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 if (attr->ia_valid & ATTR_SIZE) {
361 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
362 attr->ia_valid &= ~ATTR_SIZE;
365 /* Optimization: if the end result is no change, don't RPC */
366 attr->ia_valid &= NFS_VALID_ATTRS;
367 if (attr->ia_valid == 0)
371 /* Write all dirty data */
372 if (S_ISREG(inode->i_mode)) {
373 filemap_write_and_wait(inode->i_mapping);
377 * Return any delegations if we're going to change ACLs
379 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
380 nfs_inode_return_delegation(inode);
381 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
383 nfs_refresh_inode(inode, &fattr);
389 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
390 * @inode: pointer to struct inode
391 * @attr: pointer to struct iattr
393 * Note: we do this in the *proc.c in order to ensure that
394 * it works for things like exclusive creates too.
396 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
398 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
399 if ((attr->ia_valid & ATTR_MODE) != 0) {
400 int mode = attr->ia_mode & S_IALLUGO;
401 mode |= inode->i_mode & ~S_IALLUGO;
402 inode->i_mode = mode;
404 if ((attr->ia_valid & ATTR_UID) != 0)
405 inode->i_uid = attr->ia_uid;
406 if ((attr->ia_valid & ATTR_GID) != 0)
407 inode->i_gid = attr->ia_gid;
408 spin_lock(&inode->i_lock);
409 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
410 spin_unlock(&inode->i_lock);
412 if ((attr->ia_valid & ATTR_SIZE) != 0) {
413 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
414 inode->i_size = attr->ia_size;
415 vmtruncate(inode, attr->ia_size);
419 static int nfs_wait_schedule(void *word)
421 if (signal_pending(current))
428 * Wait for the inode to get unlocked.
430 static int nfs_wait_on_inode(struct inode *inode)
432 struct rpc_clnt *clnt = NFS_CLIENT(inode);
433 struct nfs_inode *nfsi = NFS_I(inode);
437 rpc_clnt_sigmask(clnt, &oldmask);
438 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
439 nfs_wait_schedule, TASK_INTERRUPTIBLE);
440 rpc_clnt_sigunmask(clnt, &oldmask);
445 static void nfs_wake_up_inode(struct inode *inode)
447 struct nfs_inode *nfsi = NFS_I(inode);
449 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
450 smp_mb__after_clear_bit();
451 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
454 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
456 struct inode *inode = dentry->d_inode;
457 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
460 /* Flush out writes to the server in order to update c/mtime */
461 if (S_ISREG(inode->i_mode))
462 nfs_wb_nocommit(inode);
465 * We may force a getattr if the user cares about atime.
467 * Note that we only have to check the vfsmount flags here:
468 * - NFS always sets S_NOATIME by so checking it would give a
470 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
471 * no point in checking those.
473 if ((mnt->mnt_flags & MNT_NOATIME) ||
474 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
478 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
480 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
482 generic_fillattr(inode, stat);
483 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
488 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
490 struct nfs_open_context *ctx;
492 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
494 ctx->path.dentry = dget(dentry);
495 ctx->path.mnt = mntget(mnt);
496 ctx->cred = get_rpccred(cred);
498 ctx->lockowner = current->files;
501 atomic_set(&ctx->count, 1);
506 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
509 atomic_inc(&ctx->count);
513 void put_nfs_open_context(struct nfs_open_context *ctx)
515 struct inode *inode = ctx->path.dentry->d_inode;
517 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
519 list_del(&ctx->list);
520 spin_unlock(&inode->i_lock);
521 if (ctx->state != NULL)
522 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
523 if (ctx->cred != NULL)
524 put_rpccred(ctx->cred);
525 dput(ctx->path.dentry);
526 mntput(ctx->path.mnt);
531 * Ensure that mmap has a recent RPC credential for use when writing out
534 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
536 struct inode *inode = filp->f_path.dentry->d_inode;
537 struct nfs_inode *nfsi = NFS_I(inode);
539 filp->private_data = get_nfs_open_context(ctx);
540 spin_lock(&inode->i_lock);
541 list_add(&ctx->list, &nfsi->open_files);
542 spin_unlock(&inode->i_lock);
546 * Given an inode, search for an open context with the desired characteristics
548 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
550 struct nfs_inode *nfsi = NFS_I(inode);
551 struct nfs_open_context *pos, *ctx = NULL;
553 spin_lock(&inode->i_lock);
554 list_for_each_entry(pos, &nfsi->open_files, list) {
555 if (cred != NULL && pos->cred != cred)
557 if ((pos->mode & mode) == mode) {
558 ctx = get_nfs_open_context(pos);
562 spin_unlock(&inode->i_lock);
566 static void nfs_file_clear_open_context(struct file *filp)
568 struct inode *inode = filp->f_path.dentry->d_inode;
569 struct nfs_open_context *ctx = nfs_file_open_context(filp);
572 filp->private_data = NULL;
573 spin_lock(&inode->i_lock);
574 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
575 spin_unlock(&inode->i_lock);
576 put_nfs_open_context(ctx);
581 * These allocate and release file read/write context information.
