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)
373 /* Write all dirty data */
374 if (S_ISREG(inode->i_mode)) {
375 filemap_write_and_wait(inode->i_mapping);
379 * Return any delegations if we're going to change ACLs
381 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
382 nfs_inode_return_delegation(inode);
383 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
385 nfs_refresh_inode(inode, &fattr);
390 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
391 * @inode: inode of the file used
392 * @offset: file offset to start truncating
394 * This is a copy of the common vmtruncate, but with the locking
395 * corrected to take into account the fact that NFS requires
396 * inode->i_size to be updated under the inode->i_lock.
398 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
400 if (i_size_read(inode) < offset) {
403 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
404 if (limit != RLIM_INFINITY && offset > limit)
406 if (offset > inode->i_sb->s_maxbytes)
408 spin_lock(&inode->i_lock);
409 i_size_write(inode, offset);
410 spin_unlock(&inode->i_lock);
412 struct address_space *mapping = inode->i_mapping;
415 * truncation of in-use swapfiles is disallowed - it would
416 * cause subsequent swapout to scribble on the now-freed
419 if (IS_SWAPFILE(inode))
421 spin_lock(&inode->i_lock);
422 i_size_write(inode, offset);
423 spin_unlock(&inode->i_lock);
426 * unmap_mapping_range is called twice, first simply for
427 * efficiency so that truncate_inode_pages does fewer
428 * single-page unmaps. However after this first call, and
429 * before truncate_inode_pages finishes, it is possible for
430 * private pages to be COWed, which remain after
431 * truncate_inode_pages finishes, hence the second
432 * unmap_mapping_range call must be made for correctness.
434 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
435 truncate_inode_pages(mapping, offset);
436 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
440 send_sig(SIGXFSZ, current, 0);
446 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
447 * @inode: pointer to struct inode
448 * @attr: pointer to struct iattr
450 * Note: we do this in the *proc.c in order to ensure that
451 * it works for things like exclusive creates too.
453 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
455 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
456 if ((attr->ia_valid & ATTR_MODE) != 0) {
457 int mode = attr->ia_mode & S_IALLUGO;
458 mode |= inode->i_mode & ~S_IALLUGO;
459 inode->i_mode = mode;
461 if ((attr->ia_valid & ATTR_UID) != 0)
462 inode->i_uid = attr->ia_uid;
463 if ((attr->ia_valid & ATTR_GID) != 0)
464 inode->i_gid = attr->ia_gid;
465 spin_lock(&inode->i_lock);
466 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
467 spin_unlock(&inode->i_lock);
469 if ((attr->ia_valid & ATTR_SIZE) != 0) {
470 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
471 nfs_vmtruncate(inode, attr->ia_size);
475 static int nfs_wait_schedule(void *word)
477 if (signal_pending(current))
484 * Wait for the inode to get unlocked.
486 static int nfs_wait_on_inode(struct inode *inode)
488 struct nfs_inode *nfsi = NFS_I(inode);
491 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
492 nfs_wait_schedule, TASK_KILLABLE);
497 static void nfs_wake_up_inode(struct inode *inode)
499 struct nfs_inode *nfsi = NFS_I(inode);
501 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
502 smp_mb__after_clear_bit();
503 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
506 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
508 struct inode *inode = dentry->d_inode;
509 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
513 * Flush out writes to the server in order to update c/mtime.
515 * Hold the i_mutex to suspend application writes temporarily;
516 * this prevents long-running writing applications from blocking
519 if (S_ISREG(inode->i_mode)) {
520 mutex_lock(&inode->i_mutex);
521 nfs_wb_nocommit(inode);
522 mutex_unlock(&inode->i_mutex);
526 * We may force a getattr if the user cares about atime.
528 * Note that we only have to check the vfsmount flags here:
529 * - NFS always sets S_NOATIME by so checking it would give a
531 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
532 * no point in checking those.
