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.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>
42 #include <asm/uaccess.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);
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
78 int nfs_wait_bit_killable(struct wait_bit_key *key)
80 if (fatal_signal_pending(current))
82 freezable_schedule_unsafe();
85 EXPORT_SYMBOL_GPL(nfs_wait_bit_killable);
88 * nfs_compat_user_ino64 - returns the user-visible inode number
89 * @fileid: 64-bit fileid
91 * This function returns a 32-bit inode number if the boot parameter
92 * nfs.enable_ino64 is zero.
94 u64 nfs_compat_user_ino64(u64 fileid)
105 if (sizeof(ino) < sizeof(fileid))
106 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
110 int nfs_drop_inode(struct inode *inode)
112 return NFS_STALE(inode) || generic_drop_inode(inode);
114 EXPORT_SYMBOL_GPL(nfs_drop_inode);
116 void nfs_clear_inode(struct inode *inode)
119 * The following should never happen...
121 WARN_ON_ONCE(nfs_have_writebacks(inode));
122 WARN_ON_ONCE(!list_empty(&NFS_I(inode)->open_files));
123 nfs_zap_acl_cache(inode);
124 nfs_access_zap_cache(inode);
125 nfs_fscache_clear_inode(inode);
127 EXPORT_SYMBOL_GPL(nfs_clear_inode);
129 void nfs_evict_inode(struct inode *inode)
131 truncate_inode_pages_final(&inode->i_data);
133 nfs_clear_inode(inode);
137 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
139 int nfs_sync_mapping(struct address_space *mapping)
143 if (mapping->nrpages != 0) {
144 unmap_mapping_range(mapping, 0, 0, 0);
145 ret = nfs_wb_all(mapping->host);
150 static void nfs_set_cache_invalid(struct inode *inode, unsigned long flags)
152 struct nfs_inode *nfsi = NFS_I(inode);
154 if (inode->i_mapping->nrpages == 0)
155 flags &= ~NFS_INO_INVALID_DATA;
156 nfsi->cache_validity |= flags;
157 if (flags & NFS_INO_INVALID_DATA)
158 nfs_fscache_invalidate(inode);
162 * Invalidate the local caches
164 static void nfs_zap_caches_locked(struct inode *inode)
166 struct nfs_inode *nfsi = NFS_I(inode);
167 int mode = inode->i_mode;
169 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
171 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
172 nfsi->attrtimeo_timestamp = jiffies;
174 memset(NFS_I(inode)->cookieverf, 0, sizeof(NFS_I(inode)->cookieverf));
175 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) {
176 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
177 | NFS_INO_INVALID_DATA
178 | NFS_INO_INVALID_ACCESS
179 | NFS_INO_INVALID_ACL
180 | NFS_INO_REVAL_PAGECACHE);
182 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
183 | NFS_INO_INVALID_ACCESS
184 | NFS_INO_INVALID_ACL
185 | NFS_INO_REVAL_PAGECACHE);
186 nfs_zap_label_cache_locked(nfsi);
189 void nfs_zap_caches(struct inode *inode)
191 spin_lock(&inode->i_lock);
192 nfs_zap_caches_locked(inode);
193 spin_unlock(&inode->i_lock);
196 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
198 if (mapping->nrpages != 0) {
199 spin_lock(&inode->i_lock);
200 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
201 spin_unlock(&inode->i_lock);
205 void nfs_zap_acl_cache(struct inode *inode)
207 void (*clear_acl_cache)(struct inode *);
209 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
210 if (clear_acl_cache != NULL)
211 clear_acl_cache(inode);
212 spin_lock(&inode->i_lock);
213 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
214 spin_unlock(&inode->i_lock);
216 EXPORT_SYMBOL_GPL(nfs_zap_acl_cache);
218 void nfs_invalidate_atime(struct inode *inode)
220 spin_lock(&inode->i_lock);
221 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATIME);
222 spin_unlock(&inode->i_lock);
224 EXPORT_SYMBOL_GPL(nfs_invalidate_atime);
227 * Invalidate, but do not unhash, the inode.
228 * NB: must be called with inode->i_lock held!
230 static void nfs_invalidate_inode(struct inode *inode)
232 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
233 nfs_zap_caches_locked(inode);
236 struct nfs_find_desc {
238 struct nfs_fattr *fattr;
242 * In NFSv3 we can have 64bit inode numbers. In order to support
243 * this, and re-exported directories (also seen in NFSv2)
244 * we are forced to allow 2 different inodes to have the same
248 nfs_find_actor(struct inode *inode, void *opaque)
250 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
251 struct nfs_fh *fh = desc->fh;
252 struct nfs_fattr *fattr = desc->fattr;
254 if (NFS_FILEID(inode) != fattr->fileid)
256 if ((S_IFMT & inode->i_mode) != (S_IFMT & fattr->mode))
258 if (nfs_compare_fh(NFS_FH(inode), fh))
260 if (is_bad_inode(inode) || NFS_STALE(inode))
266 nfs_init_locked(struct inode *inode, void *opaque)
268 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
269 struct nfs_fattr *fattr = desc->fattr;
271 set_nfs_fileid(inode, fattr->fileid);
272 nfs_copy_fh(NFS_FH(inode), desc->fh);
276 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
277 static void nfs_clear_label_invalid(struct inode *inode)
279 spin_lock(&inode->i_lock);
280 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_LABEL;
281 spin_unlock(&inode->i_lock);
284 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
285 struct nfs4_label *label)
292 if ((fattr->valid & NFS_ATTR_FATTR_V4_SECURITY_LABEL) && inode->i_security) {
293 error = security_inode_notifysecctx(inode, label->label,
296 printk(KERN_ERR "%s() %s %d "
297 "security_inode_notifysecctx() %d\n",
299 (char *)label->label,
301 nfs_clear_label_invalid(inode);
305 struct nfs4_label *nfs4_label_alloc(struct nfs_server *server, gfp_t flags)
307 struct nfs4_label *label = NULL;
308 int minor_version = server->nfs_client->cl_minorversion;
310 if (minor_version < 2)
313 if (!(server->caps & NFS_CAP_SECURITY_LABEL))
316 label = kzalloc(sizeof(struct nfs4_label), flags);
318 return ERR_PTR(-ENOMEM);
320 label->label = kzalloc(NFS4_MAXLABELLEN, flags);
321 if (label->label == NULL) {
323 return ERR_PTR(-ENOMEM);
325 label->len = NFS4_MAXLABELLEN;
329 EXPORT_SYMBOL_GPL(nfs4_label_alloc);
331 void nfs_setsecurity(struct inode *inode, struct nfs_fattr *fattr,
332 struct nfs4_label *label)
336 EXPORT_SYMBOL_GPL(nfs_setsecurity);
339 * This is our front-end to iget that looks up inodes by file handle
340 * instead of inode number.
