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_vmtruncate - unmap mappings "freed" by truncate() syscall
393 * @inode: inode of the file used
394 * @offset: file offset to start truncating
396 * This is a copy of the common vmtruncate, but with the locking
397 * corrected to take into account the fact that NFS requires
398 * inode->i_size to be updated under the inode->i_lock.
400 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
402 if (i_size_read(inode) < offset) {
405 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
406 if (limit != RLIM_INFINITY && offset > limit)
408 if (offset > inode->i_sb->s_maxbytes)
410 spin_lock(&inode->i_lock);
411 i_size_write(inode, offset);
412 spin_unlock(&inode->i_lock);
414 struct address_space *mapping = inode->i_mapping;
417 * truncation of in-use swapfiles is disallowed - it would
418 * cause subsequent swapout to scribble on the now-freed
421 if (IS_SWAPFILE(inode))
423 spin_lock(&inode->i_lock);
424 i_size_write(inode, offset);
425 spin_unlock(&inode->i_lock);
428 * unmap_mapping_range is called twice, first simply for
429 * efficiency so that truncate_inode_pages does fewer
430 * single-page unmaps. However after this first call, and
431 * before truncate_inode_pages finishes, it is possible for
432 * private pages to be COWed, which remain after
433 * truncate_inode_pages finishes, hence the second
434 * unmap_mapping_range call must be made for correctness.
436 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
437 truncate_inode_pages(mapping, offset);
438 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
442 send_sig(SIGXFSZ, current, 0);
448 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
449 * @inode: pointer to struct inode
450 * @attr: pointer to struct iattr
452 * Note: we do this in the *proc.c in order to ensure that
453 * it works for things like exclusive creates too.
455 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
457 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
458 if ((attr->ia_valid & ATTR_MODE) != 0) {
459 int mode = attr->ia_mode & S_IALLUGO;
460 mode |= inode->i_mode & ~S_IALLUGO;
461 inode->i_mode = mode;
463 if ((attr->ia_valid & ATTR_UID) != 0)
464 inode->i_uid = attr->ia_uid;
465 if ((attr->ia_valid & ATTR_GID) != 0)
466 inode->i_gid = attr->ia_gid;
467 spin_lock(&inode->i_lock);
468 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
469 spin_unlock(&inode->i_lock);
471 if ((attr->ia_valid & ATTR_SIZE) != 0) {
472 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
473 nfs_vmtruncate(inode, attr->ia_size);
477 static int nfs_wait_schedule(void *word)
479 if (signal_pending(current))
486 * Wait for the inode to get unlocked.
488 static int nfs_wait_on_inode(struct inode *inode)
490 struct nfs_inode *nfsi = NFS_I(inode);
493 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
494 nfs_wait_schedule, TASK_KILLABLE);
499 static void nfs_wake_up_inode(struct inode *inode)
501 struct nfs_inode *nfsi = NFS_I(inode);
503 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
504 smp_mb__after_clear_bit();
505 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
508 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
510 struct inode *inode = dentry->d_inode;
511 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
515 * Flush out writes to the server in order to update c/mtime.
517 * Hold the i_mutex to suspend application writes temporarily;
518 * this prevents long-running writing applications from blocking
521 if (S_ISREG(inode->i_mode)) {
522 mutex_lock(&inode->i_mutex);
523 nfs_wb_nocommit(inode);
524 mutex_unlock(&inode->i_mutex);
528 * We may force a getattr if the user cares about atime.
530 * Note that we only have to check the vfsmount flags here:
531 * - NFS always sets S_NOATIME by so checking it would give a
533 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
534 * no point in checking those.
