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 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
54 /* Default is to see 64-bit inode numbers */
55 static int enable_ino64
= NFS_64_BIT_INODE_NUMBERS_ENABLED
;
57 static void nfs_invalidate_inode(struct inode
*);
58 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
60 static void nfs_zap_acl_cache(struct inode
*);
62 static struct kmem_cache
* nfs_inode_cachep
;
64 static inline unsigned long
65 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
67 return nfs_fileid_to_ino_t(fattr
->fileid
);
71 * nfs_compat_user_ino64 - returns the user-visible inode number
72 * @fileid: 64-bit fileid
74 * This function returns a 32-bit inode number if the boot parameter
75 * nfs.enable_ino64 is zero.
77 u64
nfs_compat_user_ino64(u64 fileid
)
84 if (sizeof(ino
) < sizeof(fileid
))
85 ino
^= fileid
>> (sizeof(fileid
)-sizeof(ino
)) * 8;
89 int nfs_write_inode(struct inode
*inode
, int sync
)
94 ret
= filemap_fdatawait(inode
->i_mapping
);
96 ret
= nfs_commit_inode(inode
, FLUSH_SYNC
);
98 ret
= nfs_commit_inode(inode
, 0);
101 __mark_inode_dirty(inode
, I_DIRTY_DATASYNC
);
105 void nfs_clear_inode(struct inode
*inode
)
108 * The following should never happen...
110 BUG_ON(nfs_have_writebacks(inode
));
111 BUG_ON(!list_empty(&NFS_I(inode
)->open_files
));
112 nfs_zap_acl_cache(inode
);
113 nfs_access_zap_cache(inode
);
117 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
119 int nfs_sync_mapping(struct address_space
*mapping
)
123 if (mapping
->nrpages
== 0)
125 unmap_mapping_range(mapping
, 0, 0, 0);
126 ret
= filemap_write_and_wait(mapping
);
129 ret
= nfs_wb_all(mapping
->host
);
135 * Invalidate the local caches
137 static void nfs_zap_caches_locked(struct inode
*inode
)
139 struct nfs_inode
*nfsi
= NFS_I(inode
);
140 int mode
= inode
->i_mode
;
142 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
144 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
145 nfsi
->attrtimeo_timestamp
= jiffies
;
147 memset(NFS_COOKIEVERF(inode
), 0, sizeof(NFS_COOKIEVERF(inode
)));
148 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))
149 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
151 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
|NFS_INO_REVAL_PAGECACHE
;
154 void nfs_zap_caches(struct inode
*inode
)
156 spin_lock(&inode
->i_lock
);
157 nfs_zap_caches_locked(inode
);
158 spin_unlock(&inode
->i_lock
);
161 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
163 if (mapping
->nrpages
!= 0) {
164 spin_lock(&inode
->i_lock
);
165 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
166 spin_unlock(&inode
->i_lock
);
170 static void nfs_zap_acl_cache(struct inode
*inode
)
172 void (*clear_acl_cache
)(struct inode
*);
174 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
175 if (clear_acl_cache
!= NULL
)
176 clear_acl_cache(inode
);
177 spin_lock(&inode
->i_lock
);
178 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
179 spin_unlock(&inode
->i_lock
);
182 void nfs_invalidate_atime(struct inode
*inode
)
184 spin_lock(&inode
->i_lock
);
185 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
186 spin_unlock(&inode
->i_lock
);
190 * Invalidate, but do not unhash, the inode.
191 * NB: must be called with inode->i_lock held!
193 static void nfs_invalidate_inode(struct inode
*inode
)
195 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
196 nfs_zap_caches_locked(inode
);
199 struct nfs_find_desc
{
201 struct nfs_fattr
*fattr
;
205 * In NFSv3 we can have 64bit inode numbers. In order to support
206 * this, and re-exported directories (also seen in NFSv2)
207 * we are forced to allow 2 different inodes to have the same
211 nfs_find_actor(struct inode
*inode
, void *opaque
)
213 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
214 struct nfs_fh
*fh
= desc
->fh
;
215 struct nfs_fattr
*fattr
= desc
->fattr
;
217 if (NFS_FILEID(inode
) != fattr
->fileid
)
219 if (nfs_compare_fh(NFS_FH(inode
), fh
))
221 if (is_bad_inode(inode
) || NFS_STALE(inode
))
227 nfs_init_locked(struct inode
*inode
, void *opaque
)
229 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
230 struct nfs_fattr
*fattr
= desc
->fattr
;
232 set_nfs_fileid(inode
, fattr
->fileid
);
233 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
237 /* Don't use READDIRPLUS on directories that we believe are too large */
238 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
241 * This is our front-end to iget that looks up inodes by file handle
242 * instead of inode number.