343 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr, struct nfs4_label *label)
345 struct nfs_find_desc desc = {
349 struct inode *inode = ERR_PTR(-ENOENT);
352 nfs_attr_check_mountpoint(sb, fattr);
354 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
355 !nfs_attr_use_mounted_on_fileid(fattr))
357 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
360 hash = nfs_fattr_to_ino_t(fattr);
362 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
364 inode = ERR_PTR(-ENOMEM);
368 if (inode->i_state & I_NEW) {
369 struct nfs_inode *nfsi = NFS_I(inode);
370 unsigned long now = jiffies;
372 /* We set i_ino for the few things that still rely on it,
376 /* We can't support update_atime(), since the server will reset it */
377 inode->i_flags |= S_NOATIME|S_NOCMTIME;
378 inode->i_mode = fattr->mode;
379 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
380 && nfs_server_capable(inode, NFS_CAP_MODE))
381 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
382 /* Why so? Because we want revalidate for devices/FIFOs, and
383 * that's precisely what we have in nfs_file_inode_operations.
385 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
386 if (S_ISREG(inode->i_mode)) {
387 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
388 inode->i_data.a_ops = &nfs_file_aops;
389 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
390 } else if (S_ISDIR(inode->i_mode)) {
391 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
392 inode->i_fop = &nfs_dir_operations;
393 inode->i_data.a_ops = &nfs_dir_aops;
394 /* Deal with crossing mountpoints */
395 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
396 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
397 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
398 inode->i_op = &nfs_referral_inode_operations;
400 inode->i_op = &nfs_mountpoint_inode_operations;
402 inode->i_flags |= S_AUTOMOUNT;
404 } else if (S_ISLNK(inode->i_mode))
405 inode->i_op = &nfs_symlink_inode_operations;
407 init_special_inode(inode, inode->i_mode, fattr->rdev);
409 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
410 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
411 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
412 inode->i_version = 0;
415 inode->i_uid = make_kuid(&init_user_ns, -2);
416 inode->i_gid = make_kgid(&init_user_ns, -2);
418 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
422 nfsi->read_cache_jiffies = fattr->time_start;
423 nfsi->attr_gencount = fattr->gencount;
424 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
425 inode->i_atime = fattr->atime;
426 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
427 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
428 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
429 inode->i_mtime = fattr->mtime;
430 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
431 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
432 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
433 inode->i_ctime = fattr->ctime;
434 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
435 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
436 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
437 inode->i_version = fattr->change_attr;
438 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
439 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
440 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
441 inode->i_size = nfs_size_to_loff_t(fattr->size);
443 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR
444 | NFS_INO_REVAL_PAGECACHE);
445 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
446 set_nlink(inode, fattr->nlink);
447 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
448 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
449 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
450 inode->i_uid = fattr->uid;
451 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
452 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
453 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
454 inode->i_gid = fattr->gid;
455 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
456 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ATTR);
457 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
458 inode->i_blocks = fattr->du.nfs2.blocks;
459 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
461 * report the blocks in 512byte units
463 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
466 nfs_setsecurity(inode, fattr, label);
468 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
469 nfsi->attrtimeo_timestamp = now;
470 nfsi->access_cache = RB_ROOT;
472 nfs_fscache_init_inode(inode);
474 unlock_new_inode(inode);
476 nfs_refresh_inode(inode, fattr);
477 dprintk("NFS: nfs_fhget(%s/%Lu fh_crc=0x%08x ct=%d)\n",
479 (unsigned long long)NFS_FILEID(inode),
480 nfs_display_fhandle_hash(fh),
481 atomic_read(&inode->i_count));
487 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
490 EXPORT_SYMBOL_GPL(nfs_fhget);
492 #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)
495 nfs_setattr(struct dentry *dentry, struct iattr *attr)
497 struct inode *inode = dentry->d_inode;
498 struct nfs_fattr *fattr;
501 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
503 /* skip mode change if it's just for clearing setuid/setgid */
504 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
505 attr->ia_valid &= ~ATTR_MODE;
507 if (attr->ia_valid & ATTR_SIZE) {
508 BUG_ON(!S_ISREG(inode->i_mode));
510 if (attr->ia_size == i_size_read(inode))
511 attr->ia_valid &= ~ATTR_SIZE;
514 /* Optimization: if the end result is no change, don't RPC */
515 attr->ia_valid &= NFS_VALID_ATTRS;
516 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
519 trace_nfs_setattr_enter(inode);
521 /* Write all dirty data */
522 if (S_ISREG(inode->i_mode)) {
523 nfs_inode_dio_wait(inode);
527 fattr = nfs_alloc_fattr();
531 * Return any delegations if we're going to change ACLs
533 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
534 NFS_PROTO(inode)->return_delegation(inode);
535 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
537 error = nfs_refresh_inode(inode, fattr);
538 nfs_free_fattr(fattr);
540 trace_nfs_setattr_exit(inode, error);
543 EXPORT_SYMBOL_GPL(nfs_setattr);
546 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
547 * @inode: inode of the file used
548 * @offset: file offset to start truncating
550 * This is a copy of the common vmtruncate, but with the locking
551 * corrected to take into account the fact that NFS requires
552 * inode->i_size to be updated under the inode->i_lock.
