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>
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/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
52 static void nfs_invalidate_inode(struct inode *);
53 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
55 static void nfs_zap_acl_cache(struct inode *);
57 static struct kmem_cache * nfs_inode_cachep;
59 static inline unsigned long
60 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
62 return nfs_fileid_to_ino_t(fattr->fileid);
65 int nfs_write_inode(struct inode *inode, int sync)
70 ret = filemap_fdatawait(inode->i_mapping);
72 ret = nfs_commit_inode(inode, FLUSH_SYNC);
74 ret = nfs_commit_inode(inode, 0);
77 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
81 void nfs_clear_inode(struct inode *inode)
84 * The following should never happen...
86 BUG_ON(nfs_have_writebacks(inode));
87 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
88 BUG_ON(atomic_read(&NFS_I(inode)->data_updates) != 0);
89 nfs_zap_acl_cache(inode);
90 nfs_access_zap_cache(inode);
94 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
96 int nfs_sync_mapping(struct address_space *mapping)
100 if (mapping->nrpages == 0)
102 unmap_mapping_range(mapping, 0, 0, 0);
103 ret = filemap_write_and_wait(mapping);
106 ret = nfs_wb_all(mapping->host);
112 * Invalidate the local caches
114 static void nfs_zap_caches_locked(struct inode *inode)
116 struct nfs_inode *nfsi = NFS_I(inode);
117 int mode = inode->i_mode;
119 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
121 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
122 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
124 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
125 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
126 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
128 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
131 void nfs_zap_caches(struct inode *inode)
133 spin_lock(&inode->i_lock);
134 nfs_zap_caches_locked(inode);
135 spin_unlock(&inode->i_lock);
138 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
140 if (mapping->nrpages != 0) {
141 spin_lock(&inode->i_lock);
142 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
143 spin_unlock(&inode->i_lock);
147 static void nfs_zap_acl_cache(struct inode *inode)
149 void (*clear_acl_cache)(struct inode *);
151 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
152 if (clear_acl_cache != NULL)
153 clear_acl_cache(inode);
154 spin_lock(&inode->i_lock);
155 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
156 spin_unlock(&inode->i_lock);
159 void nfs_invalidate_atime(struct inode *inode)
161 spin_lock(&inode->i_lock);
162 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
163 spin_unlock(&inode->i_lock);
167 * Invalidate, but do not unhash, the inode.
168 * NB: must be called with inode->i_lock held!
170 static void nfs_invalidate_inode(struct inode *inode)
172 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
173 nfs_zap_caches_locked(inode);
176 struct nfs_find_desc {
178 struct nfs_fattr *fattr;
182 * In NFSv3 we can have 64bit inode numbers. In order to support
183 * this, and re-exported directories (also seen in NFSv2)
184 * we are forced to allow 2 different inodes to have the same
188 nfs_find_actor(struct inode *inode, void *opaque)
190 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
191 struct nfs_fh *fh = desc->fh;
192 struct nfs_fattr *fattr = desc->fattr;
194 if (NFS_FILEID(inode) != fattr->fileid)
196 if (nfs_compare_fh(NFS_FH(inode), fh))
198 if (is_bad_inode(inode) || NFS_STALE(inode))
204 nfs_init_locked(struct inode *inode, void *opaque)
206 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
207 struct nfs_fattr *fattr = desc->fattr;
209 NFS_FILEID(inode) = fattr->fileid;
210 nfs_copy_fh(NFS_FH(inode), desc->fh);
214 /* Don't use READDIRPLUS on directories that we believe are too large */
215 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
218 * This is our front-end to iget that looks up inodes by file handle
219 * instead of inode number.
