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 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
78 if (signal_pending_state(mode, current))
82 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
85 * nfs_compat_user_ino64 - returns the user-visible inode number
86 * @fileid: 64-bit fileid
88 * This function returns a 32-bit inode number if the boot parameter
89 * nfs.enable_ino64 is zero.
91 u64 nfs_compat_user_ino64(u64 fileid)
102 if (sizeof(ino) < sizeof(fileid))
103 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
107 int nfs_drop_inode(struct inode *inode)
109 return NFS_STALE(inode) || generic_drop_inode(inode);
111 EXPORT_SYMBOL_GPL(nfs_drop_inode);
113 void nfs_clear_inode(struct inode *inode)
116 * The following should never happen...
118 WARN_ON_ONCE(nfs_have_writebacks(inode));
119 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
120 nfs_zap_acl_cache(inode);
121 nfs_access_zap_cache(inode);
122 nfs_fscache_clear_inode(inode);
124 EXPORT_SYMBOL_GPL(nfs_clear_inode);
126 void nfs_evict_inode(struct inode *inode)
128 truncate_inode_pages_final(&inode->i_data);
130 nfs_clear_inode(inode);
133 int nfs_sync_inode(struct inode *inode)
135 inode_dio_wait(inode);
136 return nfs_wb_all(inode);
138 EXPORT_SYMBOL_GPL(nfs_sync_inode);
141 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
142 * @mapping: pointer to struct address_space
144 int nfs_sync_mapping(struct address_space *mapping)
148 if (mapping->nrpages != 0) {
149 unmap_mapping_range(mapping, 0, 0, 0);
150 ret = nfs_wb_all(mapping->host);
155 static int nfs_attribute_timeout(struct inode *inode)
157 struct nfs_inode *nfsi = NFS_I(inode);
159 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
162 static bool nfs_check_cache_flags_invalid(struct inode *inode,
165 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
167 return (cache_validity & flags) != 0;
170 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
172 if (nfs_check_cache_flags_invalid(inode, flags))
174 return nfs_attribute_cache_expired(inode);
176 EXPORT_SYMBOL_GPL(nfs_check_cache_invalid);
178 #ifdef CONFIG_NFS_V4_2
179 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
181 return nfsi->xattr_cache != NULL;
184 static bool nfs_has_xattr_cache(const struct nfs_inode *nfsi)
190 void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
192 struct nfs_inode *nfsi = NFS_I(inode);
193 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
195 if (have_delegation) {
196 if (!(flags & NFS_INO_REVAL_FORCED))
197 flags &= ~(NFS_INO_INVALID_MODE |
198 NFS_INO_INVALID_OTHER |
199 NFS_INO_INVALID_XATTR);
200 flags &= ~(NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
203 if (!nfs_has_xattr_cache(nfsi))
204 flags &= ~NFS_INO_INVALID_XATTR;
205 if (flags & NFS_INO_INVALID_DATA)
206 nfs_fscache_invalidate(inode, 0);
207 flags &= ~NFS_INO_REVAL_FORCED;
209 nfsi->cache_validity |= flags;
211 if (inode->i_mapping->nrpages == 0)
212 nfsi->cache_validity &= ~(NFS_INO_INVALID_DATA |
213 NFS_INO_DATA_INVAL_DEFER);
214 else if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
215 nfsi->cache_validity &= ~NFS_INO_DATA_INVAL_DEFER;
216 trace_nfs_set_cache_invalid(inode, 0);
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);
240 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR |
241 NFS_INO_INVALID_ACCESS |
242 NFS_INO_INVALID_ACL |
243 NFS_INO_INVALID_XATTR);
244 nfs_zap_label_cache_locked(nfsi);
247 void nfs_zap_caches(struct inode *inode)
249 spin_lock(&inode->i_lock);
250 nfs_zap_caches_locked(inode);
251 spin_unlock(&inode->i_lock);
254 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
256 if (mapping->nrpages != 0) {
257 spin_lock(&inode->i_lock);
258 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
259 spin_unlock(&inode->i_lock);
263 void nfs_zap_acl_cache(struct inode *inode)
265 void (*clear_acl_cache)(struct inode *);
267 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
268 if (clear_acl_cache != NULL)
269 clear_acl_cache(inode);
270 spin_lock(&inode->i_lock);
271 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
272 spin_unlock(&inode->i_lock);
274 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
276 void nfs_invalidate_atime(struct inode *inode)
278 spin_lock(&inode->i_lock);
279 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
280 spin_unlock(&inode->i_lock);
282 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
285 * Invalidate, but do not unhash, the inode.
286 * NB: must be called with inode->i_lock held!
288 static void nfs_set_inode_stale_locked(struct inode *inode)
290 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
291 nfs_zap_caches_locked(inode);
292 trace_nfs_set_inode_stale(inode);
295 void nfs_set_inode_stale(struct inode *inode)
297 spin_lock(&inode->i_lock);
298 nfs_set_inode_stale_locked(inode);
299 spin_unlock(&inode->i_lock);
302 struct nfs_find_desc {
304 struct nfs_fattr *fattr;
308 * In NFSv3 we can have 64bit inode numbers. In order to support
309 * this, and re-exported directories (also seen in NFSv2)
310 * we are forced to allow 2 different inodes to have the same
314 nfs_find_actor(struct inode *inode, void *opaque)
316 struct nfs_find_desc *desc = opaque;
317 struct nfs_fh *fh = desc->fh;
318 struct nfs_fattr *fattr = desc->fattr;
320 if (NFS_FILEID(inode) != fattr->fileid)
322 if (inode_wrong_type(inode, fattr->mode))
324 if (nfs_compare_fh(NFS_FH(inode), fh))
326 if (is_bad_inode(inode) || NFS_STALE(inode))
332 nfs_init_locked(struct inode *inode, void *opaque)
334 struct nfs_find_desc *desc = opaque;
335 struct nfs_fattr *fattr = desc->fattr;
337 set_nfs_fileid(inode, fattr->fileid);
338 inode->i_mode = fattr->mode;
339 nfs_copy_fh(NFS_FH(inode), desc->fh);
343 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
344 static void nfs_clear_label_invalid(struct inode *inode)
346 spin_lock(&inode->i_lock);
347 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
348 spin_unlock(&inode->i_lock);
351 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
355 if (fattr->label == NULL)
358 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
359 error = security_inode_notifysecctx(inode, fattr->label->label,
362 printk(KERN_ERR "%s() %s %d "
363 "security_inode_notifysecctx() %d\n",
365 (char *)fattr->label->label,
366 fattr->label->len, error);
367 nfs_clear_label_invalid(inode);
371 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
373 struct nfs4_label *label;
375 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
378 label = kzalloc(sizeof(struct nfs4_label), flags);
380 return ERR_PTR(-ENOMEM);
382 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
383 if (label->label == NULL) {
385 return ERR_PTR(-ENOMEM);
387 label->len = NFS4_MAXLABELLEN;
391 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
393 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr)
397 EXPORT_SYMBOL_GPL(nfs_setsecurity);
399 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
401 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
403 struct nfs_find_desc desc = {
410 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
411 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
414 hash = nfs_fattr_to_ino_t(fattr);
415 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
417 dprintk("%s: returning %p\n", __func__, inode);
421 static void nfs_inode_init_regular(struct nfs_inode *nfsi)
423 atomic_long_set(&nfsi->nrequests, 0);
424 atomic_long_set(&nfsi->redirtied_pages, 0);
425 INIT_LIST_HEAD(&nfsi->commit_info.list);
426 atomic_long_set(&nfsi->commit_info.ncommit, 0);
427 atomic_set(&nfsi->commit_info.rpcs_out, 0);
428 mutex_init(&nfsi->commit_mutex);
431 static void nfs_inode_init_dir(struct nfs_inode *nfsi)
433 nfsi->cache_change_attribute = 0;
434 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
435 init_rwsem(&nfsi->rmdir_sem);
439 * This is our front-end to iget that looks up inodes by file handle
440 * instead of inode number.
