4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.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/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/inet.h>
39 #include <linux/nfs_xdr.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
50 #define NFSDBG_FACILITY NFSDBG_VFS
51 #define NFS_PARANOIA 1
53 static void nfs_invalidate_inode(struct inode *);
54 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
56 static void nfs_zap_acl_cache(struct inode *);
58 static struct kmem_cache * nfs_inode_cachep;
60 static inline unsigned long
61 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
63 return nfs_fileid_to_ino_t(fattr->fileid);
66 int nfs_write_inode(struct inode *inode, int sync)
71 ret = filemap_fdatawait(inode->i_mapping);
73 ret = nfs_commit_inode(inode, FLUSH_SYNC);
75 ret = nfs_commit_inode(inode, 0);
78 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
82 void nfs_clear_inode(struct inode *inode)
85 * The following should never happen...
87 BUG_ON(nfs_have_writebacks(inode));
88 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
89 BUG_ON(atomic_read(&NFS_I(inode)->data_updates) != 0);
90 nfs_zap_acl_cache(inode);
91 nfs_access_zap_cache(inode);
95 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
97 int nfs_sync_mapping(struct address_space *mapping)
101 if (mapping->nrpages == 0)
103 unmap_mapping_range(mapping, 0, 0, 0);
104 ret = filemap_write_and_wait(mapping);
107 ret = nfs_wb_all(mapping->host);
113 * Invalidate the local caches
115 static void nfs_zap_caches_locked(struct inode *inode)
117 struct nfs_inode *nfsi = NFS_I(inode);
118 int mode = inode->i_mode;
120 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
122 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
123 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
125 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
126 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
127 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
129 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
132 void nfs_zap_caches(struct inode *inode)
134 spin_lock(&inode->i_lock);
135 nfs_zap_caches_locked(inode);
136 spin_unlock(&inode->i_lock);
139 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
141 if (mapping->nrpages != 0) {
142 spin_lock(&inode->i_lock);
143 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
144 spin_unlock(&inode->i_lock);
148 static void nfs_zap_acl_cache(struct inode *inode)
150 void (*clear_acl_cache)(struct inode *);
152 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
153 if (clear_acl_cache != NULL)
154 clear_acl_cache(inode);
155 spin_lock(&inode->i_lock);
156 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
157 spin_unlock(&inode->i_lock);
161 * Invalidate, but do not unhash, the inode.
162 * NB: must be called with inode->i_lock held!
164 static void nfs_invalidate_inode(struct inode *inode)
166 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
167 nfs_zap_caches_locked(inode);
170 struct nfs_find_desc {
172 struct nfs_fattr *fattr;
176 * In NFSv3 we can have 64bit inode numbers. In order to support
177 * this, and re-exported directories (also seen in NFSv2)
178 * we are forced to allow 2 different inodes to have the same
182 nfs_find_actor(struct inode *inode, void *opaque)
184 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
185 struct nfs_fh *fh = desc->fh;
186 struct nfs_fattr *fattr = desc->fattr;
188 if (NFS_FILEID(inode) != fattr->fileid)
190 if (nfs_compare_fh(NFS_FH(inode), fh))
192 if (is_bad_inode(inode) || NFS_STALE(inode))
198 nfs_init_locked(struct inode *inode, void *opaque)
200 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
201 struct nfs_fattr *fattr = desc->fattr;
203 NFS_FILEID(inode) = fattr->fileid;
204 nfs_copy_fh(NFS_FH(inode), desc->fh);
208 /* Don't use READDIRPLUS on directories that we believe are too large */
209 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
212 * This is our front-end to iget that looks up inodes by file handle
213 * instead of inode number.
216 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
218 struct nfs_find_desc desc = {
222 struct inode *inode = ERR_PTR(-ENOENT);
225 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
229 printk("NFS: Buggy server - nlink == 0!\n");
233 hash = nfs_fattr_to_ino_t(fattr);
235 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
237 inode = ERR_PTR(-ENOMEM);
241 if (inode->i_state & I_NEW) {
242 struct nfs_inode *nfsi = NFS_I(inode);
243 unsigned long now = jiffies;
245 /* We set i_ino for the few things that still rely on it,
249 /* We can't support update_atime(), since the server will reset it */
250 inode->i_flags |= S_NOATIME|S_NOCMTIME;
251 inode->i_mode = fattr->mode;
252 /* Why so? Because we want revalidate for devices/FIFOs, and
253 * that's precisely what we have in nfs_file_inode_operations.