583 int nfs_open(struct inode *inode, struct file *filp)
585 struct nfs_open_context *ctx;
586 struct rpc_cred *cred;
588 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
590 return PTR_ERR(cred);
591 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
595 ctx->mode = filp->f_mode;
596 nfs_file_set_open_context(filp, ctx);
597 put_nfs_open_context(ctx);
601 int nfs_release(struct inode *inode, struct file *filp)
603 nfs_file_clear_open_context(filp);
608 * This function is called whenever some part of NFS notices that
609 * the cached attributes have to be refreshed.
612 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
614 int status = -ESTALE;
615 struct nfs_fattr fattr;
616 struct nfs_inode *nfsi = NFS_I(inode);
618 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
619 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
621 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
623 if (is_bad_inode(inode))
625 if (NFS_STALE(inode))
628 status = nfs_wait_on_inode(inode);
633 if (NFS_STALE(inode))
636 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
638 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
640 (long long)NFS_FILEID(inode), status);
641 if (status == -ESTALE) {
642 nfs_zap_caches(inode);
643 if (!S_ISDIR(inode->i_mode))
644 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
649 spin_lock(&inode->i_lock);
650 status = nfs_update_inode(inode, &fattr);
652 spin_unlock(&inode->i_lock);
653 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
655 (long long)NFS_FILEID(inode), status);
658 spin_unlock(&inode->i_lock);
660 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
661 nfs_zap_acl_cache(inode);
663 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
665 (long long)NFS_FILEID(inode));
668 nfs_wake_up_inode(inode);
675 int nfs_attribute_timeout(struct inode *inode)
677 struct nfs_inode *nfsi = NFS_I(inode);
679 if (nfs_have_delegation(inode, FMODE_READ))
681 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
685 * nfs_revalidate_inode - Revalidate the inode attributes
686 * @server - pointer to nfs_server struct
687 * @inode - pointer to inode struct
689 * Updates inode attribute information by retrieving the data from the server.
691 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
693 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
694 && !nfs_attribute_timeout(inode))
695 return NFS_STALE(inode) ? -ESTALE : 0;
696 return __nfs_revalidate_inode(server, inode);
699 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
701 struct nfs_inode *nfsi = NFS_I(inode);
703 if (mapping->nrpages != 0) {
704 int ret = invalidate_inode_pages2(mapping);
708 spin_lock(&inode->i_lock);
709 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
710 if (S_ISDIR(inode->i_mode))
711 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
712 spin_unlock(&inode->i_lock);
713 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
714 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
715 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
719 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
723 mutex_lock(&inode->i_mutex);
724 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
725 ret = nfs_sync_mapping(mapping);
727 ret = nfs_invalidate_mapping_nolock(inode, mapping);
729 mutex_unlock(&inode->i_mutex);
734 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
735 * @inode - pointer to host inode
736 * @mapping - pointer to mapping
738 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
740 struct nfs_inode *nfsi = NFS_I(inode);
743 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
744 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
745 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
749 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
750 ret = nfs_invalidate_mapping_nolock(inode, mapping);
756 * nfs_revalidate_mapping - Revalidate the pagecache
757 * @inode - pointer to host inode
758 * @mapping - pointer to mapping
760 * This version of the function will take the inode->i_mutex and attempt to
761 * flush out all dirty data if it needs to invalidate the page cache.
