4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched/signal.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41 #include <linux/uaccess.h>
42 #include <linux/iversion.h>
46 #include "delegation.h"
56 #define NFSDBG_FACILITY NFSDBG_VFS
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
66 static struct kmem_cache * nfs_inode_cachep;
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
71 return nfs_fileid_to_ino_t(fattr->fileid);
74 static int nfs_wait_killable(int mode)
76 freezable_schedule_unsafe();
77 if (signal_pending_state(mode, current))
82 int nfs_wait_bit_killable(struct wait_bit_key *key, int mode)
84 return nfs_wait_killable(mode);
86 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
89 * nfs_compat_user_ino64 - returns the user-visible inode number
90 * @fileid: 64-bit fileid
92 * This function returns a 32-bit inode number if the boot parameter
93 * nfs.enable_ino64 is zero.
95 u64 nfs_compat_user_ino64(u64 fileid)
106 if (sizeof(ino) < sizeof(fileid))
107 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
111 int nfs_drop_inode(struct inode *inode)
113 return NFS_STALE(inode) || generic_drop_inode(inode);
115 EXPORT_SYMBOL_GPL(nfs_drop_inode);
117 void nfs_clear_inode(struct inode *inode)
120 * The following should never happen...
122 WARN_ON_ONCE(nfs_have_writebacks(inode));
123 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
124 nfs_zap_acl_cache(inode);
125 nfs_access_zap_cache(inode);
126 nfs_fscache_clear_inode(inode);
128 EXPORT_SYMBOL_GPL(nfs_clear_inode);
130 void nfs_evict_inode(struct inode *inode)
132 truncate_inode_pages_final(&inode->i_data);
134 nfs_clear_inode(inode);
137 int nfs_sync_inode(struct inode *inode)
139 inode_dio_wait(inode);
140 return nfs_wb_all(inode);
142 EXPORT_SYMBOL_GPL(nfs_sync_inode);
145 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
147 int nfs_sync_mapping(struct address_space *mapping)
151 if (mapping->nrpages != 0) {
152 unmap_mapping_range(mapping, 0, 0, 0);
153 ret = nfs_wb_all(mapping->host);
158 static int nfs_attribute_timeout(struct inode *inode)
160 struct nfs_inode *nfsi = NFS_I(inode);
162 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
165 static bool nfs_check_cache_invalid_delegated(struct inode *inode, unsigned long flags)
167 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
169 /* Special case for the pagecache or access cache */
170 if (flags == NFS_INO_REVAL_PAGECACHE &&
171 !(cache_validity & NFS_INO_REVAL_FORCED))
173 return (cache_validity & flags) != 0;
176 static bool nfs_check_cache_invalid_not_delegated(struct inode *inode, unsigned long flags)
178 unsigned long cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
180 if ((cache_validity & flags) != 0)
182 if (nfs_attribute_timeout(inode))
187 bool nfs_check_cache_invalid(struct inode *inode, unsigned long flags)
189 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
190 return nfs_check_cache_invalid_delegated(inode, flags);
192 return nfs_check_cache_invalid_not_delegated(inode, flags);
195 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
197 struct nfs_inode *nfsi = NFS_I(inode);
198 bool have_delegation = NFS_PROTO(inode)->have_delegation(inode, FMODE_READ);
200 if (have_delegation) {
201 if (!(flags & NFS_INO_REVAL_FORCED))
202 flags &= ~NFS_INO_INVALID_OTHER;
203 flags &= ~(NFS_INO_INVALID_CHANGE
204 | NFS_INO_INVALID_SIZE
205 | NFS_INO_REVAL_PAGECACHE);
208 if (inode->i_mapping->nrpages == 0)
209 flags &= ~NFS_INO_INVALID_DATA;
210 nfsi->cache_validity |= flags;
211 if (flags & NFS_INO_INVALID_DATA)
212 nfs_fscache_invalidate(inode);
216 * Invalidate the local caches
218 static void nfs_zap_caches_locked(struct inode *inode)
220 struct nfs_inode *nfsi = NFS_I(inode);
221 int mode = inode->i_mode;
223 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
225 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
226 nfsi->attrtimeo_timestamp = jiffies;
228 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
229 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
230 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
231 | NFS_INO_INVALID_DATA
232 | NFS_INO_INVALID_ACCESS
233 | NFS_INO_INVALID_ACL
234 | NFS_INO_REVAL_PAGECACHE);
236 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
237 | NFS_INO_INVALID_ACCESS
238 | NFS_INO_INVALID_ACL
239 | NFS_INO_REVAL_PAGECACHE);
240 nfs_zap_label_cache_locked(nfsi);
243 void nfs_zap_caches(struct inode *inode)
245 spin_lock(&inode->i_lock);
246 nfs_zap_caches_locked(inode);
247 spin_unlock(&inode->i_lock);
250 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
252 if (mapping->nrpages != 0) {
253 spin_lock(&inode->i_lock);
254 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
255 spin_unlock(&inode->i_lock);
259 void nfs_zap_acl_cache(struct inode *inode)
261 void (*clear_acl_cache)(struct inode *);
263 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
264 if (clear_acl_cache != NULL)
265 clear_acl_cache(inode);
266 spin_lock(&inode->i_lock);
267 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
268 spin_unlock(&inode->i_lock);
270 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
272 void nfs_invalidate_atime(struct inode *inode)
274 spin_lock(&inode->i_lock);
275 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
276 spin_unlock(&inode->i_lock);
278 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
281 * Invalidate, but do not unhash, the inode.
282 * NB: must be called with inode->i_lock held!
284 static void nfs_invalidate_inode(struct inode *inode)
286 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
287 nfs_zap_caches_locked(inode);
290 struct nfs_find_desc {
292 struct nfs_fattr *fattr;
296 * In NFSv3 we can have 64bit inode numbers. In order to support
297 * this, and re-exported directories (also seen in NFSv2)
298 * we are forced to allow 2 different inodes to have the same
302 nfs_find_actor(struct inode *inode, void *opaque)
304 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
305 struct nfs_fh *fh = desc->fh;
306 struct nfs_fattr *fattr = desc->fattr;
308 if (NFS_FILEID(inode) != fattr->fileid)
310 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
312 if (nfs_compare_fh(NFS_FH(inode), fh))
314 if (is_bad_inode(inode) || NFS_STALE(inode))
320 nfs_init_locked(struct inode *inode, void *opaque)
322 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
323 struct nfs_fattr *fattr = desc->fattr;
325 set_nfs_fileid(inode, fattr->fileid);
326 inode->i_mode = fattr->mode;
327 nfs_copy_fh(NFS_FH(inode), desc->fh);
331 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
332 static void nfs_clear_label_invalid(struct inode *inode)
334 spin_lock(&inode->i_lock);
335 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
336 spin_unlock(&inode->i_lock);
339 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
340 struct nfs4_label *label)
347 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
348 error = security_inode_notifysecctx(inode, label->label,
351 printk(KERN_ERR "%s() %s %d "
352 "security_inode_notifysecctx() %d\n",
354 (char *)label->label,
356 nfs_clear_label_invalid(inode);
360 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
362 struct nfs4_label *label = NULL;
363 int minor_version = server->nfs_client->cl_minorversion;
365 if (minor_version < 2)
368 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
371 label = kzalloc(sizeof(struct nfs4_label), flags);
373 return ERR_PTR(-ENOMEM);
375 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
376 if (label->label == NULL) {
378 return ERR_PTR(-ENOMEM);
380 label->len = NFS4_MAXLABELLEN;
384 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
386 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
387 struct nfs4_label *label)
391 EXPORT_SYMBOL_GPL(nfs_setsecurity);
393 /* Search for inode identified by fh, fileid and i_mode in inode cache. */
395 nfs_ilookup(struct super_block *sb, struct nfs_fattr *fattr, struct nfs_fh *fh)
397 struct nfs_find_desc desc = {
404 if (!(fattr->valid & NFS_ATTR_FATTR_FILEID) ||
405 !(fattr->valid & NFS_ATTR_FATTR_TYPE))
408 hash = nfs_fattr_to_ino_t(fattr);
409 inode = ilookup5(sb, hash, nfs_find_actor, &desc);
411 dprintk("%s: returning %p\n", __func__, inode);
416 * This is our front-end to iget that looks up inodes by file handle
417 * instead of inode number.
