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_FLAGS(inode
));
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 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_FLAGS(inode
));
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
;
303 } else if (S_ISLNK(inode
->i_mode
))
304 inode
->i_op
= &nfs_symlink_inode_operations
;
306 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
308 nfsi
->read_cache_jiffies
= fattr
->time_start
;
309 nfsi
->last_updated
= now
;
310 nfsi
->cache_change_attribute
= now
;
311 inode
->i_atime
= fattr
->atime
;
312 inode
->i_mtime
= fattr
->mtime
;
313 inode
->i_ctime
= fattr
->ctime
;
314 if (fattr
->valid
& NFS_ATTR_FATTR_V4
)
315 nfsi
->change_attr
= fattr
->change_attr
;
316 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
317 inode
->i_nlink
= fattr
->nlink
;
318 inode
->i_uid
= fattr
->uid
;
319 inode
->i_gid
= fattr
->gid
;
320 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
322 * report the blocks in 512byte units
324 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
326 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
328 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
329 nfsi
->attrtimeo_timestamp
= now
;
330 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
331 nfsi
->access_cache
= RB_ROOT
;
333 unlock_new_inode(inode
);
335 nfs_refresh_inode(inode
, fattr
);
336 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
338 (long long)NFS_FILEID(inode
),
339 atomic_read(&inode
->i_count
));
345 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
349 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
352 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
354 struct inode
*inode
= dentry
->d_inode
;
355 struct nfs_fattr fattr
;
358 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
360 /* skip mode change if it's just for clearing setuid/setgid */
361 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
362 attr
->ia_valid
&= ~ATTR_MODE
;
364 if (attr
->ia_valid
& ATTR_SIZE
) {
365 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
366 attr
->ia_valid
&= ~ATTR_SIZE
;
369 /* Optimization: if the end result is no change, don't RPC */
370 attr
->ia_valid
&= NFS_VALID_ATTRS
;
371 if (attr
->ia_valid
== 0)
375 /* Write all dirty data */
376 if (S_ISREG(inode
->i_mode
)) {
377 filemap_write_and_wait(inode
->i_mapping
);
381 * Return any delegations if we're going to change ACLs
383 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
384 nfs_inode_return_delegation(inode
);
385 error
= NFS_PROTO(inode
)->setattr(dentry
, &fattr
, attr
);
387 nfs_refresh_inode(inode
, &fattr
);
393 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
394 * @inode: pointer to struct inode
395 * @attr: pointer to struct iattr
397 * Note: we do this in the *proc.c in order to ensure that
398 * it works for things like exclusive creates too.
400 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
402 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
403 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
404 int mode
= attr
->ia_mode
& S_IALLUGO
;
405 mode
|= inode
->i_mode
& ~S_IALLUGO
;
406 inode
->i_mode
= mode
;
408 if ((attr
->ia_valid
& ATTR_UID
) != 0)
409 inode
->i_uid
= attr
->ia_uid
;
410 if ((attr
->ia_valid
& ATTR_GID
) != 0)
411 inode
->i_gid
= attr
->ia_gid
;
412 spin_lock(&inode
->i_lock
);
413 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
414 spin_unlock(&inode
->i_lock
);
416 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
417 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
418 inode
->i_size
= attr
->ia_size
;
419 vmtruncate(inode
, attr
->ia_size
);
423 static int nfs_wait_schedule(void *word
)
425 if (signal_pending(current
))
432 * Wait for the inode to get unlocked.
