1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs inode and superblock handling functions
10 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
47 #include "delegation.h"
58 #define NFSDBG_FACILITY NFSDBG_VFS
60 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
62 /* Default is to see 64-bit inode numbers */
63 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
65 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
67 static struct kmem_cache * nfs_inode_cachep;
69 static inline unsigned long
70 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
72 return nfs_fileid_to_ino_t(fattr->fileid);
75 static int nfs_wait_killable(int mode)
77 freezable_schedule_unsafe();
78 if (signal_pending_state(mode, current))
83 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
85 return nfs_wait_killable(mode);
87 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
90 * nfs_compat_user_ino64 - returns the user-visible inode number
91 * @fileid: 64-bit fileid
93 * This function returns a 32-bit inode number if the boot parameter
94 * nfs.enable_ino64 is zero.
96 u64 nfs_compat_user_ino64(u64 fileid)
107 if (sizeof(ino) < sizeof(fileid))
108 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
112 int nfs_drop_inode(struct inode *inode)
114 return NFS_STALE(inode) || generic_drop_inode(inode);
116 EXPORT_SYMBOL_GPL(nfs_drop_inode);
118 void nfs_clear_inode(struct inode *inode)
121 * The following should never happen...
123 WARN_ON_ONCE(nfs_have_writebacks(inode));
124 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
125 nfs_zap_acl_cache(inode);
126 nfs_access_zap_cache(inode);
127 nfs_fscache_clear_inode(inode);
129 EXPORT_SYMBOL_GPL(nfs_clear_inode);
131 void nfs_evict_inode(struct inode *inode)
133 truncate_inode_pages_final(&inode->i_data);
135 nfs_clear_inode(inode);
138 int nfs_sync_inode(struct inode *inode)
140 inode_dio_wait(inode);
141 return nfs_wb_all(inode);
143 EXPORT_SYMBOL_GPL(nfs_sync_inode);
146 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 * @mapping: pointer to struct address_space
149 int nfs_sync_mapping(struct address_space *mapping)
153 if (mapping->nrpages != 0) {
154 unmap_mapping_range(mapping, 0, 0, 0);
155 ret = nfs_wb_all(mapping->host);
160 static int nfs_attribute_timeout(struct inode *inode)
162 struct nfs_inode *nfsi = NFS_I(inode);
164 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
167 static bool nfs_check_cache_flags_invalid(struct inode *inode,
170 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
172 return (cache_validity & flags) != 0;
175 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
177 if (nfs_check_cache_flags_invalid(inode, flags))
179 return nfs_attribute_cache_expired(inode);
181 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
183 #ifdef CONFIG_NFS_V4_2
184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
186 return nfsi->xattr_cache != NULL;
189 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
195 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
197 struct nfs_inode *nfsi = NFS_I(inode);
198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
200 if (have_delegation) {
201 if (!(flags & NFS_INO_REVAL_FORCED))
202 flags &= ~(NFS_INO_INVALID_MODE |
203 NFS_INO_INVALID_OTHER |
204 NFS_INO_INVALID_XATTR);
205 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
206 } else if (flags & NFS_INO_REVAL_PAGECACHE)
207 flags |= NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE;
209 if (!nfs_has_xattr_cache(nfsi))
210 flags &= ~NFS_INO_INVALID_XATTR;
211 if (flags & NFS_INO_INVALID_DATA)
212 nfs_fscache_invalidate(inode);
213 flags &= ~(NFS_INO_REVAL_PAGECACHE | NFS_INO_REVAL_FORCED);
215 nfsi->cache_validity |= flags;
217 if (inode->i_mapping->nrpages == 0)
218 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA |
219 NFS_INO_DATA_INVAL_DEFER);
220 else if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
221 nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER;
223 EXPORT_SYMBOL_GPL(nfs_set_cache_invalid);
226 * Invalidate the local caches
228 static void nfs_zap_caches_locked(struct inode *inode)
230 struct nfs_inode *nfsi = NFS_I(inode);
231 int mode = inode->i_mode;
233 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
235 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
236 nfsi->attrtimeo_timestamp = jiffies;
238 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
239 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
240 | NFS_INO_INVALID_DATA
241 | NFS_INO_INVALID_ACCESS
242 | NFS_INO_INVALID_ACL
243 | NFS_INO_INVALID_XATTR
244 | NFS_INO_REVAL_PAGECACHE);
246 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
247 | NFS_INO_INVALID_ACCESS
248 | NFS_INO_INVALID_ACL
249 | NFS_INO_INVALID_XATTR
250 | NFS_INO_REVAL_PAGECACHE);
251 nfs_zap_label_cache_locked(nfsi);
254 void nfs_zap_caches(struct inode *inode)
256 spin_lock(&inode->i_lock);
257 nfs_zap_caches_locked(inode);
258 spin_unlock(&inode->i_lock);
261 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
263 if (mapping->nrpages != 0) {
264 spin_lock(&inode->i_lock);
265 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
266 spin_unlock(&inode->i_lock);
270 void nfs_zap_acl_cache(struct inode *inode)
272 void (*clear_acl_cache)(struct inode *);
274 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
275 if (clear_acl_cache != NULL)
276 clear_acl_cache(inode);
277 spin_lock(&inode->i_lock);
278 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
279 spin_unlock(&inode->i_lock);
281 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
283 void nfs_invalidate_atime(struct inode *inode)
285 spin_lock(&inode->i_lock);
286 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
287 spin_unlock(&inode->i_lock);
289 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
292 * Invalidate, but do not unhash, the inode.
293 * NB: must be called with inode->i_lock held!
