4 * Copyright (C) 1992 Rick Sladkey
6 * nfs inode and superblock handling functions
8 * Modularised by Alan Cox <alan@lxorguk.ukuu.org.uk>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
19 #include <linux/time.h>
20 #include <linux/kernel.h>
22 #include <linux/string.h>
23 #include <linux/stat.h>
24 #include <linux/errno.h>
25 #include <linux/unistd.h>
26 #include <linux/sunrpc/clnt.h>
27 #include <linux/sunrpc/stats.h>
28 #include <linux/sunrpc/metrics.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/nfs4_mount.h>
32 #include <linux/lockd/bind.h>
33 #include <linux/smp_lock.h>
34 #include <linux/seq_file.h>
35 #include <linux/mount.h>
36 #include <linux/nfs_idmap.h>
37 #include <linux/vfs.h>
38 #include <linux/inet.h>
39 #include <linux/nfs_xdr.h>
41 #include <asm/system.h>
42 #include <asm/uaccess.h>
46 #include "delegation.h"
50 #define NFSDBG_FACILITY NFSDBG_VFS
52 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
54 /* Default is to see 64-bit inode numbers */
55 static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
57 static void nfs_invalidate_inode(struct inode *);
58 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
60 static struct kmem_cache * nfs_inode_cachep;
62 static inline unsigned long
63 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
65 return nfs_fileid_to_ino_t(fattr->fileid);
69 * nfs_compat_user_ino64 - returns the user-visible inode number
70 * @fileid: 64-bit fileid
72 * This function returns a 32-bit inode number if the boot parameter
73 * nfs.enable_ino64 is zero.
75 u64 nfs_compat_user_ino64(u64 fileid)
82 if (sizeof(ino) < sizeof(fileid))
83 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
87 int nfs_write_inode(struct inode *inode, int sync)
92 ret = filemap_fdatawait(inode->i_mapping);
94 ret = nfs_commit_inode(inode, FLUSH_SYNC);
96 ret = nfs_commit_inode(inode, 0);
99 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
103 void nfs_clear_inode(struct inode *inode)
106 * The following should never happen...
108 BUG_ON(nfs_have_writebacks(inode));
109 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
110 nfs_zap_acl_cache(inode);
111 nfs_access_zap_cache(inode);
115 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
117 int nfs_sync_mapping(struct address_space *mapping)
121 if (mapping->nrpages == 0)
123 unmap_mapping_range(mapping, 0, 0, 0);
124 ret = filemap_write_and_wait(mapping);
127 ret = nfs_wb_all(mapping->host);
133 * Invalidate the local caches
135 static void nfs_zap_caches_locked(struct inode *inode)
137 struct nfs_inode *nfsi = NFS_I(inode);
138 int mode = inode->i_mode;
140 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
142 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
143 nfsi->attrtimeo_timestamp = jiffies;
145 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
146 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
147 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
149 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
152 void nfs_zap_caches(struct inode *inode)
154 spin_lock(&inode->i_lock);
155 nfs_zap_caches_locked(inode);
156 spin_unlock(&inode->i_lock);
159 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
161 if (mapping->nrpages != 0) {
162 spin_lock(&inode->i_lock);
163 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
164 spin_unlock(&inode->i_lock);
168 void nfs_zap_acl_cache(struct inode *inode)
170 void (*clear_acl_cache)(struct inode *);
172 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
173 if (clear_acl_cache != NULL)
174 clear_acl_cache(inode);
175 spin_lock(&inode->i_lock);
176 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
177 spin_unlock(&inode->i_lock);
180 void nfs_invalidate_atime(struct inode *inode)
182 spin_lock(&inode->i_lock);
183 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
184 spin_unlock(&inode->i_lock);
188 * Invalidate, but do not unhash, the inode.
189 * NB: must be called with inode->i_lock held!