434 static int nfs_wait_on_inode(struct inode
*inode
)
436 struct nfs_inode
*nfsi
= NFS_I(inode
);
439 error
= wait_on_bit_lock(&nfsi
->flags
, NFS_INO_REVALIDATING
,
440 nfs_wait_schedule
, TASK_KILLABLE
);
445 static void nfs_wake_up_inode(struct inode
*inode
)
447 struct nfs_inode
*nfsi
= NFS_I(inode
);
449 clear_bit(NFS_INO_REVALIDATING
, &nfsi
->flags
);
450 smp_mb__after_clear_bit();
451 wake_up_bit(&nfsi
->flags
, NFS_INO_REVALIDATING
);
454 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
456 struct inode
*inode
= dentry
->d_inode
;
457 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
460 /* Flush out writes to the server in order to update c/mtime */
461 if (S_ISREG(inode
->i_mode
))
462 nfs_wb_nocommit(inode
);
465 * We may force a getattr if the user cares about atime.
467 * Note that we only have to check the vfsmount flags here:
468 * - NFS always sets S_NOATIME by so checking it would give a
470 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
471 * no point in checking those.
473 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
474 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
478 err
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
480 err
= nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
482 generic_fillattr(inode
, stat
);
483 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
488 static struct nfs_open_context
*alloc_nfs_open_context(struct vfsmount
*mnt
, struct dentry
*dentry
, struct rpc_cred
*cred
)
490 struct nfs_open_context
*ctx
;
492 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
494 ctx
->path
.dentry
= dget(dentry
);
495 ctx
->path
.mnt
= mntget(mnt
);
496 ctx
->cred
= get_rpccred(cred
);
498 ctx
->lockowner
= current
->files
;
501 atomic_set(&ctx
->count
, 1);
506 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
509 atomic_inc(&ctx
->count
);
513 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int wait
)
515 struct inode
*inode
= ctx
->path
.dentry
->d_inode
;
517 if (!atomic_dec_and_lock(&ctx
->count
, &inode
->i_lock
))
519 list_del(&ctx
->list
);
520 spin_unlock(&inode
->i_lock
);
521 if (ctx
->state
!= NULL
) {
523 nfs4_close_sync(&ctx
->path
, ctx
->state
, ctx
->mode
);
525 nfs4_close_state(&ctx
->path
, ctx
->state
, ctx
->mode
);
527 if (ctx
->cred
!= NULL
)
528 put_rpccred(ctx
->cred
);
529 dput(ctx
->path
.dentry
);
530 mntput(ctx
->path
.mnt
);
534 void put_nfs_open_context(struct nfs_open_context
*ctx
)
536 __put_nfs_open_context(ctx
, 0);
539 static void put_nfs_open_context_sync(struct nfs_open_context
*ctx
)
541 __put_nfs_open_context(ctx
, 1);
545 * Ensure that mmap has a recent RPC credential for use when writing out
548 static void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
550 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
551 struct nfs_inode
*nfsi
= NFS_I(inode
);
553 filp
->private_data
= get_nfs_open_context(ctx
);
554 spin_lock(&inode
->i_lock
);
555 list_add(&ctx
->list
, &nfsi
->open_files
);
556 spin_unlock(&inode
->i_lock
);
560 * Given an inode, search for an open context with the desired characteristics
562 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, int mode
)
564 struct nfs_inode
*nfsi
= NFS_I(inode
);
565 struct nfs_open_context
*pos
, *ctx
= NULL
;
567 spin_lock(&inode
->i_lock
);
568 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
569 if (cred
!= NULL
&& pos
->cred
!= cred
)
571 if ((pos
->mode
& mode
) == mode
) {
572 ctx
= get_nfs_open_context(pos
);
576 spin_unlock(&inode
->i_lock
);
580 static void nfs_file_clear_open_context(struct file
*filp
)
582 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
583 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
586 filp
->private_data
= NULL
;
587 spin_lock(&inode
->i_lock
);
588 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
589 spin_unlock(&inode
->i_lock
);
590 put_nfs_open_context_sync(ctx
);
595 * These allocate and release file read/write context information.