255 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
256 if (S_ISREG(inode->i_mode)) {
257 inode->i_fop = &nfs_file_operations;
258 inode->i_data.a_ops = &nfs_file_aops;
259 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
260 } else if (S_ISDIR(inode->i_mode)) {
261 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
262 inode->i_fop = &nfs_dir_operations;
263 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
264 && fattr->size <= NFS_LIMIT_READDIRPLUS)
265 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
266 /* Deal with crossing mountpoints */
267 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
268 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
269 inode->i_op = &nfs_referral_inode_operations;
271 inode->i_op = &nfs_mountpoint_inode_operations;
274 } else if (S_ISLNK(inode->i_mode))
275 inode->i_op = &nfs_symlink_inode_operations;
277 init_special_inode(inode, inode->i_mode, fattr->rdev);
279 nfsi->read_cache_jiffies = fattr->time_start;
280 nfsi->last_updated = now;
281 nfsi->cache_change_attribute = now;
282 inode->i_atime = fattr->atime;
283 inode->i_mtime = fattr->mtime;
284 inode->i_ctime = fattr->ctime;
285 if (fattr->valid & NFS_ATTR_FATTR_V4)
286 nfsi->change_attr = fattr->change_attr;
287 inode->i_size = nfs_size_to_loff_t(fattr->size);
288 inode->i_nlink = fattr->nlink;
289 inode->i_uid = fattr->uid;
290 inode->i_gid = fattr->gid;
291 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
293 * report the blocks in 512byte units
295 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
297 inode->i_blocks = fattr->du.nfs2.blocks;
299 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
300 nfsi->attrtimeo_timestamp = now;
301 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
302 nfsi->access_cache = RB_ROOT;
304 unlock_new_inode(inode);
306 nfs_refresh_inode(inode, fattr);
307 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
309 (long long)NFS_FILEID(inode),
310 atomic_read(&inode->i_count));
316 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
320 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
323 nfs_setattr(struct dentry *dentry, struct iattr *attr)
325 struct inode *inode = dentry->d_inode;
326 struct nfs_fattr fattr;
329 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
331 if (attr->ia_valid & ATTR_SIZE) {
332 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
333 attr->ia_valid &= ~ATTR_SIZE;
336 /* Optimization: if the end result is no change, don't RPC */
337 attr->ia_valid &= NFS_VALID_ATTRS;
338 if (attr->ia_valid == 0)
342 nfs_begin_data_update(inode);
343 /* Write all dirty data */
344 filemap_write_and_wait(inode->i_mapping);
347 * Return any delegations if we're going to change ACLs
349 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
350 nfs_inode_return_delegation(inode);
351 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
353 nfs_refresh_inode(inode, &fattr);
354 nfs_end_data_update(inode);
360 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
361 * @inode: pointer to struct inode
362 * @attr: pointer to struct iattr
364 * Note: we do this in the *proc.c in order to ensure that
365 * it works for things like exclusive creates too.
367 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
369 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
370 if ((attr->ia_valid & ATTR_MODE) != 0) {
371 int mode = attr->ia_mode & S_IALLUGO;
372 mode |= inode->i_mode & ~S_IALLUGO;
373 inode->i_mode = mode;
375 if ((attr->ia_valid & ATTR_UID) != 0)
376 inode->i_uid = attr->ia_uid;
377 if ((attr->ia_valid & ATTR_GID) != 0)
378 inode->i_gid = attr->ia_gid;
379 spin_lock(&inode->i_lock);
380 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
381 spin_unlock(&inode->i_lock);
383 if ((attr->ia_valid & ATTR_SIZE) != 0) {
384 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
385 inode->i_size = attr->ia_size;
386 vmtruncate(inode, attr->ia_size);
390 static int nfs_wait_schedule(void *word)
392 if (signal_pending(current))
399 * Wait for the inode to get unlocked.
