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 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
54 /* Default is to see 64-bit inode numbers */
55 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
57 static void nfs_invalidate_inode(struct inode *);
58 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
60 static struct kmem_cache * nfs_inode_cachep;
62 static inline unsigned long
63 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
65 return nfs_fileid_to_ino_t(fattr->fileid);
69 * nfs_compat_user_ino64 - returns the user-visible inode number
70 * @fileid: 64-bit fileid
72 * This function returns a 32-bit inode number if the boot parameter
73 * nfs.enable_ino64 is zero.
75 u64 nfs_compat_user_ino64(u64 fileid)
82 if (sizeof(ino) < sizeof(fileid))
83 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
87 int nfs_write_inode(struct inode *inode, int sync)
92 ret = filemap_fdatawait(inode->i_mapping);
94 ret = nfs_commit_inode(inode, FLUSH_SYNC);
96 ret = nfs_commit_inode(inode, 0);
99 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
103 void nfs_clear_inode(struct inode *inode)
106 * The following should never happen...
108 BUG_ON(nfs_have_writebacks(inode));
109 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
110 nfs_zap_acl_cache(inode);
111 nfs_access_zap_cache(inode);
115 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
117 int nfs_sync_mapping(struct address_space *mapping)
121 if (mapping->nrpages == 0)
123 unmap_mapping_range(mapping, 0, 0, 0);
124 ret = filemap_write_and_wait(mapping);
127 ret = nfs_wb_all(mapping->host);
133 * Invalidate the local caches
135 static void nfs_zap_caches_locked(struct inode *inode)
137 struct nfs_inode *nfsi = NFS_I(inode);
138 int mode = inode->i_mode;
140 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
142 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
143 nfsi->attrtimeo_timestamp = jiffies;
145 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
146 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
147 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
149 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
152 void nfs_zap_caches(struct inode *inode)
154 spin_lock(&inode->i_lock);
155 nfs_zap_caches_locked(inode);
156 spin_unlock(&inode->i_lock);
159 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
161 if (mapping->nrpages != 0) {
162 spin_lock(&inode->i_lock);
163 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
164 spin_unlock(&inode->i_lock);
168 void nfs_zap_acl_cache(struct inode *inode)
170 void (*clear_acl_cache)(struct inode *);
172 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
173 if (clear_acl_cache != NULL)
174 clear_acl_cache(inode);
175 spin_lock(&inode->i_lock);
176 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
177 spin_unlock(&inode->i_lock);
180 void nfs_invalidate_atime(struct inode *inode)
182 spin_lock(&inode->i_lock);
183 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
184 spin_unlock(&inode->i_lock);
188 * Invalidate, but do not unhash, the inode.
189 * NB: must be called with inode->i_lock held!
191 static void nfs_invalidate_inode(struct inode *inode)
193 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
194 nfs_zap_caches_locked(inode);
197 struct nfs_find_desc {
199 struct nfs_fattr *fattr;
203 * In NFSv3 we can have 64bit inode numbers. In order to support
204 * this, and re-exported directories (also seen in NFSv2)
205 * we are forced to allow 2 different inodes to have the same
209 nfs_find_actor(struct inode *inode, void *opaque)
211 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
212 struct nfs_fh *fh = desc->fh;
213 struct nfs_fattr *fattr = desc->fattr;
215 if (NFS_FILEID(inode) != fattr->fileid)
217 if (nfs_compare_fh(NFS_FH(inode), fh))
219 if (is_bad_inode(inode) || NFS_STALE(inode))
225 nfs_init_locked(struct inode *inode, void *opaque)
227 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
228 struct nfs_fattr *fattr = desc->fattr;
230 set_nfs_fileid(inode, fattr->fileid);
231 nfs_copy_fh(NFS_FH(inode), desc->fh);
235 /* Don't use READDIRPLUS on directories that we believe are too large */
236 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
239 * This is our front-end to iget that looks up inodes by file handle
240 * instead of inode number.
