1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs inode and superblock handling functions
9 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
10 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
12 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
13 * J.S.Peatfield@damtp.cam.ac.uk
17 #include <linux/module.h>
18 #include <linux/init.h>
19 #include <linux/sched/signal.h>
20 #include <linux/time.h>
21 #include <linux/kernel.h>
23 #include <linux/string.h>
24 #include <linux/stat.h>
25 #include <linux/errno.h>
26 #include <linux/unistd.h>
27 #include <linux/sunrpc/clnt.h>
28 #include <linux/sunrpc/stats.h>
29 #include <linux/sunrpc/metrics.h>
30 #include <linux/nfs_fs.h>
31 #include <linux/nfs_mount.h>
32 #include <linux/nfs4_mount.h>
33 #include <linux/lockd/bind.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/uaccess.h>
43 #include <linux/iversion.h>
47 #include "delegation.h"
57 #define NFSDBG_FACILITY NFSDBG_VFS
59 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
61 /* Default is to see 64-bit inode numbers */
62 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
64 static void nfs_invalidate_inode(struct inode *);
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_invalid_delegated(struct inode *inode, unsigned long flags)
169 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
171 /* Special case for the pagecache or access cache */
172 if (flags == NFS_INO_REVAL_PAGECACHE &&
173 !(cache_validity & NFS_INO_REVAL_FORCED))
175 return (cache_validity & flags) != 0;
178 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
180 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
182 if ((cache_validity & flags) != 0)
184 if (nfs_attribute_timeout(inode))
189 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
191 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
192 return nfs_check_cache_invalid_delegated(inode, flags);
194 return nfs_check_cache_invalid_not_delegated(inode, flags);
197 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
199 struct nfs_inode *nfsi = NFS_I(inode);
200 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
202 if (have_delegation) {
203 if (!(flags & NFS_INO_REVAL_FORCED))
204 flags &= ~NFS_INO_INVALID_OTHER;
205 flags &= ~(NFS_INO_INVALID_CHANGE
206 | NFS_INO_INVALID_SIZE
207 | NFS_INO_REVAL_PAGECACHE);
210 if (inode->i_mapping->nrpages == 0)
211 flags &= ~NFS_INO_INVALID_DATA;
212 nfsi->cache_validity |= flags;
213 if (flags & NFS_INO_INVALID_DATA)
214 nfs_fscache_invalidate(inode);
218 * Invalidate the local caches
220 static void nfs_zap_caches_locked(struct inode *inode)
222 struct nfs_inode *nfsi = NFS_I(inode);
223 int mode = inode->i_mode;
225 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
227 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
228 nfsi->attrtimeo_timestamp = jiffies;
230 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
231 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
232 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
233 | NFS_INO_INVALID_DATA
234 | NFS_INO_INVALID_ACCESS
235 | NFS_INO_INVALID_ACL
236 | NFS_INO_REVAL_PAGECACHE);
238 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
239 | NFS_INO_INVALID_ACCESS
240 | NFS_INO_INVALID_ACL
241 | NFS_INO_REVAL_PAGECACHE);
242 nfs_zap_label_cache_locked(nfsi);
245 void nfs_zap_caches(struct inode *inode)
247 spin_lock(&inode->i_lock);
248 nfs_zap_caches_locked(inode);
249 spin_unlock(&inode->i_lock);
252 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
254 if (mapping->nrpages != 0) {
255 spin_lock(&inode->i_lock);
256 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
257 spin_unlock(&inode->i_lock);
261 void nfs_zap_acl_cache(struct inode *inode)
263 void (*clear_acl_cache)(struct inode *);
265 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
266 if (clear_acl_cache != NULL)
267 clear_acl_cache(inode);
268 spin_lock(&inode->i_lock);
269 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
270 spin_unlock(&inode->i_lock);
272 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
274 void nfs_invalidate_atime(struct inode *inode)
276 spin_lock(&inode->i_lock);
277 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
278 spin_unlock(&inode->i_lock);
280 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
283 * Invalidate, but do not unhash, the inode.
284 * NB: must be called with inode->i_lock held!
286 static void nfs_invalidate_inode(struct inode *inode)
288 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
289 nfs_zap_caches_locked(inode);
292 struct nfs_find_desc {
294 struct nfs_fattr *fattr;
298 * In NFSv3 we can have 64bit inode numbers. In order to support
299 * this, and re-exported directories (also seen in NFSv2)
300 * we are forced to allow 2 different inodes to have the same
304 nfs_find_actor(struct inode *inode, void *opaque)
306 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
307 struct nfs_fh *fh = desc->fh;
308 struct nfs_fattr *fattr = desc->fattr;
310 if (NFS_FILEID(inode) != fattr->fileid)
312 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
314 if (nfs_compare_fh(NFS_FH(inode), fh))
316 if (is_bad_inode(inode) || NFS_STALE(inode))
322 nfs_init_locked(struct inode *inode, void *opaque)
324 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
325 struct nfs_fattr *fattr = desc->fattr;
327 set_nfs_fileid(inode, fattr->fileid);
328 inode->i_mode = fattr->mode;
329 nfs_copy_fh(NFS_FH(inode), desc->fh);
333 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
334 static void nfs_clear_label_invalid(struct inode *inode)
336 spin_lock(&inode->i_lock);
337 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
338 spin_unlock(&inode->i_lock);
341 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
342 struct nfs4_label *label)
349 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
350 error = security_inode_notifysecctx(inode, label->label,
353 printk(KERN_ERR "%s() %s %d "
354 "security_inode_notifysecctx() %d\n",
356 (char *)label->label,
358 nfs_clear_label_invalid(inode);
362 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
364 struct nfs4_label *label = NULL;
365 int minor_version = server->nfs_client->cl_minorversion;
367 if (minor_version < 2)
370 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
373 label = kzalloc(sizeof(struct nfs4_label), flags);
375 return ERR_PTR(-ENOMEM);
377 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
378 if (label->label == NULL) {
380 return ERR_PTR(-ENOMEM);
382 label->len = NFS4_MAXLABELLEN;
386 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
388 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
389 struct nfs4_label *label)
393 EXPORT_SYMBOL_GPL(nfs_setsecurity);
395 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
397 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
399 struct nfs_find_desc desc = {
406 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
407 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
410 hash = nfs_fattr_to_ino_t(fattr);
411 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
413 dprintk("%s: returning %p\n", __func__, inode);
418 * This is our front-end to iget that looks up inodes by file handle
419 * instead of inode number.