443 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
445 struct nfs_find_desc desc = {
449 struct inode *inode = ERR_PTR(-ENOENT);
450 u64 fattr_supported = NFS_SB(sb)->fattr_valid;
453 nfs_attr_check_mountpoint(sb, fattr);
455 if (nfs_attr_use_mounted_on_fileid(fattr))
456 fattr->fileid = fattr->mounted_on_fileid;
457 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
459 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
462 hash = nfs_fattr_to_ino_t(fattr);
464 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
466 inode = ERR_PTR(-ENOMEM);
470 if (inode->i_state & I_NEW) {
471 struct nfs_inode *nfsi = NFS_I(inode);
472 unsigned long now = jiffies;
474 /* We set i_ino for the few things that still rely on it,
478 /* We can't support update_atime(), since the server will reset it */
479 inode->i_flags |= S_NOATIME|S_NOCMTIME;
480 inode->i_mode = fattr->mode;
481 nfsi->cache_validity = 0;
482 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
483 && (fattr_supported & NFS_ATTR_FATTR_MODE))
484 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MODE);
485 /* Why so? Because we want revalidate for devices/FIFOs, and
486 * that's precisely what we have in nfs_file_inode_operations.
488 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
489 if (S_ISREG(inode->i_mode)) {
490 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
491 inode->i_data.a_ops = &nfs_file_aops;
492 nfs_inode_init_regular(nfsi);
493 } else if (S_ISDIR(inode->i_mode)) {
494 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
495 inode->i_fop = &nfs_dir_operations;
496 inode->i_data.a_ops = &nfs_dir_aops;
497 nfs_inode_init_dir(nfsi);
498 /* Deal with crossing mountpoints */
499 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
500 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
501 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
502 inode->i_op = &nfs_referral_inode_operations;
504 inode->i_op = &nfs_mountpoint_inode_operations;
506 inode->i_flags |= S_AUTOMOUNT;
508 } else if (S_ISLNK(inode->i_mode)) {
509 inode->i_op = &nfs_symlink_inode_operations;
510 inode_nohighmem(inode);
512 init_special_inode(inode, inode->i_mode, fattr->rdev);
514 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
515 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
516 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
517 inode_set_iversion_raw(inode, 0);
520 inode->i_uid = make_kuid(&init_user_ns, -2);
521 inode->i_gid = make_kgid(&init_user_ns, -2);
526 nfsi->read_cache_jiffies = fattr->time_start;
527 nfsi->attr_gencount = fattr->gencount;
528 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
529 inode->i_atime = fattr->atime;
530 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
531 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
532 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
533 inode->i_mtime = fattr->mtime;
534 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
535 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
536 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
537 inode->i_ctime = fattr->ctime;
538 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
539 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
540 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
541 inode_set_iversion_raw(inode, fattr->change_attr);
543 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
544 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
545 inode->i_size = nfs_size_to_loff_t(fattr->size);
547 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
548 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
549 set_nlink(inode, fattr->nlink);
550 else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
551 nfs_set_cache_invalid(inode, NFS_INO_INVALID_NLINK);
552 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
553 inode->i_uid = fattr->uid;
554 else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
555 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
556 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
557 inode->i_gid = fattr->gid;
558 else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
559 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
560 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
561 inode->i_blocks = fattr->du.nfs2.blocks;
562 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED &&
564 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
565 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
567 * report the blocks in 512byte units
569 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
570 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED &&
572 nfs_set_cache_invalid(inode, NFS_INO_INVALID_BLOCKS);
574 nfs_setsecurity(inode, fattr);
576 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
577 nfsi->attrtimeo_timestamp = now;
578 nfsi->access_cache = RB_ROOT;
580 nfs_fscache_init_inode(inode);
582 unlock_new_inode(inode);
584 int err = nfs_refresh_inode(inode, fattr);
587 inode = ERR_PTR(err);
591 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
593 (unsigned long long)NFS_FILEID(inode),
594 nfs_display_fhandle_hash(fh),
595 atomic_read(&inode->i_count));
601 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
604 EXPORT_SYMBOL_GPL(nfs_fhget);
606 #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)
609 nfs_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
612 struct inode *inode = d_inode(dentry);
613 struct nfs_fattr *fattr;
616 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
618 /* skip mode change if it's just for clearing setuid/setgid */
619 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
620 attr->ia_valid &= ~ATTR_MODE;
622 if (attr->ia_valid & ATTR_SIZE) {
623 BUG_ON(!S_ISREG(inode->i_mode));
625 error = inode_newsize_ok(inode, attr->ia_size);
629 if (attr->ia_size == i_size_read(inode))
630 attr->ia_valid &= ~ATTR_SIZE;
633 /* Optimization: if the end result is no change, don't RPC */
634 if (((attr->ia_valid & NFS_VALID_ATTRS) & ~(ATTR_FILE|ATTR_OPEN)) == 0)
637 trace_nfs_setattr_enter(inode);
639 /* Write all dirty data */
640 if (S_ISREG(inode->i_mode))
641 nfs_sync_inode(inode);
643 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
649 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
651 error = nfs_refresh_inode(inode, fattr);
652 nfs_free_fattr(fattr);
654 trace_nfs_setattr_exit(inode, error);
657 EXPORT_SYMBOL_GPL(nfs_setattr);
660 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
661 * @inode: inode of the file used
662 * @offset: file offset to start truncating
664 * This is a copy of the common vmtruncate, but with the locking
665 * corrected to take into account the fact that NFS requires
666 * inode->i_size to be updated under the inode->i_lock.
667 * Note: must be called with inode->i_lock held!
669 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
673 err = inode_newsize_ok(inode, offset);
677 trace_nfs_size_truncate(inode, offset);
678 i_size_write(inode, offset);
681 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_DATA |
682 NFS_INO_DATA_INVAL_DEFER);
683 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
685 spin_unlock(&inode->i_lock);
686 truncate_pagecache(inode, offset);
687 spin_lock(&inode->i_lock);
693 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
694 * @inode: pointer to struct inode
695 * @attr: pointer to struct iattr
696 * @fattr: pointer to struct nfs_fattr
698 * Note: we do this in the *proc.c in order to ensure that
699 * it works for things like exclusive creates too.