191 static void nfs_invalidate_inode(struct inode *inode)
193 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
194 nfs_zap_caches_locked(inode);
197 struct nfs_find_desc {
199 struct nfs_fattr *fattr;
203 * In NFSv3 we can have 64bit inode numbers. In order to support
204 * this, and re-exported directories (also seen in NFSv2)
205 * we are forced to allow 2 different inodes to have the same
209 nfs_find_actor(struct inode *inode, void *opaque)
211 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
212 struct nfs_fh *fh = desc->fh;
213 struct nfs_fattr *fattr = desc->fattr;
215 if (NFS_FILEID(inode) != fattr->fileid)
217 if (nfs_compare_fh(NFS_FH(inode), fh))
219 if (is_bad_inode(inode) || NFS_STALE(inode))
225 nfs_init_locked(struct inode *inode, void *opaque)
227 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
228 struct nfs_fattr *fattr = desc->fattr;
230 set_nfs_fileid(inode, fattr->fileid);
231 nfs_copy_fh(NFS_FH(inode), desc->fh);
235 /* Don't use READDIRPLUS on directories that we believe are too large */
236 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
239 * This is our front-end to iget that looks up inodes by file handle
240 * instead of inode number.
243 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
245 struct nfs_find_desc desc = {
249 struct inode *inode = ERR_PTR(-ENOENT);
252 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
256 printk("NFS: Buggy server - nlink == 0!\n");
260 hash = nfs_fattr_to_ino_t(fattr);
262 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
264 inode = ERR_PTR(-ENOMEM);
268 if (inode->i_state & I_NEW) {
269 struct nfs_inode *nfsi = NFS_I(inode);
270 unsigned long now = jiffies;
272 /* We set i_ino for the few things that still rely on it,
276 /* We can't support update_atime(), since the server will reset it */
277 inode->i_flags |= S_NOATIME|S_NOCMTIME;
278 inode->i_mode = fattr->mode;
279 /* Why so? Because we want revalidate for devices/FIFOs, and
280 * that's precisely what we have in nfs_file_inode_operations.
282 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
283 if (S_ISREG(inode->i_mode)) {
284 inode->i_fop = &nfs_file_operations;
285 inode->i_data.a_ops = &nfs_file_aops;
286 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
287 } else if (S_ISDIR(inode->i_mode)) {
288 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
289 inode->i_fop = &nfs_dir_operations;
290 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
291 && fattr->size <= NFS_LIMIT_READDIRPLUS)
292 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
293 /* Deal with crossing mountpoints */
294 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
295 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
296 inode->i_op = &nfs_referral_inode_operations;
298 inode->i_op = &nfs_mountpoint_inode_operations;
300 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
302 } else if (S_ISLNK(inode->i_mode))
303 inode->i_op = &nfs_symlink_inode_operations;
305 init_special_inode(inode, inode->i_mode, fattr->rdev);
307 nfsi->read_cache_jiffies = fattr->time_start;
308 nfsi->attr_gencount = fattr->gencount;
309 inode->i_atime = fattr->atime;
310 inode->i_mtime = fattr->mtime;
311 inode->i_ctime = fattr->ctime;
312 if (fattr->valid & NFS_ATTR_FATTR_V4)
313 nfsi->change_attr = fattr->change_attr;
314 inode->i_size = nfs_size_to_loff_t(fattr->size);
315 inode->i_nlink = fattr->nlink;
316 inode->i_uid = fattr->uid;
317 inode->i_gid = fattr->gid;
318 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
320 * report the blocks in 512byte units
322 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
324 inode->i_blocks = fattr->du.nfs2.blocks;
326 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
327 nfsi->attrtimeo_timestamp = now;
328 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
329 nfsi->access_cache = RB_ROOT;
331 unlock_new_inode(inode);
333 nfs_refresh_inode(inode, fattr);
334 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
336 (long long)NFS_FILEID(inode),
337 atomic_read(&inode->i_count));
343 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
347 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
350 nfs_setattr(struct dentry *dentry, struct iattr *attr)
352 struct inode *inode = dentry->d_inode;
353 struct nfs_fattr fattr;
356 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
358 /* skip mode change if it's just for clearing setuid/setgid */
359 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
360 attr->ia_valid &= ~ATTR_MODE;
362 if (attr->ia_valid & ATTR_SIZE) {
363 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
364 attr->ia_valid &= ~ATTR_SIZE;
367 /* Optimization: if the end result is no change, don't RPC */
368 attr->ia_valid &= NFS_VALID_ATTRS;
369 if ((attr->ia_valid & ~ATTR_FILE) == 0)
372 /* Write all dirty data */
373 if (S_ISREG(inode->i_mode)) {
374 filemap_write_and_wait(inode->i_mapping);
378 * Return any delegations if we're going to change ACLs
380 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
381 nfs_inode_return_delegation(inode);
382 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
384 nfs_refresh_inode(inode, &fattr);
389 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
390 * @inode: inode of the file used
391 * @offset: file offset to start truncating
393 * This is a copy of the common vmtruncate, but with the locking
394 * corrected to take into account the fact that NFS requires
395 * inode->i_size to be updated under the inode->i_lock.
