4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, 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/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
56 #define NFSDBG_FACILITY NFSDBG_VFS
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64
= NFS_64_BIT_INODE_NUMBERS_ENABLED
;
63 static void nfs_invalidate_inode(struct inode
*);
64 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
66 static struct kmem_cache
* nfs_inode_cachep
;
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
71 return nfs_fileid_to_ino_t(fattr
->fileid
);
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
78 int nfs_wait_bit_killable(void *word
)
80 if (fatal_signal_pending(current
))
82 freezable_schedule_unsafe();
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable
);
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
94 u64
nfs_compat_user_ino64(u64 fileid
)
105 if (sizeof(ino
) < sizeof(fileid
))
106 ino
^= fileid
>> (sizeof(fileid
)-sizeof(ino
)) * 8;
110 int nfs_drop_inode(struct inode
*inode
)
112 return NFS_STALE(inode
) || generic_drop_inode(inode
);
114 EXPORT_SYMBOL_GPL(nfs_drop_inode
);
116 void nfs_clear_inode(struct inode
*inode
)
119 * The following should never happen...
121 WARN_ON_ONCE(nfs_have_writebacks(inode
));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode
)->open_files
));
123 nfs_zap_acl_cache(inode
);
124 nfs_access_zap_cache(inode
);
125 nfs_fscache_clear_inode(inode
);
127 EXPORT_SYMBOL_GPL(nfs_clear_inode
);
129 void nfs_evict_inode(struct inode
*inode
)
131 truncate_inode_pages_final(&inode
->i_data
);
133 nfs_clear_inode(inode
);
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
139 int nfs_sync_mapping(struct address_space
*mapping
)
143 if (mapping
->nrpages
!= 0) {
144 unmap_mapping_range(mapping
, 0, 0, 0);
145 ret
= nfs_wb_all(mapping
->host
);
151 * Invalidate the local caches
153 static void nfs_zap_caches_locked(struct inode
*inode
)
155 struct nfs_inode
*nfsi
= NFS_I(inode
);
156 int mode
= inode
->i_mode
;
158 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
160 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
161 nfsi
->attrtimeo_timestamp
= jiffies
;
163 memset(NFS_I(inode
)->cookieverf
, 0, sizeof(NFS_I(inode
)->cookieverf
));
164 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
165 nfs_fscache_invalidate(inode
);
166 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
167 | NFS_INO_INVALID_DATA
168 | NFS_INO_INVALID_ACCESS
169 | NFS_INO_INVALID_ACL
170 | NFS_INO_REVAL_PAGECACHE
;
172 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
173 | NFS_INO_INVALID_ACCESS
174 | NFS_INO_INVALID_ACL
175 | NFS_INO_REVAL_PAGECACHE
;
176 nfs_zap_label_cache_locked(nfsi
);
179 void nfs_zap_caches(struct inode
*inode
)
181 spin_lock(&inode
->i_lock
);
182 nfs_zap_caches_locked(inode
);
183 spin_unlock(&inode
->i_lock
);
186 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
188 if (mapping
->nrpages
!= 0) {
189 spin_lock(&inode
->i_lock
);
190 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_DATA
;
191 nfs_fscache_invalidate(inode
);
192 spin_unlock(&inode
->i_lock
);
196 void nfs_zap_acl_cache(struct inode
*inode
)
198 void (*clear_acl_cache
)(struct inode
*);
200 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
201 if (clear_acl_cache
!= NULL
)
202 clear_acl_cache(inode
);
203 spin_lock(&inode
->i_lock
);
204 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
205 spin_unlock(&inode
->i_lock
);
207 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache
);
209 void nfs_invalidate_atime(struct inode
*inode
)
211 spin_lock(&inode
->i_lock
);
212 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ATIME
;
213 spin_unlock(&inode
->i_lock
);
215 EXPORT_SYMBOL_GPL(nfs_invalidate_atime
);
218 * Invalidate, but do not unhash, the inode.
219 * NB: must be called with inode->i_lock held!
221 static void nfs_invalidate_inode(struct inode
*inode
)
223 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
224 nfs_zap_caches_locked(inode
);
227 struct nfs_find_desc
{
229 struct nfs_fattr
*fattr
;
233 * In NFSv3 we can have 64bit inode numbers. In order to support
234 * this, and re-exported directories (also seen in NFSv2)
235 * we are forced to allow 2 different inodes to have the same
239 nfs_find_actor(struct inode
*inode
, void *opaque
)
241 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
242 struct nfs_fh
*fh
= desc
->fh
;
243 struct nfs_fattr
*fattr
= desc
->fattr
;
245 if (NFS_FILEID(inode
) != fattr
->fileid
)
247 if ((S_IFMT
& inode
->i_mode
) != (S_IFMT
& fattr
->mode
))
249 if (nfs_compare_fh(NFS_FH(inode
), fh
))
251 if (is_bad_inode(inode
) || NFS_STALE(inode
))
257 nfs_init_locked(struct inode
*inode
, void *opaque
)
259 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
260 struct nfs_fattr
*fattr
= desc
->fattr
;
262 set_nfs_fileid(inode
, fattr
->fileid
);
263 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
267 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
268 static void nfs_clear_label_invalid(struct inode
*inode
)
270 spin_lock(&inode
->i_lock
);
271 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_LABEL
;
272 spin_unlock(&inode
->i_lock
);
275 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
276 struct nfs4_label
*label
)
283 if ((fattr
->valid
& NFS_ATTR_FATTR_V4_SECURITY_LABEL
) && inode
->i_security
) {
284 error
= security_inode_notifysecctx(inode
, label
->label
,
287 printk(KERN_ERR
"%s() %s %d "
288 "security_inode_notifysecctx() %d\n",
290 (char *)label
->label
,
292 nfs_clear_label_invalid(inode
);
296 struct nfs4_label
*nfs4_label_alloc(struct nfs_server
*server
, gfp_t flags
)
298 struct nfs4_label
*label
= NULL
;
299 int minor_version
= server
->nfs_client
->cl_minorversion
;
301 if (minor_version
< 2)
304 if (!(server
->caps
& NFS_CAP_SECURITY_LABEL
))
307 label
= kzalloc(sizeof(struct nfs4_label
), flags
);
309 return ERR_PTR(-ENOMEM
);
311 label
->label
= kzalloc(NFS4_MAXLABELLEN
, flags
);
312 if (label
->label
== NULL
) {
314 return ERR_PTR(-ENOMEM
);
316 label
->len
= NFS4_MAXLABELLEN
;
320 EXPORT_SYMBOL_GPL(nfs4_label_alloc
);
322 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
323 struct nfs4_label
*label
)
327 EXPORT_SYMBOL_GPL(nfs_setsecurity
);
330 * This is our front-end to iget that looks up inodes by file handle
331 * instead of inode number.