243 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
245 struct nfs_find_desc desc = {
249 struct inode *inode = ERR_PTR(-ENOENT);
252 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
256 printk("NFS: Buggy server - nlink == 0!\n");
260 hash = nfs_fattr_to_ino_t(fattr);
262 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
264 inode = ERR_PTR(-ENOMEM);
268 if (inode->i_state & I_NEW) {
269 struct nfs_inode *nfsi = NFS_I(inode);
270 unsigned long now = jiffies;
272 /* We set i_ino for the few things that still rely on it,
276 /* We can't support update_atime(), since the server will reset it */
277 inode->i_flags |= S_NOATIME|S_NOCMTIME;
278 inode->i_mode = fattr->mode;
279 /* Why so? Because we want revalidate for devices/FIFOs, and
280 * that's precisely what we have in nfs_file_inode_operations.
282 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
283 if (S_ISREG(inode->i_mode)) {
284 inode->i_fop = &nfs_file_operations;
285 inode->i_data.a_ops = &nfs_file_aops;
286 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
287 } else if (S_ISDIR(inode->i_mode)) {
288 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
289 inode->i_fop = &nfs_dir_operations;
290 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
291 && fattr->size <= NFS_LIMIT_READDIRPLUS)
292 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
293 /* Deal with crossing mountpoints */
294 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
295 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
296 inode->i_op = &nfs_referral_inode_operations;
298 inode->i_op = &nfs_mountpoint_inode_operations;
300 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
302 } else if (S_ISLNK(inode->i_mode))
303 inode->i_op = &nfs_symlink_inode_operations;
305 init_special_inode(inode, inode->i_mode, fattr->rdev);
307 nfsi->read_cache_jiffies = fattr->time_start;
308 nfsi->last_updated = now;
309 nfsi->cache_change_attribute = now;
310 inode->i_atime = fattr->atime;
311 inode->i_mtime = fattr->mtime;
312 inode->i_ctime = fattr->ctime;
313 if (fattr->valid & NFS_ATTR_FATTR_V4)
314 nfsi->change_attr = fattr->change_attr;
315 inode->i_size = nfs_size_to_loff_t(fattr->size);
316 inode->i_nlink = fattr->nlink;
317 inode->i_uid = fattr->uid;
318 inode->i_gid = fattr->gid;
319 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
321 * report the blocks in 512byte units
323 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
325 inode->i_blocks = fattr->du.nfs2.blocks;
327 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
328 nfsi->attrtimeo_timestamp = now;
329 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
330 nfsi->access_cache = RB_ROOT;
332 unlock_new_inode(inode);
334 nfs_refresh_inode(inode, fattr);
335 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
337 (long long)NFS_FILEID(inode),
338 atomic_read(&inode->i_count));
344 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
348 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
351 nfs_setattr(struct dentry *dentry, struct iattr *attr)
353 struct inode *inode = dentry->d_inode;
354 struct nfs_fattr fattr;
357 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
359 /* skip mode change if it's just for clearing setuid/setgid */
360 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
361 attr->ia_valid &= ~ATTR_MODE;
363 if (attr->ia_valid & ATTR_SIZE) {
364 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
365 attr->ia_valid &= ~ATTR_SIZE;
368 /* Optimization: if the end result is no change, don't RPC */
369 attr->ia_valid &= NFS_VALID_ATTRS;
370 if ((attr->ia_valid & ~ATTR_FILE) == 0)
374 /* Write all dirty data */
375 if (S_ISREG(inode->i_mode)) {
376 filemap_write_and_wait(inode->i_mapping);
380 * Return any delegations if we're going to change ACLs
382 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
383 nfs_inode_return_delegation(inode);
384 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
386 nfs_refresh_inode(inode, &fattr);
392 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
393 * @inode: pointer to struct inode
394 * @attr: pointer to struct iattr
396 * Note: we do this in the *proc.c in order to ensure that
397 * it works for things like exclusive creates too.
399 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
401 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
402 if ((attr->ia_valid & ATTR_MODE) != 0) {
403 int mode = attr->ia_mode & S_IALLUGO;
404 mode |= inode->i_mode & ~S_IALLUGO;
405 inode->i_mode = mode;
407 if ((attr->ia_valid & ATTR_UID) != 0)
408 inode->i_uid = attr->ia_uid;
409 if ((attr->ia_valid & ATTR_GID) != 0)
410 inode->i_gid = attr->ia_gid;
411 spin_lock(&inode->i_lock);
412 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
413 spin_unlock(&inode->i_lock);
415 if ((attr->ia_valid & ATTR_SIZE) != 0) {
416 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
417 inode->i_size = attr->ia_size;
418 vmtruncate(inode, attr->ia_size);
422 static int nfs_wait_schedule(void *word)
424 if (signal_pending(current))
431 * Wait for the inode to get unlocked.