397 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
399 if (i_size_read(inode) < offset) {
402 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
403 if (limit != RLIM_INFINITY && offset > limit)
405 if (offset > inode->i_sb->s_maxbytes)
407 spin_lock(&inode->i_lock);
408 i_size_write(inode, offset);
409 spin_unlock(&inode->i_lock);
411 struct address_space *mapping = inode->i_mapping;
414 * truncation of in-use swapfiles is disallowed - it would
415 * cause subsequent swapout to scribble on the now-freed
418 if (IS_SWAPFILE(inode))
420 spin_lock(&inode->i_lock);
421 i_size_write(inode, offset);
422 spin_unlock(&inode->i_lock);
425 * unmap_mapping_range is called twice, first simply for
426 * efficiency so that truncate_inode_pages does fewer
427 * single-page unmaps. However after this first call, and
428 * before truncate_inode_pages finishes, it is possible for
429 * private pages to be COWed, which remain after
430 * truncate_inode_pages finishes, hence the second
431 * unmap_mapping_range call must be made for correctness.
433 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
434 truncate_inode_pages(mapping, offset);
435 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
439 send_sig(SIGXFSZ, current, 0);
445 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
446 * @inode: pointer to struct inode
447 * @attr: pointer to struct iattr
449 * Note: we do this in the *proc.c in order to ensure that
450 * it works for things like exclusive creates too.
452 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
454 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
455 spin_lock(&inode->i_lock);
456 if ((attr->ia_valid & ATTR_MODE) != 0) {
457 int mode = attr->ia_mode & S_IALLUGO;
458 mode |= inode->i_mode & ~S_IALLUGO;
459 inode->i_mode = mode;
461 if ((attr->ia_valid & ATTR_UID) != 0)
462 inode->i_uid = attr->ia_uid;
463 if ((attr->ia_valid & ATTR_GID) != 0)
464 inode->i_gid = attr->ia_gid;
465 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
466 spin_unlock(&inode->i_lock);
468 if ((attr->ia_valid & ATTR_SIZE) != 0) {
469 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
470 nfs_vmtruncate(inode, attr->ia_size);
474 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
476 struct inode *inode = dentry->d_inode;
477 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
481 * Flush out writes to the server in order to update c/mtime.
483 * Hold the i_mutex to suspend application writes temporarily;
484 * this prevents long-running writing applications from blocking
487 if (S_ISREG(inode->i_mode)) {
488 mutex_lock(&inode->i_mutex);
489 nfs_wb_nocommit(inode);
490 mutex_unlock(&inode->i_mutex);
494 * We may force a getattr if the user cares about atime.
496 * Note that we only have to check the vfsmount flags here:
497 * - NFS always sets S_NOATIME by so checking it would give a
499 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
500 * no point in checking those.