534 if ((mnt->mnt_flags & MNT_NOATIME) ||
535 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
539 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
541 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
543 generic_fillattr(inode, stat);
544 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
549 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
551 struct nfs_open_context *ctx;
553 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
555 ctx->path.dentry = dget(dentry);
556 ctx->path.mnt = mntget(mnt);
557 ctx->cred = get_rpccred(cred);
559 ctx->lockowner = current->files;
563 atomic_set(&ctx->count, 1);
568 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
571 atomic_inc(&ctx->count);
575 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
582 inode = ctx->path.dentry->d_inode;
583 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
585 list_del(&ctx->list);
586 spin_unlock(&inode->i_lock);
587 if (ctx->state != NULL) {
589 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
591 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
593 if (ctx->cred != NULL)
594 put_rpccred(ctx->cred);
595 path_put(&ctx->path);
599 void put_nfs_open_context(struct nfs_open_context *ctx)
601 __put_nfs_open_context(ctx, 0);
604 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
606 __put_nfs_open_context(ctx, 1);
610 * Ensure that mmap has a recent RPC credential for use when writing out
613 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
615 struct inode *inode = filp->f_path.dentry->d_inode;
616 struct nfs_inode *nfsi = NFS_I(inode);
618 filp->private_data = get_nfs_open_context(ctx);
619 spin_lock(&inode->i_lock);
620 list_add(&ctx->list, &nfsi->open_files);
621 spin_unlock(&inode->i_lock);
625 * Given an inode, search for an open context with the desired characteristics
627 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
629 struct nfs_inode *nfsi = NFS_I(inode);
630 struct nfs_open_context *pos, *ctx = NULL;
632 spin_lock(&inode->i_lock);
633 list_for_each_entry(pos, &nfsi->open_files, list) {
634 if (cred != NULL && pos->cred != cred)
636 if ((pos->mode & mode) == mode) {
637 ctx = get_nfs_open_context(pos);
641 spin_unlock(&inode->i_lock);
645 static void nfs_file_clear_open_context(struct file *filp)
647 struct inode *inode = filp->f_path.dentry->d_inode;
648 struct nfs_open_context *ctx = nfs_file_open_context(filp);
651 filp->private_data = NULL;
652 spin_lock(&inode->i_lock);
653 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
654 spin_unlock(&inode->i_lock);
655 put_nfs_open_context_sync(ctx);
660 * These allocate and release file read/write context information.
662 int nfs_open(struct inode *inode, struct file *filp)
664 struct nfs_open_context *ctx;
665 struct rpc_cred *cred;
667 cred = rpc_lookup_cred();
669 return PTR_ERR(cred);
670 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
674 ctx->mode = filp->f_mode;
675 nfs_file_set_open_context(filp, ctx);
676 put_nfs_open_context(ctx);
680 int nfs_release(struct inode *inode, struct file *filp)
682 nfs_file_clear_open_context(filp);
687 * This function is called whenever some part of NFS notices that
688 * the cached attributes have to be refreshed.
691 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
693 int status = -ESTALE;
694 struct nfs_fattr fattr;
695 struct nfs_inode *nfsi = NFS_I(inode);
697 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
698 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
700 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
701 if (is_bad_inode(inode))
703 if (NFS_STALE(inode))
706 status = nfs_wait_on_inode(inode);
711 if (NFS_STALE(inode))
714 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
716 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
718 (long long)NFS_FILEID(inode), status);
719 if (status == -ESTALE) {
720 nfs_zap_caches(inode);
721 if (!S_ISDIR(inode->i_mode))
722 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
727 status = nfs_refresh_inode(inode, &fattr);
729 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
731 (long long)NFS_FILEID(inode), status);
735 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
736 nfs_zap_acl_cache(inode);
738 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
740 (long long)NFS_FILEID(inode));
743 nfs_wake_up_inode(inode);
749 int nfs_attribute_timeout(struct inode *inode)
751 struct nfs_inode *nfsi = NFS_I(inode);
753 if (nfs_have_delegation(inode, FMODE_READ))
756 * Special case: if the attribute timeout is set to 0, then always
757 * treat the cache as having expired (unless holding
760 if (nfsi->attrtimeo == 0)
762 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
766 * nfs_revalidate_inode - Revalidate the inode attributes
767 * @server - pointer to nfs_server struct
768 * @inode - pointer to inode struct
770 * Updates inode attribute information by retrieving the data from the server.