222 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
224 struct nfs_find_desc desc = {
228 struct inode *inode = ERR_PTR(-ENOENT);
231 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
235 printk("NFS: Buggy server - nlink == 0!\n");
239 hash = nfs_fattr_to_ino_t(fattr);
241 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
243 inode = ERR_PTR(-ENOMEM);
247 if (inode->i_state & I_NEW) {
248 struct nfs_inode *nfsi = NFS_I(inode);
249 unsigned long now = jiffies;
251 /* We set i_ino for the few things that still rely on it,
255 /* We can't support update_atime(), since the server will reset it */
256 inode->i_flags |= S_NOATIME|S_NOCMTIME;
257 inode->i_mode = fattr->mode;
258 /* Why so? Because we want revalidate for devices/FIFOs, and
259 * that's precisely what we have in nfs_file_inode_operations.
261 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
262 if (S_ISREG(inode->i_mode)) {
263 inode->i_fop = &nfs_file_operations;
264 inode->i_data.a_ops = &nfs_file_aops;
265 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
266 } else if (S_ISDIR(inode->i_mode)) {
267 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
268 inode->i_fop = &nfs_dir_operations;
269 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
270 && fattr->size <= NFS_LIMIT_READDIRPLUS)
271 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
272 /* Deal with crossing mountpoints */
273 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
274 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
275 inode->i_op = &nfs_referral_inode_operations;
277 inode->i_op = &nfs_mountpoint_inode_operations;
280 } else if (S_ISLNK(inode->i_mode))
281 inode->i_op = &nfs_symlink_inode_operations;
283 init_special_inode(inode, inode->i_mode, fattr->rdev);
285 nfsi->read_cache_jiffies = fattr->time_start;
286 nfsi->last_updated = now;
287 nfsi->cache_change_attribute = now;
288 inode->i_atime = fattr->atime;
289 inode->i_mtime = fattr->mtime;
290 inode->i_ctime = fattr->ctime;
291 if (fattr->valid & NFS_ATTR_FATTR_V4)
292 nfsi->change_attr = fattr->change_attr;
293 inode->i_size = nfs_size_to_loff_t(fattr->size);
294 inode->i_nlink = fattr->nlink;
295 inode->i_uid = fattr->uid;
296 inode->i_gid = fattr->gid;
297 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
299 * report the blocks in 512byte units
301 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
303 inode->i_blocks = fattr->du.nfs2.blocks;
305 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
306 nfsi->attrtimeo_timestamp = now;
307 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
308 nfsi->access_cache = RB_ROOT;
310 unlock_new_inode(inode);
312 nfs_refresh_inode(inode, fattr);
313 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
315 (long long)NFS_FILEID(inode),
316 atomic_read(&inode->i_count));
322 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
326 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
329 nfs_setattr(struct dentry *dentry, struct iattr *attr)
331 struct inode *inode = dentry->d_inode;
332 struct nfs_fattr fattr;
335 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
337 if (attr->ia_valid & ATTR_SIZE) {
338 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
339 attr->ia_valid &= ~ATTR_SIZE;
342 /* Optimization: if the end result is no change, don't RPC */
343 attr->ia_valid &= NFS_VALID_ATTRS;
344 if (attr->ia_valid == 0)
348 nfs_begin_data_update(inode);
349 /* Write all dirty data */
350 if (S_ISREG(inode->i_mode)) {
351 filemap_write_and_wait(inode->i_mapping);
355 * Return any delegations if we're going to change ACLs
357 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
358 nfs_inode_return_delegation(inode);
359 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
361 nfs_refresh_inode(inode, &fattr);
362 nfs_end_data_update(inode);
368 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
369 * @inode: pointer to struct inode
370 * @attr: pointer to struct iattr
372 * Note: we do this in the *proc.c in order to ensure that
373 * it works for things like exclusive creates too.