502 if ((mnt->mnt_flags & MNT_NOATIME) ||
503 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
507 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
509 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
511 generic_fillattr(inode, stat);
512 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
517 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
519 struct nfs_open_context *ctx;
521 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
523 ctx->path.dentry = dget(dentry);
524 ctx->path.mnt = mntget(mnt);
525 ctx->cred = get_rpccred(cred);
527 ctx->lockowner = current->files;
531 atomic_set(&ctx->count, 1);
536 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
539 atomic_inc(&ctx->count);
543 static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
550 inode = ctx->path.dentry->d_inode;
551 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
553 list_del(&ctx->list);
554 spin_unlock(&inode->i_lock);
555 if (ctx->state != NULL) {
557 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
559 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
561 if (ctx->cred != NULL)
562 put_rpccred(ctx->cred);
563 path_put(&ctx->path);
567 void put_nfs_open_context(struct nfs_open_context *ctx)
569 __put_nfs_open_context(ctx, 0);
572 static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
574 __put_nfs_open_context(ctx, 1);
578 * Ensure that mmap has a recent RPC credential for use when writing out
581 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
583 struct inode *inode = filp->f_path.dentry->d_inode;
584 struct nfs_inode *nfsi = NFS_I(inode);
586 filp->private_data = get_nfs_open_context(ctx);
587 spin_lock(&inode->i_lock);
588 list_add(&ctx->list, &nfsi->open_files);
589 spin_unlock(&inode->i_lock);
593 * Given an inode, search for an open context with the desired characteristics
595 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
597 struct nfs_inode *nfsi = NFS_I(inode);
598 struct nfs_open_context *pos, *ctx = NULL;
600 spin_lock(&inode->i_lock);
601 list_for_each_entry(pos, &nfsi->open_files, list) {
602 if (cred != NULL && pos->cred != cred)
604 if ((pos->mode & mode) == mode) {
605 ctx = get_nfs_open_context(pos);
609 spin_unlock(&inode->i_lock);
613 static void nfs_file_clear_open_context(struct file *filp)
615 struct inode *inode = filp->f_path.dentry->d_inode;
616 struct nfs_open_context *ctx = nfs_file_open_context(filp);
619 filp->private_data = NULL;
620 spin_lock(&inode->i_lock);
621 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
622 spin_unlock(&inode->i_lock);
623 put_nfs_open_context_sync(ctx);
628 * These allocate and release file read/write context information.
630 int nfs_open(struct inode *inode, struct file *filp)
632 struct nfs_open_context *ctx;
633 struct rpc_cred *cred;
635 cred = rpc_lookup_cred();
637 return PTR_ERR(cred);
638 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
642 ctx->mode = filp->f_mode;
643 nfs_file_set_open_context(filp, ctx);
644 put_nfs_open_context(ctx);
648 int nfs_release(struct inode *inode, struct file *filp)
650 nfs_file_clear_open_context(filp);
655 * This function is called whenever some part of NFS notices that
656 * the cached attributes have to be refreshed.
659 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
661 int status = -ESTALE;
662 struct nfs_fattr fattr;
663 struct nfs_inode *nfsi = NFS_I(inode);
665 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
666 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
668 if (is_bad_inode(inode))
670 if (NFS_STALE(inode))
673 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
674 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
676 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
678 (long long)NFS_FILEID(inode), status);
679 if (status == -ESTALE) {
680 nfs_zap_caches(inode);
681 if (!S_ISDIR(inode->i_mode))
682 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
687 status = nfs_refresh_inode(inode, &fattr);
689 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
691 (long long)NFS_FILEID(inode), status);
695 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
696 nfs_zap_acl_cache(inode);
698 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
700 (long long)NFS_FILEID(inode));
706 int nfs_attribute_timeout(struct inode *inode)
708 struct nfs_inode *nfsi = NFS_I(inode);
710 if (nfs_have_delegation(inode, FMODE_READ))
712 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
716 * nfs_revalidate_inode - Revalidate the inode attributes
717 * @server - pointer to nfs_server struct
718 * @inode - pointer to inode struct
720 * Updates inode attribute information by retrieving the data from the server.
