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 if (list_empty(&ctx
->list
))
1105 nfs_inode_attach_open_context(ctx
);
1107 EXPORT_SYMBOL_GPL(nfs_file_set_open_context
);
1110 * Given an inode, search for an open context with the desired characteristics
1112 struct nfs_open_context
*nfs_find_open_context(struct inode
*inode
, const struct cred
*cred
, fmode_t mode
)
1114 struct nfs_inode
*nfsi
= NFS_I(inode
);
1115 struct nfs_open_context
*pos
, *ctx
= NULL
;
1118 list_for_each_entry_rcu(pos
, &nfsi
->open_files
, list
) {
1119 if (cred
!= NULL
&& cred_fscmp(pos
->cred
, cred
) != 0)
1121 if ((pos
->mode
& (FMODE_READ
|FMODE_WRITE
)) != mode
)
1123 ctx
= get_nfs_open_context(pos
);
1131 void nfs_file_clear_open_context(struct file
*filp
)
1133 struct nfs_open_context
*ctx
= nfs_file_open_context(filp
);
1136 struct inode
*inode
= d_inode(ctx
->dentry
);
1139 * We fatal error on write before. Try to writeback
1143 invalidate_inode_pages2(inode
->i_mapping
);
1144 filp
->private_data
= NULL
;
1145 put_nfs_open_context_sync(ctx
);
1150 * These allocate and release file read/write context information.
1152 int nfs_open(struct inode
*inode
, struct file
*filp
)
1154 struct nfs_open_context
*ctx
;
1156 ctx
= alloc_nfs_open_context(file_dentry(filp
), filp
->f_mode
, filp
);
1158 return PTR_ERR(ctx
);
1159 nfs_file_set_open_context(filp
, ctx
);
1160 put_nfs_open_context(ctx
);
1161 nfs_fscache_open_file(inode
, filp
);
1164 EXPORT_SYMBOL_GPL(nfs_open
);
1167 * This function is called whenever some part of NFS notices that
1168 * the cached attributes have to be refreshed.
1171 __nfs_revalidate_inode(struct nfs_server
*server
, struct inode
*inode
)
1173 int status
= -ESTALE
;
1174 struct nfs4_label
*label
= NULL
;
1175 struct nfs_fattr
*fattr
= NULL
;
1176 struct nfs_inode
*nfsi
= NFS_I(inode
);
1178 dfprintk(PAGECACHE
, "NFS: revalidating (%s/%Lu)\n",
1179 inode
->i_sb
->s_id
, (unsigned long long)NFS_FILEID(inode
));
1181 trace_nfs_revalidate_inode_enter(inode
);
1183 if (is_bad_inode(inode
))
1185 if (NFS_STALE(inode
))
1188 /* pNFS: Attributes aren't updated until we layoutcommit */
1189 if (S_ISREG(inode
->i_mode
)) {
1190 status
= pnfs_sync_inode(inode
, false);
1196 fattr
= nfs_alloc_fattr();
1200 nfs_inc_stats(inode
, NFSIOS_INODEREVALIDATE
);
1202 label
= nfs4_label_alloc(NFS_SERVER(inode
), GFP_KERNEL
);
1203 if (IS_ERR(label
)) {
1204 status
= PTR_ERR(label
);
1208 status
= NFS_PROTO(inode
)->getattr(server
, NFS_FH(inode
), fattr
,
1211 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1213 (unsigned long long)NFS_FILEID(inode
), status
);
1216 /* A soft timeout occurred. Use cached information? */
1217 if (server
->flags
& NFS_MOUNT_SOFTREVAL
)
1221 if (!S_ISDIR(inode
->i_mode
))
1222 nfs_set_inode_stale(inode
);
1224 nfs_zap_caches(inode
);
1229 status
= nfs_refresh_inode(inode
, fattr
);
1231 dfprintk(PAGECACHE
, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1233 (unsigned long long)NFS_FILEID(inode
), status
);
1237 if (nfsi
->cache_validity
& NFS_INO_INVALID_ACL
)
1238 nfs_zap_acl_cache(inode
);
1240 nfs_setsecurity(inode
, fattr
, label
);
1242 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) revalidation complete\n",
1244 (unsigned long long)NFS_FILEID(inode
));
1247 nfs4_label_free(label
);
1249 nfs_free_fattr(fattr
);
1250 trace_nfs_revalidate_inode_exit(inode
, status
);
1254 int nfs_attribute_cache_expired(struct inode
*inode
)
1256 if (nfs_have_delegated_attributes(inode
))
1258 return nfs_attribute_timeout(inode
);
1262 * nfs_revalidate_inode - Revalidate the inode attributes
1263 * @inode: pointer to inode struct
1264 * @flags: cache flags to check
1266 * Updates inode attribute information by retrieving the data from the server.
1268 int nfs_revalidate_inode(struct inode
*inode
, unsigned long flags
)
1270 if (!nfs_check_cache_invalid(inode
, flags
))
1271 return NFS_STALE(inode
) ? -ESTALE
: 0;
1272 return __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1274 EXPORT_SYMBOL_GPL(nfs_revalidate_inode
);
1276 static int nfs_invalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1280 if (mapping
->nrpages
!= 0) {
1281 if (S_ISREG(inode
->i_mode
)) {
1282 ret
= nfs_sync_mapping(mapping
);
1286 ret
= invalidate_inode_pages2(mapping
);
1290 nfs_inc_stats(inode
, NFSIOS_DATAINVALIDATE
);
1291 nfs_fscache_wait_on_invalidate(inode
);
1293 dfprintk(PAGECACHE
, "NFS: (%s/%Lu) data cache invalidated\n",
1295 (unsigned long long)NFS_FILEID(inode
));
1300 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1301 * @mapping: pointer to mapping
1303 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1305 int nfs_clear_invalid_mapping(struct address_space
*mapping
)
1307 struct inode
*inode
= mapping
->host
;
1308 struct nfs_inode
*nfsi
= NFS_I(inode
);
1309 unsigned long *bitlock
= &nfsi
->flags
;
1313 * We must clear NFS_INO_INVALID_DATA first to ensure that
1314 * invalidations that come in while we're shooting down the mappings
1315 * are respected. But, that leaves a race window where one revalidator
1316 * can clear the flag, and then another checks it before the mapping
1317 * gets invalidated. Fix that by serializing access to this part of
1320 * At the same time, we need to allow other tasks to see whether we
1321 * might be in the middle of invalidating the pages, so we only set
1322 * the bit lock here if it looks like we're going to be doing that.
