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 if (inode
->i_mapping
->nrpages
== 0)
214 flags
&= ~(NFS_INO_INVALID_DATA
|NFS_INO_DATA_INVAL_DEFER
);
215 flags
&= ~(NFS_INO_REVAL_PAGECACHE
| NFS_INO_REVAL_FORCED
);
216 nfsi
->cache_validity
|= flags
;
218 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid
);
221 * Invalidate the local caches
223 static void nfs_zap_caches_locked(struct inode
*inode
)
225 struct nfs_inode
*nfsi
= NFS_I(inode
);
226 int mode
= inode
->i_mode
;
228 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
230 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
231 nfsi
->attrtimeo_timestamp
= jiffies
;
233 if (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)) {
234 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATTR
235 | NFS_INO_INVALID_DATA
236 | NFS_INO_INVALID_ACCESS
237 | NFS_INO_INVALID_ACL
238 | NFS_INO_INVALID_XATTR
239 | NFS_INO_REVAL_PAGECACHE
);
241 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATTR
242 | NFS_INO_INVALID_ACCESS
243 | NFS_INO_INVALID_ACL
244 | NFS_INO_INVALID_XATTR
245 | NFS_INO_REVAL_PAGECACHE
);
246 nfs_zap_label_cache_locked(nfsi
);
249 void nfs_zap_caches(struct inode
*inode
)
251 spin_lock(&inode
->i_lock
);
252 nfs_zap_caches_locked(inode
);
253 spin_unlock(&inode
->i_lock
);
256 void nfs_zap_mapping(struct inode
*inode
, struct address_space
*mapping
)
258 if (mapping
->nrpages
!= 0) {
259 spin_lock(&inode
->i_lock
);
260 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
261 spin_unlock(&inode
->i_lock
);
265 void nfs_zap_acl_cache(struct inode
*inode
)
267 void (*clear_acl_cache
)(struct inode
*);
269 clear_acl_cache
= NFS_PROTO(inode
)->clear_acl_cache
;
270 if (clear_acl_cache
!= NULL
)
271 clear_acl_cache(inode
);
272 spin_lock(&inode
->i_lock
);
273 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_ACL
;
274 spin_unlock(&inode
->i_lock
);
276 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache
);
278 void nfs_invalidate_atime(struct inode
*inode
)
280 spin_lock(&inode
->i_lock
);
281 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
282 spin_unlock(&inode
->i_lock
);
284 EXPORT_SYMBOL_GPL(nfs_invalidate_atime
);
287 * Invalidate, but do not unhash, the inode.
288 * NB: must be called with inode->i_lock held!
290 static void nfs_set_inode_stale_locked(struct inode
*inode
)
292 set_bit(NFS_INO_STALE
, &NFS_I(inode
)->flags
);
293 nfs_zap_caches_locked(inode
);
294 trace_nfs_set_inode_stale(inode
);
297 void nfs_set_inode_stale(struct inode
*inode
)
299 spin_lock(&inode
->i_lock
);
300 nfs_set_inode_stale_locked(inode
);
301 spin_unlock(&inode
->i_lock
);
304 struct nfs_find_desc
{
306 struct nfs_fattr
*fattr
;
310 * In NFSv3 we can have 64bit inode numbers. In order to support
311 * this, and re-exported directories (also seen in NFSv2)
312 * we are forced to allow 2 different inodes to have the same
316 nfs_find_actor(struct inode
*inode
, void *opaque
)
318 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
319 struct nfs_fh
*fh
= desc
->fh
;
320 struct nfs_fattr
*fattr
= desc
->fattr
;
322 if (NFS_FILEID(inode
) != fattr
->fileid
)
324 if (inode_wrong_type(inode
, fattr
->mode
))
326 if (nfs_compare_fh(NFS_FH(inode
), fh
))
328 if (is_bad_inode(inode
) || NFS_STALE(inode
))
334 nfs_init_locked(struct inode
*inode
, void *opaque
)
336 struct nfs_find_desc
*desc
= (struct nfs_find_desc
*)opaque
;
337 struct nfs_fattr
*fattr
= desc
->fattr
;
339 set_nfs_fileid(inode
, fattr
->fileid
);
340 inode
->i_mode
= fattr
->mode
;
341 nfs_copy_fh(NFS_FH(inode
), desc
->fh
);
345 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
346 static void nfs_clear_label_invalid(struct inode
*inode
)
348 spin_lock(&inode
->i_lock
);
349 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_LABEL
;
350 spin_unlock(&inode
->i_lock
);
353 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
354 struct nfs4_label
*label
)
361 if ((fattr
->valid
& NFS_ATTR_FATTR_V4_SECURITY_LABEL
) && inode
->i_security
) {
362 error
= security_inode_notifysecctx(inode
, label
->label
,
365 printk(KERN_ERR
"%s() %s %d "
366 "security_inode_notifysecctx() %d\n",
368 (char *)label
->label
,
370 nfs_clear_label_invalid(inode
);
374 struct nfs4_label
*nfs4_label_alloc(struct nfs_server
*server
, gfp_t flags
)
376 struct nfs4_label
*label
= NULL
;
377 int minor_version
= server
->nfs_client
->cl_minorversion
;
379 if (minor_version
< 2)
382 if (!(server
->caps
& NFS_CAP_SECURITY_LABEL
))
385 label
= kzalloc(sizeof(struct nfs4_label
), flags
);
387 return ERR_PTR(-ENOMEM
);
389 label
->label
= kzalloc(NFS4_MAXLABELLEN
, flags
);
390 if (label
->label
== NULL
) {
392 return ERR_PTR(-ENOMEM
);
394 label
->len
= NFS4_MAXLABELLEN
;
398 EXPORT_SYMBOL_GPL(nfs4_label_alloc
);
400 void nfs_setsecurity(struct inode
*inode
, struct nfs_fattr
*fattr
,
401 struct nfs4_label
*label
)
405 EXPORT_SYMBOL_GPL(nfs_setsecurity
);
407 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
409 nfs_ilookup(struct super_block
*sb
, struct nfs_fattr
*fattr
, struct nfs_fh
*fh
)
411 struct nfs_find_desc desc
= {
418 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
) ||
419 !(fattr
->valid
& NFS_ATTR_FATTR_TYPE
))
422 hash
= nfs_fattr_to_ino_t(fattr
);
423 inode
= ilookup5(sb
, hash
, nfs_find_actor
, &desc
);
425 dprintk("%s: returning %p\n", __func__
, inode
);
430 * This is our front-end to iget that looks up inodes by file handle
431 * instead of inode number.
434 nfs_fhget(struct super_block
*sb
, struct nfs_fh
*fh
, struct nfs_fattr
*fattr
, struct nfs4_label
*label
)
436 struct nfs_find_desc desc
= {
440 struct inode
*inode
= ERR_PTR(-ENOENT
);
441 u64 fattr_supported
= NFS_SB(sb
)->fattr_valid
;
444 nfs_attr_check_mountpoint(sb
, fattr
);
446 if (nfs_attr_use_mounted_on_fileid(fattr
))
447 fattr
->fileid
= fattr
->mounted_on_fileid
;
448 else if ((fattr
->valid
& NFS_ATTR_FATTR_FILEID
) == 0)
450 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) == 0)
453 hash
= nfs_fattr_to_ino_t(fattr
);
455 inode
= iget5_locked(sb
, hash
, nfs_find_actor
, nfs_init_locked
, &desc
);
457 inode
= ERR_PTR(-ENOMEM
);
461 if (inode
->i_state
& I_NEW
) {
462 struct nfs_inode
*nfsi
= NFS_I(inode
);
463 unsigned long now
= jiffies
;
465 /* We set i_ino for the few things that still rely on it,
469 /* We can't support update_atime(), since the server will reset it */
470 inode
->i_flags
|= S_NOATIME
|S_NOCMTIME
;
471 inode
->i_mode
= fattr
->mode
;
472 nfsi
->cache_validity
= 0;
473 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) == 0
474 && (fattr_supported
& NFS_ATTR_FATTR_MODE
))
475 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MODE
);
476 /* Why so? Because we want revalidate for devices/FIFOs, and
477 * that's precisely what we have in nfs_file_inode_operations.
