1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs inode and superblock handling functions
9 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13 * J.S.Peatfield@damtp.cam.ac.uk
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
47 #include "delegation.h"
58 #define NFSDBG_FACILITY NFSDBG_VFS
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64
= NFS_64_BIT_INODE_NUMBERS_ENABLED
;
65 static int nfs_update_inode(struct inode
*, struct nfs_fattr
*);
67 static struct kmem_cache
* nfs_inode_cachep
;
69 static inline unsigned long
70 nfs_fattr_to_ino_t(struct nfs_fattr
*fattr
)
72 return nfs_fileid_to_ino_t(fattr
->fileid
);
75 static int nfs_wait_killable(int mode
)
77 freezable_schedule_unsafe();
78 if (signal_pending_state(mode
, current
))
83 int nfs_wait_bit_killable(struct wait_bit_key
*key
, int mode
)
85 return nfs_wait_killable(mode
);
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable
);
90 * nfs_compat_user_ino64 - returns the user-visible inode number
91 * @fileid: 64-bit fileid
93 * This function returns a 32-bit inode number if the boot parameter
94 * nfs.enable_ino64 is zero.
96 u64
nfs_compat_user_ino64(u64 fileid
)
107 if (sizeof(ino
) < sizeof(fileid
))
108 ino
^= fileid
>> (sizeof(fileid
)-sizeof(ino
)) * 8;
112 int nfs_drop_inode(struct inode
*inode
)
114 return NFS_STALE(inode
) || generic_drop_inode(inode
);
116 EXPORT_SYMBOL_GPL(nfs_drop_inode
);
118 void nfs_clear_inode(struct inode
*inode
)
121 * The following should never happen...
123 WARN_ON_ONCE(nfs_have_writebacks(inode
));
124 WARN_ON_ONCE(!list_empty(&NFS_I(inode
)->open_files
));
125 nfs_zap_acl_cache(inode
);
126 nfs_access_zap_cache(inode
);
127 nfs_fscache_clear_inode(inode
);
129 EXPORT_SYMBOL_GPL(nfs_clear_inode
);
131 void nfs_evict_inode(struct inode
*inode
)
133 truncate_inode_pages_final(&inode
->i_data
);
135 nfs_clear_inode(inode
);
138 int nfs_sync_inode(struct inode
*inode
)
140 inode_dio_wait(inode
);
141 return nfs_wb_all(inode
);
143 EXPORT_SYMBOL_GPL(nfs_sync_inode
);
146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 * @mapping: pointer to struct address_space
149 int nfs_sync_mapping(struct address_space
*mapping
)
153 if (mapping
->nrpages
!= 0) {
154 unmap_mapping_range(mapping
, 0, 0, 0);
155 ret
= nfs_wb_all(mapping
->host
);
160 static int nfs_attribute_timeout(struct inode
*inode
)
162 struct nfs_inode
*nfsi
= NFS_I(inode
);
164 return !time_in_range_open(jiffies
, nfsi
->read_cache_jiffies
, nfsi
->read_cache_jiffies
+ nfsi
->attrtimeo
);
167 static bool nfs_check_cache_flags_invalid(struct inode
*inode
,
170 unsigned long cache_validity
= READ_ONCE(NFS_I(inode
)->cache_validity
);
172 return (cache_validity
& flags
) != 0;
175 bool nfs_check_cache_invalid(struct inode
*inode
, unsigned long flags
)
177 if (nfs_check_cache_flags_invalid(inode
, flags
))
179 return nfs_attribute_cache_expired(inode
);
181 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid
);
183 #ifdef CONFIG_NFS_V4_2
184 static bool nfs_has_xattr_cache(const struct nfs_inode
*nfsi
)
186 return nfsi
->xattr_cache
!= NULL
;
189 static bool nfs_has_xattr_cache(const struct nfs_inode
*nfsi
)
195 void nfs_set_cache_invalid(struct inode
*inode
, unsigned long flags
)
197 struct nfs_inode
*nfsi
= NFS_I(inode
);
198 bool have_delegation
= NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
);
200 if (have_delegation
) {
201 if (!(flags
& NFS_INO_REVAL_FORCED
))
202 flags
&= ~(NFS_INO_INVALID_MODE
|
203 NFS_INO_INVALID_OTHER
|
204 NFS_INO_INVALID_XATTR
);
205 flags
&= ~(NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_SIZE
);
206 } else if (flags
& NFS_INO_REVAL_PAGECACHE
)
207 flags
|= NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_SIZE
;
209 if (!nfs_has_xattr_cache(nfsi
))
210 flags
&= ~NFS_INO_INVALID_XATTR
;
211 if (flags
& NFS_INO_INVALID_DATA
)
212 nfs_fscache_invalidate(inode
);
213 flags
&= ~(NFS_INO_REVAL_PAGECACHE
| NFS_INO_REVAL_FORCED
);
215 nfsi
->cache_validity
|= flags
;
217 if (inode
->i_mapping
->nrpages
== 0)
218 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_DATA
|
219 NFS_INO_DATA_INVAL_DEFER
);
220 else if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
221 nfsi
->cache_validity
&= ~NFS_INO_DATA_INVAL_DEFER
;
223 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid
);
226 * Invalidate the local caches
228 static void nfs_zap_caches_locked(struct inode
*inode
)
230 struct nfs_inode
*nfsi
= NFS_I(inode
);
231 int mode
= inode
->i_mode
;
233 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
235 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
236 nfsi
->attrtimeo_timestamp
= jiffies
;
238 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
239 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATTR
240 | NFS_INO_INVALID_DATA
241 | NFS_INO_INVALID_ACCESS
242 | NFS_INO_INVALID_ACL
243 | NFS_INO_INVALID_XATTR
244 | NFS_INO_REVAL_PAGECACHE
);
246 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATTR
247 | NFS_INO_INVALID_ACCESS
248 | NFS_INO_INVALID_ACL
249 | NFS_INO_INVALID_XATTR
250 | NFS_INO_REVAL_PAGECACHE
);
251 nfs_zap_label_cache_locked(nfsi
);
254 void nfs_zap_caches(struct inode
*inode
)
256 spin_lock(&inode
->i_lock
);
257 nfs_zap_caches_locked(inode
);
258 spin_unlock(&inode
->i_lock
);
261 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
263 if (mapping
->nrpages
!= 0) {
264 spin_lock(&inode
->i_lock
);
265 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
266 spin_unlock(&inode
->i_lock
);
270 void nfs_zap_acl_cache(struct inode
*inode
)
272 void (*clear_acl_cache
)(struct inode
*);
274 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
275 if (clear_acl_cache
!= NULL
)
276 clear_acl_cache(inode
);
277 spin_lock(&inode
->i_lock
);
278 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
279 spin_unlock(&inode
->i_lock
);
281 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache
);
283 void nfs_invalidate_atime(struct inode
*inode
)
285 spin_lock(&inode
->i_lock
);
286 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
287 spin_unlock(&inode
->i_lock
);
289 EXPORT_SYMBOL_GPL(nfs_invalidate_atime
);
292 * Invalidate, but do not unhash, the inode.
293 * NB: must be called with inode->i_lock held!
295 static void nfs_set_inode_stale_locked(struct inode
*inode
)
297 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
298 nfs_zap_caches_locked(inode
);
299 trace_nfs_set_inode_stale(inode
);
302 void nfs_set_inode_stale(struct inode
*inode
)
304 spin_lock(&inode
->i_lock
);
305 nfs_set_inode_stale_locked(inode
);
306 spin_unlock(&inode
->i_lock
);
309 struct nfs_find_desc
{
311 struct nfs_fattr
*fattr
;
315 * In NFSv3 we can have 64bit inode numbers. In order to support
316 * this, and re-exported directories (also seen in NFSv2)
317 * we are forced to allow 2 different inodes to have the same
321 nfs_find_actor(struct inode
*inode
, void *opaque
)
323 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
324 struct nfs_fh
*fh
= desc
->fh
;
325 struct nfs_fattr
*fattr
= desc
->fattr
;
327 if (NFS_FILEID(inode
) != fattr
->fileid
)
329 if (inode_wrong_type(inode
, fattr
->mode
))
331 if (nfs_compare_fh(NFS_FH(inode
), fh
))
333 if (is_bad_inode(inode
) || NFS_STALE(inode
))
339 nfs_init_locked(struct inode
*inode
, void *opaque
)
341 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
342 struct nfs_fattr
*fattr
= desc
->fattr
;
344 set_nfs_fileid(inode
, fattr
->fileid
);
345 inode
->i_mode
= fattr
->mode
;
346 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
350 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
351 static void nfs_clear_label_invalid(struct inode
*inode
)
353 spin_lock(&inode
->i_lock
);
354 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_LABEL
;
355 spin_unlock(&inode
->i_lock
);
358 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
359 struct nfs4_label
*label
)
366 if ((fattr
->valid
& NFS_ATTR_FATTR_V4_SECURITY_LABEL
) && inode
->i_security
) {
367 error
= security_inode_notifysecctx(inode
, label
->label
,
370 printk(KERN_ERR
"%s() %s %d "
371 "security_inode_notifysecctx() %d\n",
373 (char *)label
->label
,
375 nfs_clear_label_invalid(inode
);
379 struct nfs4_label
*nfs4_label_alloc(struct nfs_server
*server
, gfp_t flags
)
381 struct nfs4_label
*label
= NULL
;
382 int minor_version
= server
->nfs_client
->cl_minorversion
;
384 if (minor_version
< 2)
387 if (!(server
->caps
& NFS_CAP_SECURITY_LABEL
))
390 label
= kzalloc(sizeof(struct nfs4_label
), flags
);
392 return ERR_PTR(-ENOMEM
);
394 label
->label
= kzalloc(NFS4_MAXLABELLEN
, flags
);
395 if (label
->label
== NULL
) {
397 return ERR_PTR(-ENOMEM
);
399 label
->len
= NFS4_MAXLABELLEN
;
403 EXPORT_SYMBOL_GPL(nfs4_label_alloc
);
405 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
406 struct nfs4_label
*label
)
410 EXPORT_SYMBOL_GPL(nfs_setsecurity
);
412 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
414 nfs_ilookup(struct super_block
*sb
, struct nfs_fattr
*fattr
, struct nfs_fh
*fh
)
416 struct nfs_find_desc desc
= {
423 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
) ||
424 !(fattr
->valid
& NFS_ATTR_FATTR_TYPE
))
427 hash
= nfs_fattr_to_ino_t(fattr
);
428 inode
= ilookup5(sb
, hash
, nfs_find_actor
, &desc
);
430 dprintk("%s: returning %p\n", __func__
, inode
);
435 * This is our front-end to iget that looks up inodes by file handle
436 * instead of inode number.