375 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
377 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
378 if ((attr->ia_valid & ATTR_MODE) != 0) {
379 int mode = attr->ia_mode & S_IALLUGO;
380 mode |= inode->i_mode & ~S_IALLUGO;
381 inode->i_mode = mode;
383 if ((attr->ia_valid & ATTR_UID) != 0)
384 inode->i_uid = attr->ia_uid;
385 if ((attr->ia_valid & ATTR_GID) != 0)
386 inode->i_gid = attr->ia_gid;
387 spin_lock(&inode->i_lock);
388 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
389 spin_unlock(&inode->i_lock);
391 if ((attr->ia_valid & ATTR_SIZE) != 0) {
392 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
393 inode->i_size = attr->ia_size;
394 vmtruncate(inode, attr->ia_size);
398 static int nfs_wait_schedule(void *word)
400 if (signal_pending(current))
407 * Wait for the inode to get unlocked.
409 static int nfs_wait_on_inode(struct inode *inode)
411 struct rpc_clnt *clnt = NFS_CLIENT(inode);
412 struct nfs_inode *nfsi = NFS_I(inode);
416 rpc_clnt_sigmask(clnt, &oldmask);
417 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
418 nfs_wait_schedule, TASK_INTERRUPTIBLE);
419 rpc_clnt_sigunmask(clnt, &oldmask);
424 static void nfs_wake_up_inode(struct inode *inode)
426 struct nfs_inode *nfsi = NFS_I(inode);
428 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
429 smp_mb__after_clear_bit();
430 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
433 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
435 struct inode *inode = dentry->d_inode;
436 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
439 /* Flush out writes to the server in order to update c/mtime */
440 if (S_ISREG(inode->i_mode))
441 nfs_wb_nocommit(inode);
444 * We may force a getattr if the user cares about atime.
446 * Note that we only have to check the vfsmount flags here:
447 * - NFS always sets S_NOATIME by so checking it would give a
449 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
450 * no point in checking those.
452 if ((mnt->mnt_flags & MNT_NOATIME) ||
453 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
457 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
459 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
461 generic_fillattr(inode, stat);
462 stat->ino = NFS_FILEID(inode);
467 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
469 struct nfs_open_context *ctx;
471 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
473 ctx->path.dentry = dget(dentry);
474 ctx->path.mnt = mntget(mnt);
475 ctx->cred = get_rpccred(cred);
477 ctx->lockowner = current->files;
480 atomic_set(&ctx->count, 1);
485 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
488 atomic_inc(&ctx->count);
492 void put_nfs_open_context(struct nfs_open_context *ctx)
494 struct inode *inode = ctx->path.dentry->d_inode;
496 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
498 list_del(&ctx->list);
499 spin_unlock(&inode->i_lock);
500 if (ctx->state != NULL)
501 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
502 if (ctx->cred != NULL)
503 put_rpccred(ctx->cred);
504 dput(ctx->path.dentry);
505 mntput(ctx->path.mnt);
510 * Ensure that mmap has a recent RPC credential for use when writing out
513 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
515 struct inode *inode = filp->f_path.dentry->d_inode;
516 struct nfs_inode *nfsi = NFS_I(inode);
518 filp->private_data = get_nfs_open_context(ctx);
519 spin_lock(&inode->i_lock);
520 list_add(&ctx->list, &nfsi->open_files);
521 spin_unlock(&inode->i_lock);
525 * Given an inode, search for an open context with the desired characteristics
527 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
529 struct nfs_inode *nfsi = NFS_I(inode);
530 struct nfs_open_context *pos, *ctx = NULL;
532 spin_lock(&inode->i_lock);
533 list_for_each_entry(pos, &nfsi->open_files, list) {
534 if (cred != NULL && pos->cred != cred)
536 if ((pos->mode & mode) == mode) {
537 ctx = get_nfs_open_context(pos);
541 spin_unlock(&inode->i_lock);
545 static void nfs_file_clear_open_context(struct file *filp)
547 struct inode *inode = filp->f_path.dentry->d_inode;
548 struct nfs_open_context *ctx = nfs_file_open_context(filp);
551 filp->private_data = NULL;
552 spin_lock(&inode->i_lock);
553 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
554 spin_unlock(&inode->i_lock);
555 put_nfs_open_context(ctx);
560 * These allocate and release file read/write context information.
