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_I(inode)->flags);
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 set_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_I(inode)->flags);
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;
302 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
304 } else if (S_ISLNK(inode->i_mode))
305 inode->i_op = &nfs_symlink_inode_operations;
307 init_special_inode(inode, inode->i_mode, fattr->rdev);
309 nfsi->read_cache_jiffies = fattr->time_start;
310 nfsi->last_updated = now;
311 nfsi->cache_change_attribute = now;
312 inode->i_atime = fattr->atime;
313 inode->i_mtime = fattr->mtime;
314 inode->i_ctime = fattr->ctime;
315 if (fattr->valid & NFS_ATTR_FATTR_V4)
316 nfsi->change_attr = fattr->change_attr;
317 inode->i_size = nfs_size_to_loff_t(fattr->size);
318 inode->i_nlink = fattr->nlink;
319 inode->i_uid = fattr->uid;
320 inode->i_gid = fattr->gid;
321 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
323 * report the blocks in 512byte units
325 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
327 inode->i_blocks = fattr->du.nfs2.blocks;
329 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
330 nfsi->attrtimeo_timestamp = now;
331 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
332 nfsi->access_cache = RB_ROOT;
334 unlock_new_inode(inode);
336 nfs_refresh_inode(inode, fattr);
337 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
339 (long long)NFS_FILEID(inode),
340 atomic_read(&inode->i_count));
346 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
350 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
353 nfs_setattr(struct dentry *dentry, struct iattr *attr)
355 struct inode *inode = dentry->d_inode;
356 struct nfs_fattr fattr;
359 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
361 /* skip mode change if it's just for clearing setuid/setgid */
362 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
363 attr->ia_valid &= ~ATTR_MODE;
365 if (attr->ia_valid & ATTR_SIZE) {
366 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
367 attr->ia_valid &= ~ATTR_SIZE;
370 /* Optimization: if the end result is no change, don't RPC */
371 attr->ia_valid &= NFS_VALID_ATTRS;
372 if (attr->ia_valid == 0)
376 /* Write all dirty data */
377 if (S_ISREG(inode->i_mode)) {
378 filemap_write_and_wait(inode->i_mapping);
382 * Return any delegations if we're going to change ACLs
384 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
385 nfs_inode_return_delegation(inode);
386 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
388 nfs_refresh_inode(inode, &fattr);
394 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
395 * @inode: pointer to struct inode
396 * @attr: pointer to struct iattr
398 * Note: we do this in the *proc.c in order to ensure that
399 * it works for things like exclusive creates too.
401 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
403 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
404 if ((attr->ia_valid & ATTR_MODE) != 0) {
405 int mode = attr->ia_mode & S_IALLUGO;
406 mode |= inode->i_mode & ~S_IALLUGO;
407 inode->i_mode = mode;
409 if ((attr->ia_valid & ATTR_UID) != 0)
410 inode->i_uid = attr->ia_uid;
411 if ((attr->ia_valid & ATTR_GID) != 0)
412 inode->i_gid = attr->ia_gid;
413 spin_lock(&inode->i_lock);
414 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
415 spin_unlock(&inode->i_lock);
417 if ((attr->ia_valid & ATTR_SIZE) != 0) {
418 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
419 inode->i_size = attr->ia_size;
420 vmtruncate(inode, attr->ia_size);
424 static int nfs_wait_schedule(void *word)
426 if (signal_pending(current))
433 * Wait for the inode to get unlocked.
435 static int nfs_wait_on_inode(struct inode *inode)
437 struct nfs_inode *nfsi = NFS_I(inode);
440 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
441 nfs_wait_schedule, TASK_KILLABLE);
446 static void nfs_wake_up_inode(struct inode *inode)
448 struct nfs_inode *nfsi = NFS_I(inode);
450 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
451 smp_mb__after_clear_bit();
452 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
455 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
457 struct inode *inode = dentry->d_inode;
458 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
462 * Flush out writes to the server in order to update c/mtime.