334 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
, struct nfs4_label
*label
)
336 struct nfs_find_desc desc
= {
340 struct inode
*inode
= ERR_PTR(-ENOENT
);
343 nfs_attr_check_mountpoint(sb
, fattr
);
345 if (((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) == 0) &&
346 !nfs_attr_use_mounted_on_fileid(fattr
))
348 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) == 0)
351 hash
= nfs_fattr_to_ino_t(fattr
);
353 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
355 inode
= ERR_PTR(-ENOMEM
);
359 if (inode
->i_state
& I_NEW
) {
360 struct nfs_inode
*nfsi
= NFS_I(inode
);
361 unsigned long now
= jiffies
;
363 /* We set i_ino for the few things that still rely on it,
367 /* We can't support update_atime(), since the server will reset it */
368 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
369 inode
->i_mode
= fattr
->mode
;
370 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) == 0
371 && nfs_server_capable(inode
, NFS_CAP_MODE
))
372 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
373 /* Why so? Because we want revalidate for devices/FIFOs, and
374 * that's precisely what we have in nfs_file_inode_operations.
376 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
377 if (S_ISREG(inode
->i_mode
)) {
378 inode
->i_fop
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_ops
;
379 inode
->i_data
.a_ops
= &nfs_file_aops
;
380 inode
->i_data
.backing_dev_info
= &NFS_SB(sb
)->backing_dev_info
;
381 } else if (S_ISDIR(inode
->i_mode
)) {
382 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
383 inode
->i_fop
= &nfs_dir_operations
;
384 inode
->i_data
.a_ops
= &nfs_dir_aops
;
385 /* Deal with crossing mountpoints */
386 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTPOINT
||
387 fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
) {
388 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
389 inode
->i_op
= &nfs_referral_inode_operations
;
391 inode
->i_op
= &nfs_mountpoint_inode_operations
;
393 inode
->i_flags
|= S_AUTOMOUNT
;
395 } else if (S_ISLNK(inode
->i_mode
))
396 inode
->i_op
= &nfs_symlink_inode_operations
;
398 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
400 memset(&inode
->i_atime
, 0, sizeof(inode
->i_atime
));
401 memset(&inode
->i_mtime
, 0, sizeof(inode
->i_mtime
));
402 memset(&inode
->i_ctime
, 0, sizeof(inode
->i_ctime
));
403 inode
->i_version
= 0;
406 inode
->i_uid
= make_kuid(&init_user_ns
, -2);
407 inode
->i_gid
= make_kgid(&init_user_ns
, -2);
409 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
413 nfsi
->read_cache_jiffies
= fattr
->time_start
;
414 nfsi
->attr_gencount
= fattr
->gencount
;
415 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
416 inode
->i_atime
= fattr
->atime
;
417 else if (nfs_server_capable(inode
, NFS_CAP_ATIME
))
418 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
419 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
420 inode
->i_mtime
= fattr
->mtime
;
421 else if (nfs_server_capable(inode
, NFS_CAP_MTIME
))
422 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
423 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
424 inode
->i_ctime
= fattr
->ctime
;
425 else if (nfs_server_capable(inode
, NFS_CAP_CTIME
))
426 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
427 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
428 inode
->i_version
= fattr
->change_attr
;
429 else if (nfs_server_capable(inode
, NFS_CAP_CHANGE_ATTR
))
430 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
431 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
432 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
434 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
435 | NFS_INO_REVAL_PAGECACHE
;
436 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
)
437 set_nlink(inode
, fattr
->nlink
);
438 else if (nfs_server_capable(inode
, NFS_CAP_NLINK
))
439 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
440 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
)
441 inode
->i_uid
= fattr
->uid
;
442 else if (nfs_server_capable(inode
, NFS_CAP_OWNER
))
443 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
444 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
)
445 inode
->i_gid
= fattr
->gid
;
446 else if (nfs_server_capable(inode
, NFS_CAP_OWNER_GROUP
))
447 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
;
448 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
449 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
450 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
452 * report the blocks in 512byte units
454 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
457 nfs_setsecurity(inode
, fattr
, label
);
459 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
460 nfsi
->attrtimeo_timestamp
= now
;
461 nfsi
->access_cache
= RB_ROOT
;
463 nfs_fscache_init_inode(inode
);
465 unlock_new_inode(inode
);
467 nfs_refresh_inode(inode
, fattr
);
468 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
470 (unsigned long long)NFS_FILEID(inode
),
471 nfs_display_fhandle_hash(fh
),
472 atomic_read(&inode
->i_count
));
478 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
481 EXPORT_SYMBOL_GPL(nfs_fhget
);
483 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
486 nfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
488 struct inode
*inode
= dentry
->d_inode
;
489 struct nfs_fattr
*fattr
;
492 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
494 /* skip mode change if it's just for clearing setuid/setgid */
495 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
496 attr
->ia_valid
&= ~ATTR_MODE
;
498 if (attr
->ia_valid
& ATTR_SIZE
) {
499 if (!S_ISREG(inode
->i_mode
) || attr
->ia_size
== i_size_read(inode
))
500 attr
->ia_valid
&= ~ATTR_SIZE
;
503 /* Optimization: if the end result is no change, don't RPC */
504 attr
->ia_valid
&= NFS_VALID_ATTRS
;
505 if ((attr
->ia_valid
& ~(ATTR_FILE
|ATTR_OPEN
)) == 0)
508 trace_nfs_setattr_enter(inode
);
510 /* Write all dirty data */
511 if (S_ISREG(inode
->i_mode
)) {
512 nfs_inode_dio_wait(inode
);
516 fattr
= nfs_alloc_fattr();
520 * Return any delegations if we're going to change ACLs
522 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0)
523 NFS_PROTO(inode
)->return_delegation(inode
);
524 error
= NFS_PROTO(inode
)->setattr(dentry
, fattr
, attr
);
526 error
= nfs_refresh_inode(inode
, fattr
);
527 nfs_free_fattr(fattr
);
529 trace_nfs_setattr_exit(inode
, error
);
532 EXPORT_SYMBOL_GPL(nfs_setattr
);
535 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
536 * @inode: inode of the file used
537 * @offset: file offset to start truncating
539 * This is a copy of the common vmtruncate, but with the locking
540 * corrected to take into account the fact that NFS requires
541 * inode->i_size to be updated under the inode->i_lock.
543 static int nfs_vmtruncate(struct inode
* inode
, loff_t offset
)
547 err
= inode_newsize_ok(inode
, offset
);
551 spin_lock(&inode
->i_lock
);
552 i_size_write(inode
, offset
);
553 spin_unlock(&inode
->i_lock
);
555 truncate_pagecache(inode
, offset
);
561 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
562 * @inode: pointer to struct inode
563 * @attr: pointer to struct iattr
565 * Note: we do this in the *proc.c in order to ensure that
566 * it works for things like exclusive creates too.