295 static void nfs_set_inode_stale_locked(struct inode *inode)
297 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
298 nfs_zap_caches_locked(inode);
299 trace_nfs_set_inode_stale(inode);
302 void nfs_set_inode_stale(struct inode *inode)
304 spin_lock(&inode->i_lock);
305 nfs_set_inode_stale_locked(inode);
306 spin_unlock(&inode->i_lock);
309 struct nfs_find_desc {
311 struct nfs_fattr *fattr;
315 * In NFSv3 we can have 64bit inode numbers. In order to support
316 * this, and re-exported directories (also seen in NFSv2)
317 * we are forced to allow 2 different inodes to have the same
321 nfs_find_actor(struct inode *inode, void *opaque)
323 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
324 struct nfs_fh *fh = desc->fh;
325 struct nfs_fattr *fattr = desc->fattr;
327 if (NFS_FILEID(inode) != fattr->fileid)
329 if (inode_wrong_type(inode, fattr->mode))
331 if (nfs_compare_fh(NFS_FH(inode), fh))
333 if (is_bad_inode(inode) || NFS_STALE(inode))
339 nfs_init_locked(struct inode *inode, void *opaque)
341 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
342 struct nfs_fattr *fattr = desc->fattr;
344 set_nfs_fileid(inode, fattr->fileid);
345 inode->i_mode = fattr->mode;
346 nfs_copy_fh(NFS_FH(inode), desc->fh);
350 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
351 static void nfs_clear_label_invalid(struct inode *inode)
353 spin_lock(&inode->i_lock);
354 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
355 spin_unlock(&inode->i_lock);
358 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
362 if (fattr->label == NULL)
365 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
366 error = security_inode_notifysecctx(inode, fattr->label->label,
369 printk(KERN_ERR "%s() %s %d "
370 "security_inode_notifysecctx() %d\n",
372 (char *)fattr->label->label,
373 fattr->label->len, error);
374 nfs_clear_label_invalid(inode);
378 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
380 struct nfs4_label *label;
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)
404 EXPORT_SYMBOL_GPL(nfs_setsecurity);
406 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
408 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
410 struct nfs_find_desc desc = {
417 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
418 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
421 hash = nfs_fattr_to_ino_t(fattr);
422 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
424 dprintk("%s: returning %p\n", __func__, inode);
428 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
430 atomic_long_set(&nfsi->nrequests, 0);
431 INIT_LIST_HEAD(&nfsi->commit_info.list);
432 atomic_long_set(&nfsi->commit_info.ncommit, 0);
433 atomic_set(&nfsi->commit_info.rpcs_out, 0);
434 mutex_init(&nfsi->commit_mutex);
437 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
439 nfsi->cache_change_attribute = 0;
440 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
441 init_rwsem(&nfsi->rmdir_sem);
445 * This is our front-end to iget that looks up inodes by file handle
446 * instead of inode number.
449 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
451 struct nfs_find_desc desc = {
455 struct inode *inode = ERR_PTR(-ENOENT);
456 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
459 nfs_attr_check_mountpoint(sb, fattr);
461 if (nfs_attr_use_mounted_on_fileid(fattr))
462 fattr->fileid = fattr->mounted_on_fileid;
463 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
465 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
468 hash = nfs_fattr_to_ino_t(fattr);
470 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
472 inode = ERR_PTR(-ENOMEM);
476 if (inode->i_state & I_NEW) {
477 struct nfs_inode *nfsi = NFS_I(inode);
478 unsigned long now = jiffies;
480 /* We set i_ino for the few things that still rely on it,
484 /* We can't support update_atime(), since the server will reset it */
485 inode->i_flags |= S_NOATIME|S_NOCMTIME;
486 inode->i_mode = fattr->mode;
487 nfsi->cache_validity = 0;
488 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
489 && (fattr_supported & NFS_ATTR_FATTR_MODE))
490 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
491 /* Why so? Because we want revalidate for devices/FIFOs, and
492 * that's precisely what we have in nfs_file_inode_operations.
494 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
495 if (S_ISREG(inode->i_mode)) {
496 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
497 inode->i_data.a_ops = &nfs_file_aops;
498 nfs_inode_init_regular(nfsi);
499 } else if (S_ISDIR(inode->i_mode)) {
500 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
501 inode->i_fop = &nfs_dir_operations;
502 inode->i_data.a_ops = &nfs_dir_aops;
503 nfs_inode_init_dir(nfsi);
504 /* Deal with crossing mountpoints */
505 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
506 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
507 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
508 inode->i_op = &nfs_referral_inode_operations;
510 inode->i_op = &nfs_mountpoint_inode_operations;
512 inode->i_flags |= S_AUTOMOUNT;
514 } else if (S_ISLNK(inode->i_mode)) {
515 inode->i_op = &nfs_symlink_inode_operations;
516 inode_nohighmem(inode);
518 init_special_inode(inode, inode->i_mode, fattr->rdev);
520 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
521 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
522 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
523 inode_set_iversion_raw(inode, 0);
526 inode->i_uid = make_kuid(&init_user_ns, -2);
527 inode->i_gid = make_kgid(&init_user_ns, -2);
532 nfsi->read_cache_jiffies = fattr->time_start;
533 nfsi->attr_gencount = fattr->gencount;
534 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
535 inode->i_atime = fattr->atime;
536 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
537 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
538 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
539 inode->i_mtime = fattr->mtime;
540 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
541 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
542 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
543 inode->i_ctime = fattr->ctime;
544 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
545 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
546 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
547 inode_set_iversion_raw(inode, fattr->change_attr);
549 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
550 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
551 inode->i_size = nfs_size_to_loff_t(fattr->size);
553 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
554 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
555 set_nlink(inode, fattr->nlink);
556 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
557 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
558 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
559 inode->i_uid = fattr->uid;
560 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
561 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
562 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
563 inode->i_gid = fattr->gid;
564 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
565 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
566 if (nfs_server_capable(inode, NFS_CAP_XATTR))
567 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
568 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
569 inode->i_blocks = fattr->du.nfs2.blocks;
570 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
572 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
573 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
575 * report the blocks in 512byte units
577 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
578 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
580 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
582 nfs_setsecurity(inode, fattr);
584 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
585 nfsi->attrtimeo_timestamp = now;
586 nfsi->access_cache = RB_ROOT;
588 nfs_fscache_init_inode(inode);
590 unlock_new_inode(inode);
592 int err = nfs_refresh_inode(inode, fattr);
595 inode = ERR_PTR(err);
599 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
601 (unsigned long long)NFS_FILEID(inode),
602 nfs_display_fhandle_hash(fh),
603 atomic_read(&inode->i_count));
609 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
612 EXPORT_SYMBOL_GPL(nfs_fhget);
614 #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)
617 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
620 struct inode *inode = d_inode(dentry);
621 struct nfs_fattr *fattr;
624 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
626 /* skip mode change if it's just for clearing setuid/setgid */
627 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
628 attr->ia_valid &= ~ATTR_MODE;
630 if (attr->ia_valid & ATTR_SIZE) {
631 BUG_ON(!S_ISREG(inode->i_mode));
633 error = inode_newsize_ok(inode, attr->ia_size);
637 if (attr->ia_size == i_size_read(inode))
638 attr->ia_valid &= ~ATTR_SIZE;
641 /* Optimization: if the end result is no change, don't RPC */
642 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
645 trace_nfs_setattr_enter(inode);
647 /* Write all dirty data */
648 if (S_ISREG(inode->i_mode))
649 nfs_sync_inode(inode);
651 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
657 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
659 error = nfs_refresh_inode(inode, fattr);
660 nfs_free_fattr(fattr);
662 trace_nfs_setattr_exit(inode, error);
665 EXPORT_SYMBOL_GPL(nfs_setattr);
668 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
669 * @inode: inode of the file used
670 * @offset: file offset to start truncating
672 * This is a copy of the common vmtruncate, but with the locking
673 * corrected to take into account the fact that NFS requires
674 * inode->i_size to be updated under the inode->i_lock.
