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>
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
51 #define NFS_PARANOIA 1
53 static void nfs_invalidate_inode(struct inode
*);
54 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
56 static void nfs_zap_acl_cache(struct inode
*);
58 static struct kmem_cache
* nfs_inode_cachep
;
60 static inline unsigned long
61 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
63 return nfs_fileid_to_ino_t(fattr
->fileid
);
66 int nfs_write_inode(struct inode
*inode
, int sync
)
68 int flags
= sync
? FLUSH_SYNC
: 0;
71 ret
= nfs_commit_inode(inode
, flags
);
77 void nfs_clear_inode(struct inode
*inode
)
80 * The following should never happen...
82 BUG_ON(nfs_have_writebacks(inode
));
83 BUG_ON(!list_empty(&NFS_I(inode
)->open_files
));
84 BUG_ON(atomic_read(&NFS_I(inode
)->data_updates
) != 0);
85 nfs_zap_acl_cache(inode
);
86 nfs_access_zap_cache(inode
);
90 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
92 int nfs_sync_mapping(struct address_space
*mapping
)
96 if (mapping
->nrpages
== 0)
98 unmap_mapping_range(mapping
, 0, 0, 0);
99 ret
= filemap_write_and_wait(mapping
);
102 ret
= nfs_wb_all(mapping
->host
);
108 * Invalidate the local caches
110 static void nfs_zap_caches_locked(struct inode
*inode
)
112 struct nfs_inode
*nfsi
= NFS_I(inode
);
113 int mode
= inode
->i_mode
;
115 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
117 NFS_ATTRTIMEO(inode
) = NFS_MINATTRTIMEO(inode
);
118 NFS_ATTRTIMEO_UPDATE(inode
) = jiffies
;
120 memset(NFS_COOKIEVERF(inode
), 0, sizeof(NFS_COOKIEVERF(inode
)));
121 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))
122 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
124 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
127 void nfs_zap_caches(struct inode
*inode
)
129 spin_lock(&inode
->i_lock
);
130 nfs_zap_caches_locked(inode
);
131 spin_unlock(&inode
->i_lock
);
134 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
136 if (mapping
->nrpages
!= 0) {
137 spin_lock(&inode
->i_lock
);
138 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
139 spin_unlock(&inode
->i_lock
);
143 static void nfs_zap_acl_cache(struct inode
*inode
)
145 void (*clear_acl_cache
)(struct inode
*);
147 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
148 if (clear_acl_cache
!= NULL
)
149 clear_acl_cache(inode
);
150 spin_lock(&inode
->i_lock
);
151 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
152 spin_unlock(&inode
->i_lock
);
156 * Invalidate, but do not unhash, the inode.
157 * NB: must be called with inode->i_lock held!
159 static void nfs_invalidate_inode(struct inode
*inode
)
161 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
162 nfs_zap_caches_locked(inode
);
165 struct nfs_find_desc
{
167 struct nfs_fattr
*fattr
;
171 * In NFSv3 we can have 64bit inode numbers. In order to support
172 * this, and re-exported directories (also seen in NFSv2)
173 * we are forced to allow 2 different inodes to have the same
177 nfs_find_actor(struct inode
*inode
, void *opaque
)
179 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
180 struct nfs_fh
*fh
= desc
->fh
;
181 struct nfs_fattr
*fattr
= desc
->fattr
;
183 if (NFS_FILEID(inode
) != fattr
->fileid
)
185 if (nfs_compare_fh(NFS_FH(inode
), fh
))
187 if (is_bad_inode(inode
) || NFS_STALE(inode
))
193 nfs_init_locked(struct inode
*inode
, void *opaque
)
195 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
196 struct nfs_fattr
*fattr
= desc
->fattr
;
198 NFS_FILEID(inode
) = fattr
->fileid
;
199 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
203 /* Don't use READDIRPLUS on directories that we believe are too large */
204 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
207 * This is our front-end to iget that looks up inodes by file handle
208 * instead of inode number.