562 int nfs_open(struct inode *inode, struct file *filp)
564 struct nfs_open_context *ctx;
565 struct rpc_cred *cred;
567 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
569 return PTR_ERR(cred);
570 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
574 ctx->mode = filp->f_mode;
575 nfs_file_set_open_context(filp, ctx);
576 put_nfs_open_context(ctx);
580 int nfs_release(struct inode *inode, struct file *filp)
582 nfs_file_clear_open_context(filp);
587 * This function is called whenever some part of NFS notices that
588 * the cached attributes have to be refreshed.
591 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
593 int status = -ESTALE;
594 struct nfs_fattr fattr;
595 struct nfs_inode *nfsi = NFS_I(inode);
597 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
598 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
600 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
602 if (is_bad_inode(inode))
604 if (NFS_STALE(inode))
607 status = nfs_wait_on_inode(inode);
610 if (NFS_STALE(inode)) {
612 /* Do we trust the cached ESTALE? */
613 if (NFS_ATTRTIMEO(inode) != 0) {
614 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
621 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
623 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
625 (long long)NFS_FILEID(inode), status);
626 if (status == -ESTALE) {
627 nfs_zap_caches(inode);
628 if (!S_ISDIR(inode->i_mode))
629 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
634 spin_lock(&inode->i_lock);
635 status = nfs_update_inode(inode, &fattr);
637 spin_unlock(&inode->i_lock);
638 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
640 (long long)NFS_FILEID(inode), status);
643 spin_unlock(&inode->i_lock);
645 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
646 nfs_zap_acl_cache(inode);
648 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
650 (long long)NFS_FILEID(inode));
653 nfs_wake_up_inode(inode);
660 int nfs_attribute_timeout(struct inode *inode)
662 struct nfs_inode *nfsi = NFS_I(inode);
664 if (nfs_have_delegation(inode, FMODE_READ))
666 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
670 * nfs_revalidate_inode - Revalidate the inode attributes
671 * @server - pointer to nfs_server struct
672 * @inode - pointer to inode struct
674 * Updates inode attribute information by retrieving the data from the server.
676 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
678 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
679 && !nfs_attribute_timeout(inode))
680 return NFS_STALE(inode) ? -ESTALE : 0;
681 return __nfs_revalidate_inode(server, inode);
684 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
686 struct nfs_inode *nfsi = NFS_I(inode);
688 if (mapping->nrpages != 0) {
689 int ret = invalidate_inode_pages2(mapping);
693 spin_lock(&inode->i_lock);
694 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
695 if (S_ISDIR(inode->i_mode)) {
696 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
697 /* This ensures we revalidate child dentries */
698 nfsi->cache_change_attribute = jiffies;
700 spin_unlock(&inode->i_lock);
701 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
702 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
703 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
707 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
711 mutex_lock(&inode->i_mutex);
712 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
713 ret = nfs_sync_mapping(mapping);
715 ret = nfs_invalidate_mapping_nolock(inode, mapping);
717 mutex_unlock(&inode->i_mutex);
722 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
723 * @inode - pointer to host inode
724 * @mapping - pointer to mapping
726 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
728 struct nfs_inode *nfsi = NFS_I(inode);
731 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
732 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
733 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
737 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
738 ret = nfs_invalidate_mapping_nolock(inode, mapping);
744 * nfs_revalidate_mapping - Revalidate the pagecache
745 * @inode - pointer to host inode
746 * @mapping - pointer to mapping
748 * This version of the function will take the inode->i_mutex and attempt to
749 * flush out all dirty data if it needs to invalidate the page cache.