433 static int nfs_wait_on_inode(struct inode *inode)
435 struct nfs_inode *nfsi = NFS_I(inode);
438 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
439 nfs_wait_schedule, TASK_KILLABLE);
444 static void nfs_wake_up_inode(struct inode *inode)
446 struct nfs_inode *nfsi = NFS_I(inode);
448 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
449 smp_mb__after_clear_bit();
450 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
453 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
455 struct inode *inode = dentry->d_inode;
456 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
460 * Flush out writes to the server in order to update c/mtime.
462 * Hold the i_mutex to suspend application writes temporarily;
463 * this prevents long-running writing applications from blocking
466 if (S_ISREG(inode->i_mode)) {
467 mutex_lock(&inode->i_mutex);
468 nfs_wb_nocommit(inode);
469 mutex_unlock(&inode->i_mutex);
473 * We may force a getattr if the user cares about atime.
475 * Note that we only have to check the vfsmount flags here:
476 * - NFS always sets S_NOATIME by so checking it would give a
478 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
479 * no point in checking those.
481 if ((mnt->mnt_flags & MNT_NOATIME) ||
482 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
486 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
488 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
490 generic_fillattr(inode, stat);
491 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
496 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
498 struct nfs_open_context *ctx;
500 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
502 ctx->path.dentry = dget(dentry);
503 ctx->path.mnt = mntget(mnt);
504 ctx->cred = get_rpccred(cred);
506 ctx->lockowner = current->files;
510 atomic_set(&ctx->count, 1);
515 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
518 atomic_inc(&ctx->count);
522 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
529 inode = ctx->path.dentry->d_inode;
530 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
532 list_del(&ctx->list);
533 spin_unlock(&inode->i_lock);
534 if (ctx->state != NULL) {
536 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
538 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
540 if (ctx->cred != NULL)
541 put_rpccred(ctx->cred);
542 path_put(&ctx->path);
546 void put_nfs_open_context(struct nfs_open_context *ctx)
548 __put_nfs_open_context(ctx, 0);
551 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
553 __put_nfs_open_context(ctx, 1);
557 * Ensure that mmap has a recent RPC credential for use when writing out
560 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
562 struct inode *inode = filp->f_path.dentry->d_inode;
563 struct nfs_inode *nfsi = NFS_I(inode);
565 filp->private_data = get_nfs_open_context(ctx);
566 spin_lock(&inode->i_lock);
567 list_add(&ctx->list, &nfsi->open_files);
568 spin_unlock(&inode->i_lock);
572 * Given an inode, search for an open context with the desired characteristics
574 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
576 struct nfs_inode *nfsi = NFS_I(inode);
577 struct nfs_open_context *pos, *ctx = NULL;
579 spin_lock(&inode->i_lock);
580 list_for_each_entry(pos, &nfsi->open_files, list) {
581 if (cred != NULL && pos->cred != cred)
583 if ((pos->mode & mode) == mode) {
584 ctx = get_nfs_open_context(pos);
588 spin_unlock(&inode->i_lock);
592 static void nfs_file_clear_open_context(struct file *filp)
594 struct inode *inode = filp->f_path.dentry->d_inode;
595 struct nfs_open_context *ctx = nfs_file_open_context(filp);
598 filp->private_data = NULL;
599 spin_lock(&inode->i_lock);
600 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
601 spin_unlock(&inode->i_lock);
602 put_nfs_open_context_sync(ctx);
607 * These allocate and release file read/write context information.
609 int nfs_open(struct inode *inode, struct file *filp)
611 struct nfs_open_context *ctx;
612 struct rpc_cred *cred;
614 cred = rpc_lookup_cred();
616 return PTR_ERR(cred);
617 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
621 ctx->mode = filp->f_mode;
622 nfs_file_set_open_context(filp, ctx);
623 put_nfs_open_context(ctx);
627 int nfs_release(struct inode *inode, struct file *filp)
629 nfs_file_clear_open_context(filp);
634 * This function is called whenever some part of NFS notices that
635 * the cached attributes have to be refreshed.