245 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
)
247 struct nfs_find_desc desc
= {
251 struct inode
*inode
= ERR_PTR(-ENOENT
);
254 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
258 printk("NFS: Buggy server - nlink == 0!\n");
262 hash
= nfs_fattr_to_ino_t(fattr
);
264 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
266 inode
= ERR_PTR(-ENOMEM
);
270 if (inode
->i_state
& I_NEW
) {
271 struct nfs_inode
*nfsi
= NFS_I(inode
);
272 unsigned long now
= jiffies
;
274 /* We set i_ino for the few things that still rely on it,
278 /* We can't support update_atime(), since the server will reset it */
279 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
280 inode
->i_mode
= fattr
->mode
;
281 /* Why so? Because we want revalidate for devices/FIFOs, and
282 * that's precisely what we have in nfs_file_inode_operations.
284 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
285 if (S_ISREG(inode
->i_mode
)) {
286 inode
->i_fop
= &nfs_file_operations
;
287 inode
->i_data
.a_ops
= &nfs_file_aops
;
288 inode
->i_data
.backing_dev_info
= &NFS_SB(sb
)->backing_dev_info
;
289 } else if (S_ISDIR(inode
->i_mode
)) {
290 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
291 inode
->i_fop
= &nfs_dir_operations
;
292 if (nfs_server_capable(inode
, NFS_CAP_READDIRPLUS
)
293 && fattr
->size
<= NFS_LIMIT_READDIRPLUS
)
294 set_bit(NFS_INO_ADVISE_RDPLUS
, &NFS_I(inode
)->flags
);
295 /* Deal with crossing mountpoints */
296 if (!nfs_fsid_equal(&NFS_SB(sb
)->fsid
, &fattr
->fsid
)) {
297 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
298 inode
->i_op
= &nfs_referral_inode_operations
;
300 inode
->i_op
= &nfs_mountpoint_inode_operations
;
302 set_bit(NFS_INO_MOUNTPOINT
, &nfsi
->flags
);
304 } else if (S_ISLNK(inode
->i_mode
))
305 inode
->i_op
= &nfs_symlink_inode_operations
;
307 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
309 nfsi
->read_cache_jiffies
= fattr
->time_start
;
310 nfsi
->last_updated
= now
;
311 nfsi
->cache_change_attribute
= now
;
312 inode
->i_atime
= fattr
->atime
;
313 inode
->i_mtime
= fattr
->mtime
;
314 inode
->i_ctime
= fattr
->ctime
;
315 if (fattr
->valid
& NFS_ATTR_FATTR_V4
)
316 nfsi
->change_attr
= fattr
->change_attr
;
317 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
318 inode
->i_nlink
= fattr
->nlink
;
319 inode
->i_uid
= fattr
->uid
;
320 inode
->i_gid
= fattr
->gid
;
321 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
323 * report the blocks in 512byte units
325 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
327 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
329 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
330 nfsi
->attrtimeo_timestamp
= now
;
331 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
332 nfsi
->access_cache
= RB_ROOT
;
334 unlock_new_inode(inode
);
336 nfs_refresh_inode(inode
, fattr
);
337 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
339 (long long)NFS_FILEID(inode
),
340 atomic_read(&inode
->i_count
));
346 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
350 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
353 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
355 struct inode
*inode
= dentry
->d_inode
;
356 struct nfs_fattr fattr
;
359 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
361 /* skip mode change if it's just for clearing setuid/setgid */
362 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
363 attr
->ia_valid
&= ~ATTR_MODE
;
365 if (attr
->ia_valid
& ATTR_SIZE
) {
366 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
367 attr
->ia_valid
&= ~ATTR_SIZE
;
370 /* Optimization: if the end result is no change, don't RPC */
371 attr
->ia_valid
&= NFS_VALID_ATTRS
;
372 if (attr
->ia_valid
== 0)
376 /* Write all dirty data */
377 if (S_ISREG(inode
->i_mode
)) {
378 filemap_write_and_wait(inode
->i_mapping
);
382 * Return any delegations if we're going to change ACLs
384 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
385 nfs_inode_return_delegation(inode
);
386 error
= NFS_PROTO(inode
)->setattr(dentry
, &fattr
, attr
);
388 nfs_refresh_inode(inode
, &fattr
);
394 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
395 * @inode: pointer to struct inode
396 * @attr: pointer to struct iattr
398 * Note: we do this in the *proc.c in order to ensure that
399 * it works for things like exclusive creates too.