420 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
422 struct nfs_find_desc desc = {
426 struct inode *inode = ERR_PTR(-ENOENT);
429 nfs_attr_check_mountpoint(sb, fattr);
431 if (nfs_attr_use_mounted_on_fileid(fattr))
432 fattr->fileid = fattr->mounted_on_fileid;
433 else if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
435 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
438 hash = nfs_fattr_to_ino_t(fattr);
440 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
442 inode = ERR_PTR(-ENOMEM);
446 if (inode->i_state & I_NEW) {
447 struct nfs_inode *nfsi = NFS_I(inode);
448 unsigned long now = jiffies;
450 /* We set i_ino for the few things that still rely on it,
454 /* We can't support update_atime(), since the server will reset it */
455 inode->i_flags |= S_NOATIME|S_NOCMTIME;
456 inode->i_mode = fattr->mode;
457 nfsi->cache_validity = 0;
458 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
459 && nfs_server_capable(inode, NFS_CAP_MODE))
460 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
461 /* Why so? Because we want revalidate for devices/FIFOs, and
462 * that's precisely what we have in nfs_file_inode_operations.
464 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
465 if (S_ISREG(inode->i_mode)) {
466 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
467 inode->i_data.a_ops = &nfs_file_aops;
468 } else if (S_ISDIR(inode->i_mode)) {
469 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
470 inode->i_fop = &nfs_dir_operations;
471 inode->i_data.a_ops = &nfs_dir_aops;
472 /* Deal with crossing mountpoints */
473 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
474 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
475 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
476 inode->i_op = &nfs_referral_inode_operations;
478 inode->i_op = &nfs_mountpoint_inode_operations;
480 inode->i_flags |= S_AUTOMOUNT;
482 } else if (S_ISLNK(inode->i_mode)) {
483 inode->i_op = &nfs_symlink_inode_operations;
484 inode_nohighmem(inode);
486 init_special_inode(inode, inode->i_mode, fattr->rdev);
488 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
489 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
490 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
491 inode_set_iversion_raw(inode, 0);
494 inode->i_uid = make_kuid(&init_user_ns, -2);
495 inode->i_gid = make_kgid(&init_user_ns, -2);
497 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
501 nfsi->read_cache_jiffies = fattr->time_start;
502 nfsi->attr_gencount = fattr->gencount;
503 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
504 inode->i_atime = timespec_to_timespec64(fattr->atime);
505 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
506 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
507 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
508 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
509 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
510 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
511 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
512 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
513 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
514 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CTIME);
515 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
516 inode_set_iversion_raw(inode, fattr->change_attr);
518 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
519 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
520 inode->i_size = nfs_size_to_loff_t(fattr->size);
522 nfs_set_cache_invalid(inode, NFS_INO_INVALID_SIZE);
523 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
524 set_nlink(inode, fattr->nlink);
525 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
526 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
527 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
528 inode->i_uid = fattr->uid;
529 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
530 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
531 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
532 inode->i_gid = fattr->gid;
533 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
534 nfs_set_cache_invalid(inode, NFS_INO_INVALID_OTHER);
535 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
536 inode->i_blocks = fattr->du.nfs2.blocks;
537 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
539 * report the blocks in 512byte units
541 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
544 if (nfsi->cache_validity != 0)
545 nfsi->cache_validity |= NFS_INO_REVAL_FORCED;
547 nfs_setsecurity(inode, fattr, label);
549 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
550 nfsi->attrtimeo_timestamp = now;
551 nfsi->access_cache = RB_ROOT;
553 nfs_fscache_init_inode(inode);
555 unlock_new_inode(inode);
557 int err = nfs_refresh_inode(inode, fattr);
560 inode = ERR_PTR(err);
564 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
566 (unsigned long long)NFS_FILEID(inode),
567 nfs_display_fhandle_hash(fh),
568 atomic_read(&inode->i_count));
574 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
577 EXPORT_SYMBOL_GPL(nfs_fhget);
579 #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)
582 nfs_setattr(struct dentry *dentry, struct iattr *attr)
584 struct inode *inode = d_inode(dentry);
585 struct nfs_fattr *fattr;
588 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
590 /* skip mode change if it's just for clearing setuid/setgid */
591 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
592 attr->ia_valid &= ~ATTR_MODE;
594 if (attr->ia_valid & ATTR_SIZE) {
595 BUG_ON(!S_ISREG(inode->i_mode));
597 error = inode_newsize_ok(inode, attr->ia_size);
601 if (attr->ia_size == i_size_read(inode))
602 attr->ia_valid &= ~ATTR_SIZE;
605 /* Optimization: if the end result is no change, don't RPC */
606 attr->ia_valid &= NFS_VALID_ATTRS;
607 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
610 trace_nfs_setattr_enter(inode);
612 /* Write all dirty data */
613 if (S_ISREG(inode->i_mode))
614 nfs_sync_inode(inode);
616 fattr = nfs_alloc_fattr();
622 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
624 error = nfs_refresh_inode(inode, fattr);
625 nfs_free_fattr(fattr);
627 trace_nfs_setattr_exit(inode, error);
630 EXPORT_SYMBOL_GPL(nfs_setattr);
633 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
634 * @inode: inode of the file used
635 * @offset: file offset to start truncating
637 * This is a copy of the common vmtruncate, but with the locking
638 * corrected to take into account the fact that NFS requires
639 * inode->i_size to be updated under the inode->i_lock.
640 * Note: must be called with inode->i_lock held!
642 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
646 err = inode_newsize_ok(inode, offset);
650 i_size_write(inode, offset);
653 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
654 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_SIZE;
656 spin_unlock(&inode->i_lock);
657 truncate_pagecache(inode, offset);
658 spin_lock(&inode->i_lock);
664 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
665 * @inode: pointer to struct inode
666 * @attr: pointer to struct iattr
667 * @fattr: pointer to struct nfs_fattr
669 * Note: we do this in the *proc.c in order to ensure that
670 * it works for things like exclusive creates too.