638 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
640 int status = -ESTALE;
641 struct nfs_fattr fattr;
642 struct nfs_inode *nfsi = NFS_I(inode);
644 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
645 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
647 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
649 if (is_bad_inode(inode))
651 if (NFS_STALE(inode))
654 status = nfs_wait_on_inode(inode);
659 if (NFS_STALE(inode))
662 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
664 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
666 (long long)NFS_FILEID(inode), status);
667 if (status == -ESTALE) {
668 nfs_zap_caches(inode);
669 if (!S_ISDIR(inode->i_mode))
670 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
675 spin_lock(&inode->i_lock);
676 status = nfs_update_inode(inode, &fattr);
678 spin_unlock(&inode->i_lock);
679 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
681 (long long)NFS_FILEID(inode), status);
684 spin_unlock(&inode->i_lock);
686 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
687 nfs_zap_acl_cache(inode);
689 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
691 (long long)NFS_FILEID(inode));
694 nfs_wake_up_inode(inode);
701 int nfs_attribute_timeout(struct inode *inode)
703 struct nfs_inode *nfsi = NFS_I(inode);
705 if (nfs_have_delegation(inode, FMODE_READ))
708 * Special case: if the attribute timeout is set to 0, then always
709 * treat the cache as having expired (unless holding
712 if (nfsi->attrtimeo == 0)
714 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
718 * nfs_revalidate_inode - Revalidate the inode attributes
719 * @server - pointer to nfs_server struct
720 * @inode - pointer to inode struct
722 * Updates inode attribute information by retrieving the data from the server.
724 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
726 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
727 && !nfs_attribute_timeout(inode))
728 return NFS_STALE(inode) ? -ESTALE : 0;
729 return __nfs_revalidate_inode(server, inode);
732 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
734 struct nfs_inode *nfsi = NFS_I(inode);
736 if (mapping->nrpages != 0) {
737 int ret = invalidate_inode_pages2(mapping);
741 spin_lock(&inode->i_lock);
742 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
743 if (S_ISDIR(inode->i_mode))
744 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
745 spin_unlock(&inode->i_lock);
746 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
747 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
748 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
752 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
756 mutex_lock(&inode->i_mutex);
757 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
758 ret = nfs_sync_mapping(mapping);
760 ret = nfs_invalidate_mapping_nolock(inode, mapping);
762 mutex_unlock(&inode->i_mutex);
767 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
768 * @inode - pointer to host inode
769 * @mapping - pointer to mapping
771 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
773 struct nfs_inode *nfsi = NFS_I(inode);
776 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
777 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
778 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
782 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
783 ret = nfs_invalidate_mapping_nolock(inode, mapping);
789 * nfs_revalidate_mapping - Revalidate the pagecache
790 * @inode - pointer to host inode
791 * @mapping - pointer to mapping
793 * This version of the function will take the inode->i_mutex and attempt to
794 * flush out all dirty data if it needs to invalidate the page cache.
796 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
798 struct nfs_inode *nfsi = NFS_I(inode);
801 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
802 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
803 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
807 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
808 ret = nfs_invalidate_mapping(inode, mapping);
813 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
815 struct nfs_inode *nfsi = NFS_I(inode);
817 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
818 nfsi->change_attr == fattr->pre_change_attr) {
819 nfsi->change_attr = fattr->change_attr;
820 if (S_ISDIR(inode->i_mode))
821 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
823 /* If we have atomic WCC data, we may update some attributes */
824 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
825 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
826 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
827 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
828 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
829 if (S_ISDIR(inode->i_mode))
830 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
832 if (inode->i_size == nfs_size_to_loff_t(fattr->pre_size) &&
834 inode->i_size = nfs_size_to_loff_t(fattr->size);
839 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
840 * @inode - pointer to inode
841 * @fattr - updated attributes
843 * Verifies the attribute cache. If we have just changed the attributes,
844 * so that fattr carries weak cache consistency data, then it may
845 * also update the ctime/mtime/change_attribute.