597 int nfs_open(struct inode
*inode
, struct file
*filp
)
599 struct nfs_open_context
*ctx
;
600 struct rpc_cred
*cred
;
602 cred
= rpcauth_lookupcred(NFS_CLIENT(inode
)->cl_auth
, 0);
604 return PTR_ERR(cred
);
605 ctx
= alloc_nfs_open_context(filp
->f_path
.mnt
, filp
->f_path
.dentry
, cred
);
609 ctx
->mode
= filp
->f_mode
;
610 nfs_file_set_open_context(filp
, ctx
);
611 put_nfs_open_context(ctx
);
615 int nfs_release(struct inode
*inode
, struct file
*filp
)
617 nfs_file_clear_open_context(filp
);
622 * This function is called whenever some part of NFS notices that
623 * the cached attributes have to be refreshed.
626 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
628 int status
= -ESTALE
;
629 struct nfs_fattr fattr
;
630 struct nfs_inode
*nfsi
= NFS_I(inode
);
632 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Ld)\n",
633 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
635 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
637 if (is_bad_inode(inode
))
639 if (NFS_STALE(inode
))
642 status
= nfs_wait_on_inode(inode
);
647 if (NFS_STALE(inode
))
650 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), &fattr
);
652 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
654 (long long)NFS_FILEID(inode
), status
);
655 if (status
== -ESTALE
) {
656 nfs_zap_caches(inode
);
657 if (!S_ISDIR(inode
->i_mode
))
658 set_bit(NFS_INO_STALE
, &NFS_FLAGS(inode
));
663 spin_lock(&inode
->i_lock
);
664 status
= nfs_update_inode(inode
, &fattr
);
666 spin_unlock(&inode
->i_lock
);
667 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
669 (long long)NFS_FILEID(inode
), status
);
672 spin_unlock(&inode
->i_lock
);
674 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
675 nfs_zap_acl_cache(inode
);
677 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) revalidation complete\n",
679 (long long)NFS_FILEID(inode
));
682 nfs_wake_up_inode(inode
);
689 int nfs_attribute_timeout(struct inode
*inode
)
691 struct nfs_inode
*nfsi
= NFS_I(inode
);
693 if (nfs_have_delegation(inode
, FMODE_READ
))
695 return !time_in_range(jiffies
, nfsi
->read_cache_jiffies
, nfsi
->read_cache_jiffies
+ nfsi
->attrtimeo
);
699 * nfs_revalidate_inode - Revalidate the inode attributes
700 * @server - pointer to nfs_server struct
701 * @inode - pointer to inode struct
703 * Updates inode attribute information by retrieving the data from the server.
705 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
707 if (!(NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATTR
)
708 && !nfs_attribute_timeout(inode
))
709 return NFS_STALE(inode
) ? -ESTALE
: 0;
710 return __nfs_revalidate_inode(server
, inode
);
713 static int nfs_invalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
715 struct nfs_inode
*nfsi
= NFS_I(inode
);
717 if (mapping
->nrpages
!= 0) {
718 int ret
= invalidate_inode_pages2(mapping
);
722 spin_lock(&inode
->i_lock
);
723 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
724 if (S_ISDIR(inode
->i_mode
))
725 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
726 spin_unlock(&inode
->i_lock
);
727 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
728 dfprintk(PAGECACHE
, "NFS: (%s/%Ld) data cache invalidated\n",
729 inode
->i_sb
->s_id
, (long long)NFS_FILEID(inode
));
733 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
737 mutex_lock(&inode
->i_mutex
);
738 if (NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_DATA
) {
739 ret
= nfs_sync_mapping(mapping
);
741 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
743 mutex_unlock(&inode
->i_mutex
);
748 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
749 * @inode - pointer to host inode
750 * @mapping - pointer to mapping
752 int nfs_revalidate_mapping_nolock(struct inode
*inode
, struct address_space
*mapping
)
754 struct nfs_inode
*nfsi
= NFS_I(inode
);
757 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
758 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
759 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
763 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
764 ret
= nfs_invalidate_mapping_nolock(inode
, mapping
);
770 * nfs_revalidate_mapping - Revalidate the pagecache
771 * @inode - pointer to host inode
772 * @mapping - pointer to mapping
774 * This version of the function will take the inode->i_mutex and attempt to
775 * flush out all dirty data if it needs to invalidate the page cache.