422 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
424 struct nfs_find_desc desc = {
428 struct inode *inode = ERR_PTR(-ENOENT);
431 nfs_attr_check_mountpoint(sb, fattr);
433 if (nfs_attr_use_mounted_on_fileid(fattr))
434 fattr->fileid = fattr->mounted_on_fileid;
435 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
437 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
440 hash = nfs_fattr_to_ino_t(fattr);
442 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
444 inode = ERR_PTR(-ENOMEM);
448 if (inode->i_state & I_NEW) {
449 struct nfs_inode *nfsi = NFS_I(inode);
450 unsigned long now = jiffies;
452 /* We set i_ino for the few things that still rely on it,
456 /* We can't support update_atime(), since the server will reset it */
457 inode->i_flags |= S_NOATIME|S_NOCMTIME;
458 inode->i_mode = fattr->mode;
459 nfsi->cache_validity = 0;
460 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
461 && nfs_server_capable(inode, NFS_CAP_MODE))
462 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
463 /* Why so? Because we want revalidate for devices/FIFOs, and
464 * that's precisely what we have in nfs_file_inode_operations.
466 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
467 if (S_ISREG(inode->i_mode)) {
468 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
469 inode->i_data.a_ops = &nfs_file_aops;
470 } else if (S_ISDIR(inode->i_mode)) {
471 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
472 inode->i_fop = &nfs_dir_operations;
473 inode->i_data.a_ops = &nfs_dir_aops;
474 /* Deal with crossing mountpoints */
475 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
476 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
477 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
478 inode->i_op = &nfs_referral_inode_operations;
480 inode->i_op = &nfs_mountpoint_inode_operations;
482 inode->i_flags |= S_AUTOMOUNT;
484 } else if (S_ISLNK(inode->i_mode)) {
485 inode->i_op = &nfs_symlink_inode_operations;
486 inode_nohighmem(inode);
488 init_special_inode(inode, inode->i_mode, fattr->rdev);
490 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
491 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
492 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
493 inode_set_iversion_raw(inode, 0);
496 inode->i_uid = make_kuid(&init_user_ns, -2);
497 inode->i_gid = make_kgid(&init_user_ns, -2);
499 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
503 nfsi->read_cache_jiffies = fattr->time_start;
504 nfsi->attr_gencount = fattr->gencount;
505 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
506 inode->i_atime = timespec_to_timespec64(fattr->atime);
507 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
508 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
509 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
510 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
511 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
512 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
513 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
514 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
515 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
516 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
517 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
518 inode_set_iversion_raw(inode, fattr->change_attr);
520 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
521 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
522 inode->i_size = nfs_size_to_loff_t(fattr->size);
524 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
525 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
526 set_nlink(inode, fattr->nlink);
527 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
528 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
529 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
530 inode->i_uid = fattr->uid;
531 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
532 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
533 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
534 inode->i_gid = fattr->gid;
535 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
536 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
537 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
538 inode->i_blocks = fattr->du.nfs2.blocks;
539 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
541 * report the blocks in 512byte units
543 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
546 if (nfsi->cache_validity != 0)
547 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
549 nfs_setsecurity(inode, fattr, label);
551 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
552 nfsi->attrtimeo_timestamp = now;
553 nfsi->access_cache = RB_ROOT;
555 nfs_fscache_init_inode(inode);
557 unlock_new_inode(inode);
559 int err = nfs_refresh_inode(inode, fattr);
562 inode = ERR_PTR(err);
566 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
568 (unsigned long long)NFS_FILEID(inode),
569 nfs_display_fhandle_hash(fh),
570 atomic_read(&inode->i_count));
576 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
579 EXPORT_SYMBOL_GPL(nfs_fhget);
581 #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)
584 nfs_setattr(struct dentry *dentry, struct iattr *attr)
586 struct inode *inode = d_inode(dentry);
587 struct nfs_fattr *fattr;
590 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
592 /* skip mode change if it's just for clearing setuid/setgid */
593 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
594 attr->ia_valid &= ~ATTR_MODE;
596 if (attr->ia_valid & ATTR_SIZE) {
597 BUG_ON(!S_ISREG(inode->i_mode));
599 error = inode_newsize_ok(inode, attr->ia_size);
603 if (attr->ia_size == i_size_read(inode))
604 attr->ia_valid &= ~ATTR_SIZE;
607 /* Optimization: if the end result is no change, don't RPC */
608 attr->ia_valid &= NFS_VALID_ATTRS;
609 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
612 trace_nfs_setattr_enter(inode);
614 /* Write all dirty data */
615 if (S_ISREG(inode->i_mode))
616 nfs_sync_inode(inode);
618 fattr = nfs_alloc_fattr();
624 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
626 error = nfs_refresh_inode(inode, fattr);
627 nfs_free_fattr(fattr);
629 trace_nfs_setattr_exit(inode, error);
632 EXPORT_SYMBOL_GPL(nfs_setattr);
635 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
636 * @inode: inode of the file used
637 * @offset: file offset to start truncating
639 * This is a copy of the common vmtruncate, but with the locking
640 * corrected to take into account the fact that NFS requires
641 * inode->i_size to be updated under the inode->i_lock.
642 * Note: must be called with inode->i_lock held!
644 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
648 err = inode_newsize_ok(inode, offset);
652 i_size_write(inode, offset);
655 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
656 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
658 spin_unlock(&inode->i_lock);
659 truncate_pagecache(inode, offset);
660 spin_lock(&inode->i_lock);
666 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
667 * @inode: pointer to struct inode
668 * @attr: pointer to struct iattr
669 * @fattr: pointer to struct nfs_fattr
671 * Note: we do this in the *proc.c in order to ensure that
672 * it works for things like exclusive creates too.