722 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
724 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
725 && !nfs_attribute_timeout(inode))
726 return NFS_STALE(inode) ? -ESTALE : 0;
727 return __nfs_revalidate_inode(server, inode);
730 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
732 struct nfs_inode *nfsi = NFS_I(inode);
734 if (mapping->nrpages != 0) {
735 int ret = invalidate_inode_pages2(mapping);
739 spin_lock(&inode->i_lock);
740 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
741 if (S_ISDIR(inode->i_mode))
742 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
743 spin_unlock(&inode->i_lock);
744 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
745 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
746 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
750 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
754 mutex_lock(&inode->i_mutex);
755 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
756 ret = nfs_sync_mapping(mapping);
758 ret = nfs_invalidate_mapping_nolock(inode, mapping);
760 mutex_unlock(&inode->i_mutex);
765 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
766 * @inode - pointer to host inode
767 * @mapping - pointer to mapping
769 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
771 struct nfs_inode *nfsi = NFS_I(inode);
774 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
775 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
776 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
780 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
781 ret = nfs_invalidate_mapping_nolock(inode, mapping);
787 * nfs_revalidate_mapping - Revalidate the pagecache
788 * @inode - pointer to host inode
789 * @mapping - pointer to mapping
791 * This version of the function will take the inode->i_mutex and attempt to
792 * flush out all dirty data if it needs to invalidate the page cache.
794 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
796 struct nfs_inode *nfsi = NFS_I(inode);
799 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
800 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
801 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
805 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
806 ret = nfs_invalidate_mapping(inode, mapping);
811 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
813 struct nfs_inode *nfsi = NFS_I(inode);
815 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
816 nfsi->change_attr == fattr->pre_change_attr) {
817 nfsi->change_attr = fattr->change_attr;
818 if (S_ISDIR(inode->i_mode))
819 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
821 /* If we have atomic WCC data, we may update some attributes */
822 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
823 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
824 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
825 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
826 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
827 if (S_ISDIR(inode->i_mode))
828 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
830 if (i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) &&
832 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
837 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
838 * @inode - pointer to inode
839 * @fattr - updated attributes
841 * Verifies the attribute cache. If we have just changed the attributes,
842 * so that fattr carries weak cache consistency data, then it may
843 * also update the ctime/mtime/change_attribute.
845 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
847 struct nfs_inode *nfsi = NFS_I(inode);
848 loff_t cur_size, new_isize;
849 unsigned long invalid = 0;
852 /* Has the inode gone and changed behind our back? */
853 if (nfsi->fileid != fattr->fileid
854 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
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->read_cache_jiffies = fattr->time_start;
891 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
893 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
896 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
898 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
901 static atomic_long_t nfs_attr_generation_counter;
903 static unsigned long nfs_read_attr_generation_counter(void)
905 return atomic_long_read(&nfs_attr_generation_counter);
908 unsigned long nfs_inc_attr_generation_counter(void)
910 return atomic_long_inc_return(&nfs_attr_generation_counter);
913 void nfs_fattr_init(struct nfs_fattr *fattr)
916 fattr->time_start = jiffies;
917 fattr->gencount = nfs_inc_attr_generation_counter();
921 * nfs_inode_attrs_need_update - check if the inode attributes need updating
922 * @inode - pointer to inode
923 * @fattr - attributes
925 * Attempt to divine whether or not an RPC call reply carrying stale
926 * attributes got scheduled after another call carrying updated ones.
928 * To do so, the function first assumes that a more recent ctime means
929 * that the attributes in fattr are newer, however it also attempt to
930 * catch the case where ctime either didn't change, or went backwards
931 * (if someone reset the clock on the server) by looking at whether
932 * or not this RPC call was started after the inode was last updated.
933 * Note also the check for wraparound of 'attr_gencount'
935 * The function returns 'true' if it thinks the attributes in 'fattr' are
936 * more recent than the ones cached in the inode.