401 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
403 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
404 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
405 int mode
= attr
->ia_mode
& S_IALLUGO
;
406 mode
|= inode
->i_mode
& ~S_IALLUGO
;
407 inode
->i_mode
= mode
;
409 if ((attr
->ia_valid
& ATTR_UID
) != 0)
410 inode
->i_uid
= attr
->ia_uid
;
411 if ((attr
->ia_valid
& ATTR_GID
) != 0)
412 inode
->i_gid
= attr
->ia_gid
;
413 spin_lock(&inode
->i_lock
);
414 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
415 spin_unlock(&inode
->i_lock
);
417 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
418 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
419 inode
->i_size
= attr
->ia_size
;
420 vmtruncate(inode
, attr
->ia_size
);
424 static int nfs_wait_schedule(void *word
)
426 if (signal_pending(current
))
433 * Wait for the inode to get unlocked.
435 static int nfs_wait_on_inode(struct inode
*inode
)
437 struct nfs_inode
*nfsi
= NFS_I(inode
);
440 error
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_REVALIDATING
,
441 nfs_wait_schedule
, TASK_KILLABLE
);
446 static void nfs_wake_up_inode(struct inode
*inode
)
448 struct nfs_inode
*nfsi
= NFS_I(inode
);
450 clear_bit(NFS_INO_REVALIDATING
, &nfsi
->flags
);
451 smp_mb__after_clear_bit();
452 wake_up_bit(&nfsi
->flags
, NFS_INO_REVALIDATING
);
455 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
457 struct inode
*inode
= dentry
->d_inode
;
458 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
462 * Flush out writes to the server in order to update c/mtime.
464 * Hold the i_mutex to suspend application writes temporarily;
465 * this prevents long-running writing applications from blocking
468 if (S_ISREG(inode
->i_mode
)) {
469 mutex_lock(&inode
->i_mutex
);
470 nfs_wb_nocommit(inode
);
471 mutex_unlock(&inode
->i_mutex
);
475 * We may force a getattr if the user cares about atime.
477 * Note that we only have to check the vfsmount flags here:
478 * - NFS always sets S_NOATIME by so checking it would give a
480 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
481 * no point in checking those.