479 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_inode_ops
;
480 if (S_ISREG(inode
->i_mode
)) {
481 inode
->i_fop
= NFS_SB(sb
)->nfs_client
->rpc_ops
->file_ops
;
482 inode
->i_data
.a_ops
= &nfs_file_aops
;
483 } else if (S_ISDIR(inode
->i_mode
)) {
484 inode
->i_op
= NFS_SB(sb
)->nfs_client
->rpc_ops
->dir_inode_ops
;
485 inode
->i_fop
= &nfs_dir_operations
;
486 inode
->i_data
.a_ops
= &nfs_dir_aops
;
487 /* Deal with crossing mountpoints */
488 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTPOINT
||
489 fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
) {
490 if (fattr
->valid
& NFS_ATTR_FATTR_V4_REFERRAL
)
491 inode
->i_op
= &nfs_referral_inode_operations
;
493 inode
->i_op
= &nfs_mountpoint_inode_operations
;
495 inode
->i_flags
|= S_AUTOMOUNT
;
497 } else if (S_ISLNK(inode
->i_mode
)) {
498 inode
->i_op
= &nfs_symlink_inode_operations
;
499 inode_nohighmem(inode
);
501 init_special_inode(inode
, inode
->i_mode
, fattr
->rdev
);
503 memset(&inode
->i_atime
, 0, sizeof(inode
->i_atime
));
504 memset(&inode
->i_mtime
, 0, sizeof(inode
->i_mtime
));
505 memset(&inode
->i_ctime
, 0, sizeof(inode
->i_ctime
));
506 inode_set_iversion_raw(inode
, 0);
509 inode
->i_uid
= make_kuid(&init_user_ns
, -2);
510 inode
->i_gid
= make_kgid(&init_user_ns
, -2);
512 memset(nfsi
->cookieverf
, 0, sizeof(nfsi
->cookieverf
));
516 nfsi
->read_cache_jiffies
= fattr
->time_start
;
517 nfsi
->attr_gencount
= fattr
->gencount
;
518 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
519 inode
->i_atime
= fattr
->atime
;
520 else if (fattr_supported
& NFS_ATTR_FATTR_ATIME
)
521 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
522 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
523 inode
->i_mtime
= fattr
->mtime
;
524 else if (fattr_supported
& NFS_ATTR_FATTR_MTIME
)
525 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
);
526 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
527 inode
->i_ctime
= fattr
->ctime
;
528 else if (fattr_supported
& NFS_ATTR_FATTR_CTIME
)
529 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CTIME
);
530 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
531 inode_set_iversion_raw(inode
, fattr
->change_attr
);
533 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
);
534 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
535 inode
->i_size
= nfs_size_to_loff_t(fattr
->size
);
537 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_SIZE
);
538 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
)
539 set_nlink(inode
, fattr
->nlink
);
540 else if (fattr_supported
& NFS_ATTR_FATTR_NLINK
)
541 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_NLINK
);
542 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
)
543 inode
->i_uid
= fattr
->uid
;
544 else if (fattr_supported
& NFS_ATTR_FATTR_OWNER
)
545 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_OTHER
);
546 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
)
547 inode
->i_gid
= fattr
->gid
;
548 else if (fattr_supported
& NFS_ATTR_FATTR_GROUP
)
549 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_OTHER
);
550 if (nfs_server_capable(inode
, NFS_CAP_XATTR
))
551 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_XATTR
);
552 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
553 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
554 else if (fattr_supported
& NFS_ATTR_FATTR_BLOCKS_USED
&&
556 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_BLOCKS
);
557 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
559 * report the blocks in 512byte units
561 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
562 } else if (fattr_supported
& NFS_ATTR_FATTR_SPACE_USED
&&
564 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_BLOCKS
);
566 nfs_setsecurity(inode
, fattr
, label
);
568 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
569 nfsi
->attrtimeo_timestamp
= now
;
570 nfsi
->access_cache
= RB_ROOT
;
572 nfs_fscache_init_inode(inode
);
574 unlock_new_inode(inode
);
576 int err
= nfs_refresh_inode(inode
, fattr
);
579 inode
= ERR_PTR(err
);
583 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
585 (unsigned long long)NFS_FILEID(inode
),
586 nfs_display_fhandle_hash(fh
),
587 atomic_read(&inode
->i_count
));
593 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode
));
596 EXPORT_SYMBOL_GPL(nfs_fhget
);
598 #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)
601 nfs_setattr(struct user_namespace
*mnt_userns
, struct dentry
*dentry
,
604 struct inode
*inode
= d_inode(dentry
);
605 struct nfs_fattr
*fattr
;
608 nfs_inc_stats(inode
, NFSIOS_VFSSETATTR
);
610 /* skip mode change if it's just for clearing setuid/setgid */
611 if (attr
->ia_valid
& (ATTR_KILL_SUID
| ATTR_KILL_SGID
))
612 attr
->ia_valid
&= ~ATTR_MODE
;
614 if (attr
->ia_valid
& ATTR_SIZE
) {
615 BUG_ON(!S_ISREG(inode
->i_mode
));
617 error
= inode_newsize_ok(inode
, attr
->ia_size
);
621 if (attr
->ia_size
== i_size_read(inode
))
622 attr
->ia_valid
&= ~ATTR_SIZE
;
625 /* Optimization: if the end result is no change, don't RPC */
626 if (((attr
->ia_valid
& NFS_VALID_ATTRS
) & ~(ATTR_FILE
|ATTR_OPEN
)) == 0)
629 trace_nfs_setattr_enter(inode
);
631 /* Write all dirty data */
632 if (S_ISREG(inode
->i_mode
))
633 nfs_sync_inode(inode
);
635 fattr
= nfs_alloc_fattr();
641 error
= NFS_PROTO(inode
)->setattr(dentry
, fattr
, attr
);
643 error
= nfs_refresh_inode(inode
, fattr
);
644 nfs_free_fattr(fattr
);
646 trace_nfs_setattr_exit(inode
, error
);
649 EXPORT_SYMBOL_GPL(nfs_setattr
);
652 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
653 * @inode: inode of the file used
654 * @offset: file offset to start truncating
656 * This is a copy of the common vmtruncate, but with the locking
657 * corrected to take into account the fact that NFS requires
658 * inode->i_size to be updated under the inode->i_lock.
659 * Note: must be called with inode->i_lock held!
661 static int nfs_vmtruncate(struct inode
* inode
, loff_t offset
)
665 err
= inode_newsize_ok(inode
, offset
);
669 i_size_write(inode
, offset
);
672 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_DATA
|
673 NFS_INO_DATA_INVAL_DEFER
);
674 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_SIZE
;
676 spin_unlock(&inode
->i_lock
);
677 truncate_pagecache(inode
, offset
);
678 spin_lock(&inode
->i_lock
);
684 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
685 * @inode: pointer to struct inode
686 * @attr: pointer to struct iattr
687 * @fattr: pointer to struct nfs_fattr
689 * Note: we do this in the *proc.c in order to ensure that
690 * it works for things like exclusive creates too.
692 void nfs_setattr_update_inode(struct inode
*inode
, struct iattr
*attr
,
693 struct nfs_fattr
*fattr
)
695 /* Barrier: bump the attribute generation count. */
696 nfs_fattr_set_barrier(fattr
);
698 spin_lock(&inode
->i_lock
);
699 NFS_I(inode
)->attr_gencount
= fattr
->gencount
;
700 if ((attr
->ia_valid
& ATTR_SIZE
) != 0) {
701 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
|
702 NFS_INO_INVALID_BLOCKS
);
703 nfs_inc_stats(inode
, NFSIOS_SETATTRTRUNC
);
704 nfs_vmtruncate(inode
, attr
->ia_size
);
706 if ((attr
->ia_valid
& (ATTR_MODE
|ATTR_UID
|ATTR_GID
)) != 0) {
707 NFS_I(inode
)->cache_validity
&= ~NFS_INO_INVALID_CTIME
;
708 if ((attr
->ia_valid
& ATTR_KILL_SUID
) != 0 &&
709 inode
->i_mode
& S_ISUID
)
710 inode
->i_mode
&= ~S_ISUID
;
711 if ((attr
->ia_valid
& ATTR_KILL_SGID
) != 0 &&
712 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) ==
714 inode
->i_mode
&= ~S_ISGID
;
715 if ((attr
->ia_valid
& ATTR_MODE
) != 0) {
716 int mode
= attr
->ia_mode
& S_IALLUGO
;
717 mode
|= inode
->i_mode
& ~S_IALLUGO
;
718 inode
->i_mode
= mode
;
720 if ((attr
->ia_valid
& ATTR_UID
) != 0)
721 inode
->i_uid
= attr
->ia_uid
;
722 if ((attr
->ia_valid
& ATTR_GID
) != 0)
723 inode
->i_gid
= attr
->ia_gid
;
724 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
725 inode
->i_ctime
= fattr
->ctime
;
727 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
728 | NFS_INO_INVALID_CTIME
);