763 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
765 struct nfs_inode *nfsi = NFS_I(inode);
768 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
769 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
770 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
774 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
775 ret = nfs_invalidate_mapping(inode, mapping);
780 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
782 struct nfs_inode *nfsi = NFS_I(inode);
784 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
785 nfsi->change_attr == fattr->pre_change_attr) {
786 nfsi->change_attr = fattr->change_attr;
787 if (S_ISDIR(inode->i_mode))
788 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
790 /* If we have atomic WCC data, we may update some attributes */
791 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
792 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
793 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
794 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
795 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
796 if (S_ISDIR(inode->i_mode))
797 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
799 if (inode->i_size == fattr->pre_size && nfsi->npages == 0)
800 inode->i_size = fattr->size;
805 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
806 * @inode - pointer to inode
807 * @fattr - updated attributes
809 * Verifies the attribute cache. If we have just changed the attributes,
810 * so that fattr carries weak cache consistency data, then it may
811 * also update the ctime/mtime/change_attribute.
813 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
815 struct nfs_inode *nfsi = NFS_I(inode);
816 loff_t cur_size, new_isize;
817 unsigned long invalid = 0;
820 /* Has the inode gone and changed behind our back? */
821 if (nfsi->fileid != fattr->fileid
822 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
826 /* Do atomic weak cache consistency updates */
827 nfs_wcc_update_inode(inode, fattr);
829 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
830 nfsi->change_attr != fattr->change_attr)
831 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
833 /* Verify a few of the more important attributes */
834 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
835 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
837 cur_size = i_size_read(inode);
838 new_isize = nfs_size_to_loff_t(fattr->size);
839 if (cur_size != new_isize && nfsi->npages == 0)
840 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
842 /* Have any file permissions changed? */
843 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
844 || inode->i_uid != fattr->uid
845 || inode->i_gid != fattr->gid)
846 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
848 /* Has the link count changed? */
849 if (inode->i_nlink != fattr->nlink)
850 invalid |= NFS_INO_INVALID_ATTR;
852 if (!timespec_equal(&inode->i_atime, &fattr->atime))
853 invalid |= NFS_INO_INVALID_ATIME;
856 nfsi->cache_validity |= invalid;
858 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
859 | NFS_INO_INVALID_ATIME
860 | NFS_INO_REVAL_PAGECACHE);
862 nfsi->read_cache_jiffies = fattr->time_start;
867 * nfs_refresh_inode - try to update the inode attribute cache
868 * @inode - pointer to inode
869 * @fattr - updated attributes
871 * Check that an RPC call that returned attributes has not overlapped with
872 * other recent updates of the inode metadata, then decide whether it is
873 * safe to do a full update of the inode attributes, or whether just to
874 * call nfs_check_inode_attributes.
876 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
878 struct nfs_inode *nfsi = NFS_I(inode);
881 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
883 spin_lock(&inode->i_lock);
884 if (time_after(fattr->time_start, nfsi->last_updated))
885 status = nfs_update_inode(inode, fattr);
887 status = nfs_check_inode_attributes(inode, fattr);
889 spin_unlock(&inode->i_lock);
894 * nfs_post_op_update_inode - try to update the inode attribute cache
895 * @inode - pointer to inode
896 * @fattr - updated attributes
898 * After an operation that has changed the inode metadata, mark the
899 * attribute cache as being invalid, then try to update it.
901 * NB: if the server didn't return any post op attributes, this
902 * function will force the retrieval of attributes before the next
903 * NFS request. Thus it should be used only for operations that
904 * are expected to change one or more attributes, to avoid
905 * unnecessary NFS requests and trips through nfs_update_inode().
907 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
909 struct nfs_inode *nfsi = NFS_I(inode);
911 spin_lock(&inode->i_lock);
912 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
913 if (S_ISDIR(inode->i_mode))
914 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
915 spin_unlock(&inode->i_lock);
916 return nfs_refresh_inode(inode, fattr);
920 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
921 * @inode - pointer to inode
922 * @fattr - updated attributes
924 * After an operation that has changed the inode metadata, mark the
925 * attribute cache as being invalid, then try to update it. Fake up
926 * weak cache consistency data, if none exist.