401 static int nfs_wait_on_inode(struct inode *inode)
403 struct rpc_clnt *clnt = NFS_CLIENT(inode);
404 struct nfs_inode *nfsi = NFS_I(inode);
408 rpc_clnt_sigmask(clnt, &oldmask);
409 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
410 nfs_wait_schedule, TASK_INTERRUPTIBLE);
411 rpc_clnt_sigunmask(clnt, &oldmask);
416 static void nfs_wake_up_inode(struct inode *inode)
418 struct nfs_inode *nfsi = NFS_I(inode);
420 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
421 smp_mb__after_clear_bit();
422 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
425 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
427 struct inode *inode = dentry->d_inode;
428 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
431 /* Flush out writes to the server in order to update c/mtime */
432 if (S_ISREG(inode->i_mode))
433 nfs_sync_mapping_range(inode->i_mapping, 0, 0, FLUSH_NOCOMMIT);
436 * We may force a getattr if the user cares about atime.
438 * Note that we only have to check the vfsmount flags here:
439 * - NFS always sets S_NOATIME by so checking it would give a
441 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
442 * no point in checking those.
444 if ((mnt->mnt_flags & MNT_NOATIME) ||
445 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
449 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
451 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
453 generic_fillattr(inode, stat);
457 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
459 struct nfs_open_context *ctx;
461 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
463 atomic_set(&ctx->count, 1);
464 ctx->dentry = dget(dentry);
465 ctx->vfsmnt = mntget(mnt);
466 ctx->cred = get_rpccred(cred);
468 ctx->lockowner = current->files;
475 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
478 atomic_inc(&ctx->count);
482 void put_nfs_open_context(struct nfs_open_context *ctx)
484 if (atomic_dec_and_test(&ctx->count)) {
485 if (!list_empty(&ctx->list)) {
486 struct inode *inode = ctx->dentry->d_inode;
487 spin_lock(&inode->i_lock);
488 list_del(&ctx->list);
489 spin_unlock(&inode->i_lock);
491 if (ctx->state != NULL)
492 nfs4_close_state(ctx->state, ctx->mode);
493 if (ctx->cred != NULL)
494 put_rpccred(ctx->cred);
502 * Ensure that mmap has a recent RPC credential for use when writing out
505 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
507 struct inode *inode = filp->f_path.dentry->d_inode;
508 struct nfs_inode *nfsi = NFS_I(inode);
510 filp->private_data = get_nfs_open_context(ctx);
511 spin_lock(&inode->i_lock);
512 list_add(&ctx->list, &nfsi->open_files);
513 spin_unlock(&inode->i_lock);
517 * Given an inode, search for an open context with the desired characteristics
519 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
521 struct nfs_inode *nfsi = NFS_I(inode);
522 struct nfs_open_context *pos, *ctx = NULL;
524 spin_lock(&inode->i_lock);
525 list_for_each_entry(pos, &nfsi->open_files, list) {
526 if (cred != NULL && pos->cred != cred)
528 if ((pos->mode & mode) == mode) {
529 ctx = get_nfs_open_context(pos);
533 spin_unlock(&inode->i_lock);
537 static void nfs_file_clear_open_context(struct file *filp)
539 struct inode *inode = filp->f_path.dentry->d_inode;
540 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
543 filp->private_data = NULL;
544 spin_lock(&inode->i_lock);
545 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
546 spin_unlock(&inode->i_lock);
547 put_nfs_open_context(ctx);
552 * These allocate and release file read/write context information.
554 int nfs_open(struct inode *inode, struct file *filp)
556 struct nfs_open_context *ctx;
557 struct rpc_cred *cred;
559 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
561 return PTR_ERR(cred);
562 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
566 ctx->mode = filp->f_mode;
567 nfs_file_set_open_context(filp, ctx);
568 put_nfs_open_context(ctx);
572 int nfs_release(struct inode *inode, struct file *filp)
574 nfs_file_clear_open_context(filp);
579 * This function is called whenever some part of NFS notices that
580 * the cached attributes have to be refreshed.