211 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
213 struct nfs_find_desc desc
= {
217 struct inode
*inode
= ERR_PTR(-ENOENT
);
220 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
224 printk("NFS: Buggy server - nlink == 0!\n");
228 hash
= nfs_fattr_to_ino_t(fattr
);
230 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
232 inode
= ERR_PTR(-ENOMEM
);
236 if (inode
->i_state
& I_NEW
) {
237 struct nfs_inode
*nfsi
= NFS_I(inode
);
239 /* We set i_ino for the few things that still rely on it,
243 /* We can't support update_atime(), since the server will reset it */
244 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
245 inode
->i_mode
= fattr
->mode
;
246 /* Why so? Because we want revalidate for devices/FIFOs, and
247 * that's precisely what we have in nfs_file_inode_operations.
249 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
250 if (S_ISREG(inode
->i_mode
)) {
251 inode
->i_fop
= &nfs_file_operations
;
252 inode
->i_data
.a_ops
= &nfs_file_aops
;
253 inode
->i_data
.backing_dev_info
= &NFS_SB(sb
)->backing_dev_info
;
254 } else if (S_ISDIR(inode
->i_mode
)) {
255 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
256 inode
->i_fop
= &nfs_dir_operations
;
257 if (nfs_server_capable(inode
, NFS_CAP_READDIRPLUS
)
258 && fattr
->size
<= NFS_LIMIT_READDIRPLUS
)
259 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_FLAGS(inode
));
260 /* Deal with crossing mountpoints */
261 if (!nfs_fsid_equal(&NFS_SB(sb
)->fsid
, &fattr
->fsid
)) {
262 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
263 inode
->i_op
= &nfs_referral_inode_operations
;
265 inode
->i_op
= &nfs_mountpoint_inode_operations
;
268 } else if (S_ISLNK(inode
->i_mode
))
269 inode
->i_op
= &nfs_symlink_inode_operations
;
271 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
273 nfsi
->read_cache_jiffies
= fattr
->time_start
;
274 nfsi
->last_updated
= jiffies
;
275 inode
->i_atime
= fattr
->atime
;
276 inode
->i_mtime
= fattr
->mtime
;
277 inode
->i_ctime
= fattr
->ctime
;
278 if (fattr
->valid
& NFS_ATTR_FATTR_V4
)
279 nfsi
->change_attr
= fattr
->change_attr
;
280 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
281 inode
->i_nlink
= fattr
->nlink
;
282 inode
->i_uid
= fattr
->uid
;
283 inode
->i_gid
= fattr
->gid
;
284 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
286 * report the blocks in 512byte units
288 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
290 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
292 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
293 nfsi
->attrtimeo_timestamp
= jiffies
;
294 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
295 nfsi
->access_cache
= RB_ROOT
;
297 unlock_new_inode(inode
);
299 nfs_refresh_inode(inode
, fattr
);
300 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
302 (long long)NFS_FILEID(inode
),
303 atomic_read(&inode
->i_count
));
309 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
313 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
316 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
318 struct inode
*inode
= dentry
->d_inode
;
319 struct nfs_fattr fattr
;
322 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
324 if (attr
->ia_valid
& ATTR_SIZE
) {
325 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
326 attr
->ia_valid
&= ~ATTR_SIZE
;
329 /* Optimization: if the end result is no change, don't RPC */
330 attr
->ia_valid
&= NFS_VALID_ATTRS
;
331 if (attr
->ia_valid
== 0)
335 nfs_begin_data_update(inode
);
336 /* Write all dirty data */
337 filemap_write_and_wait(inode
->i_mapping
);
340 * Return any delegations if we're going to change ACLs
342 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
343 nfs_inode_return_delegation(inode
);
344 error
= NFS_PROTO(inode
)->setattr(dentry
, &fattr
, attr
);
346 nfs_refresh_inode(inode
, &fattr
);
347 nfs_end_data_update(inode
);
353 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
354 * @inode: pointer to struct inode
355 * @attr: pointer to struct iattr
357 * Note: we do this in the *proc.c in order to ensure that
358 * it works for things like exclusive creates too.