464 * Hold the i_mutex to suspend application writes temporarily;
465 * this prevents long-running writing applications from blocking
468 if (S_ISREG(inode->i_mode)) {
469 mutex_lock(&inode->i_mutex);
470 nfs_wb_nocommit(inode);
471 mutex_unlock(&inode->i_mutex);
475 * We may force a getattr if the user cares about atime.
477 * Note that we only have to check the vfsmount flags here:
478 * - NFS always sets S_NOATIME by so checking it would give a
480 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
481 * no point in checking those.
483 if ((mnt->mnt_flags & MNT_NOATIME) ||
484 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
488 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
490 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
492 generic_fillattr(inode, stat);
493 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
498 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
500 struct nfs_open_context *ctx;
502 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
504 ctx->path.dentry = dget(dentry);
505 ctx->path.mnt = mntget(mnt);
506 ctx->cred = get_rpccred(cred);
508 ctx->lockowner = current->files;
511 atomic_set(&ctx->count, 1);
516 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
519 atomic_inc(&ctx->count);
523 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
530 inode = ctx->path.dentry->d_inode;
531 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
533 list_del(&ctx->list);
534 spin_unlock(&inode->i_lock);
535 if (ctx->state != NULL) {
537 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
539 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
541 if (ctx->cred != NULL)
542 put_rpccred(ctx->cred);
543 dput(ctx->path.dentry);
544 mntput(ctx->path.mnt);
548 void put_nfs_open_context(struct nfs_open_context *ctx)
550 __put_nfs_open_context(ctx, 0);
553 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
555 __put_nfs_open_context(ctx, 1);
559 * Ensure that mmap has a recent RPC credential for use when writing out
562 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
564 struct inode *inode = filp->f_path.dentry->d_inode;
565 struct nfs_inode *nfsi = NFS_I(inode);
567 filp->private_data = get_nfs_open_context(ctx);
568 spin_lock(&inode->i_lock);
569 list_add(&ctx->list, &nfsi->open_files);
570 spin_unlock(&inode->i_lock);
574 * Given an inode, search for an open context with the desired characteristics
576 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
578 struct nfs_inode *nfsi = NFS_I(inode);
579 struct nfs_open_context *pos, *ctx = NULL;
581 spin_lock(&inode->i_lock);
582 list_for_each_entry(pos, &nfsi->open_files, list) {
583 if (cred != NULL && pos->cred != cred)
585 if ((pos->mode & mode) == mode) {
586 ctx = get_nfs_open_context(pos);
590 spin_unlock(&inode->i_lock);
594 static void nfs_file_clear_open_context(struct file *filp)
596 struct inode *inode = filp->f_path.dentry->d_inode;
597 struct nfs_open_context *ctx = nfs_file_open_context(filp);
600 filp->private_data = NULL;
601 spin_lock(&inode->i_lock);
602 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
603 spin_unlock(&inode->i_lock);
604 put_nfs_open_context_sync(ctx);
609 * These allocate and release file read/write context information.
611 int nfs_open(struct inode *inode, struct file *filp)
613 struct nfs_open_context *ctx;
614 struct rpc_cred *cred;
616 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
618 return PTR_ERR(cred);
619 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
623 ctx->mode = filp->f_mode;
624 nfs_file_set_open_context(filp, ctx);
625 put_nfs_open_context(ctx);
629 int nfs_release(struct inode *inode, struct file *filp)
631 nfs_file_clear_open_context(filp);
636 * This function is called whenever some part of NFS notices that
637 * the cached attributes have to be refreshed.