1325 ret
= wait_on_bit_action(bitlock
, NFS_INO_INVALIDATING
,
1326 nfs_wait_bit_killable
, TASK_KILLABLE
);
1329 spin_lock(&inode
->i_lock
);
1330 if (test_bit(NFS_INO_INVALIDATING
, bitlock
)) {
1331 spin_unlock(&inode
->i_lock
);
1334 if (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
)
1336 spin_unlock(&inode
->i_lock
);
1340 set_bit(NFS_INO_INVALIDATING
, bitlock
);
1342 nfsi
->cache_validity
&=
1343 ~(NFS_INO_INVALID_DATA
| NFS_INO_DATA_INVAL_DEFER
);
1344 spin_unlock(&inode
->i_lock
);
1345 trace_nfs_invalidate_mapping_enter(inode
);
1346 ret
= nfs_invalidate_mapping(inode
, mapping
);
1347 trace_nfs_invalidate_mapping_exit(inode
, ret
);
1349 clear_bit_unlock(NFS_INO_INVALIDATING
, bitlock
);
1350 smp_mb__after_atomic();
1351 wake_up_bit(bitlock
, NFS_INO_INVALIDATING
);
1356 bool nfs_mapping_need_revalidate_inode(struct inode
*inode
)
1358 return nfs_check_cache_invalid(inode
, NFS_INO_INVALID_CHANGE
) ||
1362 int nfs_revalidate_mapping_rcu(struct inode
*inode
)
1364 struct nfs_inode
*nfsi
= NFS_I(inode
);
1365 unsigned long *bitlock
= &nfsi
->flags
;
1368 if (IS_SWAPFILE(inode
))
1370 if (nfs_mapping_need_revalidate_inode(inode
)) {
1374 spin_lock(&inode
->i_lock
);
1375 if (test_bit(NFS_INO_INVALIDATING
, bitlock
) ||
1376 (nfsi
->cache_validity
& NFS_INO_INVALID_DATA
))
1378 spin_unlock(&inode
->i_lock
);
1384 * nfs_revalidate_mapping - Revalidate the pagecache
1385 * @inode: pointer to host inode
1386 * @mapping: pointer to mapping
1388 int nfs_revalidate_mapping(struct inode
*inode
, struct address_space
*mapping
)
1390 /* swapfiles are not supposed to be shared. */
1391 if (IS_SWAPFILE(inode
))
1394 if (nfs_mapping_need_revalidate_inode(inode
)) {
1395 int ret
= __nfs_revalidate_inode(NFS_SERVER(inode
), inode
);
1400 return nfs_clear_invalid_mapping(mapping
);
1403 static bool nfs_file_has_writers(struct nfs_inode
*nfsi
)
1405 struct inode
*inode
= &nfsi
->vfs_inode
;
1407 if (!S_ISREG(inode
->i_mode
))
1409 if (list_empty(&nfsi
->open_files
))
1411 return inode_is_open_for_write(inode
);
1414 static bool nfs_file_has_buffered_writers(struct nfs_inode
*nfsi
)
1416 return nfs_file_has_writers(nfsi
) && nfs_file_io_is_buffered(nfsi
);
1419 static void nfs_wcc_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1421 struct timespec64 ts
;
1423 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
)
1424 && (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
)
1425 && inode_eq_iversion_raw(inode
, fattr
->pre_change_attr
)) {
1426 inode_set_iversion_raw(inode
, fattr
->change_attr
);
1427 if (S_ISDIR(inode
->i_mode
))
1428 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1429 else if (nfs_server_capable(inode
, NFS_CAP_XATTR
))
1430 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_XATTR
);
1432 /* If we have atomic WCC data, we may update some attributes */
1433 ts
= inode
->i_ctime
;
1434 if ((fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
)
1435 && (fattr
->valid
& NFS_ATTR_FATTR_CTIME
)
1436 && timespec64_equal(&ts
, &fattr
->pre_ctime
)) {
1437 inode
->i_ctime
= fattr
->ctime
;
1440 ts
= inode
->i_mtime
;
1441 if ((fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
)
1442 && (fattr
->valid
& NFS_ATTR_FATTR_MTIME
)
1443 && timespec64_equal(&ts
, &fattr
->pre_mtime
)) {
1444 inode
->i_mtime
= fattr
->mtime
;
1445 if (S_ISDIR(inode
->i_mode
))
1446 nfs_set_cache_invalid(inode
, NFS_INO_INVALID_DATA
);
1448 if ((fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
)
1449 && (fattr
->valid
& NFS_ATTR_FATTR_SIZE
)
1450 && i_size_read(inode
) == nfs_size_to_loff_t(fattr
->pre_size
)
1451 && !nfs_have_writebacks(inode
)) {
1452 i_size_write(inode
, nfs_size_to_loff_t(fattr
->size
));
1457 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1458 * @inode: pointer to inode
1459 * @fattr: updated attributes
1461 * Verifies the attribute cache. If we have just changed the attributes,
1462 * so that fattr carries weak cache consistency data, then it may
1463 * also update the ctime/mtime/change_attribute.
