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 void nfs_zap_acl_cache(struct inode *);
62 static struct kmem_cache * nfs_inode_cachep;
64 static inline unsigned long
65 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
67 return nfs_fileid_to_ino_t(fattr->fileid);
71 * nfs_compat_user_ino64 - returns the user-visible inode number
72 * @fileid: 64-bit fileid
74 * This function returns a 32-bit inode number if the boot parameter
75 * nfs.enable_ino64 is zero.
77 u64 nfs_compat_user_ino64(u64 fileid)
84 if (sizeof(ino) < sizeof(fileid))
85 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
89 int nfs_write_inode(struct inode *inode, int sync)
94 ret = filemap_fdatawait(inode->i_mapping);
96 ret = nfs_commit_inode(inode, FLUSH_SYNC);
98 ret = nfs_commit_inode(inode, 0);
101 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
105 void nfs_clear_inode(struct inode *inode)
108 * The following should never happen...
110 BUG_ON(nfs_have_writebacks(inode));
111 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
112 nfs_zap_acl_cache(inode);
113 nfs_access_zap_cache(inode);
117 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
119 int nfs_sync_mapping(struct address_space *mapping)
123 if (mapping->nrpages == 0)
125 unmap_mapping_range(mapping, 0, 0, 0);
126 ret = filemap_write_and_wait(mapping);
129 ret = nfs_wb_all(mapping->host);
135 * Invalidate the local caches
137 static void nfs_zap_caches_locked(struct inode *inode)
139 struct nfs_inode *nfsi = NFS_I(inode);
140 int mode = inode->i_mode;
142 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
144 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
145 nfsi->attrtimeo_timestamp = jiffies;
147 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
148 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
149 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
151 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
154 void nfs_zap_caches(struct inode *inode)
156 spin_lock(&inode->i_lock);
157 nfs_zap_caches_locked(inode);
158 spin_unlock(&inode->i_lock);
161 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
163 if (mapping->nrpages != 0) {
164 spin_lock(&inode->i_lock);
165 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
166 spin_unlock(&inode->i_lock);
170 static void nfs_zap_acl_cache(struct inode *inode)
172 void (*clear_acl_cache)(struct inode *);
174 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
175 if (clear_acl_cache != NULL)
176 clear_acl_cache(inode);
177 spin_lock(&inode->i_lock);
178 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
179 spin_unlock(&inode->i_lock);
182 void nfs_invalidate_atime(struct inode *inode)
184 spin_lock(&inode->i_lock);
185 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
186 spin_unlock(&inode->i_lock);
190 * Invalidate, but do not unhash, the inode.
191 * NB: must be called with inode->i_lock held!
193 static void nfs_invalidate_inode(struct inode *inode)
195 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
196 nfs_zap_caches_locked(inode);
199 struct nfs_find_desc {
201 struct nfs_fattr *fattr;
205 * In NFSv3 we can have 64bit inode numbers. In order to support
206 * this, and re-exported directories (also seen in NFSv2)
207 * we are forced to allow 2 different inodes to have the same
211 nfs_find_actor(struct inode *inode, void *opaque)
213 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
214 struct nfs_fh *fh = desc->fh;
215 struct nfs_fattr *fattr = desc->fattr;
217 if (NFS_FILEID(inode) != fattr->fileid)
219 if (nfs_compare_fh(NFS_FH(inode), fh))
221 if (is_bad_inode(inode) || NFS_STALE(inode))
227 nfs_init_locked(struct inode *inode, void *opaque)
229 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
230 struct nfs_fattr *fattr = desc->fattr;
232 NFS_FILEID(inode) = fattr->fileid;
233 nfs_copy_fh(NFS_FH(inode), desc->fh);
237 /* Don't use READDIRPLUS on directories that we believe are too large */
238 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
241 * This is our front-end to iget that looks up inodes by file handle
242 * instead of inode number.