751 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
753 struct nfs_inode *nfsi = NFS_I(inode);
756 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
757 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
758 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
762 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
763 ret = nfs_invalidate_mapping(inode, mapping);
769 * nfs_begin_data_update
770 * @inode - pointer to inode
771 * Declare that a set of operations will update file data on the server
773 void nfs_begin_data_update(struct inode *inode)
775 atomic_inc(&NFS_I(inode)->data_updates);
779 * nfs_end_data_update
780 * @inode - pointer to inode
781 * Declare end of the operations that will update file data
782 * This will mark the inode as immediately needing revalidation
783 * of its attribute cache.
785 void nfs_end_data_update(struct inode *inode)
787 struct nfs_inode *nfsi = NFS_I(inode);
789 /* Directories: invalidate page cache */
790 if (S_ISDIR(inode->i_mode)) {
791 spin_lock(&inode->i_lock);
792 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
793 spin_unlock(&inode->i_lock);
795 atomic_dec(&nfsi->data_updates);
798 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
800 struct nfs_inode *nfsi = NFS_I(inode);
802 /* If we have atomic WCC data, we may update some attributes */
803 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
804 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
805 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
806 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
807 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
808 if (S_ISDIR(inode->i_mode))
809 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
811 if (inode->i_size == fattr->pre_size && nfsi->npages == 0)
812 inode->i_size = fattr->size;
817 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
818 * @inode - pointer to inode
819 * @fattr - updated attributes
821 * Verifies the attribute cache. If we have just changed the attributes,
822 * so that fattr carries weak cache consistency data, then it may
823 * also update the ctime/mtime/change_attribute.
825 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
827 struct nfs_inode *nfsi = NFS_I(inode);
828 loff_t cur_size, new_isize;
832 /* Has the inode gone and changed behind our back? */
833 if (nfsi->fileid != fattr->fileid
834 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
838 /* Are we in the process of updating data on the server? */
839 data_unstable = nfs_caches_unstable(inode);
841 /* Do atomic weak cache consistency updates */
842 nfs_wcc_update_inode(inode, fattr);
844 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
845 nfsi->change_attr != fattr->change_attr)
846 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
848 /* Verify a few of the more important attributes */
849 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
850 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
852 cur_size = i_size_read(inode);
853 new_isize = nfs_size_to_loff_t(fattr->size);
854 if (cur_size != new_isize && nfsi->npages == 0)
855 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
857 /* Have any file permissions changed? */
858 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
859 || inode->i_uid != fattr->uid
860 || inode->i_gid != fattr->gid)
861 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
863 /* Has the link count changed? */
864 if (inode->i_nlink != fattr->nlink)
865 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
867 if (!timespec_equal(&inode->i_atime, &fattr->atime))
868 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
870 nfsi->read_cache_jiffies = fattr->time_start;
875 * nfs_refresh_inode - try to update the inode attribute cache
876 * @inode - pointer to inode
877 * @fattr - updated attributes
879 * Check that an RPC call that returned attributes has not overlapped with
880 * other recent updates of the inode metadata, then decide whether it is
881 * safe to do a full update of the inode attributes, or whether just to
882 * call nfs_check_inode_attributes.
884 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
886 struct nfs_inode *nfsi = NFS_I(inode);
889 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
891 spin_lock(&inode->i_lock);
892 if (time_after(fattr->time_start, nfsi->last_updated))
893 status = nfs_update_inode(inode, fattr);
895 status = nfs_check_inode_attributes(inode, fattr);
897 spin_unlock(&inode->i_lock);
902 * nfs_post_op_update_inode - try to update the inode attribute cache
903 * @inode - pointer to inode
904 * @fattr - updated attributes
906 * After an operation that has changed the inode metadata, mark the
907 * attribute cache as being invalid, then try to update it.
909 * NB: if the server didn't return any post op attributes, this
910 * function will force the retrieval of attributes before the next
911 * NFS request. Thus it should be used only for operations that
912 * are expected to change one or more attributes, to avoid
913 * unnecessary NFS requests and trips through nfs_update_inode().