675 * Note: must be called with inode->i_lock held!
677 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
681 err = inode_newsize_ok(inode, offset);
685 trace_nfs_size_truncate(inode, offset);
686 i_size_write(inode, offset);
689 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
690 NFS_INO_DATA_INVAL_DEFER);
691 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
693 spin_unlock(&inode->i_lock);
694 truncate_pagecache(inode, offset);
695 spin_lock(&inode->i_lock);
701 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
702 * @inode: pointer to struct inode
703 * @attr: pointer to struct iattr
704 * @fattr: pointer to struct nfs_fattr
706 * Note: we do this in the *proc.c in order to ensure that
707 * it works for things like exclusive creates too.
709 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
710 struct nfs_fattr *fattr)
712 /* Barrier: bump the attribute generation count. */
713 nfs_fattr_set_barrier(fattr);
715 spin_lock(&inode->i_lock);
716 NFS_I(inode)->attr_gencount = fattr->gencount;
717 if ((attr->ia_valid & ATTR_SIZE) != 0) {
718 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
719 NFS_INO_INVALID_BLOCKS);
720 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
721 nfs_vmtruncate(inode, attr->ia_size);
723 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
724 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
725 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
726 inode->i_mode & S_ISUID)
727 inode->i_mode &= ~S_ISUID;
728 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 &&
729 (inode->i_mode & (S_ISGID | S_IXGRP)) ==
731 inode->i_mode &= ~S_ISGID;
732 if ((attr->ia_valid & ATTR_MODE) != 0) {
733 int mode = attr->ia_mode & S_IALLUGO;
734 mode |= inode->i_mode & ~S_IALLUGO;
735 inode->i_mode = mode;
737 if ((attr->ia_valid & ATTR_UID) != 0)
738 inode->i_uid = attr->ia_uid;
739 if ((attr->ia_valid & ATTR_GID) != 0)
740 inode->i_gid = attr->ia_gid;
741 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
742 inode->i_ctime = fattr->ctime;
744 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
745 | NFS_INO_INVALID_CTIME);
746 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
747 | NFS_INO_INVALID_ACL);
749 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
750 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
751 | NFS_INO_INVALID_CTIME);
752 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
753 inode->i_atime = fattr->atime;
754 else if (attr->ia_valid & ATTR_ATIME_SET)
755 inode->i_atime = attr->ia_atime;
757 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
759 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
760 inode->i_ctime = fattr->ctime;
762 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
763 | NFS_INO_INVALID_CTIME);
765 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
766 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
767 | NFS_INO_INVALID_CTIME);
768 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
769 inode->i_mtime = fattr->mtime;
770 else if (attr->ia_valid & ATTR_MTIME_SET)
771 inode->i_mtime = attr->ia_mtime;
773 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
775 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
776 inode->i_ctime = fattr->ctime;
778 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
779 | NFS_INO_INVALID_CTIME);
782 nfs_update_inode(inode, fattr);
783 spin_unlock(&inode->i_lock);
785 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
787 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
789 struct dentry *parent;
791 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
793 parent = dget_parent(dentry);
794 nfs_force_use_readdirplus(d_inode(parent));
798 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
800 struct dentry *parent;
802 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
804 parent = dget_parent(dentry);
805 nfs_advise_use_readdirplus(d_inode(parent));
809 static u32 nfs_get_valid_attrmask(struct inode *inode)
811 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
812 u32 reply_mask = STATX_INO | STATX_TYPE;
814 if (!(cache_validity & NFS_INO_INVALID_ATIME))
815 reply_mask |= STATX_ATIME;
816 if (!(cache_validity & NFS_INO_INVALID_CTIME))
817 reply_mask |= STATX_CTIME;
818 if (!(cache_validity & NFS_INO_INVALID_MTIME))
819 reply_mask |= STATX_MTIME;
820 if (!(cache_validity & NFS_INO_INVALID_SIZE))
821 reply_mask |= STATX_SIZE;
822 if (!(cache_validity & NFS_INO_INVALID_NLINK))
823 reply_mask |= STATX_NLINK;
824 if (!(cache_validity & NFS_INO_INVALID_MODE))
825 reply_mask |= STATX_MODE;
826 if (!(cache_validity & NFS_INO_INVALID_OTHER))
827 reply_mask |= STATX_UID | STATX_GID;
828 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
829 reply_mask |= STATX_BLOCKS;
833 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
834 struct kstat *stat, u32 request_mask, unsigned int query_flags)
836 struct inode *inode = d_inode(path->dentry);
837 struct nfs_server *server = NFS_SERVER(inode);
838 unsigned long cache_validity;
840 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
841 bool do_update = false;
843 trace_nfs_getattr_enter(inode);
845 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
846 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
847 STATX_INO | STATX_SIZE | STATX_BLOCKS;
849 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
850 nfs_readdirplus_parent_cache_hit(path->dentry);
851 goto out_no_revalidate;
854 /* Flush out writes to the server in order to update c/mtime. */
855 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
856 S_ISREG(inode->i_mode)) {
857 err = filemap_write_and_wait(inode->i_mapping);
863 * We may force a getattr if the user cares about atime.
865 * Note that we only have to check the vfsmount flags here:
866 * - NFS always sets S_NOATIME by so checking it would give a
868 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
869 * no point in checking those.