674 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
675 struct nfs_fattr *fattr)
677 /* Barrier: bump the attribute generation count. */
678 nfs_fattr_set_barrier(fattr);
680 spin_lock(&inode->i_lock);
681 NFS_I(inode)->attr_gencount = fattr->gencount;
682 if ((attr->ia_valid & ATTR_SIZE) != 0) {
683 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
684 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
685 nfs_vmtruncate(inode, attr->ia_size);
687 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
688 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
689 if ((attr->ia_valid & ATTR_MODE) != 0) {
690 int mode = attr->ia_mode & S_IALLUGO;
691 mode |= inode->i_mode & ~S_IALLUGO;
692 inode->i_mode = mode;
694 if ((attr->ia_valid & ATTR_UID) != 0)
695 inode->i_uid = attr->ia_uid;
696 if ((attr->ia_valid & ATTR_GID) != 0)
697 inode->i_gid = attr->ia_gid;
698 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
699 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
701 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
702 | NFS_INO_INVALID_CTIME);
703 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
704 | NFS_INO_INVALID_ACL);
706 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
707 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
708 | NFS_INO_INVALID_CTIME);
709 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
710 inode->i_atime = timespec_to_timespec64(fattr->atime);
711 else if (attr->ia_valid & ATTR_ATIME_SET)
712 inode->i_atime = attr->ia_atime;
714 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
716 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
717 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
719 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
720 | NFS_INO_INVALID_CTIME);
722 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
723 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
724 | NFS_INO_INVALID_CTIME);
725 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
726 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
727 else if (attr->ia_valid & ATTR_MTIME_SET)
728 inode->i_mtime = attr->ia_mtime;
730 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
732 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
733 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
735 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
736 | NFS_INO_INVALID_CTIME);
739 nfs_update_inode(inode, fattr);
740 spin_unlock(&inode->i_lock);
742 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
744 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
746 struct dentry *parent;
748 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
750 parent = dget_parent(dentry);
751 nfs_force_use_readdirplus(d_inode(parent));
755 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
757 struct dentry *parent;
759 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
761 parent = dget_parent(dentry);
762 nfs_advise_use_readdirplus(d_inode(parent));
766 static bool nfs_need_revalidate_inode(struct inode *inode)
768 if (NFS_I(inode)->cache_validity &
769 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
771 if (nfs_attribute_cache_expired(inode))
776 int nfs_getattr(const struct path *path, struct kstat *stat,
777 u32 request_mask, unsigned int query_flags)
779 struct inode *inode = d_inode(path->dentry);
780 struct nfs_server *server = NFS_SERVER(inode);
781 unsigned long cache_validity;
783 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
784 bool do_update = false;
786 trace_nfs_getattr_enter(inode);
788 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
791 /* Flush out writes to the server in order to update c/mtime. */
792 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
793 S_ISREG(inode->i_mode)) {
794 err = filemap_write_and_wait(inode->i_mapping);
800 * We may force a getattr if the user cares about atime.
802 * Note that we only have to check the vfsmount flags here:
803 * - NFS always sets S_NOATIME by so checking it would give a
805 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
806 * no point in checking those.
808 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
809 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
810 request_mask &= ~STATX_ATIME;
812 /* Is the user requesting attributes that might need revalidation? */
813 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
814 STATX_MTIME|STATX_UID|STATX_GID|
815 STATX_SIZE|STATX_BLOCKS)))
816 goto out_no_revalidate;
818 /* Check whether the cached attributes are stale */
819 do_update |= force_sync || nfs_attribute_cache_expired(inode);
820 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
821 do_update |= cache_validity &
822 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
823 if (request_mask & STATX_ATIME)
824 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
825 if (request_mask & (STATX_CTIME|STATX_MTIME))
826 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
828 /* Update the attribute cache */
829 if (!(server->flags & NFS_MOUNT_NOAC))
830 nfs_readdirplus_parent_cache_miss(path->dentry);
832 nfs_readdirplus_parent_cache_hit(path->dentry);
833 err = __nfs_revalidate_inode(server, inode);
837 nfs_readdirplus_parent_cache_hit(path->dentry);
839 /* Only return attributes that were revalidated. */
840 stat->result_mask &= request_mask;
842 generic_fillattr(inode, stat);
843 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
844 if (S_ISDIR(inode->i_mode))
845 stat->blksize = NFS_SERVER(inode)->dtsize;
847 trace_nfs_getattr_exit(inode, err);
850 EXPORT_SYMBOL_GPL(nfs_getattr);
852 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
854 refcount_set(&l_ctx->count, 1);
855 l_ctx->lockowner = current->files;
856 INIT_LIST_HEAD(&l_ctx->list);
857 atomic_set(&l_ctx->io_count, 0);
860 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
862 struct nfs_lock_context *pos;
864 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
865 if (pos->lockowner != current->files)
867 if (refcount_inc_not_zero(&pos->count))
873 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
875 struct nfs_lock_context *res, *new = NULL;
876 struct inode *inode = d_inode(ctx->dentry);
879 res = __nfs_find_lock_context(ctx);
882 new = kmalloc(sizeof(*new), GFP_KERNEL);
884 return ERR_PTR(-ENOMEM);
885 nfs_init_lock_context(new);
886 spin_lock(&inode->i_lock);
887 res = __nfs_find_lock_context(ctx);
889 new->open_context = get_nfs_open_context(ctx);
890 if (new->open_context) {
891 list_add_tail_rcu(&new->list,
892 &ctx->lock_context.list);
896 res = ERR_PTR(-EBADF);
898 spin_unlock(&inode->i_lock);
903 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
905 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
907 struct nfs_open_context *ctx = l_ctx->open_context;
908 struct inode *inode = d_inode(ctx->dentry);
910 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
912 list_del_rcu(&l_ctx->list);
913 spin_unlock(&inode->i_lock);
914 put_nfs_open_context(ctx);
915 kfree_rcu(l_ctx, rcu_head);
917 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
920 * nfs_close_context - Common close_context() routine NFSv2/v3
921 * @ctx: pointer to context
922 * @is_sync: is this a synchronous close
924 * Ensure that the attributes are up to date if we're mounted
925 * with close-to-open semantics and we have cached data that will
926 * need to be revalidated on open.