245 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
247 struct nfs_find_desc desc = {
251 struct inode *inode = ERR_PTR(-ENOENT);
254 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
258 printk("NFS: Buggy server - nlink == 0!\n");
262 hash = nfs_fattr_to_ino_t(fattr);
264 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
266 inode = ERR_PTR(-ENOMEM);
270 if (inode->i_state & I_NEW) {
271 struct nfs_inode *nfsi = NFS_I(inode);
272 unsigned long now = jiffies;
274 /* We set i_ino for the few things that still rely on it,
278 /* We can't support update_atime(), since the server will reset it */
279 inode->i_flags |= S_NOATIME|S_NOCMTIME;
280 inode->i_mode = fattr->mode;
281 /* Why so? Because we want revalidate for devices/FIFOs, and
282 * that's precisely what we have in nfs_file_inode_operations.
284 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
285 if (S_ISREG(inode->i_mode)) {
286 inode->i_fop = &nfs_file_operations;
287 inode->i_data.a_ops = &nfs_file_aops;
288 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
289 } else if (S_ISDIR(inode->i_mode)) {
290 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
291 inode->i_fop = &nfs_dir_operations;
292 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
293 && fattr->size <= NFS_LIMIT_READDIRPLUS)
294 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
295 /* Deal with crossing mountpoints */
296 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
297 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
298 inode->i_op = &nfs_referral_inode_operations;
300 inode->i_op = &nfs_mountpoint_inode_operations;
303 } else if (S_ISLNK(inode->i_mode))
304 inode->i_op = &nfs_symlink_inode_operations;
306 init_special_inode(inode, inode->i_mode, fattr->rdev);
308 nfsi->read_cache_jiffies = fattr->time_start;
309 nfsi->last_updated = now;
310 nfsi->cache_change_attribute = now;
311 inode->i_atime = fattr->atime;
312 inode->i_mtime = fattr->mtime;
313 inode->i_ctime = fattr->ctime;
314 if (fattr->valid & NFS_ATTR_FATTR_V4)
315 nfsi->change_attr = fattr->change_attr;
316 inode->i_size = nfs_size_to_loff_t(fattr->size);
317 inode->i_nlink = fattr->nlink;
318 inode->i_uid = fattr->uid;
319 inode->i_gid = fattr->gid;
320 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
322 * report the blocks in 512byte units
324 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
326 inode->i_blocks = fattr->du.nfs2.blocks;
328 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
329 nfsi->attrtimeo_timestamp = now;
330 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
331 nfsi->access_cache = RB_ROOT;
333 unlock_new_inode(inode);
335 nfs_refresh_inode(inode, fattr);
336 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
338 (long long)NFS_FILEID(inode),
339 atomic_read(&inode->i_count));
345 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
349 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
352 nfs_setattr(struct dentry *dentry, struct iattr *attr)
354 struct inode *inode = dentry->d_inode;
355 struct nfs_fattr fattr;
358 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
360 /* skip mode change if it's just for clearing setuid/setgid */
361 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
362 attr->ia_valid &= ~ATTR_MODE;
364 if (attr->ia_valid & ATTR_SIZE) {
365 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
366 attr->ia_valid &= ~ATTR_SIZE;
369 /* Optimization: if the end result is no change, don't RPC */
370 attr->ia_valid &= NFS_VALID_ATTRS;
371 if (attr->ia_valid == 0)
375 /* Write all dirty data */
376 if (S_ISREG(inode->i_mode)) {
377 filemap_write_and_wait(inode->i_mapping);
381 * Return any delegations if we're going to change ACLs
383 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
384 nfs_inode_return_delegation(inode);
385 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
387 nfs_refresh_inode(inode, &fattr);
393 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
394 * @inode: pointer to struct inode
395 * @attr: pointer to struct iattr
397 * Note: we do this in the *proc.c in order to ensure that
398 * it works for things like exclusive creates too.
