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 atomic_set(&ctx
->count
, 1);
465 ctx
->dentry
= dget(dentry
);
466 ctx
->vfsmnt
= mntget(mnt
);
467 ctx
->cred
= get_rpccred(cred
);
469 ctx
->lockowner
= current
->files
;
476 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
479 atomic_inc(&ctx
->count
);
483 void put_nfs_open_context(struct nfs_open_context
*ctx
)
485 if (atomic_dec_and_test(&ctx
->count
)) {
486 if (!list_empty(&ctx
->list
)) {
487 struct inode
*inode
= ctx
->dentry
->d_inode
;
488 spin_lock(&inode
->i_lock
);
489 list_del(&ctx
->list
);
490 spin_unlock(&inode
->i_lock
);
492 if (ctx
->state
!= NULL
)
493 nfs4_close_state(ctx
->state
, ctx
->mode
);
494 if (ctx
->cred
!= NULL
)
495 put_rpccred(ctx
->cred
);
503 * Ensure that mmap has a recent RPC credential for use when writing out
506 static void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
508 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
509 struct nfs_inode
*nfsi
= NFS_I(inode
);
511 filp
->private_data
= get_nfs_open_context(ctx
);
512 spin_lock(&inode
->i_lock
);
513 list_add(&ctx
->list
, &nfsi
->open_files
);
514 spin_unlock(&inode
->i_lock
);
518 * Given an inode, search for an open context with the desired characteristics
520 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, int mode
)
522 struct nfs_inode
*nfsi
= NFS_I(inode
);
523 struct nfs_open_context
*pos
, *ctx
= NULL
;
525 spin_lock(&inode
->i_lock
);
526 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
527 if (cred
!= NULL
&& pos
->cred
!= cred
)
529 if ((pos
->mode
& mode
) == mode
) {
530 ctx
= get_nfs_open_context(pos
);
534 spin_unlock(&inode
->i_lock
);
538 static void nfs_file_clear_open_context(struct file
*filp
)
540 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
541 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)filp
->private_data
;
544 filp
->private_data
= NULL
;
545 spin_lock(&inode
->i_lock
);
546 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
547 spin_unlock(&inode
->i_lock
);
548 put_nfs_open_context(ctx
);
553 * These allocate and release file read/write context information.
555 int nfs_open(struct inode
*inode
, struct file
*filp
)
557 struct nfs_open_context
*ctx
;
558 struct rpc_cred
*cred
;
560 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
562 return PTR_ERR(cred
);
563 ctx
= alloc_nfs_open_context(filp
->f_path
.mnt
, filp
->f_path
.dentry
, cred
);
567 ctx
->mode
= filp
->f_mode
;
568 nfs_file_set_open_context(filp
, ctx
);
569 put_nfs_open_context(ctx
);
573 int nfs_release(struct inode
*inode
, struct file
*filp
)
575 nfs_file_clear_open_context(filp
);
580 * This function is called whenever some part of NFS notices that
581 * the cached attributes have to be refreshed.
584 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
586 int status
= -ESTALE
;
587 struct nfs_fattr fattr
;
588 struct nfs_inode
*nfsi
= NFS_I(inode
);
590 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
591 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
593 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
595 if (is_bad_inode(inode
))
597 if (NFS_STALE(inode
))
600 status
= nfs_wait_on_inode(inode
);
603 if (NFS_STALE(inode
)) {
605 /* Do we trust the cached ESTALE? */
606 if (NFS_ATTRTIMEO(inode
) != 0) {
607 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
)) {
614 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), &fattr
);
616 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
618 (long long)NFS_FILEID(inode
), status
);
619 if (status
== -ESTALE
) {
620 nfs_zap_caches(inode
);
621 if (!S_ISDIR(inode
->i_mode
))
622 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
627 spin_lock(&inode
->i_lock
);
628 status
= nfs_update_inode(inode
, &fattr
);
630 spin_unlock(&inode
->i_lock
);
631 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
633 (long long)NFS_FILEID(inode
), status
);
636 spin_unlock(&inode
->i_lock
);
638 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
639 nfs_zap_acl_cache(inode
);
641 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
643 (long long)NFS_FILEID(inode
));
646 nfs_wake_up_inode(inode
);
653 int nfs_attribute_timeout(struct inode
*inode
)
655 struct nfs_inode
*nfsi
= NFS_I(inode
);
657 if (nfs_have_delegation(inode
, FMODE_READ
))
659 return time_after(jiffies
, nfsi
->read_cache_jiffies
+nfsi
->attrtimeo
);
663 * nfs_revalidate_inode - Revalidate the inode attributes
664 * @server - pointer to nfs_server struct
665 * @inode - pointer to inode struct
667 * Updates inode attribute information by retrieving the data from the server.