1465 static int nfs_check_inode_attributes(struct inode
*inode
, struct nfs_fattr
*fattr
)
1467 struct nfs_inode
*nfsi
= NFS_I(inode
);
1468 loff_t cur_size
, new_isize
;
1469 unsigned long invalid
= 0;
1470 struct timespec64 ts
;
1472 if (NFS_PROTO(inode
)->have_delegation(inode
, FMODE_READ
))
1475 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1476 /* Only a mounted-on-fileid? Just exit */
1477 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1479 /* Has the inode gone and changed behind our back? */
1480 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1481 /* Is this perhaps the mounted-on fileid? */
1482 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1483 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1487 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
))
1491 if (!nfs_file_has_buffered_writers(nfsi
)) {
1492 /* Verify a few of the more important attributes */
1493 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 && !inode_eq_iversion_raw(inode
, fattr
->change_attr
))
1494 invalid
|= NFS_INO_INVALID_CHANGE
;
1496 ts
= inode
->i_mtime
;
1497 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) && !timespec64_equal(&ts
, &fattr
->mtime
))
1498 invalid
|= NFS_INO_INVALID_MTIME
;
1500 ts
= inode
->i_ctime
;
1501 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) && !timespec64_equal(&ts
, &fattr
->ctime
))
1502 invalid
|= NFS_INO_INVALID_CTIME
;
1504 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
1505 cur_size
= i_size_read(inode
);
1506 new_isize
= nfs_size_to_loff_t(fattr
->size
);
1507 if (cur_size
!= new_isize
)
1508 invalid
|= NFS_INO_INVALID_SIZE
;
1512 /* Have any file permissions changed? */
1513 if ((fattr
->valid
& NFS_ATTR_FATTR_MODE
) && (inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
))
1514 invalid
|= NFS_INO_INVALID_MODE
;
1515 if ((fattr
->valid
& NFS_ATTR_FATTR_OWNER
) && !uid_eq(inode
->i_uid
, fattr
->uid
))
1516 invalid
|= NFS_INO_INVALID_OTHER
;
1517 if ((fattr
->valid
& NFS_ATTR_FATTR_GROUP
) && !gid_eq(inode
->i_gid
, fattr
->gid
))
1518 invalid
|= NFS_INO_INVALID_OTHER
;
1520 /* Has the link count changed? */
1521 if ((fattr
->valid
& NFS_ATTR_FATTR_NLINK
) && inode
->i_nlink
!= fattr
->nlink
)
1522 invalid
|= NFS_INO_INVALID_NLINK
;
1524 ts
= inode
->i_atime
;
1525 if ((fattr
->valid
& NFS_ATTR_FATTR_ATIME
) && !timespec64_equal(&ts
, &fattr
->atime
))
1526 invalid
|= NFS_INO_INVALID_ATIME
;
1529 nfs_set_cache_invalid(inode
, invalid
);
1531 nfsi
->read_cache_jiffies
= fattr
->time_start
;
1535 static atomic_long_t nfs_attr_generation_counter
;
1537 static unsigned long nfs_read_attr_generation_counter(void)
1539 return atomic_long_read(&nfs_attr_generation_counter
);
1542 unsigned long nfs_inc_attr_generation_counter(void)
1544 return atomic_long_inc_return(&nfs_attr_generation_counter
);
1546 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter
);
1548 void nfs_fattr_init(struct nfs_fattr
*fattr
)
1551 fattr
->time_start
= jiffies
;
1552 fattr
->gencount
= nfs_inc_attr_generation_counter();
1553 fattr
->owner_name
= NULL
;
1554 fattr
->group_name
= NULL
;
1556 EXPORT_SYMBOL_GPL(nfs_fattr_init
);
1559 * nfs_fattr_set_barrier
1560 * @fattr: attributes
1562 * Used to set a barrier after an attribute was updated. This
1563 * barrier ensures that older attributes from RPC calls that may
1564 * have raced with our update cannot clobber these new values.
1565 * Note that you are still responsible for ensuring that other
1566 * operations which change the attribute on the server do not
1569 void nfs_fattr_set_barrier(struct nfs_fattr
*fattr
)
1571 fattr
->gencount
= nfs_inc_attr_generation_counter();
1574 struct nfs_fattr
*nfs_alloc_fattr(void)
1576 struct nfs_fattr
*fattr
;
1578 fattr
= kmalloc(sizeof(*fattr
), GFP_NOFS
);
1580 nfs_fattr_init(fattr
);
1583 EXPORT_SYMBOL_GPL(nfs_alloc_fattr
);
1585 struct nfs_fh
*nfs_alloc_fhandle(void)
1589 fh
= kmalloc(sizeof(struct nfs_fh
), GFP_NOFS
);
1594 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle
);
1598 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1599 * in the same way that wireshark does
1603 * For debugging only.
1605 u32
_nfs_display_fhandle_hash(const struct nfs_fh
*fh
)
1607 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1608 * not on the result */
1609 return nfs_fhandle_hash(fh
);
1611 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash
);
1614 * _nfs_display_fhandle - display an NFS file handle on the console
1616 * @fh: file handle to display
1617 * @caption: display caption
1619 * For debugging only.
1621 void _nfs_display_fhandle(const struct nfs_fh
*fh
, const char *caption
)
1625 if (fh
== NULL
|| fh
->size
== 0) {
1626 printk(KERN_DEFAULT
"%s at %p is empty\n", caption
, fh
);
1630 printk(KERN_DEFAULT
"%s at %p is %u bytes, crc: 0x%08x:\n",
1631 caption
, fh
, fh
->size
, _nfs_display_fhandle_hash(fh
));
1632 for (i
= 0; i
< fh
->size
; i
+= 16) {
1633 __be32
*pos
= (__be32
*)&fh
->data
[i
];
1635 switch ((fh
->size
- i
- 1) >> 2) {
1637 printk(KERN_DEFAULT
" %08x\n",
1641 printk(KERN_DEFAULT
" %08x %08x\n",
1642 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1));
1645 printk(KERN_DEFAULT
" %08x %08x %08x\n",
1646 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1647 be32_to_cpup(pos
+ 2));
1650 printk(KERN_DEFAULT
" %08x %08x %08x %08x\n",
1651 be32_to_cpup(pos
), be32_to_cpup(pos
+ 1),
1652 be32_to_cpup(pos
+ 2), be32_to_cpup(pos
+ 3));
1656 EXPORT_SYMBOL_GPL(_nfs_display_fhandle
);
1660 * nfs_inode_attrs_cmp_generic - compare attributes
1661 * @fattr: attributes
1662 * @inode: pointer to inode
1664 * Attempt to divine whether or not an RPC call reply carrying stale
1665 * attributes got scheduled after another call carrying updated ones.