871 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
872 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
873 request_mask &= ~STATX_ATIME;
875 /* Is the user requesting attributes that might need revalidation? */
876 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
877 STATX_MTIME|STATX_UID|STATX_GID|
878 STATX_SIZE|STATX_BLOCKS)))
879 goto out_no_revalidate;
881 /* Check whether the cached attributes are stale */
882 do_update |= force_sync || nfs_attribute_cache_expired(inode);
883 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
884 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
885 if (request_mask & STATX_ATIME)
886 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
887 if (request_mask & STATX_CTIME)
888 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
889 if (request_mask & STATX_MTIME)
890 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
891 if (request_mask & STATX_SIZE)
892 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
893 if (request_mask & STATX_NLINK)
894 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
895 if (request_mask & STATX_MODE)
896 do_update |= cache_validity & NFS_INO_INVALID_MODE;
897 if (request_mask & (STATX_UID | STATX_GID))
898 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
899 if (request_mask & STATX_BLOCKS)
900 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
903 /* Update the attribute cache */
904 if (!(server->flags & NFS_MOUNT_NOAC))
905 nfs_readdirplus_parent_cache_miss(path->dentry);
907 nfs_readdirplus_parent_cache_hit(path->dentry);
908 err = __nfs_revalidate_inode(server, inode);
912 nfs_readdirplus_parent_cache_hit(path->dentry);
914 /* Only return attributes that were revalidated. */
915 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
917 generic_fillattr(&init_user_ns, inode, stat);
918 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
919 if (S_ISDIR(inode->i_mode))
920 stat->blksize = NFS_SERVER(inode)->dtsize;
922 trace_nfs_getattr_exit(inode, err);
925 EXPORT_SYMBOL_GPL(nfs_getattr);
927 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
929 refcount_set(&l_ctx->count, 1);
930 l_ctx->lockowner = current->files;
931 INIT_LIST_HEAD(&l_ctx->list);
932 atomic_set(&l_ctx->io_count, 0);
935 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
937 struct nfs_lock_context *pos;
939 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
940 if (pos->lockowner != current->files)
942 if (refcount_inc_not_zero(&pos->count))
948 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
950 struct nfs_lock_context *res, *new = NULL;
951 struct inode *inode = d_inode(ctx->dentry);
954 res = __nfs_find_lock_context(ctx);
957 new = kmalloc(sizeof(*new), GFP_KERNEL);
959 return ERR_PTR(-ENOMEM);
960 nfs_init_lock_context(new);
961 spin_lock(&inode->i_lock);
962 res = __nfs_find_lock_context(ctx);
964 new->open_context = get_nfs_open_context(ctx);
965 if (new->open_context) {
966 list_add_tail_rcu(&new->list,
967 &ctx->lock_context.list);
971 res = ERR_PTR(-EBADF);
973 spin_unlock(&inode->i_lock);
978 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
980 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
982 struct nfs_open_context *ctx = l_ctx->open_context;
983 struct inode *inode = d_inode(ctx->dentry);
985 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
987 list_del_rcu(&l_ctx->list);
988 spin_unlock(&inode->i_lock);
989 put_nfs_open_context(ctx);
990 kfree_rcu(l_ctx, rcu_head);
992 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
995 * nfs_close_context - Common close_context() routine NFSv2/v3
996 * @ctx: pointer to context
997 * @is_sync: is this a synchronous close
999 * Ensure that the attributes are up to date if we're mounted
1000 * with close-to-open semantics and we have cached data that will
1001 * need to be revalidated on open.
1003 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
1005 struct nfs_inode *nfsi;
1006 struct inode *inode;
1008 if (!(ctx->mode & FMODE_WRITE))
1012 inode = d_inode(ctx->dentry);
1013 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1015 nfsi = NFS_I(inode);
1016 if (inode->i_mapping->nrpages == 0)
1018 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1020 if (!list_empty(&nfsi->open_files))
1022 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1024 nfs_revalidate_inode(inode,
1025 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1027 EXPORT_SYMBOL_GPL(nfs_close_context);
1029 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1033 struct nfs_open_context *ctx;
1035 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1037 return ERR_PTR(-ENOMEM);
1038 nfs_sb_active(dentry->d_sb);
1039 ctx->dentry = dget(dentry);
1041 ctx->cred = get_cred(filp->f_cred);
1043 ctx->cred = get_current_cred();
1044 rcu_assign_pointer(ctx->ll_cred, NULL);
1049 ctx->flock_owner = (fl_owner_t)filp;
1050 nfs_init_lock_context(&ctx->lock_context);
1051 ctx->lock_context.open_context = ctx;
1052 INIT_LIST_HEAD(&ctx->list);
1053 ctx->mdsthreshold = NULL;
1056 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1058 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1060 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1064 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1066 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1068 struct inode *inode = d_inode(ctx->dentry);
1069 struct super_block *sb = ctx->dentry->d_sb;
1071 if (!refcount_dec_and_test(&ctx->lock_context.count))
1073 if (!list_empty(&ctx->list)) {
1074 spin_lock(&inode->i_lock);
1075 list_del_rcu(&ctx->list);
1076 spin_unlock(&inode->i_lock);
1079 NFS_PROTO(inode)->close_context(ctx, is_sync);
1080 put_cred(ctx->cred);
1082 nfs_sb_deactive(sb);
1083 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1084 kfree(ctx->mdsthreshold);
1085 kfree_rcu(ctx, rcu_head);
1088 void put_nfs_open_context(struct nfs_open_context *ctx)
1090 __put_nfs_open_context(ctx, 0);
1092 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1094 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1096 __put_nfs_open_context(ctx, 1);
1100 * Ensure that mmap has a recent RPC credential for use when writing out
1103 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1105 struct inode *inode = d_inode(ctx->dentry);
1106 struct nfs_inode *nfsi = NFS_I(inode);
1108 spin_lock(&inode->i_lock);
1109 if (list_empty(&nfsi->open_files) &&
1110 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1111 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1112 NFS_INO_REVAL_FORCED);
1113 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1114 spin_unlock(&inode->i_lock);
1116 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1118 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1120 filp->private_data = get_nfs_open_context(ctx);
1121 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1122 if (list_empty(&ctx->list))
1123 nfs_inode_attach_open_context(ctx);
1125 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1128 * Given an inode, search for an open context with the desired characteristics
1130 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1132 struct nfs_inode *nfsi = NFS_I(inode);
1133 struct nfs_open_context *pos, *ctx = NULL;
1136 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1137 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1139 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1141 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1143 ctx = get_nfs_open_context(pos);
1151 void nfs_file_clear_open_context(struct file *filp)
1153 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1156 struct inode *inode = d_inode(ctx->dentry);
1158 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1160 * We fatal error on write before. Try to writeback
1164 invalidate_inode_pages2(inode->i_mapping);
1165 filp->private_data = NULL;
1166 put_nfs_open_context_sync(ctx);
1171 * These allocate and release file read/write context information.
1173 int nfs_open(struct inode *inode, struct file *filp)
1175 struct nfs_open_context *ctx;
1177 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1179 return PTR_ERR(ctx);
1180 nfs_file_set_open_context(filp, ctx);
1181 put_nfs_open_context(ctx);
1182 nfs_fscache_open_file(inode, filp);
1185 EXPORT_SYMBOL_GPL(nfs_open);
1188 * This function is called whenever some part of NFS notices that
1189 * the cached attributes have to be refreshed.
