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>
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
51 #define NFS_PARANOIA 1
53 static void nfs_invalidate_inode(struct inode *);
54 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
56 static void nfs_zap_acl_cache(struct inode *);
58 static struct kmem_cache * nfs_inode_cachep;
60 static inline unsigned long
61 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
63 return nfs_fileid_to_ino_t(fattr->fileid);
66 int nfs_write_inode(struct inode *inode, int sync)
71 ret = filemap_fdatawait(inode->i_mapping);
73 ret = nfs_commit_inode(inode, FLUSH_SYNC);
75 ret = nfs_commit_inode(inode, 0);
78 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
82 void nfs_clear_inode(struct inode *inode)
85 * The following should never happen...
87 BUG_ON(nfs_have_writebacks(inode));
88 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
89 BUG_ON(atomic_read(&NFS_I(inode)->data_updates) != 0);
90 nfs_zap_acl_cache(inode);
91 nfs_access_zap_cache(inode);
95 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
97 int nfs_sync_mapping(struct address_space *mapping)
101 if (mapping->nrpages == 0)
103 unmap_mapping_range(mapping, 0, 0, 0);
104 ret = filemap_write_and_wait(mapping);
107 ret = nfs_wb_all(mapping->host);
113 * Invalidate the local caches
115 static void nfs_zap_caches_locked(struct inode *inode)
117 struct nfs_inode *nfsi = NFS_I(inode);
118 int mode = inode->i_mode;
120 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
122 NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
123 NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
125 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
126 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
127 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
129 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
132 void nfs_zap_caches(struct inode *inode)
134 spin_lock(&inode->i_lock);
135 nfs_zap_caches_locked(inode);
136 spin_unlock(&inode->i_lock);
139 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
141 if (mapping->nrpages != 0) {
142 spin_lock(&inode->i_lock);
143 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
144 spin_unlock(&inode->i_lock);
148 static void nfs_zap_acl_cache(struct inode *inode)
150 void (*clear_acl_cache)(struct inode *);
152 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
153 if (clear_acl_cache != NULL)
154 clear_acl_cache(inode);
155 spin_lock(&inode->i_lock);
156 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
157 spin_unlock(&inode->i_lock);
161 * Invalidate, but do not unhash, the inode.
162 * NB: must be called with inode->i_lock held!
164 static void nfs_invalidate_inode(struct inode *inode)
166 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
167 nfs_zap_caches_locked(inode);
170 struct nfs_find_desc {
172 struct nfs_fattr *fattr;
176 * In NFSv3 we can have 64bit inode numbers. In order to support
177 * this, and re-exported directories (also seen in NFSv2)
178 * we are forced to allow 2 different inodes to have the same
182 nfs_find_actor(struct inode *inode, void *opaque)
184 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
185 struct nfs_fh *fh = desc->fh;
186 struct nfs_fattr *fattr = desc->fattr;
188 if (NFS_FILEID(inode) != fattr->fileid)
190 if (nfs_compare_fh(NFS_FH(inode), fh))
192 if (is_bad_inode(inode) || NFS_STALE(inode))
198 nfs_init_locked(struct inode *inode, void *opaque)
200 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
201 struct nfs_fattr *fattr = desc->fattr;
203 NFS_FILEID(inode) = fattr->fileid;
204 nfs_copy_fh(NFS_FH(inode), desc->fh);
208 /* Don't use READDIRPLUS on directories that we believe are too large */
209 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
212 * This is our front-end to iget that looks up inodes by file handle
213 * instead of inode number.
216 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
218 struct nfs_find_desc desc = {
222 struct inode *inode = ERR_PTR(-ENOENT);
225 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
229 printk("NFS: Buggy server - nlink == 0!\n");
233 hash = nfs_fattr_to_ino_t(fattr);
235 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
237 inode = ERR_PTR(-ENOMEM);
241 if (inode->i_state & I_NEW) {
242 struct nfs_inode *nfsi = NFS_I(inode);
244 /* We set i_ino for the few things that still rely on it,
248 /* We can't support update_atime(), since the server will reset it */
249 inode->i_flags |= S_NOATIME|S_NOCMTIME;
250 inode->i_mode = fattr->mode;
251 /* Why so? Because we want revalidate for devices/FIFOs, and
252 * that's precisely what we have in nfs_file_inode_operations.