554 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
558 err = inode_newsize_ok(inode, offset);
562 spin_lock(&inode->i_lock);
563 i_size_write(inode, offset);
566 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_DATA;
567 spin_unlock(&inode->i_lock);
569 truncate_pagecache(inode, offset);
575 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
576 * @inode: pointer to struct inode
577 * @attr: pointer to struct iattr
579 * Note: we do this in the *proc.c in order to ensure that
580 * it works for things like exclusive creates too.
582 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
584 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
585 spin_lock(&inode->i_lock);
586 if ((attr->ia_valid & ATTR_MODE) != 0) {
587 int mode = attr->ia_mode & S_IALLUGO;
588 mode |= inode->i_mode & ~S_IALLUGO;
589 inode->i_mode = mode;
591 if ((attr->ia_valid & ATTR_UID) != 0)
592 inode->i_uid = attr->ia_uid;
593 if ((attr->ia_valid & ATTR_GID) != 0)
594 inode->i_gid = attr->ia_gid;
595 nfs_set_cache_invalid(inode, NFS_INO_INVALID_ACCESS
596 | NFS_INO_INVALID_ACL);
597 spin_unlock(&inode->i_lock);
599 if ((attr->ia_valid & ATTR_SIZE) != 0) {
600 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
601 nfs_vmtruncate(inode, attr->ia_size);
604 EXPORT_SYMBOL_GPL(nfs_setattr_update_inode);
606 static void nfs_request_parent_use_readdirplus(struct dentry *dentry)
608 struct dentry *parent;
610 parent = dget_parent(dentry);
611 nfs_force_use_readdirplus(parent->d_inode);
615 static bool nfs_need_revalidate_inode(struct inode *inode)
617 if (NFS_I(inode)->cache_validity &
618 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
620 if (nfs_attribute_cache_expired(inode))
625 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
627 struct inode *inode = dentry->d_inode;
628 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
631 trace_nfs_getattr_enter(inode);
632 /* Flush out writes to the server in order to update c/mtime. */
633 if (S_ISREG(inode->i_mode)) {
634 nfs_inode_dio_wait(inode);
635 err = filemap_write_and_wait(inode->i_mapping);
641 * We may force a getattr if the user cares about atime.
643 * Note that we only have to check the vfsmount flags here:
644 * - NFS always sets S_NOATIME by so checking it would give a
646 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
647 * no point in checking those.
649 if ((mnt->mnt_flags & MNT_NOATIME) ||
650 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
653 if (need_atime || nfs_need_revalidate_inode(inode)) {
654 struct nfs_server *server = NFS_SERVER(inode);
656 if (server->caps & NFS_CAP_READDIRPLUS)
657 nfs_request_parent_use_readdirplus(dentry);
658 err = __nfs_revalidate_inode(server, inode);
661 generic_fillattr(inode, stat);
662 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
665 trace_nfs_getattr_exit(inode, err);
668 EXPORT_SYMBOL_GPL(nfs_getattr);
670 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
672 atomic_set(&l_ctx->count, 1);
673 l_ctx->lockowner.l_owner = current->files;
674 l_ctx->lockowner.l_pid = current->tgid;
675 INIT_LIST_HEAD(&l_ctx->list);
676 nfs_iocounter_init(&l_ctx->io_count);
679 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
681 struct nfs_lock_context *head = &ctx->lock_context;
682 struct nfs_lock_context *pos = head;
685 if (pos->lockowner.l_owner != current->files)
687 if (pos->lockowner.l_pid != current->tgid)
689 atomic_inc(&pos->count);
691 } while ((pos = list_entry(pos->list.next, typeof(*pos), list)) != head);
695 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
697 struct nfs_lock_context *res, *new = NULL;
698 struct inode *inode = ctx->dentry->d_inode;
700 spin_lock(&inode->i_lock);
701 res = __nfs_find_lock_context(ctx);
703 spin_unlock(&inode->i_lock);
704 new = kmalloc(sizeof(*new), GFP_KERNEL);
706 return ERR_PTR(-ENOMEM);
707 nfs_init_lock_context(new);
708 spin_lock(&inode->i_lock);
709 res = __nfs_find_lock_context(ctx);
711 list_add_tail(&new->list, &ctx->lock_context.list);
712 new->open_context = ctx;
717 spin_unlock(&inode->i_lock);
721 EXPORT_SYMBOL_GPL(nfs_get_lock_context);
723 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
725 struct nfs_open_context *ctx = l_ctx->open_context;
726 struct inode *inode = ctx->dentry->d_inode;
728 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
730 list_del(&l_ctx->list);
731 spin_unlock(&inode->i_lock);
734 EXPORT_SYMBOL_GPL(nfs_put_lock_context);
737 * nfs_close_context - Common close_context() routine NFSv2/v3
738 * @ctx: pointer to context
739 * @is_sync: is this a synchronous close
741 * always ensure that the attributes are up to date if we're mounted
742 * with close-to-open semantics
744 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
747 struct nfs_server *server;
749 if (!(ctx->mode & FMODE_WRITE))
753 inode = ctx->dentry->d_inode;
754 if (!list_empty(&NFS_I(inode)->open_files))
756 server = NFS_SERVER(inode);
757 if (server->flags & NFS_MOUNT_NOCTO)
759 nfs_revalidate_inode(server, inode);
761 EXPORT_SYMBOL_GPL(nfs_close_context);
763 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
765 struct nfs_open_context *ctx;
766 struct rpc_cred *cred = rpc_lookup_cred();
768 return ERR_CAST(cred);
770 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
773 return ERR_PTR(-ENOMEM);
775 nfs_sb_active(dentry->d_sb);
776 ctx->dentry = dget(dentry);
782 nfs_init_lock_context(&ctx->lock_context);
783 ctx->lock_context.open_context = ctx;
784 INIT_LIST_HEAD(&ctx->list);
785 ctx->mdsthreshold = NULL;
788 EXPORT_SYMBOL_GPL(alloc_nfs_open_context);
790 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
793 atomic_inc(&ctx->lock_context.count);
796 EXPORT_SYMBOL_GPL(get_nfs_open_context);
798 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
800 struct inode *inode = ctx->dentry->d_inode;
801 struct super_block *sb = ctx->dentry->d_sb;
803 if (!list_empty(&ctx->list)) {
804 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
806 list_del(&ctx->list);
807 spin_unlock(&inode->i_lock);
808 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
811 NFS_PROTO(inode)->close_context(ctx, is_sync);
812 if (ctx->cred != NULL)
813 put_rpccred(ctx->cred);
816 kfree(ctx->mdsthreshold);
820 void put_nfs_open_context(struct nfs_open_context *ctx)
822 __put_nfs_open_context(ctx, 0);
824 EXPORT_SYMBOL_GPL(put_nfs_open_context);
827 * Ensure that mmap has a recent RPC credential for use when writing out