1192 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1194 int status = -ESTALE;
1195 struct nfs_fattr *fattr = NULL;
1196 struct nfs_inode *nfsi = NFS_I(inode);
1198 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1199 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1201 trace_nfs_revalidate_inode_enter(inode);
1203 if (is_bad_inode(inode))
1205 if (NFS_STALE(inode))
1208 /* pNFS: Attributes aren't updated until we layoutcommit */
1209 if (S_ISREG(inode->i_mode)) {
1210 status = pnfs_sync_inode(inode, false);
1216 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1220 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1222 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1224 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1226 (unsigned long long)NFS_FILEID(inode), status);
1229 /* A soft timeout occurred. Use cached information? */
1230 if (server->flags & NFS_MOUNT_SOFTREVAL)
1234 if (!S_ISDIR(inode->i_mode))
1235 nfs_set_inode_stale(inode);
1237 nfs_zap_caches(inode);
1242 status = nfs_refresh_inode(inode, fattr);
1244 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1246 (unsigned long long)NFS_FILEID(inode), status);
1250 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1251 nfs_zap_acl_cache(inode);
1253 nfs_setsecurity(inode, fattr);
1255 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1257 (unsigned long long)NFS_FILEID(inode));
1260 nfs_free_fattr(fattr);
1261 trace_nfs_revalidate_inode_exit(inode, status);
1265 int nfs_attribute_cache_expired(struct inode *inode)
1267 if (nfs_have_delegated_attributes(inode))
1269 return nfs_attribute_timeout(inode);
1273 * nfs_revalidate_inode - Revalidate the inode attributes
1274 * @inode: pointer to inode struct
1275 * @flags: cache flags to check
1277 * Updates inode attribute information by retrieving the data from the server.
1279 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1281 if (!nfs_check_cache_invalid(inode, flags))
1282 return NFS_STALE(inode) ? -ESTALE : 0;
1283 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1285 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1287 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1291 if (mapping->nrpages != 0) {
1292 if (S_ISREG(inode->i_mode)) {
1293 ret = nfs_sync_mapping(mapping);
1297 ret = invalidate_inode_pages2(mapping);
1301 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1302 nfs_fscache_wait_on_invalidate(inode);
1304 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1306 (unsigned long long)NFS_FILEID(inode));
1311 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1312 * @mapping: pointer to mapping
1314 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1316 int nfs_clear_invalid_mapping(struct address_space *mapping)
1318 struct inode *inode = mapping->host;
1319 struct nfs_inode *nfsi = NFS_I(inode);
1320 unsigned long *bitlock = &nfsi->flags;
1324 * We must clear NFS_INO_INVALID_DATA first to ensure that
1325 * invalidations that come in while we're shooting down the mappings
1326 * are respected. But, that leaves a race window where one revalidator
1327 * can clear the flag, and then another checks it before the mapping
1328 * gets invalidated. Fix that by serializing access to this part of
1331 * At the same time, we need to allow other tasks to see whether we
1332 * might be in the middle of invalidating the pages, so we only set
1333 * the bit lock here if it looks like we're going to be doing that.
1336 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1337 nfs_wait_bit_killable, TASK_KILLABLE);
1340 spin_lock(&inode->i_lock);
1341 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1342 spin_unlock(&inode->i_lock);
1345 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1347 spin_unlock(&inode->i_lock);
1351 set_bit(NFS_INO_INVALIDATING, bitlock);
1353 nfsi->cache_validity &=
1354 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1355 spin_unlock(&inode->i_lock);
1356 trace_nfs_invalidate_mapping_enter(inode);
1357 ret = nfs_invalidate_mapping(inode, mapping);
1358 trace_nfs_invalidate_mapping_exit(inode, ret);
1360 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1361 smp_mb__after_atomic();
1362 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1367 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1369 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1373 int nfs_revalidate_mapping_rcu(struct inode *inode)
1375 struct nfs_inode *nfsi = NFS_I(inode);
1376 unsigned long *bitlock = &nfsi->flags;
1379 if (IS_SWAPFILE(inode))
1381 if (nfs_mapping_need_revalidate_inode(inode)) {
1385 spin_lock(&inode->i_lock);
1386 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1387 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1389 spin_unlock(&inode->i_lock);
1395 * nfs_revalidate_mapping - Revalidate the pagecache
1396 * @inode: pointer to host inode
1397 * @mapping: pointer to mapping
1399 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1401 /* swapfiles are not supposed to be shared. */
1402 if (IS_SWAPFILE(inode))
1405 if (nfs_mapping_need_revalidate_inode(inode)) {
1406 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1411 return nfs_clear_invalid_mapping(mapping);
1414 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1416 struct inode *inode = &nfsi->vfs_inode;
1418 if (!S_ISREG(inode->i_mode))
1420 if (list_empty(&nfsi->open_files))
1422 return inode_is_open_for_write(inode);
1425 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1427 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1430 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1432 struct timespec64 ts;
1434 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1435 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1436 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1437 inode_set_iversion_raw(inode, fattr->change_attr);
1438 if (S_ISDIR(inode->i_mode))
1439 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1440 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1441 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1443 /* If we have atomic WCC data, we may update some attributes */
1444 ts = inode->i_ctime;
1445 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1446 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1447 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1448 inode->i_ctime = fattr->ctime;
1451 ts = inode->i_mtime;
1452 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1453 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1454 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1455 inode->i_mtime = fattr->mtime;
1457 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1458 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1459 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1460 && !nfs_have_writebacks(inode)) {
1461 trace_nfs_size_wcc(inode, fattr->size);
1462 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1467 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1468 * @inode: pointer to inode
1469 * @fattr: updated attributes
1471 * Verifies the attribute cache. If we have just changed the attributes,
1472 * so that fattr carries weak cache consistency data, then it may
1473 * also update the ctime/mtime/change_attribute.
1475 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1477 struct nfs_inode *nfsi = NFS_I(inode);
1478 loff_t cur_size, new_isize;
1479 unsigned long invalid = 0;
1480 struct timespec64 ts;
1482 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1485 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1486 /* Only a mounted-on-fileid? Just exit */
1487 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1489 /* Has the inode gone and changed behind our back? */
1490 } else if (nfsi->fileid != fattr->fileid) {
1491 /* Is this perhaps the mounted-on fileid? */
1492 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1493 nfsi->fileid == fattr->mounted_on_fileid)
1497 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1501 if (!nfs_file_has_buffered_writers(nfsi)) {
1502 /* Verify a few of the more important attributes */
1503 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1504 invalid |= NFS_INO_INVALID_CHANGE;
1506 ts = inode->i_mtime;
1507 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1508 invalid |= NFS_INO_INVALID_MTIME;
1510 ts = inode->i_ctime;
1511 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1512 invalid |= NFS_INO_INVALID_CTIME;
1514 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1515 cur_size = i_size_read(inode);
1516 new_isize = nfs_size_to_loff_t(fattr->size);
1517 if (cur_size != new_isize)
1518 invalid |= NFS_INO_INVALID_SIZE;
1522 /* Have any file permissions changed? */
1523 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1524 invalid |= NFS_INO_INVALID_MODE;
1525 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1526 invalid |= NFS_INO_INVALID_OTHER;
1527 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1528 invalid |= NFS_INO_INVALID_OTHER;
1530 /* Has the link count changed? */
1531 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1532 invalid |= NFS_INO_INVALID_NLINK;
1534 ts = inode->i_atime;
1535 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1536 invalid |= NFS_INO_INVALID_ATIME;
1539 nfs_set_cache_invalid(inode, invalid);
1541 nfsi->read_cache_jiffies = fattr->time_start;
1545 static atomic_long_t nfs_attr_generation_counter;
1547 static unsigned long nfs_read_attr_generation_counter(void)
1549 return atomic_long_read(&nfs_attr_generation_counter);
1552 unsigned long nfs_inc_attr_generation_counter(void)
1554 return atomic_long_inc_return(&nfs_attr_generation_counter);
1556 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1558 void nfs_fattr_init(struct nfs_fattr *fattr)
1561 fattr->time_start = jiffies;
1562 fattr->gencount = nfs_inc_attr_generation_counter();
1563 fattr->owner_name = NULL;
1564 fattr->group_name = NULL;
1566 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1569 * nfs_fattr_set_barrier
1570 * @fattr: attributes
1572 * Used to set a barrier after an attribute was updated. This
1573 * barrier ensures that older attributes from RPC calls that may
1574 * have raced with our update cannot clobber these new values.