640 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
642 int status = -ESTALE;
643 struct nfs_fattr fattr;
644 struct nfs_inode *nfsi = NFS_I(inode);
646 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
647 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
649 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
651 if (is_bad_inode(inode))
653 if (NFS_STALE(inode))
656 status = nfs_wait_on_inode(inode);
661 if (NFS_STALE(inode))
664 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
666 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
668 (long long)NFS_FILEID(inode), status);
669 if (status == -ESTALE) {
670 nfs_zap_caches(inode);
671 if (!S_ISDIR(inode->i_mode))
672 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
677 spin_lock(&inode->i_lock);
678 status = nfs_update_inode(inode, &fattr);
680 spin_unlock(&inode->i_lock);
681 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
683 (long long)NFS_FILEID(inode), status);
686 spin_unlock(&inode->i_lock);
688 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
689 nfs_zap_acl_cache(inode);
691 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
693 (long long)NFS_FILEID(inode));
696 nfs_wake_up_inode(inode);
703 int nfs_attribute_timeout(struct inode *inode)
705 struct nfs_inode *nfsi = NFS_I(inode);
707 if (nfs_have_delegation(inode, FMODE_READ))
709 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
713 * nfs_revalidate_inode - Revalidate the inode attributes
714 * @server - pointer to nfs_server struct
715 * @inode - pointer to inode struct
717 * Updates inode attribute information by retrieving the data from the server.
719 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
721 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
722 && !nfs_attribute_timeout(inode))
723 return NFS_STALE(inode) ? -ESTALE : 0;
724 return __nfs_revalidate_inode(server, inode);
727 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
729 struct nfs_inode *nfsi = NFS_I(inode);
731 if (mapping->nrpages != 0) {
732 int ret = invalidate_inode_pages2(mapping);
736 spin_lock(&inode->i_lock);
737 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
738 if (S_ISDIR(inode->i_mode))
739 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
740 spin_unlock(&inode->i_lock);
741 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
742 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
743 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
747 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
751 mutex_lock(&inode->i_mutex);
752 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
753 ret = nfs_sync_mapping(mapping);
755 ret = nfs_invalidate_mapping_nolock(inode, mapping);
757 mutex_unlock(&inode->i_mutex);
762 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
763 * @inode - pointer to host inode
764 * @mapping - pointer to mapping
766 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
768 struct nfs_inode *nfsi = NFS_I(inode);
771 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
772 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
773 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
777 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
778 ret = nfs_invalidate_mapping_nolock(inode, mapping);
784 * nfs_revalidate_mapping - Revalidate the pagecache
785 * @inode - pointer to host inode
786 * @mapping - pointer to mapping
788 * This version of the function will take the inode->i_mutex and attempt to
789 * flush out all dirty data if it needs to invalidate the page cache.
791 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
793 struct nfs_inode *nfsi = NFS_I(inode);
796 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
797 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
798 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
802 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
803 ret = nfs_invalidate_mapping(inode, mapping);
808 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
810 struct nfs_inode *nfsi = NFS_I(inode);
812 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
813 nfsi->change_attr == fattr->pre_change_attr) {
814 nfsi->change_attr = fattr->change_attr;
815 if (S_ISDIR(inode->i_mode))
816 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
818 /* If we have atomic WCC data, we may update some attributes */
819 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
820 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
821 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
822 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
823 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
824 if (S_ISDIR(inode->i_mode))
825 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
827 if (inode->i_size == nfs_size_to_loff_t(fattr->pre_size) &&
829 inode->i_size = nfs_size_to_loff_t(fattr->size);
834 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
835 * @inode - pointer to inode
836 * @fattr - updated attributes
838 * Verifies the attribute cache. If we have just changed the attributes,
839 * so that fattr carries weak cache consistency data, then it may
840 * also update the ctime/mtime/change_attribute.