701 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
702 struct nfs_fattr *fattr)
704 /* Barrier: bump the attribute generation count. */
705 nfs_fattr_set_barrier(fattr);
707 spin_lock(&inode->i_lock);
708 NFS_I(inode)->attr_gencount = fattr->gencount;
709 if ((attr->ia_valid & ATTR_SIZE) != 0) {
710 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME |
711 NFS_INO_INVALID_BLOCKS);
712 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
713 nfs_vmtruncate(inode, attr->ia_size);
715 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
716 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
717 if ((attr->ia_valid & ATTR_KILL_SUID) != 0 &&
718 inode->i_mode & S_ISUID)
719 inode->i_mode &= ~S_ISUID;
720 if ((attr->ia_valid & ATTR_KILL_SGID) != 0 &&
721 (inode->i_mode & (S_ISGID | S_IXGRP)) ==
723 inode->i_mode &= ~S_ISGID;
724 if ((attr->ia_valid & ATTR_MODE) != 0) {
725 int mode = attr->ia_mode & S_IALLUGO;
726 mode |= inode->i_mode & ~S_IALLUGO;
727 inode->i_mode = mode;
729 if ((attr->ia_valid & ATTR_UID) != 0)
730 inode->i_uid = attr->ia_uid;
731 if ((attr->ia_valid & ATTR_GID) != 0)
732 inode->i_gid = attr->ia_gid;
733 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
734 inode->i_ctime = fattr->ctime;
736 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
737 | NFS_INO_INVALID_CTIME);
738 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
739 | NFS_INO_INVALID_ACL);
741 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
742 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
743 | NFS_INO_INVALID_CTIME);
744 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
745 inode->i_atime = fattr->atime;
746 else if (attr->ia_valid & ATTR_ATIME_SET)
747 inode->i_atime = attr->ia_atime;
749 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
751 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
752 inode->i_ctime = fattr->ctime;
754 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
755 | NFS_INO_INVALID_CTIME);
757 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
758 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
759 | NFS_INO_INVALID_CTIME);
760 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
761 inode->i_mtime = fattr->mtime;
762 else if (attr->ia_valid & ATTR_MTIME_SET)
763 inode->i_mtime = attr->ia_mtime;
765 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
767 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
768 inode->i_ctime = fattr->ctime;
770 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
771 | NFS_INO_INVALID_CTIME);
774 nfs_update_inode(inode, fattr);
775 spin_unlock(&inode->i_lock);
777 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
780 * Don't request help from readdirplus if the file is being written to,
781 * or if attribute caching is turned off
783 static bool nfs_getattr_readdirplus_enable(const struct inode *inode)
785 return nfs_server_capable(inode, NFS_CAP_READDIRPLUS) &&
786 !nfs_have_writebacks(inode) && NFS_MAXATTRTIMEO(inode) > 5 * HZ;
789 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
791 if (!IS_ROOT(dentry)) {
792 struct dentry *parent = dget_parent(dentry);
793 nfs_readdir_record_entry_cache_miss(d_inode(parent));
798 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
800 if (!IS_ROOT(dentry)) {
801 struct dentry *parent = dget_parent(dentry);
802 nfs_readdir_record_entry_cache_hit(d_inode(parent));
807 static u32 nfs_get_valid_attrmask(struct inode *inode)
809 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
810 u32 reply_mask = STATX_INO | STATX_TYPE;
812 if (!(cache_validity & NFS_INO_INVALID_ATIME))
813 reply_mask |= STATX_ATIME;
814 if (!(cache_validity & NFS_INO_INVALID_CTIME))
815 reply_mask |= STATX_CTIME;
816 if (!(cache_validity & NFS_INO_INVALID_MTIME))
817 reply_mask |= STATX_MTIME;
818 if (!(cache_validity & NFS_INO_INVALID_SIZE))
819 reply_mask |= STATX_SIZE;
820 if (!(cache_validity & NFS_INO_INVALID_NLINK))
821 reply_mask |= STATX_NLINK;
822 if (!(cache_validity & NFS_INO_INVALID_MODE))
823 reply_mask |= STATX_MODE;
824 if (!(cache_validity & NFS_INO_INVALID_OTHER))
825 reply_mask |= STATX_UID | STATX_GID;
826 if (!(cache_validity & NFS_INO_INVALID_BLOCKS))
827 reply_mask |= STATX_BLOCKS;
831 int nfs_getattr(struct user_namespace *mnt_userns, const struct path *path,
832 struct kstat *stat, u32 request_mask, unsigned int query_flags)
834 struct inode *inode = d_inode(path->dentry);
835 struct nfs_server *server = NFS_SERVER(inode);
836 unsigned long cache_validity;
838 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
839 bool do_update = false;
840 bool readdirplus_enabled = nfs_getattr_readdirplus_enable(inode);
842 trace_nfs_getattr_enter(inode);
844 request_mask &= STATX_TYPE | STATX_MODE | STATX_NLINK | STATX_UID |
845 STATX_GID | STATX_ATIME | STATX_MTIME | STATX_CTIME |
846 STATX_INO | STATX_SIZE | STATX_BLOCKS;
848 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync) {
849 if (readdirplus_enabled)
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 filemap_write_and_wait(inode->i_mapping);
860 * We may force a getattr if the user cares about atime.
862 * Note that we only have to check the vfsmount flags here:
863 * - NFS always sets S_NOATIME by so checking it would give a
865 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
866 * no point in checking those.
868 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
869 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
870 request_mask &= ~STATX_ATIME;
872 /* Is the user requesting attributes that might need revalidation? */
873 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
874 STATX_MTIME|STATX_UID|STATX_GID|
875 STATX_SIZE|STATX_BLOCKS)))
876 goto out_no_revalidate;
878 /* Check whether the cached attributes are stale */
879 do_update |= force_sync || nfs_attribute_cache_expired(inode);
880 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
881 do_update |= cache_validity & NFS_INO_INVALID_CHANGE;
882 if (request_mask & STATX_ATIME)
883 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
884 if (request_mask & STATX_CTIME)
885 do_update |= cache_validity & NFS_INO_INVALID_CTIME;
886 if (request_mask & STATX_MTIME)
887 do_update |= cache_validity & NFS_INO_INVALID_MTIME;
888 if (request_mask & STATX_SIZE)
889 do_update |= cache_validity & NFS_INO_INVALID_SIZE;
890 if (request_mask & STATX_NLINK)
891 do_update |= cache_validity & NFS_INO_INVALID_NLINK;
892 if (request_mask & STATX_MODE)
893 do_update |= cache_validity & NFS_INO_INVALID_MODE;
894 if (request_mask & (STATX_UID | STATX_GID))
895 do_update |= cache_validity & NFS_INO_INVALID_OTHER;
896 if (request_mask & STATX_BLOCKS)
897 do_update |= cache_validity & NFS_INO_INVALID_BLOCKS;
900 if (readdirplus_enabled)
901 nfs_readdirplus_parent_cache_miss(path->dentry);
902 err = __nfs_revalidate_inode(server, inode);
905 } else if (readdirplus_enabled)
906 nfs_readdirplus_parent_cache_hit(path->dentry);
908 /* Only return attributes that were revalidated. */
909 stat->result_mask = nfs_get_valid_attrmask(inode) | request_mask;
911 generic_fillattr(&init_user_ns, inode, stat);
912 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
913 if (S_ISDIR(inode->i_mode))
914 stat->blksize = NFS_SERVER(inode)->dtsize;
916 trace_nfs_getattr_exit(inode, err);
919 EXPORT_SYMBOL_GPL(nfs_getattr);
921 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
923 refcount_set(&l_ctx->count, 1);
924 l_ctx->lockowner = current->files;
925 INIT_LIST_HEAD(&l_ctx->list);
926 atomic_set(&l_ctx->io_count, 0);
929 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
931 struct nfs_lock_context *pos;
933 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
934 if (pos->lockowner != current->files)
936 if (refcount_inc_not_zero(&pos->count))
942 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
944 struct nfs_lock_context *res, *new = NULL;
945 struct inode *inode = d_inode(ctx->dentry);
948 res = __nfs_find_lock_context(ctx);
951 new = kmalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
953 return ERR_PTR(-ENOMEM);
954 nfs_init_lock_context(new);
955 spin_lock(&inode->i_lock);
956 res = __nfs_find_lock_context(ctx);
958 new->open_context = get_nfs_open_context(ctx);
959 if (new->open_context) {
960 list_add_tail_rcu(&new->list,
961 &ctx->lock_context.list);
965 res = ERR_PTR(-EBADF);
967 spin_unlock(&inode->i_lock);
972 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
974 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
976 struct nfs_open_context *ctx = l_ctx->open_context;
977 struct inode *inode = d_inode(ctx->dentry);
979 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
981 list_del_rcu(&l_ctx->list);
982 spin_unlock(&inode->i_lock);
983 put_nfs_open_context(ctx);
984 kfree_rcu(l_ctx, rcu_head);
986 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
989 * nfs_close_context - Common close_context() routine NFSv2/v3
990 * @ctx: pointer to context
991 * @is_sync: is this a synchronous close
993 * Ensure that the attributes are up to date if we're mounted
994 * with close-to-open semantics and we have cached data that will
995 * need to be revalidated on open.