777 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
779 struct nfs_inode
*nfsi
= NFS_I(inode
);
782 if ((nfsi
->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
783 || nfs_attribute_timeout(inode
) || NFS_STALE(inode
)) {
784 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
788 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
789 ret
= nfs_invalidate_mapping(inode
, mapping
);
794 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
796 struct nfs_inode
*nfsi
= NFS_I(inode
);
798 if ((fattr
->valid
& NFS_ATTR_WCC_V4
) != 0 &&
799 nfsi
->change_attr
== fattr
->pre_change_attr
) {
800 nfsi
->change_attr
= fattr
->change_attr
;
801 if (S_ISDIR(inode
->i_mode
))
802 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
804 /* If we have atomic WCC data, we may update some attributes */
805 if ((fattr
->valid
& NFS_ATTR_WCC
) != 0) {
806 if (timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
))
807 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
808 if (timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
809 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
810 if (S_ISDIR(inode
->i_mode
))
811 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
813 if (inode
->i_size
== fattr
->pre_size
&& nfsi
->npages
== 0)
814 inode
->i_size
= fattr
->size
;
819 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
820 * @inode - pointer to inode
821 * @fattr - updated attributes
823 * Verifies the attribute cache. If we have just changed the attributes,
824 * so that fattr carries weak cache consistency data, then it may
825 * also update the ctime/mtime/change_attribute.
827 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
829 struct nfs_inode
*nfsi
= NFS_I(inode
);
830 loff_t cur_size
, new_isize
;
831 unsigned long invalid
= 0;
834 /* Has the inode gone and changed behind our back? */
835 if (nfsi
->fileid
!= fattr
->fileid
836 || (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
840 /* Do atomic weak cache consistency updates */
841 nfs_wcc_update_inode(inode
, fattr
);
843 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
844 nfsi
->change_attr
!= fattr
->change_attr
)
845 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
847 /* Verify a few of the more important attributes */
848 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
849 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
851 cur_size
= i_size_read(inode
);
852 new_isize
= nfs_size_to_loff_t(fattr
->size
);
853 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
854 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
856 /* Have any file permissions changed? */
857 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)
858 || inode
->i_uid
!= fattr
->uid
859 || inode
->i_gid
!= fattr
->gid
)
860 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
862 /* Has the link count changed? */
863 if (inode
->i_nlink
!= fattr
->nlink
)
864 invalid
|= NFS_INO_INVALID_ATTR
;
866 if (!timespec_equal(&inode
->i_atime
, &fattr
->atime
))
867 invalid
|= NFS_INO_INVALID_ATIME
;
870 nfsi
->cache_validity
|= invalid
;
872 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
873 | NFS_INO_INVALID_ATIME
874 | NFS_INO_REVAL_PAGECACHE
);
876 nfsi
->read_cache_jiffies
= fattr
->time_start
;
881 * nfs_refresh_inode - try to update the inode attribute cache
882 * @inode - pointer to inode
883 * @fattr - updated attributes
885 * Check that an RPC call that returned attributes has not overlapped with
886 * other recent updates of the inode metadata, then decide whether it is
887 * safe to do a full update of the inode attributes, or whether just to
888 * call nfs_check_inode_attributes.
890 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
892 struct nfs_inode
*nfsi
= NFS_I(inode
);
895 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
897 spin_lock(&inode
->i_lock
);
898 if (time_after(fattr
->time_start
, nfsi
->last_updated
))
899 status
= nfs_update_inode(inode
, fattr
);
901 status
= nfs_check_inode_attributes(inode
, fattr
);
903 spin_unlock(&inode
->i_lock
);
908 * nfs_post_op_update_inode - try to update the inode attribute cache
909 * @inode - pointer to inode
910 * @fattr - updated attributes
912 * After an operation that has changed the inode metadata, mark the
913 * attribute cache as being invalid, then try to update it.