360 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
362 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
363 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
364 int mode
= attr
->ia_mode
& S_IALLUGO
;
365 mode
|= inode
->i_mode
& ~S_IALLUGO
;
366 inode
->i_mode
= mode
;
368 if ((attr
->ia_valid
& ATTR_UID
) != 0)
369 inode
->i_uid
= attr
->ia_uid
;
370 if ((attr
->ia_valid
& ATTR_GID
) != 0)
371 inode
->i_gid
= attr
->ia_gid
;
372 spin_lock(&inode
->i_lock
);
373 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
374 spin_unlock(&inode
->i_lock
);
376 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
377 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
378 inode
->i_size
= attr
->ia_size
;
379 vmtruncate(inode
, attr
->ia_size
);
383 static int nfs_wait_schedule(void *word
)
385 if (signal_pending(current
))
392 * Wait for the inode to get unlocked.
394 static int nfs_wait_on_inode(struct inode
*inode
)
396 struct rpc_clnt
*clnt
= NFS_CLIENT(inode
);
397 struct nfs_inode
*nfsi
= NFS_I(inode
);
401 rpc_clnt_sigmask(clnt
, &oldmask
);
402 error
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_REVALIDATING
,
403 nfs_wait_schedule
, TASK_INTERRUPTIBLE
);
404 rpc_clnt_sigunmask(clnt
, &oldmask
);
409 static void nfs_wake_up_inode(struct inode
*inode
)
411 struct nfs_inode
*nfsi
= NFS_I(inode
);
413 clear_bit(NFS_INO_REVALIDATING
, &nfsi
->flags
);
414 smp_mb__after_clear_bit();
415 wake_up_bit(&nfsi
->flags
, NFS_INO_REVALIDATING
);
418 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
420 struct inode
*inode
= dentry
->d_inode
;
421 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
424 /* Flush out writes to the server in order to update c/mtime */
425 nfs_sync_mapping_range(inode
->i_mapping
, 0, 0, FLUSH_NOCOMMIT
);
428 * We may force a getattr if the user cares about atime.
430 * Note that we only have to check the vfsmount flags here:
431 * - NFS always sets S_NOATIME by so checking it would give a
433 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
434 * no point in checking those.
436 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
437 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
441 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
443 err
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
445 generic_fillattr(inode
, stat
);
449 static struct nfs_open_context
*alloc_nfs_open_context(struct vfsmount
*mnt
, struct dentry
*dentry
, struct rpc_cred
*cred
)
451 struct nfs_open_context
*ctx
;
453 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
455 atomic_set(&ctx
->count
, 1);
456 ctx
->dentry
= dget(dentry
);
457 ctx
->vfsmnt
= mntget(mnt
);
458 ctx
->cred
= get_rpccred(cred
);
460 ctx
->lockowner
= current
->files
;
467 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
470 atomic_inc(&ctx
->count
);
474 void put_nfs_open_context(struct nfs_open_context
*ctx
)
476 if (atomic_dec_and_test(&ctx
->count
)) {
477 if (!list_empty(&ctx
->list
)) {
478 struct inode
*inode
= ctx
->dentry
->d_inode
;
479 spin_lock(&inode
->i_lock
);
480 list_del(&ctx
->list
);
481 spin_unlock(&inode
->i_lock
);
483 if (ctx
->state
!= NULL
)
484 nfs4_close_state(ctx
->state
, ctx
->mode
);
485 if (ctx
->cred
!= NULL
)
486 put_rpccred(ctx
->cred
);
494 * Ensure that mmap has a recent RPC credential for use when writing out
497 static void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
499 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
500 struct nfs_inode
*nfsi
= NFS_I(inode
);
502 filp
->private_data
= get_nfs_open_context(ctx
);