928 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
930 struct nfs_inode *nfsi;
932 struct nfs_server *server;
934 if (!(ctx->mode & FMODE_WRITE))
938 inode = d_inode(ctx->dentry);
939 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
942 if (inode->i_mapping->nrpages == 0)
944 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
946 if (!list_empty(&nfsi->open_files))
948 server = NFS_SERVER(inode);
949 if (server->flags & NFS_MOUNT_NOCTO)
951 nfs_revalidate_inode(server, inode);
953 EXPORT_SYMBOL_GPL(nfs_close_context);
955 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
959 struct nfs_open_context *ctx;
960 const struct cred *cred = get_current_cred();
962 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
965 return ERR_PTR(-ENOMEM);
967 nfs_sb_active(dentry->d_sb);
968 ctx->dentry = dget(dentry);
975 ctx->flock_owner = (fl_owner_t)filp;
976 nfs_init_lock_context(&ctx->lock_context);
977 ctx->lock_context.open_context = ctx;
978 INIT_LIST_HEAD(&ctx->list);
979 ctx->mdsthreshold = NULL;
982 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
984 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
986 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
990 EXPORT_SYMBOL_GPL(get_nfs_open_context);
992 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
994 struct inode *inode = d_inode(ctx->dentry);
995 struct super_block *sb = ctx->dentry->d_sb;
997 if (!refcount_dec_and_test(&ctx->lock_context.count))
999 if (!list_empty(&ctx->list)) {
1000 spin_lock(&inode->i_lock);
1001 list_del_rcu(&ctx->list);
1002 spin_unlock(&inode->i_lock);
1005 NFS_PROTO(inode)->close_context(ctx, is_sync);
1006 put_cred(ctx->cred);
1008 nfs_sb_deactive(sb);
1009 put_rpccred(ctx->ll_cred);
1010 kfree(ctx->mdsthreshold);
1011 kfree_rcu(ctx, rcu_head);
1014 void put_nfs_open_context(struct nfs_open_context *ctx)
1016 __put_nfs_open_context(ctx, 0);
1018 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1020 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1022 __put_nfs_open_context(ctx, 1);
1026 * Ensure that mmap has a recent RPC credential for use when writing out
1029 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1031 struct inode *inode = d_inode(ctx->dentry);
1032 struct nfs_inode *nfsi = NFS_I(inode);
1034 spin_lock(&inode->i_lock);
1035 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1036 spin_unlock(&inode->i_lock);
1038 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1040 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1042 filp->private_data = get_nfs_open_context(ctx);
1043 if (list_empty(&ctx->list))
1044 nfs_inode_attach_open_context(ctx);
1046 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1049 * Given an inode, search for an open context with the desired characteristics
1051 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1053 struct nfs_inode *nfsi = NFS_I(inode);
1054 struct nfs_open_context *pos, *ctx = NULL;
1057 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1058 if (cred != NULL && pos->cred != cred)
1060 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1062 ctx = get_nfs_open_context(pos);
1070 void nfs_file_clear_open_context(struct file *filp)
1072 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1075 struct inode *inode = d_inode(ctx->dentry);
1078 * We fatal error on write before. Try to writeback
1082 invalidate_inode_pages2(inode->i_mapping);
1083 filp->private_data = NULL;
1084 put_nfs_open_context_sync(ctx);
1089 * These allocate and release file read/write context information.
1091 int nfs_open(struct inode *inode, struct file *filp)
1093 struct nfs_open_context *ctx;
1095 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1097 return PTR_ERR(ctx);
1098 nfs_file_set_open_context(filp, ctx);
1099 put_nfs_open_context(ctx);
1100 nfs_fscache_open_file(inode, filp);
1103 EXPORT_SYMBOL_GPL(nfs_open);
1106 * This function is called whenever some part of NFS notices that
1107 * the cached attributes have to be refreshed.
1110 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1112 int status = -ESTALE;
1113 struct nfs4_label *label = NULL;
1114 struct nfs_fattr *fattr = NULL;
1115 struct nfs_inode *nfsi = NFS_I(inode);
1117 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1118 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1120 trace_nfs_revalidate_inode_enter(inode);
1122 if (is_bad_inode(inode))
1124 if (NFS_STALE(inode))
1127 /* pNFS: Attributes aren't updated until we layoutcommit */
1128 if (S_ISREG(inode->i_mode)) {
1129 status = pnfs_sync_inode(inode, false);
1135 fattr = nfs_alloc_fattr();
1139 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1141 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1142 if (IS_ERR(label)) {
1143 status = PTR_ERR(label);
1147 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1150 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1152 (unsigned long long)NFS_FILEID(inode), status);
1153 if (status == -ESTALE) {
1154 nfs_zap_caches(inode);
1155 if (!S_ISDIR(inode->i_mode))
1156 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1161 status = nfs_refresh_inode(inode, fattr);
1163 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1165 (unsigned long long)NFS_FILEID(inode), status);
1169 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1170 nfs_zap_acl_cache(inode);
1172 nfs_setsecurity(inode, fattr, label);
1174 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1176 (unsigned long long)NFS_FILEID(inode));
1179 nfs4_label_free(label);
1181 nfs_free_fattr(fattr);
1182 trace_nfs_revalidate_inode_exit(inode, status);
1186 int nfs_attribute_cache_expired(struct inode *inode)
1188 if (nfs_have_delegated_attributes(inode))
1190 return nfs_attribute_timeout(inode);
1194 * nfs_revalidate_inode - Revalidate the inode attributes
1195 * @server: pointer to nfs_server struct
1196 * @inode: pointer to inode struct
1198 * Updates inode attribute information by retrieving the data from the server.
1200 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1202 if (!nfs_need_revalidate_inode(inode))
1203 return NFS_STALE(inode) ? -ESTALE : 0;
1204 return __nfs_revalidate_inode(server, inode);
1206 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1208 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1210 struct nfs_inode *nfsi = NFS_I(inode);
1213 if (mapping->nrpages != 0) {
1214 if (S_ISREG(inode->i_mode)) {
1215 ret = nfs_sync_mapping(mapping);
1219 ret = invalidate_inode_pages2(mapping);
1223 if (S_ISDIR(inode->i_mode)) {
1224 spin_lock(&inode->i_lock);
1225 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1226 spin_unlock(&inode->i_lock);
1228 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1229 nfs_fscache_wait_on_invalidate(inode);
1231 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1233 (unsigned long long)NFS_FILEID(inode));
1237 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1239 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1243 int nfs_revalidate_mapping_rcu(struct inode *inode)
1245 struct nfs_inode *nfsi = NFS_I(inode);
1246 unsigned long *bitlock = &nfsi->flags;
1249 if (IS_SWAPFILE(inode))
1251 if (nfs_mapping_need_revalidate_inode(inode)) {
1255 spin_lock(&inode->i_lock);
1256 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1257 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1259 spin_unlock(&inode->i_lock);
1265 * nfs_revalidate_mapping - Revalidate the pagecache
1266 * @inode: pointer to host inode
1267 * @mapping: pointer to mapping
1269 int nfs_revalidate_mapping(struct inode *inode,
1270 struct address_space *mapping)
1272 struct nfs_inode *nfsi = NFS_I(inode);
1273 unsigned long *bitlock = &nfsi->flags;
1276 /* swapfiles are not supposed to be shared. */
1277 if (IS_SWAPFILE(inode))
1280 if (nfs_mapping_need_revalidate_inode(inode)) {
1281 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1287 * We must clear NFS_INO_INVALID_DATA first to ensure that
1288 * invalidations that come in while we're shooting down the mappings
1289 * are respected. But, that leaves a race window where one revalidator
1290 * can clear the flag, and then another checks it before the mapping
1291 * gets invalidated. Fix that by serializing access to this part of
1294 * At the same time, we need to allow other tasks to see whether we
1295 * might be in the middle of invalidating the pages, so we only set
1296 * the bit lock here if it looks like we're going to be doing that.