939 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
941 const struct nfs_inode *nfsi = NFS_I(inode);
943 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
944 nfs_ctime_need_update(inode, fattr) ||
945 nfs_size_need_update(inode, fattr) ||
946 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
949 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
951 if (nfs_inode_attrs_need_update(inode, fattr))
952 return nfs_update_inode(inode, fattr);
953 return nfs_check_inode_attributes(inode, fattr);
957 * nfs_refresh_inode - try to update the inode attribute cache
958 * @inode - pointer to inode
959 * @fattr - updated attributes
961 * Check that an RPC call that returned attributes has not overlapped with
962 * other recent updates of the inode metadata, then decide whether it is
963 * safe to do a full update of the inode attributes, or whether just to
964 * call nfs_check_inode_attributes.
966 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
970 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
972 spin_lock(&inode->i_lock);
973 status = nfs_refresh_inode_locked(inode, fattr);
974 spin_unlock(&inode->i_lock);
978 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
980 struct nfs_inode *nfsi = NFS_I(inode);
982 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
983 if (S_ISDIR(inode->i_mode))
984 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
985 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
987 return nfs_refresh_inode_locked(inode, fattr);
991 * nfs_post_op_update_inode - try to update the inode attribute cache
992 * @inode - pointer to inode
993 * @fattr - updated attributes
995 * After an operation that has changed the inode metadata, mark the
996 * attribute cache as being invalid, then try to update it.
998 * NB: if the server didn't return any post op attributes, this
999 * function will force the retrieval of attributes before the next
1000 * NFS request. Thus it should be used only for operations that
1001 * are expected to change one or more attributes, to avoid
1002 * unnecessary NFS requests and trips through nfs_update_inode().
1004 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1008 spin_lock(&inode->i_lock);
1009 status = nfs_post_op_update_inode_locked(inode, fattr);
1010 spin_unlock(&inode->i_lock);
1015 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1016 * @inode - pointer to inode
1017 * @fattr - updated attributes
1019 * After an operation that has changed the inode metadata, mark the
1020 * attribute cache as being invalid, then try to update it. Fake up
1021 * weak cache consistency data, if none exist.
1023 * This function is mainly designed to be used by the ->write_done() functions.
1025 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1029 spin_lock(&inode->i_lock);
1030 /* Don't do a WCC update if these attributes are already stale */
1031 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1032 !nfs_inode_attrs_need_update(inode, fattr)) {
1033 fattr->valid &= ~(NFS_ATTR_WCC_V4|NFS_ATTR_WCC);
1036 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1037 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
1038 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1039 fattr->valid |= NFS_ATTR_WCC_V4;
1041 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
1042 (fattr->valid & NFS_ATTR_WCC) == 0) {
1043 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1044 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1045 fattr->pre_size = i_size_read(inode);
1046 fattr->valid |= NFS_ATTR_WCC;
1049 status = nfs_post_op_update_inode_locked(inode, fattr);
1050 spin_unlock(&inode->i_lock);
1055 * Many nfs protocol calls return the new file attributes after
1056 * an operation. Here we update the inode to reflect the state
1057 * of the server's inode.
1059 * This is a bit tricky because we have to make sure all dirty pages
1060 * have been sent off to the server before calling invalidate_inode_pages.
1061 * To make sure no other process adds more write requests while we try
1062 * our best to flush them, we make them sleep during the attribute refresh.
1064 * A very similar scenario holds for the dir cache.
1066 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1068 struct nfs_server *server;
1069 struct nfs_inode *nfsi = NFS_I(inode);
1070 loff_t cur_isize, new_isize;
1071 unsigned long invalid = 0;
1072 unsigned long now = jiffies;
1074 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
1075 __func__, inode->i_sb->s_id, inode->i_ino,
1076 atomic_read(&inode->i_count), fattr->valid);
1078 if (nfsi->fileid != fattr->fileid)
1082 * Make sure the inode's type hasn't changed.
1084 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1087 server = NFS_SERVER(inode);
1088 /* Update the fsid? */
1089 if (S_ISDIR(inode->i_mode) &&
1090 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1091 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
1092 server->fsid = fattr->fsid;
1095 * Update the read time so we don't revalidate too often.