503 spin_lock(&inode
->i_lock
);
504 list_add(&ctx
->list
, &nfsi
->open_files
);
505 spin_unlock(&inode
->i_lock
);
509 * Given an inode, search for an open context with the desired characteristics
511 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, int mode
)
513 struct nfs_inode
*nfsi
= NFS_I(inode
);
514 struct nfs_open_context
*pos
, *ctx
= NULL
;
516 spin_lock(&inode
->i_lock
);
517 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
518 if (cred
!= NULL
&& pos
->cred
!= cred
)
520 if ((pos
->mode
& mode
) == mode
) {
521 ctx
= get_nfs_open_context(pos
);
525 spin_unlock(&inode
->i_lock
);
529 static void nfs_file_clear_open_context(struct file
*filp
)
531 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
532 struct nfs_open_context
*ctx
= (struct nfs_open_context
*)filp
->private_data
;
535 filp
->private_data
= NULL
;
536 spin_lock(&inode
->i_lock
);
537 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
538 spin_unlock(&inode
->i_lock
);
539 put_nfs_open_context(ctx
);
544 * These allocate and release file read/write context information.
546 int nfs_open(struct inode
*inode
, struct file
*filp
)
548 struct nfs_open_context
*ctx
;
549 struct rpc_cred
*cred
;
551 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
553 return PTR_ERR(cred
);
554 ctx
= alloc_nfs_open_context(filp
->f_path
.mnt
, filp
->f_path
.dentry
, cred
);
558 ctx
->mode
= filp
->f_mode
;
559 nfs_file_set_open_context(filp
, ctx
);
560 put_nfs_open_context(ctx
);
564 int nfs_release(struct inode
*inode
, struct file
*filp
)
566 nfs_file_clear_open_context(filp
);
571 * This function is called whenever some part of NFS notices that
572 * the cached attributes have to be refreshed.
575 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
577 int status
= -ESTALE
;
578 struct nfs_fattr fattr
;
579 struct nfs_inode
*nfsi
= NFS_I(inode
);
581 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
582 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
584 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
586 if (is_bad_inode(inode
))
588 if (NFS_STALE(inode
))
591 status
= nfs_wait_on_inode(inode
);
594 if (NFS_STALE(inode
)) {
596 /* Do we trust the cached ESTALE? */
597 if (NFS_ATTRTIMEO(inode
) != 0) {
598 if (nfsi
->cache_validity
& (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
)) {
605 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), &fattr
);
607 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
609 (long long)NFS_FILEID(inode
), status
);
610 if (status
== -ESTALE
) {
611 nfs_zap_caches(inode
);
612 if (!S_ISDIR(inode
->i_mode
))
613 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
618 spin_lock(&inode
->i_lock
);
619 status
= nfs_update_inode(inode
, &fattr
);
621 spin_unlock(&inode
->i_lock
);
622 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
624 (long long)NFS_FILEID(inode
), status
);
627 spin_unlock(&inode
->i_lock
);
629 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
630 nfs_zap_acl_cache(inode
);
632 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
634 (long long)NFS_FILEID(inode
));
637 nfs_wake_up_inode(inode
);
644 int nfs_attribute_timeout(struct inode
*inode
)
646 struct nfs_inode
*nfsi
= NFS_I(inode
);
648 if (nfs_have_delegation(inode
, FMODE_READ
))
650 return time_after(jiffies
, nfsi
->read_cache_jiffies
+nfsi
->attrtimeo
);
654 * nfs_revalidate_inode - Revalidate the inode attributes
655 * @server - pointer to nfs_server struct
656 * @inode - pointer to inode struct
658 * Updates inode attribute information by retrieving the data from the server.