400 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
402 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
403 if ((attr->ia_valid & ATTR_MODE) != 0) {
404 int mode = attr->ia_mode & S_IALLUGO;
405 mode |= inode->i_mode & ~S_IALLUGO;
406 inode->i_mode = mode;
408 if ((attr->ia_valid & ATTR_UID) != 0)
409 inode->i_uid = attr->ia_uid;
410 if ((attr->ia_valid & ATTR_GID) != 0)
411 inode->i_gid = attr->ia_gid;
412 spin_lock(&inode->i_lock);
413 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
414 spin_unlock(&inode->i_lock);
416 if ((attr->ia_valid & ATTR_SIZE) != 0) {
417 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
418 inode->i_size = attr->ia_size;
419 vmtruncate(inode, attr->ia_size);
423 static int nfs_wait_schedule(void *word)
425 if (signal_pending(current))
432 * Wait for the inode to get unlocked.
434 static int nfs_wait_on_inode(struct inode *inode)
436 struct rpc_clnt *clnt = NFS_CLIENT(inode);
437 struct nfs_inode *nfsi = NFS_I(inode);
441 rpc_clnt_sigmask(clnt, &oldmask);
442 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
443 nfs_wait_schedule, TASK_INTERRUPTIBLE);
444 rpc_clnt_sigunmask(clnt, &oldmask);
449 static void nfs_wake_up_inode(struct inode *inode)
451 struct nfs_inode *nfsi = NFS_I(inode);
453 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
454 smp_mb__after_clear_bit();
455 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
458 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
460 struct inode *inode = dentry->d_inode;
461 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
464 /* Flush out writes to the server in order to update c/mtime */
465 if (S_ISREG(inode->i_mode))
466 nfs_wb_nocommit(inode);
469 * We may force a getattr if the user cares about atime.
471 * Note that we only have to check the vfsmount flags here:
472 * - NFS always sets S_NOATIME by so checking it would give a
474 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
475 * no point in checking those.
477 if ((mnt->mnt_flags & MNT_NOATIME) ||
478 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
482 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
484 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
486 generic_fillattr(inode, stat);
487 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
492 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
494 struct nfs_open_context *ctx;
496 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
498 ctx->path.dentry = dget(dentry);
499 ctx->path.mnt = mntget(mnt);
500 ctx->cred = get_rpccred(cred);
502 ctx->lockowner = current->files;
505 atomic_set(&ctx->count, 1);
510 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
513 atomic_inc(&ctx->count);
517 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
519 struct inode *inode = ctx->path.dentry->d_inode;
521 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
523 list_del(&ctx->list);
524 spin_unlock(&inode->i_lock);
525 if (ctx->state != NULL) {
527 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
529 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
531 if (ctx->cred != NULL)
532 put_rpccred(ctx->cred);
533 dput(ctx->path.dentry);
534 mntput(ctx->path.mnt);
538 void put_nfs_open_context(struct nfs_open_context *ctx)
540 __put_nfs_open_context(ctx, 0);
543 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
545 __put_nfs_open_context(ctx, 1);
549 * Ensure that mmap has a recent RPC credential for use when writing out
552 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
554 struct inode *inode = filp->f_path.dentry->d_inode;
555 struct nfs_inode *nfsi = NFS_I(inode);
557 filp->private_data = get_nfs_open_context(ctx);
558 spin_lock(&inode->i_lock);
559 list_add(&ctx->list, &nfsi->open_files);
560 spin_unlock(&inode->i_lock);
564 * Given an inode, search for an open context with the desired characteristics
566 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
568 struct nfs_inode *nfsi = NFS_I(inode);
569 struct nfs_open_context *pos, *ctx = NULL;
571 spin_lock(&inode->i_lock);
572 list_for_each_entry(pos, &nfsi->open_files, list) {
573 if (cred != NULL && pos->cred != cred)
575 if ((pos->mode & mode) == mode) {
576 ctx = get_nfs_open_context(pos);
580 spin_unlock(&inode->i_lock);
584 static void nfs_file_clear_open_context(struct file *filp)
586 struct inode *inode = filp->f_path.dentry->d_inode;
587 struct nfs_open_context *ctx = nfs_file_open_context(filp);
590 filp->private_data = NULL;
591 spin_lock(&inode->i_lock);
592 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
593 spin_unlock(&inode->i_lock);
594 put_nfs_open_context_sync(ctx);
599 * These allocate and release file read/write context information.
