspagecache();
1250 #ifdef CONFIG_PROC_FS
1251 rpc_proc_unregister("nfs");
1253 unregister_nfs_fs();
1257 /* Not quite true; I just maintain it */
1258 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1259 MODULE_LICENSE("GPL");
1261 module_init(init_nfs_fs
)
1262 module_exit(exit_nfs_fs
)