830 void nfs_inode_attach_open_context(struct nfs_open_context *ctx)
832 struct inode *inode = ctx->dentry->d_inode;
833 struct nfs_inode *nfsi = NFS_I(inode);
835 spin_lock(&inode->i_lock);
836 list_add(&ctx->list, &nfsi->open_files);
837 spin_unlock(&inode->i_lock);
839 EXPORT_SYMBOL_GPL(nfs_inode_attach_open_context);
841 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
843 filp->private_data = get_nfs_open_context(ctx);
844 if (list_empty(&ctx->list))
845 nfs_inode_attach_open_context(ctx);
847 EXPORT_SYMBOL_GPL(nfs_file_set_open_context);
850 * Given an inode, search for an open context with the desired characteristics
852 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
854 struct nfs_inode *nfsi = NFS_I(inode);
855 struct nfs_open_context *pos, *ctx = NULL;
857 spin_lock(&inode->i_lock);
858 list_for_each_entry(pos, &nfsi->open_files, list) {
859 if (cred != NULL && pos->cred != cred)
861 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
863 ctx = get_nfs_open_context(pos);
866 spin_unlock(&inode->i_lock);
870 static void nfs_file_clear_open_context(struct file *filp)
872 struct nfs_open_context *ctx = nfs_file_open_context(filp);
875 struct inode *inode = ctx->dentry->d_inode;
877 filp->private_data = NULL;
878 spin_lock(&inode->i_lock);
879 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
880 spin_unlock(&inode->i_lock);
881 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
886 * These allocate and release file read/write context information.
888 int nfs_open(struct inode *inode, struct file *filp)
890 struct nfs_open_context *ctx;
892 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
895 nfs_file_set_open_context(filp, ctx);
896 put_nfs_open_context(ctx);
897 nfs_fscache_open_file(inode, filp);
901 int nfs_release(struct inode *inode, struct file *filp)
903 nfs_file_clear_open_context(filp);
908 * This function is called whenever some part of NFS notices that
909 * the cached attributes have to be refreshed.
912 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
914 int status = -ESTALE;
915 struct nfs4_label *label = NULL;
916 struct nfs_fattr *fattr = NULL;
917 struct nfs_inode *nfsi = NFS_I(inode);
919 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Lu)\n",
920 inode->i_sb->s_id, (unsigned long long)NFS_FILEID(inode));
922 trace_nfs_revalidate_inode_enter(inode);
924 if (is_bad_inode(inode))
926 if (NFS_STALE(inode))
930 fattr = nfs_alloc_fattr();
934 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
936 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
938 status = PTR_ERR(label);
942 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr, label);
944 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) getattr failed, error=%d\n",
946 (unsigned long long)NFS_FILEID(inode), status);
947 if (status == -ESTALE) {
948 nfs_zap_caches(inode);
949 if (!S_ISDIR(inode->i_mode))
950 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
955 status = nfs_refresh_inode(inode, fattr);
957 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Lu) refresh failed, error=%d\n",
959 (unsigned long long)NFS_FILEID(inode), status);
963 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
964 nfs_zap_acl_cache(inode);
966 nfs_setsecurity(inode, fattr, label);
968 dfprintk(PAGECACHE, "NFS: (%s/%Lu) revalidation complete\n",
970 (unsigned long long)NFS_FILEID(inode));
973 nfs4_label_free(label);
975 nfs_free_fattr(fattr);
976 trace_nfs_revalidate_inode_exit(inode, status);
980 int nfs_attribute_timeout(struct inode *inode)
982 struct nfs_inode *nfsi = NFS_I(inode);
984 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
987 int nfs_attribute_cache_expired(struct inode *inode)
989 if (nfs_have_delegated_attributes(inode))
991 return nfs_attribute_timeout(inode);
995 * nfs_revalidate_inode - Revalidate the inode attributes
996 * @server - pointer to nfs_server struct
997 * @inode - pointer to inode struct
999 * Updates inode attribute information by retrieving the data from the server.
1001 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
1003 if (!nfs_need_revalidate_inode(inode))
1004 return NFS_STALE(inode) ? -ESTALE : 0;
1005 return __nfs_revalidate_inode(server, inode);
1007 EXPORT_SYMBOL_GPL(nfs_revalidate_inode);
1009 int nfs_revalidate_inode_rcu(struct nfs_server *server, struct inode *inode)
1011 if (!(NFS_I(inode)->cache_validity &
1012 (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_LABEL))
1013 && !nfs_attribute_cache_expired(inode))
1014 return NFS_STALE(inode) ? -ESTALE : 0;
1018 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
1020 struct nfs_inode *nfsi = NFS_I(inode);
1023 if (mapping->nrpages != 0) {
1024 if (S_ISREG(inode->i_mode)) {
1025 ret = nfs_sync_mapping(mapping);
1029 ret = invalidate_inode_pages2(mapping);
1033 if (S_ISDIR(inode->i_mode)) {
1034 spin_lock(&inode->i_lock);
1035 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1036 spin_unlock(&inode->i_lock);
1038 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
1039 nfs_fscache_wait_on_invalidate(inode);
1041 dfprintk(PAGECACHE, "NFS: (%s/%Lu) data cache invalidated\n",
1043 (unsigned long long)NFS_FILEID(inode));
1047 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
1049 if (nfs_have_delegated_attributes(inode))
1051 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
1052 || nfs_attribute_timeout(inode)
1053 || NFS_STALE(inode);
1057 * nfs_revalidate_mapping - Revalidate the pagecache
1058 * @inode - pointer to host inode
1059 * @mapping - pointer to mapping
1061 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
1063 struct nfs_inode *nfsi = NFS_I(inode);
1064 unsigned long *bitlock = &nfsi->flags;
1067 /* swapfiles are not supposed to be shared. */
1068 if (IS_SWAPFILE(inode))
1071 if (nfs_mapping_need_revalidate_inode(inode)) {
1072 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1078 * We must clear NFS_INO_INVALID_DATA first to ensure that
1079 * invalidations that come in while we're shooting down the mappings
1080 * are respected. But, that leaves a race window where one revalidator
1081 * can clear the flag, and then another checks it before the mapping
1082 * gets invalidated. Fix that by serializing access to this part of
1085 * At the same time, we need to allow other tasks to see whether we
1086 * might be in the middle of invalidating the pages, so we only set
1087 * the bit lock here if it looks like we're going to be doing that.