583 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
585 int status = -ESTALE;
586 struct nfs_fattr fattr;
587 struct nfs_inode *nfsi = NFS_I(inode);
589 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
590 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
592 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
594 if (is_bad_inode(inode))
596 if (NFS_STALE(inode))
599 status = nfs_wait_on_inode(inode);
602 if (NFS_STALE(inode)) {
604 /* Do we trust the cached ESTALE? */
605 if (NFS_ATTRTIMEO(inode) != 0) {
606 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
613 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
615 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
617 (long long)NFS_FILEID(inode), status);
618 if (status == -ESTALE) {
619 nfs_zap_caches(inode);
620 if (!S_ISDIR(inode->i_mode))
621 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
626 spin_lock(&inode->i_lock);
627 status = nfs_update_inode(inode, &fattr);
629 spin_unlock(&inode->i_lock);
630 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
632 (long long)NFS_FILEID(inode), status);
635 spin_unlock(&inode->i_lock);
637 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
638 nfs_zap_acl_cache(inode);
640 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
642 (long long)NFS_FILEID(inode));
645 nfs_wake_up_inode(inode);
652 int nfs_attribute_timeout(struct inode *inode)
654 struct nfs_inode *nfsi = NFS_I(inode);
656 if (nfs_have_delegation(inode, FMODE_READ))
658 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
662 * nfs_revalidate_inode - Revalidate the inode attributes
663 * @server - pointer to nfs_server struct
664 * @inode - pointer to inode struct
666 * Updates inode attribute information by retrieving the data from the server.
668 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
670 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
671 && !nfs_attribute_timeout(inode))
672 return NFS_STALE(inode) ? -ESTALE : 0;
673 return __nfs_revalidate_inode(server, inode);
676 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
678 struct nfs_inode *nfsi = NFS_I(inode);
680 if (mapping->nrpages != 0) {
681 int ret = invalidate_inode_pages2(mapping);
685 spin_lock(&inode->i_lock);
686 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
687 if (S_ISDIR(inode->i_mode)) {
688 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
689 /* This ensures we revalidate child dentries */
690 nfsi->cache_change_attribute = jiffies;
692 spin_unlock(&inode->i_lock);
693 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
694 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
695 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
699 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
703 mutex_lock(&inode->i_mutex);
704 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
705 ret = nfs_sync_mapping(mapping);
707 ret = nfs_invalidate_mapping_nolock(inode, mapping);
709 mutex_unlock(&inode->i_mutex);
714 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
715 * @inode - pointer to host inode
716 * @mapping - pointer to mapping
718 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
720 struct nfs_inode *nfsi = NFS_I(inode);
723 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
724 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
725 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
729 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
730 ret = nfs_invalidate_mapping_nolock(inode, mapping);
736 * nfs_revalidate_mapping - Revalidate the pagecache
737 * @inode - pointer to host inode
738 * @mapping - pointer to mapping
740 * This version of the function will take the inode->i_mutex and attempt to
741 * flush out all dirty data if it needs to invalidate the page cache.
743 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
745 struct nfs_inode *nfsi = NFS_I(inode);
748 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
749 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
750 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
754 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
755 ret = nfs_invalidate_mapping(inode, mapping);
761 * nfs_begin_data_update
762 * @inode - pointer to inode
763 * Declare that a set of operations will update file data on the server
765 void nfs_begin_data_update(struct inode *inode)
767 atomic_inc(&NFS_I(inode)->data_updates);
771 * nfs_end_data_update
772 * @inode - pointer to inode
773 * Declare end of the operations that will update file data
774 * This will mark the inode as immediately needing revalidation
775 * of its attribute cache.