842 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
844 struct nfs_inode *nfsi = NFS_I(inode);
845 loff_t cur_size, new_isize;
846 unsigned long invalid = 0;
849 /* Has the inode gone and changed behind our back? */
850 if (nfsi->fileid != fattr->fileid
851 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
855 /* Do atomic weak cache consistency updates */
856 nfs_wcc_update_inode(inode, fattr);
858 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
859 nfsi->change_attr != fattr->change_attr)
860 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
862 /* Verify a few of the more important attributes */
863 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
864 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
866 cur_size = i_size_read(inode);
867 new_isize = nfs_size_to_loff_t(fattr->size);
868 if (cur_size != new_isize && nfsi->npages == 0)
869 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
871 /* Have any file permissions changed? */
872 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
873 || inode->i_uid != fattr->uid
874 || inode->i_gid != fattr->gid)
875 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
877 /* Has the link count changed? */
878 if (inode->i_nlink != fattr->nlink)
879 invalid |= NFS_INO_INVALID_ATTR;
881 if (!timespec_equal(&inode->i_atime, &fattr->atime))
882 invalid |= NFS_INO_INVALID_ATIME;
885 nfsi->cache_validity |= invalid;
887 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
888 | NFS_INO_INVALID_ATIME
889 | NFS_INO_REVAL_PAGECACHE);
891 nfsi->read_cache_jiffies = fattr->time_start;
896 * nfs_refresh_inode - try to update the inode attribute cache
897 * @inode - pointer to inode
898 * @fattr - updated attributes
900 * Check that an RPC call that returned attributes has not overlapped with
901 * other recent updates of the inode metadata, then decide whether it is
902 * safe to do a full update of the inode attributes, or whether just to
903 * call nfs_check_inode_attributes.
905 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
907 struct nfs_inode *nfsi = NFS_I(inode);
910 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
912 spin_lock(&inode->i_lock);
913 if (time_after(fattr->time_start, nfsi->last_updated))
914 status = nfs_update_inode(inode, fattr);
916 status = nfs_check_inode_attributes(inode, fattr);
918 spin_unlock(&inode->i_lock);
923 * nfs_post_op_update_inode - try to update the inode attribute cache
924 * @inode - pointer to inode
925 * @fattr - updated attributes
927 * After an operation that has changed the inode metadata, mark the
928 * attribute cache as being invalid, then try to update it.
930 * NB: if the server didn't return any post op attributes, this
931 * function will force the retrieval of attributes before the next
932 * NFS request. Thus it should be used only for operations that
933 * are expected to change one or more attributes, to avoid
934 * unnecessary NFS requests and trips through nfs_update_inode().
936 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
938 struct nfs_inode *nfsi = NFS_I(inode);
940 spin_lock(&inode->i_lock);
941 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
942 if (S_ISDIR(inode->i_mode))
943 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
944 spin_unlock(&inode->i_lock);
945 return nfs_refresh_inode(inode, fattr);
949 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
950 * @inode - pointer to inode
951 * @fattr - updated attributes
953 * After an operation that has changed the inode metadata, mark the
954 * attribute cache as being invalid, then try to update it. Fake up
955 * weak cache consistency data, if none exist.
957 * This function is mainly designed to be used by the ->write_done() functions.
959 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
961 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
962 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
963 fattr->pre_change_attr = NFS_I(inode)->change_attr;
964 fattr->valid |= NFS_ATTR_WCC_V4;
966 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
967 (fattr->valid & NFS_ATTR_WCC) == 0) {
968 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
969 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
970 fattr->pre_size = inode->i_size;
971 fattr->valid |= NFS_ATTR_WCC;
973 return nfs_post_op_update_inode(inode, fattr);
977 * Many nfs protocol calls return the new file attributes after
978 * an operation. Here we update the inode to reflect the state
979 * of the server's inode.
981 * This is a bit tricky because we have to make sure all dirty pages
982 * have been sent off to the server before calling invalidate_inode_pages.
983 * To make sure no other process adds more write requests while we try
984 * our best to flush them, we make them sleep during the attribute refresh.
986 * A very similar scenario holds for the dir cache.
988 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
990 struct nfs_server *server;
991 struct nfs_inode *nfsi = NFS_I(inode);
992 loff_t cur_isize, new_isize;
993 unsigned long invalid = 0;
994 unsigned long now = jiffies;
996 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
997 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
998 atomic_read(&inode->i_count), fattr->valid);
1000 if (nfsi->fileid != fattr->fileid)
1004 * Make sure the inode's type hasn't changed.