1575 * Note that you are still responsible for ensuring that other
1576 * operations which change the attribute on the server do not
1579 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1581 fattr->gencount = nfs_inc_attr_generation_counter();
1584 struct nfs_fattr *nfs_alloc_fattr(void)
1586 struct nfs_fattr *fattr;
1588 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1589 if (fattr != NULL) {
1590 nfs_fattr_init(fattr);
1591 fattr->label = NULL;
1595 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1597 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1599 struct nfs_fattr *fattr = nfs_alloc_fattr();
1604 fattr->label = nfs4_label_alloc(server, GFP_NOFS);
1605 if (IS_ERR(fattr->label)) {
1612 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1614 struct nfs_fh *nfs_alloc_fhandle(void)
1618 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1623 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1627 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1628 * in the same way that wireshark does
1632 * For debugging only.
1634 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1636 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1637 * not on the result */
1638 return nfs_fhandle_hash(fh);
1640 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1643 * _nfs_display_fhandle - display an NFS file handle on the console
1645 * @fh: file handle to display
1646 * @caption: display caption
1648 * For debugging only.
1650 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1654 if (fh == NULL || fh->size == 0) {
1655 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1659 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1660 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1661 for (i = 0; i < fh->size; i += 16) {
1662 __be32 *pos = (__be32 *)&fh->data[i];
1664 switch ((fh->size - i - 1) >> 2) {
1666 printk(KERN_DEFAULT " %08x\n",
1670 printk(KERN_DEFAULT " %08x %08x\n",
1671 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1674 printk(KERN_DEFAULT " %08x %08x %08x\n",
1675 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1676 be32_to_cpup(pos + 2));
1679 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1680 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1681 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1685 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1689 * nfs_inode_attrs_cmp_generic - compare attributes
1690 * @fattr: attributes
1691 * @inode: pointer to inode
1693 * Attempt to divine whether or not an RPC call reply carrying stale
1694 * attributes got scheduled after another call carrying updated ones.
1695 * Note also the check for wraparound of 'attr_gencount'
1697 * The function returns '1' if it thinks the attributes in @fattr are
1698 * more recent than the ones cached in @inode. Otherwise it returns
1701 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1702 const struct inode *inode)
1704 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1706 return (long)(fattr->gencount - attr_gencount) > 0 ||
1707 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1711 * nfs_inode_attrs_cmp_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 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 '0' indicates no measurable change
1723 * A return value of '-1' means that the attributes in @inode are
1726 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1727 const struct inode *inode)
1729 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1732 return diff == 0 ? 0 : -1;
1736 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1737 * @fattr: attributes
1738 * @inode: pointer to inode
1740 * Attempt to divine whether or not an RPC call reply carrying stale
1741 * attributes got scheduled after another call carrying updated ones.
1743 * We assume that the server observes strictly monotonic semantics for
1744 * the change attribute, so a larger value means that the attributes in
1745 * @fattr are more recent, in which case the function returns the
1747 * A return value of '-1' means that the attributes in @inode are
1748 * more recent or unchanged.
1750 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1751 const struct inode *inode)
1753 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1757 * nfs_inode_attrs_cmp - compare attributes
1758 * @fattr: attributes
1759 * @inode: pointer to inode
1761 * This function returns '1' if it thinks the attributes in @fattr are
1762 * more recent than the ones cached in @inode. It returns '-1' if
1763 * the attributes in @inode are more recent than the ones in @fattr,
1764 * and it returns 0 if not sure.
1766 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1767 const struct inode *inode)
1769 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1771 switch (NFS_SERVER(inode)->change_attr_type) {
1772 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1774 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1775 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1777 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1779 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1781 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1787 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1788 * @fattr: attributes
1789 * @inode: pointer to inode
1791 * Returns '1' if the last attribute update left the inode cached
1792 * attributes in a partially unrevalidated state, and @fattr
1793 * matches the change attribute of that partial update.
1794 * Otherwise returns '0'.
1796 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1797 const struct inode *inode)
1799 const unsigned long check_valid =
1800 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1801 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1802 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1803 NFS_INO_INVALID_NLINK;
1804 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1805 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1807 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1808 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1809 (cache_validity & check_valid) != 0 &&
1810 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1811 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1816 static int nfs_refresh_inode_locked(struct inode *inode,
1817 struct nfs_fattr *fattr)
1819 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1822 trace_nfs_refresh_inode_enter(inode);
1824 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1825 ret = nfs_update_inode(inode, fattr);
1826 else if (attr_cmp == 0)
1827 ret = nfs_check_inode_attributes(inode, fattr);
1829 trace_nfs_refresh_inode_exit(inode, ret);
1834 * nfs_refresh_inode - try to update the inode attribute cache
1835 * @inode: pointer to inode
1836 * @fattr: updated attributes
1838 * Check that an RPC call that returned attributes has not overlapped with
1839 * other recent updates of the inode metadata, then decide whether it is
1840 * safe to do a full update of the inode attributes, or whether just to
1841 * call nfs_check_inode_attributes.