997 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
999 struct nfs_inode *nfsi;
1000 struct inode *inode;
1002 if (!(ctx->mode & FMODE_WRITE))
1006 inode = d_inode(ctx->dentry);
1007 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1009 nfsi = NFS_I(inode);
1010 if (inode->i_mapping->nrpages == 0)
1012 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1014 if (!list_empty(&nfsi->open_files))
1016 if (NFS_SERVER(inode)->flags & NFS_MOUNT_NOCTO)
1018 nfs_revalidate_inode(inode,
1019 NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_SIZE);
1021 EXPORT_SYMBOL_GPL(nfs_close_context);
1023 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
1027 struct nfs_open_context *ctx;
1029 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
1031 return ERR_PTR(-ENOMEM);
1032 nfs_sb_active(dentry->d_sb);
1033 ctx->dentry = dget(dentry);
1035 ctx->cred = get_cred(filp->f_cred);
1037 ctx->cred = get_current_cred();
1038 rcu_assign_pointer(ctx->ll_cred, NULL);
1043 ctx->flock_owner = (fl_owner_t)filp;
1044 nfs_init_lock_context(&ctx->lock_context);
1045 ctx->lock_context.open_context = ctx;
1046 INIT_LIST_HEAD(&ctx->list);
1047 ctx->mdsthreshold = NULL;
1050 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
1052 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
1054 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
1058 EXPORT_SYMBOL_GPL(get_nfs_open_context);
1060 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
1062 struct inode *inode = d_inode(ctx->dentry);
1063 struct super_block *sb = ctx->dentry->d_sb;
1065 if (!refcount_dec_and_test(&ctx->lock_context.count))
1067 if (!list_empty(&ctx->list)) {
1068 spin_lock(&inode->i_lock);
1069 list_del_rcu(&ctx->list);
1070 spin_unlock(&inode->i_lock);
1073 NFS_PROTO(inode)->close_context(ctx, is_sync);
1074 put_cred(ctx->cred);
1076 nfs_sb_deactive(sb);
1077 put_rpccred(rcu_dereference_protected(ctx->ll_cred, 1));
1078 kfree(ctx->mdsthreshold);
1079 kfree_rcu(ctx, rcu_head);
1082 void put_nfs_open_context(struct nfs_open_context *ctx)
1084 __put_nfs_open_context(ctx, 0);
1086 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1088 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1090 __put_nfs_open_context(ctx, 1);
1094 * Ensure that mmap has a recent RPC credential for use when writing out
1097 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1099 struct inode *inode = d_inode(ctx->dentry);
1100 struct nfs_inode *nfsi = NFS_I(inode);
1102 spin_lock(&inode->i_lock);
1103 if (list_empty(&nfsi->open_files) &&
1104 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
1105 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA |
1106 NFS_INO_REVAL_FORCED);
1107 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1108 spin_unlock(&inode->i_lock);
1110 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1112 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1114 filp->private_data = get_nfs_open_context(ctx);
1115 set_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1116 if (list_empty(&ctx->list))
1117 nfs_inode_attach_open_context(ctx);
1119 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1122 * Given an inode, search for an open context with the desired characteristics
1124 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1126 struct nfs_inode *nfsi = NFS_I(inode);
1127 struct nfs_open_context *pos, *ctx = NULL;
1130 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1131 if (cred != NULL && cred_fscmp(pos->cred, cred) != 0)
1133 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1135 if (!test_bit(NFS_CONTEXT_FILE_OPEN, &pos->flags))
1137 ctx = get_nfs_open_context(pos);
1145 void nfs_file_clear_open_context(struct file *filp)
1147 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1150 struct inode *inode = d_inode(ctx->dentry);
1152 clear_bit(NFS_CONTEXT_FILE_OPEN, &ctx->flags);
1154 * We fatal error on write before. Try to writeback
1158 invalidate_inode_pages2(inode->i_mapping);
1159 filp->private_data = NULL;
1160 put_nfs_open_context_sync(ctx);
1165 * These allocate and release file read/write context information.
1167 int nfs_open(struct inode *inode, struct file *filp)
1169 struct nfs_open_context *ctx;
1171 ctx = alloc_nfs_open_context(file_dentry(filp),
1172 flags_to_mode(filp->f_flags), filp);
1174 return PTR_ERR(ctx);
1175 nfs_file_set_open_context(filp, ctx);
1176 put_nfs_open_context(ctx);
1177 nfs_fscache_open_file(inode, filp);
1182 * This function is called whenever some part of NFS notices that
1183 * the cached attributes have to be refreshed.
1186 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1188 int status = -ESTALE;
1189 struct nfs_fattr *fattr = NULL;
1190 struct nfs_inode *nfsi = NFS_I(inode);
1192 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1193 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1195 trace_nfs_revalidate_inode_enter(inode);
1197 if (is_bad_inode(inode))
1199 if (NFS_STALE(inode))
1202 /* pNFS: Attributes aren't updated until we layoutcommit */
1203 if (S_ISREG(inode->i_mode)) {
1204 status = pnfs_sync_inode(inode, false);
1210 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1214 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1216 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, inode);
1218 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1220 (unsigned long long)NFS_FILEID(inode), status);
1223 /* A soft timeout occurred. Use cached information? */
1224 if (server->flags & NFS_MOUNT_SOFTREVAL)
1228 if (!S_ISDIR(inode->i_mode))
1229 nfs_set_inode_stale(inode);
1231 nfs_zap_caches(inode);
1236 status = nfs_refresh_inode(inode, fattr);
1238 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1240 (unsigned long long)NFS_FILEID(inode), status);
1244 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1245 nfs_zap_acl_cache(inode);
1247 nfs_setsecurity(inode, fattr);
1249 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1251 (unsigned long long)NFS_FILEID(inode));
1254 nfs_free_fattr(fattr);
1255 trace_nfs_revalidate_inode_exit(inode, status);
1259 int nfs_attribute_cache_expired(struct inode *inode)
1261 if (nfs_have_delegated_attributes(inode))
1263 return nfs_attribute_timeout(inode);
1267 * nfs_revalidate_inode - Revalidate the inode attributes
1268 * @inode: pointer to inode struct
1269 * @flags: cache flags to check
1271 * Updates inode attribute information by retrieving the data from the server.
1273 int nfs_revalidate_inode(struct inode *inode, unsigned long flags)
1275 if (!nfs_check_cache_invalid(inode, flags))
1276 return NFS_STALE(inode) ? -ESTALE : 0;
1277 return __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1279 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1281 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1285 nfs_fscache_invalidate(inode, 0);
1286 if (mapping->nrpages != 0) {
1287 if (S_ISREG(inode->i_mode)) {
1288 ret = nfs_sync_mapping(mapping);
1292 ret = invalidate_inode_pages2(mapping);
1296 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1298 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1300 (unsigned long long)NFS_FILEID(inode));
1305 * nfs_clear_invalid_mapping - Conditionally clear a mapping
1306 * @mapping: pointer to mapping
1308 * If the NFS_INO_INVALID_DATA inode flag is set, clear the mapping.
1310 int nfs_clear_invalid_mapping(struct address_space *mapping)
1312 struct inode *inode = mapping->host;
1313 struct nfs_inode *nfsi = NFS_I(inode);
1314 unsigned long *bitlock = &nfsi->flags;
1318 * We must clear NFS_INO_INVALID_DATA first to ensure that
1319 * invalidations that come in while we're shooting down the mappings
1320 * are respected. But, that leaves a race window where one revalidator
1321 * can clear the flag, and then another checks it before the mapping
1322 * gets invalidated. Fix that by serializing access to this part of
1325 * At the same time, we need to allow other tasks to see whether we
1326 * might be in the middle of invalidating the pages, so we only set
1327 * the bit lock here if it looks like we're going to be doing that.