1097 nfsi->read_cache_jiffies = fattr->time_start;
1099 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1100 | NFS_INO_REVAL_PAGECACHE);
1102 /* Do atomic weak cache consistency updates */
1103 nfs_wcc_update_inode(inode, fattr);
1105 /* More cache consistency checks */
1106 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1107 /* NFSv2/v3: Check if the mtime agrees */
1108 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1109 dprintk("NFS: mtime change on server for file %s/%ld\n",
1110 inode->i_sb->s_id, inode->i_ino);
1111 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1112 if (S_ISDIR(inode->i_mode))
1113 nfs_force_lookup_revalidate(inode);
1115 /* If ctime has changed we should definitely clear access+acl caches */
1116 if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
1117 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1118 /* and probably clear data for a directory too as utimes can cause
1119 * havoc with our cache.
1121 if (S_ISDIR(inode->i_mode)) {
1122 invalid |= NFS_INO_INVALID_DATA;
1123 nfs_force_lookup_revalidate(inode);
1126 } else if (nfsi->change_attr != fattr->change_attr) {
1127 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1128 inode->i_sb->s_id, inode->i_ino);
1129 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1130 if (S_ISDIR(inode->i_mode))
1131 nfs_force_lookup_revalidate(inode);
1134 /* Check if our cached file size is stale */
1135 new_isize = nfs_size_to_loff_t(fattr->size);
1136 cur_isize = i_size_read(inode);
1137 if (new_isize != cur_isize) {
1138 /* Do we perhaps have any outstanding writes, or has
1139 * the file grown beyond our last write? */
1140 if (nfsi->npages == 0 || new_isize > cur_isize) {
1141 i_size_write(inode, new_isize);
1142 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1144 dprintk("NFS: isize change on server for file %s/%ld\n",
1145 inode->i_sb->s_id, inode->i_ino);
1149 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1150 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1151 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1152 nfsi->change_attr = fattr->change_attr;
1154 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1155 inode->i_uid != fattr->uid ||
1156 inode->i_gid != fattr->gid)
1157 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1159 if (inode->i_nlink != fattr->nlink) {
1160 invalid |= NFS_INO_INVALID_ATTR;
1161 if (S_ISDIR(inode->i_mode))
1162 invalid |= NFS_INO_INVALID_DATA;
1165 inode->i_mode = fattr->mode;
1166 inode->i_nlink = fattr->nlink;
1167 inode->i_uid = fattr->uid;
1168 inode->i_gid = fattr->gid;
1170 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1172 * report the blocks in 512byte units
1174 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1176 inode->i_blocks = fattr->du.nfs2.blocks;
1179 /* Update attrtimeo value if we're out of the unstable period */
1180 if (invalid & NFS_INO_INVALID_ATTR) {
1181 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1182 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1183 nfsi->attrtimeo_timestamp = now;
1184 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1186 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1187 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1188 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1189 nfsi->attrtimeo_timestamp = now;
1192 invalid &= ~NFS_INO_INVALID_ATTR;
1193 /* Don't invalidate the data if we were to blame */
1194 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1195 || S_ISLNK(inode->i_mode)))
1196 invalid &= ~NFS_INO_INVALID_DATA;
1197 if (!nfs_have_delegation(inode, FMODE_READ) ||
1198 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1199 nfsi->cache_validity |= invalid;
1200 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1205 * Big trouble! The inode has become a different object.
1207 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1208 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1211 * No need to worry about unhashing the dentry, as the
1212 * lookup validation will know that the inode is bad.
1213 * (But we fall through to invalidate the caches.)