729 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ACCESS
730 | NFS_INO_INVALID_ACL
);
732 if (attr
->ia_valid
& (ATTR_ATIME_SET
|ATTR_ATIME
)) {
733 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_ATIME
734 | NFS_INO_INVALID_CTIME
);
735 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
736 inode
->i_atime
= fattr
->atime
;
737 else if (attr
->ia_valid
& ATTR_ATIME_SET
)
738 inode
->i_atime
= attr
->ia_atime
;
740 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_ATIME
);
742 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
743 inode
->i_ctime
= fattr
->ctime
;
745 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
746 | NFS_INO_INVALID_CTIME
);
748 if (attr
->ia_valid
& (ATTR_MTIME_SET
|ATTR_MTIME
)) {
749 NFS_I(inode
)->cache_validity
&= ~(NFS_INO_INVALID_MTIME
750 | NFS_INO_INVALID_CTIME
);
751 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
752 inode
->i_mtime
= fattr
->mtime
;
753 else if (attr
->ia_valid
& ATTR_MTIME_SET
)
754 inode
->i_mtime
= attr
->ia_mtime
;
756 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_MTIME
);
758 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
759 inode
->i_ctime
= fattr
->ctime
;
761 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
762 | NFS_INO_INVALID_CTIME
);
765 nfs_update_inode(inode
, fattr
);
766 spin_unlock(&inode
->i_lock
);
768 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode
);
770 static void nfs_readdirplus_parent_cache_miss(struct dentry
*dentry
)
772 struct dentry
*parent
;
774 if (!nfs_server_capable(d_inode(dentry
), NFS_CAP_READDIRPLUS
))
776 parent
= dget_parent(dentry
);
777 nfs_force_use_readdirplus(d_inode(parent
));
781 static void nfs_readdirplus_parent_cache_hit(struct dentry
*dentry
)
783 struct dentry
*parent
;
785 if (!nfs_server_capable(d_inode(dentry
), NFS_CAP_READDIRPLUS
))
787 parent
= dget_parent(dentry
);
788 nfs_advise_use_readdirplus(d_inode(parent
));
792 static u32
nfs_get_valid_attrmask(struct inode
*inode
)
794 unsigned long cache_validity
= READ_ONCE(NFS_I(inode
)->cache_validity
);
795 u32 reply_mask
= STATX_INO
| STATX_TYPE
;
797 if (!(cache_validity
& NFS_INO_INVALID_ATIME
))
798 reply_mask
|= STATX_ATIME
;
799 if (!(cache_validity
& NFS_INO_INVALID_CTIME
))
800 reply_mask
|= STATX_CTIME
;
801 if (!(cache_validity
& NFS_INO_INVALID_MTIME
))
802 reply_mask
|= STATX_MTIME
;
803 if (!(cache_validity
& NFS_INO_INVALID_SIZE
))
804 reply_mask
|= STATX_SIZE
;
805 if (!(cache_validity
& NFS_INO_INVALID_NLINK
))
806 reply_mask
|= STATX_NLINK
;
807 if (!(cache_validity
& NFS_INO_INVALID_MODE
))
808 reply_mask
|= STATX_MODE
;
809 if (!(cache_validity
& NFS_INO_INVALID_OTHER
))
810 reply_mask
|= STATX_UID
| STATX_GID
;
811 if (!(cache_validity
& NFS_INO_INVALID_BLOCKS
))
812 reply_mask
|= STATX_BLOCKS
;
816 int nfs_getattr(struct user_namespace
*mnt_userns
, const struct path
*path
,
817 struct kstat
*stat
, u32 request_mask
, unsigned int query_flags
)
819 struct inode
*inode
= d_inode(path
->dentry
);
820 struct nfs_server
*server
= NFS_SERVER(inode
);
821 unsigned long cache_validity
;
823 bool force_sync
= query_flags
& AT_STATX_FORCE_SYNC
;
824 bool do_update
= false;
826 trace_nfs_getattr_enter(inode
);
828 request_mask
&= STATX_TYPE
| STATX_MODE
| STATX_NLINK
| STATX_UID
|
829 STATX_GID
| STATX_ATIME
| STATX_MTIME
| STATX_CTIME
|
830 STATX_INO
| STATX_SIZE
| STATX_BLOCKS
;
832 if ((query_flags
& AT_STATX_DONT_SYNC
) && !force_sync
) {
833 nfs_readdirplus_parent_cache_hit(path
->dentry
);
834 goto out_no_revalidate
;
837 /* Flush out writes to the server in order to update c/mtime. */
838 if ((request_mask
& (STATX_CTIME
|STATX_MTIME
)) &&
839 S_ISREG(inode
->i_mode
)) {
840 err
= filemap_write_and_wait(inode
->i_mapping
);
846 * We may force a getattr if the user cares about atime.
848 * Note that we only have to check the vfsmount flags here:
849 * - NFS always sets S_NOATIME by so checking it would give a
851 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
852 * no point in checking those.
854 if ((path
->mnt
->mnt_flags
& MNT_NOATIME
) ||
855 ((path
->mnt
->mnt_flags
& MNT_NODIRATIME
) && S_ISDIR(inode
->i_mode
)))
856 request_mask
&= ~STATX_ATIME
;
858 /* Is the user requesting attributes that might need revalidation? */
859 if (!(request_mask
& (STATX_MODE
|STATX_NLINK
|STATX_ATIME
|STATX_CTIME
|
860 STATX_MTIME
|STATX_UID
|STATX_GID
|
861 STATX_SIZE
|STATX_BLOCKS
)))
862 goto out_no_revalidate
;
864 /* Check whether the cached attributes are stale */
865 do_update
|= force_sync
|| nfs_attribute_cache_expired(inode
);
866 cache_validity
= READ_ONCE(NFS_I(inode
)->cache_validity
);
867 do_update
|= cache_validity
& NFS_INO_INVALID_CHANGE
;
868 if (request_mask
& STATX_ATIME
)
869 do_update
|= cache_validity
& NFS_INO_INVALID_ATIME
;
870 if (request_mask
& STATX_CTIME
)
871 do_update
|= cache_validity
& NFS_INO_INVALID_CTIME
;
872 if (request_mask
& STATX_MTIME
)
873 do_update
|= cache_validity
& NFS_INO_INVALID_MTIME
;
874 if (request_mask
& STATX_SIZE
)
875 do_update
|= cache_validity
& NFS_INO_INVALID_SIZE
;
876 if (request_mask
& STATX_NLINK
)
877 do_update
|= cache_validity
& NFS_INO_INVALID_NLINK
;
878 if (request_mask
& STATX_MODE
)
879 do_update
|= cache_validity
& NFS_INO_INVALID_MODE
;
880 if (request_mask
& (STATX_UID
| STATX_GID
))
881 do_update
|= cache_validity
& NFS_INO_INVALID_OTHER
;
882 if (request_mask
& STATX_BLOCKS
)
883 do_update
|= cache_validity
& NFS_INO_INVALID_BLOCKS
;
886 /* Update the attribute cache */
887 if (!(server
->flags
& NFS_MOUNT_NOAC
))
888 nfs_readdirplus_parent_cache_miss(path
->dentry
);
890 nfs_readdirplus_parent_cache_hit(path
->dentry
);
891 err
= __nfs_revalidate_inode(server
, inode
);
895 nfs_readdirplus_parent_cache_hit(path
->dentry
);
897 /* Only return attributes that were revalidated. */
898 stat
->result_mask
= nfs_get_valid_attrmask(inode
) | request_mask
;
900 generic_fillattr(&init_user_ns
, inode
, stat
);
901 stat
->ino
= nfs_compat_user_ino64(NFS_FILEID(inode
));
902 if (S_ISDIR(inode
->i_mode
))
903 stat
->blksize
= NFS_SERVER(inode
)->dtsize
;
905 trace_nfs_getattr_exit(inode
, err
);
908 EXPORT_SYMBOL_GPL(nfs_getattr
);
910 static void nfs_init_lock_context(struct nfs_lock_context
*l_ctx
)
912 refcount_set(&l_ctx
->count
, 1);
913 l_ctx
->lockowner
= current
->files
;
914 INIT_LIST_HEAD(&l_ctx
->list
);
915 atomic_set(&l_ctx
->io_count
, 0);
918 static struct nfs_lock_context
*__nfs_find_lock_context(struct nfs_open_context
*ctx
)
920 struct nfs_lock_context
*pos
;
922 list_for_each_entry_rcu(pos
, &ctx
->lock_context
.list
, list
) {
923 if (pos
->lockowner
!= current
->files
)
925 if (refcount_inc_not_zero(&pos
->count
))
931 struct nfs_lock_context
*nfs_get_lock_context(struct nfs_open_context
*ctx
)
933 struct nfs_lock_context
*res
, *new = NULL
;
934 struct inode
*inode
= d_inode(ctx
->dentry
);
937 res
= __nfs_find_lock_context(ctx
);
940 new = kmalloc(sizeof(*new), GFP_KERNEL
);
942 return ERR_PTR(-ENOMEM
);
943 nfs_init_lock_context(new);
944 spin_lock(&inode
->i_lock
);
945 res
= __nfs_find_lock_context(ctx
);
947 new->open_context
= get_nfs_open_context(ctx
);
948 if (new->open_context
) {
949 list_add_tail_rcu(&new->list
,
950 &ctx
->lock_context
.list
);
954 res
= ERR_PTR(-EBADF
);
956 spin_unlock(&inode
->i_lock
);
961 EXPORT_SYMBOL_GPL(nfs_get_lock_context
);
963 void nfs_put_lock_context(struct nfs_lock_context
*l_ctx
)
965 struct nfs_open_context
*ctx
= l_ctx
->open_context
;
966 struct inode
*inode
= d_inode(ctx
->dentry
);
968 if (!refcount_dec_and_lock(&l_ctx
->count
, &inode
->i_lock
))
970 list_del_rcu(&l_ctx
->list
);
971 spin_unlock(&inode
->i_lock
);
972 put_nfs_open_context(ctx
);
973 kfree_rcu(l_ctx
, rcu_head
);
975 EXPORT_SYMBOL_GPL(nfs_put_lock_context
);
978 * nfs_close_context - Common close_context() routine NFSv2/v3
979 * @ctx: pointer to context
980 * @is_sync: is this a synchronous close
982 * Ensure that the attributes are up to date if we're mounted
983 * with close-to-open semantics and we have cached data that will
984 * need to be revalidated on open.