915 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
917 struct nfs_inode *nfsi = NFS_I(inode);
920 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
921 spin_lock(&inode->i_lock);
922 nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
923 nfsi->cache_change_attribute = jiffies;
924 spin_unlock(&inode->i_lock);
927 status = nfs_refresh_inode(inode, fattr);
933 * Many nfs protocol calls return the new file attributes after
934 * an operation. Here we update the inode to reflect the state
935 * of the server's inode.
937 * This is a bit tricky because we have to make sure all dirty pages
938 * have been sent off to the server before calling invalidate_inode_pages.
939 * To make sure no other process adds more write requests while we try
940 * our best to flush them, we make them sleep during the attribute refresh.
942 * A very similar scenario holds for the dir cache.
944 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
946 struct nfs_server *server;
947 struct nfs_inode *nfsi = NFS_I(inode);
948 loff_t cur_isize, new_isize;
949 unsigned int invalid = 0;
950 unsigned long now = jiffies;
953 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
954 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
955 atomic_read(&inode->i_count), fattr->valid);
957 if (nfsi->fileid != fattr->fileid)
961 * Make sure the inode's type hasn't changed.
963 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
966 server = NFS_SERVER(inode);
967 /* Update the fsid? */
968 if (S_ISDIR(inode->i_mode)
969 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
970 server->fsid = fattr->fsid;
973 * Update the read time so we don't revalidate too often.
975 nfsi->read_cache_jiffies = fattr->time_start;
976 nfsi->last_updated = now;
978 /* Are we racing with known updates of the metadata on the server? */
979 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
980 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
981 | NFS_INO_REVAL_PAGECACHE);
983 /* Do atomic weak cache consistency updates */
984 nfs_wcc_update_inode(inode, fattr);
986 /* More cache consistency checks */
987 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
988 /* NFSv2/v3: Check if the mtime agrees */
989 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
990 dprintk("NFS: mtime change on server for file %s/%ld\n",
991 inode->i_sb->s_id, inode->i_ino);
992 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
993 nfsi->cache_change_attribute = now;
995 /* If ctime has changed we should definitely clear access+acl caches */
996 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
997 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
998 nfsi->cache_change_attribute = now;
1000 } else if (nfsi->change_attr != fattr->change_attr) {
1001 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1002 inode->i_sb->s_id, inode->i_ino);
1003 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1004 nfsi->cache_change_attribute = now;
1007 /* Check if our cached file size is stale */
1008 new_isize = nfs_size_to_loff_t(fattr->size);
1009 cur_isize = i_size_read(inode);
1010 if (new_isize != cur_isize) {
1011 /* Do we perhaps have any outstanding writes? */
1012 if (nfsi->npages == 0) {
1013 /* No, but did we race with nfs_end_data_update()? */
1015 inode->i_size = new_isize;
1016 invalid |= NFS_INO_INVALID_DATA;
1018 invalid |= NFS_INO_INVALID_ATTR;
1019 } else if (new_isize > cur_isize) {
1020 inode->i_size = new_isize;
1021 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1023 nfsi->cache_change_attribute = now;
1024 dprintk("NFS: isize change on server for file %s/%ld\n",
1025 inode->i_sb->s_id, inode->i_ino);
1029 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1030 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1031 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1032 nfsi->change_attr = fattr->change_attr;
1034 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1035 inode->i_uid != fattr->uid ||
1036 inode->i_gid != fattr->gid)
1037 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1039 inode->i_mode = fattr->mode;
1040 inode->i_nlink = fattr->nlink;
1041 inode->i_uid = fattr->uid;
1042 inode->i_gid = fattr->gid;
1044 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1046 * report the blocks in 512byte units
1048 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1050 inode->i_blocks = fattr->du.nfs2.blocks;
1053 /* Update attrtimeo value if we're out of the unstable period */
1054 if (invalid & NFS_INO_INVALID_ATTR) {
1055 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1056 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1057 nfsi->attrtimeo_timestamp = now;
1058 } else if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1059 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1060 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1061 nfsi->attrtimeo_timestamp = now;
1063 invalid &= ~NFS_INO_INVALID_ATTR;
1064 /* Don't invalidate the data if we were to blame */
1065 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1066 || S_ISLNK(inode->i_mode)))
1067 invalid &= ~NFS_INO_INVALID_DATA;
1068 if (!nfs_have_delegation(inode, FMODE_READ) ||
1069 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1070 nfsi->cache_validity |= invalid;
1071 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1076 * Big trouble! The inode has become a different object.