777 void nfs_end_data_update(struct inode *inode)
779 struct nfs_inode *nfsi = NFS_I(inode);
781 /* Directories: invalidate page cache */
782 if (S_ISDIR(inode->i_mode)) {
783 spin_lock(&inode->i_lock);
784 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
785 spin_unlock(&inode->i_lock);
787 nfsi->cache_change_attribute = jiffies;
788 atomic_dec(&nfsi->data_updates);
791 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
793 struct nfs_inode *nfsi = NFS_I(inode);
794 unsigned long now = jiffies;
796 /* If we have atomic WCC data, we may update some attributes */
797 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
798 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
799 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
800 nfsi->cache_change_attribute = now;
802 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
803 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
804 nfsi->cache_change_attribute = now;
806 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
807 inode->i_size = fattr->size;
808 nfsi->cache_change_attribute = now;
814 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
815 * @inode - pointer to inode
816 * @fattr - updated attributes
818 * Verifies the attribute cache. If we have just changed the attributes,
819 * so that fattr carries weak cache consistency data, then it may
820 * also update the ctime/mtime/change_attribute.
822 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
824 struct nfs_inode *nfsi = NFS_I(inode);
825 loff_t cur_size, new_isize;
829 /* Has the inode gone and changed behind our back? */
830 if (nfsi->fileid != fattr->fileid
831 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
835 /* Are we in the process of updating data on the server? */
836 data_unstable = nfs_caches_unstable(inode);
838 /* Do atomic weak cache consistency updates */
839 nfs_wcc_update_inode(inode, fattr);
841 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
842 nfsi->change_attr != fattr->change_attr)
843 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
845 /* Verify a few of the more important attributes */
846 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
847 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
849 cur_size = i_size_read(inode);
850 new_isize = nfs_size_to_loff_t(fattr->size);
851 if (cur_size != new_isize && nfsi->npages == 0)
852 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
854 /* Have any file permissions changed? */
855 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
856 || inode->i_uid != fattr->uid
857 || inode->i_gid != fattr->gid)
858 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
860 /* Has the link count changed? */
861 if (inode->i_nlink != fattr->nlink)
862 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
864 if (!timespec_equal(&inode->i_atime, &fattr->atime))
865 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
867 nfsi->read_cache_jiffies = fattr->time_start;
872 * nfs_refresh_inode - try to update the inode attribute cache
873 * @inode - pointer to inode
874 * @fattr - updated attributes
876 * Check that an RPC call that returned attributes has not overlapped with
877 * other recent updates of the inode metadata, then decide whether it is
878 * safe to do a full update of the inode attributes, or whether just to
879 * call nfs_check_inode_attributes.
881 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
883 struct nfs_inode *nfsi = NFS_I(inode);
886 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
888 spin_lock(&inode->i_lock);
889 if (time_after(fattr->time_start, nfsi->last_updated))
890 status = nfs_update_inode(inode, fattr);
892 status = nfs_check_inode_attributes(inode, fattr);
894 spin_unlock(&inode->i_lock);
899 * nfs_post_op_update_inode - try to update the inode attribute cache
900 * @inode - pointer to inode
901 * @fattr - updated attributes
903 * After an operation that has changed the inode metadata, mark the
904 * attribute cache as being invalid, then try to update it.
906 * NB: if the server didn't return any post op attributes, this
907 * function will force the retrieval of attributes before the next
908 * NFS request. Thus it should be used only for operations that
909 * are expected to change one or more attributes, to avoid
910 * unnecessary NFS requests and trips through nfs_update_inode().
912 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
914 struct nfs_inode *nfsi = NFS_I(inode);
917 spin_lock(&inode->i_lock);
918 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
919 nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
922 status = nfs_update_inode(inode, fattr);
924 spin_unlock(&inode->i_lock);
929 * Many nfs protocol calls return the new file attributes after
930 * an operation. Here we update the inode to reflect the state
931 * of the server's inode.
933 * This is a bit tricky because we have to make sure all dirty pages
934 * have been sent off to the server before calling invalidate_inode_pages.
935 * To make sure no other process adds more write requests while we try
936 * our best to flush them, we make them sleep during the attribute refresh.
938 * A very similar scenario holds for the dir cache.