1843 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1847 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1849 spin_lock(&inode->i_lock);
1850 status = nfs_refresh_inode_locked(inode, fattr);
1851 spin_unlock(&inode->i_lock);
1855 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1857 static int nfs_post_op_update_inode_locked(struct inode *inode,
1858 struct nfs_fattr *fattr, unsigned int invalid)
1860 if (S_ISDIR(inode->i_mode))
1861 invalid |= NFS_INO_INVALID_DATA;
1862 nfs_set_cache_invalid(inode, invalid);
1863 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1865 return nfs_refresh_inode_locked(inode, fattr);
1869 * nfs_post_op_update_inode - try to update the inode attribute cache
1870 * @inode: pointer to inode
1871 * @fattr: updated attributes
1873 * After an operation that has changed the inode metadata, mark the
1874 * attribute cache as being invalid, then try to update it.
1876 * NB: if the server didn't return any post op attributes, this
1877 * function will force the retrieval of attributes before the next
1878 * NFS request. Thus it should be used only for operations that
1879 * are expected to change one or more attributes, to avoid
1880 * unnecessary NFS requests and trips through nfs_update_inode().
1882 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1886 spin_lock(&inode->i_lock);
1887 nfs_fattr_set_barrier(fattr);
1888 status = nfs_post_op_update_inode_locked(inode, fattr,
1889 NFS_INO_INVALID_CHANGE
1890 | NFS_INO_INVALID_CTIME
1891 | NFS_INO_REVAL_FORCED);
1892 spin_unlock(&inode->i_lock);
1896 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1899 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1900 * @inode: pointer to inode
1901 * @fattr: updated attributes
1903 * After an operation that has changed the inode metadata, mark the
1904 * attribute cache as being invalid, then try to update it. Fake up
1905 * weak cache consistency data, if none exist.
1907 * This function is mainly designed to be used by the ->write_done() functions.
1909 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1911 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1914 /* Don't do a WCC update if these attributes are already stale */
1917 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1918 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1919 | NFS_ATTR_FATTR_PRESIZE
1920 | NFS_ATTR_FATTR_PREMTIME
1921 | NFS_ATTR_FATTR_PRECTIME);
1924 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1925 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1926 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1927 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1929 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1930 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1931 fattr->pre_ctime = inode->i_ctime;
1932 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1934 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1935 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1936 fattr->pre_mtime = inode->i_mtime;
1937 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1939 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1940 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1941 fattr->pre_size = i_size_read(inode);
1942 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1945 status = nfs_post_op_update_inode_locked(inode, fattr,
1946 NFS_INO_INVALID_CHANGE
1947 | NFS_INO_INVALID_CTIME
1948 | NFS_INO_INVALID_MTIME
1949 | NFS_INO_INVALID_BLOCKS);
1954 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1955 * @inode: pointer to inode
1956 * @fattr: updated attributes
1958 * After an operation that has changed the inode metadata, mark the
1959 * attribute cache as being invalid, then try to update it. Fake up
1960 * weak cache consistency data, if none exist.
1962 * This function is mainly designed to be used by the ->write_done() functions.
1964 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1968 spin_lock(&inode->i_lock);
1969 nfs_fattr_set_barrier(fattr);
1970 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1971 spin_unlock(&inode->i_lock);
1974 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1978 * Many nfs protocol calls return the new file attributes after
1979 * an operation. Here we update the inode to reflect the state
1980 * of the server's inode.
1982 * This is a bit tricky because we have to make sure all dirty pages
1983 * have been sent off to the server before calling invalidate_inode_pages.
1984 * To make sure no other process adds more write requests while we try
1985 * our best to flush them, we make them sleep during the attribute refresh.
1987 * A very similar scenario holds for the dir cache.
1989 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1991 struct nfs_server *server = NFS_SERVER(inode);
1992 struct nfs_inode *nfsi = NFS_I(inode);
1993 loff_t cur_isize, new_isize;
1994 u64 fattr_supported = server->fattr_valid;
1995 unsigned long invalid = 0;
1996 unsigned long now = jiffies;
1997 unsigned long save_cache_validity;
1998 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1999 bool cache_revalidated = true;
2000 bool attr_changed = false;
2001 bool have_delegation;
2003 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
2004 __func__, inode->i_sb->s_id, inode->i_ino,
2005 nfs_display_fhandle_hash(NFS_FH(inode)),
2006 atomic_read(&inode->i_count), fattr->valid);
2008 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2009 /* Only a mounted-on-fileid? Just exit */
2010 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2012 /* Has the inode gone and changed behind our back? */
2013 } else if (nfsi->fileid != fattr->fileid) {
2014 /* Is this perhaps the mounted-on fileid? */
2015 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2016 nfsi->fileid == fattr->mounted_on_fileid)
2018 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2019 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2020 NFS_SERVER(inode)->nfs_client->cl_hostname,
2021 inode->i_sb->s_id, (long long)nfsi->fileid,
2022 (long long)fattr->fileid);
2027 * Make sure the inode's type hasn't changed.
2029 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2031 * Big trouble! The inode has become a different object.
2033 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2034 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2038 /* Update the fsid? */
2039 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2040 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2041 !IS_AUTOMOUNT(inode))
2042 server->fsid = fattr->fsid;
2044 /* Save the delegation state before clearing cache_validity */
2045 have_delegation = nfs_have_delegated_attributes(inode);
2048 * Update the read time so we don't revalidate too often.