1330 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1331 nfs_wait_bit_killable,
1332 TASK_KILLABLE|TASK_FREEZABLE_UNSAFE);
1335 spin_lock(&inode->i_lock);
1336 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1337 spin_unlock(&inode->i_lock);
1340 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1342 spin_unlock(&inode->i_lock);
1346 set_bit(NFS_INO_INVALIDATING, bitlock);
1348 nfsi->cache_validity &=
1349 ~(NFS_INO_INVALID_DATA | NFS_INO_DATA_INVAL_DEFER);
1350 spin_unlock(&inode->i_lock);
1351 trace_nfs_invalidate_mapping_enter(inode);
1352 ret = nfs_invalidate_mapping(inode, mapping);
1353 trace_nfs_invalidate_mapping_exit(inode, ret);
1355 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1356 smp_mb__after_atomic();
1357 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1362 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1364 return nfs_check_cache_invalid(inode, NFS_INO_INVALID_CHANGE) ||
1368 int nfs_revalidate_mapping_rcu(struct inode *inode)
1370 struct nfs_inode *nfsi = NFS_I(inode);
1371 unsigned long *bitlock = &nfsi->flags;
1374 if (IS_SWAPFILE(inode))
1376 if (nfs_mapping_need_revalidate_inode(inode)) {
1380 spin_lock(&inode->i_lock);
1381 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1382 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1384 spin_unlock(&inode->i_lock);
1390 * nfs_revalidate_mapping - Revalidate the pagecache
1391 * @inode: pointer to host inode
1392 * @mapping: pointer to mapping
1394 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1396 /* swapfiles are not supposed to be shared. */
1397 if (IS_SWAPFILE(inode))
1400 if (nfs_mapping_need_revalidate_inode(inode)) {
1401 int ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1406 return nfs_clear_invalid_mapping(mapping);
1409 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1411 struct inode *inode = &nfsi->vfs_inode;
1413 if (!S_ISREG(inode->i_mode))
1415 if (list_empty(&nfsi->open_files))
1417 return inode_is_open_for_write(inode);
1420 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1422 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1425 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1427 struct timespec64 ts;
1429 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1430 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1431 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1432 inode_set_iversion_raw(inode, fattr->change_attr);
1433 if (S_ISDIR(inode->i_mode))
1434 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1435 else if (nfs_server_capable(inode, NFS_CAP_XATTR))
1436 nfs_set_cache_invalid(inode, NFS_INO_INVALID_XATTR);
1438 /* If we have atomic WCC data, we may update some attributes */
1439 ts = inode->i_ctime;
1440 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1441 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1442 && timespec64_equal(&ts, &fattr->pre_ctime)) {
1443 inode->i_ctime = fattr->ctime;
1446 ts = inode->i_mtime;
1447 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1448 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1449 && timespec64_equal(&ts, &fattr->pre_mtime)) {
1450 inode->i_mtime = fattr->mtime;
1452 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1453 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1454 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1455 && !nfs_have_writebacks(inode)) {
1456 trace_nfs_size_wcc(inode, fattr->size);
1457 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1462 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1463 * @inode: pointer to inode
1464 * @fattr: updated attributes
1466 * Verifies the attribute cache. If we have just changed the attributes,
1467 * so that fattr carries weak cache consistency data, then it may
1468 * also update the ctime/mtime/change_attribute.
1470 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1472 struct nfs_inode *nfsi = NFS_I(inode);
1473 loff_t cur_size, new_isize;
1474 unsigned long invalid = 0;
1475 struct timespec64 ts;
1477 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1480 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
1481 /* Only a mounted-on-fileid? Just exit */
1482 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1484 /* Has the inode gone and changed behind our back? */
1485 } else if (nfsi->fileid != fattr->fileid) {
1486 /* Is this perhaps the mounted-on fileid? */
1487 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
1488 nfsi->fileid == fattr->mounted_on_fileid)
1492 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode))
1496 if (!nfs_file_has_buffered_writers(nfsi)) {
1497 /* Verify a few of the more important attributes */
1498 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1499 invalid |= NFS_INO_INVALID_CHANGE;
1501 ts = inode->i_mtime;
1502 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec64_equal(&ts, &fattr->mtime))
1503 invalid |= NFS_INO_INVALID_MTIME;
1505 ts = inode->i_ctime;
1506 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec64_equal(&ts, &fattr->ctime))
1507 invalid |= NFS_INO_INVALID_CTIME;
1509 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1510 cur_size = i_size_read(inode);
1511 new_isize = nfs_size_to_loff_t(fattr->size);
1512 if (cur_size != new_isize)
1513 invalid |= NFS_INO_INVALID_SIZE;
1517 /* Have any file permissions changed? */
1518 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1519 invalid |= NFS_INO_INVALID_MODE;
1520 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1521 invalid |= NFS_INO_INVALID_OTHER;
1522 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1523 invalid |= NFS_INO_INVALID_OTHER;
1525 /* Has the link count changed? */
1526 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1527 invalid |= NFS_INO_INVALID_NLINK;
1529 ts = inode->i_atime;
1530 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec64_equal(&ts, &fattr->atime))
1531 invalid |= NFS_INO_INVALID_ATIME;
1534 nfs_set_cache_invalid(inode, invalid);
1536 nfsi->read_cache_jiffies = fattr->time_start;
1540 static atomic_long_t nfs_attr_generation_counter;
1542 static unsigned long nfs_read_attr_generation_counter(void)
1544 return atomic_long_read(&nfs_attr_generation_counter);
1547 unsigned long nfs_inc_attr_generation_counter(void)
1549 return atomic_long_inc_return(&nfs_attr_generation_counter);
1551 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1553 void nfs_fattr_init(struct nfs_fattr *fattr)
1556 fattr->time_start = jiffies;
1557 fattr->gencount = nfs_inc_attr_generation_counter();
1558 fattr->owner_name = NULL;
1559 fattr->group_name = NULL;
1561 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1564 * nfs_fattr_set_barrier
1565 * @fattr: attributes
1567 * Used to set a barrier after an attribute was updated. This
1568 * barrier ensures that older attributes from RPC calls that may
1569 * have raced with our update cannot clobber these new values.
1570 * Note that you are still responsible for ensuring that other
1571 * operations which change the attribute on the server do not
1574 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1576 fattr->gencount = nfs_inc_attr_generation_counter();
1579 struct nfs_fattr *nfs_alloc_fattr(void)
1581 struct nfs_fattr *fattr;
1583 fattr = kmalloc(sizeof(*fattr), GFP_KERNEL);
1584 if (fattr != NULL) {
1585 nfs_fattr_init(fattr);
1586 fattr->label = NULL;
1590 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1592 struct nfs_fattr *nfs_alloc_fattr_with_label(struct nfs_server *server)
1594 struct nfs_fattr *fattr = nfs_alloc_fattr();
1599 fattr->label = nfs4_label_alloc(server, GFP_KERNEL);
1600 if (IS_ERR(fattr->label)) {
1607 EXPORT_SYMBOL_GPL(nfs_alloc_fattr_with_label);
1609 struct nfs_fh *nfs_alloc_fhandle(void)
1613 fh = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
1618 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1622 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1623 * in the same way that wireshark does
1627 * For debugging only.
1629 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1631 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1632 * not on the result */
1633 return nfs_fhandle_hash(fh);
1635 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1638 * _nfs_display_fhandle - display an NFS file handle on the console
1640 * @fh: file handle to display
1641 * @caption: display caption
1643 * For debugging only.