254 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
255 if (S_ISREG(inode->i_mode)) {
256 inode->i_fop = &nfs_file_operations;
257 inode->i_data.a_ops = &nfs_file_aops;
258 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
259 } else if (S_ISDIR(inode->i_mode)) {
260 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
261 inode->i_fop = &nfs_dir_operations;
262 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
263 && fattr->size <= NFS_LIMIT_READDIRPLUS)
264 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
265 /* Deal with crossing mountpoints */
266 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
267 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
268 inode->i_op = &nfs_referral_inode_operations;
270 inode->i_op = &nfs_mountpoint_inode_operations;
273 } else if (S_ISLNK(inode->i_mode))
274 inode->i_op = &nfs_symlink_inode_operations;
276 init_special_inode(inode, inode->i_mode, fattr->rdev);
278 nfsi->read_cache_jiffies = fattr->time_start;
279 nfsi->last_updated = jiffies;
280 inode->i_atime = fattr->atime;
281 inode->i_mtime = fattr->mtime;
282 inode->i_ctime = fattr->ctime;
283 if (fattr->valid & NFS_ATTR_FATTR_V4)
284 nfsi->change_attr = fattr->change_attr;
285 inode->i_size = nfs_size_to_loff_t(fattr->size);
286 inode->i_nlink = fattr->nlink;
287 inode->i_uid = fattr->uid;
288 inode->i_gid = fattr->gid;
289 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
291 * report the blocks in 512byte units
293 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
295 inode->i_blocks = fattr->du.nfs2.blocks;
297 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
298 nfsi->attrtimeo_timestamp = jiffies;
299 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
300 nfsi->access_cache = RB_ROOT;
302 unlock_new_inode(inode);
304 nfs_refresh_inode(inode, fattr);
305 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
307 (long long)NFS_FILEID(inode),
308 atomic_read(&inode->i_count));
314 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
318 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
321 nfs_setattr(struct dentry *dentry, struct iattr *attr)
323 struct inode *inode = dentry->d_inode;
324 struct nfs_fattr fattr;
327 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
329 if (attr->ia_valid & ATTR_SIZE) {
330 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
331 attr->ia_valid &= ~ATTR_SIZE;
334 /* Optimization: if the end result is no change, don't RPC */
335 attr->ia_valid &= NFS_VALID_ATTRS;
336 if (attr->ia_valid == 0)
340 nfs_begin_data_update(inode);
341 /* Write all dirty data */
342 filemap_write_and_wait(inode->i_mapping);
345 * Return any delegations if we're going to change ACLs
347 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
348 nfs_inode_return_delegation(inode);
349 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
351 nfs_refresh_inode(inode, &fattr);
352 nfs_end_data_update(inode);
358 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
359 * @inode: pointer to struct inode
360 * @attr: pointer to struct iattr
362 * Note: we do this in the *proc.c in order to ensure that
363 * it works for things like exclusive creates too.
365 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
367 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
368 if ((attr->ia_valid & ATTR_MODE) != 0) {
369 int mode = attr->ia_mode & S_IALLUGO;
370 mode |= inode->i_mode & ~S_IALLUGO;
371 inode->i_mode = mode;
373 if ((attr->ia_valid & ATTR_UID) != 0)
374 inode->i_uid = attr->ia_uid;
375 if ((attr->ia_valid & ATTR_GID) != 0)
376 inode->i_gid = attr->ia_gid;
377 spin_lock(&inode->i_lock);
378 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
379 spin_unlock(&inode->i_lock);
381 if ((attr->ia_valid & ATTR_SIZE) != 0) {
382 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
383 inode->i_size = attr->ia_size;
384 vmtruncate(inode, attr->ia_size);
388 static int nfs_wait_schedule(void *word)
390 if (signal_pending(current))
397 * Wait for the inode to get unlocked.