483 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
484 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
488 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
490 err
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
492 generic_fillattr(inode
, stat
);
493 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
498 static struct nfs_open_context
*alloc_nfs_open_context(struct vfsmount
*mnt
, struct dentry
*dentry
, struct rpc_cred
*cred
)
500 struct nfs_open_context
*ctx
;
502 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
504 ctx
->path
.dentry
= dget(dentry
);
505 ctx
->path
.mnt
= mntget(mnt
);
506 ctx
->cred
= get_rpccred(cred
);
508 ctx
->lockowner
= current
->files
;
512 atomic_set(&ctx
->count
, 1);
517 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
520 atomic_inc(&ctx
->count
);
524 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int wait
)
531 inode
= ctx
->path
.dentry
->d_inode
;
532 if (!atomic_dec_and_lock(&ctx
->count
, &inode
->i_lock
))
534 list_del(&ctx
->list
);
535 spin_unlock(&inode
->i_lock
);
536 if (ctx
->state
!= NULL
) {
538 nfs4_close_sync(&ctx
->path
, ctx
->state
, ctx
->mode
);
540 nfs4_close_state(&ctx
->path
, ctx
->state
, ctx
->mode
);
542 if (ctx
->cred
!= NULL
)
543 put_rpccred(ctx
->cred
);
544 path_put(&ctx
->path
);
548 void put_nfs_open_context(struct nfs_open_context
*ctx
)
550 __put_nfs_open_context(ctx
, 0);
553 static void put_nfs_open_context_sync(struct nfs_open_context
*ctx
)
555 __put_nfs_open_context(ctx
, 1);
559 * Ensure that mmap has a recent RPC credential for use when writing out
562 static void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
564 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
565 struct nfs_inode
*nfsi
= NFS_I(inode
);
567 filp
->private_data
= get_nfs_open_context(ctx
);
568 spin_lock(&inode
->i_lock
);
569 list_add(&ctx
->list
, &nfsi
->open_files
);
570 spin_unlock(&inode
->i_lock
);
574 * Given an inode, search for an open context with the desired characteristics
576 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, int mode
)
578 struct nfs_inode
*nfsi
= NFS_I(inode
);
579 struct nfs_open_context
*pos
, *ctx
= NULL
;
581 spin_lock(&inode
->i_lock
);
582 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
583 if (cred
!= NULL
&& pos
->cred
!= cred
)
585 if ((pos
->mode
& mode
) == mode
) {
586 ctx
= get_nfs_open_context(pos
);
590 spin_unlock(&inode
->i_lock
);
594 static void nfs_file_clear_open_context(struct file
*filp
)
596 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
597 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
600 filp
->private_data
= NULL
;
601 spin_lock(&inode
->i_lock
);
602 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
603 spin_unlock(&inode
->i_lock
);
604 put_nfs_open_context_sync(ctx
);
609 * These allocate and release file read/write context information.
611 int nfs_open(struct inode
*inode
, struct file
*filp
)
613 struct nfs_open_context
*ctx
;
614 struct rpc_cred
*cred
;
616 cred
= rpc_lookup_cred();
618 return PTR_ERR(cred
);
619 ctx
= alloc_nfs_open_context(filp
->f_path
.mnt
, filp
->f_path
.dentry
, cred
);
623 ctx
->mode
= filp
->f_mode
;
624 nfs_file_set_open_context(filp
, ctx
);
625 put_nfs_open_context(ctx
);
629 int nfs_release(struct inode
*inode
, struct file
*filp
)
631 nfs_file_clear_open_context(filp
);
636 * This function is called whenever some part of NFS notices that
637 * the cached attributes have to be refreshed.
640 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
642 int status
= -ESTALE
;
643 struct nfs_fattr fattr
;
644 struct nfs_inode
*nfsi
= NFS_I(inode
);
646 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
647 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
649 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
651 if (is_bad_inode(inode
))
653 if (NFS_STALE(inode
))
656 status
= nfs_wait_on_inode(inode
);
661 if (NFS_STALE(inode
))
664 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), &fattr
);
666 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
668 (long long)NFS_FILEID(inode
), status
);
669 if (status
== -ESTALE
) {
670 nfs_zap_caches(inode
);
671 if (!S_ISDIR(inode
->i_mode
))
672 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
677 spin_lock(&inode
->i_lock
);
678 status
= nfs_update_inode(inode
, &fattr
);
680 spin_unlock(&inode
->i_lock
);
681 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
683 (long long)NFS_FILEID(inode
), status
);
686 spin_unlock(&inode
->i_lock
);
688 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
689 nfs_zap_acl_cache(inode
);
691 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
693 (long long)NFS_FILEID(inode
));
696 nfs_wake_up_inode(inode
);
703 int nfs_attribute_timeout(struct inode
*inode
)
705 struct nfs_inode
*nfsi
= NFS_I(inode
);
707 if (nfs_have_delegation(inode
, FMODE_READ
))
710 * Special case: if the attribute timeout is set to 0, then always
711 * treat the cache as having expired (unless holding
714 if (nfsi
->attrtimeo
== 0)
716 return !time_in_range(jiffies
, nfsi
->read_cache_jiffies
, nfsi
->read_cache_jiffies
+ nfsi
->attrtimeo
);
720 * nfs_revalidate_inode - Revalidate the inode attributes
721 * @server - pointer to nfs_server struct
722 * @inode - pointer to inode struct
724 * Updates inode attribute information by retrieving the data from the server.