1090 ret = wait_on_bit_action(bitlock, NFS_INO_INVALIDATING,
1091 nfs_wait_bit_killable, TASK_KILLABLE);
1094 spin_lock(&inode->i_lock);
1095 if (test_bit(NFS_INO_INVALIDATING, bitlock)) {
1096 spin_unlock(&inode->i_lock);
1099 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1101 spin_unlock(&inode->i_lock);
1105 set_bit(NFS_INO_INVALIDATING, bitlock);
1107 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
1108 spin_unlock(&inode->i_lock);
1109 trace_nfs_invalidate_mapping_enter(inode);
1110 ret = nfs_invalidate_mapping(inode, mapping);
1111 trace_nfs_invalidate_mapping_exit(inode, ret);
1113 clear_bit_unlock(NFS_INO_INVALIDATING, bitlock);
1114 smp_mb__after_atomic();
1115 wake_up_bit(bitlock, NFS_INO_INVALIDATING);
1120 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1122 struct nfs_inode *nfsi = NFS_I(inode);
1123 unsigned long ret = 0;
1125 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
1126 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
1127 && inode->i_version == fattr->pre_change_attr) {
1128 inode->i_version = fattr->change_attr;
1129 if (S_ISDIR(inode->i_mode))
1130 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1131 ret |= NFS_INO_INVALID_ATTR;
1133 /* If we have atomic WCC data, we may update some attributes */
1134 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
1135 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
1136 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
1137 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1138 ret |= NFS_INO_INVALID_ATTR;
1141 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
1142 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
1143 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
1144 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1145 if (S_ISDIR(inode->i_mode))
1146 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1147 ret |= NFS_INO_INVALID_ATTR;
1149 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
1150 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
1151 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
1152 && nfsi->npages == 0) {
1153 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
1154 ret |= NFS_INO_INVALID_ATTR;
1161 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1162 * @inode - pointer to inode
1163 * @fattr - updated attributes
1165 * Verifies the attribute cache. If we have just changed the attributes,
1166 * so that fattr carries weak cache consistency data, then it may
1167 * also update the ctime/mtime/change_attribute.
1169 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1171 struct nfs_inode *nfsi = NFS_I(inode);
1172 loff_t cur_size, new_isize;
1173 unsigned long invalid = 0;
1176 if (nfs_have_delegated_attributes(inode))
1178 /* Has the inode gone and changed behind our back? */
1179 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
1181 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1184 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1185 inode->i_version != fattr->change_attr)
1186 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1188 /* Verify a few of the more important attributes */
1189 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
1190 invalid |= NFS_INO_INVALID_ATTR;
1192 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1193 cur_size = i_size_read(inode);
1194 new_isize = nfs_size_to_loff_t(fattr->size);
1195 if (cur_size != new_isize && nfsi->npages == 0)
1196 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1199 /* Have any file permissions changed? */
1200 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
1201 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1202 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && !uid_eq(inode->i_uid, fattr->uid))
1203 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1204 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && !gid_eq(inode->i_gid, fattr->gid))
1205 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1207 /* Has the link count changed? */
1208 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
1209 invalid |= NFS_INO_INVALID_ATTR;
1211 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
1212 invalid |= NFS_INO_INVALID_ATIME;
1215 nfs_set_cache_invalid(inode, invalid);
1217 nfsi->read_cache_jiffies = fattr->time_start;
1221 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1223 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1225 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1228 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1230 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1232 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1235 static atomic_long_t nfs_attr_generation_counter;
1237 static unsigned long nfs_read_attr_generation_counter(void)
1239 return atomic_long_read(&nfs_attr_generation_counter);
1242 unsigned long nfs_inc_attr_generation_counter(void)
1244 return atomic_long_inc_return(&nfs_attr_generation_counter);
1247 void nfs_fattr_init(struct nfs_fattr *fattr)
1250 fattr->time_start = jiffies;
1251 fattr->gencount = nfs_inc_attr_generation_counter();
1252 fattr->owner_name = NULL;
1253 fattr->group_name = NULL;
1255 EXPORT_SYMBOL_GPL(nfs_fattr_init);
1257 struct nfs_fattr *nfs_alloc_fattr(void)
1259 struct nfs_fattr *fattr;
1261 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1263 nfs_fattr_init(fattr);
1266 EXPORT_SYMBOL_GPL(nfs_alloc_fattr);
1268 struct nfs_fh *nfs_alloc_fhandle(void)
1272 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1277 EXPORT_SYMBOL_GPL(nfs_alloc_fhandle);
1281 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1282 * in the same way that wireshark does
1286 * For debugging only.
1288 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1290 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1291 * not on the result */
1292 return nfs_fhandle_hash(fh);
1294 EXPORT_SYMBOL_GPL(_nfs_display_fhandle_hash);
1297 * _nfs_display_fhandle - display an NFS file handle on the console
1299 * @fh: file handle to display
1300 * @caption: display caption
1302 * For debugging only.