1006 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1009 server = NFS_SERVER(inode);
1010 /* Update the fsid? */
1011 if (S_ISDIR(inode->i_mode) &&
1012 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1013 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1014 server->fsid = fattr->fsid;
1017 * Update the read time so we don't revalidate too often.
1019 nfsi->read_cache_jiffies = fattr->time_start;
1021 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1022 | NFS_INO_REVAL_PAGECACHE);
1024 /* Do atomic weak cache consistency updates */
1025 nfs_wcc_update_inode(inode, fattr);
1027 /* More cache consistency checks */
1028 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1029 /* NFSv2/v3: Check if the mtime agrees */
1030 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1031 dprintk("NFS: mtime change on server for file %s/%ld\n",
1032 inode->i_sb->s_id, inode->i_ino);
1033 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1034 if (S_ISDIR(inode->i_mode))
1035 nfs_force_lookup_revalidate(inode);
1037 /* If ctime has changed we should definitely clear access+acl caches */
1038 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
1039 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1040 } else if (nfsi->change_attr != fattr->change_attr) {
1041 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1042 inode->i_sb->s_id, inode->i_ino);
1043 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1044 if (S_ISDIR(inode->i_mode))
1045 nfs_force_lookup_revalidate(inode);
1048 /* Check if our cached file size is stale */
1049 new_isize = nfs_size_to_loff_t(fattr->size);
1050 cur_isize = i_size_read(inode);
1051 if (new_isize != cur_isize) {
1052 /* Do we perhaps have any outstanding writes, or has
1053 * the file grown beyond our last write? */
1054 if (nfsi->npages == 0 || new_isize > cur_isize) {
1055 inode->i_size = new_isize;
1056 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1058 dprintk("NFS: isize change on server for file %s/%ld\n",
1059 inode->i_sb->s_id, inode->i_ino);
1063 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1064 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1065 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1066 nfsi->change_attr = fattr->change_attr;
1068 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1069 inode->i_uid != fattr->uid ||
1070 inode->i_gid != fattr->gid)
1071 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1073 inode->i_mode = fattr->mode;
1074 inode->i_nlink = fattr->nlink;
1075 inode->i_uid = fattr->uid;
1076 inode->i_gid = fattr->gid;
1078 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1080 * report the blocks in 512byte units
1082 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1084 inode->i_blocks = fattr->du.nfs2.blocks;
1087 /* Update attrtimeo value if we're out of the unstable period */
1088 if (invalid & NFS_INO_INVALID_ATTR) {
1089 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1090 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1091 nfsi->attrtimeo_timestamp = now;
1092 nfsi->last_updated = now;
1094 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1095 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1096 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1097 nfsi->attrtimeo_timestamp = now;
1100 * Avoid jiffy wraparound issues with nfsi->last_updated
1102 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1103 nfsi->last_updated = nfsi->read_cache_jiffies;
1105 invalid &= ~NFS_INO_INVALID_ATTR;
1106 /* Don't invalidate the data if we were to blame */
1107 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1108 || S_ISLNK(inode->i_mode)))
1109 invalid &= ~NFS_INO_INVALID_DATA;
1110 if (!nfs_have_delegation(inode, FMODE_READ) ||
1111 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1112 nfsi->cache_validity |= invalid;
1113 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1118 * Big trouble! The inode has become a different object.
1120 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1121 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1124 * No need to worry about unhashing the dentry, as the
1125 * lookup validation will know that the inode is bad.
1126 * (But we fall through to invalidate the caches.)