928 * This function is mainly designed to be used by the ->write_done() functions.
930 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
932 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
933 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
934 fattr->pre_change_attr = NFS_I(inode)->change_attr;
935 fattr->valid |= NFS_ATTR_WCC_V4;
937 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
938 (fattr->valid & NFS_ATTR_WCC) == 0) {
939 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
940 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
941 fattr->pre_size = inode->i_size;
942 fattr->valid |= NFS_ATTR_WCC;
944 return nfs_post_op_update_inode(inode, fattr);
948 * Many nfs protocol calls return the new file attributes after
949 * an operation. Here we update the inode to reflect the state
950 * of the server's inode.
952 * This is a bit tricky because we have to make sure all dirty pages
953 * have been sent off to the server before calling invalidate_inode_pages.
954 * To make sure no other process adds more write requests while we try
955 * our best to flush them, we make them sleep during the attribute refresh.
957 * A very similar scenario holds for the dir cache.
959 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
961 struct nfs_server *server;
962 struct nfs_inode *nfsi = NFS_I(inode);
963 loff_t cur_isize, new_isize;
964 unsigned long invalid = 0;
965 unsigned long now = jiffies;
967 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
968 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
969 atomic_read(&inode->i_count), fattr->valid);
971 if (nfsi->fileid != fattr->fileid)
975 * Make sure the inode's type hasn't changed.
977 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
980 server = NFS_SERVER(inode);
981 /* Update the fsid? */
982 if (S_ISDIR(inode->i_mode)
983 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
984 server->fsid = fattr->fsid;
987 * Update the read time so we don't revalidate too often.
989 nfsi->read_cache_jiffies = fattr->time_start;
991 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
992 | NFS_INO_REVAL_PAGECACHE);
994 /* Do atomic weak cache consistency updates */
995 nfs_wcc_update_inode(inode, fattr);
997 /* More cache consistency checks */
998 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
999 /* NFSv2/v3: Check if the mtime agrees */
1000 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1001 dprintk("NFS: mtime change on server for file %s/%ld\n",
1002 inode->i_sb->s_id, inode->i_ino);
1003 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1004 nfsi->cache_change_attribute = now;
1006 /* If ctime has changed we should definitely clear access+acl caches */
1007 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1008 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1009 } else if (nfsi->change_attr != fattr->change_attr) {
1010 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1011 inode->i_sb->s_id, inode->i_ino);
1012 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1013 nfsi->cache_change_attribute = now;
1016 /* Check if our cached file size is stale */
1017 new_isize = nfs_size_to_loff_t(fattr->size);
1018 cur_isize = i_size_read(inode);
1019 if (new_isize != cur_isize) {
1020 /* Do we perhaps have any outstanding writes, or has
1021 * the file grown beyond our last write? */
1022 if (nfsi->npages == 0 || new_isize > cur_isize) {
1023 inode->i_size = new_isize;
1024 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1026 dprintk("NFS: isize change on server for file %s/%ld\n",
1027 inode->i_sb->s_id, inode->i_ino);
1031 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1032 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1033 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1034 nfsi->change_attr = fattr->change_attr;
1036 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1037 inode->i_uid != fattr->uid ||
1038 inode->i_gid != fattr->gid)
1039 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1041 inode->i_mode = fattr->mode;
1042 inode->i_nlink = fattr->nlink;
1043 inode->i_uid = fattr->uid;
1044 inode->i_gid = fattr->gid;
1046 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1048 * report the blocks in 512byte units
1050 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1052 inode->i_blocks = fattr->du.nfs2.blocks;
1055 /* Update attrtimeo value if we're out of the unstable period */
1056 if (invalid & NFS_INO_INVALID_ATTR) {
1057 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1058 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1059 nfsi->attrtimeo_timestamp = now;
1060 nfsi->last_updated = now;
1062 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1063 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1064 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1065 nfsi->attrtimeo_timestamp = now;
1068 * Avoid jiffy wraparound issues with nfsi->last_updated
1070 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1071 nfsi->last_updated = nfsi->read_cache_jiffies;
1073 invalid &= ~NFS_INO_INVALID_ATTR;
1074 /* Don't invalidate the data if we were to blame */
1075 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1076 || S_ISLNK(inode->i_mode)))
1077 invalid &= ~NFS_INO_INVALID_DATA;
1078 if (!nfs_have_delegation(inode, FMODE_READ) ||
1079 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1080 nfsi->cache_validity |= invalid;
1081 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1086 * Big trouble! The inode has become a different object.