1215 nfs_invalidate_inode(inode);
1219 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1220 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1221 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1222 (long long)nfsi->fileid, (long long)fattr->fileid);
1227 #ifdef CONFIG_NFS_V4
1230 * Clean out any remaining NFSv4 state that might be left over due
1231 * to open() calls that passed nfs_atomic_lookup, but failed to call
1234 void nfs4_clear_inode(struct inode *inode)
1236 /* If we are holding a delegation, return it! */
1237 nfs_inode_return_delegation_noreclaim(inode);
1238 /* First call standard NFS clear_inode() code */
1239 nfs_clear_inode(inode);
1243 struct inode *nfs_alloc_inode(struct super_block *sb)
1245 struct nfs_inode *nfsi;
1246 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1250 nfsi->cache_validity = 0UL;
1251 #ifdef CONFIG_NFS_V3_ACL
1252 nfsi->acl_access = ERR_PTR(-EAGAIN);
1253 nfsi->acl_default = ERR_PTR(-EAGAIN);
1255 #ifdef CONFIG_NFS_V4
1256 nfsi->nfs4_acl = NULL;
1257 #endif /* CONFIG_NFS_V4 */
1258 return &nfsi->vfs_inode;
1261 void nfs_destroy_inode(struct inode *inode)
1263 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1266 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1268 #ifdef CONFIG_NFS_V4
1269 INIT_LIST_HEAD(&nfsi->open_states);
1270 nfsi->delegation = NULL;
1271 nfsi->delegation_state = 0;
1272 init_rwsem(&nfsi->rwsem);
1276 static void init_once(void *foo)
1278 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1280 inode_init_once(&nfsi->vfs_inode);
1281 INIT_LIST_HEAD(&nfsi->open_files);
1282 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1283 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1284 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1287 atomic_set(&nfsi->silly_count, 1);
1288 INIT_HLIST_HEAD(&nfsi->silly_list);
1289 init_waitqueue_head(&nfsi->waitqueue);
1290 nfs4_init_once(nfsi);
1293 static int __init nfs_init_inodecache(void)
1295 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1296 sizeof(struct nfs_inode),
1297 0, (SLAB_RECLAIM_ACCOUNT|
1300 if (nfs_inode_cachep == NULL)
1306 static void nfs_destroy_inodecache(void)
1308 kmem_cache_destroy(nfs_inode_cachep);
1311 struct workqueue_struct *nfsiod_workqueue;
1314 * start up the nfsiod workqueue
1316 static int nfsiod_start(void)
1318 struct workqueue_struct *wq;
1319 dprintk("RPC: creating workqueue nfsiod\n");
1320 wq = create_singlethread_workqueue("nfsiod");
1323 nfsiod_workqueue = wq;
1328 * Destroy the nfsiod workqueue
1330 static void nfsiod_stop(void)
1332 struct workqueue_struct *wq;
1334 wq = nfsiod_workqueue;
1337 nfsiod_workqueue = NULL;
1338 destroy_workqueue(wq);
1344 static int __init init_nfs_fs(void)
1348 err = nfsiod_start();
1352 err = nfs_fs_proc_init();
1356 err = nfs_init_nfspagecache();
1360 err = nfs_init_inodecache();
1364 err = nfs_init_readpagecache();
1368 err = nfs_init_writepagecache();
1372 err = nfs_init_directcache();
1376 #ifdef CONFIG_PROC_FS
1377 rpc_proc_register(&nfs_rpcstat);
1379 if ((err = register_nfs_fs()) != 0)
1383 #ifdef CONFIG_PROC_FS
1384 rpc_proc_unregister("nfs");
1386 nfs_destroy_directcache();
1388 nfs_destroy_writepagecache();
1390 nfs_destroy_readpagecache();
1392 nfs_destroy_inodecache();
1394 nfs_destroy_nfspagecache();
1403 static void __exit exit_nfs_fs(void)
1405 nfs_destroy_directcache();
1406 nfs_destroy_writepagecache();
1407 nfs_destroy_readpagecache();
1408 nfs_destroy_inodecache();
1409 nfs_destroy_nfspagecache();
1410 #ifdef CONFIG_PROC_FS
1411 rpc_proc_unregister("nfs");
1413 unregister_nfs_fs();
1418 /* Not quite true; I just maintain it */
1419 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1420 MODULE_LICENSE("GPL");
1421 module_param(enable_ino64, bool, 0644);
1423 module_init(init_nfs_fs)
1424 module_exit(exit_nfs_fs)