669 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
671 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
672 && !nfs_attribute_timeout(inode
))
673 return NFS_STALE(inode
) ? -ESTALE
: 0;
674 return __nfs_revalidate_inode(server
, inode
);
677 static int nfs_invalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
679 struct nfs_inode
*nfsi
= NFS_I(inode
);
681 if (mapping
->nrpages
!= 0) {
682 int ret
= invalidate_inode_pages2(mapping
);
686 spin_lock(&inode
->i_lock
);
687 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
688 if (S_ISDIR(inode
->i_mode
)) {
689 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
690 /* This ensures we revalidate child dentries */
691 nfsi
->cache_change_attribute
= jiffies
;
693 spin_unlock(&inode
->i_lock
);
694 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
695 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
696 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
700 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
704 mutex_lock(&inode
->i_mutex
);
705 if (NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_DATA
) {
706 ret
= nfs_sync_mapping(mapping
);
708 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
710 mutex_unlock(&inode
->i_mutex
);
715 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
716 * @inode - pointer to host inode
717 * @mapping - pointer to mapping
719 int nfs_revalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
721 struct nfs_inode
*nfsi
= NFS_I(inode
);
724 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
725 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
726 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
730 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
731 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
737 * nfs_revalidate_mapping - Revalidate the pagecache
738 * @inode - pointer to host inode
739 * @mapping - pointer to mapping
741 * This version of the function will take the inode->i_mutex and attempt to
742 * flush out all dirty data if it needs to invalidate the page cache.
744 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
746 struct nfs_inode
*nfsi
= NFS_I(inode
);
749 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
750 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
751 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
755 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
756 ret
= nfs_invalidate_mapping(inode
, mapping
);
762 * nfs_begin_data_update
763 * @inode - pointer to inode
764 * Declare that a set of operations will update file data on the server
766 void nfs_begin_data_update(struct inode
*inode
)
768 atomic_inc(&NFS_I(inode
)->data_updates
);
772 * nfs_end_data_update
773 * @inode - pointer to inode
774 * Declare end of the operations that will update file data
775 * This will mark the inode as immediately needing revalidation
776 * of its attribute cache.
778 void nfs_end_data_update(struct inode
*inode
)
780 struct nfs_inode
*nfsi
= NFS_I(inode
);
782 /* Directories: invalidate page cache */
783 if (S_ISDIR(inode
->i_mode
)) {
784 spin_lock(&inode
->i_lock
);
785 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
786 spin_unlock(&inode
->i_lock
);
788 nfsi
->cache_change_attribute
= jiffies
;
789 atomic_dec(&nfsi
->data_updates
);
792 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
794 struct nfs_inode
*nfsi
= NFS_I(inode
);
795 unsigned long now
= jiffies
;
797 /* If we have atomic WCC data, we may update some attributes */
798 if ((fattr
->valid
& NFS_ATTR_WCC
) != 0) {
799 if (timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
)) {
800 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
801 nfsi
->cache_change_attribute
= now
;
803 if (timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
804 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
805 nfsi
->cache_change_attribute
= now
;
807 if (inode
->i_size
== fattr
->pre_size
&& nfsi
->npages
== 0) {
808 inode
->i_size
= fattr
->size
;
809 nfsi
->cache_change_attribute
= now
;
815 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
816 * @inode - pointer to inode
817 * @fattr - updated attributes
819 * Verifies the attribute cache. If we have just changed the attributes,
820 * so that fattr carries weak cache consistency data, then it may
821 * also update the ctime/mtime/change_attribute.