439 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
, struct nfs4_label
*label
)
441 struct nfs_find_desc desc
= {
445 struct inode
*inode
= ERR_PTR(-ENOENT
);
446 u64 fattr_supported
= NFS_SB(sb
)->fattr_valid
;
449 nfs_attr_check_mountpoint(sb
, fattr
);
451 if (nfs_attr_use_mounted_on_fileid(fattr
))
452 fattr
->fileid
= fattr
->mounted_on_fileid
;
453 else if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) == 0)
455 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) == 0)
458 hash
= nfs_fattr_to_ino_t(fattr
);
460 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
462 inode
= ERR_PTR(-ENOMEM
);
466 if (inode
->i_state
& I_NEW
) {
467 struct nfs_inode
*nfsi
= NFS_I(inode
);
468 unsigned long now
= jiffies
;
470 /* We set i_ino for the few things that still rely on it,
474 /* We can't support update_atime(), since the server will reset it */
475 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
476 inode
->i_mode
= fattr
->mode
;
477 nfsi
->cache_validity
= 0;
478 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) == 0
479 && (fattr_supported
& NFS_ATTR_FATTR_MODE
))
480 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MODE
);
481 /* Why so? Because we want revalidate for devices/FIFOs, and
482 * that's precisely what we have in nfs_file_inode_operations.
484 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
485 if (S_ISREG(inode
->i_mode
)) {
486 inode
->i_fop
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_ops
;
487 inode
->i_data
.a_ops
= &nfs_file_aops
;
488 } else if (S_ISDIR(inode
->i_mode
)) {
489 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
490 inode
->i_fop
= &nfs_dir_operations
;
491 inode
->i_data
.a_ops
= &nfs_dir_aops
;
492 /* Deal with crossing mountpoints */
493 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTPOINT
||
494 fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
) {
495 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
496 inode
->i_op
= &nfs_referral_inode_operations
;
498 inode
->i_op
= &nfs_mountpoint_inode_operations
;
500 inode
->i_flags
|= S_AUTOMOUNT
;
502 } else if (S_ISLNK(inode
->i_mode
)) {
503 inode
->i_op
= &nfs_symlink_inode_operations
;
504 inode_nohighmem(inode
);
506 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
508 memset(&inode
->i_atime
, 0, sizeof(inode
->i_atime
));
509 memset(&inode
->i_mtime
, 0, sizeof(inode
->i_mtime
));
510 memset(&inode
->i_ctime
, 0, sizeof(inode
->i_ctime
));
511 inode_set_iversion_raw(inode
, 0);
514 inode
->i_uid
= make_kuid(&init_user_ns
, -2);
515 inode
->i_gid
= make_kgid(&init_user_ns
, -2);
517 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
521 nfsi
->read_cache_jiffies
= fattr
->time_start
;
522 nfsi
->attr_gencount
= fattr
->gencount
;
523 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
524 inode
->i_atime
= fattr
->atime
;
525 else if (fattr_supported
& NFS_ATTR_FATTR_ATIME
)
526 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
527 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
528 inode
->i_mtime
= fattr
->mtime
;
529 else if (fattr_supported
& NFS_ATTR_FATTR_MTIME
)
530 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
);
531 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
532 inode
->i_ctime
= fattr
->ctime
;
533 else if (fattr_supported
& NFS_ATTR_FATTR_CTIME
)
534 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CTIME
);
535 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
536 inode_set_iversion_raw(inode
, fattr
->change_attr
);
538 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
);
539 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
540 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
542 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_SIZE
);
543 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
)
544 set_nlink(inode
, fattr
->nlink
);
545 else if (fattr_supported
& NFS_ATTR_FATTR_NLINK
)
546 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_NLINK
);
547 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
)
548 inode
->i_uid
= fattr
->uid
;
549 else if (fattr_supported
& NFS_ATTR_FATTR_OWNER
)
550 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_OTHER
);
551 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
)
552 inode
->i_gid
= fattr
->gid
;
553 else if (fattr_supported
& NFS_ATTR_FATTR_GROUP
)
554 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_OTHER
);
555 if (nfs_server_capable(inode
, NFS_CAP_XATTR
))
556 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_XATTR
);
557 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
558 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
559 else if (fattr_supported
& NFS_ATTR_FATTR_BLOCKS_USED
&&
561 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_BLOCKS
);
562 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
564 * report the blocks in 512byte units
566 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
567 } else if (fattr_supported
& NFS_ATTR_FATTR_SPACE_USED
&&
569 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_BLOCKS
);
571 nfs_setsecurity(inode
, fattr
, label
);
573 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
574 nfsi
->attrtimeo_timestamp
= now
;
575 nfsi
->access_cache
= RB_ROOT
;
577 nfs_fscache_init_inode(inode
);
579 unlock_new_inode(inode
);
581 int err
= nfs_refresh_inode(inode
, fattr
);
584 inode
= ERR_PTR(err
);
588 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
590 (unsigned long long)NFS_FILEID(inode
),
591 nfs_display_fhandle_hash(fh
),
592 atomic_read(&inode
->i_count
));
598 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
601 EXPORT_SYMBOL_GPL(nfs_fhget
);
603 #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)
606 nfs_setattr(struct user_namespace
*mnt_userns
, struct dentry
*dentry
,
609 struct inode
*inode
= d_inode(dentry
);
610 struct nfs_fattr
*fattr
;
613 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
615 /* skip mode change if it's just for clearing setuid/setgid */
616 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
617 attr
->ia_valid
&= ~ATTR_MODE
;
619 if (attr
->ia_valid
& ATTR_SIZE
) {
620 BUG_ON(!S_ISREG(inode
->i_mode
));
622 error
= inode_newsize_ok(inode
, attr
->ia_size
);
626 if (attr
->ia_size
== i_size_read(inode
))
627 attr
->ia_valid
&= ~ATTR_SIZE
;
630 /* Optimization: if the end result is no change, don't RPC */
631 if (((attr
->ia_valid
& NFS_VALID_ATTRS
) & ~(ATTR_FILE
|ATTR_OPEN
)) == 0)
634 trace_nfs_setattr_enter(inode
);
636 /* Write all dirty data */
637 if (S_ISREG(inode
->i_mode
))
638 nfs_sync_inode(inode
);
640 fattr
= nfs_alloc_fattr();
646 error
= NFS_PROTO(inode
)->setattr(dentry
, fattr
, attr
);
648 error
= nfs_refresh_inode(inode
, fattr
);
649 nfs_free_fattr(fattr
);
651 trace_nfs_setattr_exit(inode
, error
);
654 EXPORT_SYMBOL_GPL(nfs_setattr
);
657 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
658 * @inode: inode of the file used
659 * @offset: file offset to start truncating
661 * This is a copy of the common vmtruncate, but with the locking
662 * corrected to take into account the fact that NFS requires
663 * inode->i_size to be updated under the inode->i_lock.
664 * Note: must be called with inode->i_lock held!
666 static int nfs_vmtruncate(struct inode
* inode
, loff_t offset
)
670 err
= inode_newsize_ok(inode
, offset
);
674 i_size_write(inode
, offset
);
677 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_DATA
|
678 NFS_INO_DATA_INVAL_DEFER
);
679 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_SIZE
;
681 spin_unlock(&inode
->i_lock
);
682 truncate_pagecache(inode
, offset
);
683 spin_lock(&inode
->i_lock
);
689 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
690 * @inode: pointer to struct inode
691 * @attr: pointer to struct iattr
692 * @fattr: pointer to struct nfs_fattr
694 * Note: we do this in the *proc.c in order to ensure that
695 * it works for things like exclusive creates too.