601 int nfs_open(struct inode *inode, struct file *filp)
603 struct nfs_open_context *ctx;
604 struct rpc_cred *cred;
606 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
608 return PTR_ERR(cred);
609 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
613 ctx->mode = filp->f_mode;
614 nfs_file_set_open_context(filp, ctx);
615 put_nfs_open_context(ctx);
619 int nfs_release(struct inode *inode, struct file *filp)
621 nfs_file_clear_open_context(filp);
626 * This function is called whenever some part of NFS notices that
627 * the cached attributes have to be refreshed.
630 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
632 int status = -ESTALE;
633 struct nfs_fattr fattr;
634 struct nfs_inode *nfsi = NFS_I(inode);
636 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
637 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
639 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
641 if (is_bad_inode(inode))
643 if (NFS_STALE(inode))
646 status = nfs_wait_on_inode(inode);
651 if (NFS_STALE(inode))
654 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
656 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
658 (long long)NFS_FILEID(inode), status);
659 if (status == -ESTALE) {
660 nfs_zap_caches(inode);
661 if (!S_ISDIR(inode->i_mode))
662 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
667 spin_lock(&inode->i_lock);
668 status = nfs_update_inode(inode, &fattr);
670 spin_unlock(&inode->i_lock);
671 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
673 (long long)NFS_FILEID(inode), status);
676 spin_unlock(&inode->i_lock);
678 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
679 nfs_zap_acl_cache(inode);
681 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
683 (long long)NFS_FILEID(inode));
686 nfs_wake_up_inode(inode);
693 int nfs_attribute_timeout(struct inode *inode)
695 struct nfs_inode *nfsi = NFS_I(inode);
697 if (nfs_have_delegation(inode, FMODE_READ))
699 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
703 * nfs_revalidate_inode - Revalidate the inode attributes
704 * @server - pointer to nfs_server struct
705 * @inode - pointer to inode struct
707 * Updates inode attribute information by retrieving the data from the server.
709 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
711 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
712 && !nfs_attribute_timeout(inode))
713 return NFS_STALE(inode) ? -ESTALE : 0;
714 return __nfs_revalidate_inode(server, inode);
717 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
719 struct nfs_inode *nfsi = NFS_I(inode);
721 if (mapping->nrpages != 0) {
722 int ret = invalidate_inode_pages2(mapping);
726 spin_lock(&inode->i_lock);
727 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
728 if (S_ISDIR(inode->i_mode))
729 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
730 spin_unlock(&inode->i_lock);
731 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
732 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
733 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
737 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
741 mutex_lock(&inode->i_mutex);
742 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
743 ret = nfs_sync_mapping(mapping);
745 ret = nfs_invalidate_mapping_nolock(inode, mapping);
747 mutex_unlock(&inode->i_mutex);
752 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
753 * @inode - pointer to host inode
754 * @mapping - pointer to mapping
756 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
758 struct nfs_inode *nfsi = NFS_I(inode);
761 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
762 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
763 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
767 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
768 ret = nfs_invalidate_mapping_nolock(inode, mapping);
774 * nfs_revalidate_mapping - Revalidate the pagecache
775 * @inode - pointer to host inode
776 * @mapping - pointer to mapping
778 * This version of the function will take the inode->i_mutex and attempt to
779 * flush out all dirty data if it needs to invalidate the page cache.
781 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
783 struct nfs_inode *nfsi = NFS_I(inode);
786 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
787 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
788 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
792 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
793 ret = nfs_invalidate_mapping(inode, mapping);
798 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
800 struct nfs_inode *nfsi = NFS_I(inode);
802 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
803 nfsi->change_attr == fattr->pre_change_attr) {
804 nfsi->change_attr = fattr->change_attr;
805 if (S_ISDIR(inode->i_mode))
806 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
808 /* If we have atomic WCC data, we may update some attributes */
809 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
810 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
811 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
812 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
813 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
814 if (S_ISDIR(inode->i_mode))
815 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
817 if (inode->i_size == nfs_size_to_loff_t(fattr->pre_size) &&
819 inode->i_size = nfs_size_to_loff_t(fattr->size);
824 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
825 * @inode - pointer to inode
826 * @fattr - updated attributes
828 * Verifies the attribute cache. If we have just changed the attributes,
829 * so that fattr carries weak cache consistency data, then it may
830 * also update the ctime/mtime/change_attribute.