726 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
728 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
729 && !nfs_attribute_timeout(inode
))
730 return NFS_STALE(inode
) ? -ESTALE
: 0;
731 return __nfs_revalidate_inode(server
, inode
);
734 static int nfs_invalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
736 struct nfs_inode
*nfsi
= NFS_I(inode
);
738 if (mapping
->nrpages
!= 0) {
739 int ret
= invalidate_inode_pages2(mapping
);
743 spin_lock(&inode
->i_lock
);
744 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
745 if (S_ISDIR(inode
->i_mode
))
746 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
747 spin_unlock(&inode
->i_lock
);
748 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
749 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
750 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
754 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
758 mutex_lock(&inode
->i_mutex
);
759 if (NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_DATA
) {
760 ret
= nfs_sync_mapping(mapping
);
762 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
764 mutex_unlock(&inode
->i_mutex
);
769 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
770 * @inode - pointer to host inode
771 * @mapping - pointer to mapping
773 int nfs_revalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
775 struct nfs_inode
*nfsi
= NFS_I(inode
);
778 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
779 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
780 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
784 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
785 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
791 * nfs_revalidate_mapping - Revalidate the pagecache
792 * @inode - pointer to host inode
793 * @mapping - pointer to mapping
795 * This version of the function will take the inode->i_mutex and attempt to
796 * flush out all dirty data if it needs to invalidate the page cache.
798 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
800 struct nfs_inode
*nfsi
= NFS_I(inode
);
803 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
804 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
805 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
809 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
810 ret
= nfs_invalidate_mapping(inode
, mapping
);
815 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
817 struct nfs_inode
*nfsi
= NFS_I(inode
);
819 if ((fattr
->valid
& NFS_ATTR_WCC_V4
) != 0 &&
820 nfsi
->change_attr
== fattr
->pre_change_attr
) {
821 nfsi
->change_attr
= fattr
->change_attr
;
822 if (S_ISDIR(inode
->i_mode
))
823 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
825 /* If we have atomic WCC data, we may update some attributes */
826 if ((fattr
->valid
& NFS_ATTR_WCC
) != 0) {
827 if (timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
))
828 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
829 if (timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
830 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
831 if (S_ISDIR(inode
->i_mode
))
832 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
834 if (inode
->i_size
== nfs_size_to_loff_t(fattr
->pre_size
) &&
836 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
841 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
842 * @inode - pointer to inode
843 * @fattr - updated attributes
845 * Verifies the attribute cache. If we have just changed the attributes,
846 * so that fattr carries weak cache consistency data, then it may
847 * also update the ctime/mtime/change_attribute.
849 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
851 struct nfs_inode
*nfsi
= NFS_I(inode
);
852 loff_t cur_size
, new_isize
;
853 unsigned long invalid
= 0;
856 /* Has the inode gone and changed behind our back? */
857 if (nfsi
->fileid
!= fattr
->fileid
858 || (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
862 /* Do atomic weak cache consistency updates */
863 nfs_wcc_update_inode(inode
, fattr
);
865 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
866 nfsi
->change_attr
!= fattr
->change_attr
)
867 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
869 /* Verify a few of the more important attributes */
870 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
871 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
873 cur_size
= i_size_read(inode
);
874 new_isize
= nfs_size_to_loff_t(fattr
->size
);
875 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
876 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
878 /* Have any file permissions changed? */
879 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)
880 || inode
->i_uid
!= fattr
->uid
881 || inode
->i_gid
!= fattr
->gid
)
882 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
884 /* Has the link count changed? */
885 if (inode
->i_nlink
!= fattr
->nlink
)
886 invalid
|= NFS_INO_INVALID_ATTR
;
888 if (!timespec_equal(&inode
->i_atime
, &fattr
->atime
))
889 invalid
|= NFS_INO_INVALID_ATIME
;
892 nfsi
->cache_validity
|= invalid
;
894 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
895 | NFS_INO_INVALID_ATIME
896 | NFS_INO_REVAL_PAGECACHE
);
898 nfsi
->read_cache_jiffies
= fattr
->time_start
;
903 * nfs_refresh_inode - try to update the inode attribute cache
904 * @inode - pointer to inode
905 * @fattr - updated attributes
907 * Check that an RPC call that returned attributes has not overlapped with
908 * other recent updates of the inode metadata, then decide whether it is
909 * safe to do a full update of the inode attributes, or whether just to
910 * call nfs_check_inode_attributes.