697 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
,
698 struct nfs_fattr
*fattr
)
700 /* Barrier: bump the attribute generation count. */
701 nfs_fattr_set_barrier(fattr
);
703 spin_lock(&inode
->i_lock
);
704 NFS_I(inode
)->attr_gencount
= fattr
->gencount
;
705 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
706 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
|
707 NFS_INO_INVALID_BLOCKS
);
708 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
709 nfs_vmtruncate(inode
, attr
->ia_size
);
711 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
712 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_CTIME
;
713 if ((attr
->ia_valid
& ATTR_KILL_SUID
) != 0 &&
714 inode
->i_mode
& S_ISUID
)
715 inode
->i_mode
&= ~S_ISUID
;
716 if ((attr
->ia_valid
& ATTR_KILL_SGID
) != 0 &&
717 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) ==
719 inode
->i_mode
&= ~S_ISGID
;
720 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
721 int mode
= attr
->ia_mode
& S_IALLUGO
;
722 mode
|= inode
->i_mode
& ~S_IALLUGO
;
723 inode
->i_mode
= mode
;
725 if ((attr
->ia_valid
& ATTR_UID
) != 0)
726 inode
->i_uid
= attr
->ia_uid
;
727 if ((attr
->ia_valid
& ATTR_GID
) != 0)
728 inode
->i_gid
= attr
->ia_gid
;
729 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
730 inode
->i_ctime
= fattr
->ctime
;
732 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
733 | NFS_INO_INVALID_CTIME
);
734 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
735 | NFS_INO_INVALID_ACL
);
737 if (attr
->ia_valid
& (ATTR_ATIME_SET
|ATTR_ATIME
)) {
738 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_ATIME
739 | NFS_INO_INVALID_CTIME
);
740 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
741 inode
->i_atime
= fattr
->atime
;
742 else if (attr
->ia_valid
& ATTR_ATIME_SET
)
743 inode
->i_atime
= attr
->ia_atime
;
745 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
747 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
748 inode
->i_ctime
= fattr
->ctime
;
750 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
751 | NFS_INO_INVALID_CTIME
);
753 if (attr
->ia_valid
& (ATTR_MTIME_SET
|ATTR_MTIME
)) {
754 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_MTIME
755 | NFS_INO_INVALID_CTIME
);
756 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
757 inode
->i_mtime
= fattr
->mtime
;
758 else if (attr
->ia_valid
& ATTR_MTIME_SET
)
759 inode
->i_mtime
= attr
->ia_mtime
;
761 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
);
763 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
764 inode
->i_ctime
= fattr
->ctime
;
766 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
767 | NFS_INO_INVALID_CTIME
);
770 nfs_update_inode(inode
, fattr
);
771 spin_unlock(&inode
->i_lock
);
773 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode
);
775 static void nfs_readdirplus_parent_cache_miss(struct dentry
*dentry
)
777 struct dentry
*parent
;
779 if (!nfs_server_capable(d_inode(dentry
), NFS_CAP_READDIRPLUS
))
781 parent
= dget_parent(dentry
);
782 nfs_force_use_readdirplus(d_inode(parent
));
786 static void nfs_readdirplus_parent_cache_hit(struct dentry
*dentry
)
788 struct dentry
*parent
;
790 if (!nfs_server_capable(d_inode(dentry
), NFS_CAP_READDIRPLUS
))
792 parent
= dget_parent(dentry
);
793 nfs_advise_use_readdirplus(d_inode(parent
));
797 static u32
nfs_get_valid_attrmask(struct inode
*inode
)
799 unsigned long cache_validity
= READ_ONCE(NFS_I(inode
)->cache_validity
);
800 u32 reply_mask
= STATX_INO
| STATX_TYPE
;
802 if (!(cache_validity
& NFS_INO_INVALID_ATIME
))
803 reply_mask
|= STATX_ATIME
;
804 if (!(cache_validity
& NFS_INO_INVALID_CTIME
))
805 reply_mask
|= STATX_CTIME
;
806 if (!(cache_validity
& NFS_INO_INVALID_MTIME
))
807 reply_mask
|= STATX_MTIME
;
808 if (!(cache_validity
& NFS_INO_INVALID_SIZE
))
809 reply_mask
|= STATX_SIZE
;
810 if (!(cache_validity
& NFS_INO_INVALID_NLINK
))
811 reply_mask
|= STATX_NLINK
;
812 if (!(cache_validity
& NFS_INO_INVALID_MODE
))
813 reply_mask
|= STATX_MODE
;
814 if (!(cache_validity
& NFS_INO_INVALID_OTHER
))
815 reply_mask
|= STATX_UID
| STATX_GID
;
816 if (!(cache_validity
& NFS_INO_INVALID_BLOCKS
))
817 reply_mask
|= STATX_BLOCKS
;
821 int nfs_getattr(struct user_namespace
*mnt_userns
, const struct path
*path
,
822 struct kstat
*stat
, u32 request_mask
, unsigned int query_flags
)
824 struct inode
*inode
= d_inode(path
->dentry
);
825 struct nfs_server
*server
= NFS_SERVER(inode
);
826 unsigned long cache_validity
;
828 bool force_sync
= query_flags
& AT_STATX_FORCE_SYNC
;
829 bool do_update
= false;
831 trace_nfs_getattr_enter(inode
);
833 request_mask
&= STATX_TYPE
| STATX_MODE
| STATX_NLINK
| STATX_UID
|
834 STATX_GID
| STATX_ATIME
| STATX_MTIME
| STATX_CTIME
|
835 STATX_INO
| STATX_SIZE
| STATX_BLOCKS
;
837 if ((query_flags
& AT_STATX_DONT_SYNC
) && !force_sync
) {
838 nfs_readdirplus_parent_cache_hit(path
->dentry
);
839 goto out_no_revalidate
;
842 /* Flush out writes to the server in order to update c/mtime. */
843 if ((request_mask
& (STATX_CTIME
|STATX_MTIME
)) &&
844 S_ISREG(inode
->i_mode
)) {
845 err
= filemap_write_and_wait(inode
->i_mapping
);
851 * We may force a getattr if the user cares about atime.
853 * Note that we only have to check the vfsmount flags here:
854 * - NFS always sets S_NOATIME by so checking it would give a
856 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
857 * no point in checking those.
859 if ((path
->mnt
->mnt_flags
& MNT_NOATIME
) ||
860 ((path
->mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
861 request_mask
&= ~STATX_ATIME
;
863 /* Is the user requesting attributes that might need revalidation? */
864 if (!(request_mask
& (STATX_MODE
|STATX_NLINK
|STATX_ATIME
|STATX_CTIME
|
865 STATX_MTIME
|STATX_UID
|STATX_GID
|
866 STATX_SIZE
|STATX_BLOCKS
)))
867 goto out_no_revalidate
;
869 /* Check whether the cached attributes are stale */
870 do_update
|= force_sync
|| nfs_attribute_cache_expired(inode
);
871 cache_validity
= READ_ONCE(NFS_I(inode
)->cache_validity
);
872 do_update
|= cache_validity
& NFS_INO_INVALID_CHANGE
;
873 if (request_mask
& STATX_ATIME
)
874 do_update
|= cache_validity
& NFS_INO_INVALID_ATIME
;
875 if (request_mask
& STATX_CTIME
)
876 do_update
|= cache_validity
& NFS_INO_INVALID_CTIME
;
877 if (request_mask
& STATX_MTIME
)
878 do_update
|= cache_validity
& NFS_INO_INVALID_MTIME
;
879 if (request_mask
& STATX_SIZE
)
880 do_update
|= cache_validity
& NFS_INO_INVALID_SIZE
;
881 if (request_mask
& STATX_NLINK
)
882 do_update
|= cache_validity
& NFS_INO_INVALID_NLINK
;
883 if (request_mask
& STATX_MODE
)
884 do_update
|= cache_validity
& NFS_INO_INVALID_MODE
;
885 if (request_mask
& (STATX_UID
| STATX_GID
))
886 do_update
|= cache_validity
& NFS_INO_INVALID_OTHER
;
887 if (request_mask
& STATX_BLOCKS
)
888 do_update
|= cache_validity
& NFS_INO_INVALID_BLOCKS
;
891 /* Update the attribute cache */
892 if (!(server
->flags
& NFS_MOUNT_NOAC
))
893 nfs_readdirplus_parent_cache_miss(path
->dentry
);
895 nfs_readdirplus_parent_cache_hit(path
->dentry
);
896 err
= __nfs_revalidate_inode(server
, inode
);
900 nfs_readdirplus_parent_cache_hit(path
->dentry
);
902 /* Only return attributes that were revalidated. */
903 stat
->result_mask
= nfs_get_valid_attrmask(inode
) | request_mask
;
905 generic_fillattr(&init_user_ns
, inode
, stat
);
906 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
907 if (S_ISDIR(inode
->i_mode
))
908 stat
->blksize
= NFS_SERVER(inode
)->dtsize
;
910 trace_nfs_getattr_exit(inode
, err
);
913 EXPORT_SYMBOL_GPL(nfs_getattr
);
915 static void nfs_init_lock_context(struct nfs_lock_context
*l_ctx
)
917 refcount_set(&l_ctx
->count
, 1);
918 l_ctx
->lockowner
= current
->files
;
919 INIT_LIST_HEAD(&l_ctx
->list
);
920 atomic_set(&l_ctx
->io_count
, 0);
923 static struct nfs_lock_context
*__nfs_find_lock_context(struct nfs_open_context
*ctx
)
925 struct nfs_lock_context
*pos
;
927 list_for_each_entry_rcu(pos
, &ctx
->lock_context
.list
, list
) {
928 if (pos
->lockowner
!= current
->files
)
930 if (refcount_inc_not_zero(&pos
->count
))
936 struct nfs_lock_context
*nfs_get_lock_context(struct nfs_open_context
*ctx
)
938 struct nfs_lock_context
*res
, *new = NULL
;
939 struct inode
*inode
= d_inode(ctx
->dentry
);
942 res
= __nfs_find_lock_context(ctx
);
945 new = kmalloc(sizeof(*new), GFP_KERNEL
);
947 return ERR_PTR(-ENOMEM
);
948 nfs_init_lock_context(new);
949 spin_lock(&inode
->i_lock
);
950 res
= __nfs_find_lock_context(ctx
);
952 new->open_context
= get_nfs_open_context(ctx
);
953 if (new->open_context
) {
954 list_add_tail_rcu(&new->list
,
955 &ctx
->lock_context
.list
);
959 res
= ERR_PTR(-EBADF
);
961 spin_unlock(&inode
->i_lock
);
966 EXPORT_SYMBOL_GPL(nfs_get_lock_context
);
968 void nfs_put_lock_context(struct nfs_lock_context
*l_ctx
)
970 struct nfs_open_context
*ctx
= l_ctx
->open_context
;
971 struct inode
*inode
= d_inode(ctx
->dentry
);
973 if (!refcount_dec_and_lock(&l_ctx
->count
, &inode
->i_lock
))
975 list_del_rcu(&l_ctx
->list
);
976 spin_unlock(&inode
->i_lock
);
977 put_nfs_open_context(ctx
);
978 kfree_rcu(l_ctx
, rcu_head
);
980 EXPORT_SYMBOL_GPL(nfs_put_lock_context
);
983 * nfs_close_context - Common close_context() routine NFSv2/v3
984 * @ctx: pointer to context
985 * @is_sync: is this a synchronous close
987 * Ensure that the attributes are up to date if we're mounted
988 * with close-to-open semantics and we have cached data that will
989 * need to be revalidated on open.