568 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
)
570 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
571 spin_lock(&inode
->i_lock
);
572 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
573 int mode
= attr
->ia_mode
& S_IALLUGO
;
574 mode
|= inode
->i_mode
& ~S_IALLUGO
;
575 inode
->i_mode
= mode
;
577 if ((attr
->ia_valid
& ATTR_UID
) != 0)
578 inode
->i_uid
= attr
->ia_uid
;
579 if ((attr
->ia_valid
& ATTR_GID
) != 0)
580 inode
->i_gid
= attr
->ia_gid
;
581 NFS_I(inode
)->cache_validity
|= NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
582 spin_unlock(&inode
->i_lock
);
584 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
585 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
586 nfs_vmtruncate(inode
, attr
->ia_size
);
589 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode
);
591 static void nfs_request_parent_use_readdirplus(struct dentry
*dentry
)
593 struct dentry
*parent
;
595 parent
= dget_parent(dentry
);
596 nfs_force_use_readdirplus(parent
->d_inode
);
600 static bool nfs_need_revalidate_inode(struct inode
*inode
)
602 if (NFS_I(inode
)->cache_validity
&
603 (NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_LABEL
))
605 if (nfs_attribute_cache_expired(inode
))
610 int nfs_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
, struct kstat
*stat
)
612 struct inode
*inode
= dentry
->d_inode
;
613 int need_atime
= NFS_I(inode
)->cache_validity
& NFS_INO_INVALID_ATIME
;
616 trace_nfs_getattr_enter(inode
);
617 /* Flush out writes to the server in order to update c/mtime. */
618 if (S_ISREG(inode
->i_mode
)) {
619 nfs_inode_dio_wait(inode
);
620 err
= filemap_write_and_wait(inode
->i_mapping
);
626 * We may force a getattr if the user cares about atime.
628 * Note that we only have to check the vfsmount flags here:
629 * - NFS always sets S_NOATIME by so checking it would give a
631 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
632 * no point in checking those.
634 if ((mnt
->mnt_flags
& MNT_NOATIME
) ||
635 ((mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
638 if (need_atime
|| nfs_need_revalidate_inode(inode
)) {
639 struct nfs_server
*server
= NFS_SERVER(inode
);
641 if (server
->caps
& NFS_CAP_READDIRPLUS
)
642 nfs_request_parent_use_readdirplus(dentry
);
643 err
= __nfs_revalidate_inode(server
, inode
);
646 generic_fillattr(inode
, stat
);
647 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
650 trace_nfs_getattr_exit(inode
, err
);
653 EXPORT_SYMBOL_GPL(nfs_getattr
);
655 static void nfs_init_lock_context(struct nfs_lock_context
*l_ctx
)
657 atomic_set(&l_ctx
->count
, 1);
658 l_ctx
->lockowner
.l_owner
= current
->files
;
659 l_ctx
->lockowner
.l_pid
= current
->tgid
;
660 INIT_LIST_HEAD(&l_ctx
->list
);
661 nfs_iocounter_init(&l_ctx
->io_count
);
664 static struct nfs_lock_context
*__nfs_find_lock_context(struct nfs_open_context
*ctx
)
666 struct nfs_lock_context
*head
= &ctx
->lock_context
;
667 struct nfs_lock_context
*pos
= head
;
670 if (pos
->lockowner
.l_owner
!= current
->files
)
672 if (pos
->lockowner
.l_pid
!= current
->tgid
)
674 atomic_inc(&pos
->count
);
676 } while ((pos
= list_entry(pos
->list
.next
, typeof(*pos
), list
)) != head
);
680 struct nfs_lock_context
*nfs_get_lock_context(struct nfs_open_context
*ctx
)
682 struct nfs_lock_context
*res
, *new = NULL
;
683 struct inode
*inode
= ctx
->dentry
->d_inode
;
685 spin_lock(&inode
->i_lock
);
686 res
= __nfs_find_lock_context(ctx
);
688 spin_unlock(&inode
->i_lock
);
689 new = kmalloc(sizeof(*new), GFP_KERNEL
);
691 return ERR_PTR(-ENOMEM
);
692 nfs_init_lock_context(new);
693 spin_lock(&inode
->i_lock
);
694 res
= __nfs_find_lock_context(ctx
);
696 list_add_tail(&new->list
, &ctx
->lock_context
.list
);
697 new->open_context
= ctx
;
702 spin_unlock(&inode
->i_lock
);
707 void nfs_put_lock_context(struct nfs_lock_context
*l_ctx
)
709 struct nfs_open_context
*ctx
= l_ctx
->open_context
;
710 struct inode
*inode
= ctx
->dentry
->d_inode
;
712 if (!atomic_dec_and_lock(&l_ctx
->count
, &inode
->i_lock
))
714 list_del(&l_ctx
->list
);
715 spin_unlock(&inode
->i_lock
);
720 * nfs_close_context - Common close_context() routine NFSv2/v3
721 * @ctx: pointer to context
722 * @is_sync: is this a synchronous close
724 * always ensure that the attributes are up to date if we're mounted
725 * with close-to-open semantics
727 void nfs_close_context(struct nfs_open_context
*ctx
, int is_sync
)
730 struct nfs_server
*server
;
732 if (!(ctx
->mode
& FMODE_WRITE
))
736 inode
= ctx
->dentry
->d_inode
;
737 if (!list_empty(&NFS_I(inode
)->open_files
))
739 server
= NFS_SERVER(inode
);
740 if (server
->flags
& NFS_MOUNT_NOCTO
)
742 nfs_revalidate_inode(server
, inode
);
744 EXPORT_SYMBOL_GPL(nfs_close_context
);
746 struct nfs_open_context
*alloc_nfs_open_context(struct dentry
*dentry
, fmode_t f_mode
)
748 struct nfs_open_context
*ctx
;
749 struct rpc_cred
*cred
= rpc_lookup_cred();
751 return ERR_CAST(cred
);
753 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
756 return ERR_PTR(-ENOMEM
);
758 nfs_sb_active(dentry
->d_sb
);
759 ctx
->dentry
= dget(dentry
);
765 nfs_init_lock_context(&ctx
->lock_context
);
766 ctx
->lock_context
.open_context
= ctx
;
767 INIT_LIST_HEAD(&ctx
->list
);
768 ctx
->mdsthreshold
= NULL
;
771 EXPORT_SYMBOL_GPL(alloc_nfs_open_context
);
773 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
776 atomic_inc(&ctx
->lock_context
.count
);
779 EXPORT_SYMBOL_GPL(get_nfs_open_context
);
781 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int is_sync
)
783 struct inode
*inode
= ctx
->dentry
->d_inode
;
784 struct super_block
*sb
= ctx
->dentry
->d_sb
;
786 if (!list_empty(&ctx
->list
)) {
787 if (!atomic_dec_and_lock(&ctx
->lock_context
.count
, &inode
->i_lock
))
789 list_del(&ctx
->list
);
790 spin_unlock(&inode
->i_lock
);
791 } else if (!atomic_dec_and_test(&ctx
->lock_context
.count
))
794 NFS_PROTO(inode
)->close_context(ctx
, is_sync
);
795 if (ctx
->cred
!= NULL
)
796 put_rpccred(ctx
->cred
);
799 kfree(ctx
->mdsthreshold
);
803 void put_nfs_open_context(struct nfs_open_context
*ctx
)
805 __put_nfs_open_context(ctx
, 0);
807 EXPORT_SYMBOL_GPL(put_nfs_open_context
);
810 * Ensure that mmap has a recent RPC credential for use when writing out
813 void nfs_inode_attach_open_context(struct nfs_open_context
*ctx
)
815 struct inode
*inode
= ctx
->dentry
->d_inode
;
816 struct nfs_inode
*nfsi
= NFS_I(inode
);
818 spin_lock(&inode
->i_lock
);
819 list_add(&ctx
->list
, &nfsi
->open_files
);
820 spin_unlock(&inode
->i_lock
);
822 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context
);
824 void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
826 filp
->private_data
= get_nfs_open_context(ctx
);
827 if (list_empty(&ctx
->list
))
828 nfs_inode_attach_open_context(ctx
);
830 EXPORT_SYMBOL_GPL(nfs_file_set_open_context
);
833 * Given an inode, search for an open context with the desired characteristics
835 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, struct rpc_cred
*cred
, fmode_t mode
)
837 struct nfs_inode
*nfsi
= NFS_I(inode
);
838 struct nfs_open_context
*pos
, *ctx
= NULL
;
840 spin_lock(&inode
->i_lock
);
841 list_for_each_entry(pos
, &nfsi
->open_files
, list
) {
842 if (cred
!= NULL
&& pos
->cred
!= cred
)
844 if ((pos
->mode
& (FMODE_READ
|FMODE_WRITE
)) != mode
)
846 ctx
= get_nfs_open_context(pos
);
849 spin_unlock(&inode
->i_lock
);
853 static void nfs_file_clear_open_context(struct file
*filp
)
855 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
858 struct inode
*inode
= ctx
->dentry
->d_inode
;
860 filp
->private_data
= NULL
;
861 spin_lock(&inode
->i_lock
);
862 list_move_tail(&ctx
->list
, &NFS_I(inode
)->open_files
);
863 spin_unlock(&inode
->i_lock
);
864 __put_nfs_open_context(ctx
, filp
->f_flags
& O_DIRECT
? 0 : 1);
869 * These allocate and release file read/write context information.