823 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
825 struct nfs_inode
*nfsi
= NFS_I(inode
);
826 loff_t cur_size
, new_isize
;
830 /* Has the inode gone and changed behind our back? */
831 if (nfsi
->fileid
!= fattr
->fileid
832 || (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
836 /* Are we in the process of updating data on the server? */
837 data_unstable
= nfs_caches_unstable(inode
);
839 /* Do atomic weak cache consistency updates */
840 nfs_wcc_update_inode(inode
, fattr
);
842 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
843 nfsi
->change_attr
!= fattr
->change_attr
)
844 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
846 /* Verify a few of the more important attributes */
847 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
848 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
850 cur_size
= i_size_read(inode
);
851 new_isize
= nfs_size_to_loff_t(fattr
->size
);
852 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
853 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
855 /* Have any file permissions changed? */
856 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)
857 || inode
->i_uid
!= fattr
->uid
858 || inode
->i_gid
!= fattr
->gid
)
859 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
861 /* Has the link count changed? */
862 if (inode
->i_nlink
!= fattr
->nlink
)
863 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
865 if (!timespec_equal(&inode
->i_atime
, &fattr
->atime
))
866 nfsi
->cache_validity
|= NFS_INO_INVALID_ATIME
;
868 nfsi
->read_cache_jiffies
= fattr
->time_start
;
873 * nfs_refresh_inode - try to update the inode attribute cache
874 * @inode - pointer to inode
875 * @fattr - updated attributes
877 * Check that an RPC call that returned attributes has not overlapped with
878 * other recent updates of the inode metadata, then decide whether it is
879 * safe to do a full update of the inode attributes, or whether just to
880 * call nfs_check_inode_attributes.
882 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
884 struct nfs_inode
*nfsi
= NFS_I(inode
);
887 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
889 spin_lock(&inode
->i_lock
);
890 if (time_after(fattr
->time_start
, nfsi
->last_updated
))
891 status
= nfs_update_inode(inode
, fattr
);
893 status
= nfs_check_inode_attributes(inode
, fattr
);
895 spin_unlock(&inode
->i_lock
);
900 * nfs_post_op_update_inode - try to update the inode attribute cache
901 * @inode - pointer to inode
902 * @fattr - updated attributes
904 * After an operation that has changed the inode metadata, mark the
905 * attribute cache as being invalid, then try to update it.
907 * NB: if the server didn't return any post op attributes, this
908 * function will force the retrieval of attributes before the next
909 * NFS request. Thus it should be used only for operations that
910 * are expected to change one or more attributes, to avoid
911 * unnecessary NFS requests and trips through nfs_update_inode().
913 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
915 struct nfs_inode
*nfsi
= NFS_I(inode
);
918 spin_lock(&inode
->i_lock
);
919 if (unlikely((fattr
->valid
& NFS_ATTR_FATTR
) == 0)) {
920 nfsi
->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
923 status
= nfs_update_inode(inode
, fattr
);
925 spin_unlock(&inode
->i_lock
);
930 * Many nfs protocol calls return the new file attributes after
931 * an operation. Here we update the inode to reflect the state
932 * of the server's inode.
934 * This is a bit tricky because we have to make sure all dirty pages
935 * have been sent off to the server before calling invalidate_inode_pages.
936 * To make sure no other process adds more write requests while we try
937 * our best to flush them, we make them sleep during the attribute refresh.
939 * A very similar scenario holds for the dir cache.
941 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
943 struct nfs_server
*server
;
944 struct nfs_inode
*nfsi
= NFS_I(inode
);
945 loff_t cur_isize
, new_isize
;
946 unsigned int invalid
= 0;
947 unsigned long now
= jiffies
;
950 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
951 __FUNCTION__
, inode
->i_sb
->s_id
, inode
->i_ino
,
952 atomic_read(&inode
->i_count
), fattr
->valid
);
954 if (nfsi
->fileid
!= fattr
->fileid
)
958 * Make sure the inode's type hasn't changed.