772 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
774 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
775 && !nfs_attribute_timeout(inode))
776 return NFS_STALE(inode) ? -ESTALE : 0;
777 return __nfs_revalidate_inode(server, inode);
780 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
782 struct nfs_inode *nfsi = NFS_I(inode);
784 if (mapping->nrpages != 0) {
785 int ret = invalidate_inode_pages2(mapping);
789 spin_lock(&inode->i_lock);
790 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
791 if (S_ISDIR(inode->i_mode))
792 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
793 spin_unlock(&inode->i_lock);
794 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
795 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
796 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
800 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
804 mutex_lock(&inode->i_mutex);
805 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
806 ret = nfs_sync_mapping(mapping);
808 ret = nfs_invalidate_mapping_nolock(inode, mapping);
810 mutex_unlock(&inode->i_mutex);
815 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
816 * @inode - pointer to host inode
817 * @mapping - pointer to mapping
819 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
821 struct nfs_inode *nfsi = NFS_I(inode);
824 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
825 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
826 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
830 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
831 ret = nfs_invalidate_mapping_nolock(inode, mapping);
837 * nfs_revalidate_mapping - Revalidate the pagecache
838 * @inode - pointer to host inode
839 * @mapping - pointer to mapping
841 * This version of the function will take the inode->i_mutex and attempt to
842 * flush out all dirty data if it needs to invalidate the page cache.
844 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
846 struct nfs_inode *nfsi = NFS_I(inode);
849 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
850 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
851 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
855 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
856 ret = nfs_invalidate_mapping(inode, mapping);
861 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
863 struct nfs_inode *nfsi = NFS_I(inode);
865 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
866 nfsi->change_attr == fattr->pre_change_attr) {
867 nfsi->change_attr = fattr->change_attr;
868 if (S_ISDIR(inode->i_mode))
869 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
871 /* If we have atomic WCC data, we may update some attributes */
872 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
873 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
874 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
875 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
876 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
877 if (S_ISDIR(inode->i_mode))
878 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
880 if (i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) &&
882 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
887 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
888 * @inode - pointer to inode
889 * @fattr - updated attributes
891 * Verifies the attribute cache. If we have just changed the attributes,
892 * so that fattr carries weak cache consistency data, then it may
893 * also update the ctime/mtime/change_attribute.
895 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
897 struct nfs_inode *nfsi = NFS_I(inode);
898 loff_t cur_size, new_isize;
899 unsigned long invalid = 0;
902 /* Has the inode gone and changed behind our back? */
903 if (nfsi->fileid != fattr->fileid
904 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
908 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
909 nfsi->change_attr != fattr->change_attr)
910 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
912 /* Verify a few of the more important attributes */
913 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
914 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
916 cur_size = i_size_read(inode);
917 new_isize = nfs_size_to_loff_t(fattr->size);
918 if (cur_size != new_isize && nfsi->npages == 0)
919 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
921 /* Have any file permissions changed? */
922 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
923 || inode->i_uid != fattr->uid
924 || inode->i_gid != fattr->gid)
925 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
927 /* Has the link count changed? */
928 if (inode->i_nlink != fattr->nlink)
929 invalid |= NFS_INO_INVALID_ATTR;
931 if (!timespec_equal(&inode->i_atime, &fattr->atime))
932 invalid |= NFS_INO_INVALID_ATIME;
935 nfsi->cache_validity |= invalid;
937 nfsi->read_cache_jiffies = fattr->time_start;
941 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
943 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
946 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
948 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
952 * nfs_inode_attrs_need_update - check if the inode attributes need updating
953 * @inode - pointer to inode
954 * @fattr - attributes
956 * Attempt to divine whether or not an RPC call reply carrying stale
957 * attributes got scheduled after another call carrying updated ones.
959 * To do so, the function first assumes that a more recent ctime means
960 * that the attributes in fattr are newer, however it also attempt to
961 * catch the case where ctime either didn't change, or went backwards
962 * (if someone reset the clock on the server) by looking at whether
963 * or not this RPC call was started after the inode was last updated.