1304 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1308 if (fh == NULL || fh->size == 0) {
1309 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1313 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1314 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1315 for (i = 0; i < fh->size; i += 16) {
1316 __be32 *pos = (__be32 *)&fh->data[i];
1318 switch ((fh->size - i - 1) >> 2) {
1320 printk(KERN_DEFAULT " %08x\n",
1324 printk(KERN_DEFAULT " %08x %08x\n",
1325 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1328 printk(KERN_DEFAULT " %08x %08x %08x\n",
1329 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1330 be32_to_cpup(pos + 2));
1333 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1334 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1335 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1339 EXPORT_SYMBOL_GPL(_nfs_display_fhandle);
1343 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1344 * @inode - pointer to inode
1345 * @fattr - attributes
1347 * Attempt to divine whether or not an RPC call reply carrying stale
1348 * attributes got scheduled after another call carrying updated ones.
1350 * To do so, the function first assumes that a more recent ctime means
1351 * that the attributes in fattr are newer, however it also attempt to
1352 * catch the case where ctime either didn't change, or went backwards
1353 * (if someone reset the clock on the server) by looking at whether
1354 * or not this RPC call was started after the inode was last updated.
1355 * Note also the check for wraparound of 'attr_gencount'
1357 * The function returns 'true' if it thinks the attributes in 'fattr' are
1358 * more recent than the ones cached in the inode.
1361 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1363 const struct nfs_inode *nfsi = NFS_I(inode);
1365 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1366 nfs_ctime_need_update(inode, fattr) ||
1367 nfs_size_need_update(inode, fattr) ||
1368 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1372 * Don't trust the change_attribute, mtime, ctime or size if
1373 * a pnfs LAYOUTCOMMIT is outstanding
1375 static void nfs_inode_attrs_handle_layoutcommit(struct inode *inode,
1376 struct nfs_fattr *fattr)
1378 if (pnfs_layoutcommit_outstanding(inode))
1379 fattr->valid &= ~(NFS_ATTR_FATTR_CHANGE |
1380 NFS_ATTR_FATTR_MTIME |
1381 NFS_ATTR_FATTR_CTIME |
1382 NFS_ATTR_FATTR_SIZE);
1385 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1389 trace_nfs_refresh_inode_enter(inode);
1391 nfs_inode_attrs_handle_layoutcommit(inode, fattr);
1393 if (nfs_inode_attrs_need_update(inode, fattr))
1394 ret = nfs_update_inode(inode, fattr);
1396 ret = nfs_check_inode_attributes(inode, fattr);
1398 trace_nfs_refresh_inode_exit(inode, ret);
1403 * nfs_refresh_inode - try to update the inode attribute cache
1404 * @inode - pointer to inode
1405 * @fattr - updated attributes
1407 * Check that an RPC call that returned attributes has not overlapped with
1408 * other recent updates of the inode metadata, then decide whether it is
1409 * safe to do a full update of the inode attributes, or whether just to
1410 * call nfs_check_inode_attributes.
1412 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1416 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1418 spin_lock(&inode->i_lock);
1419 status = nfs_refresh_inode_locked(inode, fattr);
1420 spin_unlock(&inode->i_lock);
1424 EXPORT_SYMBOL_GPL(nfs_refresh_inode);
1426 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1428 unsigned long invalid = NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1430 if (S_ISDIR(inode->i_mode))
1431 invalid |= NFS_INO_INVALID_DATA;
1432 nfs_set_cache_invalid(inode, invalid);
1433 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1435 return nfs_refresh_inode_locked(inode, fattr);
1439 * nfs_post_op_update_inode - try to update the inode attribute cache
1440 * @inode - pointer to inode
1441 * @fattr - updated attributes
1443 * After an operation that has changed the inode metadata, mark the
1444 * attribute cache as being invalid, then try to update it.
1446 * NB: if the server didn't return any post op attributes, this
1447 * function will force the retrieval of attributes before the next
1448 * NFS request. Thus it should be used only for operations that
1449 * are expected to change one or more attributes, to avoid
1450 * unnecessary NFS requests and trips through nfs_update_inode().
1452 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1456 spin_lock(&inode->i_lock);
1457 status = nfs_post_op_update_inode_locked(inode, fattr);
1458 spin_unlock(&inode->i_lock);
1462 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode);
1465 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1466 * @inode - pointer to inode
1467 * @fattr - updated attributes
1469 * After an operation that has changed the inode metadata, mark the
1470 * attribute cache as being invalid, then try to update it. Fake up
1471 * weak cache consistency data, if none exist.
1473 * This function is mainly designed to be used by the ->write_done() functions.
1475 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1479 spin_lock(&inode->i_lock);
1480 /* Don't do a WCC update if these attributes are already stale */
1481 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1482 !nfs_inode_attrs_need_update(inode, fattr)) {
1483 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1484 | NFS_ATTR_FATTR_PRESIZE
1485 | NFS_ATTR_FATTR_PREMTIME
1486 | NFS_ATTR_FATTR_PRECTIME);
1489 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1490 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1491 fattr->pre_change_attr = inode->i_version;
1492 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1494 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1495 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1496 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1497 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1499 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1500 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1501 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1502 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1504 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1505 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1506 fattr->pre_size = i_size_read(inode);
1507 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1510 status = nfs_post_op_update_inode_locked(inode, fattr);
1511 spin_unlock(&inode->i_lock);
1514 EXPORT_SYMBOL_GPL(nfs_post_op_update_inode_force_wcc);
1517 * Many nfs protocol calls return the new file attributes after
1518 * an operation. Here we update the inode to reflect the state
1519 * of the server's inode.
1521 * This is a bit tricky because we have to make sure all dirty pages
1522 * have been sent off to the server before calling invalidate_inode_pages.
1523 * To make sure no other process adds more write requests while we try
1524 * our best to flush them, we make them sleep during the attribute refresh.
1526 * A very similar scenario holds for the dir cache.