672 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr,
673 struct nfs_fattr *fattr)
675 /* Barrier: bump the attribute generation count. */
676 nfs_fattr_set_barrier(fattr);
678 spin_lock(&inode->i_lock);
679 NFS_I(inode)->attr_gencount = fattr->gencount;
680 if ((attr->ia_valid & ATTR_SIZE) != 0) {
681 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
682 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
683 nfs_vmtruncate(inode, attr->ia_size);
685 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
686 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_CTIME;
687 if ((attr->ia_valid & ATTR_MODE) != 0) {
688 int mode = attr->ia_mode & S_IALLUGO;
689 mode |= inode->i_mode & ~S_IALLUGO;
690 inode->i_mode = mode;
692 if ((attr->ia_valid & ATTR_UID) != 0)
693 inode->i_uid = attr->ia_uid;
694 if ((attr->ia_valid & ATTR_GID) != 0)
695 inode->i_gid = attr->ia_gid;
696 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
697 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
699 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
700 | NFS_INO_INVALID_CTIME);
701 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
702 | NFS_INO_INVALID_ACL);
704 if (attr->ia_valid & (ATTR_ATIME_SET|ATTR_ATIME)) {
705 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_ATIME
706 | NFS_INO_INVALID_CTIME);
707 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
708 inode->i_atime = timespec_to_timespec64(fattr->atime);
709 else if (attr->ia_valid & ATTR_ATIME_SET)
710 inode->i_atime = attr->ia_atime;
712 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
714 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
715 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
717 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
718 | NFS_INO_INVALID_CTIME);
720 if (attr->ia_valid & (ATTR_MTIME_SET|ATTR_MTIME)) {
721 NFS_I(inode)->cache_validity &= ~(NFS_INO_INVALID_MTIME
722 | NFS_INO_INVALID_CTIME);
723 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
724 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
725 else if (attr->ia_valid & ATTR_MTIME_SET)
726 inode->i_mtime = attr->ia_mtime;
728 nfs_set_cache_invalid(inode, NFS_INO_INVALID_MTIME);
730 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
731 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
733 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE
734 | NFS_INO_INVALID_CTIME);
737 nfs_update_inode(inode, fattr);
738 spin_unlock(&inode->i_lock);
740 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
742 static void nfs_readdirplus_parent_cache_miss(struct dentry *dentry)
744 struct dentry *parent;
746 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
748 parent = dget_parent(dentry);
749 nfs_force_use_readdirplus(d_inode(parent));
753 static void nfs_readdirplus_parent_cache_hit(struct dentry *dentry)
755 struct dentry *parent;
757 if (!nfs_server_capable(d_inode(dentry), NFS_CAP_READDIRPLUS))
759 parent = dget_parent(dentry);
760 nfs_advise_use_readdirplus(d_inode(parent));
764 static bool nfs_need_revalidate_inode(struct inode *inode)
766 if (NFS_I(inode)->cache_validity &
767 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
769 if (nfs_attribute_cache_expired(inode))
774 int nfs_getattr(const struct path *path, struct kstat *stat,
775 u32 request_mask, unsigned int query_flags)
777 struct inode *inode = d_inode(path->dentry);
778 struct nfs_server *server = NFS_SERVER(inode);
779 unsigned long cache_validity;
781 bool force_sync = query_flags & AT_STATX_FORCE_SYNC;
782 bool do_update = false;
784 trace_nfs_getattr_enter(inode);
786 if ((query_flags & AT_STATX_DONT_SYNC) && !force_sync)
789 /* Flush out writes to the server in order to update c/mtime. */
790 if ((request_mask & (STATX_CTIME|STATX_MTIME)) &&
791 S_ISREG(inode->i_mode)) {
792 err = filemap_write_and_wait(inode->i_mapping);
798 * We may force a getattr if the user cares about atime.
800 * Note that we only have to check the vfsmount flags here:
801 * - NFS always sets S_NOATIME by so checking it would give a
803 * - NFS never sets SB_NOATIME or SB_NODIRATIME so there is
804 * no point in checking those.
806 if ((path->mnt->mnt_flags & MNT_NOATIME) ||
807 ((path->mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
808 request_mask &= ~STATX_ATIME;
810 /* Is the user requesting attributes that might need revalidation? */
811 if (!(request_mask & (STATX_MODE|STATX_NLINK|STATX_ATIME|STATX_CTIME|
812 STATX_MTIME|STATX_UID|STATX_GID|
813 STATX_SIZE|STATX_BLOCKS)))
814 goto out_no_revalidate;
816 /* Check whether the cached attributes are stale */
817 do_update |= force_sync || nfs_attribute_cache_expired(inode);
818 cache_validity = READ_ONCE(NFS_I(inode)->cache_validity);
819 do_update |= cache_validity &
820 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL);
821 if (request_mask & STATX_ATIME)
822 do_update |= cache_validity & NFS_INO_INVALID_ATIME;
823 if (request_mask & (STATX_CTIME|STATX_MTIME))
824 do_update |= cache_validity & NFS_INO_REVAL_PAGECACHE;
826 /* Update the attribute cache */
827 if (!(server->flags & NFS_MOUNT_NOAC))
828 nfs_readdirplus_parent_cache_miss(path->dentry);
830 nfs_readdirplus_parent_cache_hit(path->dentry);
831 err = __nfs_revalidate_inode(server, inode);
835 nfs_readdirplus_parent_cache_hit(path->dentry);
837 /* Only return attributes that were revalidated. */
838 stat->result_mask &= request_mask;
840 generic_fillattr(inode, stat);
841 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
842 if (S_ISDIR(inode->i_mode))
843 stat->blksize = NFS_SERVER(inode)->dtsize;
845 trace_nfs_getattr_exit(inode, err);
848 EXPORT_SYMBOL_GPL(nfs_getattr);
850 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
852 refcount_set(&l_ctx->count, 1);
853 l_ctx->lockowner = current->files;
854 INIT_LIST_HEAD(&l_ctx->list);
855 atomic_set(&l_ctx->io_count, 0);
858 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
860 struct nfs_lock_context *pos;
862 list_for_each_entry_rcu(pos, &ctx->lock_context.list, list) {
863 if (pos->lockowner != current->files)
865 if (refcount_inc_not_zero(&pos->count))
871 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
873 struct nfs_lock_context *res, *new = NULL;
874 struct inode *inode = d_inode(ctx->dentry);
877 res = __nfs_find_lock_context(ctx);
880 new = kmalloc(sizeof(*new), GFP_KERNEL);
882 return ERR_PTR(-ENOMEM);
883 nfs_init_lock_context(new);
884 spin_lock(&inode->i_lock);
885 res = __nfs_find_lock_context(ctx);
887 list_add_tail_rcu(&new->list, &ctx->lock_context.list);
888 new->open_context = ctx;
892 spin_unlock(&inode->i_lock);
897 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
899 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
901 struct nfs_open_context *ctx = l_ctx->open_context;
902 struct inode *inode = d_inode(ctx->dentry);
904 if (!refcount_dec_and_lock(&l_ctx->count, &inode->i_lock))
906 list_del_rcu(&l_ctx->list);
907 spin_unlock(&inode->i_lock);
908 kfree_rcu(l_ctx, rcu_head);
910 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
913 * nfs_close_context - Common close_context() routine NFSv2/v3
914 * @ctx: pointer to context
915 * @is_sync: is this a synchronous close
917 * Ensure that the attributes are up to date if we're mounted
918 * with close-to-open semantics and we have cached data that will
919 * need to be revalidated on open.