1666 * Note also the check for wraparound of 'attr_gencount'
1668 * The function returns '1' if it thinks the attributes in @fattr are
1669 * more recent than the ones cached in @inode. Otherwise it returns
1672 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr
*fattr
,
1673 const struct inode
*inode
)
1675 unsigned long attr_gencount
= NFS_I(inode
)->attr_gencount
;
1677 return (long)(fattr
->gencount
- attr_gencount
) > 0 ||
1678 (long)(attr_gencount
- nfs_read_attr_generation_counter()) > 0;
1682 * nfs_inode_attrs_cmp_monotonic - compare attributes
1683 * @fattr: attributes
1684 * @inode: pointer to inode
1686 * Attempt to divine whether or not an RPC call reply carrying stale
1687 * attributes got scheduled after another call carrying updated ones.
1689 * We assume that the server observes monotonic semantics for
1690 * the change attribute, so a larger value means that the attributes in
1691 * @fattr are more recent, in which case the function returns the
1693 * A return value of '0' indicates no measurable change
1694 * A return value of '-1' means that the attributes in @inode are
1697 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr
*fattr
,
1698 const struct inode
*inode
)
1700 s64 diff
= fattr
->change_attr
- inode_peek_iversion_raw(inode
);
1703 return diff
== 0 ? 0 : -1;
1707 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1708 * @fattr: attributes
1709 * @inode: pointer to inode
1711 * Attempt to divine whether or not an RPC call reply carrying stale
1712 * attributes got scheduled after another call carrying updated ones.
1714 * We assume that the server observes strictly monotonic semantics for
1715 * the change attribute, so a larger value means that the attributes in
1716 * @fattr are more recent, in which case the function returns the
1718 * A return value of '-1' means that the attributes in @inode are
1719 * more recent or unchanged.
1721 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr
*fattr
,
1722 const struct inode
*inode
)
1724 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
) > 0 ? 1 : -1;
1728 * nfs_inode_attrs_cmp - compare attributes
1729 * @fattr: attributes
1730 * @inode: pointer to inode
1732 * This function returns '1' if it thinks the attributes in @fattr are
1733 * more recent than the ones cached in @inode. It returns '-1' if
1734 * the attributes in @inode are more recent than the ones in @fattr,
1735 * and it returns 0 if not sure.
1737 static int nfs_inode_attrs_cmp(const struct nfs_fattr
*fattr
,
1738 const struct inode
*inode
)
1740 if (nfs_inode_attrs_cmp_generic(fattr
, inode
) > 0)
1742 switch (NFS_SERVER(inode
)->change_attr_type
) {
1743 case NFS4_CHANGE_TYPE_IS_UNDEFINED
:
1745 case NFS4_CHANGE_TYPE_IS_TIME_METADATA
:
1746 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1748 return nfs_inode_attrs_cmp_monotonic(fattr
, inode
);
1750 if (!(fattr
->valid
& NFS_ATTR_FATTR_CHANGE
))
1752 return nfs_inode_attrs_cmp_strict_monotonic(fattr
, inode
);
1758 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1759 * @fattr: attributes
1760 * @inode: pointer to inode
1762 * Returns '1' if the last attribute update left the inode cached
1763 * attributes in a partially unrevalidated state, and @fattr
1764 * matches the change attribute of that partial update.
1765 * Otherwise returns '0'.
1767 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr
*fattr
,
1768 const struct inode
*inode
)
1770 const unsigned long check_valid
=
1771 NFS_INO_INVALID_ATIME
| NFS_INO_INVALID_CTIME
|
1772 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
1773 NFS_INO_INVALID_BLOCKS
| NFS_INO_INVALID_OTHER
|
1774 NFS_INO_INVALID_NLINK
;
1775 unsigned long cache_validity
= NFS_I(inode
)->cache_validity
;
1777 if (!(cache_validity
& NFS_INO_INVALID_CHANGE
) &&
1778 (cache_validity
& check_valid
) != 0 &&
1779 (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1780 nfs_inode_attrs_cmp_monotonic(fattr
, inode
) == 0)
1785 static int nfs_refresh_inode_locked(struct inode
*inode
,
1786 struct nfs_fattr
*fattr
)
1788 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1791 trace_nfs_refresh_inode_enter(inode
);
1793 if (attr_cmp
> 0 || nfs_inode_finish_partial_attr_update(fattr
, inode
))
1794 ret
= nfs_update_inode(inode
, fattr
);
1795 else if (attr_cmp
== 0)
1796 ret
= nfs_check_inode_attributes(inode
, fattr
);
1798 trace_nfs_refresh_inode_exit(inode
, ret
);
1803 * nfs_refresh_inode - try to update the inode attribute cache
1804 * @inode: pointer to inode
1805 * @fattr: updated attributes
1807 * Check that an RPC call that returned attributes has not overlapped with
1808 * other recent updates of the inode metadata, then decide whether it is
1809 * safe to do a full update of the inode attributes, or whether just to
1810 * call nfs_check_inode_attributes.
1812 int nfs_refresh_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1816 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1818 spin_lock(&inode
->i_lock
);
1819 status
= nfs_refresh_inode_locked(inode
, fattr
);
1820 spin_unlock(&inode
->i_lock
);
1824 EXPORT_SYMBOL_GPL(nfs_refresh_inode
);
1826 static int nfs_post_op_update_inode_locked(struct inode
*inode
,
1827 struct nfs_fattr
*fattr
, unsigned int invalid
)
1829 if (S_ISDIR(inode
->i_mode
))
1830 invalid
|= NFS_INO_INVALID_DATA
;
1831 nfs_set_cache_invalid(inode
, invalid
);
1832 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0)
1834 return nfs_refresh_inode_locked(inode
, fattr
);
1838 * nfs_post_op_update_inode - try to update the inode attribute cache
1839 * @inode: pointer to inode
1840 * @fattr: updated attributes
1842 * After an operation that has changed the inode metadata, mark the
1843 * attribute cache as being invalid, then try to update it.