1128 nfs_invalidate_inode(inode);
1132 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1133 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1134 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1135 (long long)nfsi->fileid, (long long)fattr->fileid);
1140 #ifdef CONFIG_NFS_V4
1143 * Clean out any remaining NFSv4 state that might be left over due
1144 * to open() calls that passed nfs_atomic_lookup, but failed to call
1147 void nfs4_clear_inode(struct inode *inode)
1149 /* If we are holding a delegation, return it! */
1150 nfs_inode_return_delegation_noreclaim(inode);
1151 /* First call standard NFS clear_inode() code */
1152 nfs_clear_inode(inode);
1156 struct inode *nfs_alloc_inode(struct super_block *sb)
1158 struct nfs_inode *nfsi;
1159 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1163 nfsi->cache_validity = 0UL;
1164 #ifdef CONFIG_NFS_V3_ACL
1165 nfsi->acl_access = ERR_PTR(-EAGAIN);
1166 nfsi->acl_default = ERR_PTR(-EAGAIN);
1168 #ifdef CONFIG_NFS_V4
1169 nfsi->nfs4_acl = NULL;
1170 #endif /* CONFIG_NFS_V4 */
1171 return &nfsi->vfs_inode;
1174 void nfs_destroy_inode(struct inode *inode)
1176 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1179 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1181 #ifdef CONFIG_NFS_V4
1182 INIT_LIST_HEAD(&nfsi->open_states);
1183 nfsi->delegation = NULL;
1184 nfsi->delegation_state = 0;
1185 init_rwsem(&nfsi->rwsem);
1189 static void init_once(struct kmem_cache * cachep, void *foo)
1191 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1193 inode_init_once(&nfsi->vfs_inode);
1194 INIT_LIST_HEAD(&nfsi->open_files);
1195 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1196 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1197 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1200 atomic_set(&nfsi->silly_count, 1);
1201 INIT_HLIST_HEAD(&nfsi->silly_list);
1202 init_waitqueue_head(&nfsi->waitqueue);
1203 nfs4_init_once(nfsi);
1206 static int __init nfs_init_inodecache(void)
1208 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1209 sizeof(struct nfs_inode),
1210 0, (SLAB_RECLAIM_ACCOUNT|
1213 if (nfs_inode_cachep == NULL)
1219 static void nfs_destroy_inodecache(void)
1221 kmem_cache_destroy(nfs_inode_cachep);
1224 struct workqueue_struct *nfsiod_workqueue;
1227 * start up the nfsiod workqueue
1229 static int nfsiod_start(void)
1231 struct workqueue_struct *wq;
1232 dprintk("RPC: creating workqueue nfsiod\n");
1233 wq = create_singlethread_workqueue("nfsiod");
1236 nfsiod_workqueue = wq;
1241 * Destroy the nfsiod workqueue
1243 static void nfsiod_stop(void)
1245 struct workqueue_struct *wq;
1247 wq = nfsiod_workqueue;
1250 nfsiod_workqueue = NULL;
1251 destroy_workqueue(wq);
1257 static int __init init_nfs_fs(void)
1261 err = nfsiod_start();
1265 err = nfs_fs_proc_init();
1269 err = nfs_init_nfspagecache();
1273 err = nfs_init_inodecache();
1277 err = nfs_init_readpagecache();
1281 err = nfs_init_writepagecache();
1285 err = nfs_init_directcache();
1289 #ifdef CONFIG_PROC_FS
1290 rpc_proc_register(&nfs_rpcstat);
1292 if ((err = register_nfs_fs()) != 0)
1296 #ifdef CONFIG_PROC_FS
1297 rpc_proc_unregister("nfs");
1299 nfs_destroy_directcache();
1301 nfs_destroy_writepagecache();
1303 nfs_destroy_readpagecache();
1305 nfs_destroy_inodecache();
1307 nfs_destroy_nfspagecache();
1316 static void __exit exit_nfs_fs(void)
1318 nfs_destroy_directcache();
1319 nfs_destroy_writepagecache();
1320 nfs_destroy_readpagecache();
1321 nfs_destroy_inodecache();
1322 nfs_destroy_nfspagecache();
1323 #ifdef CONFIG_PROC_FS
1324 rpc_proc_unregister("nfs");
1326 unregister_nfs_fs();
1331 /* Not quite true; I just maintain it */
1332 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1333 MODULE_LICENSE("GPL");
1334 module_param(enable_ino64, bool, 0644);
1336 module_init(init_nfs_fs)
1337 module_exit(exit_nfs_fs)