921 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
923 struct nfs_inode *nfsi;
925 struct nfs_server *server;
927 if (!(ctx->mode & FMODE_WRITE))
931 inode = d_inode(ctx->dentry);
932 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
935 if (inode->i_mapping->nrpages == 0)
937 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
939 if (!list_empty(&nfsi->open_files))
941 server = NFS_SERVER(inode);
942 if (server->flags & NFS_MOUNT_NOCTO)
944 nfs_revalidate_inode(server, inode);
946 EXPORT_SYMBOL_GPL(nfs_close_context);
948 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry,
952 struct nfs_open_context *ctx;
953 const struct cred *cred = get_current_cred();
955 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
958 return ERR_PTR(-ENOMEM);
960 nfs_sb_active(dentry->d_sb);
961 ctx->dentry = dget(dentry);
968 ctx->flock_owner = (fl_owner_t)filp;
969 nfs_init_lock_context(&ctx->lock_context);
970 ctx->lock_context.open_context = ctx;
971 INIT_LIST_HEAD(&ctx->list);
972 ctx->mdsthreshold = NULL;
975 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
977 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
979 if (ctx != NULL && refcount_inc_not_zero(&ctx->lock_context.count))
983 EXPORT_SYMBOL_GPL(get_nfs_open_context);
985 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
987 struct inode *inode = d_inode(ctx->dentry);
988 struct super_block *sb = ctx->dentry->d_sb;
990 if (!refcount_dec_and_test(&ctx->lock_context.count))
992 if (!list_empty(&ctx->list)) {
993 spin_lock(&inode->i_lock);
994 list_del_rcu(&ctx->list);
995 spin_unlock(&inode->i_lock);
998 NFS_PROTO(inode)->close_context(ctx, is_sync);
1001 nfs_sb_deactive(sb);
1002 put_rpccred(ctx->ll_cred);
1003 kfree(ctx->mdsthreshold);
1004 kfree_rcu(ctx, rcu_head);
1007 void put_nfs_open_context(struct nfs_open_context *ctx)
1009 __put_nfs_open_context(ctx, 0);
1011 EXPORT_SYMBOL_GPL(put_nfs_open_context);
1013 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
1015 __put_nfs_open_context(ctx, 1);
1019 * Ensure that mmap has a recent RPC credential for use when writing out
1022 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
1024 struct inode *inode = d_inode(ctx->dentry);
1025 struct nfs_inode *nfsi = NFS_I(inode);
1027 spin_lock(&inode->i_lock);
1028 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
1029 spin_unlock(&inode->i_lock);
1031 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
1033 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1035 filp->private_data = get_nfs_open_context(ctx);
1036 if (list_empty(&ctx->list))
1037 nfs_inode_attach_open_context(ctx);
1039 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
1042 * Given an inode, search for an open context with the desired characteristics
1044 struct nfs_open_context *nfs_find_open_context(struct inode *inode, const struct cred *cred, fmode_t mode)
1046 struct nfs_inode *nfsi = NFS_I(inode);
1047 struct nfs_open_context *pos, *ctx = NULL;
1050 list_for_each_entry_rcu(pos, &nfsi->open_files, list) {
1051 if (cred != NULL && pos->cred != cred)
1053 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
1055 ctx = get_nfs_open_context(pos);
1063 void nfs_file_clear_open_context(struct file *filp)
1065 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1068 struct inode *inode = d_inode(ctx->dentry);
1071 * We fatal error on write before. Try to writeback
1075 invalidate_inode_pages2(inode->i_mapping);
1076 filp->private_data = NULL;
1077 put_nfs_open_context_sync(ctx);
1082 * These allocate and release file read/write context information.
1084 int nfs_open(struct inode *inode, struct file *filp)
1086 struct nfs_open_context *ctx;
1088 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
1090 return PTR_ERR(ctx);
1091 nfs_file_set_open_context(filp, ctx);
1092 put_nfs_open_context(ctx);
1093 nfs_fscache_open_file(inode, filp);
1098 * This function is called whenever some part of NFS notices that
1099 * the cached attributes have to be refreshed.
1102 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1104 int status = -ESTALE;
1105 struct nfs4_label *label = NULL;
1106 struct nfs_fattr *fattr = NULL;
1107 struct nfs_inode *nfsi = NFS_I(inode);
1109 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
1110 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
1112 trace_nfs_revalidate_inode_enter(inode);
1114 if (is_bad_inode(inode))
1116 if (NFS_STALE(inode))
1119 /* pNFS: Attributes aren't updated until we layoutcommit */
1120 if (S_ISREG(inode->i_mode)) {
1121 status = pnfs_sync_inode(inode, false);
1127 fattr = nfs_alloc_fattr();
1131 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1133 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1134 if (IS_ERR(label)) {
1135 status = PTR_ERR(label);
1139 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr,
1142 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
1144 (unsigned long long)NFS_FILEID(inode), status);
1145 if (status == -ESTALE) {
1146 nfs_zap_caches(inode);
1147 if (!S_ISDIR(inode->i_mode))
1148 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1153 status = nfs_refresh_inode(inode, fattr);
1155 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
1157 (unsigned long long)NFS_FILEID(inode), status);
1161 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
1162 nfs_zap_acl_cache(inode);
1164 nfs_setsecurity(inode, fattr, label);
1166 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
1168 (unsigned long long)NFS_FILEID(inode));
1171 nfs4_label_free(label);
1173 nfs_free_fattr(fattr);
1174 trace_nfs_revalidate_inode_exit(inode, status);
1178 int nfs_attribute_cache_expired(struct inode *inode)
1180 if (nfs_have_delegated_attributes(inode))
1182 return nfs_attribute_timeout(inode);
1186 * nfs_revalidate_inode - Revalidate the inode attributes
1187 * @server - pointer to nfs_server struct
1188 * @inode - pointer to inode struct
1190 * Updates inode attribute information by retrieving the data from the server.