1845 * NB: if the server didn't return any post op attributes, this
1846 * function will force the retrieval of attributes before the next
1847 * NFS request. Thus it should be used only for operations that
1848 * are expected to change one or more attributes, to avoid
1849 * unnecessary NFS requests and trips through nfs_update_inode().
1851 int nfs_post_op_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1855 spin_lock(&inode
->i_lock
);
1856 nfs_fattr_set_barrier(fattr
);
1857 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1858 NFS_INO_INVALID_CHANGE
1859 | NFS_INO_INVALID_CTIME
1860 | NFS_INO_REVAL_FORCED
);
1861 spin_unlock(&inode
->i_lock
);
1865 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode
);
1868 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1869 * @inode: pointer to inode
1870 * @fattr: updated attributes
1872 * After an operation that has changed the inode metadata, mark the
1873 * attribute cache as being invalid, then try to update it. Fake up
1874 * weak cache consistency data, if none exist.
1876 * This function is mainly designed to be used by the ->write_done() functions.
1878 int nfs_post_op_update_inode_force_wcc_locked(struct inode
*inode
, struct nfs_fattr
*fattr
)
1880 int attr_cmp
= nfs_inode_attrs_cmp(fattr
, inode
);
1883 /* Don't do a WCC update if these attributes are already stale */
1886 if ((fattr
->valid
& NFS_ATTR_FATTR
) == 0 || !attr_cmp
) {
1887 fattr
->valid
&= ~(NFS_ATTR_FATTR_PRECHANGE
1888 | NFS_ATTR_FATTR_PRESIZE
1889 | NFS_ATTR_FATTR_PREMTIME
1890 | NFS_ATTR_FATTR_PRECTIME
);
1893 if ((fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) != 0 &&
1894 (fattr
->valid
& NFS_ATTR_FATTR_PRECHANGE
) == 0) {
1895 fattr
->pre_change_attr
= inode_peek_iversion_raw(inode
);
1896 fattr
->valid
|= NFS_ATTR_FATTR_PRECHANGE
;
1898 if ((fattr
->valid
& NFS_ATTR_FATTR_CTIME
) != 0 &&
1899 (fattr
->valid
& NFS_ATTR_FATTR_PRECTIME
) == 0) {
1900 fattr
->pre_ctime
= inode
->i_ctime
;
1901 fattr
->valid
|= NFS_ATTR_FATTR_PRECTIME
;
1903 if ((fattr
->valid
& NFS_ATTR_FATTR_MTIME
) != 0 &&
1904 (fattr
->valid
& NFS_ATTR_FATTR_PREMTIME
) == 0) {
1905 fattr
->pre_mtime
= inode
->i_mtime
;
1906 fattr
->valid
|= NFS_ATTR_FATTR_PREMTIME
;
1908 if ((fattr
->valid
& NFS_ATTR_FATTR_SIZE
) != 0 &&
1909 (fattr
->valid
& NFS_ATTR_FATTR_PRESIZE
) == 0) {
1910 fattr
->pre_size
= i_size_read(inode
);
1911 fattr
->valid
|= NFS_ATTR_FATTR_PRESIZE
;
1914 status
= nfs_post_op_update_inode_locked(inode
, fattr
,
1915 NFS_INO_INVALID_CHANGE
1916 | NFS_INO_INVALID_CTIME
1917 | NFS_INO_INVALID_MTIME
1918 | NFS_INO_INVALID_BLOCKS
);
1923 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1924 * @inode: pointer to inode
1925 * @fattr: updated attributes
1927 * After an operation that has changed the inode metadata, mark the
1928 * attribute cache as being invalid, then try to update it. Fake up
1929 * weak cache consistency data, if none exist.
1931 * This function is mainly designed to be used by the ->write_done() functions.
1933 int nfs_post_op_update_inode_force_wcc(struct inode
*inode
, struct nfs_fattr
*fattr
)
1937 spin_lock(&inode
->i_lock
);
1938 nfs_fattr_set_barrier(fattr
);
1939 status
= nfs_post_op_update_inode_force_wcc_locked(inode
, fattr
);
1940 spin_unlock(&inode
->i_lock
);
1943 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc
);
1947 * Many nfs protocol calls return the new file attributes after
1948 * an operation. Here we update the inode to reflect the state
1949 * of the server's inode.
1951 * This is a bit tricky because we have to make sure all dirty pages
1952 * have been sent off to the server before calling invalidate_inode_pages.
1953 * To make sure no other process adds more write requests while we try
1954 * our best to flush them, we make them sleep during the attribute refresh.
1956 * A very similar scenario holds for the dir cache.
1958 static int nfs_update_inode(struct inode
*inode
, struct nfs_fattr
*fattr
)
1960 struct nfs_server
*server
= NFS_SERVER(inode
);
1961 struct nfs_inode
*nfsi
= NFS_I(inode
);
1962 loff_t cur_isize
, new_isize
;
1963 u64 fattr_supported
= server
->fattr_valid
;
1964 unsigned long invalid
= 0;
1965 unsigned long now
= jiffies
;
1966 unsigned long save_cache_validity
;
1967 bool have_writers
= nfs_file_has_buffered_writers(nfsi
);
1968 bool cache_revalidated
= true;
1969 bool attr_changed
= false;
1970 bool have_delegation
;
1972 dfprintk(VFS
, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1973 __func__
, inode
->i_sb
->s_id
, inode
->i_ino
,
1974 nfs_display_fhandle_hash(NFS_FH(inode
)),
1975 atomic_read(&inode
->i_count
), fattr
->valid
);
1977 if (!(fattr
->valid
& NFS_ATTR_FATTR_FILEID
)) {
1978 /* Only a mounted-on-fileid? Just exit */
1979 if (fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
)
1981 /* Has the inode gone and changed behind our back? */
1982 } else if (nfsi
->fileid
!= fattr
->fileid
) {
1983 /* Is this perhaps the mounted-on fileid? */
1984 if ((fattr
->valid
& NFS_ATTR_FATTR_MOUNTED_ON_FILEID
) &&
1985 nfsi
->fileid
== fattr
->mounted_on_fileid
)
1987 printk(KERN_ERR
"NFS: server %s error: fileid changed\n"
1988 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1989 NFS_SERVER(inode
)->nfs_client
->cl_hostname
,
1990 inode
->i_sb
->s_id
, (long long)nfsi
->fileid
,
1991 (long long)fattr
->fileid
);
1996 * Make sure the inode's type hasn't changed.