986 void nfs_close_context(struct nfs_open_context
*ctx
, int is_sync
)
988 struct nfs_inode
*nfsi
;
991 if (!(ctx
->mode
& FMODE_WRITE
))
995 inode
= d_inode(ctx
->dentry
);
996 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
))
999 if (inode
->i_mapping
->nrpages
== 0)
1001 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1003 if (!list_empty(&nfsi
->open_files
))
1005 if (NFS_SERVER(inode
)->flags
& NFS_MOUNT_NOCTO
)
1007 nfs_revalidate_inode(inode
,
1008 NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_SIZE
);
1010 EXPORT_SYMBOL_GPL(nfs_close_context
);
1012 struct nfs_open_context
*alloc_nfs_open_context(struct dentry
*dentry
,
1016 struct nfs_open_context
*ctx
;
1018 ctx
= kmalloc(sizeof(*ctx
), GFP_KERNEL
);
1020 return ERR_PTR(-ENOMEM
);
1021 nfs_sb_active(dentry
->d_sb
);
1022 ctx
->dentry
= dget(dentry
);
1024 ctx
->cred
= get_cred(filp
->f_cred
);
1026 ctx
->cred
= get_current_cred();
1027 ctx
->ll_cred
= NULL
;
1032 ctx
->flock_owner
= (fl_owner_t
)filp
;
1033 nfs_init_lock_context(&ctx
->lock_context
);
1034 ctx
->lock_context
.open_context
= ctx
;
1035 INIT_LIST_HEAD(&ctx
->list
);
1036 ctx
->mdsthreshold
= NULL
;
1039 EXPORT_SYMBOL_GPL(alloc_nfs_open_context
);
1041 struct nfs_open_context
*get_nfs_open_context(struct nfs_open_context
*ctx
)
1043 if (ctx
!= NULL
&& refcount_inc_not_zero(&ctx
->lock_context
.count
))
1047 EXPORT_SYMBOL_GPL(get_nfs_open_context
);
1049 static void __put_nfs_open_context(struct nfs_open_context
*ctx
, int is_sync
)
1051 struct inode
*inode
= d_inode(ctx
->dentry
);
1052 struct super_block
*sb
= ctx
->dentry
->d_sb
;
1054 if (!refcount_dec_and_test(&ctx
->lock_context
.count
))
1056 if (!list_empty(&ctx
->list
)) {
1057 spin_lock(&inode
->i_lock
);
1058 list_del_rcu(&ctx
->list
);
1059 spin_unlock(&inode
->i_lock
);
1062 NFS_PROTO(inode
)->close_context(ctx
, is_sync
);
1063 put_cred(ctx
->cred
);
1065 nfs_sb_deactive(sb
);
1066 put_rpccred(ctx
->ll_cred
);
1067 kfree(ctx
->mdsthreshold
);
1068 kfree_rcu(ctx
, rcu_head
);
1071 void put_nfs_open_context(struct nfs_open_context
*ctx
)
1073 __put_nfs_open_context(ctx
, 0);
1075 EXPORT_SYMBOL_GPL(put_nfs_open_context
);
1077 static void put_nfs_open_context_sync(struct nfs_open_context
*ctx
)
1079 __put_nfs_open_context(ctx
, 1);
1083 * Ensure that mmap has a recent RPC credential for use when writing out
1086 void nfs_inode_attach_open_context(struct nfs_open_context
*ctx
)
1088 struct inode
*inode
= d_inode(ctx
->dentry
);
1089 struct nfs_inode
*nfsi
= NFS_I(inode
);
1091 spin_lock(&inode
->i_lock
);
1092 if (list_empty(&nfsi
->open_files
) &&
1093 (nfsi
->cache_validity
& NFS_INO_DATA_INVAL_DEFER
))
1094 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
|
1095 NFS_INO_REVAL_FORCED
);
1096 list_add_tail_rcu(&ctx
->list
, &nfsi
->open_files
);
1097 spin_unlock(&inode
->i_lock
);
1099 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context
);
1101 void nfs_file_set_open_context(struct file
*filp
, struct nfs_open_context
*ctx
)
1103 filp
->private_data
= get_nfs_open_context(ctx
);
1104 set_bit(NFS_CONTEXT_FILE_OPEN
, &ctx
->flags
);
1105 if (list_empty(&ctx
->list
))
1106 nfs_inode_attach_open_context(ctx
);
1108 EXPORT_SYMBOL_GPL(nfs_file_set_open_context
);
1111 * Given an inode, search for an open context with the desired characteristics
1113 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, const struct cred
*cred
, fmode_t mode
)
1115 struct nfs_inode
*nfsi
= NFS_I(inode
);
1116 struct nfs_open_context
*pos
, *ctx
= NULL
;
1119 list_for_each_entry_rcu(pos
, &nfsi
->open_files
, list
) {
1120 if (cred
!= NULL
&& cred_fscmp(pos
->cred
, cred
) != 0)
1122 if ((pos
->mode
& (FMODE_READ
|FMODE_WRITE
)) != mode
)
1124 if (!test_bit(NFS_CONTEXT_FILE_OPEN
, &pos
->flags
))
1126 ctx
= get_nfs_open_context(pos
);
1134 void nfs_file_clear_open_context(struct file
*filp
)
1136 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
1139 struct inode
*inode
= d_inode(ctx
->dentry
);
1141 clear_bit(NFS_CONTEXT_FILE_OPEN
, &ctx
->flags
);
1143 * We fatal error on write before. Try to writeback
1147 invalidate_inode_pages2(inode
->i_mapping
);
1148 filp
->private_data
= NULL
;
1149 put_nfs_open_context_sync(ctx
);
1154 * These allocate and release file read/write context information.
1156 int nfs_open(struct inode
*inode
, struct file
*filp
)
1158 struct nfs_open_context
*ctx
;
1160 ctx
= alloc_nfs_open_context(file_dentry(filp
), filp
->f_mode
, filp
);
1162 return PTR_ERR(ctx
);
1163 nfs_file_set_open_context(filp
, ctx
);
1164 put_nfs_open_context(ctx
);
1165 nfs_fscache_open_file(inode
, filp
);
1168 EXPORT_SYMBOL_GPL(nfs_open
);
1171 * This function is called whenever some part of NFS notices that
1172 * the cached attributes have to be refreshed.
1175 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
1177 int status
= -ESTALE
;
1178 struct nfs4_label
*label
= NULL
;
1179 struct nfs_fattr
*fattr
= NULL
;
1180 struct nfs_inode
*nfsi
= NFS_I(inode
);
1182 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Lu)\n",
1183 inode
->i_sb
->s_id
, (unsigned long long)NFS_FILEID(inode
));
1185 trace_nfs_revalidate_inode_enter(inode
);
1187 if (is_bad_inode(inode
))
1189 if (NFS_STALE(inode
))
1192 /* pNFS: Attributes aren't updated until we layoutcommit */
1193 if (S_ISREG(inode
->i_mode
)) {
1194 status
= pnfs_sync_inode(inode
, false);
1200 fattr
= nfs_alloc_fattr();
1204 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
1206 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1207 if (IS_ERR(label
)) {
1208 status
= PTR_ERR(label
);
1212 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), fattr
,
1215 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1217 (unsigned long long)NFS_FILEID(inode
), status
);
1220 /* A soft timeout occurred. Use cached information? */
1221 if (server
->flags
& NFS_MOUNT_SOFTREVAL
)
1225 if (!S_ISDIR(inode
->i_mode
))
1226 nfs_set_inode_stale(inode
);
1228 nfs_zap_caches(inode
);
1233 status
= nfs_refresh_inode(inode
, fattr
);
1235 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1237 (unsigned long long)NFS_FILEID(inode
), status
);
1241 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
1242 nfs_zap_acl_cache(inode
);
1244 nfs_setsecurity(inode
, fattr
, label
);
1246 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) revalidation complete\n",
1248 (unsigned long long)NFS_FILEID(inode
));
1251 nfs4_label_free(label
);
1253 nfs_free_fattr(fattr
);
1254 trace_nfs_revalidate_inode_exit(inode
, status
);
1258 int nfs_attribute_cache_expired(struct inode
*inode
)
1260 if (nfs_have_delegated_attributes(inode
))
1262 return nfs_attribute_timeout(inode
);
1266 * nfs_revalidate_inode - Revalidate the inode attributes
1267 * @inode: pointer to inode struct
1268 * @flags: cache flags to check
1270 * Updates inode attribute information by retrieving the data from the server.