1088 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1089 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1092 * No need to worry about unhashing the dentry, as the
1093 * lookup validation will know that the inode is bad.
1094 * (But we fall through to invalidate the caches.)
1096 nfs_invalidate_inode(inode);
1100 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1101 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1102 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1103 (long long)nfsi->fileid, (long long)fattr->fileid);
1108 #ifdef CONFIG_NFS_V4
1111 * Clean out any remaining NFSv4 state that might be left over due
1112 * to open() calls that passed nfs_atomic_lookup, but failed to call
1115 void nfs4_clear_inode(struct inode *inode)
1117 /* If we are holding a delegation, return it! */
1118 nfs_inode_return_delegation(inode);
1119 /* First call standard NFS clear_inode() code */
1120 nfs_clear_inode(inode);
1124 struct inode *nfs_alloc_inode(struct super_block *sb)
1126 struct nfs_inode *nfsi;
1127 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1131 nfsi->cache_validity = 0UL;
1132 #ifdef CONFIG_NFS_V3_ACL
1133 nfsi->acl_access = ERR_PTR(-EAGAIN);
1134 nfsi->acl_default = ERR_PTR(-EAGAIN);
1136 #ifdef CONFIG_NFS_V4
1137 nfsi->nfs4_acl = NULL;
1138 #endif /* CONFIG_NFS_V4 */
1139 return &nfsi->vfs_inode;
1142 void nfs_destroy_inode(struct inode *inode)
1144 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1147 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1149 #ifdef CONFIG_NFS_V4
1150 INIT_LIST_HEAD(&nfsi->open_states);
1151 nfsi->delegation = NULL;
1152 nfsi->delegation_state = 0;
1153 init_rwsem(&nfsi->rwsem);
1157 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1159 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1161 inode_init_once(&nfsi->vfs_inode);
1162 INIT_LIST_HEAD(&nfsi->open_files);
1163 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1164 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1165 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1168 nfs4_init_once(nfsi);
1171 static int __init nfs_init_inodecache(void)
1173 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1174 sizeof(struct nfs_inode),
1175 0, (SLAB_RECLAIM_ACCOUNT|
1178 if (nfs_inode_cachep == NULL)
1184 static void nfs_destroy_inodecache(void)
1186 kmem_cache_destroy(nfs_inode_cachep);
1192 static int __init init_nfs_fs(void)
1196 err = nfs_fs_proc_init();
1200 err = nfs_init_nfspagecache();
1204 err = nfs_init_inodecache();
1208 err = nfs_init_readpagecache();
1212 err = nfs_init_writepagecache();
1216 err = nfs_init_directcache();
1220 #ifdef CONFIG_PROC_FS
1221 rpc_proc_register(&nfs_rpcstat);
1223 if ((err = register_nfs_fs()) != 0)
1227 #ifdef CONFIG_PROC_FS
1228 rpc_proc_unregister("nfs");
1230 nfs_destroy_directcache();
1232 nfs_destroy_writepagecache();
1234 nfs_destroy_readpagecache();
1236 nfs_destroy_inodecache();
1238 nfs_destroy_nfspagecache();
1245 static void __exit exit_nfs_fs(void)
1247 nfs_destroy_directcache();
1248 nfs_destroy_writepagecache();
1249 nfs_destroy_readpagecache();
1250 nfs_destroy_inodecache();
1251 nfs_destroy_nfspagecache();
1252 #ifdef CONFIG_PROC_FS
1253 rpc_proc_unregister("nfs");
1255 unregister_nfs_fs();
1259 /* Not quite true; I just maintain it */
1260 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1261 MODULE_LICENSE("GPL");
1262 module_param(enable_ino64, bool, 0644);
1264 module_init(init_nfs_fs)
1265 module_exit(exit_nfs_fs)