1078 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1079 __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 /* If we are holding a delegation, return it! */
1108 nfs_inode_return_delegation(inode);
1109 /* First call standard NFS clear_inode() code */
1110 nfs_clear_inode(inode);
1114 struct inode *nfs_alloc_inode(struct super_block *sb)
1116 struct nfs_inode *nfsi;
1117 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1121 nfsi->cache_validity = 0UL;
1122 #ifdef CONFIG_NFS_V3_ACL
1123 nfsi->acl_access = ERR_PTR(-EAGAIN);
1124 nfsi->acl_default = ERR_PTR(-EAGAIN);
1126 #ifdef CONFIG_NFS_V4
1127 nfsi->nfs4_acl = NULL;
1128 #endif /* CONFIG_NFS_V4 */
1129 return &nfsi->vfs_inode;
1132 void nfs_destroy_inode(struct inode *inode)
1134 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1137 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1139 #ifdef CONFIG_NFS_V4
1140 INIT_LIST_HEAD(&nfsi->open_states);
1141 nfsi->delegation = NULL;
1142 nfsi->delegation_state = 0;
1143 init_rwsem(&nfsi->rwsem);
1147 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1149 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1151 inode_init_once(&nfsi->vfs_inode);
1152 INIT_LIST_HEAD(&nfsi->open_files);
1153 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1154 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1155 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1156 atomic_set(&nfsi->data_updates, 0);
1159 nfs4_init_once(nfsi);
1162 static int __init nfs_init_inodecache(void)
1164 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1165 sizeof(struct nfs_inode),
1166 0, (SLAB_RECLAIM_ACCOUNT|
1169 if (nfs_inode_cachep == NULL)
1175 static void nfs_destroy_inodecache(void)
1177 kmem_cache_destroy(nfs_inode_cachep);
1183 static int __init init_nfs_fs(void)
1187 err = nfs_fs_proc_init();
1191 err = nfs_init_nfspagecache();
1195 err = nfs_init_inodecache();
1199 err = nfs_init_readpagecache();
1203 err = nfs_init_writepagecache();
1207 err = nfs_init_directcache();
1211 #ifdef CONFIG_PROC_FS
1212 rpc_proc_register(&nfs_rpcstat);
1214 if ((err = register_nfs_fs()) != 0)
1218 #ifdef CONFIG_PROC_FS
1219 rpc_proc_unregister("nfs");
1221 nfs_destroy_directcache();
1223 nfs_destroy_writepagecache();
1225 nfs_destroy_readpagecache();
1227 nfs_destroy_inodecache();
1229 nfs_destroy_nfspagecache();
1236 static void __exit exit_nfs_fs(void)
1238 nfs_destroy_directcache();
1239 nfs_destroy_writepagecache();
1240 nfs_destroy_readpagecache();
1241 nfs_destroy_inodecache();
1242 nfs_destroy_nfspagecache();
1243 #ifdef CONFIG_PROC_FS
1244 rpc_proc_unregister("nfs");
1246 unregister_nfs_fs();
1250 /* Not quite true; I just maintain it */
1251 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1252 MODULE_LICENSE("GPL");
1254 module_init(init_nfs_fs)
1255 module_exit(exit_nfs_fs)