1528 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1530 struct nfs_server *server;
1531 struct nfs_inode *nfsi = NFS_I(inode);
1532 loff_t cur_isize, new_isize;
1533 unsigned long invalid = 0;
1534 unsigned long now = jiffies;
1535 unsigned long save_cache_validity;
1537 dfprintk(VFS, "NFS: %s(%s/%lu fh_crc=0x%08x ct=%d info=0x%x)\n",
1538 __func__, inode->i_sb->s_id, inode->i_ino,
1539 nfs_display_fhandle_hash(NFS_FH(inode)),
1540 atomic_read(&inode->i_count), fattr->valid);
1542 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1543 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1544 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1545 NFS_SERVER(inode)->nfs_client->cl_hostname,
1546 inode->i_sb->s_id, (long long)nfsi->fileid,
1547 (long long)fattr->fileid);
1552 * Make sure the inode's type hasn't changed.
1554 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1556 * Big trouble! The inode has become a different object.
1558 printk(KERN_DEBUG "NFS: %s: inode %lu mode changed, %07o to %07o\n",
1559 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1563 server = NFS_SERVER(inode);
1564 /* Update the fsid? */
1565 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1566 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1567 !IS_AUTOMOUNT(inode))
1568 server->fsid = fattr->fsid;
1571 * Update the read time so we don't revalidate too often.
1573 nfsi->read_cache_jiffies = fattr->time_start;
1575 save_cache_validity = nfsi->cache_validity;
1576 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1577 | NFS_INO_INVALID_ATIME
1578 | NFS_INO_REVAL_FORCED
1579 | NFS_INO_REVAL_PAGECACHE);
1581 /* Do atomic weak cache consistency updates */
1582 invalid |= nfs_wcc_update_inode(inode, fattr);
1584 /* More cache consistency checks */
1585 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1586 if (inode->i_version != fattr->change_attr) {
1587 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1588 inode->i_sb->s_id, inode->i_ino);
1589 invalid |= NFS_INO_INVALID_ATTR
1590 | NFS_INO_INVALID_DATA
1591 | NFS_INO_INVALID_ACCESS
1592 | NFS_INO_INVALID_ACL
1593 | NFS_INO_REVAL_PAGECACHE;
1594 if (S_ISDIR(inode->i_mode))
1595 nfs_force_lookup_revalidate(inode);
1596 inode->i_version = fattr->change_attr;
1598 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1599 nfsi->cache_validity |= save_cache_validity;
1601 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1602 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1603 } else if (server->caps & NFS_CAP_MTIME)
1604 nfsi->cache_validity |= save_cache_validity &
1605 (NFS_INO_INVALID_ATTR
1606 | NFS_INO_REVAL_FORCED);
1608 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1609 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1610 } else if (server->caps & NFS_CAP_CTIME)
1611 nfsi->cache_validity |= save_cache_validity &
1612 (NFS_INO_INVALID_ATTR
1613 | NFS_INO_REVAL_FORCED);
1615 /* Check if our cached file size is stale */
1616 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1617 new_isize = nfs_size_to_loff_t(fattr->size);
1618 cur_isize = i_size_read(inode);
1619 if (new_isize != cur_isize) {
1620 /* Do we perhaps have any outstanding writes, or has
1621 * the file grown beyond our last write? */
1622 if ((nfsi->npages == 0) || new_isize > cur_isize) {
1623 i_size_write(inode, new_isize);
1624 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1625 invalid &= ~NFS_INO_REVAL_PAGECACHE;
1627 dprintk("NFS: isize change on server for file %s/%ld "
1631 (long long)cur_isize,
1632 (long long)new_isize);
1635 nfsi->cache_validity |= save_cache_validity &
1636 (NFS_INO_INVALID_ATTR
1637 | NFS_INO_REVAL_PAGECACHE
1638 | NFS_INO_REVAL_FORCED);
1641 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1642 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1643 else if (server->caps & NFS_CAP_ATIME)
1644 nfsi->cache_validity |= save_cache_validity &
1645 (NFS_INO_INVALID_ATIME
1646 | NFS_INO_REVAL_FORCED);
1648 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1649 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1650 umode_t newmode = inode->i_mode & S_IFMT;
1651 newmode |= fattr->mode & S_IALLUGO;
1652 inode->i_mode = newmode;
1653 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1655 } else if (server->caps & NFS_CAP_MODE)
1656 nfsi->cache_validity |= save_cache_validity &
1657 (NFS_INO_INVALID_ATTR
1658 | NFS_INO_INVALID_ACCESS
1659 | NFS_INO_INVALID_ACL
1660 | NFS_INO_REVAL_FORCED);
1662 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1663 if (!uid_eq(inode->i_uid, fattr->uid)) {
1664 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1665 inode->i_uid = fattr->uid;
1667 } else if (server->caps & NFS_CAP_OWNER)
1668 nfsi->cache_validity |= save_cache_validity &
1669 (NFS_INO_INVALID_ATTR
1670 | NFS_INO_INVALID_ACCESS
1671 | NFS_INO_INVALID_ACL
1672 | NFS_INO_REVAL_FORCED);
1674 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1675 if (!gid_eq(inode->i_gid, fattr->gid)) {
1676 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1677 inode->i_gid = fattr->gid;
1679 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1680 nfsi->cache_validity |= save_cache_validity &
1681 (NFS_INO_INVALID_ATTR
1682 | NFS_INO_INVALID_ACCESS
1683 | NFS_INO_INVALID_ACL
1684 | NFS_INO_REVAL_FORCED);
1686 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1687 if (inode->i_nlink != fattr->nlink) {
1688 invalid |= NFS_INO_INVALID_ATTR;
1689 if (S_ISDIR(inode->i_mode))
1690 invalid |= NFS_INO_INVALID_DATA;
1691 set_nlink(inode, fattr->nlink);
1693 } else if (server->caps & NFS_CAP_NLINK)
1694 nfsi->cache_validity |= save_cache_validity &
1695 (NFS_INO_INVALID_ATTR
1696 | NFS_INO_REVAL_FORCED);
1698 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1700 * report the blocks in 512byte units
1702 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1704 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1705 inode->i_blocks = fattr->du.nfs2.blocks;
1707 /* Update attrtimeo value if we're out of the unstable period */
1708 if (invalid & NFS_INO_INVALID_ATTR) {
1709 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1710 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1711 nfsi->attrtimeo_timestamp = now;
1712 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1714 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1715 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1716 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1717 nfsi->attrtimeo_timestamp = now;
1720 invalid &= ~NFS_INO_INVALID_ATTR;
1721 /* Don't invalidate the data if we were to blame */
1722 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1723 || S_ISLNK(inode->i_mode)))
1724 invalid &= ~NFS_INO_INVALID_DATA;
1725 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1726 (save_cache_validity & NFS_INO_REVAL_FORCED))
1727 nfs_set_cache_invalid(inode, invalid);
1732 * No need to worry about unhashing the dentry, as the
1733 * lookup validation will know that the inode is bad.