871 int nfs_open(struct inode
*inode
, struct file
*filp
)
873 struct nfs_open_context
*ctx
;
875 ctx
= alloc_nfs_open_context(filp
->f_path
.dentry
, filp
->f_mode
);
878 nfs_file_set_open_context(filp
, ctx
);
879 put_nfs_open_context(ctx
);
880 nfs_fscache_open_file(inode
, filp
);
884 int nfs_release(struct inode
*inode
, struct file
*filp
)
886 nfs_file_clear_open_context(filp
);
891 * This function is called whenever some part of NFS notices that
892 * the cached attributes have to be refreshed.
895 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
897 int status
= -ESTALE
;
898 struct nfs4_label
*label
= NULL
;
899 struct nfs_fattr
*fattr
= NULL
;
900 struct nfs_inode
*nfsi
= NFS_I(inode
);
902 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Lu)\n",
903 inode
->i_sb
->s_id
, (unsigned long long)NFS_FILEID(inode
));
905 trace_nfs_revalidate_inode_enter(inode
);
907 if (is_bad_inode(inode
))
909 if (NFS_STALE(inode
))
913 fattr
= nfs_alloc_fattr();
917 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
919 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
921 status
= PTR_ERR(label
);
925 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), fattr
, label
);
927 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
929 (unsigned long long)NFS_FILEID(inode
), status
);
930 if (status
== -ESTALE
) {
931 nfs_zap_caches(inode
);
932 if (!S_ISDIR(inode
->i_mode
))
933 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
938 status
= nfs_refresh_inode(inode
, fattr
);
940 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
942 (unsigned long long)NFS_FILEID(inode
), status
);
946 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
947 nfs_zap_acl_cache(inode
);
949 nfs_setsecurity(inode
, fattr
, label
);
951 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) revalidation complete\n",
953 (unsigned long long)NFS_FILEID(inode
));
956 nfs4_label_free(label
);
958 nfs_free_fattr(fattr
);
959 trace_nfs_revalidate_inode_exit(inode
, status
);
963 int nfs_attribute_timeout(struct inode
*inode
)
965 struct nfs_inode
*nfsi
= NFS_I(inode
);
967 return !time_in_range_open(jiffies
, nfsi
->read_cache_jiffies
, nfsi
->read_cache_jiffies
+ nfsi
->attrtimeo
);
970 int nfs_attribute_cache_expired(struct inode
*inode
)
972 if (nfs_have_delegated_attributes(inode
))
974 return nfs_attribute_timeout(inode
);
978 * nfs_revalidate_inode - Revalidate the inode attributes
979 * @server - pointer to nfs_server struct
980 * @inode - pointer to inode struct
982 * Updates inode attribute information by retrieving the data from the server.
984 int nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
986 if (!nfs_need_revalidate_inode(inode
))
987 return NFS_STALE(inode
) ? -ESTALE
: 0;
988 return __nfs_revalidate_inode(server
, inode
);
990 EXPORT_SYMBOL_GPL(nfs_revalidate_inode
);
992 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
994 struct nfs_inode
*nfsi
= NFS_I(inode
);
997 if (mapping
->nrpages
!= 0) {
998 if (S_ISREG(inode
->i_mode
)) {
999 ret
= nfs_sync_mapping(mapping
);
1003 ret
= invalidate_inode_pages2(mapping
);
1007 if (S_ISDIR(inode
->i_mode
)) {
1008 spin_lock(&inode
->i_lock
);
1009 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
1010 spin_unlock(&inode
->i_lock
);
1012 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
1013 nfs_fscache_wait_on_invalidate(inode
);
1015 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) data cache invalidated\n",
1017 (unsigned long long)NFS_FILEID(inode
));
1021 static bool nfs_mapping_need_revalidate_inode(struct inode
*inode
)
1023 if (nfs_have_delegated_attributes(inode
))
1025 return (NFS_I(inode
)->cache_validity
& NFS_INO_REVAL_PAGECACHE
)
1026 || nfs_attribute_timeout(inode
)
1027 || NFS_STALE(inode
);
1031 * nfs_revalidate_mapping - Revalidate the pagecache
1032 * @inode - pointer to host inode
1033 * @mapping - pointer to mapping
1035 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1037 struct nfs_inode
*nfsi
= NFS_I(inode
);
1038 unsigned long *bitlock
= &nfsi
->flags
;
1041 /* swapfiles are not supposed to be shared. */
1042 if (IS_SWAPFILE(inode
))
1045 if (nfs_mapping_need_revalidate_inode(inode
)) {
1046 ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1052 * We must clear NFS_INO_INVALID_DATA first to ensure that
1053 * invalidations that come in while we're shooting down the mappings
1054 * are respected. But, that leaves a race window where one revalidator
1055 * can clear the flag, and then another checks it before the mapping
1056 * gets invalidated. Fix that by serializing access to this part of
1059 * At the same time, we need to allow other tasks to see whether we
1060 * might be in the middle of invalidating the pages, so we only set
1061 * the bit lock here if it looks like we're going to be doing that.