940 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
942 struct nfs_server *server;
943 struct nfs_inode *nfsi = NFS_I(inode);
944 loff_t cur_isize, new_isize;
945 unsigned int invalid = 0;
946 unsigned long now = jiffies;
949 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
950 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
951 atomic_read(&inode->i_count), fattr->valid);
953 if (nfsi->fileid != fattr->fileid)
957 * Make sure the inode's type hasn't changed.
959 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
962 server = NFS_SERVER(inode);
963 /* Update the fsid if and only if this is the root directory */
964 if (inode == inode->i_sb->s_root->d_inode
965 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
966 server->fsid = fattr->fsid;
969 * Update the read time so we don't revalidate too often.
971 nfsi->read_cache_jiffies = fattr->time_start;
972 nfsi->last_updated = now;
974 /* Fix a wraparound issue with nfsi->cache_change_attribute */
975 if (time_before(now, nfsi->cache_change_attribute))
976 nfsi->cache_change_attribute = now - 600*HZ;
978 /* Are we racing with known updates of the metadata on the server? */
979 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
981 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
983 /* Do atomic weak cache consistency updates */
984 nfs_wcc_update_inode(inode, fattr);
986 /* Check if our cached file size is stale */
987 new_isize = nfs_size_to_loff_t(fattr->size);
988 cur_isize = i_size_read(inode);
989 if (new_isize != cur_isize) {
990 /* Do we perhaps have any outstanding writes? */
991 if (nfsi->npages == 0) {
992 /* No, but did we race with nfs_end_data_update()? */
994 inode->i_size = new_isize;
995 invalid |= NFS_INO_INVALID_DATA;
997 invalid |= NFS_INO_INVALID_ATTR;
998 } else if (new_isize > cur_isize) {
999 inode->i_size = new_isize;
1000 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1002 nfsi->cache_change_attribute = now;
1003 dprintk("NFS: isize change on server for file %s/%ld\n",
1004 inode->i_sb->s_id, inode->i_ino);
1007 /* Check if the mtime agrees */
1008 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1009 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1010 dprintk("NFS: mtime change on server for file %s/%ld\n",
1011 inode->i_sb->s_id, inode->i_ino);
1012 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1013 nfsi->cache_change_attribute = now;
1016 /* If ctime has changed we should definitely clear access+acl caches */
1017 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1018 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1019 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1020 nfsi->cache_change_attribute = now;
1022 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1024 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1025 inode->i_uid != fattr->uid ||
1026 inode->i_gid != fattr->gid)
1027 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1029 inode->i_mode = fattr->mode;
1030 inode->i_nlink = fattr->nlink;
1031 inode->i_uid = fattr->uid;
1032 inode->i_gid = fattr->gid;
1034 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1036 * report the blocks in 512byte units
1038 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1040 inode->i_blocks = fattr->du.nfs2.blocks;
1043 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1044 nfsi->change_attr != fattr->change_attr) {
1045 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1046 inode->i_sb->s_id, inode->i_ino);
1047 nfsi->change_attr = fattr->change_attr;
1048 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1049 nfsi->cache_change_attribute = now;
1052 /* Update attrtimeo value if we're out of the unstable period */
1053 if (invalid & NFS_INO_INVALID_ATTR) {
1054 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1055 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1056 nfsi->attrtimeo_timestamp = now;
1057 } else if (time_after(now, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1058 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1059 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1060 nfsi->attrtimeo_timestamp = now;
1062 /* Don't invalidate the data if we were to blame */
1063 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1064 || S_ISLNK(inode->i_mode)))
1065 invalid &= ~NFS_INO_INVALID_DATA;
1067 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1068 if (!nfs_have_delegation(inode, FMODE_READ))
1069 nfsi->cache_validity |= invalid;
1074 * Big trouble! The inode has become a different object.
1077 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1078 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1082 * No need to worry about unhashing the dentry, as the
1083 * lookup validation will know that the inode is bad.
1084 * (But we fall through to invalidate the caches.)