1272 int nfs_revalidate_inode(struct inode
*inode
, unsigned long flags
)
1274 if (!nfs_check_cache_invalid(inode
, flags
))
1275 return NFS_STALE(inode
) ? -ESTALE
: 0;
1276 return __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1278 EXPORT_SYMBOL_GPL(nfs_revalidate_inode
);
1280 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1284 if (mapping
->nrpages
!= 0) {
1285 if (S_ISREG(inode
->i_mode
)) {
1286 ret
= nfs_sync_mapping(mapping
);
1290 ret
= invalidate_inode_pages2(mapping
);
1294 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
1295 nfs_fscache_wait_on_invalidate(inode
);
1297 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) data cache invalidated\n",
1299 (unsigned long long)NFS_FILEID(inode
));
1304 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1305 * @mapping: pointer to mapping
1307 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1309 int nfs_clear_invalid_mapping(struct address_space
*mapping
)
1311 struct inode
*inode
= mapping
->host
;
1312 struct nfs_inode
*nfsi
= NFS_I(inode
);
1313 unsigned long *bitlock
= &nfsi
->flags
;
1317 * We must clear NFS_INO_INVALID_DATA first to ensure that
1318 * invalidations that come in while we're shooting down the mappings
1319 * are respected. But, that leaves a race window where one revalidator
1320 * can clear the flag, and then another checks it before the mapping
1321 * gets invalidated. Fix that by serializing access to this part of
1324 * At the same time, we need to allow other tasks to see whether we
1325 * might be in the middle of invalidating the pages, so we only set
1326 * the bit lock here if it looks like we're going to be doing that.
1329 ret
= wait_on_bit_action(bitlock
, NFS_INO_INVALIDATING
,
1330 nfs_wait_bit_killable
, TASK_KILLABLE
);
1333 spin_lock(&inode
->i_lock
);
1334 if (test_bit(NFS_INO_INVALIDATING
, bitlock
)) {
1335 spin_unlock(&inode
->i_lock
);
1338 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1340 spin_unlock(&inode
->i_lock
);
1344 set_bit(NFS_INO_INVALIDATING
, bitlock
);
1346 nfsi
->cache_validity
&=
1347 ~(NFS_INO_INVALID_DATA
| NFS_INO_DATA_INVAL_DEFER
);
1348 spin_unlock(&inode
->i_lock
);
1349 trace_nfs_invalidate_mapping_enter(inode
);
1350 ret
= nfs_invalidate_mapping(inode
, mapping
);
1351 trace_nfs_invalidate_mapping_exit(inode
, ret
);
1353 clear_bit_unlock(NFS_INO_INVALIDATING
, bitlock
);
1354 smp_mb__after_atomic();
1355 wake_up_bit(bitlock
, NFS_INO_INVALIDATING
);
1360 bool nfs_mapping_need_revalidate_inode(struct inode
*inode
)
1362 return nfs_check_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
) ||
1366 int nfs_revalidate_mapping_rcu(struct inode
*inode
)
1368 struct nfs_inode
*nfsi
= NFS_I(inode
);
1369 unsigned long *bitlock
= &nfsi
->flags
;
1372 if (IS_SWAPFILE(inode
))
1374 if (nfs_mapping_need_revalidate_inode(inode
)) {
1378 spin_lock(&inode
->i_lock
);
1379 if (test_bit(NFS_INO_INVALIDATING
, bitlock
) ||
1380 (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
))
1382 spin_unlock(&inode
->i_lock
);
1388 * nfs_revalidate_mapping - Revalidate the pagecache
1389 * @inode: pointer to host inode
1390 * @mapping: pointer to mapping
1392 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1394 /* swapfiles are not supposed to be shared. */
1395 if (IS_SWAPFILE(inode
))
1398 if (nfs_mapping_need_revalidate_inode(inode
)) {
1399 int ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1404 return nfs_clear_invalid_mapping(mapping
);
1407 static bool nfs_file_has_writers(struct nfs_inode
*nfsi
)
1409 struct inode
*inode
= &nfsi
->vfs_inode
;
1411 if (!S_ISREG(inode
->i_mode
))
1413 if (list_empty(&nfsi
->open_files
))
1415 return inode_is_open_for_write(inode
);
1418 static bool nfs_file_has_buffered_writers(struct nfs_inode
*nfsi
)
1420 return nfs_file_has_writers(nfsi
) && nfs_file_io_is_buffered(nfsi
);
1423 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1425 struct timespec64 ts
;
1427 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
)
1428 && (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
1429 && inode_eq_iversion_raw(inode
, fattr
->pre_change_attr
)) {
1430 inode_set_iversion_raw(inode
, fattr
->change_attr
);
1431 if (S_ISDIR(inode
->i_mode
))
1432 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1433 else if (nfs_server_capable(inode
, NFS_CAP_XATTR
))
1434 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_XATTR
);
1436 /* If we have atomic WCC data, we may update some attributes */
1437 ts
= inode
->i_ctime
;
1438 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
)
1439 && (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
1440 && timespec64_equal(&ts
, &fattr
->pre_ctime
)) {
1441 inode
->i_ctime
= fattr
->ctime
;
1444 ts
= inode
->i_mtime
;
1445 if ((fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
)
1446 && (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
1447 && timespec64_equal(&ts
, &fattr
->pre_mtime
)) {
1448 inode
->i_mtime
= fattr
->mtime
;
1449 if (S_ISDIR(inode
->i_mode
))
1450 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1452 if ((fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
)
1453 && (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
1454 && i_size_read(inode
) == nfs_size_to_loff_t(fattr
->pre_size
)
1455 && !nfs_have_writebacks(inode
)) {
1456 i_size_write(inode
, nfs_size_to_loff_t(fattr
->size
));
1461 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1462 * @inode: pointer to inode
1463 * @fattr: updated attributes
1465 * Verifies the attribute cache. If we have just changed the attributes,
1466 * so that fattr carries weak cache consistency data, then it may
1467 * also update the ctime/mtime/change_attribute.
1469 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
1471 struct nfs_inode
*nfsi
= NFS_I(inode
);
1472 loff_t cur_size
, new_isize
;
1473 unsigned long invalid
= 0;
1474 struct timespec64 ts
;
1476 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
))
1479 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1480 /* Only a mounted-on-fileid? Just exit */
1481 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1483 /* Has the inode gone and changed behind our back? */
1484 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1485 /* Is this perhaps the mounted-on fileid? */
1486 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1487 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1491 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
))
1495 if (!nfs_file_has_buffered_writers(nfsi
)) {
1496 /* Verify a few of the more important attributes */
1497 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 && !inode_eq_iversion_raw(inode
, fattr
->change_attr
))
1498 invalid
|= NFS_INO_INVALID_CHANGE
;
1500 ts
= inode
->i_mtime
;
1501 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) && !timespec64_equal(&ts
, &fattr
->mtime
))
1502 invalid
|= NFS_INO_INVALID_MTIME
;
1504 ts
= inode
->i_ctime
;
1505 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) && !timespec64_equal(&ts
, &fattr
->ctime
))
1506 invalid
|= NFS_INO_INVALID_CTIME
;
1508 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1509 cur_size
= i_size_read(inode
);
1510 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1511 if (cur_size
!= new_isize
)
1512 invalid
|= NFS_INO_INVALID_SIZE
;
1516 /* Have any file permissions changed? */
1517 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) && (inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
))
1518 invalid
|= NFS_INO_INVALID_MODE
;
1519 if ((fattr
->valid
& NFS_ATTR_FATTR_OWNER
) && !uid_eq(inode
->i_uid
, fattr
->uid
))
1520 invalid
|= NFS_INO_INVALID_OTHER
;
1521 if ((fattr
->valid
& NFS_ATTR_FATTR_GROUP
) && !gid_eq(inode
->i_gid
, fattr
->gid
))
1522 invalid
|= NFS_INO_INVALID_OTHER
;
1524 /* Has the link count changed? */
1525 if ((fattr
->valid
& NFS_ATTR_FATTR_NLINK
) && inode
->i_nlink
!= fattr
->nlink
)
1526 invalid
|= NFS_INO_INVALID_NLINK
;
1528 ts
= inode
->i_atime
;
1529 if ((fattr
->valid
& NFS_ATTR_FATTR_ATIME
) && !timespec64_equal(&ts
, &fattr
->atime
))
1530 invalid
|= NFS_INO_INVALID_ATIME
;
1533 nfs_set_cache_invalid(inode
, invalid
);
1535 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1539 static atomic_long_t nfs_attr_generation_counter
;
1541 static unsigned long nfs_read_attr_generation_counter(void)
1543 return atomic_long_read(&nfs_attr_generation_counter
);
1546 unsigned long nfs_inc_attr_generation_counter(void)
1548 return atomic_long_inc_return(&nfs_attr_generation_counter
);
1550 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter
);
1552 void nfs_fattr_init(struct nfs_fattr
*fattr
)
1555 fattr
->time_start
= jiffies
;
1556 fattr
->gencount
= nfs_inc_attr_generation_counter();
1557 fattr
->owner_name
= NULL
;
1558 fattr
->group_name
= NULL
;
1560 EXPORT_SYMBOL_GPL(nfs_fattr_init
);
1563 * nfs_fattr_set_barrier
1564 * @fattr: attributes
1566 * Used to set a barrier after an attribute was updated. This
1567 * barrier ensures that older attributes from RPC calls that may
1568 * have raced with our update cannot clobber these new values.