832 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
834 struct nfs_inode *nfsi = NFS_I(inode);
835 loff_t cur_size, new_isize;
836 unsigned long invalid = 0;
839 /* Has the inode gone and changed behind our back? */
840 if (nfsi->fileid != fattr->fileid
841 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
845 /* Do atomic weak cache consistency updates */
846 nfs_wcc_update_inode(inode, fattr);
848 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
849 nfsi->change_attr != fattr->change_attr)
850 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
852 /* Verify a few of the more important attributes */
853 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
854 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
856 cur_size = i_size_read(inode);
857 new_isize = nfs_size_to_loff_t(fattr->size);
858 if (cur_size != new_isize && nfsi->npages == 0)
859 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
861 /* Have any file permissions changed? */
862 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
863 || inode->i_uid != fattr->uid
864 || inode->i_gid != fattr->gid)
865 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
867 /* Has the link count changed? */
868 if (inode->i_nlink != fattr->nlink)
869 invalid |= NFS_INO_INVALID_ATTR;
871 if (!timespec_equal(&inode->i_atime, &fattr->atime))
872 invalid |= NFS_INO_INVALID_ATIME;
875 nfsi->cache_validity |= invalid;
877 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
878 | NFS_INO_INVALID_ATIME
879 | NFS_INO_REVAL_PAGECACHE);
881 nfsi->read_cache_jiffies = fattr->time_start;
886 * nfs_refresh_inode - try to update the inode attribute cache
887 * @inode - pointer to inode
888 * @fattr - updated attributes
890 * Check that an RPC call that returned attributes has not overlapped with
891 * other recent updates of the inode metadata, then decide whether it is
892 * safe to do a full update of the inode attributes, or whether just to
893 * call nfs_check_inode_attributes.
895 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
897 struct nfs_inode *nfsi = NFS_I(inode);
900 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
902 spin_lock(&inode->i_lock);
903 if (time_after(fattr->time_start, nfsi->last_updated))
904 status = nfs_update_inode(inode, fattr);
906 status = nfs_check_inode_attributes(inode, fattr);
908 spin_unlock(&inode->i_lock);
913 * nfs_post_op_update_inode - try to update the inode attribute cache
914 * @inode - pointer to inode
915 * @fattr - updated attributes
917 * After an operation that has changed the inode metadata, mark the
918 * attribute cache as being invalid, then try to update it.
920 * NB: if the server didn't return any post op attributes, this
921 * function will force the retrieval of attributes before the next
922 * NFS request. Thus it should be used only for operations that
923 * are expected to change one or more attributes, to avoid
924 * unnecessary NFS requests and trips through nfs_update_inode().
926 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
928 struct nfs_inode *nfsi = NFS_I(inode);
930 spin_lock(&inode->i_lock);
931 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
932 if (S_ISDIR(inode->i_mode))
933 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
934 spin_unlock(&inode->i_lock);
935 return nfs_refresh_inode(inode, fattr);
939 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
940 * @inode - pointer to inode
941 * @fattr - updated attributes
943 * After an operation that has changed the inode metadata, mark the
944 * attribute cache as being invalid, then try to update it. Fake up
945 * weak cache consistency data, if none exist.
947 * This function is mainly designed to be used by the ->write_done() functions.
949 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
951 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
952 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
953 fattr->pre_change_attr = NFS_I(inode)->change_attr;
954 fattr->valid |= NFS_ATTR_WCC_V4;
956 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
957 (fattr->valid & NFS_ATTR_WCC) == 0) {
958 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
959 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
960 fattr->pre_size = inode->i_size;
961 fattr->valid |= NFS_ATTR_WCC;
963 return nfs_post_op_update_inode(inode, fattr);
967 * Many nfs protocol calls return the new file attributes after
968 * an operation. Here we update the inode to reflect the state
969 * of the server's inode.