399 static int nfs_wait_on_inode(struct inode *inode)
401 struct rpc_clnt *clnt = NFS_CLIENT(inode);
402 struct nfs_inode *nfsi = NFS_I(inode);
406 rpc_clnt_sigmask(clnt, &oldmask);
407 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
408 nfs_wait_schedule, TASK_INTERRUPTIBLE);
409 rpc_clnt_sigunmask(clnt, &oldmask);
414 static void nfs_wake_up_inode(struct inode *inode)
416 struct nfs_inode *nfsi = NFS_I(inode);
418 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
419 smp_mb__after_clear_bit();
420 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
423 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
425 struct inode *inode = dentry->d_inode;
426 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
429 /* Flush out writes to the server in order to update c/mtime */
430 nfs_sync_mapping_range(inode->i_mapping, 0, 0, FLUSH_NOCOMMIT);
433 * We may force a getattr if the user cares about atime.
435 * Note that we only have to check the vfsmount flags here:
436 * - NFS always sets S_NOATIME by so checking it would give a
438 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
439 * no point in checking those.
441 if ((mnt->mnt_flags & MNT_NOATIME) ||
442 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
446 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
448 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
450 generic_fillattr(inode, stat);
454 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
456 struct nfs_open_context *ctx;
458 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
460 atomic_set(&ctx->count, 1);
461 ctx->dentry = dget(dentry);
462 ctx->vfsmnt = mntget(mnt);
463 ctx->cred = get_rpccred(cred);
465 ctx->lockowner = current->files;
472 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
475 atomic_inc(&ctx->count);
479 void put_nfs_open_context(struct nfs_open_context *ctx)
481 if (atomic_dec_and_test(&ctx->count)) {
482 if (!list_empty(&ctx->list)) {
483 struct inode *inode = ctx->dentry->d_inode;
484 spin_lock(&inode->i_lock);
485 list_del(&ctx->list);
486 spin_unlock(&inode->i_lock);
488 if (ctx->state != NULL)
489 nfs4_close_state(ctx->state, ctx->mode);
490 if (ctx->cred != NULL)
491 put_rpccred(ctx->cred);
499 * Ensure that mmap has a recent RPC credential for use when writing out
502 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
504 struct inode *inode = filp->f_path.dentry->d_inode;
505 struct nfs_inode *nfsi = NFS_I(inode);
507 filp->private_data = get_nfs_open_context(ctx);
508 spin_lock(&inode->i_lock);
509 list_add(&ctx->list, &nfsi->open_files);
510 spin_unlock(&inode->i_lock);
514 * Given an inode, search for an open context with the desired characteristics
516 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
518 struct nfs_inode *nfsi = NFS_I(inode);
519 struct nfs_open_context *pos, *ctx = NULL;
521 spin_lock(&inode->i_lock);
522 list_for_each_entry(pos, &nfsi->open_files, list) {
523 if (cred != NULL && pos->cred != cred)
525 if ((pos->mode & mode) == mode) {
526 ctx = get_nfs_open_context(pos);
530 spin_unlock(&inode->i_lock);
534 static void nfs_file_clear_open_context(struct file *filp)
536 struct inode *inode = filp->f_path.dentry->d_inode;
537 struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
540 filp->private_data = NULL;
541 spin_lock(&inode->i_lock);
542 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
543 spin_unlock(&inode->i_lock);
544 put_nfs_open_context(ctx);
549 * These allocate and release file read/write context information.
551 int nfs_open(struct inode *inode, struct file *filp)
553 struct nfs_open_context *ctx;
554 struct rpc_cred *cred;
556 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
558 return PTR_ERR(cred);
559 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
563 ctx->mode = filp->f_mode;
564 nfs_file_set_open_context(filp, ctx);
565 put_nfs_open_context(ctx);
569 int nfs_release(struct inode *inode, struct file *filp)
571 nfs_file_clear_open_context(filp);
576 * This function is called whenever some part of NFS notices that
577 * the cached attributes have to be refreshed.