536 if ((mnt->mnt_flags & MNT_NOATIME) ||
537 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
541 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
543 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
545 generic_fillattr(inode, stat);
546 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
551 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
553 struct nfs_open_context *ctx;
555 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
557 ctx->path.dentry = dget(dentry);
558 ctx->path.mnt = mntget(mnt);
559 ctx->cred = get_rpccred(cred);
561 ctx->lockowner = current->files;
565 atomic_set(&ctx->count, 1);
570 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
573 atomic_inc(&ctx->count);
577 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
584 inode = ctx->path.dentry->d_inode;
585 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
587 list_del(&ctx->list);
588 spin_unlock(&inode->i_lock);
589 if (ctx->state != NULL) {
591 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
593 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
595 if (ctx->cred != NULL)
596 put_rpccred(ctx->cred);
597 path_put(&ctx->path);
601 void put_nfs_open_context(struct nfs_open_context *ctx)
603 __put_nfs_open_context(ctx, 0);
606 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
608 __put_nfs_open_context(ctx, 1);
612 * Ensure that mmap has a recent RPC credential for use when writing out
615 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
617 struct inode *inode = filp->f_path.dentry->d_inode;
618 struct nfs_inode *nfsi = NFS_I(inode);
620 filp->private_data = get_nfs_open_context(ctx);
621 spin_lock(&inode->i_lock);
622 list_add(&ctx->list, &nfsi->open_files);
623 spin_unlock(&inode->i_lock);
627 * Given an inode, search for an open context with the desired characteristics
629 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
631 struct nfs_inode *nfsi = NFS_I(inode);
632 struct nfs_open_context *pos, *ctx = NULL;
634 spin_lock(&inode->i_lock);
635 list_for_each_entry(pos, &nfsi->open_files, list) {
636 if (cred != NULL && pos->cred != cred)
638 if ((pos->mode & mode) == mode) {
639 ctx = get_nfs_open_context(pos);
643 spin_unlock(&inode->i_lock);
647 static void nfs_file_clear_open_context(struct file *filp)
649 struct inode *inode = filp->f_path.dentry->d_inode;
650 struct nfs_open_context *ctx = nfs_file_open_context(filp);
653 filp->private_data = NULL;
654 spin_lock(&inode->i_lock);
655 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
656 spin_unlock(&inode->i_lock);
657 put_nfs_open_context_sync(ctx);
662 * These allocate and release file read/write context information.
664 int nfs_open(struct inode *inode, struct file *filp)
666 struct nfs_open_context *ctx;
667 struct rpc_cred *cred;
669 cred = rpc_lookup_cred();
671 return PTR_ERR(cred);
672 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
676 ctx->mode = filp->f_mode;
677 nfs_file_set_open_context(filp, ctx);
678 put_nfs_open_context(ctx);
682 int nfs_release(struct inode *inode, struct file *filp)
684 nfs_file_clear_open_context(filp);
689 * This function is called whenever some part of NFS notices that
690 * the cached attributes have to be refreshed.
693 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
695 int status = -ESTALE;
696 struct nfs_fattr fattr;
697 struct nfs_inode *nfsi = NFS_I(inode);
699 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
700 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
702 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
704 if (is_bad_inode(inode))
706 if (NFS_STALE(inode))
709 status = nfs_wait_on_inode(inode);
714 if (NFS_STALE(inode))
717 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
719 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
721 (long long)NFS_FILEID(inode), status);
722 if (status == -ESTALE) {
723 nfs_zap_caches(inode);
724 if (!S_ISDIR(inode->i_mode))
725 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
730 spin_lock(&inode->i_lock);
731 status = nfs_update_inode(inode, &fattr);
733 spin_unlock(&inode->i_lock);
734 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
736 (long long)NFS_FILEID(inode), status);
739 spin_unlock(&inode->i_lock);
741 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
742 nfs_zap_acl_cache(inode);
744 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
746 (long long)NFS_FILEID(inode));
749 nfs_wake_up_inode(inode);
756 int nfs_attribute_timeout(struct inode *inode)
758 struct nfs_inode *nfsi = NFS_I(inode);
760 if (nfs_have_delegation(inode, FMODE_READ))
763 * Special case: if the attribute timeout is set to 0, then always
764 * treat the cache as having expired (unless holding
767 if (nfsi->attrtimeo == 0)
769 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
773 * nfs_revalidate_inode - Revalidate the inode attributes
774 * @server - pointer to nfs_server struct
775 * @inode - pointer to inode struct
777 * Updates inode attribute information by retrieving the data from the server.
779 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
781 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
782 && !nfs_attribute_timeout(inode))
783 return NFS_STALE(inode) ? -ESTALE : 0;
784 return __nfs_revalidate_inode(server, inode);
787 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
789 struct nfs_inode *nfsi = NFS_I(inode);
791 if (mapping->nrpages != 0) {
792 int ret = invalidate_inode_pages2(mapping);
796 spin_lock(&inode->i_lock);
797 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
798 if (S_ISDIR(inode->i_mode))
799 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
800 spin_unlock(&inode->i_lock);
801 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
802 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
803 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
807 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
811 mutex_lock(&inode->i_mutex);
812 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
813 ret = nfs_sync_mapping(mapping);
815 ret = nfs_invalidate_mapping_nolock(inode, mapping);
817 mutex_unlock(&inode->i_mutex);
822 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
823 * @inode - pointer to host inode
824 * @mapping - pointer to mapping
826 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
828 struct nfs_inode *nfsi = NFS_I(inode);
831 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
832 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
833 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
837 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
838 ret = nfs_invalidate_mapping_nolock(inode, mapping);
844 * nfs_revalidate_mapping - Revalidate the pagecache
845 * @inode - pointer to host inode
846 * @mapping - pointer to mapping
848 * This version of the function will take the inode->i_mutex and attempt to
849 * flush out all dirty data if it needs to invalidate the page cache.