1064 ret
= wait_on_bit(bitlock
, NFS_INO_INVALIDATING
,
1065 nfs_wait_bit_killable
, TASK_KILLABLE
);
1068 spin_lock(&inode
->i_lock
);
1069 if (test_bit(NFS_INO_INVALIDATING
, bitlock
)) {
1070 spin_unlock(&inode
->i_lock
);
1073 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1075 spin_unlock(&inode
->i_lock
);
1079 set_bit(NFS_INO_INVALIDATING
, bitlock
);
1081 nfsi
->cache_validity
&= ~NFS_INO_INVALID_DATA
;
1082 spin_unlock(&inode
->i_lock
);
1083 trace_nfs_invalidate_mapping_enter(inode
);
1084 ret
= nfs_invalidate_mapping(inode
, mapping
);
1085 trace_nfs_invalidate_mapping_exit(inode
, ret
);
1087 clear_bit_unlock(NFS_INO_INVALIDATING
, bitlock
);
1088 smp_mb__after_atomic();
1089 wake_up_bit(bitlock
, NFS_INO_INVALIDATING
);
1094 static unsigned long nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1096 struct nfs_inode
*nfsi
= NFS_I(inode
);
1097 unsigned long ret
= 0;
1099 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
)
1100 && (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
1101 && inode
->i_version
== fattr
->pre_change_attr
) {
1102 inode
->i_version
= fattr
->change_attr
;
1103 if (S_ISDIR(inode
->i_mode
))
1104 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
1105 ret
|= NFS_INO_INVALID_ATTR
;
1107 /* If we have atomic WCC data, we may update some attributes */
1108 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
)
1109 && (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
1110 && timespec_equal(&inode
->i_ctime
, &fattr
->pre_ctime
)) {
1111 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1112 ret
|= NFS_INO_INVALID_ATTR
;
1115 if ((fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
)
1116 && (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
1117 && timespec_equal(&inode
->i_mtime
, &fattr
->pre_mtime
)) {
1118 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1119 if (S_ISDIR(inode
->i_mode
))
1120 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
1121 ret
|= NFS_INO_INVALID_ATTR
;
1123 if ((fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
)
1124 && (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
1125 && i_size_read(inode
) == nfs_size_to_loff_t(fattr
->pre_size
)
1126 && nfsi
->npages
== 0) {
1127 i_size_write(inode
, nfs_size_to_loff_t(fattr
->size
));
1128 ret
|= NFS_INO_INVALID_ATTR
;
1131 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1132 nfs_fscache_invalidate(inode
);
1138 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1139 * @inode - pointer to inode
1140 * @fattr - updated attributes
1142 * Verifies the attribute cache. If we have just changed the attributes,
1143 * so that fattr carries weak cache consistency data, then it may
1144 * also update the ctime/mtime/change_attribute.
1146 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
1148 struct nfs_inode
*nfsi
= NFS_I(inode
);
1149 loff_t cur_size
, new_isize
;
1150 unsigned long invalid
= 0;
1153 if (nfs_have_delegated_attributes(inode
))
1155 /* Has the inode gone and changed behind our back? */
1156 if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) && nfsi
->fileid
!= fattr
->fileid
)
1158 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
))
1161 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1162 inode
->i_version
!= fattr
->change_attr
)
1163 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
1165 /* Verify a few of the more important attributes */
1166 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) && !timespec_equal(&inode
->i_mtime
, &fattr
->mtime
))
1167 invalid
|= NFS_INO_INVALID_ATTR
;
1169 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1170 cur_size
= i_size_read(inode
);
1171 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1172 if (cur_size
!= new_isize
&& nfsi
->npages
== 0)
1173 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
1176 /* Have any file permissions changed? */
1177 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) && (inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
))
1178 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
1179 if ((fattr
->valid
& NFS_ATTR_FATTR_OWNER
) && !uid_eq(inode
->i_uid
, fattr
->uid
))
1180 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
1181 if ((fattr
->valid
& NFS_ATTR_FATTR_GROUP
) && !gid_eq(inode
->i_gid
, fattr
->gid
))
1182 invalid
|= NFS_INO_INVALID_ATTR
| NFS_INO_INVALID_ACCESS
| NFS_INO_INVALID_ACL
;
1184 /* Has the link count changed? */
1185 if ((fattr
->valid
& NFS_ATTR_FATTR_NLINK
) && inode
->i_nlink
!= fattr
->nlink
)
1186 invalid
|= NFS_INO_INVALID_ATTR
;
1188 if ((fattr
->valid
& NFS_ATTR_FATTR_ATIME
) && !timespec_equal(&inode
->i_atime
, &fattr
->atime
))
1189 invalid
|= NFS_INO_INVALID_ATIME
;
1192 nfsi
->cache_validity
|= invalid
;
1194 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1198 static int nfs_ctime_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
1200 if (!(fattr
->valid
& NFS_ATTR_FATTR_CTIME
))
1202 return timespec_compare(&fattr
->ctime
, &inode
->i_ctime
) > 0;
1205 static int nfs_size_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
1207 if (!(fattr
->valid
& NFS_ATTR_FATTR_SIZE
))
1209 return nfs_size_to_loff_t(fattr
->size
) > i_size_read(inode
);
1212 static atomic_long_t nfs_attr_generation_counter
;
1214 static unsigned long nfs_read_attr_generation_counter(void)
1216 return atomic_long_read(&nfs_attr_generation_counter
);
1219 unsigned long nfs_inc_attr_generation_counter(void)
1221 return atomic_long_inc_return(&nfs_attr_generation_counter
);
1224 void nfs_fattr_init(struct nfs_fattr
*fattr
)
1227 fattr
->time_start
= jiffies
;
1228 fattr
->gencount
= nfs_inc_attr_generation_counter();
1229 fattr
->owner_name
= NULL
;
1230 fattr
->group_name
= NULL
;
1232 EXPORT_SYMBOL_GPL(nfs_fattr_init
);
1234 struct nfs_fattr
*nfs_alloc_fattr(void)
1236 struct nfs_fattr
*fattr
;
1238 fattr
= kmalloc(sizeof(*fattr
), GFP_NOFS
);
1240 nfs_fattr_init(fattr
);
1243 EXPORT_SYMBOL_GPL(nfs_alloc_fattr
);
1245 struct nfs_fh
*nfs_alloc_fhandle(void)
1249 fh
= kmalloc(sizeof(struct nfs_fh
), GFP_NOFS
);
1254 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle
);
1258 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1259 * in the same way that wireshark does
1263 * For debugging only.
1265 u32
_nfs_display_fhandle_hash(const struct nfs_fh
*fh
)
1267 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1268 * not on the result */
1269 return nfs_fhandle_hash(fh
);
1271 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash
);
1274 * _nfs_display_fhandle - display an NFS file handle on the console
1276 * @fh: file handle to display
1277 * @caption: display caption
1279 * For debugging only.