2050 nfsi->read_cache_jiffies = fattr->time_start;
2052 save_cache_validity = nfsi->cache_validity;
2053 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2054 | NFS_INO_INVALID_ATIME
2055 | NFS_INO_REVAL_FORCED
2056 | NFS_INO_INVALID_BLOCKS);
2058 /* Do atomic weak cache consistency updates */
2059 nfs_wcc_update_inode(inode, fattr);
2061 if (pnfs_layoutcommit_outstanding(inode)) {
2062 nfsi->cache_validity |=
2063 save_cache_validity &
2064 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2065 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2066 NFS_INO_INVALID_BLOCKS);
2067 cache_revalidated = false;
2070 /* More cache consistency checks */
2071 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2072 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2073 /* Could it be a race with writeback? */
2074 if (!(have_writers || have_delegation)) {
2075 invalid |= NFS_INO_INVALID_DATA
2076 | NFS_INO_INVALID_ACCESS
2077 | NFS_INO_INVALID_ACL
2078 | NFS_INO_INVALID_XATTR;
2079 /* Force revalidate of all attributes */
2080 save_cache_validity |= NFS_INO_INVALID_CTIME
2081 | NFS_INO_INVALID_MTIME
2082 | NFS_INO_INVALID_SIZE
2083 | NFS_INO_INVALID_BLOCKS
2084 | NFS_INO_INVALID_NLINK
2085 | NFS_INO_INVALID_MODE
2086 | NFS_INO_INVALID_OTHER;
2087 if (S_ISDIR(inode->i_mode))
2088 nfs_force_lookup_revalidate(inode);
2089 attr_changed = true;
2090 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2093 } else if (!have_delegation)
2094 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2095 inode_set_iversion_raw(inode, fattr->change_attr);
2098 nfsi->cache_validity |=
2099 save_cache_validity & NFS_INO_INVALID_CHANGE;
2100 if (!have_delegation ||
2101 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2102 cache_revalidated = false;
2105 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2106 inode->i_mtime = fattr->mtime;
2107 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2108 nfsi->cache_validity |=
2109 save_cache_validity & NFS_INO_INVALID_MTIME;
2111 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2112 inode->i_ctime = fattr->ctime;
2113 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2114 nfsi->cache_validity |=
2115 save_cache_validity & NFS_INO_INVALID_CTIME;
2117 /* Check if our cached file size is stale */
2118 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2119 new_isize = nfs_size_to_loff_t(fattr->size);
2120 cur_isize = i_size_read(inode);
2121 if (new_isize != cur_isize && !have_delegation) {
2122 /* Do we perhaps have any outstanding writes, or has
2123 * the file grown beyond our last write? */
2124 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2125 trace_nfs_size_update(inode, new_isize);
2126 i_size_write(inode, new_isize);
2128 invalid |= NFS_INO_INVALID_DATA;
2131 if (new_isize == 0 &&
2132 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2133 NFS_ATTR_FATTR_BLOCKS_USED))) {
2134 fattr->du.nfs3.used = 0;
2135 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2138 nfsi->cache_validity |=
2139 save_cache_validity & NFS_INO_INVALID_SIZE;
2141 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2142 inode->i_atime = fattr->atime;
2143 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2144 nfsi->cache_validity |=
2145 save_cache_validity & NFS_INO_INVALID_ATIME;
2147 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2148 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2149 umode_t newmode = inode->i_mode & S_IFMT;
2150 newmode |= fattr->mode & S_IALLUGO;
2151 inode->i_mode = newmode;
2152 invalid |= NFS_INO_INVALID_ACCESS
2153 | NFS_INO_INVALID_ACL;
2155 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2156 nfsi->cache_validity |=
2157 save_cache_validity & NFS_INO_INVALID_MODE;
2159 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2160 if (!uid_eq(inode->i_uid, fattr->uid)) {
2161 invalid |= NFS_INO_INVALID_ACCESS
2162 | NFS_INO_INVALID_ACL;
2163 inode->i_uid = fattr->uid;
2165 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2166 nfsi->cache_validity |=
2167 save_cache_validity & NFS_INO_INVALID_OTHER;
2169 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2170 if (!gid_eq(inode->i_gid, fattr->gid)) {
2171 invalid |= NFS_INO_INVALID_ACCESS
2172 | NFS_INO_INVALID_ACL;
2173 inode->i_gid = fattr->gid;
2175 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2176 nfsi->cache_validity |=
2177 save_cache_validity & NFS_INO_INVALID_OTHER;
2179 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2180 if (inode->i_nlink != fattr->nlink)
2181 set_nlink(inode, fattr->nlink);
2182 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2183 nfsi->cache_validity |=
2184 save_cache_validity & NFS_INO_INVALID_NLINK;
2186 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2188 * report the blocks in 512byte units
2190 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2191 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2192 nfsi->cache_validity |=
2193 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2195 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2196 inode->i_blocks = fattr->du.nfs2.blocks;
2197 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2198 nfsi->cache_validity |=
2199 save_cache_validity & NFS_INO_INVALID_BLOCKS;
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 nfs4_init_once(nfsi);
2285 static int __init nfs_init_inodecache(void)
2287 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2288 sizeof(struct nfs_inode),
2289 0, (SLAB_RECLAIM_ACCOUNT|
2290 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2292 if (nfs_inode_cachep == NULL)
2298 static void nfs_destroy_inodecache(void)
2301 * Make sure all delayed rcu free inodes are flushed before we
2305 kmem_cache_destroy(nfs_inode_cachep);
2308 struct workqueue_struct *nfsiod_workqueue;
2309 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2312 * start up the nfsiod workqueue
2314 static int nfsiod_start(void)
2316 struct workqueue_struct *wq;
2317 dprintk("RPC: creating workqueue nfsiod\n");
2318 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2321 nfsiod_workqueue = wq;
2326 * Destroy the nfsiod workqueue
2328 static void nfsiod_stop(void)
2330 struct workqueue_struct *wq;
2332 wq = nfsiod_workqueue;
2335 nfsiod_workqueue = NULL;
2336 destroy_workqueue(wq);
2339 unsigned int nfs_net_id;
2340 EXPORT_SYMBOL_GPL(nfs_net_id);
2342 static int nfs_net_init(struct net *net)
2344 nfs_clients_init(net);
2345 return nfs_fs_proc_net_init(net);
2348 static void nfs_net_exit(struct net *net)
2350 nfs_fs_proc_net_exit(net);
2351 nfs_clients_exit(net);
2354 static struct pernet_operations nfs_net_ops = {
2355 .init = nfs_net_init,
2356 .exit = nfs_net_exit,
2358 .size = sizeof(struct nfs_net),
2364 static int __init init_nfs_fs(void)
2368 err = nfs_sysfs_init();
2372 err = register_pernet_subsys(&nfs_net_ops);
2376 err = nfs_fscache_register();
2380 err = nfsiod_start();
2384 err = nfs_fs_proc_init();
2388 err = nfs_init_nfspagecache();
2392 err = nfs_init_inodecache();
2396 err = nfs_init_readpagecache();
2400 err = nfs_init_writepagecache();
2404 err = nfs_init_directcache();
2408 rpc_proc_register(&init_net, &nfs_rpcstat);
2410 err = register_nfs_fs();
2416 rpc_proc_unregister(&init_net, "nfs");
2417 nfs_destroy_directcache();
2419 nfs_destroy_writepagecache();
2421 nfs_destroy_readpagecache();
2423 nfs_destroy_inodecache();
2425 nfs_destroy_nfspagecache();
2431 nfs_fscache_unregister();
2433 unregister_pernet_subsys(&nfs_net_ops);
2440 static void __exit exit_nfs_fs(void)
2442 nfs_destroy_directcache();
2443 nfs_destroy_writepagecache();
2444 nfs_destroy_readpagecache();
2445 nfs_destroy_inodecache();
2446 nfs_destroy_nfspagecache();
2447 nfs_fscache_unregister();
2448 unregister_pernet_subsys(&nfs_net_ops);
2449 rpc_proc_unregister(&init_net, "nfs");
2450 unregister_nfs_fs();
2456 /* Not quite true; I just maintain it */
2458 MODULE_LICENSE("GPL");
2459 module_param(enable_ino64, bool, 0644);
2461 module_init(init_nfs_fs)
2462 module_exit(exit_nfs_fs)