847 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
849 struct nfs_inode *nfsi = NFS_I(inode);
850 loff_t cur_size, new_isize;
851 unsigned long invalid = 0;
854 /* Has the inode gone and changed behind our back? */
855 if (nfsi->fileid != fattr->fileid
856 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
860 /* Do atomic weak cache consistency updates */
861 nfs_wcc_update_inode(inode, fattr);
863 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
864 nfsi->change_attr != fattr->change_attr)
865 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
867 /* Verify a few of the more important attributes */
868 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
869 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
871 cur_size = i_size_read(inode);
872 new_isize = nfs_size_to_loff_t(fattr->size);
873 if (cur_size != new_isize && nfsi->npages == 0)
874 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
876 /* Have any file permissions changed? */
877 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
878 || inode->i_uid != fattr->uid
879 || inode->i_gid != fattr->gid)
880 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
882 /* Has the link count changed? */
883 if (inode->i_nlink != fattr->nlink)
884 invalid |= NFS_INO_INVALID_ATTR;
886 if (!timespec_equal(&inode->i_atime, &fattr->atime))
887 invalid |= NFS_INO_INVALID_ATIME;
890 nfsi->cache_validity |= invalid;
892 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
893 | NFS_INO_INVALID_ATIME
894 | NFS_INO_REVAL_PAGECACHE);
896 nfsi->read_cache_jiffies = fattr->time_start;
901 * nfs_refresh_inode - try to update the inode attribute cache
902 * @inode - pointer to inode
903 * @fattr - updated attributes
905 * Check that an RPC call that returned attributes has not overlapped with
906 * other recent updates of the inode metadata, then decide whether it is
907 * safe to do a full update of the inode attributes, or whether just to
908 * call nfs_check_inode_attributes.
910 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
912 struct nfs_inode *nfsi = NFS_I(inode);
915 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
917 spin_lock(&inode->i_lock);
918 if (time_after(fattr->time_start, nfsi->last_updated))
919 status = nfs_update_inode(inode, fattr);
921 status = nfs_check_inode_attributes(inode, fattr);
923 spin_unlock(&inode->i_lock);
928 * nfs_post_op_update_inode - try to update the inode attribute cache
929 * @inode - pointer to inode
930 * @fattr - updated attributes
932 * After an operation that has changed the inode metadata, mark the
933 * attribute cache as being invalid, then try to update it.
935 * NB: if the server didn't return any post op attributes, this
936 * function will force the retrieval of attributes before the next
937 * NFS request. Thus it should be used only for operations that
938 * are expected to change one or more attributes, to avoid
939 * unnecessary NFS requests and trips through nfs_update_inode().
941 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
943 struct nfs_inode *nfsi = NFS_I(inode);
945 spin_lock(&inode->i_lock);
946 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
947 if (S_ISDIR(inode->i_mode))
948 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
949 spin_unlock(&inode->i_lock);
950 return nfs_refresh_inode(inode, fattr);
954 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
955 * @inode - pointer to inode
956 * @fattr - updated attributes
958 * After an operation that has changed the inode metadata, mark the
959 * attribute cache as being invalid, then try to update it. Fake up
960 * weak cache consistency data, if none exist.
962 * This function is mainly designed to be used by the ->write_done() functions.
964 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
966 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
967 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
968 fattr->pre_change_attr = NFS_I(inode)->change_attr;
969 fattr->valid |= NFS_ATTR_WCC_V4;
971 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
972 (fattr->valid & NFS_ATTR_WCC) == 0) {
973 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
974 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
975 fattr->pre_size = inode->i_size;
976 fattr->valid |= NFS_ATTR_WCC;
978 return nfs_post_op_update_inode(inode, fattr);
982 * Many nfs protocol calls return the new file attributes after
983 * an operation. Here we update the inode to reflect the state
984 * of the server's inode.
986 * This is a bit tricky because we have to make sure all dirty pages
987 * have been sent off to the server before calling invalidate_inode_pages.
988 * To make sure no other process adds more write requests while we try
989 * our best to flush them, we make them sleep during the attribute refresh.
991 * A very similar scenario holds for the dir cache.