1734 * (But we fall through to invalidate the caches.)
1736 nfs_invalidate_inode(inode);
1740 struct inode *nfs_alloc_inode(struct super_block *sb)
1742 struct nfs_inode *nfsi;
1743 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1747 nfsi->cache_validity = 0UL;
1748 #if IS_ENABLED(CONFIG_NFS_V4)
1749 nfsi->nfs4_acl = NULL;
1750 #endif /* CONFIG_NFS_V4 */
1751 return &nfsi->vfs_inode;
1753 EXPORT_SYMBOL_GPL(nfs_alloc_inode);
1755 static void nfs_i_callback(struct rcu_head *head)
1757 struct inode *inode = container_of(head, struct inode, i_rcu);
1758 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1761 void nfs_destroy_inode(struct inode *inode)
1763 call_rcu(&inode->i_rcu, nfs_i_callback);
1765 EXPORT_SYMBOL_GPL(nfs_destroy_inode);
1767 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1769 #if IS_ENABLED(CONFIG_NFS_V4)
1770 INIT_LIST_HEAD(&nfsi->open_states);
1771 nfsi->delegation = NULL;
1772 nfsi->delegation_state = 0;
1773 init_rwsem(&nfsi->rwsem);
1774 nfsi->layout = NULL;
1778 static void init_once(void *foo)
1780 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1782 inode_init_once(&nfsi->vfs_inode);
1783 INIT_LIST_HEAD(&nfsi->open_files);
1784 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1785 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1786 INIT_LIST_HEAD(&nfsi->commit_info.list);
1788 nfsi->commit_info.ncommit = 0;
1789 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1790 atomic_set(&nfsi->silly_count, 1);
1791 INIT_HLIST_HEAD(&nfsi->silly_list);
1792 init_waitqueue_head(&nfsi->waitqueue);
1793 nfs4_init_once(nfsi);
1796 static int __init nfs_init_inodecache(void)
1798 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1799 sizeof(struct nfs_inode),
1800 0, (SLAB_RECLAIM_ACCOUNT|
1803 if (nfs_inode_cachep == NULL)
1809 static void nfs_destroy_inodecache(void)
1812 * Make sure all delayed rcu free inodes are flushed before we
1816 kmem_cache_destroy(nfs_inode_cachep);
1819 struct workqueue_struct *nfsiod_workqueue;
1820 EXPORT_SYMBOL_GPL(nfsiod_workqueue);
1823 * start up the nfsiod workqueue
1825 static int nfsiod_start(void)
1827 struct workqueue_struct *wq;
1828 dprintk("RPC: creating workqueue nfsiod\n");
1829 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1832 nfsiod_workqueue = wq;
1837 * Destroy the nfsiod workqueue
1839 static void nfsiod_stop(void)
1841 struct workqueue_struct *wq;
1843 wq = nfsiod_workqueue;
1846 nfsiod_workqueue = NULL;
1847 destroy_workqueue(wq);
1851 EXPORT_SYMBOL_GPL(nfs_net_id);
1853 static int nfs_net_init(struct net *net)
1855 nfs_clients_init(net);
1856 return nfs_fs_proc_net_init(net);
1859 static void nfs_net_exit(struct net *net)
1861 nfs_fs_proc_net_exit(net);
1862 nfs_cleanup_cb_ident_idr(net);
1865 static struct pernet_operations nfs_net_ops = {
1866 .init = nfs_net_init,
1867 .exit = nfs_net_exit,
1869 .size = sizeof(struct nfs_net),
1875 static int __init init_nfs_fs(void)
1879 err = register_pernet_subsys(&nfs_net_ops);
1883 err = nfs_fscache_register();
1887 err = nfsiod_start();
1891 err = nfs_fs_proc_init();
1895 err = nfs_init_nfspagecache();
1899 err = nfs_init_inodecache();
1903 err = nfs_init_readpagecache();
1907 err = nfs_init_writepagecache();
1911 err = nfs_init_directcache();
1915 #ifdef CONFIG_PROC_FS
1916 rpc_proc_register(&init_net, &nfs_rpcstat);
1918 if ((err = register_nfs_fs()) != 0)
1923 #ifdef CONFIG_PROC_FS
1924 rpc_proc_unregister(&init_net, "nfs");
1926 nfs_destroy_directcache();
1928 nfs_destroy_writepagecache();
1930 nfs_destroy_readpagecache();
1932 nfs_destroy_inodecache();
1934 nfs_destroy_nfspagecache();
1940 nfs_fscache_unregister();
1942 unregister_pernet_subsys(&nfs_net_ops);
1947 static void __exit exit_nfs_fs(void)
1949 nfs_destroy_directcache();
1950 nfs_destroy_writepagecache();
1951 nfs_destroy_readpagecache();
1952 nfs_destroy_inodecache();
1953 nfs_destroy_nfspagecache();
1954 nfs_fscache_unregister();
1955 unregister_pernet_subsys(&nfs_net_ops);
1956 #ifdef CONFIG_PROC_FS
1957 rpc_proc_unregister(&init_net, "nfs");
1959 unregister_nfs_fs();
1964 /* Not quite true; I just maintain it */
1966 MODULE_LICENSE("GPL");
1967 module_param(enable_ino64, bool, 0644);
1969 module_init(init_nfs_fs)
1970 module_exit(exit_nfs_fs)