580 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
582 int status = -ESTALE;
583 struct nfs_fattr fattr;
584 struct nfs_inode *nfsi = NFS_I(inode);
586 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
587 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
589 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
591 if (is_bad_inode(inode))
593 if (NFS_STALE(inode))
596 status = nfs_wait_on_inode(inode);
599 if (NFS_STALE(inode)) {
601 /* Do we trust the cached ESTALE? */
602 if (NFS_ATTRTIMEO(inode) != 0) {
603 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
610 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
612 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
614 (long long)NFS_FILEID(inode), status);
615 if (status == -ESTALE) {
616 nfs_zap_caches(inode);
617 if (!S_ISDIR(inode->i_mode))
618 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
623 spin_lock(&inode->i_lock);
624 status = nfs_update_inode(inode, &fattr);
626 spin_unlock(&inode->i_lock);
627 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
629 (long long)NFS_FILEID(inode), status);
632 spin_unlock(&inode->i_lock);
634 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
635 nfs_zap_acl_cache(inode);
637 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
639 (long long)NFS_FILEID(inode));
642 nfs_wake_up_inode(inode);
649 int nfs_attribute_timeout(struct inode *inode)
651 struct nfs_inode *nfsi = NFS_I(inode);
653 if (nfs_have_delegation(inode, FMODE_READ))
655 return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
659 * nfs_revalidate_inode - Revalidate the inode attributes
660 * @server - pointer to nfs_server struct
661 * @inode - pointer to inode struct
663 * Updates inode attribute information by retrieving the data from the server.
665 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
667 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
668 && !nfs_attribute_timeout(inode))
669 return NFS_STALE(inode) ? -ESTALE : 0;
670 return __nfs_revalidate_inode(server, inode);
673 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
675 struct nfs_inode *nfsi = NFS_I(inode);
677 if (mapping->nrpages != 0) {
678 int ret = invalidate_inode_pages2(mapping);
682 spin_lock(&inode->i_lock);
683 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
684 if (S_ISDIR(inode->i_mode)) {
685 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
686 /* This ensures we revalidate child dentries */
687 nfsi->cache_change_attribute = jiffies;
689 spin_unlock(&inode->i_lock);
690 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
691 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
692 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
696 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
700 mutex_lock(&inode->i_mutex);
701 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
702 ret = nfs_sync_mapping(mapping);
704 ret = nfs_invalidate_mapping_nolock(inode, mapping);
706 mutex_unlock(&inode->i_mutex);
711 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
712 * @inode - pointer to host inode
713 * @mapping - pointer to mapping
715 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
717 struct nfs_inode *nfsi = NFS_I(inode);
720 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
721 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
722 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
726 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
727 ret = nfs_invalidate_mapping_nolock(inode, mapping);
733 * nfs_revalidate_mapping - Revalidate the pagecache
734 * @inode - pointer to host inode
735 * @mapping - pointer to mapping
737 * This version of the function will take the inode->i_mutex and attempt to
738 * flush out all dirty data if it needs to invalidate the page cache.
740 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
742 struct nfs_inode *nfsi = NFS_I(inode);
745 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
746 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
747 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
751 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
752 ret = nfs_invalidate_mapping(inode, mapping);
758 * nfs_begin_data_update
759 * @inode - pointer to inode
760 * Declare that a set of operations will update file data on the server
762 void nfs_begin_data_update(struct inode *inode)
764 atomic_inc(&NFS_I(inode)->data_updates);
768 * nfs_end_data_update
769 * @inode - pointer to inode
770 * Declare end of the operations that will update file data
771 * This will mark the inode as immediately needing revalidation
772 * of its attribute cache.