993 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
995 struct nfs_server *server;
996 struct nfs_inode *nfsi = NFS_I(inode);
997 loff_t cur_isize, new_isize;
998 unsigned long invalid = 0;
999 unsigned long now = jiffies;
1001 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1002 __func__, inode->i_sb->s_id, inode->i_ino,
1003 atomic_read(&inode->i_count), fattr->valid);
1005 if (nfsi->fileid != fattr->fileid)
1009 * Make sure the inode's type hasn't changed.
1011 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1014 server = NFS_SERVER(inode);
1015 /* Update the fsid? */
1016 if (S_ISDIR(inode->i_mode) &&
1017 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1018 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1019 server->fsid = fattr->fsid;
1022 * Update the read time so we don't revalidate too often.
1024 nfsi->read_cache_jiffies = fattr->time_start;
1026 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1027 | NFS_INO_REVAL_PAGECACHE);
1029 /* Do atomic weak cache consistency updates */
1030 nfs_wcc_update_inode(inode, fattr);
1032 /* More cache consistency checks */
1033 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1034 /* NFSv2/v3: Check if the mtime agrees */
1035 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1036 dprintk("NFS: mtime change on server for file %s/%ld\n",
1037 inode->i_sb->s_id, inode->i_ino);
1038 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1039 if (S_ISDIR(inode->i_mode))
1040 nfs_force_lookup_revalidate(inode);
1042 /* If ctime has changed we should definitely clear access+acl caches */
1043 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1044 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1045 } else if (nfsi->change_attr != fattr->change_attr) {
1046 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1047 inode->i_sb->s_id, inode->i_ino);
1048 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1049 if (S_ISDIR(inode->i_mode))
1050 nfs_force_lookup_revalidate(inode);
1053 /* Check if our cached file size is stale */
1054 new_isize = nfs_size_to_loff_t(fattr->size);
1055 cur_isize = i_size_read(inode);
1056 if (new_isize != cur_isize) {
1057 /* Do we perhaps have any outstanding writes, or has
1058 * the file grown beyond our last write? */
1059 if (nfsi->npages == 0 || new_isize > cur_isize) {
1060 inode->i_size = new_isize;
1061 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1063 dprintk("NFS: isize change on server for file %s/%ld\n",
1064 inode->i_sb->s_id, inode->i_ino);
1068 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1069 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1070 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1071 nfsi->change_attr = fattr->change_attr;
1073 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1074 inode->i_uid != fattr->uid ||
1075 inode->i_gid != fattr->gid)
1076 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1078 inode->i_mode = fattr->mode;
1079 inode->i_nlink = fattr->nlink;
1080 inode->i_uid = fattr->uid;
1081 inode->i_gid = fattr->gid;
1083 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1085 * report the blocks in 512byte units
1087 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1089 inode->i_blocks = fattr->du.nfs2.blocks;
1092 /* Update attrtimeo value if we're out of the unstable period */
1093 if (invalid & NFS_INO_INVALID_ATTR) {
1094 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1095 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1096 nfsi->attrtimeo_timestamp = now;
1097 nfsi->last_updated = now;
1099 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1100 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1101 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1102 nfsi->attrtimeo_timestamp = now;
1105 * Avoid jiffy wraparound issues with nfsi->last_updated
1107 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1108 nfsi->last_updated = nfsi->read_cache_jiffies;
1110 invalid &= ~NFS_INO_INVALID_ATTR;
1111 /* Don't invalidate the data if we were to blame */
1112 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1113 || S_ISLNK(inode->i_mode)))
1114 invalid &= ~NFS_INO_INVALID_DATA;
1115 if (!nfs_have_delegation(inode, FMODE_READ) ||
1116 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1117 nfsi->cache_validity |= invalid;
1118 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1123 * Big trouble! The inode has become a different object.
1125 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1126 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1129 * No need to worry about unhashing the dentry, as the
1130 * lookup validation will know that the inode is bad.
1131 * (But we fall through to invalidate the caches.)