1192 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1194 if (!nfs_need_revalidate_inode(inode))
1195 return NFS_STALE(inode) ? -ESTALE : 0;
1196 return __nfs_revalidate_inode(server, inode);
1198 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1200 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1202 struct nfs_inode *nfsi = NFS_I(inode);
1205 if (mapping->nrpages != 0) {
1206 if (S_ISREG(inode->i_mode)) {
1207 ret = nfs_sync_mapping(mapping);
1211 ret = invalidate_inode_pages2(mapping);
1215 if (S_ISDIR(inode->i_mode)) {
1216 spin_lock(&inode->i_lock);
1217 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1218 spin_unlock(&inode->i_lock);
1220 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1221 nfs_fscache_wait_on_invalidate(inode);
1223 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1225 (unsigned long long)NFS_FILEID(inode));
1229 bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1231 return nfs_check_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE) ||
1235 int nfs_revalidate_mapping_rcu(struct inode *inode)
1237 struct nfs_inode *nfsi = NFS_I(inode);
1238 unsigned long *bitlock = &nfsi->flags;
1241 if (IS_SWAPFILE(inode))
1243 if (nfs_mapping_need_revalidate_inode(inode)) {
1247 spin_lock(&inode->i_lock);
1248 if (test_bit(NFS_INO_INVALIDATING, bitlock) ||
1249 (nfsi->cache_validity & NFS_INO_INVALID_DATA))
1251 spin_unlock(&inode->i_lock);
1257 * nfs_revalidate_mapping - Revalidate the pagecache
1258 * @inode - pointer to host inode
1259 * @mapping - pointer to mapping
1261 int nfs_revalidate_mapping(struct inode *inode,
1262 struct address_space *mapping)
1264 struct nfs_inode *nfsi = NFS_I(inode);
1265 unsigned long *bitlock = &nfsi->flags;
1268 /* swapfiles are not supposed to be shared. */
1269 if (IS_SWAPFILE(inode))
1272 if (nfs_mapping_need_revalidate_inode(inode)) {
1273 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1279 * We must clear NFS_INO_INVALID_DATA first to ensure that
1280 * invalidations that come in while we're shooting down the mappings
1281 * are respected. But, that leaves a race window where one revalidator
1282 * can clear the flag, and then another checks it before the mapping
1283 * gets invalidated. Fix that by serializing access to this part of
1286 * At the same time, we need to allow other tasks to see whether we
1287 * might be in the middle of invalidating the pages, so we only set
1288 * the bit lock here if it looks like we're going to be doing that.
1291 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1292 nfs_wait_bit_killable, TASK_KILLABLE);
1295 spin_lock(&inode->i_lock);
1296 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1297 spin_unlock(&inode->i_lock);
1300 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1302 spin_unlock(&inode->i_lock);
1306 set_bit(NFS_INO_INVALIDATING, bitlock);
1308 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1309 spin_unlock(&inode->i_lock);
1310 trace_nfs_invalidate_mapping_enter(inode);
1311 ret = nfs_invalidate_mapping(inode, mapping);
1312 trace_nfs_invalidate_mapping_exit(inode, ret);
1314 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1315 smp_mb__after_atomic();
1316 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1321 static bool nfs_file_has_writers(struct nfs_inode *nfsi)
1323 struct inode *inode = &nfsi->vfs_inode;
1325 if (!S_ISREG(inode->i_mode))
1327 if (list_empty(&nfsi->open_files))
1329 return inode_is_open_for_write(inode);
1332 static bool nfs_file_has_buffered_writers(struct nfs_inode *nfsi)
1334 return nfs_file_has_writers(nfsi) && nfs_file_io_is_buffered(nfsi);
1337 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1341 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1342 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1343 && inode_eq_iversion_raw(inode, fattr->pre_change_attr)) {
1344 inode_set_iversion_raw(inode, fattr->change_attr);
1345 if (S_ISDIR(inode->i_mode))
1346 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1348 /* If we have atomic WCC data, we may update some attributes */
1349 ts = timespec64_to_timespec(inode->i_ctime);
1350 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1351 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1352 && timespec_equal(&ts, &fattr->pre_ctime)) {
1353 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1356 ts = timespec64_to_timespec(inode->i_mtime);
1357 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1358 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1359 && timespec_equal(&ts, &fattr->pre_mtime)) {
1360 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1361 if (S_ISDIR(inode->i_mode))
1362 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1364 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1365 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1366 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1367 && !nfs_have_writebacks(inode)) {
1368 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1373 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1374 * @inode - pointer to inode
1375 * @fattr - updated attributes
1377 * Verifies the attribute cache. If we have just changed the attributes,
1378 * so that fattr carries weak cache consistency data, then it may
1379 * also update the ctime/mtime/change_attribute.
1381 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1383 struct nfs_inode *nfsi = NFS_I(inode);
1384 loff_t cur_size, new_isize;
1385 unsigned long invalid = 0;
1388 if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1391 /* Has the inode gone and changed behind our back? */
1392 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1394 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1397 if (!nfs_file_has_buffered_writers(nfsi)) {
1398 /* Verify a few of the more important attributes */
1399 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 && !inode_eq_iversion_raw(inode, fattr->change_attr))
1400 invalid |= NFS_INO_INVALID_CHANGE
1401 | NFS_INO_REVAL_PAGECACHE;
1403 ts = timespec64_to_timespec(inode->i_mtime);
1404 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&ts, &fattr->mtime))
1405 invalid |= NFS_INO_INVALID_MTIME;
1407 ts = timespec64_to_timespec(inode->i_ctime);
1408 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) && !timespec_equal(&ts, &fattr->ctime))
1409 invalid |= NFS_INO_INVALID_CTIME;
1411 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1412 cur_size = i_size_read(inode);
1413 new_isize = nfs_size_to_loff_t(fattr->size);
1414 if (cur_size != new_isize)
1415 invalid |= NFS_INO_INVALID_SIZE
1416 | NFS_INO_REVAL_PAGECACHE;
1420 /* Have any file permissions changed? */
1421 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1422 invalid |= NFS_INO_INVALID_ACCESS
1423 | NFS_INO_INVALID_ACL
1424 | NFS_INO_INVALID_OTHER;
1425 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1426 invalid |= NFS_INO_INVALID_ACCESS
1427 | NFS_INO_INVALID_ACL
1428 | NFS_INO_INVALID_OTHER;
1429 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1430 invalid |= NFS_INO_INVALID_ACCESS
1431 | NFS_INO_INVALID_ACL
1432 | NFS_INO_INVALID_OTHER;
1434 /* Has the link count changed? */
1435 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1436 invalid |= NFS_INO_INVALID_OTHER;
1438 ts = timespec64_to_timespec(inode->i_atime);
1439 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&ts, &fattr->atime))
1440 invalid |= NFS_INO_INVALID_ATIME;
1443 nfs_set_cache_invalid(inode, invalid);
1445 nfsi->read_cache_jiffies = fattr->time_start;
1449 static atomic_long_t nfs_attr_generation_counter;
1451 static unsigned long nfs_read_attr_generation_counter(void)
1453 return atomic_long_read(&nfs_attr_generation_counter);
1456 unsigned long nfs_inc_attr_generation_counter(void)
1458 return atomic_long_inc_return(&nfs_attr_generation_counter);
1460 EXPORT_SYMBOL_GPL(nfs_inc_attr_generation_counter);
1462 void nfs_fattr_init(struct nfs_fattr *fattr)
1465 fattr->time_start = jiffies;
1466 fattr->gencount = nfs_inc_attr_generation_counter();
1467 fattr->owner_name = NULL;
1468 fattr->group_name = NULL;
1470 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1473 * nfs_fattr_set_barrier
1474 * @fattr: attributes
1476 * Used to set a barrier after an attribute was updated. This
1477 * barrier ensures that older attributes from RPC calls that may
1478 * have raced with our update cannot clobber these new values.