991 void nfs_close_context(struct nfs_open_context
*ctx
, int is_sync
)
993 struct nfs_inode
*nfsi
;
996 if (!(ctx
->mode
& FMODE_WRITE
))
1000 inode
= d_inode(ctx
->dentry
);
1001 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
))
1003 nfsi
= NFS_I(inode
);
1004 if (inode
->i_mapping
->nrpages
== 0)
1006 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1008 if (!list_empty(&nfsi
->open_files
))
1010 if (NFS_SERVER(inode
)->flags
& NFS_MOUNT_NOCTO
)
1012 nfs_revalidate_inode(inode
,
1013 NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_SIZE
);
1015 EXPORT_SYMBOL_GPL(nfs_close_context
);
1017 struct nfs_open_context
*alloc_nfs_open_context(struct dentry
*dentry
,
1021 struct nfs_open_context
*ctx
;
1023 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
1025 return ERR_PTR(-ENOMEM
);
1026 nfs_sb_active(dentry
->d_sb
);
1027 ctx
->dentry
= dget(dentry
);
1029 ctx
->cred
= get_cred(filp
->f_cred
);
1031 ctx
->cred
= get_current_cred();
1032 ctx
->ll_cred
= NULL
;
1037 ctx
->flock_owner
= (fl_owner_t
)filp
;
1038 nfs_init_lock_context(&ctx
->lock_context
);
1039 ctx
->lock_context
.open_context
= ctx
;
1040 INIT_LIST_HEAD(&ctx
->list
);
1041 ctx
->mdsthreshold
= NULL
;
1044 EXPORT_SYMBOL_GPL(alloc_nfs_open_context
);
1046 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
1048 if (ctx
!= NULL
&& refcount_inc_not_zero(&ctx
->lock_context
.count
))
1052 EXPORT_SYMBOL_GPL(get_nfs_open_context
);
1054 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int is_sync
)
1056 struct inode
*inode
= d_inode(ctx
->dentry
);
1057 struct super_block
*sb
= ctx
->dentry
->d_sb
;
1059 if (!refcount_dec_and_test(&ctx
->lock_context
.count
))
1061 if (!list_empty(&ctx
->list
)) {
1062 spin_lock(&inode
->i_lock
);
1063 list_del_rcu(&ctx
->list
);
1064 spin_unlock(&inode
->i_lock
);
1067 NFS_PROTO(inode
)->close_context(ctx
, is_sync
);
1068 put_cred(ctx
->cred
);
1070 nfs_sb_deactive(sb
);
1071 put_rpccred(ctx
->ll_cred
);
1072 kfree(ctx
->mdsthreshold
);
1073 kfree_rcu(ctx
, rcu_head
);
1076 void put_nfs_open_context(struct nfs_open_context
*ctx
)
1078 __put_nfs_open_context(ctx
, 0);
1080 EXPORT_SYMBOL_GPL(put_nfs_open_context
);
1082 static void put_nfs_open_context_sync(struct nfs_open_context
*ctx
)
1084 __put_nfs_open_context(ctx
, 1);
1088 * Ensure that mmap has a recent RPC credential for use when writing out
1091 void nfs_inode_attach_open_context(struct nfs_open_context
*ctx
)
1093 struct inode
*inode
= d_inode(ctx
->dentry
);
1094 struct nfs_inode
*nfsi
= NFS_I(inode
);
1096 spin_lock(&inode
->i_lock
);
1097 if (list_empty(&nfsi
->open_files
) &&
1098 (nfsi
->cache_validity
& NFS_INO_DATA_INVAL_DEFER
))
1099 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
|
1100 NFS_INO_REVAL_FORCED
);
1101 list_add_tail_rcu(&ctx
->list
, &nfsi
->open_files
);
1102 spin_unlock(&inode
->i_lock
);
1104 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context
);
1106 void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
1108 filp
->private_data
= get_nfs_open_context(ctx
);
1109 set_bit(NFS_CONTEXT_FILE_OPEN
, &ctx
->flags
);
1110 if (list_empty(&ctx
->list
))
1111 nfs_inode_attach_open_context(ctx
);
1113 EXPORT_SYMBOL_GPL(nfs_file_set_open_context
);
1116 * Given an inode, search for an open context with the desired characteristics
1118 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, const struct cred
*cred
, fmode_t mode
)
1120 struct nfs_inode
*nfsi
= NFS_I(inode
);
1121 struct nfs_open_context
*pos
, *ctx
= NULL
;
1124 list_for_each_entry_rcu(pos
, &nfsi
->open_files
, list
) {
1125 if (cred
!= NULL
&& cred_fscmp(pos
->cred
, cred
) != 0)
1127 if ((pos
->mode
& (FMODE_READ
|FMODE_WRITE
)) != mode
)
1129 if (!test_bit(NFS_CONTEXT_FILE_OPEN
, &pos
->flags
))
1131 ctx
= get_nfs_open_context(pos
);
1139 void nfs_file_clear_open_context(struct file
*filp
)
1141 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
1144 struct inode
*inode
= d_inode(ctx
->dentry
);
1146 clear_bit(NFS_CONTEXT_FILE_OPEN
, &ctx
->flags
);
1148 * We fatal error on write before. Try to writeback
1152 invalidate_inode_pages2(inode
->i_mapping
);
1153 filp
->private_data
= NULL
;
1154 put_nfs_open_context_sync(ctx
);
1159 * These allocate and release file read/write context information.
1161 int nfs_open(struct inode
*inode
, struct file
*filp
)
1163 struct nfs_open_context
*ctx
;
1165 ctx
= alloc_nfs_open_context(file_dentry(filp
), filp
->f_mode
, filp
);
1167 return PTR_ERR(ctx
);
1168 nfs_file_set_open_context(filp
, ctx
);
1169 put_nfs_open_context(ctx
);
1170 nfs_fscache_open_file(inode
, filp
);
1173 EXPORT_SYMBOL_GPL(nfs_open
);
1176 * This function is called whenever some part of NFS notices that
1177 * the cached attributes have to be refreshed.
1180 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
1182 int status
= -ESTALE
;
1183 struct nfs4_label
*label
= NULL
;
1184 struct nfs_fattr
*fattr
= NULL
;
1185 struct nfs_inode
*nfsi
= NFS_I(inode
);
1187 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Lu)\n",
1188 inode
->i_sb
->s_id
, (unsigned long long)NFS_FILEID(inode
));
1190 trace_nfs_revalidate_inode_enter(inode
);
1192 if (is_bad_inode(inode
))
1194 if (NFS_STALE(inode
))
1197 /* pNFS: Attributes aren't updated until we layoutcommit */
1198 if (S_ISREG(inode
->i_mode
)) {
1199 status
= pnfs_sync_inode(inode
, false);
1205 fattr
= nfs_alloc_fattr();
1209 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
1211 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1212 if (IS_ERR(label
)) {
1213 status
= PTR_ERR(label
);
1217 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), fattr
,
1220 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1222 (unsigned long long)NFS_FILEID(inode
), status
);
1225 /* A soft timeout occurred. Use cached information? */
1226 if (server
->flags
& NFS_MOUNT_SOFTREVAL
)
1230 if (!S_ISDIR(inode
->i_mode
))
1231 nfs_set_inode_stale(inode
);
1233 nfs_zap_caches(inode
);
1238 status
= nfs_refresh_inode(inode
, fattr
);
1240 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1242 (unsigned long long)NFS_FILEID(inode
), status
);
1246 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
1247 nfs_zap_acl_cache(inode
);
1249 nfs_setsecurity(inode
, fattr
, label
);
1251 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) revalidation complete\n",
1253 (unsigned long long)NFS_FILEID(inode
));
1256 nfs4_label_free(label
);
1258 nfs_free_fattr(fattr
);
1259 trace_nfs_revalidate_inode_exit(inode
, status
);
1263 int nfs_attribute_cache_expired(struct inode
*inode
)
1265 if (nfs_have_delegated_attributes(inode
))
1267 return nfs_attribute_timeout(inode
);
1271 * nfs_revalidate_inode - Revalidate the inode attributes
1272 * @inode: pointer to inode struct
1273 * @flags: cache flags to check
1275 * Updates inode attribute information by retrieving the data from the server.