912 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
914 struct nfs_inode
*nfsi
= NFS_I(inode
);
917 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
919 spin_lock(&inode
->i_lock
);
920 if (time_after(fattr
->time_start
, nfsi
->last_updated
))
921 status
= nfs_update_inode(inode
, fattr
);
923 status
= nfs_check_inode_attributes(inode
, fattr
);
925 spin_unlock(&inode
->i_lock
);
930 * nfs_post_op_update_inode - try to update the inode attribute cache
931 * @inode - pointer to inode
932 * @fattr - updated attributes
934 * After an operation that has changed the inode metadata, mark the
935 * attribute cache as being invalid, then try to update it.
937 * NB: if the server didn't return any post op attributes, this
938 * function will force the retrieval of attributes before the next
939 * NFS request. Thus it should be used only for operations that
940 * are expected to change one or more attributes, to avoid
941 * unnecessary NFS requests and trips through nfs_update_inode().
943 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
945 struct nfs_inode
*nfsi
= NFS_I(inode
);
947 spin_lock(&inode
->i_lock
);
948 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
949 if (S_ISDIR(inode
->i_mode
))
950 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
951 spin_unlock(&inode
->i_lock
);
952 return nfs_refresh_inode(inode
, fattr
);
956 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
957 * @inode - pointer to inode
958 * @fattr - updated attributes
960 * After an operation that has changed the inode metadata, mark the
961 * attribute cache as being invalid, then try to update it. Fake up
962 * weak cache consistency data, if none exist.
964 * This function is mainly designed to be used by the ->write_done() functions.
966 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
968 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
969 (fattr
->valid
& NFS_ATTR_WCC_V4
) == 0) {
970 fattr
->pre_change_attr
= NFS_I(inode
)->change_attr
;
971 fattr
->valid
|= NFS_ATTR_WCC_V4
;
973 if ((fattr
->valid
& NFS_ATTR_FATTR
) != 0 &&
974 (fattr
->valid
& NFS_ATTR_WCC
) == 0) {
975 memcpy(&fattr
->pre_ctime
, &inode
->i_ctime
, sizeof(fattr
->pre_ctime
));
976 memcpy(&fattr
->pre_mtime
, &inode
->i_mtime
, sizeof(fattr
->pre_mtime
));
977 fattr
->pre_size
= inode
->i_size
;
978 fattr
->valid
|= NFS_ATTR_WCC
;
980 return nfs_post_op_update_inode(inode
, fattr
);
984 * Many nfs protocol calls return the new file attributes after
985 * an operation. Here we update the inode to reflect the state
986 * of the server's inode.
988 * This is a bit tricky because we have to make sure all dirty pages
989 * have been sent off to the server before calling invalidate_inode_pages.
990 * To make sure no other process adds more write requests while we try
991 * our best to flush them, we make them sleep during the attribute refresh.
993 * A very similar scenario holds for the dir cache.
995 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
997 struct nfs_server
*server
;
998 struct nfs_inode
*nfsi
= NFS_I(inode
);
999 loff_t cur_isize
, new_isize
;
1000 unsigned long invalid
= 0;
1001 unsigned long now
= jiffies
;
1003 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1004 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1005 atomic_read(&inode
->i_count
), fattr
->valid
);
1007 if (nfsi
->fileid
!= fattr
->fileid
)
1011 * Make sure the inode's type hasn't changed.
1013 if ((inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
1016 server
= NFS_SERVER(inode
);
1017 /* Update the fsid? */
1018 if (S_ISDIR(inode
->i_mode
) &&
1019 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
1020 !test_bit(NFS_INO_MOUNTPOINT
, &nfsi
->flags
))
1021 server
->fsid
= fattr
->fsid
;
1024 * Update the read time so we don't revalidate too often.