1998 if ((fattr
->valid
& NFS_ATTR_FATTR_TYPE
) && inode_wrong_type(inode
, fattr
->mode
)) {
2000 * Big trouble! The inode has become a different object.
2002 printk(KERN_DEBUG
"NFS: %s: inode %lu mode changed, %07o to %07o\n",
2003 __func__
, inode
->i_ino
, inode
->i_mode
, fattr
->mode
);
2007 /* Update the fsid? */
2008 if (S_ISDIR(inode
->i_mode
) && (fattr
->valid
& NFS_ATTR_FATTR_FSID
) &&
2009 !nfs_fsid_equal(&server
->fsid
, &fattr
->fsid
) &&
2010 !IS_AUTOMOUNT(inode
))
2011 server
->fsid
= fattr
->fsid
;
2013 /* Save the delegation state before clearing cache_validity */
2014 have_delegation
= nfs_have_delegated_attributes(inode
);
2017 * Update the read time so we don't revalidate too often.
2019 nfsi
->read_cache_jiffies
= fattr
->time_start
;
2021 save_cache_validity
= nfsi
->cache_validity
;
2022 nfsi
->cache_validity
&= ~(NFS_INO_INVALID_ATTR
2023 | NFS_INO_INVALID_ATIME
2024 | NFS_INO_REVAL_FORCED
2025 | NFS_INO_INVALID_BLOCKS
);
2027 /* Do atomic weak cache consistency updates */
2028 nfs_wcc_update_inode(inode
, fattr
);
2030 if (pnfs_layoutcommit_outstanding(inode
)) {
2031 nfsi
->cache_validity
|=
2032 save_cache_validity
&
2033 (NFS_INO_INVALID_CHANGE
| NFS_INO_INVALID_CTIME
|
2034 NFS_INO_INVALID_MTIME
| NFS_INO_INVALID_SIZE
|
2035 NFS_INO_INVALID_BLOCKS
);
2036 cache_revalidated
= false;
2039 /* More cache consistency checks */
2040 if (fattr
->valid
& NFS_ATTR_FATTR_CHANGE
) {
2041 if (!inode_eq_iversion_raw(inode
, fattr
->change_attr
)) {
2042 /* Could it be a race with writeback? */
2043 if (!(have_writers
|| have_delegation
)) {
2044 invalid
|= NFS_INO_INVALID_DATA
2045 | NFS_INO_INVALID_ACCESS
2046 | NFS_INO_INVALID_ACL
2047 | NFS_INO_INVALID_XATTR
;
2048 /* Force revalidate of all attributes */
2049 save_cache_validity
|= NFS_INO_INVALID_CTIME
2050 | NFS_INO_INVALID_MTIME
2051 | NFS_INO_INVALID_SIZE
2052 | NFS_INO_INVALID_BLOCKS
2053 | NFS_INO_INVALID_NLINK
2054 | NFS_INO_INVALID_MODE
2055 | NFS_INO_INVALID_OTHER
;
2056 if (S_ISDIR(inode
->i_mode
))
2057 nfs_force_lookup_revalidate(inode
);
2058 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2061 } else if (!have_delegation
)
2062 nfsi
->cache_validity
|= NFS_INO_DATA_INVAL_DEFER
;
2063 inode_set_iversion_raw(inode
, fattr
->change_attr
);
2064 attr_changed
= true;
2067 nfsi
->cache_validity
|=
2068 save_cache_validity
& NFS_INO_INVALID_CHANGE
;
2069 cache_revalidated
= false;
2072 if (fattr
->valid
& NFS_ATTR_FATTR_MTIME
) {
2073 inode
->i_mtime
= fattr
->mtime
;
2074 } else if (fattr_supported
& NFS_ATTR_FATTR_MTIME
) {
2075 nfsi
->cache_validity
|=
2076 save_cache_validity
& NFS_INO_INVALID_MTIME
;
2077 cache_revalidated
= false;
2080 if (fattr
->valid
& NFS_ATTR_FATTR_CTIME
) {
2081 inode
->i_ctime
= fattr
->ctime
;
2082 } else if (fattr_supported
& NFS_ATTR_FATTR_CTIME
) {
2083 nfsi
->cache_validity
|=
2084 save_cache_validity
& NFS_INO_INVALID_CTIME
;
2085 cache_revalidated
= false;
2088 /* Check if our cached file size is stale */
2089 if (fattr
->valid
& NFS_ATTR_FATTR_SIZE
) {
2090 new_isize
= nfs_size_to_loff_t(fattr
->size
);
2091 cur_isize
= i_size_read(inode
);
2092 if (new_isize
!= cur_isize
&& !have_delegation
) {
2093 /* Do we perhaps have any outstanding writes, or has
2094 * the file grown beyond our last write? */
2095 if (!nfs_have_writebacks(inode
) || new_isize
> cur_isize
) {
2096 i_size_write(inode
, new_isize
);
2098 invalid
|= NFS_INO_INVALID_DATA
;
2099 attr_changed
= true;
2101 dprintk("NFS: isize change on server for file %s/%ld "
2105 (long long)cur_isize
,
2106 (long long)new_isize
);
2108 if (new_isize
== 0 &&
2109 !(fattr
->valid
& (NFS_ATTR_FATTR_SPACE_USED
|
2110 NFS_ATTR_FATTR_BLOCKS_USED
))) {
2111 fattr
->du
.nfs3
.used
= 0;
2112 fattr
->valid
|= NFS_ATTR_FATTR_SPACE_USED
;
2115 nfsi
->cache_validity
|=
2116 save_cache_validity
& NFS_INO_INVALID_SIZE
;
2117 cache_revalidated
= false;