1299 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1300 nfs_wait_bit_killable, TASK_KILLABLE);
1303 spin_lock(&inode->i_lock);
1304 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1305 spin_unlock(&inode->i_lock);
1308 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1310 spin_unlock(&inode->i_lock);
1314 set_bit(NFS_INO_INVALIDATING, bitlock);
1316 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1317 spin_unlock(&inode->i_lock);
1318 trace_nfs_invalidate_mapping_enter(inode);
1319 ret = nfs_invalidate_mapping(inode, mapping);
1320 trace_nfs_invalidate_mapping_exit(inode, ret);
1322 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1323 smp_mb__after_atomic();
1324 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1329 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1331 struct inode *inode = &nfsi->vfs_inode;
1333 if (!S_ISREG(inode->i_mode))
1335 if (list_empty(&nfsi->open_files))
1337 return inode_is_open_for_write(inode);
1340 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1342 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1345 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1349 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1350 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1351 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1352 inode_set_iversion_raw(inode, fattr->change_attr);
1353 if (S_ISDIR(inode->i_mode))
1354 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1356 /* If we have atomic WCC data, we may update some attributes */
1357 ts = timespec64_to_timespec(inode->i_ctime);
1358 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1359 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1360 && timespec_equal(&ts, &fattr->pre_ctime)) {
1361 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1364 ts = timespec64_to_timespec(inode->i_mtime);
1365 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1366 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1367 && timespec_equal(&ts, &fattr->pre_mtime)) {
1368 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1369 if (S_ISDIR(inode->i_mode))
1370 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1372 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1373 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1374 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1375 && !nfs_have_writebacks(inode)) {
1376 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1381 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1382 * @inode: pointer to inode
1383 * @fattr: updated attributes
1385 * Verifies the attribute cache. If we have just changed the attributes,
1386 * so that fattr carries weak cache consistency data, then it may
1387 * also update the ctime/mtime/change_attribute.
1389 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1391 struct nfs_inode *nfsi = NFS_I(inode);
1392 loff_t cur_size, new_isize;
1393 unsigned long invalid = 0;
1396 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1399 /* Has the inode gone and changed behind our back? */
1400 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1402 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1405 if (!nfs_file_has_buffered_writers(nfsi)) {
1406 /* Verify a few of the more important attributes */
1407 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1408 invalid |= NFS_INO_INVALID_CHANGE
1409 | NFS_INO_REVAL_PAGECACHE;
1411 ts = timespec64_to_timespec(inode->i_mtime);
1412 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime))
1413 invalid |= NFS_INO_INVALID_MTIME;
1415 ts = timespec64_to_timespec(inode->i_ctime);
1416 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime))
1417 invalid |= NFS_INO_INVALID_CTIME;
1419 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1420 cur_size = i_size_read(inode);
1421 new_isize = nfs_size_to_loff_t(fattr->size);
1422 if (cur_size != new_isize)
1423 invalid |= NFS_INO_INVALID_SIZE
1424 | NFS_INO_REVAL_PAGECACHE;
1428 /* Have any file permissions changed? */
1429 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1430 invalid |= NFS_INO_INVALID_ACCESS
1431 | NFS_INO_INVALID_ACL
1432 | NFS_INO_INVALID_OTHER;
1433 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1434 invalid |= NFS_INO_INVALID_ACCESS
1435 | NFS_INO_INVALID_ACL
1436 | NFS_INO_INVALID_OTHER;
1437 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1438 invalid |= NFS_INO_INVALID_ACCESS
1439 | NFS_INO_INVALID_ACL
1440 | NFS_INO_INVALID_OTHER;
1442 /* Has the link count changed? */
1443 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1444 invalid |= NFS_INO_INVALID_OTHER;
1446 ts = timespec64_to_timespec(inode->i_atime);
1447 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime))
1448 invalid |= NFS_INO_INVALID_ATIME;
1451 nfs_set_cache_invalid(inode, invalid);
1453 nfsi->read_cache_jiffies = fattr->time_start;
1457 static atomic_long_t nfs_attr_generation_counter;
1459 static unsigned long nfs_read_attr_generation_counter(void)
1461 return atomic_long_read(&nfs_attr_generation_counter);
1464 unsigned long nfs_inc_attr_generation_counter(void)
1466 return atomic_long_inc_return(&nfs_attr_generation_counter);
1468 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1470 void nfs_fattr_init(struct nfs_fattr *fattr)
1473 fattr->time_start = jiffies;
1474 fattr->gencount = nfs_inc_attr_generation_counter();
1475 fattr->owner_name = NULL;
1476 fattr->group_name = NULL;
1478 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1481 * nfs_fattr_set_barrier
1482 * @fattr: attributes
1484 * Used to set a barrier after an attribute was updated. This
1485 * barrier ensures that older attributes from RPC calls that may
1486 * have raced with our update cannot clobber these new values.
1487 * Note that you are still responsible for ensuring that other
1488 * operations which change the attribute on the server do not
1491 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1493 fattr->gencount = nfs_inc_attr_generation_counter();
1496 struct nfs_fattr *nfs_alloc_fattr(void)
1498 struct nfs_fattr *fattr;
1500 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1502 nfs_fattr_init(fattr);
1505 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1507 struct nfs_fh *nfs_alloc_fhandle(void)
1511 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1516 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1520 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1521 * in the same way that wireshark does
1525 * For debugging only.