971 * This is a bit tricky because we have to make sure all dirty pages
972 * have been sent off to the server before calling invalidate_inode_pages.
973 * To make sure no other process adds more write requests while we try
974 * our best to flush them, we make them sleep during the attribute refresh.
976 * A very similar scenario holds for the dir cache.
978 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
980 struct nfs_server *server;
981 struct nfs_inode *nfsi = NFS_I(inode);
982 loff_t cur_isize, new_isize;
983 unsigned long invalid = 0;
984 unsigned long now = jiffies;
986 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
987 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
988 atomic_read(&inode->i_count), fattr->valid);
990 if (nfsi->fileid != fattr->fileid)
994 * Make sure the inode's type hasn't changed.
996 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
999 server = NFS_SERVER(inode);
1000 /* Update the fsid? */
1001 if (S_ISDIR(inode->i_mode)
1002 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
1003 server->fsid = fattr->fsid;
1006 * Update the read time so we don't revalidate too often.
1008 nfsi->read_cache_jiffies = fattr->time_start;
1010 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1011 | NFS_INO_REVAL_PAGECACHE);
1013 /* Do atomic weak cache consistency updates */
1014 nfs_wcc_update_inode(inode, fattr);
1016 /* More cache consistency checks */
1017 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1018 /* NFSv2/v3: Check if the mtime agrees */
1019 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1020 dprintk("NFS: mtime change on server for file %s/%ld\n",
1021 inode->i_sb->s_id, inode->i_ino);
1022 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1023 nfsi->cache_change_attribute = now;
1025 /* If ctime has changed we should definitely clear access+acl caches */
1026 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1027 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1028 } else if (nfsi->change_attr != fattr->change_attr) {
1029 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1030 inode->i_sb->s_id, inode->i_ino);
1031 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1032 nfsi->cache_change_attribute = now;
1035 /* Check if our cached file size is stale */
1036 new_isize = nfs_size_to_loff_t(fattr->size);
1037 cur_isize = i_size_read(inode);
1038 if (new_isize != cur_isize) {
1039 /* Do we perhaps have any outstanding writes, or has
1040 * the file grown beyond our last write? */
1041 if (nfsi->npages == 0 || new_isize > cur_isize) {
1042 inode->i_size = new_isize;
1043 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1045 dprintk("NFS: isize change on server for file %s/%ld\n",
1046 inode->i_sb->s_id, inode->i_ino);
1050 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1051 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1052 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1053 nfsi->change_attr = fattr->change_attr;
1055 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1056 inode->i_uid != fattr->uid ||
1057 inode->i_gid != fattr->gid)
1058 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1060 inode->i_mode = fattr->mode;
1061 inode->i_nlink = fattr->nlink;
1062 inode->i_uid = fattr->uid;
1063 inode->i_gid = fattr->gid;
1065 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1067 * report the blocks in 512byte units
1069 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1071 inode->i_blocks = fattr->du.nfs2.blocks;
1074 /* Update attrtimeo value if we're out of the unstable period */
1075 if (invalid & NFS_INO_INVALID_ATTR) {
1076 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1077 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1078 nfsi->attrtimeo_timestamp = now;
1079 nfsi->last_updated = now;
1081 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1082 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1083 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1084 nfsi->attrtimeo_timestamp = now;
1087 * Avoid jiffy wraparound issues with nfsi->last_updated
1089 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1090 nfsi->last_updated = nfsi->read_cache_jiffies;
1092 invalid &= ~NFS_INO_INVALID_ATTR;
1093 /* Don't invalidate the data if we were to blame */
1094 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1095 || S_ISLNK(inode->i_mode)))
1096 invalid &= ~NFS_INO_INVALID_DATA;
1097 if (!nfs_have_delegation(inode, FMODE_READ) ||
1098 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1099 nfsi->cache_validity |= invalid;
1100 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1105 * Big trouble! The inode has become a different object.