660 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
662 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
663 && !nfs_attribute_timeout(inode
))
664 return NFS_STALE(inode
) ? -ESTALE
: 0;
665 return __nfs_revalidate_inode(server
, inode
);
668 static int nfs_invalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
670 struct nfs_inode
*nfsi
= NFS_I(inode
);
672 if (mapping
->nrpages
!= 0) {
673 int ret
= invalidate_inode_pages2(mapping
);
677 spin_lock(&inode
->i_lock
);
678 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
679 if (S_ISDIR(inode
->i_mode
)) {
680 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
681 /* This ensures we revalidate child dentries */
682 nfsi
->cache_change_attribute
= jiffies
;
684 spin_unlock(&inode
->i_lock
);
685 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
686 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
687 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
691 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
695 mutex_lock(&inode
->i_mutex
);
696 if (NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_DATA
) {
697 ret
= nfs_sync_mapping(mapping
);
699 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
701 mutex_unlock(&inode
->i_mutex
);
706 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
707 * @inode - pointer to host inode
708 * @mapping - pointer to mapping
710 int nfs_revalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
712 struct nfs_inode
*nfsi
= NFS_I(inode
);
715 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
716 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
717 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
721 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
722 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
728 * nfs_revalidate_mapping - Revalidate the pagecache
729 * @inode - pointer to host inode
730 * @mapping - pointer to mapping
732 * This version of the function will take the inode->i_mutex and attempt to
733 * flush out all dirty data if it needs to invalidate the page cache.
735 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
737 struct nfs_inode
*nfsi
= NFS_I(inode
);
740 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
741 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
742 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
746 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
747 ret
= nfs_invalidate_mapping(inode
, mapping
);
753 * nfs_begin_data_update
754 * @inode - pointer to inode
755 * Declare that a set of operations will update file data on the server
757 void nfs_begin_data_update(struct inode
*inode
)
759 atomic_inc(&NFS_I(inode
)->data_updates
);
763 * nfs_end_data_update
764 * @inode - pointer to inode
765 * Declare end of the operations that will update file data
766 * This will mark the inode as immediately needing revalidation
767 * of its attribute cache.
769 void nfs_end_data_update(struct inode
*inode
)
771 struct nfs_inode
*nfsi
= NFS_I(inode
);
773 /* Directories: invalidate page cache */
774 if (S_ISDIR(inode
->i_mode
)) {
775 spin_lock(&inode
->i_lock
);
776 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
777 spin_unlock(&inode
->i_lock
);
779 nfsi
->cache_change_attribute
= jiffies
;
780 atomic_dec(&nfsi
->data_updates
);
783 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
785 struct nfs_inode
*nfsi
= NFS_I(inode
);
787 /* If we have atomic WCC data, we may update some attributes */
788 if ((fattr
->valid
& NFS_ATTR_WCC
) != 0) {
789 if (timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
)) {
790 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
791 nfsi
->cache_change_attribute
= jiffies
;
793 if (timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
794 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
795 nfsi
->cache_change_attribute
= jiffies
;
797 if (inode
->i_size
== fattr
->pre_size
&& nfsi
->npages
== 0) {
798 inode
->i_size
= fattr
->size
;
799 nfsi
->cache_change_attribute
= jiffies
;
805 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
806 * @inode - pointer to inode
807 * @fattr - updated attributes
809 * Verifies the attribute cache. If we have just changed the attributes,
810 * so that fattr carries weak cache consistency data, then it may
811 * also update the ctime/mtime/change_attribute.