915 * NB: if the server didn't return any post op attributes, this
916 * function will force the retrieval of attributes before the next
917 * NFS request. Thus it should be used only for operations that
918 * are expected to change one or more attributes, to avoid
919 * unnecessary NFS requests and trips through nfs_update_inode().
921 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
923 struct nfs_inode
*nfsi
= NFS_I(inode
);
925 spin_lock(&inode
->i_lock
);
926 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
927 if (S_ISDIR(inode
->i_mode
))
928 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
929 spin_unlock(&inode
->i_lock
);
930 return nfs_refresh_inode(inode
, fattr
);
934 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
935 * @inode - pointer to inode
936 * @fattr - updated attributes
938 * After an operation that has changed the inode metadata, mark the
939 * attribute cache as being invalid, then try to update it. Fake up
940 * weak cache consistency data, if none exist.
942 * This function is mainly designed to be used by the ->write_done() functions.
944 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
946 if ((fattr
->valid
& NFS_ATTR_FATTR_V4
) != 0 &&
947 (fattr
->valid
& NFS_ATTR_WCC_V4
) == 0) {
948 fattr
->pre_change_attr
= NFS_I(inode
)->change_attr
;
949 fattr
->valid
|= NFS_ATTR_WCC_V4
;
951 if ((fattr
->valid
& NFS_ATTR_FATTR
) != 0 &&
952 (fattr
->valid
& NFS_ATTR_WCC
) == 0) {
953 memcpy(&fattr
->pre_ctime
, &inode
->i_ctime
, sizeof(fattr
->pre_ctime
));
954 memcpy(&fattr
->pre_mtime
, &inode
->i_mtime
, sizeof(fattr
->pre_mtime
));
955 fattr
->pre_size
= inode
->i_size
;
956 fattr
->valid
|= NFS_ATTR_WCC
;
958 return nfs_post_op_update_inode(inode
, fattr
);
962 * Many nfs protocol calls return the new file attributes after
963 * an operation. Here we update the inode to reflect the state
964 * of the server's inode.
966 * This is a bit tricky because we have to make sure all dirty pages
967 * have been sent off to the server before calling invalidate_inode_pages.
968 * To make sure no other process adds more write requests while we try
969 * our best to flush them, we make them sleep during the attribute refresh.
971 * A very similar scenario holds for the dir cache.
973 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
975 struct nfs_server
*server
;
976 struct nfs_inode
*nfsi
= NFS_I(inode
);
977 loff_t cur_isize
, new_isize
;
978 unsigned long invalid
= 0;
979 unsigned long now
= jiffies
;
981 dfprintk(VFS
, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
982 __FUNCTION__
, inode
->i_sb
->s_id
, inode
->i_ino
,
983 atomic_read(&inode
->i_count
), fattr
->valid
);
985 if (nfsi
->fileid
!= fattr
->fileid
)
989 * Make sure the inode's type hasn't changed.
991 if ((inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
994 server
= NFS_SERVER(inode
);
995 /* Update the fsid? */
996 if (S_ISDIR(inode
->i_mode
)
997 && !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
))
998 server
->fsid
= fattr
->fsid
;
1001 * Update the read time so we don't revalidate too often.