851 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
853 struct nfs_inode *nfsi = NFS_I(inode);
856 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
857 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
858 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
862 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
863 ret = nfs_invalidate_mapping(inode, mapping);
868 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
870 struct nfs_inode *nfsi = NFS_I(inode);
872 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
873 nfsi->change_attr == fattr->pre_change_attr) {
874 nfsi->change_attr = fattr->change_attr;
875 if (S_ISDIR(inode->i_mode))
876 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
878 /* If we have atomic WCC data, we may update some attributes */
879 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
880 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
881 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
882 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
883 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
884 if (S_ISDIR(inode->i_mode))
885 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
887 if (i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) &&
889 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
894 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
895 * @inode - pointer to inode
896 * @fattr - updated attributes
898 * Verifies the attribute cache. If we have just changed the attributes,
899 * so that fattr carries weak cache consistency data, then it may
900 * also update the ctime/mtime/change_attribute.
902 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
904 struct nfs_inode *nfsi = NFS_I(inode);
905 loff_t cur_size, new_isize;
906 unsigned long invalid = 0;
909 /* Has the inode gone and changed behind our back? */
910 if (nfsi->fileid != fattr->fileid
911 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
915 /* Do atomic weak cache consistency updates */
916 nfs_wcc_update_inode(inode, fattr);
918 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
919 nfsi->change_attr != fattr->change_attr)
920 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
922 /* Verify a few of the more important attributes */
923 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
924 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
926 cur_size = i_size_read(inode);
927 new_isize = nfs_size_to_loff_t(fattr->size);
928 if (cur_size != new_isize && nfsi->npages == 0)
929 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
931 /* Have any file permissions changed? */
932 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
933 || inode->i_uid != fattr->uid
934 || inode->i_gid != fattr->gid)
935 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
937 /* Has the link count changed? */
938 if (inode->i_nlink != fattr->nlink)
939 invalid |= NFS_INO_INVALID_ATTR;
941 if (!timespec_equal(&inode->i_atime, &fattr->atime))
942 invalid |= NFS_INO_INVALID_ATIME;
945 nfsi->cache_validity |= invalid;
947 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
948 | NFS_INO_INVALID_ATIME
949 | NFS_INO_REVAL_PAGECACHE);
951 nfsi->read_cache_jiffies = fattr->time_start;
956 * nfs_refresh_inode - try to update the inode attribute cache
957 * @inode - pointer to inode
958 * @fattr - updated attributes
960 * Check that an RPC call that returned attributes has not overlapped with
961 * other recent updates of the inode metadata, then decide whether it is
962 * safe to do a full update of the inode attributes, or whether just to
963 * call nfs_check_inode_attributes.
965 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
967 struct nfs_inode *nfsi = NFS_I(inode);
970 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
972 spin_lock(&inode->i_lock);
973 if (time_after(fattr->time_start, nfsi->last_updated))
974 status = nfs_update_inode(inode, fattr);
976 status = nfs_check_inode_attributes(inode, fattr);
978 spin_unlock(&inode->i_lock);
983 * nfs_post_op_update_inode - try to update the inode attribute cache
984 * @inode - pointer to inode
985 * @fattr - updated attributes
987 * After an operation that has changed the inode metadata, mark the
988 * attribute cache as being invalid, then try to update it.
990 * NB: if the server didn't return any post op attributes, this
991 * function will force the retrieval of attributes before the next
992 * NFS request. Thus it should be used only for operations that
993 * are expected to change one or more attributes, to avoid
994 * unnecessary NFS requests and trips through nfs_update_inode().
996 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
998 struct nfs_inode *nfsi = NFS_I(inode);
1000 spin_lock(&inode->i_lock);
1001 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1002 if (S_ISDIR(inode->i_mode))
1003 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1004 spin_unlock(&inode->i_lock);
1005 return nfs_refresh_inode(inode, fattr);
1009 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1010 * @inode - pointer to inode
1011 * @fattr - updated attributes
1013 * After an operation that has changed the inode metadata, mark the
1014 * attribute cache as being invalid, then try to update it. Fake up
1015 * weak cache consistency data, if none exist.
1017 * This function is mainly designed to be used by the ->write_done() functions.