1527 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1529 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1530 * not on the result */
1531 return nfs_fhandle_hash(fh);
1533 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1536 * _nfs_display_fhandle - display an NFS file handle on the console
1538 * @fh: file handle to display
1539 * @caption: display caption
1541 * For debugging only.
1543 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1547 if (fh == NULL || fh->size == 0) {
1548 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1552 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1553 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1554 for (i = 0; i < fh->size; i += 16) {
1555 __be32 *pos = (__be32 *)&fh->data[i];
1557 switch ((fh->size - i - 1) >> 2) {
1559 printk(KERN_DEFAULT " %08x\n",
1563 printk(KERN_DEFAULT " %08x %08x\n",
1564 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1567 printk(KERN_DEFAULT " %08x %08x %08x\n",
1568 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1569 be32_to_cpup(pos + 2));
1572 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1573 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1574 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1578 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1582 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1583 * @inode: pointer to inode
1584 * @fattr: attributes
1586 * Attempt to divine whether or not an RPC call reply carrying stale
1587 * attributes got scheduled after another call carrying updated ones.
1589 * To do so, the function first assumes that a more recent ctime means
1590 * that the attributes in fattr are newer, however it also attempt to
1591 * catch the case where ctime either didn't change, or went backwards
1592 * (if someone reset the clock on the server) by looking at whether
1593 * or not this RPC call was started after the inode was last updated.
1594 * Note also the check for wraparound of 'attr_gencount'
1596 * The function returns 'true' if it thinks the attributes in 'fattr' are
1597 * more recent than the ones cached in the inode.
1600 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1602 const struct nfs_inode *nfsi = NFS_I(inode);
1604 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1605 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1608 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1612 trace_nfs_refresh_inode_enter(inode);
1614 if (nfs_inode_attrs_need_update(inode, fattr))
1615 ret = nfs_update_inode(inode, fattr);
1617 ret = nfs_check_inode_attributes(inode, fattr);
1619 trace_nfs_refresh_inode_exit(inode, ret);
1624 * nfs_refresh_inode - try to update the inode attribute cache
1625 * @inode: pointer to inode
1626 * @fattr: updated attributes
1628 * Check that an RPC call that returned attributes has not overlapped with
1629 * other recent updates of the inode metadata, then decide whether it is
1630 * safe to do a full update of the inode attributes, or whether just to
1631 * call nfs_check_inode_attributes.
1633 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1637 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1639 spin_lock(&inode->i_lock);
1640 status = nfs_refresh_inode_locked(inode, fattr);
1641 spin_unlock(&inode->i_lock);
1645 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1647 static int nfs_post_op_update_inode_locked(struct inode *inode,
1648 struct nfs_fattr *fattr, unsigned int invalid)
1650 if (S_ISDIR(inode->i_mode))
1651 invalid |= NFS_INO_INVALID_DATA;
1652 nfs_set_cache_invalid(inode, invalid);
1653 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1655 return nfs_refresh_inode_locked(inode, fattr);
1659 * nfs_post_op_update_inode - try to update the inode attribute cache
1660 * @inode: pointer to inode
1661 * @fattr: updated attributes
1663 * After an operation that has changed the inode metadata, mark the
1664 * attribute cache as being invalid, then try to update it.
1666 * NB: if the server didn't return any post op attributes, this
1667 * function will force the retrieval of attributes before the next
1668 * NFS request. Thus it should be used only for operations that
1669 * are expected to change one or more attributes, to avoid
1670 * unnecessary NFS requests and trips through nfs_update_inode().
1672 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1676 spin_lock(&inode->i_lock);
1677 nfs_fattr_set_barrier(fattr);
1678 status = nfs_post_op_update_inode_locked(inode, fattr,
1679 NFS_INO_INVALID_CHANGE
1680 | NFS_INO_INVALID_CTIME
1681 | NFS_INO_REVAL_FORCED);
1682 spin_unlock(&inode->i_lock);
1686 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1689 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1690 * @inode: pointer to inode
1691 * @fattr: updated attributes
1693 * After an operation that has changed the inode metadata, mark the
1694 * attribute cache as being invalid, then try to update it. Fake up
1695 * weak cache consistency data, if none exist.
1697 * This function is mainly designed to be used by the ->write_done() functions.
1699 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1703 /* Don't do a WCC update if these attributes are already stale */
1704 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1705 !nfs_inode_attrs_need_update(inode, fattr)) {
1706 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1707 | NFS_ATTR_FATTR_PRESIZE
1708 | NFS_ATTR_FATTR_PREMTIME
1709 | NFS_ATTR_FATTR_PRECTIME);
1712 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1713 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1714 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1715 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1717 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1718 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1719 fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime);
1720 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1722 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1723 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1724 fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime);
1725 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1727 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1728 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1729 fattr->pre_size = i_size_read(inode);
1730 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1733 status = nfs_post_op_update_inode_locked(inode, fattr,
1734 NFS_INO_INVALID_CHANGE
1735 | NFS_INO_INVALID_CTIME
1736 | NFS_INO_INVALID_MTIME);
1741 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1742 * @inode: pointer to inode
1743 * @fattr: updated attributes
1745 * After an operation that has changed the inode metadata, mark the
1746 * attribute cache as being invalid, then try to update it. Fake up
1747 * weak cache consistency data, if none exist.
1749 * This function is mainly designed to be used by the ->write_done() functions.
1751 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1755 spin_lock(&inode->i_lock);
1756 nfs_fattr_set_barrier(fattr);
1757 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1758 spin_unlock(&inode->i_lock);
1761 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1764 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1765 struct nfs_fattr *fattr)
1767 bool ret1 = true, ret2 = true;
1769 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1770 ret1 = (nfsi->fileid == fattr->fileid);
1771 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1772 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1773 return ret1 || ret2;
1777 * Many nfs protocol calls return the new file attributes after
1778 * an operation. Here we update the inode to reflect the state
1779 * of the server's inode.
1781 * This is a bit tricky because we have to make sure all dirty pages
1782 * have been sent off to the server before calling invalidate_inode_pages.
1783 * To make sure no other process adds more write requests while we try
1784 * our best to flush them, we make them sleep during the attribute refresh.
1786 * A very similar scenario holds for the dir cache.