1645 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1649 if (fh == NULL || fh->size == 0) {
1650 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1654 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1655 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1656 for (i = 0; i < fh->size; i += 16) {
1657 __be32 *pos = (__be32 *)&fh->data[i];
1659 switch ((fh->size - i - 1) >> 2) {
1661 printk(KERN_DEFAULT " %08x\n",
1665 printk(KERN_DEFAULT " %08x %08x\n",
1666 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1669 printk(KERN_DEFAULT " %08x %08x %08x\n",
1670 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1671 be32_to_cpup(pos + 2));
1674 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1675 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1676 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1680 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1684 * nfs_inode_attrs_cmp_generic - compare attributes
1685 * @fattr: attributes
1686 * @inode: pointer to inode
1688 * Attempt to divine whether or not an RPC call reply carrying stale
1689 * attributes got scheduled after another call carrying updated ones.
1690 * Note also the check for wraparound of 'attr_gencount'
1692 * The function returns '1' if it thinks the attributes in @fattr are
1693 * more recent than the ones cached in @inode. Otherwise it returns
1696 static int nfs_inode_attrs_cmp_generic(const struct nfs_fattr *fattr,
1697 const struct inode *inode)
1699 unsigned long attr_gencount = NFS_I(inode)->attr_gencount;
1701 return (long)(fattr->gencount - attr_gencount) > 0 ||
1702 (long)(attr_gencount - nfs_read_attr_generation_counter()) > 0;
1706 * nfs_inode_attrs_cmp_monotonic - compare attributes
1707 * @fattr: attributes
1708 * @inode: pointer to inode
1710 * Attempt to divine whether or not an RPC call reply carrying stale
1711 * attributes got scheduled after another call carrying updated ones.
1713 * We assume that the server observes monotonic semantics for
1714 * the change attribute, so a larger value means that the attributes in
1715 * @fattr are more recent, in which case the function returns the
1717 * A return value of '0' indicates no measurable change
1718 * A return value of '-1' means that the attributes in @inode are
1721 static int nfs_inode_attrs_cmp_monotonic(const struct nfs_fattr *fattr,
1722 const struct inode *inode)
1724 s64 diff = fattr->change_attr - inode_peek_iversion_raw(inode);
1727 return diff == 0 ? 0 : -1;
1731 * nfs_inode_attrs_cmp_strict_monotonic - compare attributes
1732 * @fattr: attributes
1733 * @inode: pointer to inode
1735 * Attempt to divine whether or not an RPC call reply carrying stale
1736 * attributes got scheduled after another call carrying updated ones.
1738 * We assume that the server observes strictly monotonic semantics for
1739 * the change attribute, so a larger value means that the attributes in
1740 * @fattr are more recent, in which case the function returns the
1742 * A return value of '-1' means that the attributes in @inode are
1743 * more recent or unchanged.
1745 static int nfs_inode_attrs_cmp_strict_monotonic(const struct nfs_fattr *fattr,
1746 const struct inode *inode)
1748 return nfs_inode_attrs_cmp_monotonic(fattr, inode) > 0 ? 1 : -1;
1752 * nfs_inode_attrs_cmp - compare attributes
1753 * @fattr: attributes
1754 * @inode: pointer to inode
1756 * This function returns '1' if it thinks the attributes in @fattr are
1757 * more recent than the ones cached in @inode. It returns '-1' if
1758 * the attributes in @inode are more recent than the ones in @fattr,
1759 * and it returns 0 if not sure.
1761 static int nfs_inode_attrs_cmp(const struct nfs_fattr *fattr,
1762 const struct inode *inode)
1764 if (nfs_inode_attrs_cmp_generic(fattr, inode) > 0)
1766 switch (NFS_SERVER(inode)->change_attr_type) {
1767 case NFS4_CHANGE_TYPE_IS_UNDEFINED:
1769 case NFS4_CHANGE_TYPE_IS_TIME_METADATA:
1770 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1772 return nfs_inode_attrs_cmp_monotonic(fattr, inode);
1774 if (!(fattr->valid & NFS_ATTR_FATTR_CHANGE))
1776 return nfs_inode_attrs_cmp_strict_monotonic(fattr, inode);
1782 * nfs_inode_finish_partial_attr_update - complete a previous inode update
1783 * @fattr: attributes
1784 * @inode: pointer to inode
1786 * Returns '1' if the last attribute update left the inode cached
1787 * attributes in a partially unrevalidated state, and @fattr
1788 * matches the change attribute of that partial update.
1789 * Otherwise returns '0'.
1791 static int nfs_inode_finish_partial_attr_update(const struct nfs_fattr *fattr,
1792 const struct inode *inode)
1794 const unsigned long check_valid =
1795 NFS_INO_INVALID_ATIME | NFS_INO_INVALID_CTIME |
1796 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
1797 NFS_INO_INVALID_BLOCKS | NFS_INO_INVALID_OTHER |
1798 NFS_INO_INVALID_NLINK;
1799 unsigned long cache_validity = NFS_I(inode)->cache_validity;
1800 enum nfs4_change_attr_type ctype = NFS_SERVER(inode)->change_attr_type;
1802 if (ctype != NFS4_CHANGE_TYPE_IS_UNDEFINED &&
1803 !(cache_validity & NFS_INO_INVALID_CHANGE) &&
1804 (cache_validity & check_valid) != 0 &&
1805 (fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1806 nfs_inode_attrs_cmp_monotonic(fattr, inode) == 0)
1811 static int nfs_refresh_inode_locked(struct inode *inode,
1812 struct nfs_fattr *fattr)
1814 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1817 trace_nfs_refresh_inode_enter(inode);
1819 if (attr_cmp > 0 || nfs_inode_finish_partial_attr_update(fattr, inode))
1820 ret = nfs_update_inode(inode, fattr);
1821 else if (attr_cmp == 0)
1822 ret = nfs_check_inode_attributes(inode, fattr);
1824 trace_nfs_refresh_inode_exit(inode, ret);
1829 * nfs_refresh_inode - try to update the inode attribute cache
1830 * @inode: pointer to inode
1831 * @fattr: updated attributes
1833 * Check that an RPC call that returned attributes has not overlapped with
1834 * other recent updates of the inode metadata, then decide whether it is
1835 * safe to do a full update of the inode attributes, or whether just to
1836 * call nfs_check_inode_attributes.
1838 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1842 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1844 spin_lock(&inode->i_lock);
1845 status = nfs_refresh_inode_locked(inode, fattr);
1846 spin_unlock(&inode->i_lock);
1850 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1852 static int nfs_post_op_update_inode_locked(struct inode *inode,
1853 struct nfs_fattr *fattr, unsigned int invalid)
1855 if (S_ISDIR(inode->i_mode))
1856 invalid |= NFS_INO_INVALID_DATA;
1857 nfs_set_cache_invalid(inode, invalid);
1858 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1860 return nfs_refresh_inode_locked(inode, fattr);
1864 * nfs_post_op_update_inode - try to update the inode attribute cache
1865 * @inode: pointer to inode
1866 * @fattr: updated attributes
1868 * After an operation that has changed the inode metadata, mark the
1869 * attribute cache as being invalid, then try to update it.
1871 * NB: if the server didn't return any post op attributes, this
1872 * function will force the retrieval of attributes before the next
1873 * NFS request. Thus it should be used only for operations that
1874 * are expected to change one or more attributes, to avoid
1875 * unnecessary NFS requests and trips through nfs_update_inode().
1877 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1881 spin_lock(&inode->i_lock);
1882 nfs_fattr_set_barrier(fattr);
1883 status = nfs_post_op_update_inode_locked(inode, fattr,
1884 NFS_INO_INVALID_CHANGE
1885 | NFS_INO_INVALID_CTIME
1886 | NFS_INO_REVAL_FORCED);
1887 spin_unlock(&inode->i_lock);
1891 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1894 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1895 * @inode: pointer to inode
1896 * @fattr: updated attributes
1898 * After an operation that has changed the inode metadata, mark the
1899 * attribute cache as being invalid, then try to update it. Fake up
1900 * weak cache consistency data, if none exist.
1902 * This function is mainly designed to be used by the ->write_done() functions.