1281 void _nfs_display_fhandle(const struct nfs_fh
*fh
, const char *caption
)
1285 if (fh
== NULL
|| fh
->size
== 0) {
1286 printk(KERN_DEFAULT
"%s at %p is empty\n", caption
, fh
);
1290 printk(KERN_DEFAULT
"%s at %p is %u bytes, crc: 0x%08x:\n",
1291 caption
, fh
, fh
->size
, _nfs_display_fhandle_hash(fh
));
1292 for (i
= 0; i
< fh
->size
; i
+= 16) {
1293 __be32
*pos
= (__be32
*)&fh
->data
[i
];
1295 switch ((fh
->size
- i
- 1) >> 2) {
1297 printk(KERN_DEFAULT
" %08x\n",
1301 printk(KERN_DEFAULT
" %08x %08x\n",
1302 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1));
1305 printk(KERN_DEFAULT
" %08x %08x %08x\n",
1306 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1307 be32_to_cpup(pos
+ 2));
1310 printk(KERN_DEFAULT
" %08x %08x %08x %08x\n",
1311 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1312 be32_to_cpup(pos
+ 2), be32_to_cpup(pos
+ 3));
1316 EXPORT_SYMBOL_GPL(_nfs_display_fhandle
);
1320 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1321 * @inode - pointer to inode
1322 * @fattr - attributes
1324 * Attempt to divine whether or not an RPC call reply carrying stale
1325 * attributes got scheduled after another call carrying updated ones.
1327 * To do so, the function first assumes that a more recent ctime means
1328 * that the attributes in fattr are newer, however it also attempt to
1329 * catch the case where ctime either didn't change, or went backwards
1330 * (if someone reset the clock on the server) by looking at whether
1331 * or not this RPC call was started after the inode was last updated.
1332 * Note also the check for wraparound of 'attr_gencount'
1334 * The function returns 'true' if it thinks the attributes in 'fattr' are
1335 * more recent than the ones cached in the inode.
1338 static int nfs_inode_attrs_need_update(const struct inode
*inode
, const struct nfs_fattr
*fattr
)
1340 const struct nfs_inode
*nfsi
= NFS_I(inode
);
1342 return ((long)fattr
->gencount
- (long)nfsi
->attr_gencount
) > 0 ||
1343 nfs_ctime_need_update(inode
, fattr
) ||
1344 nfs_size_need_update(inode
, fattr
) ||
1345 ((long)nfsi
->attr_gencount
- (long)nfs_read_attr_generation_counter() > 0);
1349 * Don't trust the change_attribute, mtime, ctime or size if
1350 * a pnfs LAYOUTCOMMIT is outstanding
1352 static void nfs_inode_attrs_handle_layoutcommit(struct inode
*inode
,
1353 struct nfs_fattr
*fattr
)
1355 if (pnfs_layoutcommit_outstanding(inode
))
1356 fattr
->valid
&= ~(NFS_ATTR_FATTR_CHANGE
|
1357 NFS_ATTR_FATTR_MTIME
|
1358 NFS_ATTR_FATTR_CTIME
|
1359 NFS_ATTR_FATTR_SIZE
);
1362 static int nfs_refresh_inode_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1366 trace_nfs_refresh_inode_enter(inode
);
1368 nfs_inode_attrs_handle_layoutcommit(inode
, fattr
);
1370 if (nfs_inode_attrs_need_update(inode
, fattr
))
1371 ret
= nfs_update_inode(inode
, fattr
);
1373 ret
= nfs_check_inode_attributes(inode
, fattr
);
1375 trace_nfs_refresh_inode_exit(inode
, ret
);
1380 * nfs_refresh_inode - try to update the inode attribute cache
1381 * @inode - pointer to inode
1382 * @fattr - updated attributes
1384 * Check that an RPC call that returned attributes has not overlapped with
1385 * other recent updates of the inode metadata, then decide whether it is
1386 * safe to do a full update of the inode attributes, or whether just to
1387 * call nfs_check_inode_attributes.
1389 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1393 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1395 spin_lock(&inode
->i_lock
);
1396 status
= nfs_refresh_inode_locked(inode
, fattr
);
1397 spin_unlock(&inode
->i_lock
);
1401 EXPORT_SYMBOL_GPL(nfs_refresh_inode
);
1403 static int nfs_post_op_update_inode_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1405 struct nfs_inode
*nfsi
= NFS_I(inode
);
1407 nfsi
->cache_validity
|= NFS_INO_INVALID_ATTR
|NFS_INO_REVAL_PAGECACHE
;
1408 if (S_ISDIR(inode
->i_mode
)) {
1409 nfsi
->cache_validity
|= NFS_INO_INVALID_DATA
;
1410 nfs_fscache_invalidate(inode
);
1412 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1414 return nfs_refresh_inode_locked(inode
, fattr
);
1418 * nfs_post_op_update_inode - try to update the inode attribute cache
1419 * @inode - pointer to inode
1420 * @fattr - updated attributes
1422 * After an operation that has changed the inode metadata, mark the
1423 * attribute cache as being invalid, then try to update it.
1425 * NB: if the server didn't return any post op attributes, this
1426 * function will force the retrieval of attributes before the next
1427 * NFS request. Thus it should be used only for operations that
1428 * are expected to change one or more attributes, to avoid
1429 * unnecessary NFS requests and trips through nfs_update_inode().
1431 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1435 spin_lock(&inode
->i_lock
);
1436 status
= nfs_post_op_update_inode_locked(inode
, fattr
);
1437 spin_unlock(&inode
->i_lock
);
1441 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode
);
1444 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1445 * @inode - pointer to inode
1446 * @fattr - updated attributes
1448 * After an operation that has changed the inode metadata, mark the
1449 * attribute cache as being invalid, then try to update it. Fake up
1450 * weak cache consistency data, if none exist.
1452 * This function is mainly designed to be used by the ->write_done() functions.
1454 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
1458 spin_lock(&inode
->i_lock
);
1459 /* Don't do a WCC update if these attributes are already stale */
1460 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0 ||
1461 !nfs_inode_attrs_need_update(inode
, fattr
)) {
1462 fattr
->valid
&= ~(NFS_ATTR_FATTR_PRECHANGE
1463 | NFS_ATTR_FATTR_PRESIZE
1464 | NFS_ATTR_FATTR_PREMTIME
1465 | NFS_ATTR_FATTR_PRECTIME
);
1468 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1469 (fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
) == 0) {
1470 fattr
->pre_change_attr
= inode
->i_version
;
1471 fattr
->valid
|= NFS_ATTR_FATTR_PRECHANGE
;
1473 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) != 0 &&
1474 (fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
) == 0) {
1475 memcpy(&fattr
->pre_ctime
, &inode
->i_ctime
, sizeof(fattr
->pre_ctime
));
1476 fattr
->valid
|= NFS_ATTR_FATTR_PRECTIME
;
1478 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) != 0 &&
1479 (fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
) == 0) {
1480 memcpy(&fattr
->pre_mtime
, &inode
->i_mtime
, sizeof(fattr
->pre_mtime
));
1481 fattr
->valid
|= NFS_ATTR_FATTR_PREMTIME
;
1483 if ((fattr
->valid
& NFS_ATTR_FATTR_SIZE
) != 0 &&
1484 (fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
) == 0) {
1485 fattr
->pre_size
= i_size_read(inode
);
1486 fattr
->valid
|= NFS_ATTR_FATTR_PRESIZE
;
1489 status
= nfs_post_op_update_inode_locked(inode
, fattr
);
1490 spin_unlock(&inode
->i_lock
);
1493 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc
);
1496 * Many nfs protocol calls return the new file attributes after
1497 * an operation. Here we update the inode to reflect the state
1498 * of the server's inode.
1500 * This is a bit tricky because we have to make sure all dirty pages
1501 * have been sent off to the server before calling invalidate_inode_pages.