1026 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1028 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ATIME
1029 | NFS_INO_REVAL_PAGECACHE
);
1031 /* Do atomic weak cache consistency updates */
1032 nfs_wcc_update_inode(inode
, fattr
);
1034 /* More cache consistency checks */
1035 if (!(fattr
->valid
& NFS_ATTR_FATTR_V4
)) {
1036 /* NFSv2/v3: Check if the mtime agrees */
1037 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
1038 dprintk("NFS: mtime change on server for file %s/%ld\n",
1039 inode
->i_sb
->s_id
, inode
->i_ino
);
1040 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1041 if (S_ISDIR(inode
->i_mode
))
1042 nfs_force_lookup_revalidate(inode
);
1044 /* If ctime has changed we should definitely clear access+acl caches */
1045 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
))
1046 invalid
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1047 } else if (nfsi
->change_attr
!= fattr
->change_attr
) {
1048 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1049 inode
->i_sb
->s_id
, inode
->i_ino
);
1050 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1051 if (S_ISDIR(inode
->i_mode
))
1052 nfs_force_lookup_revalidate(inode
);
1055 /* Check if our cached file size is stale */
1056 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1057 cur_isize
= i_size_read(inode
);
1058 if (new_isize
!= cur_isize
) {
1059 /* Do we perhaps have any outstanding writes, or has
1060 * the file grown beyond our last write? */
1061 if (nfsi
->npages
== 0 || new_isize
> cur_isize
) {
1062 inode
->i_size
= new_isize
;
1063 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1065 dprintk("NFS: isize change on server for file %s/%ld\n",
1066 inode
->i_sb
->s_id
, inode
->i_ino
);
1070 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1071 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1072 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1073 nfsi
->change_attr
= fattr
->change_attr
;
1075 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
) ||
1076 inode
->i_uid
!= fattr
->uid
||
1077 inode
->i_gid
!= fattr
->gid
)
1078 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1080 inode
->i_mode
= fattr
->mode
;
1081 inode
->i_nlink
= fattr
->nlink
;
1082 inode
->i_uid
= fattr
->uid
;
1083 inode
->i_gid
= fattr
->gid
;
1085 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
1087 * report the blocks in 512byte units
1089 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1091 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1094 /* Update attrtimeo value if we're out of the unstable period */
1095 if (invalid
& NFS_INO_INVALID_ATTR
) {
1096 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1097 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1098 nfsi
->attrtimeo_timestamp
= now
;
1099 nfsi
->last_updated
= now
;
1101 if (!time_in_range(now
, nfsi
->attrtimeo_timestamp
, nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
1102 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1103 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1104 nfsi
->attrtimeo_timestamp
= now
;
1107 * Avoid jiffy wraparound issues with nfsi->last_updated
1109 if (!time_in_range(nfsi
->last_updated
, nfsi
->read_cache_jiffies
, now
))
1110 nfsi
->last_updated
= nfsi
->read_cache_jiffies
;
1112 invalid
&= ~NFS_INO_INVALID_ATTR
;
1113 /* Don't invalidate the data if we were to blame */
1114 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1115 || S_ISLNK(inode
->i_mode
)))
1116 invalid
&= ~NFS_INO_INVALID_DATA
;
1117 if (!nfs_have_delegation(inode
, FMODE_READ
) ||
1118 (nfsi
->cache_validity
& NFS_INO_REVAL_FORCED
))
1119 nfsi
->cache_validity
|= invalid
;
1120 nfsi
->cache_validity
&= ~NFS_INO_REVAL_FORCED
;
1125 * Big trouble! The inode has become a different object.
1127 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1128 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1131 * No need to worry about unhashing the dentry, as the
1132 * lookup validation will know that the inode is bad.
1133 * (But we fall through to invalidate the caches.)