1019 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1021 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1022 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
1023 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1024 fattr->valid |= NFS_ATTR_WCC_V4;
1026 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
1027 (fattr->valid & NFS_ATTR_WCC) == 0) {
1028 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1029 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1030 fattr->pre_size = i_size_read(inode);
1031 fattr->valid |= NFS_ATTR_WCC;
1033 return nfs_post_op_update_inode(inode, fattr);
1037 * Many nfs protocol calls return the new file attributes after
1038 * an operation. Here we update the inode to reflect the state
1039 * of the server's inode.
1041 * This is a bit tricky because we have to make sure all dirty pages
1042 * have been sent off to the server before calling invalidate_inode_pages.
1043 * To make sure no other process adds more write requests while we try
1044 * our best to flush them, we make them sleep during the attribute refresh.
1046 * A very similar scenario holds for the dir cache.
1048 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1050 struct nfs_server *server;
1051 struct nfs_inode *nfsi = NFS_I(inode);
1052 loff_t cur_isize, new_isize;
1053 unsigned long invalid = 0;
1054 unsigned long now = jiffies;
1056 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1057 __func__, inode->i_sb->s_id, inode->i_ino,
1058 atomic_read(&inode->i_count), fattr->valid);
1060 if (nfsi->fileid != fattr->fileid)
1064 * Make sure the inode's type hasn't changed.
1066 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1069 server = NFS_SERVER(inode);
1070 /* Update the fsid? */
1071 if (S_ISDIR(inode->i_mode) &&
1072 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1073 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1074 server->fsid = fattr->fsid;
1077 * Update the read time so we don't revalidate too often.
1079 nfsi->read_cache_jiffies = fattr->time_start;
1081 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1082 | NFS_INO_REVAL_PAGECACHE);
1084 /* Do atomic weak cache consistency updates */
1085 nfs_wcc_update_inode(inode, fattr);
1087 /* More cache consistency checks */
1088 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1089 /* NFSv2/v3: Check if the mtime agrees */
1090 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1091 dprintk("NFS: mtime change on server for file %s/%ld\n",
1092 inode->i_sb->s_id, inode->i_ino);
1093 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1094 if (S_ISDIR(inode->i_mode))
1095 nfs_force_lookup_revalidate(inode);
1097 /* If ctime has changed we should definitely clear access+acl caches */
1098 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1099 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1100 } else if (nfsi->change_attr != fattr->change_attr) {
1101 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1102 inode->i_sb->s_id, inode->i_ino);
1103 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1104 if (S_ISDIR(inode->i_mode))
1105 nfs_force_lookup_revalidate(inode);
1108 /* Check if our cached file size is stale */
1109 new_isize = nfs_size_to_loff_t(fattr->size);
1110 cur_isize = i_size_read(inode);
1111 if (new_isize != cur_isize) {
1112 /* Do we perhaps have any outstanding writes, or has
1113 * the file grown beyond our last write? */
1114 if (nfsi->npages == 0 || new_isize > cur_isize) {
1115 i_size_write(inode, new_isize);
1116 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1118 dprintk("NFS: isize change on server for file %s/%ld\n",
1119 inode->i_sb->s_id, inode->i_ino);
1123 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1124 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1125 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1126 nfsi->change_attr = fattr->change_attr;
1128 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1129 inode->i_uid != fattr->uid ||
1130 inode->i_gid != fattr->gid)
1131 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1133 inode->i_mode = fattr->mode;
1134 inode->i_nlink = fattr->nlink;
1135 inode->i_uid = fattr->uid;
1136 inode->i_gid = fattr->gid;
1138 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1140 * report the blocks in 512byte units
1142 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1144 inode->i_blocks = fattr->du.nfs2.blocks;
1147 /* Update attrtimeo value if we're out of the unstable period */
1148 if (invalid & NFS_INO_INVALID_ATTR) {
1149 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1150 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1151 nfsi->attrtimeo_timestamp = now;
1152 nfsi->last_updated = now;
1154 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1155 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1156 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1157 nfsi->attrtimeo_timestamp = now;
1160 * Avoid jiffy wraparound issues with nfsi->last_updated
1162 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1163 nfsi->last_updated = nfsi->read_cache_jiffies;
1165 invalid &= ~NFS_INO_INVALID_ATTR;
1166 /* Don't invalidate the data if we were to blame */
1167 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1168 || S_ISLNK(inode->i_mode)))
1169 invalid &= ~NFS_INO_INVALID_DATA;
1170 if (!nfs_have_delegation(inode, FMODE_READ) ||
1171 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1172 nfsi->cache_validity |= invalid;
1173 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1178 * Big trouble! The inode has become a different object.
1180 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1181 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1184 * No need to worry about unhashing the dentry, as the
1185 * lookup validation will know that the inode is bad.