1479 * Note that you are still responsible for ensuring that other
1480 * operations which change the attribute on the server do not
1483 void nfs_fattr_set_barrier(struct nfs_fattr *fattr)
1485 fattr->gencount = nfs_inc_attr_generation_counter();
1488 struct nfs_fattr *nfs_alloc_fattr(void)
1490 struct nfs_fattr *fattr;
1492 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1494 nfs_fattr_init(fattr);
1497 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1499 struct nfs_fh *nfs_alloc_fhandle(void)
1503 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1508 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1512 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1513 * in the same way that wireshark does
1517 * For debugging only.
1519 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1521 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1522 * not on the result */
1523 return nfs_fhandle_hash(fh);
1525 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1528 * _nfs_display_fhandle - display an NFS file handle on the console
1530 * @fh: file handle to display
1531 * @caption: display caption
1533 * For debugging only.
1535 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1539 if (fh == NULL || fh->size == 0) {
1540 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1544 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1545 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1546 for (i = 0; i < fh->size; i += 16) {
1547 __be32 *pos = (__be32 *)&fh->data[i];
1549 switch ((fh->size - i - 1) >> 2) {
1551 printk(KERN_DEFAULT " %08x\n",
1555 printk(KERN_DEFAULT " %08x %08x\n",
1556 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1559 printk(KERN_DEFAULT " %08x %08x %08x\n",
1560 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1561 be32_to_cpup(pos + 2));
1564 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1565 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1566 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1570 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1574 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1575 * @inode - pointer to inode
1576 * @fattr - attributes
1578 * Attempt to divine whether or not an RPC call reply carrying stale
1579 * attributes got scheduled after another call carrying updated ones.
1581 * To do so, the function first assumes that a more recent ctime means
1582 * that the attributes in fattr are newer, however it also attempt to
1583 * catch the case where ctime either didn't change, or went backwards
1584 * (if someone reset the clock on the server) by looking at whether
1585 * or not this RPC call was started after the inode was last updated.
1586 * Note also the check for wraparound of 'attr_gencount'
1588 * The function returns 'true' if it thinks the attributes in 'fattr' are
1589 * more recent than the ones cached in the inode.
1592 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1594 const struct nfs_inode *nfsi = NFS_I(inode);
1596 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1597 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1600 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1604 trace_nfs_refresh_inode_enter(inode);
1606 if (nfs_inode_attrs_need_update(inode, fattr))
1607 ret = nfs_update_inode(inode, fattr);
1609 ret = nfs_check_inode_attributes(inode, fattr);
1611 trace_nfs_refresh_inode_exit(inode, ret);
1616 * nfs_refresh_inode - try to update the inode attribute cache
1617 * @inode - pointer to inode
1618 * @fattr - updated attributes
1620 * Check that an RPC call that returned attributes has not overlapped with
1621 * other recent updates of the inode metadata, then decide whether it is
1622 * safe to do a full update of the inode attributes, or whether just to
1623 * call nfs_check_inode_attributes.
1625 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1629 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1631 spin_lock(&inode->i_lock);
1632 status = nfs_refresh_inode_locked(inode, fattr);
1633 spin_unlock(&inode->i_lock);
1637 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1639 static int nfs_post_op_update_inode_locked(struct inode *inode,
1640 struct nfs_fattr *fattr, unsigned int invalid)
1642 if (S_ISDIR(inode->i_mode))
1643 invalid |= NFS_INO_INVALID_DATA;
1644 nfs_set_cache_invalid(inode, invalid);
1645 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1647 return nfs_refresh_inode_locked(inode, fattr);
1651 * nfs_post_op_update_inode - try to update the inode attribute cache
1652 * @inode - pointer to inode
1653 * @fattr - updated attributes
1655 * After an operation that has changed the inode metadata, mark the
1656 * attribute cache as being invalid, then try to update it.
1658 * NB: if the server didn't return any post op attributes, this
1659 * function will force the retrieval of attributes before the next
1660 * NFS request. Thus it should be used only for operations that
1661 * are expected to change one or more attributes, to avoid
1662 * unnecessary NFS requests and trips through nfs_update_inode().
1664 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1668 spin_lock(&inode->i_lock);
1669 nfs_fattr_set_barrier(fattr);
1670 status = nfs_post_op_update_inode_locked(inode, fattr,
1671 NFS_INO_INVALID_CHANGE
1672 | NFS_INO_INVALID_CTIME
1673 | NFS_INO_REVAL_FORCED);
1674 spin_unlock(&inode->i_lock);
1678 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1681 * nfs_post_op_update_inode_force_wcc_locked - update the inode attribute cache
1682 * @inode - pointer to inode
1683 * @fattr - updated attributes
1685 * After an operation that has changed the inode metadata, mark the
1686 * attribute cache as being invalid, then try to update it. Fake up
1687 * weak cache consistency data, if none exist.
1689 * This function is mainly designed to be used by the ->write_done() functions.
1691 int nfs_post_op_update_inode_force_wcc_locked(struct inode *inode, struct nfs_fattr *fattr)
1695 /* Don't do a WCC update if these attributes are already stale */
1696 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1697 !nfs_inode_attrs_need_update(inode, fattr)) {
1698 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1699 | NFS_ATTR_FATTR_PRESIZE
1700 | NFS_ATTR_FATTR_PREMTIME
1701 | NFS_ATTR_FATTR_PRECTIME);
1704 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1705 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1706 fattr->pre_change_attr = inode_peek_iversion_raw(inode);
1707 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1709 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1710 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1711 fattr->pre_ctime = timespec64_to_timespec(inode->i_ctime);
1712 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1714 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1715 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1716 fattr->pre_mtime = timespec64_to_timespec(inode->i_mtime);
1717 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1719 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1720 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1721 fattr->pre_size = i_size_read(inode);
1722 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1725 status = nfs_post_op_update_inode_locked(inode, fattr,
1726 NFS_INO_INVALID_CHANGE
1727 | NFS_INO_INVALID_CTIME
1728 | NFS_INO_INVALID_MTIME);
1733 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1734 * @inode - pointer to inode
1735 * @fattr - updated attributes
1737 * After an operation that has changed the inode metadata, mark the
1738 * attribute cache as being invalid, then try to update it. Fake up
1739 * weak cache consistency data, if none exist.
1741 * This function is mainly designed to be used by the ->write_done() functions.
1743 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1747 spin_lock(&inode->i_lock);
1748 nfs_fattr_set_barrier(fattr);
1749 status = nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1750 spin_unlock(&inode->i_lock);
1753 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1756 static inline bool nfs_fileid_valid(struct nfs_inode *nfsi,
1757 struct nfs_fattr *fattr)
1759 bool ret1 = true, ret2 = true;
1761 if (fattr->valid & NFS_ATTR_FATTR_FILEID)
1762 ret1 = (nfsi->fileid == fattr->fileid);
1763 if (fattr->valid & NFS_ATTR_FATTR_MOUNTED_ON_FILEID)
1764 ret2 = (nfsi->fileid == fattr->mounted_on_fileid);
1765 return ret1 || ret2;
1769 * Many nfs protocol calls return the new file attributes after
1770 * an operation. Here we update the inode to reflect the state
1771 * of the server's inode.
1773 * This is a bit tricky because we have to make sure all dirty pages
1774 * have been sent off to the server before calling invalidate_inode_pages.