813 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
815 struct nfs_inode
*nfsi
= NFS_I(inode
);
816 loff_t cur_size
, new_isize
;
820 /* Has the inode gone and changed behind our back? */
821 if (nfsi
->fileid
!= fattr
->fileid
822 || (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
826 /* Are we in the process of updating data on the server? */
827 data_unstable
= nfs_caches_unstable(inode
);
829 /* Do atomic weak cache consistency updates */
830 nfs_wcc_update_inode(inode
, fattr
);
832 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
833 nfsi
->change_attr
!= fattr
->change_attr
)
834 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
836 /* Verify a few of the more important attributes */
837 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
838 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
840 cur_size
= i_size_read(inode
);
841 new_isize
= nfs_size_to_loff_t(fattr
->size
);
842 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
843 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
845 /* Have any file permissions changed? */
846 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)
847 || inode
->i_uid
!= fattr
->uid
848 || inode
->i_gid
!= fattr
->gid
)
849 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
851 /* Has the link count changed? */
852 if (inode
->i_nlink
!= fattr
->nlink
)
853 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
855 if (!timespec_equal(&inode
->i_atime
, &fattr
->atime
))
856 nfsi
->cache_validity
|= NFS_INO_INVALID_ATIME
;
858 nfsi
->read_cache_jiffies
= fattr
->time_start
;
863 * nfs_refresh_inode - try to update the inode attribute cache
864 * @inode - pointer to inode
865 * @fattr - updated attributes
867 * Check that an RPC call that returned attributes has not overlapped with
868 * other recent updates of the inode metadata, then decide whether it is
869 * safe to do a full update of the inode attributes, or whether just to
870 * call nfs_check_inode_attributes.
872 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
874 struct nfs_inode
*nfsi
= NFS_I(inode
);
877 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
879 spin_lock(&inode
->i_lock
);
880 if (time_after(fattr
->time_start
, nfsi
->last_updated
))
881 status
= nfs_update_inode(inode
, fattr
);
883 status
= nfs_check_inode_attributes(inode
, fattr
);
885 spin_unlock(&inode
->i_lock
);
890 * nfs_post_op_update_inode - try to update the inode attribute cache
891 * @inode - pointer to inode
892 * @fattr - updated attributes
894 * After an operation that has changed the inode metadata, mark the
895 * attribute cache as being invalid, then try to update it.
897 * NB: if the server didn't return any post op attributes, this
898 * function will force the retrieval of attributes before the next
899 * NFS request. Thus it should be used only for operations that
900 * are expected to change one or more attributes, to avoid
901 * unnecessary NFS requests and trips through nfs_update_inode().
903 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
905 struct nfs_inode
*nfsi
= NFS_I(inode
);
908 spin_lock(&inode
->i_lock
);
909 if (unlikely((fattr
->valid
& NFS_ATTR_FATTR
) == 0)) {
910 nfsi
->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
913 status
= nfs_update_inode(inode
, fattr
);
915 spin_unlock(&inode
->i_lock
);
920 * Many nfs protocol calls return the new file attributes after
921 * an operation. Here we update the inode to reflect the state
922 * of the server's inode.
924 * This is a bit tricky because we have to make sure all dirty pages
925 * have been sent off to the server before calling invalidate_inode_pages.
926 * To make sure no other process adds more write requests while we try
927 * our best to flush them, we make them sleep during the attribute refresh.
929 * A very similar scenario holds for the dir cache.
931 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
933 struct nfs_server
*server
;
934 struct nfs_inode
*nfsi
= NFS_I(inode
);
935 loff_t cur_isize
, new_isize
;
936 unsigned int invalid
= 0;
939 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
940 __FUNCTION__
, inode
->i_sb
->s_id
, inode
->i_ino
,
941 atomic_read(&inode
->i_count
), fattr
->valid
);
943 if (nfsi
->fileid
!= fattr
->fileid
)
947 * Make sure the inode's type hasn't changed.
949 if ((inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
952 server
= NFS_SERVER(inode
);
953 /* Update the fsid if and only if this is the root directory */
954 if (inode
== inode
->i_sb
->s_root
->d_inode
955 && !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
956 server
->fsid
= fattr
->fsid
;
959 * Update the read time so we don't revalidate too often.