1133 nfs_invalidate_inode(inode);
1137 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1138 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1139 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1140 (long long)nfsi->fileid, (long long)fattr->fileid);
1145 #ifdef CONFIG_NFS_V4
1148 * Clean out any remaining NFSv4 state that might be left over due
1149 * to open() calls that passed nfs_atomic_lookup, but failed to call
1152 void nfs4_clear_inode(struct inode *inode)
1154 /* If we are holding a delegation, return it! */
1155 nfs_inode_return_delegation_noreclaim(inode);
1156 /* First call standard NFS clear_inode() code */
1157 nfs_clear_inode(inode);
1161 struct inode *nfs_alloc_inode(struct super_block *sb)
1163 struct nfs_inode *nfsi;
1164 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1168 nfsi->cache_validity = 0UL;
1169 #ifdef CONFIG_NFS_V3_ACL
1170 nfsi->acl_access = ERR_PTR(-EAGAIN);
1171 nfsi->acl_default = ERR_PTR(-EAGAIN);
1173 #ifdef CONFIG_NFS_V4
1174 nfsi->nfs4_acl = NULL;
1175 #endif /* CONFIG_NFS_V4 */
1176 return &nfsi->vfs_inode;
1179 void nfs_destroy_inode(struct inode *inode)
1181 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1184 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1186 #ifdef CONFIG_NFS_V4
1187 INIT_LIST_HEAD(&nfsi->open_states);
1188 nfsi->delegation = NULL;
1189 nfsi->delegation_state = 0;
1190 init_rwsem(&nfsi->rwsem);
1194 static void init_once(struct kmem_cache * cachep, void *foo)
1196 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1198 inode_init_once(&nfsi->vfs_inode);
1199 INIT_LIST_HEAD(&nfsi->open_files);
1200 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1201 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1202 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1205 atomic_set(&nfsi->silly_count, 1);
1206 INIT_HLIST_HEAD(&nfsi->silly_list);
1207 init_waitqueue_head(&nfsi->waitqueue);
1208 nfs4_init_once(nfsi);
1211 static int __init nfs_init_inodecache(void)
1213 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1214 sizeof(struct nfs_inode),
1215 0, (SLAB_RECLAIM_ACCOUNT|
1218 if (nfs_inode_cachep == NULL)
1224 static void nfs_destroy_inodecache(void)
1226 kmem_cache_destroy(nfs_inode_cachep);
1229 struct workqueue_struct *nfsiod_workqueue;
1232 * start up the nfsiod workqueue
1234 static int nfsiod_start(void)
1236 struct workqueue_struct *wq;
1237 dprintk("RPC: creating workqueue nfsiod\n");
1238 wq = create_singlethread_workqueue("nfsiod");
1241 nfsiod_workqueue = wq;
1246 * Destroy the nfsiod workqueue
1248 static void nfsiod_stop(void)
1250 struct workqueue_struct *wq;
1252 wq = nfsiod_workqueue;
1255 nfsiod_workqueue = NULL;
1256 destroy_workqueue(wq);
1262 static int __init init_nfs_fs(void)
1266 err = nfsiod_start();
1270 err = nfs_fs_proc_init();
1274 err = nfs_init_nfspagecache();
1278 err = nfs_init_inodecache();
1282 err = nfs_init_readpagecache();
1286 err = nfs_init_writepagecache();
1290 err = nfs_init_directcache();
1294 #ifdef CONFIG_PROC_FS
1295 rpc_proc_register(&nfs_rpcstat);
1297 if ((err = register_nfs_fs()) != 0)
1301 #ifdef CONFIG_PROC_FS
1302 rpc_proc_unregister("nfs");
1304 nfs_destroy_directcache();
1306 nfs_destroy_writepagecache();
1308 nfs_destroy_readpagecache();
1310 nfs_destroy_inodecache();
1312 nfs_destroy_nfspagecache();
1321 static void __exit exit_nfs_fs(void)
1323 nfs_destroy_directcache();
1324 nfs_destroy_writepagecache();
1325 nfs_destroy_readpagecache();
1326 nfs_destroy_inodecache();
1327 nfs_destroy_nfspagecache();
1328 #ifdef CONFIG_PROC_FS
1329 rpc_proc_unregister("nfs");
1331 unregister_nfs_fs();
1336 /* Not quite true; I just maintain it */
1337 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1338 MODULE_LICENSE("GPL");
1339 module_param(enable_ino64, bool, 0644);
1341 module_init(init_nfs_fs)
1342 module_exit(exit_nfs_fs)