1775 * To make sure no other process adds more write requests while we try
1776 * our best to flush them, we make them sleep during the attribute refresh.
1778 * A very similar scenario holds for the dir cache.
1780 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1782 struct nfs_server *server;
1783 struct nfs_inode *nfsi = NFS_I(inode);
1784 loff_t cur_isize, new_isize;
1785 unsigned long invalid = 0;
1786 unsigned long now = jiffies;
1787 unsigned long save_cache_validity;
1788 bool have_writers = nfs_file_has_buffered_writers(nfsi);
1789 bool cache_revalidated = true;
1790 bool attr_changed = false;
1791 bool have_delegation;
1793 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1794 __func__, inode->i_sb->s_id, inode->i_ino,
1795 nfs_display_fhandle_hash(NFS_FH(inode)),
1796 atomic_read(&inode->i_count), fattr->valid);
1798 if (!nfs_fileid_valid(nfsi, fattr)) {
1799 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1800 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1801 NFS_SERVER(inode)->nfs_client->cl_hostname,
1802 inode->i_sb->s_id, (long long)nfsi->fileid,
1803 (long long)fattr->fileid);
1808 * Make sure the inode's type hasn't changed.
1810 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1812 * Big trouble! The inode has become a different object.
1814 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1815 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1819 server = NFS_SERVER(inode);
1820 /* Update the fsid? */
1821 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1822 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1823 !IS_AUTOMOUNT(inode))
1824 server->fsid = fattr->fsid;
1826 /* Save the delegation state before clearing cache_validity */
1827 have_delegation = nfs_have_delegated_attributes(inode);
1830 * Update the read time so we don't revalidate too often.
1832 nfsi->read_cache_jiffies = fattr->time_start;
1834 save_cache_validity = nfsi->cache_validity;
1835 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1836 | NFS_INO_INVALID_ATIME
1837 | NFS_INO_REVAL_FORCED
1838 | NFS_INO_REVAL_PAGECACHE);
1840 /* Do atomic weak cache consistency updates */
1841 nfs_wcc_update_inode(inode, fattr);
1843 if (pnfs_layoutcommit_outstanding(inode)) {
1844 nfsi->cache_validity |= save_cache_validity & NFS_INO_INVALID_ATTR;
1845 cache_revalidated = false;
1848 /* More cache consistency checks */
1849 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1850 if (!inode_eq_iversion_raw(inode, fattr->change_attr)) {
1851 /* Could it be a race with writeback? */
1852 if (!(have_writers || have_delegation)) {
1853 invalid |= NFS_INO_INVALID_DATA
1854 | NFS_INO_INVALID_ACCESS
1855 | NFS_INO_INVALID_ACL;
1856 /* Force revalidate of all attributes */
1857 save_cache_validity |= NFS_INO_INVALID_CTIME
1858 | NFS_INO_INVALID_MTIME
1859 | NFS_INO_INVALID_SIZE
1860 | NFS_INO_INVALID_OTHER;
1861 if (S_ISDIR(inode->i_mode))
1862 nfs_force_lookup_revalidate(inode);
1863 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1867 inode_set_iversion_raw(inode, fattr->change_attr);
1868 attr_changed = true;
1871 nfsi->cache_validity |= save_cache_validity &
1872 (NFS_INO_INVALID_CHANGE
1873 | NFS_INO_REVAL_PAGECACHE
1874 | NFS_INO_REVAL_FORCED);
1875 cache_revalidated = false;
1878 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1879 inode->i_mtime = timespec_to_timespec64(fattr->mtime);
1880 } else if (server->caps & NFS_CAP_MTIME) {
1881 nfsi->cache_validity |= save_cache_validity &
1882 (NFS_INO_INVALID_MTIME
1883 | NFS_INO_REVAL_FORCED);
1884 cache_revalidated = false;
1887 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1888 inode->i_ctime = timespec_to_timespec64(fattr->ctime);
1889 } else if (server->caps & NFS_CAP_CTIME) {
1890 nfsi->cache_validity |= save_cache_validity &
1891 (NFS_INO_INVALID_CTIME
1892 | NFS_INO_REVAL_FORCED);
1893 cache_revalidated = false;
1896 /* Check if our cached file size is stale */
1897 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1898 new_isize = nfs_size_to_loff_t(fattr->size);
1899 cur_isize = i_size_read(inode);
1900 if (new_isize != cur_isize && !have_delegation) {
1901 /* Do we perhaps have any outstanding writes, or has
1902 * the file grown beyond our last write? */
1903 if (!nfs_have_writebacks(inode) || new_isize > cur_isize) {
1904 i_size_write(inode, new_isize);
1906 invalid |= NFS_INO_INVALID_DATA;
1907 attr_changed = true;
1909 dprintk("NFS: isize change on server for file %s/%ld "
1913 (long long)cur_isize,
1914 (long long)new_isize);
1917 nfsi->cache_validity |= save_cache_validity &
1918 (NFS_INO_INVALID_SIZE
1919 | NFS_INO_REVAL_PAGECACHE
1920 | NFS_INO_REVAL_FORCED);
1921 cache_revalidated = false;
1925 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1926 inode->i_atime = timespec_to_timespec64(fattr->atime);
1927 else if (server->caps & NFS_CAP_ATIME) {
1928 nfsi->cache_validity |= save_cache_validity &
1929 (NFS_INO_INVALID_ATIME
1930 | NFS_INO_REVAL_FORCED);
1931 cache_revalidated = false;
1934 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1935 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1936 umode_t newmode = inode->i_mode & S_IFMT;
1937 newmode |= fattr->mode & S_IALLUGO;
1938 inode->i_mode = newmode;
1939 invalid |= NFS_INO_INVALID_ACCESS
1940 | NFS_INO_INVALID_ACL;
1941 attr_changed = true;
1943 } else if (server->caps & NFS_CAP_MODE) {
1944 nfsi->cache_validity |= save_cache_validity &
1945 (NFS_INO_INVALID_OTHER
1946 | NFS_INO_REVAL_FORCED);
1947 cache_revalidated = false;
1950 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1951 if (!uid_eq(inode->i_uid, fattr->uid)) {
1952 invalid |= NFS_INO_INVALID_ACCESS
1953 | NFS_INO_INVALID_ACL;
1954 inode->i_uid = fattr->uid;
1955 attr_changed = true;
1957 } else if (server->caps & NFS_CAP_OWNER) {
1958 nfsi->cache_validity |= save_cache_validity &
1959 (NFS_INO_INVALID_OTHER
1960 | NFS_INO_REVAL_FORCED);
1961 cache_revalidated = false;
1964 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1965 if (!gid_eq(inode->i_gid, fattr->gid)) {
1966 invalid |= NFS_INO_INVALID_ACCESS
1967 | NFS_INO_INVALID_ACL;
1968 inode->i_gid = fattr->gid;
1969 attr_changed = true;
1971 } else if (server->caps & NFS_CAP_OWNER_GROUP) {
1972 nfsi->cache_validity |= save_cache_validity &
1973 (NFS_INO_INVALID_OTHER
1974 | NFS_INO_REVAL_FORCED);
1975 cache_revalidated = false;
1978 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1979 if (inode->i_nlink != fattr->nlink) {
1980 if (S_ISDIR(inode->i_mode))
1981 invalid |= NFS_INO_INVALID_DATA;
1982 set_nlink(inode, fattr->nlink);
1983 attr_changed = true;
1985 } else if (server->caps & NFS_CAP_NLINK) {
1986 nfsi->cache_validity |= save_cache_validity &
1987 (NFS_INO_INVALID_OTHER
1988 | NFS_INO_REVAL_FORCED);
1989 cache_revalidated = false;
1992 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1994 * report the blocks in 512byte units
1996 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1997 } else if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1998 inode->i_blocks = fattr->du.nfs2.blocks;
2000 cache_revalidated = false;
2002 /* Update attrtimeo value if we're out of the unstable period */
2004 invalid &= ~NFS_INO_INVALID_ATTR;
2005 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
2006 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
2007 nfsi->attrtimeo_timestamp = now;
2008 /* Set barrier to be more recent than all outstanding updates */
2009 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
2011 if (cache_revalidated) {
2012 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp,
2013 nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
2014 nfsi->attrtimeo <<= 1;
2015 if (nfsi->attrtimeo > NFS_MAXATTRTIMEO(inode))
2016 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
2018 nfsi->attrtimeo_timestamp = now;
2020 /* Set the barrier to be more recent than this fattr */
2021 if ((long)fattr->gencount - (long)nfsi->attr_gencount > 0)
2022 nfsi->attr_gencount = fattr->gencount;
2025 /* Don't invalidate the data if we were to blame */
2026 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
2027 || S_ISLNK(inode->i_mode)))
2028 invalid &= ~NFS_INO_INVALID_DATA;
2029 nfs_set_cache_invalid(inode, invalid);
2034 * No need to worry about unhashing the dentry, as the
2035 * lookup validation will know that the inode is bad.