964 * Note also the check for jiffy wraparound if the last_updated timestamp
965 * is later than 'jiffies'.
967 * The function returns 'true' if it thinks the attributes in 'fattr' are
968 * more recent than the ones cached in the inode.
971 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
973 const struct nfs_inode *nfsi = NFS_I(inode);
975 return nfs_ctime_need_update(inode, fattr) ||
976 nfs_size_need_update(inode, fattr) ||
977 time_after(fattr->time_start, nfsi->last_updated) ||
978 time_after(nfsi->last_updated, jiffies);
981 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
983 if (nfs_inode_attrs_need_update(inode, fattr))
984 return nfs_update_inode(inode, fattr);
985 return nfs_check_inode_attributes(inode, fattr);
989 * nfs_refresh_inode - try to update the inode attribute cache
990 * @inode - pointer to inode
991 * @fattr - updated attributes
993 * Check that an RPC call that returned attributes has not overlapped with
994 * other recent updates of the inode metadata, then decide whether it is
995 * safe to do a full update of the inode attributes, or whether just to
996 * call nfs_check_inode_attributes.
998 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1002 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1004 spin_lock(&inode->i_lock);
1005 status = nfs_refresh_inode_locked(inode, fattr);
1006 spin_unlock(&inode->i_lock);
1010 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1012 struct nfs_inode *nfsi = NFS_I(inode);
1014 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1015 if (S_ISDIR(inode->i_mode))
1016 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1017 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1019 return nfs_refresh_inode_locked(inode, fattr);
1023 * nfs_post_op_update_inode - try to update the inode attribute cache
1024 * @inode - pointer to inode
1025 * @fattr - updated attributes
1027 * After an operation that has changed the inode metadata, mark the
1028 * attribute cache as being invalid, then try to update it.
1030 * NB: if the server didn't return any post op attributes, this
1031 * function will force the retrieval of attributes before the next
1032 * NFS request. Thus it should be used only for operations that
1033 * are expected to change one or more attributes, to avoid
1034 * unnecessary NFS requests and trips through nfs_update_inode().
1036 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1040 spin_lock(&inode->i_lock);
1041 status = nfs_post_op_update_inode_locked(inode, fattr);
1042 spin_unlock(&inode->i_lock);
1047 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1048 * @inode - pointer to inode
1049 * @fattr - updated attributes
1051 * After an operation that has changed the inode metadata, mark the
1052 * attribute cache as being invalid, then try to update it. Fake up
1053 * weak cache consistency data, if none exist.
1055 * This function is mainly designed to be used by the ->write_done() functions.
1057 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1061 spin_lock(&inode->i_lock);
1062 /* Don't do a WCC update if these attributes are already stale */
1063 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1064 !nfs_inode_attrs_need_update(inode, fattr)) {
1065 fattr->valid &= ~(NFS_ATTR_WCC_V4|NFS_ATTR_WCC);
1068 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1069 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
1070 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1071 fattr->valid |= NFS_ATTR_WCC_V4;
1073 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
1074 (fattr->valid & NFS_ATTR_WCC) == 0) {
1075 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1076 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1077 fattr->pre_size = i_size_read(inode);
1078 fattr->valid |= NFS_ATTR_WCC;
1081 status = nfs_post_op_update_inode_locked(inode, fattr);
1082 spin_unlock(&inode->i_lock);
1087 * Many nfs protocol calls return the new file attributes after
1088 * an operation. Here we update the inode to reflect the state
1089 * of the server's inode.
1091 * This is a bit tricky because we have to make sure all dirty pages
1092 * have been sent off to the server before calling invalidate_inode_pages.
1093 * To make sure no other process adds more write requests while we try
1094 * our best to flush them, we make them sleep during the attribute refresh.
1096 * A very similar scenario holds for the dir cache.
1098 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1100 struct nfs_server *server;
1101 struct nfs_inode *nfsi = NFS_I(inode);
1102 loff_t cur_isize, new_isize;
1103 unsigned long invalid = 0;
1104 unsigned long now = jiffies;
1106 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1107 __func__, inode->i_sb->s_id, inode->i_ino,
1108 atomic_read(&inode->i_count), fattr->valid);
1110 if (nfsi->fileid != fattr->fileid)
1114 * Make sure the inode's type hasn't changed.