1904 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1906 int attr_cmp = nfs_inode_attrs_cmp(fattr, inode);
1909 /* Don't do a WCC update if these attributes are already stale */
1912 if ((fattr->valid & NFS_ATTR_FATTR) == 0 || !attr_cmp) {
1913 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1914 | NFS_ATTR_FATTR_PRESIZE
1915 | NFS_ATTR_FATTR_PREMTIME
1916 | NFS_ATTR_FATTR_PRECTIME);
1919 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1920 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1921 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1922 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1924 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1925 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1926 fattr->pre_ctime = inode->i_ctime;
1927 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1929 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1930 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1931 fattr->pre_mtime = inode->i_mtime;
1932 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1934 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1935 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1936 fattr->pre_size = i_size_read(inode);
1937 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1940 status = nfs_post_op_update_inode_locked(inode, fattr,
1941 NFS_INO_INVALID_CHANGE
1942 | NFS_INO_INVALID_CTIME
1943 | NFS_INO_INVALID_MTIME
1944 | NFS_INO_INVALID_BLOCKS);
1949 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1950 * @inode: pointer to inode
1951 * @fattr: updated attributes
1953 * After an operation that has changed the inode metadata, mark the
1954 * attribute cache as being invalid, then try to update it. Fake up
1955 * weak cache consistency data, if none exist.
1957 * This function is mainly designed to be used by the ->write_done() functions.
1959 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1963 spin_lock(&inode->i_lock);
1964 nfs_fattr_set_barrier(fattr);
1965 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1966 spin_unlock(&inode->i_lock);
1969 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1973 * Many nfs protocol calls return the new file attributes after
1974 * an operation. Here we update the inode to reflect the state
1975 * of the server's inode.
1977 * This is a bit tricky because we have to make sure all dirty pages
1978 * have been sent off to the server before calling invalidate_inode_pages.
1979 * To make sure no other process adds more write requests while we try
1980 * our best to flush them, we make them sleep during the attribute refresh.
1982 * A very similar scenario holds for the dir cache.
1984 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1986 struct nfs_server *server = NFS_SERVER(inode);
1987 struct nfs_inode *nfsi = NFS_I(inode);
1988 loff_t cur_isize, new_isize;
1989 u64 fattr_supported = server->fattr_valid;
1990 unsigned long invalid = 0;
1991 unsigned long now = jiffies;
1992 unsigned long save_cache_validity;
1993 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1994 bool cache_revalidated = true;
1995 bool attr_changed = false;
1996 bool have_delegation;
1998 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1999 __func__, inode->i_sb->s_id, inode->i_ino,
2000 nfs_display_fhandle_hash(NFS_FH(inode)),
2001 atomic_read(&inode->i_count), fattr->valid);
2003 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID)) {
2004 /* Only a mounted-on-fileid? Just exit */
2005 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
2007 /* Has the inode gone and changed behind our back? */
2008 } else if (nfsi->fileid != fattr->fileid) {
2009 /* Is this perhaps the mounted-on fileid? */
2010 if ((fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID) &&
2011 nfsi->fileid == fattr->mounted_on_fileid)
2013 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
2014 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
2015 NFS_SERVER(inode)->nfs_client->cl_hostname,
2016 inode->i_sb->s_id, (long long)nfsi->fileid,
2017 (long long)fattr->fileid);
2022 * Make sure the inode's type hasn't changed.
2024 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && inode_wrong_type(inode, fattr->mode)) {
2026 * Big trouble! The inode has become a different object.
2028 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
2029 __func__, inode->i_ino, inode->i_mode, fattr->mode);
2033 /* Update the fsid? */
2034 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
2035 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
2036 !IS_AUTOMOUNT(inode))
2037 server->fsid = fattr->fsid;
2039 /* Save the delegation state before clearing cache_validity */
2040 have_delegation = nfs_have_delegated_attributes(inode);
2043 * Update the read time so we don't revalidate too often.
2045 nfsi->read_cache_jiffies = fattr->time_start;
2047 save_cache_validity = nfsi->cache_validity;
2048 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
2049 | NFS_INO_INVALID_ATIME
2050 | NFS_INO_REVAL_FORCED
2051 | NFS_INO_INVALID_BLOCKS);
2053 /* Do atomic weak cache consistency updates */
2054 nfs_wcc_update_inode(inode, fattr);
2056 if (pnfs_layoutcommit_outstanding(inode)) {
2057 nfsi->cache_validity |=
2058 save_cache_validity &
2059 (NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
2060 NFS_INO_INVALID_MTIME | NFS_INO_INVALID_SIZE |
2061 NFS_INO_INVALID_BLOCKS);
2062 cache_revalidated = false;
2065 /* More cache consistency checks */
2066 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
2067 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
2068 /* Could it be a race with writeback? */
2069 if (!(have_writers || have_delegation)) {
2070 invalid |= NFS_INO_INVALID_DATA
2071 | NFS_INO_INVALID_ACCESS
2072 | NFS_INO_INVALID_ACL
2073 | NFS_INO_INVALID_XATTR;
2074 /* Force revalidate of all attributes */
2075 save_cache_validity |= NFS_INO_INVALID_CTIME
2076 | NFS_INO_INVALID_MTIME
2077 | NFS_INO_INVALID_SIZE
2078 | NFS_INO_INVALID_BLOCKS
2079 | NFS_INO_INVALID_NLINK
2080 | NFS_INO_INVALID_MODE
2081 | NFS_INO_INVALID_OTHER;
2082 if (S_ISDIR(inode->i_mode))
2083 nfs_force_lookup_revalidate(inode);
2084 attr_changed = true;
2085 dprintk("NFS: change_attr change on server for file %s/%ld\n",
2088 } else if (!have_delegation)
2089 nfsi->cache_validity |= NFS_INO_DATA_INVAL_DEFER;
2090 inode_set_iversion_raw(inode, fattr->change_attr);
2093 nfsi->cache_validity |=
2094 save_cache_validity & NFS_INO_INVALID_CHANGE;
2095 if (!have_delegation ||
2096 (nfsi->cache_validity & NFS_INO_INVALID_CHANGE) != 0)
2097 cache_revalidated = false;
2100 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
2101 inode->i_mtime = fattr->mtime;
2102 else if (fattr_supported & NFS_ATTR_FATTR_MTIME)
2103 nfsi->cache_validity |=
2104 save_cache_validity & NFS_INO_INVALID_MTIME;
2106 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
2107 inode->i_ctime = fattr->ctime;
2108 else if (fattr_supported & NFS_ATTR_FATTR_CTIME)
2109 nfsi->cache_validity |=
2110 save_cache_validity & NFS_INO_INVALID_CTIME;
2112 /* Check if our cached file size is stale */
2113 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
2114 new_isize = nfs_size_to_loff_t(fattr->size);
2115 cur_isize = i_size_read(inode);
2116 if (new_isize != cur_isize && !have_delegation) {
2117 /* Do we perhaps have any outstanding writes, or has
2118 * the file grown beyond our last write? */
2119 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
2120 trace_nfs_size_update(inode, new_isize);