1135 nfs_invalidate_inode(inode
);
1139 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1140 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1141 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1142 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1147 #ifdef CONFIG_NFS_V4
1150 * Clean out any remaining NFSv4 state that might be left over due
1151 * to open() calls that passed nfs_atomic_lookup, but failed to call
1154 void nfs4_clear_inode(struct inode
*inode
)
1156 /* If we are holding a delegation, return it! */
1157 nfs_inode_return_delegation_noreclaim(inode
);
1158 /* First call standard NFS clear_inode() code */
1159 nfs_clear_inode(inode
);
1163 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1165 struct nfs_inode
*nfsi
;
1166 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1170 nfsi
->cache_validity
= 0UL;
1171 #ifdef CONFIG_NFS_V3_ACL
1172 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1173 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1175 #ifdef CONFIG_NFS_V4
1176 nfsi
->nfs4_acl
= NULL
;
1177 #endif /* CONFIG_NFS_V4 */
1178 return &nfsi
->vfs_inode
;
1181 void nfs_destroy_inode(struct inode
*inode
)
1183 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1186 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1188 #ifdef CONFIG_NFS_V4
1189 INIT_LIST_HEAD(&nfsi
->open_states
);
1190 nfsi
->delegation
= NULL
;
1191 nfsi
->delegation_state
= 0;
1192 init_rwsem(&nfsi
->rwsem
);
1196 static void init_once(struct kmem_cache
* cachep
, void *foo
)
1198 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1200 inode_init_once(&nfsi
->vfs_inode
);
1201 INIT_LIST_HEAD(&nfsi
->open_files
);
1202 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1203 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1204 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1207 atomic_set(&nfsi
->silly_count
, 1);
1208 INIT_HLIST_HEAD(&nfsi
->silly_list
);
1209 init_waitqueue_head(&nfsi
->waitqueue
);
1210 nfs4_init_once(nfsi
);
1213 static int __init
nfs_init_inodecache(void)
1215 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1216 sizeof(struct nfs_inode
),
1217 0, (SLAB_RECLAIM_ACCOUNT
|
1220 if (nfs_inode_cachep
== NULL
)
1226 static void nfs_destroy_inodecache(void)
1228 kmem_cache_destroy(nfs_inode_cachep
);
1231 struct workqueue_struct
*nfsiod_workqueue
;
1234 * start up the nfsiod workqueue
1236 static int nfsiod_start(void)
1238 struct workqueue_struct
*wq
;
1239 dprintk("RPC: creating workqueue nfsiod\n");
1240 wq
= create_singlethread_workqueue("nfsiod");
1243 nfsiod_workqueue
= wq
;
1248 * Destroy the nfsiod workqueue
1250 static void nfsiod_stop(void)
1252 struct workqueue_struct
*wq
;
1254 wq
= nfsiod_workqueue
;
1257 nfsiod_workqueue
= NULL
;
1258 destroy_workqueue(wq
);
1264 static int __init
init_nfs_fs(void)
1268 err
= nfsiod_start();
1272 err
= nfs_fs_proc_init();
1276 err
= nfs_init_nfspagecache();
1280 err
= nfs_init_inodecache();
1284 err
= nfs_init_readpagecache();
1288 err
= nfs_init_writepagecache();
1292 err
= nfs_init_directcache();
1296 #ifdef CONFIG_PROC_FS
1297 rpc_proc_register(&nfs_rpcstat
);
1299 if ((err
= register_nfs_fs()) != 0)
1303 #ifdef CONFIG_PROC_FS
1304 rpc_proc_unregister("nfs");
1306 nfs_destroy_directcache();
1308 nfs_destroy_writepagecache();
1310 nfs_destroy_readpagecache();
1312 nfs_destroy_inodecache();
1314 nfs_destroy_nfspagecache();
1323 static void __exit
exit_nfs_fs(void)
1325 nfs_destroy_directcache();
1326 nfs_destroy_writepagecache();
1327 nfs_destroy_readpagecache();
1328 nfs_destroy_inodecache();
1329 nfs_destroy_nfspagecache();
1330 #ifdef CONFIG_PROC_FS
1331 rpc_proc_unregister("nfs");
1333 unregister_nfs_fs();
1338 /* Not quite true; I just maintain it */
1339 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1340 MODULE_LICENSE("GPL");
1341 module_param(enable_ino64
, bool, 0644);
1343 module_init(init_nfs_fs
)
1344 module_exit(exit_nfs_fs
)