1116 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1119 server = NFS_SERVER(inode);
1120 /* Update the fsid? */
1121 if (S_ISDIR(inode->i_mode) &&
1122 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1123 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1124 server->fsid = fattr->fsid;
1127 * Update the read time so we don't revalidate too often.
1129 nfsi->read_cache_jiffies = fattr->time_start;
1131 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1132 | NFS_INO_REVAL_PAGECACHE);
1134 /* Do atomic weak cache consistency updates */
1135 nfs_wcc_update_inode(inode, fattr);
1137 /* More cache consistency checks */
1138 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1139 /* NFSv2/v3: Check if the mtime agrees */
1140 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1141 dprintk("NFS: mtime change on server for file %s/%ld\n",
1142 inode->i_sb->s_id, inode->i_ino);
1143 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1144 if (S_ISDIR(inode->i_mode))
1145 nfs_force_lookup_revalidate(inode);
1147 /* If ctime has changed we should definitely clear access+acl caches */
1148 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1149 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1150 } else if (nfsi->change_attr != fattr->change_attr) {
1151 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1152 inode->i_sb->s_id, inode->i_ino);
1153 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1154 if (S_ISDIR(inode->i_mode))
1155 nfs_force_lookup_revalidate(inode);
1158 /* Check if our cached file size is stale */
1159 new_isize = nfs_size_to_loff_t(fattr->size);
1160 cur_isize = i_size_read(inode);
1161 if (new_isize != cur_isize) {
1162 /* Do we perhaps have any outstanding writes, or has
1163 * the file grown beyond our last write? */
1164 if (nfsi->npages == 0 || new_isize > cur_isize) {
1165 i_size_write(inode, new_isize);
1166 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1168 dprintk("NFS: isize change on server for file %s/%ld\n",
1169 inode->i_sb->s_id, inode->i_ino);
1173 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1174 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1175 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1176 nfsi->change_attr = fattr->change_attr;
1178 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1179 inode->i_uid != fattr->uid ||
1180 inode->i_gid != fattr->gid)
1181 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1183 inode->i_mode = fattr->mode;
1184 inode->i_nlink = fattr->nlink;
1185 inode->i_uid = fattr->uid;
1186 inode->i_gid = fattr->gid;
1188 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1190 * report the blocks in 512byte units
1192 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1194 inode->i_blocks = fattr->du.nfs2.blocks;
1197 /* Update attrtimeo value if we're out of the unstable period */
1198 if (invalid & NFS_INO_INVALID_ATTR) {
1199 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1200 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1201 nfsi->attrtimeo_timestamp = now;
1202 nfsi->last_updated = now;
1204 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1205 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1206 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1207 nfsi->attrtimeo_timestamp = now;
1210 * Avoid jiffy wraparound issues with nfsi->last_updated
1212 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1213 nfsi->last_updated = nfsi->read_cache_jiffies;
1215 invalid &= ~NFS_INO_INVALID_ATTR;
1216 /* Don't invalidate the data if we were to blame */
1217 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1218 || S_ISLNK(inode->i_mode)))
1219 invalid &= ~NFS_INO_INVALID_DATA;
1220 if (!nfs_have_delegation(inode, FMODE_READ) ||
1221 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1222 nfsi->cache_validity |= invalid;
1223 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1228 * Big trouble! The inode has become a different object.
1230 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1231 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1234 * No need to worry about unhashing the dentry, as the
1235 * lookup validation will know that the inode is bad.
1236 * (But we fall through to invalidate the caches.)