1569 * Note that you are still responsible for ensuring that other
1570 * operations which change the attribute on the server do not
1573 void nfs_fattr_set_barrier(struct nfs_fattr
*fattr
)
1575 fattr
->gencount
= nfs_inc_attr_generation_counter();
1578 struct nfs_fattr
*nfs_alloc_fattr(void)
1580 struct nfs_fattr
*fattr
;
1582 fattr
= kmalloc(sizeof(*fattr
), GFP_NOFS
);
1584 nfs_fattr_init(fattr
);
1587 EXPORT_SYMBOL_GPL(nfs_alloc_fattr
);
1589 struct nfs_fh
*nfs_alloc_fhandle(void)
1593 fh
= kmalloc(sizeof(struct nfs_fh
), GFP_NOFS
);
1598 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle
);
1602 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1603 * in the same way that wireshark does
1607 * For debugging only.
1609 u32
_nfs_display_fhandle_hash(const struct nfs_fh
*fh
)
1611 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1612 * not on the result */
1613 return nfs_fhandle_hash(fh
);
1615 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash
);
1618 * _nfs_display_fhandle - display an NFS file handle on the console
1620 * @fh: file handle to display
1621 * @caption: display caption
1623 * For debugging only.
1625 void _nfs_display_fhandle(const struct nfs_fh
*fh
, const char *caption
)
1629 if (fh
== NULL
|| fh
->size
== 0) {
1630 printk(KERN_DEFAULT
"%s at %p is empty\n", caption
, fh
);
1634 printk(KERN_DEFAULT
"%s at %p is %u bytes, crc: 0x%08x:\n",
1635 caption
, fh
, fh
->size
, _nfs_display_fhandle_hash(fh
));
1636 for (i
= 0; i
< fh
->size
; i
+= 16) {
1637 __be32
*pos
= (__be32
*)&fh
->data
[i
];
1639 switch ((fh
->size
- i
- 1) >> 2) {
1641 printk(KERN_DEFAULT
" %08x\n",
1645 printk(KERN_DEFAULT
" %08x %08x\n",
1646 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1));
1649 printk(KERN_DEFAULT
" %08x %08x %08x\n",
1650 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1651 be32_to_cpup(pos
+ 2));
1654 printk(KERN_DEFAULT
" %08x %08x %08x %08x\n",
1655 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1656 be32_to_cpup(pos
+ 2), be32_to_cpup(pos
+ 3));
1660 EXPORT_SYMBOL_GPL(_nfs_display_fhandle
);
1664 * nfs_inode_attrs_cmp_generic - compare attributes
1665 * @fattr: attributes
1666 * @inode: pointer to inode
1668 * Attempt to divine whether or not an RPC call reply carrying stale
1669 * attributes got scheduled after another call carrying updated ones.
1670 * Note also the check for wraparound of 'attr_gencount'
1672 * The function returns '1' if it thinks the attributes in @fattr are
1673 * more recent than the ones cached in @inode. Otherwise it returns
1676 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr
*fattr
,
1677 const struct inode
*inode
)
1679 unsigned long attr_gencount
= NFS_I(inode
)->attr_gencount
;
1681 return (long)(fattr
->gencount
- attr_gencount
) > 0 ||
1682 (long)(attr_gencount
- nfs_read_attr_generation_counter()) > 0;
1686 * nfs_inode_attrs_cmp_monotonic - compare attributes
1687 * @fattr: attributes
1688 * @inode: pointer to inode
1690 * Attempt to divine whether or not an RPC call reply carrying stale
1691 * attributes got scheduled after another call carrying updated ones.
1693 * We assume that the server observes monotonic semantics for
1694 * the change attribute, so a larger value means that the attributes in
1695 * @fattr are more recent, in which case the function returns the
1697 * A return value of '0' indicates no measurable change
1698 * A return value of '-1' means that the attributes in @inode are
1701 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr
*fattr
,
1702 const struct inode
*inode
)
1704 s64 diff
= fattr
->change_attr
- inode_peek_iversion_raw(inode
);
1707 return diff
== 0 ? 0 : -1;
1711 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1712 * @fattr: attributes
1713 * @inode: pointer to inode
1715 * Attempt to divine whether or not an RPC call reply carrying stale
1716 * attributes got scheduled after another call carrying updated ones.
1718 * We assume that the server observes strictly monotonic semantics for
1719 * the change attribute, so a larger value means that the attributes in
1720 * @fattr are more recent, in which case the function returns the
1722 * A return value of '-1' means that the attributes in @inode are
1723 * more recent or unchanged.
1725 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr
*fattr
,
1726 const struct inode
*inode
)
1728 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
) > 0 ? 1 : -1;
1732 * nfs_inode_attrs_cmp - compare attributes
1733 * @fattr: attributes
1734 * @inode: pointer to inode
1736 * This function returns '1' if it thinks the attributes in @fattr are
1737 * more recent than the ones cached in @inode. It returns '-1' if
1738 * the attributes in @inode are more recent than the ones in @fattr,
1739 * and it returns 0 if not sure.
1741 static int nfs_inode_attrs_cmp(const struct nfs_fattr
*fattr
,
1742 const struct inode
*inode
)
1744 if (nfs_inode_attrs_cmp_generic(fattr
, inode
) > 0)
1746 switch (NFS_SERVER(inode
)->change_attr_type
) {
1747 case NFS4_CHANGE_TYPE_IS_UNDEFINED
:
1749 case NFS4_CHANGE_TYPE_IS_TIME_METADATA
:
1750 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1752 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
);
1754 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1756 return nfs_inode_attrs_cmp_strict_monotonic(fattr
, inode
);
1762 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1763 * @fattr: attributes
1764 * @inode: pointer to inode
1766 * Returns '1' if the last attribute update left the inode cached
1767 * attributes in a partially unrevalidated state, and @fattr
1768 * matches the change attribute of that partial update.
1769 * Otherwise returns '0'.
1771 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr
*fattr
,
1772 const struct inode
*inode
)
1774 const unsigned long check_valid
=
1775 NFS_INO_INVALID_ATIME
| NFS_INO_INVALID_CTIME
|
1776 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
1777 NFS_INO_INVALID_BLOCKS
| NFS_INO_INVALID_OTHER
|
1778 NFS_INO_INVALID_NLINK
;
1779 unsigned long cache_validity
= NFS_I(inode
)->cache_validity
;
1781 if (!(cache_validity
& NFS_INO_INVALID_CHANGE
) &&
1782 (cache_validity
& check_valid
) != 0 &&
1783 (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1784 nfs_inode_attrs_cmp_monotonic(fattr
, inode
) == 0)
1789 static int nfs_refresh_inode_locked(struct inode
*inode
,
1790 struct nfs_fattr
*fattr
)
1792 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1795 trace_nfs_refresh_inode_enter(inode
);
1797 if (attr_cmp
> 0 || nfs_inode_finish_partial_attr_update(fattr
, inode
))
1798 ret
= nfs_update_inode(inode
, fattr
);
1799 else if (attr_cmp
== 0)
1800 ret
= nfs_check_inode_attributes(inode
, fattr
);
1802 trace_nfs_refresh_inode_exit(inode
, ret
);
1807 * nfs_refresh_inode - try to update the inode attribute cache
1808 * @inode: pointer to inode
1809 * @fattr: updated attributes
1811 * Check that an RPC call that returned attributes has not overlapped with
1812 * other recent updates of the inode metadata, then decide whether it is
1813 * safe to do a full update of the inode attributes, or whether just to
1814 * call nfs_check_inode_attributes.
1816 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1820 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1822 spin_lock(&inode
->i_lock
);
1823 status
= nfs_refresh_inode_locked(inode
, fattr
);
1824 spin_unlock(&inode
->i_lock
);
1828 EXPORT_SYMBOL_GPL(nfs_refresh_inode
);
1830 static int nfs_post_op_update_inode_locked(struct inode
*inode
,
1831 struct nfs_fattr
*fattr
, unsigned int invalid
)
1833 if (S_ISDIR(inode
->i_mode
))
1834 invalid
|= NFS_INO_INVALID_DATA
;
1835 nfs_set_cache_invalid(inode
, invalid
);
1836 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1838 return nfs_refresh_inode_locked(inode
, fattr
);
1842 * nfs_post_op_update_inode - try to update the inode attribute cache
1843 * @inode: pointer to inode
1844 * @fattr: updated attributes
1846 * After an operation that has changed the inode metadata, mark the
1847 * attribute cache as being invalid, then try to update it.
1849 * NB: if the server didn't return any post op attributes, this
1850 * function will force the retrieval of attributes before the next
1851 * NFS request. Thus it should be used only for operations that
1852 * are expected to change one or more attributes, to avoid
1853 * unnecessary NFS requests and trips through nfs_update_inode().
1855 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1859 spin_lock(&inode
->i_lock
);
1860 nfs_fattr_set_barrier(fattr
);
1861 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1862 NFS_INO_INVALID_CHANGE
1863 | NFS_INO_INVALID_CTIME
1864 | NFS_INO_REVAL_FORCED
);
1865 spin_unlock(&inode
->i_lock
);
1869 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode
);
1872 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1873 * @inode: pointer to inode
1874 * @fattr: updated attributes
1876 * After an operation that has changed the inode metadata, mark the
1877 * attribute cache as being invalid, then try to update it. Fake up
1878 * weak cache consistency data, if none exist.