1086 nfs_invalidate_inode(inode);
1090 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1091 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1092 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1093 (long long)nfsi->fileid, (long long)fattr->fileid);
1098 #ifdef CONFIG_NFS_V4
1101 * Clean out any remaining NFSv4 state that might be left over due
1102 * to open() calls that passed nfs_atomic_lookup, but failed to call
1105 void nfs4_clear_inode(struct inode *inode)
1107 struct nfs_inode *nfsi = NFS_I(inode);
1109 /* If we are holding a delegation, return it! */
1110 nfs_inode_return_delegation(inode);
1111 /* First call standard NFS clear_inode() code */
1112 nfs_clear_inode(inode);
1113 /* Now clear out any remaining state */
1114 while (!list_empty(&nfsi->open_states)) {
1115 struct nfs4_state *state;
1117 state = list_entry(nfsi->open_states.next,
1120 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1123 (long long)NFS_FILEID(inode),
1125 BUG_ON(atomic_read(&state->count) != 1);
1126 nfs4_close_state(state, state->state);
1131 struct inode *nfs_alloc_inode(struct super_block *sb)
1133 struct nfs_inode *nfsi;
1134 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1138 nfsi->cache_validity = 0UL;
1139 #ifdef CONFIG_NFS_V3_ACL
1140 nfsi->acl_access = ERR_PTR(-EAGAIN);
1141 nfsi->acl_default = ERR_PTR(-EAGAIN);
1143 #ifdef CONFIG_NFS_V4
1144 nfsi->nfs4_acl = NULL;
1145 #endif /* CONFIG_NFS_V4 */
1146 return &nfsi->vfs_inode;
1149 void nfs_destroy_inode(struct inode *inode)
1151 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1154 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1156 #ifdef CONFIG_NFS_V4
1157 INIT_LIST_HEAD(&nfsi->open_states);
1158 nfsi->delegation = NULL;
1159 nfsi->delegation_state = 0;
1160 init_rwsem(&nfsi->rwsem);
1164 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1166 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1168 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1169 SLAB_CTOR_CONSTRUCTOR) {
1170 inode_init_once(&nfsi->vfs_inode);
1171 spin_lock_init(&nfsi->req_lock);
1172 INIT_LIST_HEAD(&nfsi->dirty);
1173 INIT_LIST_HEAD(&nfsi->commit);
1174 INIT_LIST_HEAD(&nfsi->open_files);
1175 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1176 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1177 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1178 atomic_set(&nfsi->data_updates, 0);
1182 nfs4_init_once(nfsi);
1186 static int __init nfs_init_inodecache(void)
1188 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1189 sizeof(struct nfs_inode),
1190 0, (SLAB_RECLAIM_ACCOUNT|
1193 if (nfs_inode_cachep == NULL)
1199 static void nfs_destroy_inodecache(void)
1201 kmem_cache_destroy(nfs_inode_cachep);
1207 static int __init init_nfs_fs(void)
1211 err = nfs_fs_proc_init();
1215 err = nfs_init_nfspagecache();
1219 err = nfs_init_inodecache();
1223 err = nfs_init_readpagecache();
1227 err = nfs_init_writepagecache();
1231 err = nfs_init_directcache();
1235 #ifdef CONFIG_PROC_FS
1236 rpc_proc_register(&nfs_rpcstat);
1238 if ((err = register_nfs_fs()) != 0)
1242 #ifdef CONFIG_PROC_FS
1243 rpc_proc_unregister("nfs");
1245 nfs_destroy_directcache();
1247 nfs_destroy_writepagecache();
1249 nfs_destroy_readpagecache();
1251 nfs_destroy_inodecache();
1253 nfs_destroy_nfspagecache();
1260 static void __exit exit_nfs_fs(void)
1262 nfs_destroy_directcache();
1263 nfs_destroy_writepagecache();
1264 nfs_destroy_readpagecache();
1265 nfs_destroy_inodecache();
1266 nfs_destroy_nfspagecache();
1267 #ifdef CONFIG_PROC_FS
1268 rpc_proc_unregister("nfs");
1270 unregister_nfs_fs();
1274 /* Not quite true; I just maintain it */
1275 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1276 MODULE_LICENSE("GPL");
1278 module_init(init_nfs_fs)
1279 module_exit(exit_nfs_fs)