774 void nfs_end_data_update(struct inode *inode)
776 struct nfs_inode *nfsi = NFS_I(inode);
778 /* Directories: invalidate page cache */
779 if (S_ISDIR(inode->i_mode)) {
780 spin_lock(&inode->i_lock);
781 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
782 spin_unlock(&inode->i_lock);
784 nfsi->cache_change_attribute = jiffies;
785 atomic_dec(&nfsi->data_updates);
788 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
790 struct nfs_inode *nfsi = NFS_I(inode);
792 /* If we have atomic WCC data, we may update some attributes */
793 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
794 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
795 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
796 nfsi->cache_change_attribute = jiffies;
798 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
799 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
800 nfsi->cache_change_attribute = jiffies;
802 if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
803 inode->i_size = fattr->size;
804 nfsi->cache_change_attribute = jiffies;
810 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
811 * @inode - pointer to inode
812 * @fattr - updated attributes
814 * Verifies the attribute cache. If we have just changed the attributes,
815 * so that fattr carries weak cache consistency data, then it may
816 * also update the ctime/mtime/change_attribute.
818 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
820 struct nfs_inode *nfsi = NFS_I(inode);
821 loff_t cur_size, new_isize;
825 /* Has the inode gone and changed behind our back? */
826 if (nfsi->fileid != fattr->fileid
827 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
831 /* Are we in the process of updating data on the server? */
832 data_unstable = nfs_caches_unstable(inode);
834 /* Do atomic weak cache consistency updates */
835 nfs_wcc_update_inode(inode, fattr);
837 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
838 nfsi->change_attr != fattr->change_attr)
839 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
841 /* Verify a few of the more important attributes */
842 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
843 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
845 cur_size = i_size_read(inode);
846 new_isize = nfs_size_to_loff_t(fattr->size);
847 if (cur_size != new_isize && nfsi->npages == 0)
848 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
850 /* Have any file permissions changed? */
851 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
852 || inode->i_uid != fattr->uid
853 || inode->i_gid != fattr->gid)
854 nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
856 /* Has the link count changed? */
857 if (inode->i_nlink != fattr->nlink)
858 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
860 if (!timespec_equal(&inode->i_atime, &fattr->atime))
861 nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
863 nfsi->read_cache_jiffies = fattr->time_start;
868 * nfs_refresh_inode - try to update the inode attribute cache
869 * @inode - pointer to inode
870 * @fattr - updated attributes
872 * Check that an RPC call that returned attributes has not overlapped with
873 * other recent updates of the inode metadata, then decide whether it is
874 * safe to do a full update of the inode attributes, or whether just to
875 * call nfs_check_inode_attributes.
877 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
879 struct nfs_inode *nfsi = NFS_I(inode);
882 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
884 spin_lock(&inode->i_lock);
885 if (time_after(fattr->time_start, nfsi->last_updated))
886 status = nfs_update_inode(inode, fattr);
888 status = nfs_check_inode_attributes(inode, fattr);
890 spin_unlock(&inode->i_lock);
895 * nfs_post_op_update_inode - try to update the inode attribute cache
896 * @inode - pointer to inode
897 * @fattr - updated attributes
899 * After an operation that has changed the inode metadata, mark the
900 * attribute cache as being invalid, then try to update it.
902 * NB: if the server didn't return any post op attributes, this
903 * function will force the retrieval of attributes before the next
904 * NFS request. Thus it should be used only for operations that
905 * are expected to change one or more attributes, to avoid
906 * unnecessary NFS requests and trips through nfs_update_inode().
908 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
910 struct nfs_inode *nfsi = NFS_I(inode);
913 spin_lock(&inode->i_lock);
914 if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
915 nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
918 status = nfs_update_inode(inode, fattr);
920 spin_unlock(&inode->i_lock);
925 * Many nfs protocol calls return the new file attributes after
926 * an operation. Here we update the inode to reflect the state
927 * of the server's inode.
929 * This is a bit tricky because we have to make sure all dirty pages
930 * have been sent off to the server before calling invalidate_inode_pages.
931 * To make sure no other process adds more write requests while we try
932 * our best to flush them, we make them sleep during the attribute refresh.
934 * A very similar scenario holds for the dir cache.