960 if ((inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
963 server
= NFS_SERVER(inode
);
964 /* Update the fsid? */
965 if (S_ISDIR(inode
->i_mode
)
966 && !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
967 server
->fsid
= fattr
->fsid
;
970 * Update the read time so we don't revalidate too often.
972 nfsi
->read_cache_jiffies
= fattr
->time_start
;
973 nfsi
->last_updated
= now
;
975 /* Fix a wraparound issue with nfsi->cache_change_attribute */
976 if (time_before(now
, nfsi
->cache_change_attribute
))
977 nfsi
->cache_change_attribute
= now
- 600*HZ
;
979 /* Are we racing with known updates of the metadata on the server? */
980 data_stable
= nfs_verify_change_attribute(inode
, fattr
->time_start
);
982 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
|NFS_INO_INVALID_ATIME
);
984 /* Do atomic weak cache consistency updates */
985 nfs_wcc_update_inode(inode
, fattr
);
987 /* Check if our cached file size is stale */
988 new_isize
= nfs_size_to_loff_t(fattr
->size
);
989 cur_isize
= i_size_read(inode
);
990 if (new_isize
!= cur_isize
) {
991 /* Do we perhaps have any outstanding writes? */
992 if (nfsi
->npages
== 0) {
993 /* No, but did we race with nfs_end_data_update()? */
995 inode
->i_size
= new_isize
;
996 invalid
|= NFS_INO_INVALID_DATA
;
998 invalid
|= NFS_INO_INVALID_ATTR
;
999 } else if (new_isize
> cur_isize
) {
1000 inode
->i_size
= new_isize
;
1001 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1003 nfsi
->cache_change_attribute
= now
;
1004 dprintk("NFS: isize change on server for file %s/%ld\n",
1005 inode
->i_sb
->s_id
, inode
->i_ino
);
1008 /* Check if the mtime agrees */
1009 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
1010 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1011 dprintk("NFS: mtime change on server for file %s/%ld\n",
1012 inode
->i_sb
->s_id
, inode
->i_ino
);
1013 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1014 nfsi
->cache_change_attribute
= now
;
1017 /* If ctime has changed we should definitely clear access+acl caches */
1018 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
)) {
1019 invalid
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1020 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1021 nfsi
->cache_change_attribute
= now
;
1023 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1025 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
) ||
1026 inode
->i_uid
!= fattr
->uid
||
1027 inode
->i_gid
!= fattr
->gid
)
1028 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1030 inode
->i_mode
= fattr
->mode
;
1031 inode
->i_nlink
= fattr
->nlink
;
1032 inode
->i_uid
= fattr
->uid
;
1033 inode
->i_gid
= fattr
->gid
;
1035 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
1037 * report the blocks in 512byte units
1039 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1041 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1044 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
1045 nfsi
->change_attr
!= fattr
->change_attr
) {
1046 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1047 inode
->i_sb
->s_id
, inode
->i_ino
);
1048 nfsi
->change_attr
= fattr
->change_attr
;
1049 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1050 nfsi
->cache_change_attribute
= now
;
1053 /* Update attrtimeo value if we're out of the unstable period */
1054 if (invalid
& NFS_INO_INVALID_ATTR
) {
1055 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1056 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1057 nfsi
->attrtimeo_timestamp
= now
;
1058 } else if (time_after(now
, nfsi
->attrtimeo_timestamp
+nfsi
->attrtimeo
)) {
1059 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1060 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1061 nfsi
->attrtimeo_timestamp
= now
;
1063 /* Don't invalidate the data if we were to blame */
1064 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1065 || S_ISLNK(inode
->i_mode
)))
1066 invalid
&= ~NFS_INO_INVALID_DATA
;
1068 invalid
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
|NFS_INO_REVAL_PAGECACHE
);
1069 if (!nfs_have_delegation(inode
, FMODE_READ
))
1070 nfsi
->cache_validity
|= invalid
;
1075 * Big trouble! The inode has become a different object.
1077 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1078 __FUNCTION__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1081 * No need to worry about unhashing the dentry, as the
1082 * lookup validation will know that the inode is bad.
1083 * (But we fall through to invalidate the caches.)