2120 if (fattr
->valid
& NFS_ATTR_FATTR_ATIME
)
2121 inode
->i_atime
= fattr
->atime
;
2122 else if (fattr_supported
& NFS_ATTR_FATTR_ATIME
) {
2123 nfsi
->cache_validity
|=
2124 save_cache_validity
& NFS_INO_INVALID_ATIME
;
2125 cache_revalidated
= false;
2128 if (fattr
->valid
& NFS_ATTR_FATTR_MODE
) {
2129 if ((inode
->i_mode
& S_IALLUGO
) != (fattr
->mode
& S_IALLUGO
)) {
2130 umode_t newmode
= inode
->i_mode
& S_IFMT
;
2131 newmode
|= fattr
->mode
& S_IALLUGO
;
2132 inode
->i_mode
= newmode
;
2133 invalid
|= NFS_INO_INVALID_ACCESS
2134 | NFS_INO_INVALID_ACL
;
2135 attr_changed
= true;
2137 } else if (fattr_supported
& NFS_ATTR_FATTR_MODE
) {
2138 nfsi
->cache_validity
|=
2139 save_cache_validity
& NFS_INO_INVALID_MODE
;
2140 cache_revalidated
= false;
2143 if (fattr
->valid
& NFS_ATTR_FATTR_OWNER
) {
2144 if (!uid_eq(inode
->i_uid
, fattr
->uid
)) {
2145 invalid
|= NFS_INO_INVALID_ACCESS
2146 | NFS_INO_INVALID_ACL
;
2147 inode
->i_uid
= fattr
->uid
;
2148 attr_changed
= true;
2150 } else if (fattr_supported
& NFS_ATTR_FATTR_OWNER
) {
2151 nfsi
->cache_validity
|=
2152 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2153 cache_revalidated
= false;
2156 if (fattr
->valid
& NFS_ATTR_FATTR_GROUP
) {
2157 if (!gid_eq(inode
->i_gid
, fattr
->gid
)) {
2158 invalid
|= NFS_INO_INVALID_ACCESS
2159 | NFS_INO_INVALID_ACL
;
2160 inode
->i_gid
= fattr
->gid
;
2161 attr_changed
= true;
2163 } else if (fattr_supported
& NFS_ATTR_FATTR_GROUP
) {
2164 nfsi
->cache_validity
|=
2165 save_cache_validity
& NFS_INO_INVALID_OTHER
;
2166 cache_revalidated
= false;
2169 if (fattr
->valid
& NFS_ATTR_FATTR_NLINK
) {
2170 if (inode
->i_nlink
!= fattr
->nlink
) {
2171 if (S_ISDIR(inode
->i_mode
))
2172 invalid
|= NFS_INO_INVALID_DATA
;
2173 set_nlink(inode
, fattr
->nlink
);
2174 attr_changed
= true;
2176 } else if (fattr_supported
& NFS_ATTR_FATTR_NLINK
) {
2177 nfsi
->cache_validity
|=
2178 save_cache_validity
& NFS_INO_INVALID_NLINK
;
2179 cache_revalidated
= false;
2182 if (fattr
->valid
& NFS_ATTR_FATTR_SPACE_USED
) {
2184 * report the blocks in 512byte units
2186 inode
->i_blocks
= nfs_calc_block_size(fattr
->du
.nfs3
.used
);
2187 } else if (fattr_supported
& NFS_ATTR_FATTR_SPACE_USED
) {
2188 nfsi
->cache_validity
|=
2189 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2190 cache_revalidated
= false;
2193 if (fattr
->valid
& NFS_ATTR_FATTR_BLOCKS_USED
) {
2194 inode
->i_blocks
= fattr
->du
.nfs2
.blocks
;
2195 } else if (fattr_supported
& NFS_ATTR_FATTR_BLOCKS_USED
) {
2196 nfsi
->cache_validity
|=
2197 save_cache_validity
& NFS_INO_INVALID_BLOCKS
;
2198 cache_revalidated
= false;
2201 /* Update attrtimeo value if we're out of the unstable period */
2203 nfs_inc_stats(inode
, NFSIOS_ATTRINVALIDATE
);
2204 nfsi
->attrtimeo
= NFS_MINATTRTIMEO(inode
);
2205 nfsi
->attrtimeo_timestamp
= now
;
2206 /* Set barrier to be more recent than all outstanding updates */
2207 nfsi
->attr_gencount
= nfs_inc_attr_generation_counter();
2209 if (cache_revalidated
) {
2210 if (!time_in_range_open(now
, nfsi
->attrtimeo_timestamp
,
2211 nfsi
->attrtimeo_timestamp
+ nfsi
->attrtimeo
)) {
2212 nfsi
->attrtimeo
<<= 1;
2213 if (nfsi
->attrtimeo
> NFS_MAXATTRTIMEO(inode
))
2214 nfsi
->attrtimeo
= NFS_MAXATTRTIMEO(inode
);
2216 nfsi
->attrtimeo_timestamp
= now
;
2218 /* Set the barrier to be more recent than this fattr */
2219 if ((long)(fattr
->gencount
- nfsi
->attr_gencount
) > 0)
2220 nfsi
->attr_gencount
= fattr
->gencount
;
2223 /* Don't invalidate the data if we were to blame */
2224 if (!(S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
)
2225 || S_ISLNK(inode
->i_mode
)))
2226 invalid
&= ~NFS_INO_INVALID_DATA
;
2227 nfs_set_cache_invalid(inode
, invalid
);
2232 * No need to worry about unhashing the dentry, as the
2233 * lookup validation will know that the inode is bad.
2234 * (But we fall through to invalidate the caches.)