1107 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1108 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1111 * No need to worry about unhashing the dentry, as the
1112 * lookup validation will know that the inode is bad.
1113 * (But we fall through to invalidate the caches.)
1115 nfs_invalidate_inode(inode);
1119 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1120 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1121 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1122 (long long)nfsi->fileid, (long long)fattr->fileid);
1127 #ifdef CONFIG_NFS_V4
1130 * Clean out any remaining NFSv4 state that might be left over due
1131 * to open() calls that passed nfs_atomic_lookup, but failed to call
1134 void nfs4_clear_inode(struct inode *inode)
1136 /* If we are holding a delegation, return it! */
1137 nfs_inode_return_delegation(inode);
1138 /* First call standard NFS clear_inode() code */
1139 nfs_clear_inode(inode);
1143 struct inode *nfs_alloc_inode(struct super_block *sb)
1145 struct nfs_inode *nfsi;
1146 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1150 nfsi->cache_validity = 0UL;
1151 #ifdef CONFIG_NFS_V3_ACL
1152 nfsi->acl_access = ERR_PTR(-EAGAIN);
1153 nfsi->acl_default = ERR_PTR(-EAGAIN);
1155 #ifdef CONFIG_NFS_V4
1156 nfsi->nfs4_acl = NULL;
1157 #endif /* CONFIG_NFS_V4 */
1158 return &nfsi->vfs_inode;
1161 void nfs_destroy_inode(struct inode *inode)
1163 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1166 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1168 #ifdef CONFIG_NFS_V4
1169 INIT_LIST_HEAD(&nfsi->open_states);
1170 nfsi->delegation = NULL;
1171 nfsi->delegation_state = 0;
1172 init_rwsem(&nfsi->rwsem);
1176 static void init_once(struct kmem_cache * cachep, void *foo)
1178 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1180 inode_init_once(&nfsi->vfs_inode);
1181 INIT_LIST_HEAD(&nfsi->open_files);
1182 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1183 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1184 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1187 atomic_set(&nfsi->silly_count, 1);
1188 INIT_HLIST_HEAD(&nfsi->silly_list);
1189 init_waitqueue_head(&nfsi->waitqueue);
1190 nfs4_init_once(nfsi);
1193 static int __init nfs_init_inodecache(void)
1195 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1196 sizeof(struct nfs_inode),
1197 0, (SLAB_RECLAIM_ACCOUNT|
1200 if (nfs_inode_cachep == NULL)
1206 static void nfs_destroy_inodecache(void)
1208 kmem_cache_destroy(nfs_inode_cachep);
1214 static int __init init_nfs_fs(void)
1218 err = nfs_fs_proc_init();
1222 err = nfs_init_nfspagecache();
1226 err = nfs_init_inodecache();
1230 err = nfs_init_readpagecache();
1234 err = nfs_init_writepagecache();
1238 err = nfs_init_directcache();
1242 #ifdef CONFIG_PROC_FS
1243 rpc_proc_register(&nfs_rpcstat);
1245 if ((err = register_nfs_fs()) != 0)
1249 #ifdef CONFIG_PROC_FS
1250 rpc_proc_unregister("nfs");
1252 nfs_destroy_directcache();
1254 nfs_destroy_writepagecache();
1256 nfs_destroy_readpagecache();
1258 nfs_destroy_inodecache();
1260 nfs_destroy_nfspagecache();
1267 static void __exit exit_nfs_fs(void)
1269 nfs_destroy_directcache();
1270 nfs_destroy_writepagecache();
1271 nfs_destroy_readpagecache();
1272 nfs_destroy_inodecache();
1273 nfs_destroy_nfspagecache();
1274 #ifdef CONFIG_PROC_FS
1275 rpc_proc_unregister("nfs");
1277 unregister_nfs_fs();
1281 /* Not quite true; I just maintain it */
1282 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1283 MODULE_LICENSE("GPL");
1284 module_param(enable_ino64, bool, 0644);
1286 module_init(init_nfs_fs)
1287 module_exit(exit_nfs_fs)