961 nfsi
->read_cache_jiffies
= fattr
->time_start
;
962 nfsi
->last_updated
= jiffies
;
964 /* Are we racing with known updates of the metadata on the server? */
965 data_stable
= nfs_verify_change_attribute(inode
, fattr
->time_start
);
967 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
|NFS_INO_INVALID_ATIME
);
969 /* Do atomic weak cache consistency updates */
970 nfs_wcc_update_inode(inode
, fattr
);
972 /* Check if our cached file size is stale */
973 new_isize
= nfs_size_to_loff_t(fattr
->size
);
974 cur_isize
= i_size_read(inode
);
975 if (new_isize
!= cur_isize
) {
976 /* Do we perhaps have any outstanding writes? */
977 if (nfsi
->npages
== 0) {
978 /* No, but did we race with nfs_end_data_update()? */
980 inode
->i_size
= new_isize
;
981 invalid
|= NFS_INO_INVALID_DATA
;
983 invalid
|= NFS_INO_INVALID_ATTR
;
984 } else if (new_isize
> cur_isize
) {
985 inode
->i_size
= new_isize
;
986 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
988 nfsi
->cache_change_attribute
= jiffies
;
989 dprintk("NFS: isize change on server for file %s/%ld\n",
990 inode
->i_sb
->s_id
, inode
->i_ino
);
993 /* Check if the mtime agrees */
994 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
995 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
996 dprintk("NFS: mtime change on server for file %s/%ld\n",
997 inode
->i_sb
->s_id
, inode
->i_ino
);
998 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
999 nfsi
->cache_change_attribute
= jiffies
;
1002 /* If ctime has changed we should definitely clear access+acl caches */
1003 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
)) {
1004 invalid
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1005 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1006 nfsi
->cache_change_attribute
= jiffies
;
1008 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1010 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
) ||
1011 inode
->i_uid
!= fattr
->uid
||
1012 inode
->i_gid
!= fattr
->gid
)
1013 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1015 inode
->i_mode
= fattr
->mode
;
1016 inode
->i_nlink
= fattr
->nlink
;
1017 inode
->i_uid
= fattr
->uid
;
1018 inode
->i_gid
= fattr
->gid
;
1020 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
1022 * report the blocks in 512byte units
1024 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1026 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1029 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
1030 nfsi
->change_attr
!= fattr
->change_attr
) {
1031 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1032 inode
->i_sb
->s_id
, inode
->i_ino
);
1033 nfsi
->change_attr
= fattr
->change_attr
;
1034 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1035 nfsi
->cache_change_attribute
= jiffies
;
1038 /* Update attrtimeo value if we're out of the unstable period */
1039 if (invalid
& NFS_INO_INVALID_ATTR
) {
1040 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1041 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1042 nfsi
->attrtimeo_timestamp
= jiffies
;
1043 } else if (time_after(jiffies
, nfsi
->attrtimeo_timestamp
+nfsi
->attrtimeo
)) {
1044 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1045 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1046 nfsi
->attrtimeo_timestamp
= jiffies
;
1048 /* Don't invalidate the data if we were to blame */
1049 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1050 || S_ISLNK(inode
->i_mode
)))
1051 invalid
&= ~NFS_INO_INVALID_DATA
;
1053 invalid
&= ~(NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ATIME
|NFS_INO_REVAL_PAGECACHE
);
1054 if (!nfs_have_delegation(inode
, FMODE_READ
))
1055 nfsi
->cache_validity
|= invalid
;
1060 * Big trouble! The inode has become a different object.
1063 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1064 __FUNCTION__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1068 * No need to worry about unhashing the dentry, as the
1069 * lookup validation will know that the inode is bad.
1070 * (But we fall through to invalidate the caches.)