1186 * (But we fall through to invalidate the caches.)
1188 nfs_invalidate_inode(inode);
1192 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1193 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1194 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1195 (long long)nfsi->fileid, (long long)fattr->fileid);
1200 #ifdef CONFIG_NFS_V4
1203 * Clean out any remaining NFSv4 state that might be left over due
1204 * to open() calls that passed nfs_atomic_lookup, but failed to call
1207 void nfs4_clear_inode(struct inode *inode)
1209 /* If we are holding a delegation, return it! */
1210 nfs_inode_return_delegation_noreclaim(inode);
1211 /* First call standard NFS clear_inode() code */
1212 nfs_clear_inode(inode);
1216 struct inode *nfs_alloc_inode(struct super_block *sb)
1218 struct nfs_inode *nfsi;
1219 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1223 nfsi->cache_validity = 0UL;
1224 #ifdef CONFIG_NFS_V3_ACL
1225 nfsi->acl_access = ERR_PTR(-EAGAIN);
1226 nfsi->acl_default = ERR_PTR(-EAGAIN);
1228 #ifdef CONFIG_NFS_V4
1229 nfsi->nfs4_acl = NULL;
1230 #endif /* CONFIG_NFS_V4 */
1231 return &nfsi->vfs_inode;
1234 void nfs_destroy_inode(struct inode *inode)
1236 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1239 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1241 #ifdef CONFIG_NFS_V4
1242 INIT_LIST_HEAD(&nfsi->open_states);
1243 nfsi->delegation = NULL;
1244 nfsi->delegation_state = 0;
1245 init_rwsem(&nfsi->rwsem);
1249 static void init_once(struct kmem_cache * cachep, void *foo)
1251 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1253 inode_init_once(&nfsi->vfs_inode);
1254 INIT_LIST_HEAD(&nfsi->open_files);
1255 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1256 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1257 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1260 atomic_set(&nfsi->silly_count, 1);
1261 INIT_HLIST_HEAD(&nfsi->silly_list);
1262 init_waitqueue_head(&nfsi->waitqueue);
1263 nfs4_init_once(nfsi);
1266 static int __init nfs_init_inodecache(void)
1268 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1269 sizeof(struct nfs_inode),
1270 0, (SLAB_RECLAIM_ACCOUNT|
1273 if (nfs_inode_cachep == NULL)
1279 static void nfs_destroy_inodecache(void)
1281 kmem_cache_destroy(nfs_inode_cachep);
1284 struct workqueue_struct *nfsiod_workqueue;
1287 * start up the nfsiod workqueue
1289 static int nfsiod_start(void)
1291 struct workqueue_struct *wq;
1292 dprintk("RPC: creating workqueue nfsiod\n");
1293 wq = create_singlethread_workqueue("nfsiod");
1296 nfsiod_workqueue = wq;
1301 * Destroy the nfsiod workqueue
1303 static void nfsiod_stop(void)
1305 struct workqueue_struct *wq;
1307 wq = nfsiod_workqueue;
1310 nfsiod_workqueue = NULL;
1311 destroy_workqueue(wq);
1317 static int __init init_nfs_fs(void)
1321 err = nfsiod_start();
1325 err = nfs_fs_proc_init();
1329 err = nfs_init_nfspagecache();
1333 err = nfs_init_inodecache();
1337 err = nfs_init_readpagecache();
1341 err = nfs_init_writepagecache();
1345 err = nfs_init_directcache();
1349 #ifdef CONFIG_PROC_FS
1350 rpc_proc_register(&nfs_rpcstat);
1352 if ((err = register_nfs_fs()) != 0)
1356 #ifdef CONFIG_PROC_FS
1357 rpc_proc_unregister("nfs");
1359 nfs_destroy_directcache();
1361 nfs_destroy_writepagecache();
1363 nfs_destroy_readpagecache();
1365 nfs_destroy_inodecache();
1367 nfs_destroy_nfspagecache();
1376 static void __exit exit_nfs_fs(void)
1378 nfs_destroy_directcache();
1379 nfs_destroy_writepagecache();
1380 nfs_destroy_readpagecache();
1381 nfs_destroy_inodecache();
1382 nfs_destroy_nfspagecache();
1383 #ifdef CONFIG_PROC_FS
1384 rpc_proc_unregister("nfs");
1386 unregister_nfs_fs();
1391 /* Not quite true; I just maintain it */
1392 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1393 MODULE_LICENSE("GPL");
1394 module_param(enable_ino64, bool, 0644);
1396 module_init(init_nfs_fs)
1397 module_exit(exit_nfs_fs)