1502 * To make sure no other process adds more write requests while we try
1503 * our best to flush them, we make them sleep during the attribute refresh.
1505 * A very similar scenario holds for the dir cache.
1507 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1509 struct nfs_server
*server
;
1510 struct nfs_inode
*nfsi
= NFS_I(inode
);
1511 loff_t cur_isize
, new_isize
;
1512 unsigned long invalid
= 0;
1513 unsigned long now
= jiffies
;
1514 unsigned long save_cache_validity
;
1516 dfprintk(VFS
, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1517 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1518 nfs_display_fhandle_hash(NFS_FH(inode
)),
1519 atomic_read(&inode
->i_count
), fattr
->valid
);
1521 if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) && nfsi
->fileid
!= fattr
->fileid
) {
1522 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1523 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1524 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1525 inode
->i_sb
->s_id
, (long long)nfsi
->fileid
,
1526 (long long)fattr
->fileid
);
1531 * Make sure the inode's type hasn't changed.
1533 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && (inode
->i_mode
& S_IFMT
) != (fattr
->mode
& S_IFMT
)) {
1535 * Big trouble! The inode has become a different object.
1537 printk(KERN_DEBUG
"NFS: %s: inode %lu mode changed, %07o to %07o\n",
1538 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
1542 server
= NFS_SERVER(inode
);
1543 /* Update the fsid? */
1544 if (S_ISDIR(inode
->i_mode
) && (fattr
->valid
& NFS_ATTR_FATTR_FSID
) &&
1545 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
1546 !IS_AUTOMOUNT(inode
))
1547 server
->fsid
= fattr
->fsid
;
1550 * Update the read time so we don't revalidate too often.
1552 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1554 save_cache_validity
= nfsi
->cache_validity
;
1555 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
1556 | NFS_INO_INVALID_ATIME
1557 | NFS_INO_REVAL_FORCED
1558 | NFS_INO_REVAL_PAGECACHE
);
1560 /* Do atomic weak cache consistency updates */
1561 invalid
|= nfs_wcc_update_inode(inode
, fattr
);
1563 /* More cache consistency checks */
1564 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) {
1565 if (inode
->i_version
!= fattr
->change_attr
) {
1566 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1567 inode
->i_sb
->s_id
, inode
->i_ino
);
1568 invalid
|= NFS_INO_INVALID_ATTR
1569 | NFS_INO_INVALID_DATA
1570 | NFS_INO_INVALID_ACCESS
1571 | NFS_INO_INVALID_ACL
1572 | NFS_INO_REVAL_PAGECACHE
;
1573 if (S_ISDIR(inode
->i_mode
))
1574 nfs_force_lookup_revalidate(inode
);
1575 inode
->i_version
= fattr
->change_attr
;
1577 } else if (server
->caps
& NFS_CAP_CHANGE_ATTR
)
1578 nfsi
->cache_validity
|= save_cache_validity
;
1580 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
) {
1581 memcpy(&inode
->i_mtime
, &fattr
->mtime
, sizeof(inode
->i_mtime
));
1582 } else if (server
->caps
& NFS_CAP_MTIME
)
1583 nfsi
->cache_validity
|= save_cache_validity
&
1584 (NFS_INO_INVALID_ATTR
1585 | NFS_INO_REVAL_FORCED
);
1587 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
) {
1588 memcpy(&inode
->i_ctime
, &fattr
->ctime
, sizeof(inode
->i_ctime
));
1589 } else if (server
->caps
& NFS_CAP_CTIME
)
1590 nfsi
->cache_validity
|= save_cache_validity
&
1591 (NFS_INO_INVALID_ATTR
1592 | NFS_INO_REVAL_FORCED
);
1594 /* Check if our cached file size is stale */
1595 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1596 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1597 cur_isize
= i_size_read(inode
);
1598 if (new_isize
!= cur_isize
) {
1599 /* Do we perhaps have any outstanding writes, or has
1600 * the file grown beyond our last write? */
1601 if ((nfsi
->npages
== 0) || new_isize
> cur_isize
) {
1602 i_size_write(inode
, new_isize
);
1603 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_DATA
;
1605 dprintk("NFS: isize change on server for file %s/%ld "
1609 (long long)cur_isize
,
1610 (long long)new_isize
);
1613 nfsi
->cache_validity
|= save_cache_validity
&
1614 (NFS_INO_INVALID_ATTR
1615 | NFS_INO_REVAL_PAGECACHE
1616 | NFS_INO_REVAL_FORCED
);
1619 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
1620 memcpy(&inode
->i_atime
, &fattr
->atime
, sizeof(inode
->i_atime
));
1621 else if (server
->caps
& NFS_CAP_ATIME
)
1622 nfsi
->cache_validity
|= save_cache_validity
&
1623 (NFS_INO_INVALID_ATIME
1624 | NFS_INO_REVAL_FORCED
);
1626 if (fattr
->valid
& NFS_ATTR_FATTR_MODE
) {
1627 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)) {
1628 umode_t newmode
= inode
->i_mode
& S_IFMT
;
1629 newmode
|= fattr
->mode
& S_IALLUGO
;
1630 inode
->i_mode
= newmode
;
1631 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1633 } else if (server
->caps
& NFS_CAP_MODE
)
1634 nfsi
->cache_validity
|= save_cache_validity
&
1635 (NFS_INO_INVALID_ATTR
1636 | NFS_INO_INVALID_ACCESS
1637 | NFS_INO_INVALID_ACL
1638 | NFS_INO_REVAL_FORCED
);
1640 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
) {
1641 if (!uid_eq(inode
->i_uid
, fattr
->uid
)) {
1642 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1643 inode
->i_uid
= fattr
->uid
;
1645 } else if (server
->caps
& NFS_CAP_OWNER
)
1646 nfsi
->cache_validity
|= save_cache_validity
&
1647 (NFS_INO_INVALID_ATTR
1648 | NFS_INO_INVALID_ACCESS
1649 | NFS_INO_INVALID_ACL
1650 | NFS_INO_REVAL_FORCED
);
1652 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
) {
1653 if (!gid_eq(inode
->i_gid
, fattr
->gid
)) {
1654 invalid
|= NFS_INO_INVALID_ATTR
|NFS_INO_INVALID_ACCESS
|NFS_INO_INVALID_ACL
;
1655 inode
->i_gid
= fattr
->gid
;
1657 } else if (server
->caps
& NFS_CAP_OWNER_GROUP
)
1658 nfsi
->cache_validity
|= save_cache_validity
&
1659 (NFS_INO_INVALID_ATTR
1660 | NFS_INO_INVALID_ACCESS
1661 | NFS_INO_INVALID_ACL
1662 | NFS_INO_REVAL_FORCED
);
1664 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
) {
1665 if (inode
->i_nlink
!= fattr
->nlink
) {
1666 invalid
|= NFS_INO_INVALID_ATTR
;
1667 if (S_ISDIR(inode
->i_mode
))
1668 invalid
|= NFS_INO_INVALID_DATA
;
1669 set_nlink(inode
, fattr
->nlink
);
1671 } else if (server
->caps
& NFS_CAP_NLINK
)
1672 nfsi
->cache_validity
|= save_cache_validity
&
1673 (NFS_INO_INVALID_ATTR
1674 | NFS_INO_REVAL_FORCED
);
1676 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
1678 * report the blocks in 512byte units
1680 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
1682 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
1683 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
1685 /* Update attrtimeo value if we're out of the unstable period */
1686 if (invalid
& NFS_INO_INVALID_ATTR
) {
1687 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
1688 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
1689 nfsi
->attrtimeo_timestamp
= now
;
1690 nfsi
->attr_gencount
= nfs_inc_attr_generation_counter();
1692 if (!time_in_range_open(now
, nfsi
->attrtimeo_timestamp
, nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
1693 if ((nfsi
->attrtimeo
<<= 1) > NFS_MAXATTRTIMEO(inode
))
1694 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
1695 nfsi
->attrtimeo_timestamp
= now
;
1698 invalid
&= ~NFS_INO_INVALID_ATTR
;
1699 /* Don't invalidate the data if we were to blame */
1700 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
1701 || S_ISLNK(inode
->i_mode
)))
1702 invalid
&= ~NFS_INO_INVALID_DATA
;
1703 if (!NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
) ||
1704 (save_cache_validity
& NFS_INO_REVAL_FORCED
))
1705 nfsi
->cache_validity
|= invalid
;
1707 if (invalid
& NFS_INO_INVALID_DATA
)
1708 nfs_fscache_invalidate(inode
);
1713 * No need to worry about unhashing the dentry, as the
1714 * lookup validation will know that the inode is bad.