1788 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1790 struct nfs_server *server;
1791 struct nfs_inode *nfsi = NFS_I(inode);
1792 loff_t cur_isize, new_isize;
1793 unsigned long invalid = 0;
1794 unsigned long now = jiffies;
1795 unsigned long save_cache_validity;
1796 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1797 bool cache_revalidated = true;
1798 bool attr_changed = false;
1799 bool have_delegation;
1801 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1802 __func__, inode->i_sb->s_id, inode->i_ino,
1803 nfs_display_fhandle_hash(NFS_FH(inode)),
1804 atomic_read(&inode->i_count), fattr->valid);
1806 if (!nfs_fileid_valid(nfsi, fattr)) {
1807 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1808 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1809 NFS_SERVER(inode)->nfs_client->cl_hostname,
1810 inode->i_sb->s_id, (long long)nfsi->fileid,
1811 (long long)fattr->fileid);
1816 * Make sure the inode's type hasn't changed.
1818 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1820 * Big trouble! The inode has become a different object.
1822 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1823 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1827 server = NFS_SERVER(inode);
1828 /* Update the fsid? */
1829 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1830 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1831 !IS_AUTOMOUNT(inode))
1832 server->fsid = fattr->fsid;
1834 /* Save the delegation state before clearing cache_validity */
1835 have_delegation = nfs_have_delegated_attributes(inode);
1838 * Update the read time so we don't revalidate too often.
1840 nfsi->read_cache_jiffies = fattr->time_start;
1842 save_cache_validity = nfsi->cache_validity;
1843 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1844 | NFS_INO_INVALID_ATIME
1845 | NFS_INO_REVAL_FORCED
1846 | NFS_INO_REVAL_PAGECACHE);
1848 /* Do atomic weak cache consistency updates */
1849 nfs_wcc_update_inode(inode, fattr);
1851 if (pnfs_layoutcommit_outstanding(inode)) {
1852 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1853 cache_revalidated = false;
1856 /* More cache consistency checks */
1857 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1858 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1859 /* Could it be a race with writeback? */
1860 if (!(have_writers || have_delegation)) {
1861 invalid |= NFS_INO_INVALID_DATA
1862 | NFS_INO_INVALID_ACCESS
1863 | NFS_INO_INVALID_ACL;
1864 /* Force revalidate of all attributes */
1865 save_cache_validity |= NFS_INO_INVALID_CTIME
1866 | NFS_INO_INVALID_MTIME
1867 | NFS_INO_INVALID_SIZE
1868 | NFS_INO_INVALID_OTHER;
1869 if (S_ISDIR(inode->i_mode))
1870 nfs_force_lookup_revalidate(inode);
1871 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1875 inode_set_iversion_raw(inode, fattr->change_attr);
1876 attr_changed = true;
1879 nfsi->cache_validity |= save_cache_validity &
1880 (NFS_INO_INVALID_CHANGE
1881 | NFS_INO_REVAL_PAGECACHE
1882 | NFS_INO_REVAL_FORCED);
1883 cache_revalidated = false;
1886 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1887 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1888 } else if (server->caps & NFS_CAP_MTIME) {
1889 nfsi->cache_validity |= save_cache_validity &
1890 (NFS_INO_INVALID_MTIME
1891 | NFS_INO_REVAL_FORCED);
1892 cache_revalidated = false;
1895 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1896 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1897 } else if (server->caps & NFS_CAP_CTIME) {
1898 nfsi->cache_validity |= save_cache_validity &
1899 (NFS_INO_INVALID_CTIME
1900 | NFS_INO_REVAL_FORCED);
1901 cache_revalidated = false;
1904 /* Check if our cached file size is stale */
1905 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1906 new_isize = nfs_size_to_loff_t(fattr->size);
1907 cur_isize = i_size_read(inode);
1908 if (new_isize != cur_isize && !have_delegation) {
1909 /* Do we perhaps have any outstanding writes, or has
1910 * the file grown beyond our last write? */
1911 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1912 i_size_write(inode, new_isize);
1914 invalid |= NFS_INO_INVALID_DATA;
1915 attr_changed = true;
1917 dprintk("NFS: isize change on server for file %s/%ld "
1921 (long long)cur_isize,
1922 (long long)new_isize);
1925 nfsi->cache_validity |= save_cache_validity &
1926 (NFS_INO_INVALID_SIZE
1927 | NFS_INO_REVAL_PAGECACHE
1928 | NFS_INO_REVAL_FORCED);
1929 cache_revalidated = false;
1933 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1934 inode->i_atime = timespec_to_timespec64(fattr->atime);
1935 else if (server->caps & NFS_CAP_ATIME) {
1936 nfsi->cache_validity |= save_cache_validity &
1937 (NFS_INO_INVALID_ATIME
1938 | NFS_INO_REVAL_FORCED);
1939 cache_revalidated = false;
1942 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1943 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1944 umode_t newmode = inode->i_mode & S_IFMT;
1945 newmode |= fattr->mode & S_IALLUGO;
1946 inode->i_mode = newmode;
1947 invalid |= NFS_INO_INVALID_ACCESS
1948 | NFS_INO_INVALID_ACL;
1949 attr_changed = true;
1951 } else if (server->caps & NFS_CAP_MODE) {
1952 nfsi->cache_validity |= save_cache_validity &
1953 (NFS_INO_INVALID_OTHER
1954 | NFS_INO_REVAL_FORCED);
1955 cache_revalidated = false;
1958 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1959 if (!uid_eq(inode->i_uid, fattr->uid)) {
1960 invalid |= NFS_INO_INVALID_ACCESS
1961 | NFS_INO_INVALID_ACL;
1962 inode->i_uid = fattr->uid;
1963 attr_changed = true;
1965 } else if (server->caps & NFS_CAP_OWNER) {
1966 nfsi->cache_validity |= save_cache_validity &
1967 (NFS_INO_INVALID_OTHER
1968 | NFS_INO_REVAL_FORCED);
1969 cache_revalidated = false;
1972 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1973 if (!gid_eq(inode->i_gid, fattr->gid)) {
1974 invalid |= NFS_INO_INVALID_ACCESS
1975 | NFS_INO_INVALID_ACL;
1976 inode->i_gid = fattr->gid;
1977 attr_changed = true;
1979 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
1980 nfsi->cache_validity |= save_cache_validity &
1981 (NFS_INO_INVALID_OTHER
1982 | NFS_INO_REVAL_FORCED);
1983 cache_revalidated = false;
1986 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1987 if (inode->i_nlink != fattr->nlink) {
1988 if (S_ISDIR(inode->i_mode))
1989 invalid |= NFS_INO_INVALID_DATA;
1990 set_nlink(inode, fattr->nlink);
1991 attr_changed = true;
1993 } else if (server->caps & NFS_CAP_NLINK) {
1994 nfsi->cache_validity |= save_cache_validity &
1995 (NFS_INO_INVALID_OTHER
1996 | NFS_INO_REVAL_FORCED);
1997 cache_revalidated = false;
2000 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
2002 * report the blocks in 512byte units
2004 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
2005 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
2006 inode->i_blocks = fattr->du.nfs2.blocks;
2008 cache_revalidated = false;
2010 /* Update attrtimeo value if we're out of the unstable period */
2012 invalid &= ~NFS_INO_INVALID_ATTR;
2013 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2014 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2015 nfsi->attrtimeo_timestamp = now;
2016 /* Set barrier to be more recent than all outstanding updates */
2017 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2019 if (cache_revalidated) {
2020 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2021 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2022 nfsi->attrtimeo <<= 1;
2023 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2024 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2026 nfsi->attrtimeo_timestamp = now;
2028 /* Set the barrier to be more recent than this fattr */
2029 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2030 nfsi->attr_gencount = fattr->gencount;
2033 /* Don't invalidate the data if we were to blame */
2034 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2035 || S_ISLNK(inode->i_mode)))
2036 invalid &= ~NFS_INO_INVALID_DATA;
2037 nfs_set_cache_invalid(inode, invalid);
2042 * No need to worry about unhashing the dentry, as the
2043 * lookup validation will know that the inode is bad.