2121 i_size_write(inode, new_isize);
2123 invalid |= NFS_INO_INVALID_DATA;
2126 if (new_isize == 0 &&
2127 !(fattr->valid & (NFS_ATTR_FATTR_SPACE_USED |
2128 NFS_ATTR_FATTR_BLOCKS_USED))) {
2129 fattr->du.nfs3.used = 0;
2130 fattr->valid |= NFS_ATTR_FATTR_SPACE_USED;
2133 nfsi->cache_validity |=
2134 save_cache_validity & NFS_INO_INVALID_SIZE;
2136 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
2137 inode->i_atime = fattr->atime;
2138 else if (fattr_supported & NFS_ATTR_FATTR_ATIME)
2139 nfsi->cache_validity |=
2140 save_cache_validity & NFS_INO_INVALID_ATIME;
2142 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
2143 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
2144 umode_t newmode = inode->i_mode & S_IFMT;
2145 newmode |= fattr->mode & S_IALLUGO;
2146 inode->i_mode = newmode;
2147 invalid |= NFS_INO_INVALID_ACCESS
2148 | NFS_INO_INVALID_ACL;
2150 } else if (fattr_supported & NFS_ATTR_FATTR_MODE)
2151 nfsi->cache_validity |=
2152 save_cache_validity & NFS_INO_INVALID_MODE;
2154 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
2155 if (!uid_eq(inode->i_uid, fattr->uid)) {
2156 invalid |= NFS_INO_INVALID_ACCESS
2157 | NFS_INO_INVALID_ACL;
2158 inode->i_uid = fattr->uid;
2160 } else if (fattr_supported & NFS_ATTR_FATTR_OWNER)
2161 nfsi->cache_validity |=
2162 save_cache_validity & NFS_INO_INVALID_OTHER;
2164 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
2165 if (!gid_eq(inode->i_gid, fattr->gid)) {
2166 invalid |= NFS_INO_INVALID_ACCESS
2167 | NFS_INO_INVALID_ACL;
2168 inode->i_gid = fattr->gid;
2170 } else if (fattr_supported & NFS_ATTR_FATTR_GROUP)
2171 nfsi->cache_validity |=
2172 save_cache_validity & NFS_INO_INVALID_OTHER;
2174 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
2175 if (inode->i_nlink != fattr->nlink)
2176 set_nlink(inode, fattr->nlink);
2177 } else if (fattr_supported & NFS_ATTR_FATTR_NLINK)
2178 nfsi->cache_validity |=
2179 save_cache_validity & NFS_INO_INVALID_NLINK;
2181 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2183 * report the blocks in 512byte units
2185 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2186 } else if (fattr_supported & NFS_ATTR_FATTR_SPACE_USED)
2187 nfsi->cache_validity |=
2188 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2190 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2191 inode->i_blocks = fattr->du.nfs2.blocks;
2192 else if (fattr_supported & NFS_ATTR_FATTR_BLOCKS_USED)
2193 nfsi->cache_validity |=
2194 save_cache_validity & NFS_INO_INVALID_BLOCKS;
2196 /* Update attrtimeo value if we're out of the unstable period */
2198 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2199 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2200 nfsi->attrtimeo_timestamp = now;
2201 /* Set barrier to be more recent than all outstanding updates */
2202 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2204 if (cache_revalidated) {
2205 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2206 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2207 nfsi->attrtimeo <<= 1;
2208 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2209 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2211 nfsi->attrtimeo_timestamp = now;
2213 /* Set the barrier to be more recent than this fattr */
2214 if ((long)(fattr->gencount - nfsi->attr_gencount) > 0)
2215 nfsi->attr_gencount = fattr->gencount;
2218 /* Don't invalidate the data if we were to blame */
2219 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2220 || S_ISLNK(inode->i_mode)))
2221 invalid &= ~NFS_INO_INVALID_DATA;
2222 nfs_set_cache_invalid(inode, invalid);
2227 * No need to worry about unhashing the dentry, as the
2228 * lookup validation will know that the inode is bad.
2229 * (But we fall through to invalidate the caches.)
2231 nfs_set_inode_stale_locked(inode);
2235 struct inode *nfs_alloc_inode(struct super_block *sb)
2237 struct nfs_inode *nfsi;
2238 nfsi = alloc_inode_sb(sb, nfs_inode_cachep, GFP_KERNEL);
2242 nfsi->cache_validity = 0UL;
2243 #if IS_ENABLED(CONFIG_NFS_V4)
2244 nfsi->nfs4_acl = NULL;
2245 #endif /* CONFIG_NFS_V4 */
2246 #ifdef CONFIG_NFS_V4_2
2247 nfsi->xattr_cache = NULL;
2249 return &nfsi->vfs_inode;
2251 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2253 void nfs_free_inode(struct inode *inode)
2255 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2257 EXPORT_SYMBOL_GPL(nfs_free_inode);
2259 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2261 #if IS_ENABLED(CONFIG_NFS_V4)
2262 INIT_LIST_HEAD(&nfsi->open_states);
2263 nfsi->delegation = NULL;
2264 init_rwsem(&nfsi->rwsem);
2265 nfsi->layout = NULL;
2269 static void init_once(void *foo)
2271 struct nfs_inode *nfsi = foo;
2273 inode_init_once(&nfsi->vfs_inode);
2274 INIT_LIST_HEAD(&nfsi->open_files);
2275 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2276 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2277 nfs4_init_once(nfsi);
2280 static int __init nfs_init_inodecache(void)
2282 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2283 sizeof(struct nfs_inode),
2284 0, (SLAB_RECLAIM_ACCOUNT|
2285 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2287 if (nfs_inode_cachep == NULL)
2293 static void nfs_destroy_inodecache(void)
2296 * Make sure all delayed rcu free inodes are flushed before we
2300 kmem_cache_destroy(nfs_inode_cachep);
2303 struct workqueue_struct *nfsiod_workqueue;
2304 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2307 * start up the nfsiod workqueue
2309 static int nfsiod_start(void)
2311 struct workqueue_struct *wq;
2312 dprintk("RPC: creating workqueue nfsiod\n");
2313 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM | WQ_UNBOUND, 0);
2316 nfsiod_workqueue = wq;
2321 * Destroy the nfsiod workqueue
2323 static void nfsiod_stop(void)
2325 struct workqueue_struct *wq;
2327 wq = nfsiod_workqueue;
2330 nfsiod_workqueue = NULL;
2331 destroy_workqueue(wq);
2334 unsigned int nfs_net_id;
2335 EXPORT_SYMBOL_GPL(nfs_net_id);
2337 static int nfs_net_init(struct net *net)
2339 nfs_clients_init(net);
2340 return nfs_fs_proc_net_init(net);
2343 static void nfs_net_exit(struct net *net)
2345 nfs_fs_proc_net_exit(net);
2346 nfs_clients_exit(net);
2349 static struct pernet_operations nfs_net_ops = {
2350 .init = nfs_net_init,
2351 .exit = nfs_net_exit,
2353 .size = sizeof(struct nfs_net),
2359 static int __init init_nfs_fs(void)
2363 err = nfs_sysfs_init();
2367 err = register_pernet_subsys(&nfs_net_ops);
2371 err = nfsiod_start();
2375 err = nfs_fs_proc_init();
2379 err = nfs_init_nfspagecache();
2383 err = nfs_init_inodecache();
2387 err = nfs_init_readpagecache();
2391 err = nfs_init_writepagecache();
2395 err = nfs_init_directcache();
2399 rpc_proc_register(&init_net, &nfs_rpcstat);
2401 err = register_nfs_fs();
2407 rpc_proc_unregister(&init_net, "nfs");
2408 nfs_destroy_directcache();
2410 nfs_destroy_writepagecache();
2412 nfs_destroy_readpagecache();
2414 nfs_destroy_inodecache();
2416 nfs_destroy_nfspagecache();
2422 unregister_pernet_subsys(&nfs_net_ops);
2429 static void __exit exit_nfs_fs(void)
2431 nfs_destroy_directcache();
2432 nfs_destroy_writepagecache();
2433 nfs_destroy_readpagecache();
2434 nfs_destroy_inodecache();
2435 nfs_destroy_nfspagecache();
2436 unregister_pernet_subsys(&nfs_net_ops);
2437 rpc_proc_unregister(&init_net, "nfs");
2438 unregister_nfs_fs();
2444 /* Not quite true; I just maintain it */
2446 MODULE_LICENSE("GPL");
2447 module_param(enable_ino64, bool, 0644);
2449 module_init(init_nfs_fs)
2450 module_exit(exit_nfs_fs)