1277 int nfs_revalidate_inode(struct inode
*inode
, unsigned long flags
)
1279 if (!nfs_check_cache_invalid(inode
, flags
))
1280 return NFS_STALE(inode
) ? -ESTALE
: 0;
1281 return __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1283 EXPORT_SYMBOL_GPL(nfs_revalidate_inode
);
1285 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1289 if (mapping
->nrpages
!= 0) {
1290 if (S_ISREG(inode
->i_mode
)) {
1291 ret
= nfs_sync_mapping(mapping
);
1295 ret
= invalidate_inode_pages2(mapping
);
1299 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
1300 nfs_fscache_wait_on_invalidate(inode
);
1302 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) data cache invalidated\n",
1304 (unsigned long long)NFS_FILEID(inode
));
1309 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1310 * @mapping: pointer to mapping
1312 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1314 int nfs_clear_invalid_mapping(struct address_space
*mapping
)
1316 struct inode
*inode
= mapping
->host
;
1317 struct nfs_inode
*nfsi
= NFS_I(inode
);
1318 unsigned long *bitlock
= &nfsi
->flags
;
1322 * We must clear NFS_INO_INVALID_DATA first to ensure that
1323 * invalidations that come in while we're shooting down the mappings
1324 * are respected. But, that leaves a race window where one revalidator
1325 * can clear the flag, and then another checks it before the mapping
1326 * gets invalidated. Fix that by serializing access to this part of
1329 * At the same time, we need to allow other tasks to see whether we
1330 * might be in the middle of invalidating the pages, so we only set
1331 * the bit lock here if it looks like we're going to be doing that.
1334 ret
= wait_on_bit_action(bitlock
, NFS_INO_INVALIDATING
,
1335 nfs_wait_bit_killable
, TASK_KILLABLE
);
1338 spin_lock(&inode
->i_lock
);
1339 if (test_bit(NFS_INO_INVALIDATING
, bitlock
)) {
1340 spin_unlock(&inode
->i_lock
);
1343 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1345 spin_unlock(&inode
->i_lock
);
1349 set_bit(NFS_INO_INVALIDATING
, bitlock
);
1351 nfsi
->cache_validity
&=
1352 ~(NFS_INO_INVALID_DATA
| NFS_INO_DATA_INVAL_DEFER
);
1353 spin_unlock(&inode
->i_lock
);
1354 trace_nfs_invalidate_mapping_enter(inode
);
1355 ret
= nfs_invalidate_mapping(inode
, mapping
);
1356 trace_nfs_invalidate_mapping_exit(inode
, ret
);
1358 clear_bit_unlock(NFS_INO_INVALIDATING
, bitlock
);
1359 smp_mb__after_atomic();
1360 wake_up_bit(bitlock
, NFS_INO_INVALIDATING
);
1365 bool nfs_mapping_need_revalidate_inode(struct inode
*inode
)
1367 return nfs_check_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
) ||
1371 int nfs_revalidate_mapping_rcu(struct inode
*inode
)
1373 struct nfs_inode
*nfsi
= NFS_I(inode
);
1374 unsigned long *bitlock
= &nfsi
->flags
;
1377 if (IS_SWAPFILE(inode
))
1379 if (nfs_mapping_need_revalidate_inode(inode
)) {
1383 spin_lock(&inode
->i_lock
);
1384 if (test_bit(NFS_INO_INVALIDATING
, bitlock
) ||
1385 (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
))
1387 spin_unlock(&inode
->i_lock
);
1393 * nfs_revalidate_mapping - Revalidate the pagecache
1394 * @inode: pointer to host inode
1395 * @mapping: pointer to mapping
1397 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1399 /* swapfiles are not supposed to be shared. */
1400 if (IS_SWAPFILE(inode
))
1403 if (nfs_mapping_need_revalidate_inode(inode
)) {
1404 int ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1409 return nfs_clear_invalid_mapping(mapping
);
1412 static bool nfs_file_has_writers(struct nfs_inode
*nfsi
)
1414 struct inode
*inode
= &nfsi
->vfs_inode
;
1416 if (!S_ISREG(inode
->i_mode
))
1418 if (list_empty(&nfsi
->open_files
))
1420 return inode_is_open_for_write(inode
);
1423 static bool nfs_file_has_buffered_writers(struct nfs_inode
*nfsi
)
1425 return nfs_file_has_writers(nfsi
) && nfs_file_io_is_buffered(nfsi
);
1428 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1430 struct timespec64 ts
;
1432 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
)
1433 && (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
1434 && inode_eq_iversion_raw(inode
, fattr
->pre_change_attr
)) {
1435 inode_set_iversion_raw(inode
, fattr
->change_attr
);
1436 if (S_ISDIR(inode
->i_mode
))
1437 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1438 else if (nfs_server_capable(inode
, NFS_CAP_XATTR
))
1439 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_XATTR
);
1441 /* If we have atomic WCC data, we may update some attributes */
1442 ts
= inode
->i_ctime
;
1443 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
)
1444 && (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
1445 && timespec64_equal(&ts
, &fattr
->pre_ctime
)) {
1446 inode
->i_ctime
= fattr
->ctime
;
1449 ts
= inode
->i_mtime
;
1450 if ((fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
)
1451 && (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
1452 && timespec64_equal(&ts
, &fattr
->pre_mtime
)) {
1453 inode
->i_mtime
= fattr
->mtime
;
1454 if (S_ISDIR(inode
->i_mode
))
1455 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1457 if ((fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
)
1458 && (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
1459 && i_size_read(inode
) == nfs_size_to_loff_t(fattr
->pre_size
)
1460 && !nfs_have_writebacks(inode
)) {
1461 i_size_write(inode
, nfs_size_to_loff_t(fattr
->size
));
1466 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1467 * @inode: pointer to inode
1468 * @fattr: updated attributes
1470 * Verifies the attribute cache. If we have just changed the attributes,
1471 * so that fattr carries weak cache consistency data, then it may
1472 * also update the ctime/mtime/change_attribute.
1474 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
1476 struct nfs_inode
*nfsi
= NFS_I(inode
);
1477 loff_t cur_size
, new_isize
;
1478 unsigned long invalid
= 0;
1479 struct timespec64 ts
;
1481 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
))
1484 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1485 /* Only a mounted-on-fileid? Just exit */
1486 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1488 /* Has the inode gone and changed behind our back? */
1489 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1490 /* Is this perhaps the mounted-on fileid? */
1491 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1492 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1496 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
))
1500 if (!nfs_file_has_buffered_writers(nfsi
)) {
1501 /* Verify a few of the more important attributes */
1502 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 && !inode_eq_iversion_raw(inode
, fattr
->change_attr
))
1503 invalid
|= NFS_INO_INVALID_CHANGE
;
1505 ts
= inode
->i_mtime
;
1506 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) && !timespec64_equal(&ts
, &fattr
->mtime
))
1507 invalid
|= NFS_INO_INVALID_MTIME
;
1509 ts
= inode
->i_ctime
;
1510 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) && !timespec64_equal(&ts
, &fattr
->ctime
))
1511 invalid
|= NFS_INO_INVALID_CTIME
;
1513 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1514 cur_size
= i_size_read(inode
);
1515 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1516 if (cur_size
!= new_isize
)
1517 invalid
|= NFS_INO_INVALID_SIZE
;
1521 /* Have any file permissions changed? */
1522 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) && (inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
))
1523 invalid
|= NFS_INO_INVALID_MODE
;
1524 if ((fattr
->valid
& NFS_ATTR_FATTR_OWNER
) && !uid_eq(inode
->i_uid
, fattr
->uid
))
1525 invalid
|= NFS_INO_INVALID_OTHER
;
1526 if ((fattr
->valid
& NFS_ATTR_FATTR_GROUP
) && !gid_eq(inode
->i_gid
, fattr
->gid
))
1527 invalid
|= NFS_INO_INVALID_OTHER
;
1529 /* Has the link count changed? */
1530 if ((fattr
->valid
& NFS_ATTR_FATTR_NLINK
) && inode
->i_nlink
!= fattr
->nlink
)
1531 invalid
|= NFS_INO_INVALID_NLINK
;
1533 ts
= inode
->i_atime
;
1534 if ((fattr
->valid
& NFS_ATTR_FATTR_ATIME
) && !timespec64_equal(&ts
, &fattr
->atime
))
1535 invalid
|= NFS_INO_INVALID_ATIME
;
1538 nfs_set_cache_invalid(inode
, invalid
);
1540 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1544 static atomic_long_t nfs_attr_generation_counter
;
1546 static unsigned long nfs_read_attr_generation_counter(void)
1548 return atomic_long_read(&nfs_attr_generation_counter
);
1551 unsigned long nfs_inc_attr_generation_counter(void)
1553 return atomic_long_inc_return(&nfs_attr_generation_counter
);
1555 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter
);
1557 void nfs_fattr_init(struct nfs_fattr
*fattr
)
1560 fattr
->time_start
= jiffies
;
1561 fattr
->gencount
= nfs_inc_attr_generation_counter();
1562 fattr
->owner_name
= NULL
;
1563 fattr
->group_name
= NULL
;
1565 EXPORT_SYMBOL_GPL(nfs_fattr_init
);
1568 * nfs_fattr_set_barrier
1569 * @fattr: attributes
1571 * Used to set a barrier after an attribute was updated. This
1572 * barrier ensures that older attributes from RPC calls that may
1573 * have raced with our update cannot clobber these new values.