1085 nfs_invalidate_inode(inode
);
1089 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1090 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1091 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1092 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1097 #ifdef CONFIG_NFS_V4
1100 * Clean out any remaining NFSv4 state that might be left over due
1101 * to open() calls that passed nfs_atomic_lookup, but failed to call
1104 void nfs4_clear_inode(struct inode
*inode
)
1106 struct nfs_inode
*nfsi
= NFS_I(inode
);
1108 /* If we are holding a delegation, return it! */
1109 nfs_inode_return_delegation(inode
);
1110 /* First call standard NFS clear_inode() code */
1111 nfs_clear_inode(inode
);
1112 /* Now clear out any remaining state */
1113 while (!list_empty(&nfsi
->open_states
)) {
1114 struct nfs4_state
*state
;
1116 state
= list_entry(nfsi
->open_states
.next
,
1119 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1122 (long long)NFS_FILEID(inode
),
1124 BUG_ON(atomic_read(&state
->count
) != 1);
1125 nfs4_close_state(state
, state
->state
);
1130 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1132 struct nfs_inode
*nfsi
;
1133 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1137 nfsi
->cache_validity
= 0UL;
1138 #ifdef CONFIG_NFS_V3_ACL
1139 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1140 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1142 #ifdef CONFIG_NFS_V4
1143 nfsi
->nfs4_acl
= NULL
;
1144 #endif /* CONFIG_NFS_V4 */
1145 return &nfsi
->vfs_inode
;
1148 void nfs_destroy_inode(struct inode
*inode
)
1150 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1153 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1155 #ifdef CONFIG_NFS_V4
1156 INIT_LIST_HEAD(&nfsi
->open_states
);
1157 nfsi
->delegation
= NULL
;
1158 nfsi
->delegation_state
= 0;
1159 init_rwsem(&nfsi
->rwsem
);
1163 static void init_once(void * foo
, struct kmem_cache
* cachep
, unsigned long flags
)
1165 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1167 inode_init_once(&nfsi
->vfs_inode
);
1168 spin_lock_init(&nfsi
->req_lock
);
1169 INIT_LIST_HEAD(&nfsi
->dirty
);
1170 INIT_LIST_HEAD(&nfsi
->commit
);
1171 INIT_LIST_HEAD(&nfsi
->open_files
);
1172 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1173 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1174 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1175 atomic_set(&nfsi
->data_updates
, 0);
1179 nfs4_init_once(nfsi
);
1182 static int __init
nfs_init_inodecache(void)
1184 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1185 sizeof(struct nfs_inode
),
1186 0, (SLAB_RECLAIM_ACCOUNT
|
1189 if (nfs_inode_cachep
== NULL
)
1195 static void nfs_destroy_inodecache(void)
1197 kmem_cache_destroy(nfs_inode_cachep
);
1203 static int __init
init_nfs_fs(void)
1207 err
= nfs_fs_proc_init();
1211 err
= nfs_init_nfspagecache();
1215 err
= nfs_init_inodecache();
1219 err
= nfs_init_readpagecache();
1223 err
= nfs_init_writepagecache();
1227 err
= nfs_init_directcache();
1231 #ifdef CONFIG_PROC_FS
1232 rpc_proc_register(&nfs_rpcstat
);
1234 if ((err
= register_nfs_fs()) != 0)
1238 #ifdef CONFIG_PROC_FS
1239 rpc_proc_unregister("nfs");
1241 nfs_destroy_directcache();
1243 nfs_destroy_writepagecache();
1245 nfs_destroy_readpagecache();
1247 nfs_destroy_inodecache();
1249 nfs_destroy_nfspagecache();
1256 static void __exit
exit_nfs_fs(void)
1258 nfs_destroy_directcache();
1259 nfs_destroy_writepagecache();
1260 nfs_destroy_readpagecache();
1261 nfs_destroy_inodecache();
1262 nfs_destroy_nfspagecache();
1263 #ifdef CONFIG_PROC_FS
1264 rpc_proc_unregister("nfs");
1266 unregister_nfs_fs();
1270 /* Not quite true; I just maintain it */
1271 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1272 MODULE_LICENSE("GPL");
1274 module_init(init_nfs_fs
)
1275 module_exit(exit_nfs_fs
)