1880 * This function is mainly designed to be used by the ->write_done() functions.
1882 int nfs_post_op_update_inode_force_wcc_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1884 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1887 /* Don't do a WCC update if these attributes are already stale */
1890 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0 || !attr_cmp
) {
1891 fattr
->valid
&= ~(NFS_ATTR_FATTR_PRECHANGE
1892 | NFS_ATTR_FATTR_PRESIZE
1893 | NFS_ATTR_FATTR_PREMTIME
1894 | NFS_ATTR_FATTR_PRECTIME
);
1897 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1898 (fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
) == 0) {
1899 fattr
->pre_change_attr
= inode_peek_iversion_raw(inode
);
1900 fattr
->valid
|= NFS_ATTR_FATTR_PRECHANGE
;
1902 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) != 0 &&
1903 (fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
) == 0) {
1904 fattr
->pre_ctime
= inode
->i_ctime
;
1905 fattr
->valid
|= NFS_ATTR_FATTR_PRECTIME
;
1907 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) != 0 &&
1908 (fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
) == 0) {
1909 fattr
->pre_mtime
= inode
->i_mtime
;
1910 fattr
->valid
|= NFS_ATTR_FATTR_PREMTIME
;
1912 if ((fattr
->valid
& NFS_ATTR_FATTR_SIZE
) != 0 &&
1913 (fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
) == 0) {
1914 fattr
->pre_size
= i_size_read(inode
);
1915 fattr
->valid
|= NFS_ATTR_FATTR_PRESIZE
;
1918 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1919 NFS_INO_INVALID_CHANGE
1920 | NFS_INO_INVALID_CTIME
1921 | NFS_INO_INVALID_MTIME
1922 | NFS_INO_INVALID_BLOCKS
);
1927 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1928 * @inode: pointer to inode
1929 * @fattr: updated attributes
1931 * After an operation that has changed the inode metadata, mark the
1932 * attribute cache as being invalid, then try to update it. Fake up
1933 * weak cache consistency data, if none exist.
1935 * This function is mainly designed to be used by the ->write_done() functions.
1937 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
1941 spin_lock(&inode
->i_lock
);
1942 nfs_fattr_set_barrier(fattr
);
1943 status
= nfs_post_op_update_inode_force_wcc_locked(inode
, fattr
);
1944 spin_unlock(&inode
->i_lock
);
1947 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc
);
1951 * Many nfs protocol calls return the new file attributes after
1952 * an operation. Here we update the inode to reflect the state
1953 * of the server's inode.
1955 * This is a bit tricky because we have to make sure all dirty pages
1956 * have been sent off to the server before calling invalidate_inode_pages.
1957 * To make sure no other process adds more write requests while we try
1958 * our best to flush them, we make them sleep during the attribute refresh.
1960 * A very similar scenario holds for the dir cache.
1962 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1964 struct nfs_server
*server
= NFS_SERVER(inode
);
1965 struct nfs_inode
*nfsi
= NFS_I(inode
);
1966 loff_t cur_isize
, new_isize
;
1967 u64 fattr_supported
= server
->fattr_valid
;
1968 unsigned long invalid
= 0;
1969 unsigned long now
= jiffies
;
1970 unsigned long save_cache_validity
;
1971 bool have_writers
= nfs_file_has_buffered_writers(nfsi
);
1972 bool cache_revalidated
= true;
1973 bool attr_changed
= false;
1974 bool have_delegation
;
1976 dfprintk(VFS
, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1977 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1978 nfs_display_fhandle_hash(NFS_FH(inode
)),
1979 atomic_read(&inode
->i_count
), fattr
->valid
);
1981 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1982 /* Only a mounted-on-fileid? Just exit */
1983 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1985 /* Has the inode gone and changed behind our back? */
1986 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1987 /* Is this perhaps the mounted-on fileid? */
1988 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1989 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1991 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1992 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1993 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1994 inode
->i_sb
->s_id
, (long long)nfsi
->fileid
,
1995 (long long)fattr
->fileid
);
2000 * Make sure the inode's type hasn't changed.
2002 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
)) {
2004 * Big trouble! The inode has become a different object.
2006 printk(KERN_DEBUG
"NFS: %s: inode %lu mode changed, %07o to %07o\n",
2007 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
2011 /* Update the fsid? */
2012 if (S_ISDIR(inode
->i_mode
) && (fattr
->valid
& NFS_ATTR_FATTR_FSID
) &&
2013 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
2014 !IS_AUTOMOUNT(inode
))
2015 server
->fsid
= fattr
->fsid
;
2017 /* Save the delegation state before clearing cache_validity */
2018 have_delegation
= nfs_have_delegated_attributes(inode
);
2021 * Update the read time so we don't revalidate too often.
2023 nfsi
->read_cache_jiffies
= fattr
->time_start
;
2025 save_cache_validity
= nfsi
->cache_validity
;
2026 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
2027 | NFS_INO_INVALID_ATIME
2028 | NFS_INO_REVAL_FORCED
2029 | NFS_INO_INVALID_BLOCKS
);
2031 /* Do atomic weak cache consistency updates */
2032 nfs_wcc_update_inode(inode
, fattr
);
2034 if (pnfs_layoutcommit_outstanding(inode
)) {
2035 nfsi
->cache_validity
|=
2036 save_cache_validity
&
2037 (NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_CTIME
|
2038 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
2039 NFS_INO_INVALID_BLOCKS
);
2040 cache_revalidated
= false;
2043 /* More cache consistency checks */
2044 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) {
2045 if (!inode_eq_iversion_raw(inode
, fattr
->change_attr
)) {
2046 /* Could it be a race with writeback? */
2047 if (!(have_writers
|| have_delegation
)) {
2048 invalid
|= NFS_INO_INVALID_DATA
2049 | NFS_INO_INVALID_ACCESS
2050 | NFS_INO_INVALID_ACL
2051 | NFS_INO_INVALID_XATTR
;
2052 /* Force revalidate of all attributes */
2053 save_cache_validity
|= NFS_INO_INVALID_CTIME
2054 | NFS_INO_INVALID_MTIME
2055 | NFS_INO_INVALID_SIZE
2056 | NFS_INO_INVALID_BLOCKS
2057 | NFS_INO_INVALID_NLINK
2058 | NFS_INO_INVALID_MODE
2059 | NFS_INO_INVALID_OTHER
;
2060 if (S_ISDIR(inode
->i_mode
))
2061 nfs_force_lookup_revalidate(inode
);
2062 attr_changed
= true;
2063 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2066 } else if (!have_delegation
)
2067 nfsi
->cache_validity
|= NFS_INO_DATA_INVAL_DEFER
;
2068 inode_set_iversion_raw(inode
, fattr
->change_attr
);
2071 nfsi
->cache_validity
|=
2072 save_cache_validity
& NFS_INO_INVALID_CHANGE
;
2073 if (!have_delegation
||
2074 (nfsi
->cache_validity
& NFS_INO_INVALID_CHANGE
) != 0)
2075 cache_revalidated
= false;
2078 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
2079 inode
->i_mtime
= fattr
->mtime
;
2080 else if (fattr_supported
& NFS_ATTR_FATTR_MTIME
)
2081 nfsi
->cache_validity
|=
2082 save_cache_validity
& NFS_INO_INVALID_MTIME
;
2084 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
2085 inode
->i_ctime
= fattr
->ctime
;
2086 else if (fattr_supported
& NFS_ATTR_FATTR_CTIME
)
2087 nfsi
->cache_validity
|=
2088 save_cache_validity
& NFS_INO_INVALID_CTIME
;
2090 /* Check if our cached file size is stale */
2091 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
2092 new_isize
= nfs_size_to_loff_t(fattr
->size
);
2093 cur_isize
= i_size_read(inode
);
2094 if (new_isize
!= cur_isize
&& !have_delegation
) {
2095 /* Do we perhaps have any outstanding writes, or has
2096 * the file grown beyond our last write? */
2097 if (!nfs_have_writebacks(inode
) || new_isize
> cur_isize
) {
2098 i_size_write(inode
, new_isize
);
2100 invalid
|= NFS_INO_INVALID_DATA
;
2102 dprintk("NFS: isize change on server for file %s/%ld "
2106 (long long)cur_isize
,
2107 (long long)new_isize
);
2109 if (new_isize
== 0 &&
2110 !(fattr
->valid
& (NFS_ATTR_FATTR_SPACE_USED
|
2111 NFS_ATTR_FATTR_BLOCKS_USED