1003 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1005 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ATIME
1006 | NFS_INO_REVAL_PAGECACHE
);
1008 /* Do atomic weak cache consistency updates */
1009 nfs_wcc_update_inode(inode
, fattr
);
1011 /* More cache consistency checks */
1012 if (!(fattr
->valid
& NFS_ATTR_FATTR_V4
)) {
1013 /* NFSv2/v3: Check if the mtime agrees */
1014 if (!timespec_equal(&inode
->i_mtime
, &fattr
->mtime
)) {
1015 dprintk("NFS: mtime change on server for file %s/%ld\n",
1016 inode
->i_sb
->s_id
, inode
->i_ino
);
1017 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1018 nfsi
->cache_change_attribute
= now
;
1020 /* If ctime has changed we should definitely clear access+acl caches */
1021 if (!timespec_equal(&inode
->i_ctime
, &fattr
->ctime
))
1022 invalid
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1023 } else if (nfsi
->change_attr
!= fattr
->change_attr
) {
1024 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1025 inode
->i_sb
->s_id
, inode
->i_ino
);
1026 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1027 nfsi
->cache_change_attribute
= now
;
1030 /* Check if our cached file size is stale */
1031 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1032 cur_isize
= i_size_read(inode
);
1033 if (new_isize
!= cur_isize
) {
1034 /* Do we perhaps have any outstanding writes, or has
1035 * the file grown beyond our last write? */
1036 if (nfsi
->npages
== 0 || new_isize
> cur_isize
) {
1037 inode
->i_size
= new_isize
;
1038 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1040 dprintk("NFS: isize change on server for file %s/%ld\n",
1041 inode
->i_sb
->s_id
, inode
->i_ino
);
1045 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1046 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1047 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1048 nfsi
->change_attr
= fattr
->change_attr
;
1050 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
) ||
1051 inode
->i_uid
!= fattr
->uid
||
1052 inode
->i_gid
!= fattr
->gid
)
1053 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1055 inode
->i_mode
= fattr
->mode
;
1056 inode
->i_nlink
= fattr
->nlink
;
1057 inode
->i_uid
= fattr
->uid
;
1058 inode
->i_gid
= fattr
->gid
;
1060 if (fattr
->valid
& (NFS_ATTR_FATTR_V3
| NFS_ATTR_FATTR_V4
)) {
1062 * report the blocks in 512byte units
1064 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1066 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1069 /* Update attrtimeo value if we're out of the unstable period */
1070 if (invalid
& NFS_INO_INVALID_ATTR
) {
1071 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1072 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1073 nfsi
->attrtimeo_timestamp
= now
;
1074 nfsi
->last_updated
= now
;
1076 if (!time_in_range(now
, nfsi
->attrtimeo_timestamp
, nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
1077 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1078 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1079 nfsi
->attrtimeo_timestamp
= now
;
1082 * Avoid jiffy wraparound issues with nfsi->last_updated
1084 if (!time_in_range(nfsi
->last_updated
, nfsi
->read_cache_jiffies
, now
))
1085 nfsi
->last_updated
= nfsi
->read_cache_jiffies
;
1087 invalid
&= ~NFS_INO_INVALID_ATTR
;
1088 /* Don't invalidate the data if we were to blame */
1089 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1090 || S_ISLNK(inode
->i_mode
)))
1091 invalid
&= ~NFS_INO_INVALID_DATA
;
1092 if (!nfs_have_delegation(inode
, FMODE_READ
) ||
1093 (nfsi
->cache_validity
& NFS_INO_REVAL_FORCED
))
1094 nfsi
->cache_validity
|= invalid
;
1095 nfsi
->cache_validity
&= ~NFS_INO_REVAL_FORCED
;
1100 * Big trouble! The inode has become a different object.
1102 printk(KERN_DEBUG
"%s: inode %ld mode changed, %07o to %07o\n",
1103 __FUNCTION__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1106 * No need to worry about unhashing the dentry, as the
1107 * lookup validation will know that the inode is bad.
1108 * (But we fall through to invalidate the caches.)