1072 nfs_invalidate_inode(inode
);
1076 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1077 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1078 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1079 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1084 #ifdef CONFIG_NFS_V4
1087 * Clean out any remaining NFSv4 state that might be left over due
1088 * to open() calls that passed nfs_atomic_lookup, but failed to call
1091 void nfs4_clear_inode(struct inode
*inode
)
1093 struct nfs_inode
*nfsi
= NFS_I(inode
);
1095 /* If we are holding a delegation, return it! */
1096 nfs_inode_return_delegation(inode
);
1097 /* First call standard NFS clear_inode() code */
1098 nfs_clear_inode(inode
);
1099 /* Now clear out any remaining state */
1100 while (!list_empty(&nfsi
->open_states
)) {
1101 struct nfs4_state
*state
;
1103 state
= list_entry(nfsi
->open_states
.next
,
1106 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1109 (long long)NFS_FILEID(inode
),
1111 BUG_ON(atomic_read(&state
->count
) != 1);
1112 nfs4_close_state(state
, state
->state
);
1117 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1119 struct nfs_inode
*nfsi
;
1120 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1124 nfsi
->cache_validity
= 0UL;
1125 nfsi
->cache_change_attribute
= jiffies
;
1126 #ifdef CONFIG_NFS_V3_ACL
1127 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1128 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1130 #ifdef CONFIG_NFS_V4
1131 nfsi
->nfs4_acl
= NULL
;
1132 #endif /* CONFIG_NFS_V4 */
1133 return &nfsi
->vfs_inode
;
1136 void nfs_destroy_inode(struct inode
*inode
)
1138 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1141 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1143 #ifdef CONFIG_NFS_V4
1144 INIT_LIST_HEAD(&nfsi
->open_states
);
1145 nfsi
->delegation
= NULL
;
1146 nfsi
->delegation_state
= 0;
1147 init_rwsem(&nfsi
->rwsem
);
1151 static void init_once(void * foo
, struct kmem_cache
* cachep
, unsigned long flags
)
1153 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1155 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) ==
1156 SLAB_CTOR_CONSTRUCTOR
) {
1157 inode_init_once(&nfsi
->vfs_inode
);
1158 spin_lock_init(&nfsi
->req_lock
);
1159 INIT_LIST_HEAD(&nfsi
->dirty
);
1160 INIT_LIST_HEAD(&nfsi
->commit
);
1161 INIT_LIST_HEAD(&nfsi
->open_files
);
1162 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1163 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1164 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1165 atomic_set(&nfsi
->data_updates
, 0);
1169 nfs4_init_once(nfsi
);
1173 static int __init
nfs_init_inodecache(void)
1175 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1176 sizeof(struct nfs_inode
),
1177 0, (SLAB_RECLAIM_ACCOUNT
|
1180 if (nfs_inode_cachep
== NULL
)
1186 static void nfs_destroy_inodecache(void)
1188 kmem_cache_destroy(nfs_inode_cachep
);
1194 static int __init
init_nfs_fs(void)
1198 err
= nfs_fs_proc_init();
1202 err
= nfs_init_nfspagecache();
1206 err
= nfs_init_inodecache();
1210 err
= nfs_init_readpagecache();
1214 err
= nfs_init_writepagecache();
1218 err
= nfs_init_directcache();
1222 #ifdef CONFIG_PROC_FS
1223 rpc_proc_register(&nfs_rpcstat
);
1225 if ((err
= register_nfs_fs()) != 0)
1229 #ifdef CONFIG_PROC_FS
1230 rpc_proc_unregister("nfs");
1232 nfs_destroy_directcache();
1234 nfs_destroy_writepagecache();
1236 nfs_destroy_readpagecache();
1238 nfs_destroy_inodecache();
1240 nfs_destroy_nfspagecache();
1247 static void __exit
exit_nfs_fs(void)
1249 nfs_destroy_directcache();
1250 nfs_destroy_writepagecache();
1251 nfs_destroy_readpagecache();
1252 nfs_destroy_inodecache();
1253 nfs_destroy_nfspagecache();
1254 #ifdef CONFIG_PROC_FS
1255 rpc_proc_unregister("nfs");
1257 unregister_nfs_fs();
1261 /* Not quite true; I just maintain it */
1262 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1263 MODULE_LICENSE("GPL");
1265 module_init(init_nfs_fs
)
1266 module_exit(exit_nfs_fs
)