2044 * (But we fall through to invalidate the caches.)
2046 nfs_invalidate_inode(inode);
2050 struct inode *nfs_alloc_inode(struct super_block *sb)
2052 struct nfs_inode *nfsi;
2053 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2057 nfsi->cache_validity = 0UL;
2058 #if IS_ENABLED(CONFIG_NFS_V4)
2059 nfsi->nfs4_acl = NULL;
2060 #endif /* CONFIG_NFS_V4 */
2061 return &nfsi->vfs_inode;
2063 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2065 void nfs_free_inode(struct inode *inode)
2067 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2069 EXPORT_SYMBOL_GPL(nfs_free_inode);
2071 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2073 #if IS_ENABLED(CONFIG_NFS_V4)
2074 INIT_LIST_HEAD(&nfsi->open_states);
2075 nfsi->delegation = NULL;
2076 init_rwsem(&nfsi->rwsem);
2077 nfsi->layout = NULL;
2081 static void init_once(void *foo)
2083 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2085 inode_init_once(&nfsi->vfs_inode);
2086 INIT_LIST_HEAD(&nfsi->open_files);
2087 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2088 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2089 INIT_LIST_HEAD(&nfsi->commit_info.list);
2090 atomic_long_set(&nfsi->nrequests, 0);
2091 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2092 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2093 init_rwsem(&nfsi->rmdir_sem);
2094 mutex_init(&nfsi->commit_mutex);
2095 nfs4_init_once(nfsi);
2098 static int __init nfs_init_inodecache(void)
2100 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2101 sizeof(struct nfs_inode),
2102 0, (SLAB_RECLAIM_ACCOUNT|
2103 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2105 if (nfs_inode_cachep == NULL)
2111 static void nfs_destroy_inodecache(void)
2114 * Make sure all delayed rcu free inodes are flushed before we
2118 kmem_cache_destroy(nfs_inode_cachep);
2121 struct workqueue_struct *nfsiod_workqueue;
2122 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2125 * start up the nfsiod workqueue
2127 static int nfsiod_start(void)
2129 struct workqueue_struct *wq;
2130 dprintk("RPC: creating workqueue nfsiod\n");
2131 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2134 nfsiod_workqueue = wq;
2139 * Destroy the nfsiod workqueue
2141 static void nfsiod_stop(void)
2143 struct workqueue_struct *wq;
2145 wq = nfsiod_workqueue;
2148 nfsiod_workqueue = NULL;
2149 destroy_workqueue(wq);
2152 unsigned int nfs_net_id;
2153 EXPORT_SYMBOL_GPL(nfs_net_id);
2155 static int nfs_net_init(struct net *net)
2157 nfs_clients_init(net);
2158 return nfs_fs_proc_net_init(net);
2161 static void nfs_net_exit(struct net *net)
2163 struct nfs_net *nn = net_generic(net, nfs_net_id);
2165 nfs_fs_proc_net_exit(net);
2166 nfs_cleanup_cb_ident_idr(net);
2167 WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2168 WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2171 static struct pernet_operations nfs_net_ops = {
2172 .init = nfs_net_init,
2173 .exit = nfs_net_exit,
2175 .size = sizeof(struct nfs_net),
2181 static int __init init_nfs_fs(void)
2185 err = register_pernet_subsys(&nfs_net_ops);
2189 err = nfs_fscache_register();
2193 err = nfsiod_start();
2197 err = nfs_fs_proc_init();
2201 err = nfs_init_nfspagecache();
2205 err = nfs_init_inodecache();
2209 err = nfs_init_readpagecache();
2213 err = nfs_init_writepagecache();
2217 err = nfs_init_directcache();
2221 rpc_proc_register(&init_net, &nfs_rpcstat);
2223 err = register_nfs_fs();
2229 rpc_proc_unregister(&init_net, "nfs");
2230 nfs_destroy_directcache();
2232 nfs_destroy_writepagecache();
2234 nfs_destroy_readpagecache();
2236 nfs_destroy_inodecache();
2238 nfs_destroy_nfspagecache();
2244 nfs_fscache_unregister();
2246 unregister_pernet_subsys(&nfs_net_ops);
2251 static void __exit exit_nfs_fs(void)
2253 nfs_destroy_directcache();
2254 nfs_destroy_writepagecache();
2255 nfs_destroy_readpagecache();
2256 nfs_destroy_inodecache();
2257 nfs_destroy_nfspagecache();
2258 nfs_fscache_unregister();
2259 unregister_pernet_subsys(&nfs_net_ops);
2260 rpc_proc_unregister(&init_net, "nfs");
2261 unregister_nfs_fs();
2266 /* Not quite true; I just maintain it */
2267 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
2268 MODULE_LICENSE("GPL");
2269 module_param(enable_ino64, bool, 0644);
2271 module_init(init_nfs_fs)
2272 module_exit(exit_nfs_fs)