1574 * Note that you are still responsible for ensuring that other
1575 * operations which change the attribute on the server do not
1578 void nfs_fattr_set_barrier(struct nfs_fattr
*fattr
)
1580 fattr
->gencount
= nfs_inc_attr_generation_counter();
1583 struct nfs_fattr
*nfs_alloc_fattr(void)
1585 struct nfs_fattr
*fattr
;
1587 fattr
= kmalloc(sizeof(*fattr
), GFP_NOFS
);
1589 nfs_fattr_init(fattr
);
1592 EXPORT_SYMBOL_GPL(nfs_alloc_fattr
);
1594 struct nfs_fh
*nfs_alloc_fhandle(void)
1598 fh
= kmalloc(sizeof(struct nfs_fh
), GFP_NOFS
);
1603 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle
);
1607 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1608 * in the same way that wireshark does
1612 * For debugging only.
1614 u32
_nfs_display_fhandle_hash(const struct nfs_fh
*fh
)
1616 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1617 * not on the result */
1618 return nfs_fhandle_hash(fh
);
1620 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash
);
1623 * _nfs_display_fhandle - display an NFS file handle on the console
1625 * @fh: file handle to display
1626 * @caption: display caption
1628 * For debugging only.
1630 void _nfs_display_fhandle(const struct nfs_fh
*fh
, const char *caption
)
1634 if (fh
== NULL
|| fh
->size
== 0) {
1635 printk(KERN_DEFAULT
"%s at %p is empty\n", caption
, fh
);
1639 printk(KERN_DEFAULT
"%s at %p is %u bytes, crc: 0x%08x:\n",
1640 caption
, fh
, fh
->size
, _nfs_display_fhandle_hash(fh
));
1641 for (i
= 0; i
< fh
->size
; i
+= 16) {
1642 __be32
*pos
= (__be32
*)&fh
->data
[i
];
1644 switch ((fh
->size
- i
- 1) >> 2) {
1646 printk(KERN_DEFAULT
" %08x\n",
1650 printk(KERN_DEFAULT
" %08x %08x\n",
1651 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1));
1654 printk(KERN_DEFAULT
" %08x %08x %08x\n",
1655 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1656 be32_to_cpup(pos
+ 2));
1659 printk(KERN_DEFAULT
" %08x %08x %08x %08x\n",
1660 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1661 be32_to_cpup(pos
+ 2), be32_to_cpup(pos
+ 3));
1665 EXPORT_SYMBOL_GPL(_nfs_display_fhandle
);
1669 * nfs_inode_attrs_cmp_generic - compare attributes
1670 * @fattr: attributes
1671 * @inode: pointer to inode
1673 * Attempt to divine whether or not an RPC call reply carrying stale
1674 * attributes got scheduled after another call carrying updated ones.
1675 * Note also the check for wraparound of 'attr_gencount'
1677 * The function returns '1' if it thinks the attributes in @fattr are
1678 * more recent than the ones cached in @inode. Otherwise it returns
1681 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr
*fattr
,
1682 const struct inode
*inode
)
1684 unsigned long attr_gencount
= NFS_I(inode
)->attr_gencount
;
1686 return (long)(fattr
->gencount
- attr_gencount
) > 0 ||
1687 (long)(attr_gencount
- nfs_read_attr_generation_counter()) > 0;
1691 * nfs_inode_attrs_cmp_monotonic - compare attributes
1692 * @fattr: attributes
1693 * @inode: pointer to inode
1695 * Attempt to divine whether or not an RPC call reply carrying stale
1696 * attributes got scheduled after another call carrying updated ones.
1698 * We assume that the server observes monotonic semantics for
1699 * the change attribute, so a larger value means that the attributes in
1700 * @fattr are more recent, in which case the function returns the
1702 * A return value of '0' indicates no measurable change
1703 * A return value of '-1' means that the attributes in @inode are
1706 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr
*fattr
,
1707 const struct inode
*inode
)
1709 s64 diff
= fattr
->change_attr
- inode_peek_iversion_raw(inode
);
1712 return diff
== 0 ? 0 : -1;
1716 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1717 * @fattr: attributes
1718 * @inode: pointer to inode
1720 * Attempt to divine whether or not an RPC call reply carrying stale
1721 * attributes got scheduled after another call carrying updated ones.
1723 * We assume that the server observes strictly monotonic semantics for
1724 * the change attribute, so a larger value means that the attributes in
1725 * @fattr are more recent, in which case the function returns the
1727 * A return value of '-1' means that the attributes in @inode are
1728 * more recent or unchanged.
1730 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr
*fattr
,
1731 const struct inode
*inode
)
1733 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
) > 0 ? 1 : -1;
1737 * nfs_inode_attrs_cmp - compare attributes
1738 * @fattr: attributes
1739 * @inode: pointer to inode
1741 * This function returns '1' if it thinks the attributes in @fattr are
1742 * more recent than the ones cached in @inode. It returns '-1' if
1743 * the attributes in @inode are more recent than the ones in @fattr,
1744 * and it returns 0 if not sure.
1746 static int nfs_inode_attrs_cmp(const struct nfs_fattr
*fattr
,
1747 const struct inode
*inode
)
1749 if (nfs_inode_attrs_cmp_generic(fattr
, inode
) > 0)
1751 switch (NFS_SERVER(inode
)->change_attr_type
) {
1752 case NFS4_CHANGE_TYPE_IS_UNDEFINED
:
1754 case NFS4_CHANGE_TYPE_IS_TIME_METADATA
:
1755 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1757 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
);
1759 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1761 return nfs_inode_attrs_cmp_strict_monotonic(fattr
, inode
);
1767 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1768 * @fattr: attributes
1769 * @inode: pointer to inode
1771 * Returns '1' if the last attribute update left the inode cached
1772 * attributes in a partially unrevalidated state, and @fattr
1773 * matches the change attribute of that partial update.
1774 * Otherwise returns '0'.
1776 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr
*fattr
,
1777 const struct inode
*inode
)
1779 const unsigned long check_valid
=
1780 NFS_INO_INVALID_ATIME
| NFS_INO_INVALID_CTIME
|
1781 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
1782 NFS_INO_INVALID_BLOCKS
| NFS_INO_INVALID_OTHER
|
1783 NFS_INO_INVALID_NLINK
;
1784 unsigned long cache_validity
= NFS_I(inode
)->cache_validity
;
1785 enum nfs4_change_attr_type ctype
= NFS_SERVER(inode
)->change_attr_type
;
1787 if (ctype
!= NFS4_CHANGE_TYPE_IS_UNDEFINED
&&
1788 !(cache_validity
& NFS_INO_INVALID_CHANGE
) &&
1789 (cache_validity
& check_valid
) != 0 &&
1790 (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1791 nfs_inode_attrs_cmp_monotonic(fattr
, inode
) == 0)
1796 static int nfs_refresh_inode_locked(struct inode
*inode
,
1797 struct nfs_fattr
*fattr
)
1799 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1802 trace_nfs_refresh_inode_enter(inode
);
1804 if (attr_cmp
> 0 || nfs_inode_finish_partial_attr_update(fattr
, inode
))
1805 ret
= nfs_update_inode(inode
, fattr
);
1806 else if (attr_cmp
== 0)
1807 ret
= nfs_check_inode_attributes(inode
, fattr
);
1809 trace_nfs_refresh_inode_exit(inode
, ret
);
1814 * nfs_refresh_inode - try to update the inode attribute cache
1815 * @inode: pointer to inode
1816 * @fattr: updated attributes
1818 * Check that an RPC call that returned attributes has not overlapped with
1819 * other recent updates of the inode metadata, then decide whether it is
1820 * safe to do a full update of the inode attributes, or whether just to
1821 * call nfs_check_inode_attributes.
1823 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1827 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1829 spin_lock(&inode
->i_lock
);
1830 status
= nfs_refresh_inode_locked(inode
, fattr
);
1831 spin_unlock(&inode
->i_lock
);
1835 EXPORT_SYMBOL_GPL(nfs_refresh_inode
);
1837 static int nfs_post_op_update_inode_locked(struct inode
*inode
,
1838 struct nfs_fattr
*fattr
, unsigned int invalid
)
1840 if (S_ISDIR(inode
->i_mode
))
1841 invalid
|= NFS_INO_INVALID_DATA
;
1842 nfs_set_cache_invalid(inode
, invalid
);
1843 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1845 return nfs_refresh_inode_locked(inode
, fattr
);
1849 * nfs_post_op_update_inode - try to update the inode attribute cache
1850 * @inode: pointer to inode
1851 * @fattr: updated attributes
1853 * After an operation that has changed the inode metadata, mark the
1854 * attribute cache as being invalid, then try to update it.
1856 * NB: if the server didn't return any post op attributes, this
1857 * function will force the retrieval of attributes before the next
1858 * NFS request. Thus it should be used only for operations that
1859 * are expected to change one or more attributes, to avoid
1860 * unnecessary NFS requests and trips through nfs_update_inode().