2236 nfs_set_inode_stale_locked(inode
);
2240 struct inode
*nfs_alloc_inode(struct super_block
*sb
)
2242 struct nfs_inode
*nfsi
;
2243 nfsi
= kmem_cache_alloc(nfs_inode_cachep
, GFP_KERNEL
);
2247 nfsi
->cache_validity
= 0UL;
2248 #if IS_ENABLED(CONFIG_NFS_V4)
2249 nfsi
->nfs4_acl
= NULL
;
2250 #endif /* CONFIG_NFS_V4 */
2251 #ifdef CONFIG_NFS_V4_2
2252 nfsi
->xattr_cache
= NULL
;
2254 return &nfsi
->vfs_inode
;
2256 EXPORT_SYMBOL_GPL(nfs_alloc_inode
);
2258 void nfs_free_inode(struct inode
*inode
)
2260 kmem_cache_free(nfs_inode_cachep
, NFS_I(inode
));
2262 EXPORT_SYMBOL_GPL(nfs_free_inode
);
2264 static inline void nfs4_init_once(struct nfs_inode
*nfsi
)
2266 #if IS_ENABLED(CONFIG_NFS_V4)
2267 INIT_LIST_HEAD(&nfsi
->open_states
);
2268 nfsi
->delegation
= NULL
;
2269 init_rwsem(&nfsi
->rwsem
);
2270 nfsi
->layout
= NULL
;
2274 static void init_once(void *foo
)
2276 struct nfs_inode
*nfsi
= (struct nfs_inode
*) foo
;
2278 inode_init_once(&nfsi
->vfs_inode
);
2279 INIT_LIST_HEAD(&nfsi
->open_files
);
2280 INIT_LIST_HEAD(&nfsi
->access_cache_entry_lru
);
2281 INIT_LIST_HEAD(&nfsi
->access_cache_inode_lru
);
2282 INIT_LIST_HEAD(&nfsi
->commit_info
.list
);
2283 atomic_long_set(&nfsi
->nrequests
, 0);
2284 atomic_long_set(&nfsi
->commit_info
.ncommit
, 0);
2285 atomic_set(&nfsi
->commit_info
.rpcs_out
, 0);
2286 init_rwsem(&nfsi
->rmdir_sem
);
2287 mutex_init(&nfsi
->commit_mutex
);
2288 nfs4_init_once(nfsi
);
2289 nfsi
->cache_change_attribute
= 0;
2292 static int __init
nfs_init_inodecache(void)
2294 nfs_inode_cachep
= kmem_cache_create("nfs_inode_cache",
2295 sizeof(struct nfs_inode
),
2296 0, (SLAB_RECLAIM_ACCOUNT
|
2297 SLAB_MEM_SPREAD
|SLAB_ACCOUNT
),
2299 if (nfs_inode_cachep
== NULL
)
2305 static void nfs_destroy_inodecache(void)
2308 * Make sure all delayed rcu free inodes are flushed before we
2312 kmem_cache_destroy(nfs_inode_cachep
);
2315 struct workqueue_struct
*nfsiod_workqueue
;
2316 EXPORT_SYMBOL_GPL(nfsiod_workqueue
);
2319 * start up the nfsiod workqueue
2321 static int nfsiod_start(void)
2323 struct workqueue_struct
*wq
;
2324 dprintk("RPC: creating workqueue nfsiod\n");
2325 wq
= alloc_workqueue("nfsiod", WQ_MEM_RECLAIM
| WQ_UNBOUND
, 0);
2328 nfsiod_workqueue
= wq
;
2333 * Destroy the nfsiod workqueue
2335 static void nfsiod_stop(void)
2337 struct workqueue_struct
*wq
;
2339 wq
= nfsiod_workqueue
;
2342 nfsiod_workqueue
= NULL
;
2343 destroy_workqueue(wq
);
2346 unsigned int nfs_net_id
;
2347 EXPORT_SYMBOL_GPL(nfs_net_id
);
2349 static int nfs_net_init(struct net
*net
)
2351 nfs_clients_init(net
);
2352 return nfs_fs_proc_net_init(net
);
2355 static void nfs_net_exit(struct net
*net
)
2357 nfs_fs_proc_net_exit(net
);
2358 nfs_clients_exit(net
);
2361 static struct pernet_operations nfs_net_ops
= {
2362 .init
= nfs_net_init
,
2363 .exit
= nfs_net_exit
,
2365 .size
= sizeof(struct nfs_net
),
2371 static int __init
init_nfs_fs(void)
2375 err
= nfs_sysfs_init();
2379 err
= register_pernet_subsys(&nfs_net_ops
);
2383 err
= nfs_fscache_register();
2387 err
= nfsiod_start();
2391 err
= nfs_fs_proc_init();
2395 err
= nfs_init_nfspagecache();
2399 err
= nfs_init_inodecache();
2403 err
= nfs_init_readpagecache();
2407 err
= nfs_init_writepagecache();
2411 err
= nfs_init_directcache();
2415 rpc_proc_register(&init_net
, &nfs_rpcstat
);
2417 err
= register_nfs_fs();
2423 rpc_proc_unregister(&init_net
, "nfs");
2424 nfs_destroy_directcache();
2426 nfs_destroy_writepagecache();
2428 nfs_destroy_readpagecache();
2430 nfs_destroy_inodecache();
2432 nfs_destroy_nfspagecache();
2438 nfs_fscache_unregister();
2440 unregister_pernet_subsys(&nfs_net_ops
);
2447 static void __exit
exit_nfs_fs(void)
2449 nfs_destroy_directcache();
2450 nfs_destroy_writepagecache();
2451 nfs_destroy_readpagecache();
2452 nfs_destroy_inodecache();
2453 nfs_destroy_nfspagecache();
2454 nfs_fscache_unregister();
2455 unregister_pernet_subsys(&nfs_net_ops
);
2456 rpc_proc_unregister(&init_net
, "nfs");
2457 unregister_nfs_fs();
2463 /* Not quite true; I just maintain it */
2464 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2465 MODULE_LICENSE("GPL");
2466 module_param(enable_ino64
, bool, 0644);
2468 module_init(init_nfs_fs
)
2469 module_exit(exit_nfs_fs
)