2036 * (But we fall through to invalidate the caches.)
2038 nfs_invalidate_inode(inode);
2042 struct inode *nfs_alloc_inode(struct super_block *sb)
2044 struct nfs_inode *nfsi;
2045 nfsi = kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
2049 nfsi->cache_validity = 0UL;
2050 #if IS_ENABLED(CONFIG_NFS_V4)
2051 nfsi->nfs4_acl = NULL;
2052 #endif /* CONFIG_NFS_V4 */
2053 return &nfsi->vfs_inode;
2055 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
2057 static void nfs_i_callback(struct rcu_head *head)
2059 struct inode *inode = container_of(head, struct inode, i_rcu);
2060 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
2063 void nfs_destroy_inode(struct inode *inode)
2065 call_rcu(&inode->i_rcu, nfs_i_callback);
2067 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
2069 static inline void nfs4_init_once(struct nfs_inode *nfsi)
2071 #if IS_ENABLED(CONFIG_NFS_V4)
2072 INIT_LIST_HEAD(&nfsi->open_states);
2073 nfsi->delegation = NULL;
2074 init_rwsem(&nfsi->rwsem);
2075 nfsi->layout = NULL;
2079 static void init_once(void *foo)
2081 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
2083 inode_init_once(&nfsi->vfs_inode);
2084 INIT_LIST_HEAD(&nfsi->open_files);
2085 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
2086 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
2087 INIT_LIST_HEAD(&nfsi->commit_info.list);
2088 atomic_long_set(&nfsi->nrequests, 0);
2089 atomic_long_set(&nfsi->commit_info.ncommit, 0);
2090 atomic_set(&nfsi->commit_info.rpcs_out, 0);
2091 init_rwsem(&nfsi->rmdir_sem);
2092 mutex_init(&nfsi->commit_mutex);
2093 nfs4_init_once(nfsi);
2096 static int __init nfs_init_inodecache(void)
2098 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
2099 sizeof(struct nfs_inode),
2100 0, (SLAB_RECLAIM_ACCOUNT|
2101 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2103 if (nfs_inode_cachep == NULL)
2109 static void nfs_destroy_inodecache(void)
2112 * Make sure all delayed rcu free inodes are flushed before we
2116 kmem_cache_destroy(nfs_inode_cachep);
2119 struct workqueue_struct *nfsiod_workqueue;
2120 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
2123 * start up the nfsiod workqueue
2125 static int nfsiod_start(void)
2127 struct workqueue_struct *wq;
2128 dprintk("RPC: creating workqueue nfsiod\n");
2129 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
2132 nfsiod_workqueue = wq;
2137 * Destroy the nfsiod workqueue
2139 static void nfsiod_stop(void)
2141 struct workqueue_struct *wq;
2143 wq = nfsiod_workqueue;
2146 nfsiod_workqueue = NULL;
2147 destroy_workqueue(wq);
2150 unsigned int nfs_net_id;
2151 EXPORT_SYMBOL_GPL(nfs_net_id);
2153 static int nfs_net_init(struct net *net)
2155 nfs_clients_init(net);
2156 return nfs_fs_proc_net_init(net);
2159 static void nfs_net_exit(struct net *net)
2161 struct nfs_net *nn = net_generic(net, nfs_net_id);
2163 nfs_fs_proc_net_exit(net);
2164 nfs_cleanup_cb_ident_idr(net);
2165 WARN_ON_ONCE(!list_empty(&nn->nfs_client_list));
2166 WARN_ON_ONCE(!list_empty(&nn->nfs_volume_list));
2169 static struct pernet_operations nfs_net_ops = {
2170 .init = nfs_net_init,
2171 .exit = nfs_net_exit,
2173 .size = sizeof(struct nfs_net),
2179 static int __init init_nfs_fs(void)
2183 err = register_pernet_subsys(&nfs_net_ops);
2187 err = nfs_fscache_register();
2191 err = nfsiod_start();
2195 err = nfs_fs_proc_init();
2199 err = nfs_init_nfspagecache();
2203 err = nfs_init_inodecache();
2207 err = nfs_init_readpagecache();
2211 err = nfs_init_writepagecache();
2215 err = nfs_init_directcache();
2219 rpc_proc_register(&init_net, &nfs_rpcstat);
2221 err = register_nfs_fs();
2227 rpc_proc_unregister(&init_net, "nfs");
2228 nfs_destroy_directcache();
2230 nfs_destroy_writepagecache();
2232 nfs_destroy_readpagecache();
2234 nfs_destroy_inodecache();
2236 nfs_destroy_nfspagecache();
2242 nfs_fscache_unregister();
2244 unregister_pernet_subsys(&nfs_net_ops);
2249 static void __exit exit_nfs_fs(void)
2251 nfs_destroy_directcache();
2252 nfs_destroy_writepagecache();
2253 nfs_destroy_readpagecache();
2254 nfs_destroy_inodecache();
2255 nfs_destroy_nfspagecache();
2256 nfs_fscache_unregister();
2257 unregister_pernet_subsys(&nfs_net_ops);
2258 rpc_proc_unregister(&init_net, "nfs");
2259 unregister_nfs_fs();
2264 /* Not quite true; I just maintain it */
2266 MODULE_LICENSE("GPL");
2267 module_param(enable_ino64, bool, 0644);
2269 module_init(init_nfs_fs)
2270 module_exit(exit_nfs_fs)