))) {
2112 fattr
->du
.nfs3
.used
= 0;
2113 fattr
->valid
|= NFS_ATTR_FATTR_SPACE_USED
;
2116 nfsi
->cache_validity
|=
2117 save_cache_validity
& NFS_INO_INVALID_SIZE
;
2119 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
2120 inode
->i_atime
= fattr
->atime
;
2121 else if (fattr_supported
& NFS_ATTR_FATTR_ATIME
)
2122 nfsi
->cache_validity
|=
2123 save_cache_validity
& NFS_INO_INVALID_ATIME
;
2125 if (fattr
->valid
& NFS_ATTR_FATTR_MODE
) {
2126 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)) {
2127 umode_t newmode
= inode
->i_mode
& S_IFMT
;
2128 newmode
|= fattr
->mode
& S_IALLUGO
;
2129 inode
->i_mode
= newmode
;
2130 invalid
|= NFS_INO_INVALID_ACCESS
2131 | NFS_INO_INVALID_ACL
;
2133 } else if (fattr_supported
& NFS_ATTR_FATTR_MODE
)
2134 nfsi
->cache_validity
|=
2135 save_cache_validity
& NFS_INO_INVALID_MODE
;
2137 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
) {
2138 if (!uid_eq(inode
->i_uid
, fattr
->uid
)) {
2139 invalid
|= NFS_INO_INVALID_ACCESS
2140 | NFS_INO_INVALID_ACL
;
2141 inode
->i_uid
= fattr
->uid
;
2143 } else if (fattr_supported
& NFS_ATTR_FATTR_OWNER
)
2144 nfsi
->cache_validity
|=
2145 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2147 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
) {
2148 if (!gid_eq(inode
->i_gid
, fattr
->gid
)) {
2149 invalid
|= NFS_INO_INVALID_ACCESS
2150 | NFS_INO_INVALID_ACL
;
2151 inode
->i_gid
= fattr
->gid
;
2153 } else if (fattr_supported
& NFS_ATTR_FATTR_GROUP
)
2154 nfsi
->cache_validity
|=
2155 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2157 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
) {
2158 if (inode
->i_nlink
!= fattr
->nlink
) {
2159 if (S_ISDIR(inode
->i_mode
))
2160 invalid
|= NFS_INO_INVALID_DATA
;
2161 set_nlink(inode
, fattr
->nlink
);
2163 } else if (fattr_supported
& NFS_ATTR_FATTR_NLINK
)
2164 nfsi
->cache_validity
|=
2165 save_cache_validity
& NFS_INO_INVALID_NLINK
;
2167 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
2169 * report the blocks in 512byte units
2171 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
2172 } else if (fattr_supported
& NFS_ATTR_FATTR_SPACE_USED
)
2173 nfsi
->cache_validity
|=
2174 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2176 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
)
2177 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
2178 else if (fattr_supported
& NFS_ATTR_FATTR_BLOCKS_USED
)
2179 nfsi
->cache_validity
|=
2180 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2182 /* Update attrtimeo value if we're out of the unstable period */
2184 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
2185 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
2186 nfsi
->attrtimeo_timestamp
= now
;
2187 /* Set barrier to be more recent than all outstanding updates */
2188 nfsi
->attr_gencount
= nfs_inc_attr_generation_counter();
2190 if (cache_revalidated
) {
2191 if (!time_in_range_open(now
, nfsi
->attrtimeo_timestamp
,
2192 nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
2193 nfsi
->attrtimeo
<<= 1;
2194 if (nfsi
->attrtimeo
> NFS_MAXATTRTIMEO(inode
))
2195 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
2197 nfsi
->attrtimeo_timestamp
= now
;
2199 /* Set the barrier to be more recent than this fattr */
2200 if ((long)(fattr
->gencount
- nfsi
->attr_gencount
) > 0)
2201 nfsi
->attr_gencount
= fattr
->gencount
;
2204 /* Don't invalidate the data if we were to blame */
2205 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
2206 || S_ISLNK(inode
->i_mode
)))
2207 invalid
&= ~NFS_INO_INVALID_DATA
;
2208 nfs_set_cache_invalid(inode
, invalid
);
2213 * No need to worry about unhashing the dentry, as the
2214 * lookup validation will know that the inode is bad.
2215 * (But we fall through to invalidate the caches.)
2217 nfs_set_inode_stale_locked(inode
);
2221 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
2223 struct nfs_inode
*nfsi
;
2224 nfsi
= kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
2228 nfsi
->cache_validity
= 0UL;
2229 #if IS_ENABLED(CONFIG_NFS_V4)
2230 nfsi
->nfs4_acl
= NULL
;
2231 #endif /* CONFIG_NFS_V4 */
2232 #ifdef CONFIG_NFS_V4_2
2233 nfsi
->xattr_cache
= NULL
;
2235 return &nfsi
->vfs_inode
;
2237 EXPORT_SYMBOL_GPL(nfs_alloc_inode
);
2239 void nfs_free_inode(struct inode
*inode
)
2241 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
2243 EXPORT_SYMBOL_GPL(nfs_free_inode
);
2245 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
2247 #if IS_ENABLED(CONFIG_NFS_V4)
2248 INIT_LIST_HEAD(&nfsi
->open_states
);
2249 nfsi
->delegation
= NULL
;
2250 init_rwsem(&nfsi
->rwsem
);
2251 nfsi
->layout
= NULL
;
2255 static void init_once(void *foo
)
2257 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
2259 inode_init_once(&nfsi
->vfs_inode
);
2260 INIT_LIST_HEAD(&nfsi
->open_files
);
2261 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
2262 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
2263 INIT_LIST_HEAD(&nfsi
->commit_info
.list
);
2264 atomic_long_set(&nfsi
->nrequests
, 0);
2265 atomic_long_set(&nfsi
->commit_info
.ncommit
, 0);
2266 atomic_set(&nfsi
->commit_info
.rpcs_out
, 0);
2267 init_rwsem(&nfsi
->rmdir_sem
);
2268 mutex_init(&nfsi
->commit_mutex
);
2269 nfs4_init_once(nfsi
);
2270 nfsi
->cache_change_attribute
= 0;
2273 static int __init
nfs_init_inodecache(void)
2275 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
2276 sizeof(struct nfs_inode
),
2277 0, (SLAB_RECLAIM_ACCOUNT
|
2278 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
2280 if (nfs_inode_cachep
== NULL
)
2286 static void nfs_destroy_inodecache(void)
2289 * Make sure all delayed rcu free inodes are flushed before we
2293 kmem_cache_destroy(nfs_inode_cachep
);
2296 struct workqueue_struct
*nfsiod_workqueue
;
2297 EXPORT_SYMBOL_GPL(nfsiod_workqueue
);
2300 * start up the nfsiod workqueue
2302 static int nfsiod_start(void)
2304 struct workqueue_struct
*wq
;
2305 dprintk("RPC: creating workqueue nfsiod\n");
2306 wq
= alloc_workqueue("nfsiod", WQ_MEM_RECLAIM
| WQ_UNBOUND
, 0);
2309 nfsiod_workqueue
= wq
;
2314 * Destroy the nfsiod workqueue
2316 static void nfsiod_stop(void)
2318 struct workqueue_struct
*wq
;
2320 wq
= nfsiod_workqueue
;
2323 nfsiod_workqueue
= NULL
;
2324 destroy_workqueue(wq
);
2327 unsigned int nfs_net_id
;
2328 EXPORT_SYMBOL_GPL(nfs_net_id
);
2330 static int nfs_net_init(struct net
*net
)
2332 nfs_clients_init(net
);
2333 return nfs_fs_proc_net_init(net
);
2336 static void nfs_net_exit(struct net
*net
)
2338 nfs_fs_proc_net_exit(net
);
2339 nfs_clients_exit(net
);
2342 static struct pernet_operations nfs_net_ops
= {
2343 .init
= nfs_net_init
,
2344 .exit
= nfs_net_exit
,
2346 .size
= sizeof(struct nfs_net
),
2352 static int __init
init_nfs_fs(void)
2356 err
= nfs_sysfs_init();
2360 err
= register_pernet_subsys(&nfs_net_ops
);
2364 err
= nfs_fscache_register();
2368 err
= nfsiod_start();
2372 err
= nfs_fs_proc_init();
2376 err
= nfs_init_nfspagecache();
2380 err
= nfs_init_inodecache();
2384 err
= nfs_init_readpagecache();
2388 err
= nfs_init_writepagecache();
2392 err
= nfs_init_directcache();
2396 rpc_proc_register(&init_net
, &nfs_rpcstat
);
2398 err
= register_nfs_fs();
2404 rpc_proc_unregister(&init_net
, "nfs");
2405 nfs_destroy_directcache();
2407 nfs_destroy_writepagecache();
2409 nfs_destroy_readpagecache();
2411 nfs_destroy_inodecache();
2413 nfs_destroy_nfspagecache();
2419 nfs_fscache_unregister();
2421 unregister_pernet_subsys(&nfs_net_ops
);
2428 static void __exit
exit_nfs_fs(void)
2430 nfs_destroy_directcache();
2431 nfs_destroy_writepagecache();
2432 nfs_destroy_readpagecache();
2433 nfs_destroy_inodecache();
2434 nfs_destroy_nfspagecache();
2435 nfs_fscache_unregister();
2436 unregister_pernet_subsys(&nfs_net_ops
);
2437 rpc_proc_unregister(&init_net
, "nfs");
2438 unregister_nfs_fs();
2444 /* Not quite true; I just maintain it */
2445 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2446 MODULE_LICENSE("GPL");
2447 module_param(enable_ino64
, bool, 0644);
2449 module_init(init_nfs_fs
)
2450 module_exit(exit_nfs_fs
)