1862 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1866 spin_lock(&inode
->i_lock
);
1867 nfs_fattr_set_barrier(fattr
);
1868 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1869 NFS_INO_INVALID_CHANGE
1870 | NFS_INO_INVALID_CTIME
1871 | NFS_INO_REVAL_FORCED
);
1872 spin_unlock(&inode
->i_lock
);
1876 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode
);
1879 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1880 * @inode: pointer to inode
1881 * @fattr: updated attributes
1883 * After an operation that has changed the inode metadata, mark the
1884 * attribute cache as being invalid, then try to update it. Fake up
1885 * weak cache consistency data, if none exist.
1887 * This function is mainly designed to be used by the ->write_done() functions.
1889 int nfs_post_op_update_inode_force_wcc_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1891 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1894 /* Don't do a WCC update if these attributes are already stale */
1897 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0 || !attr_cmp
) {
1898 fattr
->valid
&= ~(NFS_ATTR_FATTR_PRECHANGE
1899 | NFS_ATTR_FATTR_PRESIZE
1900 | NFS_ATTR_FATTR_PREMTIME
1901 | NFS_ATTR_FATTR_PRECTIME
);
1904 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1905 (fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
) == 0) {
1906 fattr
->pre_change_attr
= inode_peek_iversion_raw(inode
);
1907 fattr
->valid
|= NFS_ATTR_FATTR_PRECHANGE
;
1909 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) != 0 &&
1910 (fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
) == 0) {
1911 fattr
->pre_ctime
= inode
->i_ctime
;
1912 fattr
->valid
|= NFS_ATTR_FATTR_PRECTIME
;
1914 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) != 0 &&
1915 (fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
) == 0) {
1916 fattr
->pre_mtime
= inode
->i_mtime
;
1917 fattr
->valid
|= NFS_ATTR_FATTR_PREMTIME
;
1919 if ((fattr
->valid
& NFS_ATTR_FATTR_SIZE
) != 0 &&
1920 (fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
) == 0) {
1921 fattr
->pre_size
= i_size_read(inode
);
1922 fattr
->valid
|= NFS_ATTR_FATTR_PRESIZE
;
1925 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1926 NFS_INO_INVALID_CHANGE
1927 | NFS_INO_INVALID_CTIME
1928 | NFS_INO_INVALID_MTIME
1929 | NFS_INO_INVALID_BLOCKS
);
1934 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1935 * @inode: pointer to inode
1936 * @fattr: updated attributes
1938 * After an operation that has changed the inode metadata, mark the
1939 * attribute cache as being invalid, then try to update it. Fake up
1940 * weak cache consistency data, if none exist.
1942 * This function is mainly designed to be used by the ->write_done() functions.
1944 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
1948 spin_lock(&inode
->i_lock
);
1949 nfs_fattr_set_barrier(fattr
);
1950 status
= nfs_post_op_update_inode_force_wcc_locked(inode
, fattr
);
1951 spin_unlock(&inode
->i_lock
);
1954 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc
);
1958 * Many nfs protocol calls return the new file attributes after
1959 * an operation. Here we update the inode to reflect the state
1960 * of the server's inode.
1962 * This is a bit tricky because we have to make sure all dirty pages
1963 * have been sent off to the server before calling invalidate_inode_pages.
1964 * To make sure no other process adds more write requests while we try
1965 * our best to flush them, we make them sleep during the attribute refresh.
1967 * A very similar scenario holds for the dir cache.
1969 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1971 struct nfs_server
*server
= NFS_SERVER(inode
);
1972 struct nfs_inode
*nfsi
= NFS_I(inode
);
1973 loff_t cur_isize
, new_isize
;
1974 u64 fattr_supported
= server
->fattr_valid
;
1975 unsigned long invalid
= 0;
1976 unsigned long now
= jiffies
;
1977 unsigned long save_cache_validity
;
1978 bool have_writers
= nfs_file_has_buffered_writers(nfsi
);
1979 bool cache_revalidated
= true;
1980 bool attr_changed
= false;
1981 bool have_delegation
;
1983 dfprintk(VFS
, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1984 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1985 nfs_display_fhandle_hash(NFS_FH(inode
)),
1986 atomic_read(&inode
->i_count
), fattr
->valid
);
1988 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1989 /* Only a mounted-on-fileid? Just exit */
1990 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1992 /* Has the inode gone and changed behind our back? */
1993 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1994 /* Is this perhaps the mounted-on fileid? */
1995 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1996 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1998 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1999 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2000 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
2001 inode
->i_sb
->s_id
, (long long)nfsi
->fileid
,
2002 (long long)fattr
->fileid
);
2007 * Make sure the inode's type hasn't changed.
2009 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
)) {
2011 * Big trouble! The inode has become a different object.
2013 printk(KERN_DEBUG
"NFS: %s: inode %lu mode changed, %07o to %07o\n",
2014 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
2018 /* Update the fsid? */
2019 if (S_ISDIR(inode
->i_mode
) && (fattr
->valid
& NFS_ATTR_FATTR_FSID
) &&
2020 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
2021 !IS_AUTOMOUNT(inode
))
2022 server
->fsid
= fattr
->fsid
;
2024 /* Save the delegation state before clearing cache_validity */
2025 have_delegation
= nfs_have_delegated_attributes(inode
);
2028 * Update the read time so we don't revalidate too often.
2030 nfsi
->read_cache_jiffies
= fattr
->time_start
;
2032 save_cache_validity
= nfsi
->cache_validity
;
2033 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
2034 | NFS_INO_INVALID_ATIME
2035 | NFS_INO_REVAL_FORCED
2036 | NFS_INO_INVALID_BLOCKS
);
2038 /* Do atomic weak cache consistency updates */
2039 nfs_wcc_update_inode(inode
, fattr
);
2041 if (pnfs_layoutcommit_outstanding(inode
)) {
2042 nfsi
->cache_validity
|=
2043 save_cache_validity
&
2044 (NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_CTIME
|
2045 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
2046 NFS_INO_INVALID_BLOCKS
);
2047 cache_revalidated
= false;
2050 /* More cache consistency checks */
2051 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) {
2052 if (!inode_eq_iversion_raw(inode
, fattr
->change_attr
)) {
2053 /* Could it be a race with writeback? */
2054 if (!(have_writers
|| have_delegation
)) {
2055 invalid
|= NFS_INO_INVALID_DATA
2056 | NFS_INO_INVALID_ACCESS
2057 | NFS_INO_INVALID_ACL
2058 | NFS_INO_INVALID_XATTR
;
2059 /* Force revalidate of all attributes */
2060 save_cache_validity
|= NFS_INO_INVALID_CTIME
2061 | NFS_INO_INVALID_MTIME
2062 | NFS_INO_INVALID_SIZE
2063 | NFS_INO_INVALID_BLOCKS
2064 | NFS_INO_INVALID_NLINK
2065 | NFS_INO_INVALID_MODE
2066 | NFS_INO_INVALID_OTHER
;
2067 if (S_ISDIR(inode
->i_mode
))
2068 nfs_force_lookup_revalidate(inode
);
2069 attr_changed
= true;
2070 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2073 } else if (!have_delegation
)
2074 nfsi
->cache_validity
|= NFS_INO_DATA_INVAL_DEFER
;
2075 inode_set_iversion_raw(inode
, fattr
->change_attr
);
2078 nfsi
->cache_validity
|=
2079 save_cache_validity
& NFS_INO_INVALID_CHANGE
;
2080 if (!have_delegation
||
2081 (nfsi
->cache_validity
& NFS_INO_INVALID_CHANGE
) != 0)
2082 cache_revalidated
= false;
2085 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
2086 inode
->i_mtime
= fattr
->mtime
;
2087 else if (fattr_supported
& NFS_ATTR_FATTR_MTIME
)
2088 nfsi
->cache_validity
|=
2089 save_cache_validity
& NFS_INO_INVALID_MTIME
;
2091 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
2092 inode
->i_ctime
= fattr
->ctime
;
2093 else if (fattr_supported
& NFS_ATTR_FATTR_CTIME
)
2094 nfsi
->cache_validity
|=
2095 save_cache_validity
& NFS_INO_INVALID_CTIME
;
2097 /* Check if our cached file size is stale */
2098 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
2099 new_isize
= nfs_size_to_loff_t(fattr
->size
);
2100 cur_isize
= i_size_read(inode
);
2101 if (new_isize
!= cur_isize
&& !have_delegation
) {
2102 /* Do we perhaps have any outstanding writes, or has
2103 * the file grown beyond our last write? */