1110 nfs_invalidate_inode(inode
);
1114 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1115 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1116 NFS_SERVER(inode
)->nfs_client
->cl_hostname
, inode
->i_sb
->s_id
,
1117 (long long)nfsi
->fileid
, (long long)fattr
->fileid
);
1122 #ifdef CONFIG_NFS_V4
1125 * Clean out any remaining NFSv4 state that might be left over due
1126 * to open() calls that passed nfs_atomic_lookup, but failed to call
1129 void nfs4_clear_inode(struct inode
*inode
)
1131 /* If we are holding a delegation, return it! */
1132 nfs_inode_return_delegation(inode
);
1133 /* First call standard NFS clear_inode() code */
1134 nfs_clear_inode(inode
);
1138 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1140 struct nfs_inode
*nfsi
;
1141 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1145 nfsi
->cache_validity
= 0UL;
1146 #ifdef CONFIG_NFS_V3_ACL
1147 nfsi
->acl_access
= ERR_PTR(-EAGAIN
);
1148 nfsi
->acl_default
= ERR_PTR(-EAGAIN
);
1150 #ifdef CONFIG_NFS_V4
1151 nfsi
->nfs4_acl
= NULL
;
1152 #endif /* CONFIG_NFS_V4 */
1153 return &nfsi
->vfs_inode
;
1156 void nfs_destroy_inode(struct inode
*inode
)
1158 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1161 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1163 #ifdef CONFIG_NFS_V4
1164 INIT_LIST_HEAD(&nfsi
->open_states
);
1165 nfsi
->delegation
= NULL
;
1166 nfsi
->delegation_state
= 0;
1167 init_rwsem(&nfsi
->rwsem
);
1171 static void init_once(struct kmem_cache
* cachep
, void *foo
)
1173 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1175 inode_init_once(&nfsi
->vfs_inode
);
1176 INIT_LIST_HEAD(&nfsi
->open_files
);
1177 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1178 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1179 INIT_RADIX_TREE(&nfsi
->nfs_page_tree
, GFP_ATOMIC
);
1182 atomic_set(&nfsi
->silly_count
, 1);
1183 INIT_HLIST_HEAD(&nfsi
->silly_list
);
1184 init_waitqueue_head(&nfsi
->waitqueue
);
1185 nfs4_init_once(nfsi
);
1188 static int __init
nfs_init_inodecache(void)
1190 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1191 sizeof(struct nfs_inode
),
1192 0, (SLAB_RECLAIM_ACCOUNT
|
1195 if (nfs_inode_cachep
== NULL
)
1201 static void nfs_destroy_inodecache(void)
1203 kmem_cache_destroy(nfs_inode_cachep
);
1209 static int __init
init_nfs_fs(void)
1213 err
= nfs_fs_proc_init();
1217 err
= nfs_init_nfspagecache();
1221 err
= nfs_init_inodecache();
1225 err
= nfs_init_readpagecache();
1229 err
= nfs_init_writepagecache();
1233 err
= nfs_init_directcache();
1237 #ifdef CONFIG_PROC_FS
1238 rpc_proc_register(&nfs_rpcstat
);
1240 if ((err
= register_nfs_fs()) != 0)
1244 #ifdef CONFIG_PROC_FS
1245 rpc_proc_unregister("nfs");
1247 nfs_destroy_directcache();
1249 nfs_destroy_writepagecache();
1251 nfs_destroy_readpagecache();
1253 nfs_destroy_inodecache();
1255 nfs_destroy_nfspagecache();
1262 static void __exit
exit_nfs_fs(void)
1264 nfs_destroy_directcache();
1265 nfs_destroy_writepagecache();
1266 nfs_destroy_readpagecache();
1267 nfs_destroy_inodecache();
1268 nfs_destroy_nfspagecache();
1269 #ifdef CONFIG_PROC_FS
1270 rpc_proc_unregister("nfs");
1272 unregister_nfs_fs();
1276 /* Not quite true; I just maintain it */
1277 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1278 MODULE_LICENSE("GPL");
1279 module_param(enable_ino64
, bool, 0644);
1281 module_init(init_nfs_fs
)
1282 module_exit(exit_nfs_fs
)