936 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
938 struct nfs_server *server;
939 struct nfs_inode *nfsi = NFS_I(inode);
940 loff_t cur_isize, new_isize;
941 unsigned int invalid = 0;
944 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
945 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
946 atomic_read(&inode->i_count), fattr->valid);
948 if (nfsi->fileid != fattr->fileid)
952 * Make sure the inode's type hasn't changed.
954 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
957 server = NFS_SERVER(inode);
958 /* Update the fsid if and only if this is the root directory */
959 if (inode == inode->i_sb->s_root->d_inode
960 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
961 server->fsid = fattr->fsid;
964 * Update the read time so we don't revalidate too often.
966 nfsi->read_cache_jiffies = fattr->time_start;
967 nfsi->last_updated = jiffies;
969 /* Are we racing with known updates of the metadata on the server? */
970 data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
972 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
974 /* Do atomic weak cache consistency updates */
975 nfs_wcc_update_inode(inode, fattr);
977 /* Check if our cached file size is stale */
978 new_isize = nfs_size_to_loff_t(fattr->size);
979 cur_isize = i_size_read(inode);
980 if (new_isize != cur_isize) {
981 /* Do we perhaps have any outstanding writes? */
982 if (nfsi->npages == 0) {
983 /* No, but did we race with nfs_end_data_update()? */
985 inode->i_size = new_isize;
986 invalid |= NFS_INO_INVALID_DATA;
988 invalid |= NFS_INO_INVALID_ATTR;
989 } else if (new_isize > cur_isize) {
990 inode->i_size = new_isize;
991 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
993 nfsi->cache_change_attribute = jiffies;
994 dprintk("NFS: isize change on server for file %s/%ld\n",
995 inode->i_sb->s_id, inode->i_ino);
998 /* Check if the mtime agrees */
999 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1000 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_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 = jiffies;
1007 /* If ctime has changed we should definitely clear access+acl caches */
1008 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1009 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1010 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1011 nfsi->cache_change_attribute = jiffies;
1013 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1015 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1016 inode->i_uid != fattr->uid ||
1017 inode->i_gid != fattr->gid)
1018 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1020 inode->i_mode = fattr->mode;
1021 inode->i_nlink = fattr->nlink;
1022 inode->i_uid = fattr->uid;
1023 inode->i_gid = fattr->gid;
1025 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1027 * report the blocks in 512byte units
1029 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1031 inode->i_blocks = fattr->du.nfs2.blocks;
1034 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1035 nfsi->change_attr != fattr->change_attr) {
1036 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1037 inode->i_sb->s_id, inode->i_ino);
1038 nfsi->change_attr = fattr->change_attr;
1039 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1040 nfsi->cache_change_attribute = jiffies;
1043 /* Update attrtimeo value if we're out of the unstable period */
1044 if (invalid & NFS_INO_INVALID_ATTR) {
1045 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1046 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1047 nfsi->attrtimeo_timestamp = jiffies;
1048 } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
1049 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1050 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1051 nfsi->attrtimeo_timestamp = jiffies;
1053 /* Don't invalidate the data if we were to blame */
1054 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1055 || S_ISLNK(inode->i_mode)))
1056 invalid &= ~NFS_INO_INVALID_DATA;
1058 invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
1059 if (!nfs_have_delegation(inode, FMODE_READ))
1060 nfsi->cache_validity |= invalid;
1065 * Big trouble! The inode has become a different object.
1068 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1069 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1073 * No need to worry about unhashing the dentry, as the
1074 * lookup validation will know that the inode is bad.
1075 * (But we fall through to invalidate the caches.)