1715 * (But we fall through to invalidate the caches.)
1717 nfs_invalidate_inode(inode
);
1721 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
1723 struct nfs_inode
*nfsi
;
1724 nfsi
= (struct nfs_inode
*)kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
1728 nfsi
->cache_validity
= 0UL;
1729 #if IS_ENABLED(CONFIG_NFS_V4)
1730 nfsi
->nfs4_acl
= NULL
;
1731 #endif /* CONFIG_NFS_V4 */
1732 return &nfsi
->vfs_inode
;
1734 EXPORT_SYMBOL_GPL(nfs_alloc_inode
);
1736 static void nfs_i_callback(struct rcu_head
*head
)
1738 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
1739 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
1742 void nfs_destroy_inode(struct inode
*inode
)
1744 call_rcu(&inode
->i_rcu
, nfs_i_callback
);
1746 EXPORT_SYMBOL_GPL(nfs_destroy_inode
);
1748 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
1750 #if IS_ENABLED(CONFIG_NFS_V4)
1751 INIT_LIST_HEAD(&nfsi
->open_states
);
1752 nfsi
->delegation
= NULL
;
1753 nfsi
->delegation_state
= 0;
1754 init_rwsem(&nfsi
->rwsem
);
1755 nfsi
->layout
= NULL
;
1759 static void init_once(void *foo
)
1761 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
1763 inode_init_once(&nfsi
->vfs_inode
);
1764 INIT_LIST_HEAD(&nfsi
->open_files
);
1765 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
1766 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
1767 INIT_LIST_HEAD(&nfsi
->commit_info
.list
);
1769 nfsi
->commit_info
.ncommit
= 0;
1770 atomic_set(&nfsi
->commit_info
.rpcs_out
, 0);
1771 atomic_set(&nfsi
->silly_count
, 1);
1772 INIT_HLIST_HEAD(&nfsi
->silly_list
);
1773 init_waitqueue_head(&nfsi
->waitqueue
);
1774 nfs4_init_once(nfsi
);
1777 static int __init
nfs_init_inodecache(void)
1779 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
1780 sizeof(struct nfs_inode
),
1781 0, (SLAB_RECLAIM_ACCOUNT
|
1784 if (nfs_inode_cachep
== NULL
)
1790 static void nfs_destroy_inodecache(void)
1793 * Make sure all delayed rcu free inodes are flushed before we
1797 kmem_cache_destroy(nfs_inode_cachep
);
1800 struct workqueue_struct
*nfsiod_workqueue
;
1801 EXPORT_SYMBOL_GPL(nfsiod_workqueue
);
1804 * start up the nfsiod workqueue
1806 static int nfsiod_start(void)
1808 struct workqueue_struct
*wq
;
1809 dprintk("RPC: creating workqueue nfsiod\n");
1810 wq
= alloc_workqueue("nfsiod", WQ_MEM_RECLAIM
, 0);
1813 nfsiod_workqueue
= wq
;
1818 * Destroy the nfsiod workqueue
1820 static void nfsiod_stop(void)
1822 struct workqueue_struct
*wq
;
1824 wq
= nfsiod_workqueue
;
1827 nfsiod_workqueue
= NULL
;
1828 destroy_workqueue(wq
);
1832 EXPORT_SYMBOL_GPL(nfs_net_id
);
1834 static int nfs_net_init(struct net
*net
)
1836 nfs_clients_init(net
);
1840 static void nfs_net_exit(struct net
*net
)
1842 nfs_cleanup_cb_ident_idr(net
);
1845 static struct pernet_operations nfs_net_ops
= {
1846 .init
= nfs_net_init
,
1847 .exit
= nfs_net_exit
,
1849 .size
= sizeof(struct nfs_net
),
1855 static int __init
init_nfs_fs(void)
1859 err
= register_pernet_subsys(&nfs_net_ops
);
1863 err
= nfs_fscache_register();
1867 err
= nfsiod_start();
1871 err
= nfs_fs_proc_init();
1875 err
= nfs_init_nfspagecache();
1879 err
= nfs_init_inodecache();
1883 err
= nfs_init_readpagecache();
1887 err
= nfs_init_writepagecache();
1891 err
= nfs_init_directcache();
1895 #ifdef CONFIG_PROC_FS
1896 rpc_proc_register(&init_net
, &nfs_rpcstat
);
1898 if ((err
= register_nfs_fs()) != 0)
1903 #ifdef CONFIG_PROC_FS
1904 rpc_proc_unregister(&init_net
, "nfs");
1906 nfs_destroy_directcache();
1908 nfs_destroy_writepagecache();
1910 nfs_destroy_readpagecache();
1912 nfs_destroy_inodecache();
1914 nfs_destroy_nfspagecache();
1920 nfs_fscache_unregister();
1922 unregister_pernet_subsys(&nfs_net_ops
);
1927 static void __exit
exit_nfs_fs(void)
1929 nfs_destroy_directcache();
1930 nfs_destroy_writepagecache();
1931 nfs_destroy_readpagecache();
1932 nfs_destroy_inodecache();
1933 nfs_destroy_nfspagecache();
1934 nfs_fscache_unregister();
1935 unregister_pernet_subsys(&nfs_net_ops
);
1936 #ifdef CONFIG_PROC_FS
1937 rpc_proc_unregister(&init_net
, "nfs");
1939 unregister_nfs_fs();
1944 /* Not quite true; I just maintain it */
1945 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1946 MODULE_LICENSE("GPL");
1947 module_param(enable_ino64
, bool, 0644);
1949 module_init(init_nfs_fs
)
1950 module_exit(exit_nfs_fs
)