1238 nfs_invalidate_inode(inode);
1242 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1243 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1244 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1245 (long long)nfsi->fileid, (long long)fattr->fileid);
1250 #ifdef CONFIG_NFS_V4
1253 * Clean out any remaining NFSv4 state that might be left over due
1254 * to open() calls that passed nfs_atomic_lookup, but failed to call
1257 void nfs4_clear_inode(struct inode *inode)
1259 /* If we are holding a delegation, return it! */
1260 nfs_inode_return_delegation_noreclaim(inode);
1261 /* First call standard NFS clear_inode() code */
1262 nfs_clear_inode(inode);
1266 struct inode *nfs_alloc_inode(struct super_block *sb)
1268 struct nfs_inode *nfsi;
1269 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1273 nfsi->cache_validity = 0UL;
1274 #ifdef CONFIG_NFS_V3_ACL
1275 nfsi->acl_access = ERR_PTR(-EAGAIN);
1276 nfsi->acl_default = ERR_PTR(-EAGAIN);
1278 #ifdef CONFIG_NFS_V4
1279 nfsi->nfs4_acl = NULL;
1280 #endif /* CONFIG_NFS_V4 */
1281 return &nfsi->vfs_inode;
1284 void nfs_destroy_inode(struct inode *inode)
1286 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1289 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1291 #ifdef CONFIG_NFS_V4
1292 INIT_LIST_HEAD(&nfsi->open_states);
1293 nfsi->delegation = NULL;
1294 nfsi->delegation_state = 0;
1295 init_rwsem(&nfsi->rwsem);
1299 static void init_once(void *foo)
1301 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1303 inode_init_once(&nfsi->vfs_inode);
1304 INIT_LIST_HEAD(&nfsi->open_files);
1305 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1306 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1307 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1310 atomic_set(&nfsi->silly_count, 1);
1311 INIT_HLIST_HEAD(&nfsi->silly_list);
1312 init_waitqueue_head(&nfsi->waitqueue);
1313 nfs4_init_once(nfsi);
1316 static int __init nfs_init_inodecache(void)
1318 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1319 sizeof(struct nfs_inode),
1320 0, (SLAB_RECLAIM_ACCOUNT|
1323 if (nfs_inode_cachep == NULL)
1329 static void nfs_destroy_inodecache(void)
1331 kmem_cache_destroy(nfs_inode_cachep);
1334 struct workqueue_struct *nfsiod_workqueue;
1337 * start up the nfsiod workqueue
1339 static int nfsiod_start(void)
1341 struct workqueue_struct *wq;
1342 dprintk("RPC: creating workqueue nfsiod\n");
1343 wq = create_singlethread_workqueue("nfsiod");
1346 nfsiod_workqueue = wq;
1351 * Destroy the nfsiod workqueue
1353 static void nfsiod_stop(void)
1355 struct workqueue_struct *wq;
1357 wq = nfsiod_workqueue;
1360 nfsiod_workqueue = NULL;
1361 destroy_workqueue(wq);
1367 static int __init init_nfs_fs(void)
1371 err = nfsiod_start();
1375 err = nfs_fs_proc_init();
1379 err = nfs_init_nfspagecache();
1383 err = nfs_init_inodecache();
1387 err = nfs_init_readpagecache();
1391 err = nfs_init_writepagecache();
1395 err = nfs_init_directcache();
1399 #ifdef CONFIG_PROC_FS
1400 rpc_proc_register(&nfs_rpcstat);
1402 if ((err = register_nfs_fs()) != 0)
1406 #ifdef CONFIG_PROC_FS
1407 rpc_proc_unregister("nfs");
1409 nfs_destroy_directcache();
1411 nfs_destroy_writepagecache();
1413 nfs_destroy_readpagecache();
1415 nfs_destroy_inodecache();
1417 nfs_destroy_nfspagecache();
1426 static void __exit exit_nfs_fs(void)
1428 nfs_destroy_directcache();
1429 nfs_destroy_writepagecache();
1430 nfs_destroy_readpagecache();
1431 nfs_destroy_inodecache();
1432 nfs_destroy_nfspagecache();
1433 #ifdef CONFIG_PROC_FS
1434 rpc_proc_unregister("nfs");
1436 unregister_nfs_fs();
1441 /* Not quite true; I just maintain it */
1442 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1443 MODULE_LICENSE("GPL");
1444 module_param(enable_ino64, bool, 0644);
1446 module_init(init_nfs_fs)
1447 module_exit(exit_nfs_fs)