1077 nfs_invalidate_inode(inode);
1081 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1082 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1083 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1084 (long long)nfsi->fileid, (long long)fattr->fileid);
1089 #ifdef CONFIG_NFS_V4
1092 * Clean out any remaining NFSv4 state that might be left over due
1093 * to open() calls that passed nfs_atomic_lookup, but failed to call
1096 void nfs4_clear_inode(struct inode *inode)
1098 struct nfs_inode *nfsi = NFS_I(inode);
1100 /* If we are holding a delegation, return it! */
1101 nfs_inode_return_delegation(inode);
1102 /* First call standard NFS clear_inode() code */
1103 nfs_clear_inode(inode);
1104 /* Now clear out any remaining state */
1105 while (!list_empty(&nfsi->open_states)) {
1106 struct nfs4_state *state;
1108 state = list_entry(nfsi->open_states.next,
1111 dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
1114 (long long)NFS_FILEID(inode),
1116 BUG_ON(atomic_read(&state->count) != 1);
1117 nfs4_close_state(state, state->state);
1122 struct inode *nfs_alloc_inode(struct super_block *sb)
1124 struct nfs_inode *nfsi;
1125 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1129 nfsi->cache_validity = 0UL;
1130 nfsi->cache_change_attribute = jiffies;
1131 #ifdef CONFIG_NFS_V3_ACL
1132 nfsi->acl_access = ERR_PTR(-EAGAIN);
1133 nfsi->acl_default = ERR_PTR(-EAGAIN);
1135 #ifdef CONFIG_NFS_V4
1136 nfsi->nfs4_acl = NULL;
1137 #endif /* CONFIG_NFS_V4 */
1138 return &nfsi->vfs_inode;
1141 void nfs_destroy_inode(struct inode *inode)
1143 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1146 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1148 #ifdef CONFIG_NFS_V4
1149 INIT_LIST_HEAD(&nfsi->open_states);
1150 nfsi->delegation = NULL;
1151 nfsi->delegation_state = 0;
1152 init_rwsem(&nfsi->rwsem);
1156 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1158 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1160 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1161 SLAB_CTOR_CONSTRUCTOR) {
1162 inode_init_once(&nfsi->vfs_inode);
1163 spin_lock_init(&nfsi->req_lock);
1164 INIT_LIST_HEAD(&nfsi->dirty);
1165 INIT_LIST_HEAD(&nfsi->commit);
1166 INIT_LIST_HEAD(&nfsi->open_files);
1167 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1168 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1169 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1170 atomic_set(&nfsi->data_updates, 0);
1174 nfs4_init_once(nfsi);
1178 static int __init nfs_init_inodecache(void)
1180 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1181 sizeof(struct nfs_inode),
1182 0, (SLAB_RECLAIM_ACCOUNT|
1185 if (nfs_inode_cachep == NULL)
1191 static void nfs_destroy_inodecache(void)
1193 kmem_cache_destroy(nfs_inode_cachep);
1199 static int __init init_nfs_fs(void)
1203 err = nfs_fs_proc_init();
1207 err = nfs_init_nfspagecache();
1211 err = nfs_init_inodecache();
1215 err = nfs_init_readpagecache();
1219 err = nfs_init_writepagecache();
1223 err = nfs_init_directcache();
1227 #ifdef CONFIG_PROC_FS
1228 rpc_proc_register(&nfs_rpcstat);
1230 if ((err = register_nfs_fs()) != 0)
1234 #ifdef CONFIG_PROC_FS
1235 rpc_proc_unregister("nfs");
1237 nfs_destroy_directcache();
1239 nfs_destroy_writepagecache();
1241 nfs_destroy_readpagecache();
1243 nfs_destroy_inodecache();
1245 nfs_destroy_nfspagecache();
1252 static void __exit exit_nfs_fs(void)
1254 nfs_destroy_directcache();
1255 nfs_destroy_writepagecache();
1256 nfs_destroy_readpagecache();
1257 nfs_destroy_inodecache();
1258 nfs_destroy_nfspagecache();
1259 #ifdef CONFIG_PROC_FS
1260 rpc_proc_unregister("nfs");
1262 unregister_nfs_fs();
1266 /* Not quite true; I just maintain it */
1267 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1268 MODULE_LICENSE("GPL");
1270 module_init(init_nfs_fs)
1271 module_exit(exit_nfs_fs)