2104 if (!nfs_have_writebacks(inode
) || new_isize
> cur_isize
) {
2105 i_size_write(inode
, new_isize
);
2107 invalid
|= NFS_INO_INVALID_DATA
;
2109 dprintk("NFS: isize change on server for file %s/%ld "
2113 (long long)cur_isize
,
2114 (long long)new_isize
);
2116 if (new_isize
== 0 &&
2117 !(fattr
->valid
& (NFS_ATTR_FATTR_SPACE_USED
|
2118 NFS_ATTR_FATTR_BLOCKS_USED
))) {
2119 fattr
->du
.nfs3
.used
= 0;
2120 fattr
->valid
|= NFS_ATTR_FATTR_SPACE_USED
;
2123 nfsi
->cache_validity
|=
2124 save_cache_validity
& NFS_INO_INVALID_SIZE
;
2126 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
2127 inode
->i_atime
= fattr
->atime
;
2128 else if (fattr_supported
& NFS_ATTR_FATTR_ATIME
)
2129 nfsi
->cache_validity
|=
2130 save_cache_validity
& NFS_INO_INVALID_ATIME
;
2132 if (fattr
->valid
& NFS_ATTR_FATTR_MODE
) {
2133 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)) {
2134 umode_t newmode
= inode
->i_mode
& S_IFMT
;
2135 newmode
|= fattr
->mode
& S_IALLUGO
;
2136 inode
->i_mode
= newmode
;
2137 invalid
|= NFS_INO_INVALID_ACCESS
2138 | NFS_INO_INVALID_ACL
;
2140 } else if (fattr_supported
& NFS_ATTR_FATTR_MODE
)
2141 nfsi
->cache_validity
|=
2142 save_cache_validity
& NFS_INO_INVALID_MODE
;
2144 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
) {
2145 if (!uid_eq(inode
->i_uid
, fattr
->uid
)) {
2146 invalid
|= NFS_INO_INVALID_ACCESS
2147 | NFS_INO_INVALID_ACL
;
2148 inode
->i_uid
= fattr
->uid
;
2150 } else if (fattr_supported
& NFS_ATTR_FATTR_OWNER
)
2151 nfsi
->cache_validity
|=
2152 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2154 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
) {
2155 if (!gid_eq(inode
->i_gid
, fattr
->gid
)) {
2156 invalid
|= NFS_INO_INVALID_ACCESS
2157 | NFS_INO_INVALID_ACL
;
2158 inode
->i_gid
= fattr
->gid
;
2160 } else if (fattr_supported
& NFS_ATTR_FATTR_GROUP
)
2161 nfsi
->cache_validity
|=
2162 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2164 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
) {
2165 if (inode
->i_nlink
!= fattr
->nlink
) {
2166 if (S_ISDIR(inode
->i_mode
))
2167 invalid
|= NFS_INO_INVALID_DATA
;
2168 set_nlink(inode
, fattr
->nlink
);
2170 } else if (fattr_supported
& NFS_ATTR_FATTR_NLINK
)
2171 nfsi
->cache_validity
|=
2172 save_cache_validity
& NFS_INO_INVALID_NLINK
;
2174 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
2176 * report the blocks in 512byte units
2178 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
2179 } else if (fattr_supported
& NFS_ATTR_FATTR_SPACE_USED
)
2180 nfsi
->cache_validity
|=
2181 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2183 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
2184 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
2185 else if (fattr_supported
& NFS_ATTR_FATTR_BLOCKS_USED
)
2186 nfsi
->cache_validity
|=
2187 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2189 /* Update attrtimeo value if we're out of the unstable period */
2191 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
2192 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
2193 nfsi
->attrtimeo_timestamp
= now
;
2194 /* Set barrier to be more recent than all outstanding updates */
2195 nfsi
->attr_gencount
= nfs_inc_attr_generation_counter();
2197 if (cache_revalidated
) {
2198 if (!time_in_range_open(now
, nfsi
->attrtimeo_timestamp
,
2199 nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
2200 nfsi
->attrtimeo
<<= 1;
2201 if (nfsi
->attrtimeo
> NFS_MAXATTRTIMEO(inode
))
2202 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
2204 nfsi
->attrtimeo_timestamp
= now
;
2206 /* Set the barrier to be more recent than this fattr */
2207 if ((long)(fattr
->gencount
- nfsi
->attr_gencount
) > 0)
2208 nfsi
->attr_gencount
= fattr
->gencount
;
2211 /* Don't invalidate the data if we were to blame */
2212 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
2213 || S_ISLNK(inode
->i_mode
)))
2214 invalid
&= ~NFS_INO_INVALID_DATA
;
2215 nfs_set_cache_invalid(inode
, invalid
);
2220 * No need to worry about unhashing the dentry, as the
2221 * lookup validation will know that the inode is bad.
2222 * (But we fall through to invalidate the caches.)
2224 nfs_set_inode_stale_locked(inode
);
2228 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
2230 struct nfs_inode
*nfsi
;
2231 nfsi
= kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
2235 nfsi
->cache_validity
= 0UL;
2236 #if IS_ENABLED(CONFIG_NFS_V4)
2237 nfsi
->nfs4_acl
= NULL
;
2238 #endif /* CONFIG_NFS_V4 */
2239 #ifdef CONFIG_NFS_V4_2
2240 nfsi
->xattr_cache
= NULL
;
2242 return &nfsi
->vfs_inode
;
2244 EXPORT_SYMBOL_GPL(nfs_alloc_inode
);
2246 void nfs_free_inode(struct inode
*inode
)
2248 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
2250 EXPORT_SYMBOL_GPL(nfs_free_inode
);
2252 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
2254 #if IS_ENABLED(CONFIG_NFS_V4)
2255 INIT_LIST_HEAD(&nfsi
->open_states
);
2256 nfsi
->delegation
= NULL
;
2257 init_rwsem(&nfsi
->rwsem
);
2258 nfsi
->layout
= NULL
;
2262 static void init_once(void *foo
)
2264 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
2266 inode_init_once(&nfsi
->vfs_inode
);
2267 INIT_LIST_HEAD(&nfsi
->open_files
);
2268 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
2269 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
2270 INIT_LIST_HEAD(&nfsi
->commit_info
.list
);
2271 atomic_long_set(&nfsi
->nrequests
, 0);
2272 atomic_long_set(&nfsi
->commit_info
.ncommit
, 0);
2273 atomic_set(&nfsi
->commit_info
.rpcs_out
, 0);
2274 init_rwsem(&nfsi
->rmdir_sem
);
2275 mutex_init(&nfsi
->commit_mutex
);
2276 nfs4_init_once(nfsi
);
2277 nfsi
->cache_change_attribute
= 0;
2280 static int __init
nfs_init_inodecache(void)
2282 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
2283 sizeof(struct nfs_inode
),
2284 0, (SLAB_RECLAIM_ACCOUNT
|
2285 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
2287 if (nfs_inode_cachep
== NULL
)
2293 static void nfs_destroy_inodecache(void)
2296 * Make sure all delayed rcu free inodes are flushed before we
2300 kmem_cache_destroy(nfs_inode_cachep
);
2303 struct workqueue_struct
*nfsiod_workqueue
;
2304 EXPORT_SYMBOL_GPL(nfsiod_workqueue
);
2307 * start up the nfsiod workqueue
2309 static int nfsiod_start(void)
2311 struct workqueue_struct
*wq
;
2312 dprintk("RPC: creating workqueue nfsiod\n");
2313 wq
= alloc_workqueue("nfsiod", WQ_MEM_RECLAIM
| WQ_UNBOUND
, 0);
2316 nfsiod_workqueue
= wq
;
2321 * Destroy the nfsiod workqueue
2323 static void nfsiod_stop(void)
2325 struct workqueue_struct
*wq
;
2327 wq
= nfsiod_workqueue
;
2330 nfsiod_workqueue
= NULL
;
2331 destroy_workqueue(wq
);
2334 unsigned int nfs_net_id
;
2335 EXPORT_SYMBOL_GPL(nfs_net_id
);
2337 static int nfs_net_init(struct net
*net
)
2339 nfs_clients_init(net
);
2340 return nfs_fs_proc_net_init(net
);
2343 static void nfs_net_exit(struct net
*net
)
2345 nfs_fs_proc_net_exit(net
);
2346 nfs_clients_exit(net
);
2349 static struct pernet_operations nfs_net_ops
= {
2350 .init
= nfs_net_init
,
2351 .exit
= nfs_net_exit
,
2353 .size
= sizeof(struct nfs_net
),
2359 static int __init
init_nfs_fs(void)
2363 err
= nfs_sysfs_init();
2367 err
= register_pernet_subsys(&nfs_net_ops
);
2371 err
= nfs_fscache_register();
2375 err
= nfsiod_start();
2379 err
= nfs_fs_proc_init();
2383 err
= nfs_init_nfspagecache();
2387 err
= nfs_init_inodecache();
2391 err
= nfs_init_readpagecache();
2395 err
= nfs_init_writepagecache();
2399 err
= nfs_init_directcache();
2403 rpc_proc_register(&init_net
, &nfs_rpcstat
);
2405 err
= register_nfs_fs();
2411 rpc_proc_unregister(&init_net
, "nfs");
2412 nfs_destroy_directcache();
2414 nfs_destroy_writepagecache();
2416 nfs_destroy_readpagecache();
2418 nfs_destroy_inodecache();
2420 nfs_destroy_nfspagecache();
2426 nfs_fscache_unregister();
2428 unregister_pernet_subsys(&nfs_net_ops
);
2435 static void __exit
exit_nfs_fs(void)
2437 nfs_destroy_directcache();
2438 nfs_destroy_writepagecache();
2439 nfs_destroy_readpagecache();
2440 nfs_destroy_inodecache();
2441 nfs_destroy_nfspagecache();
2442 nfs_fscache_unregister();
2443 unregister_pernet_subsys(&nfs_net_ops
);
2444 rpc_proc_unregister(&init_net
, "nfs");
2445 unregister_nfs_fs();
2451 /* Not quite true; I just maintain it */
2452 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2453 MODULE_LICENSE("GPL");
2454 module_param(enable_ino64
, bool, 0644);
2456 module_init(init_nfs_fs
)
2457 module_exit(exit_nfs_fs
)