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 static void nfs_invalidate_inode(struct inode *);
53 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
55 static void nfs_zap_acl_cache(struct inode *);
57 static struct kmem_cache * nfs_inode_cachep;
59 static inline unsigned long
60 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
62 return nfs_fileid_to_ino_t(fattr->fileid);
65 int nfs_write_inode(struct inode *inode, int sync)
70 ret = filemap_fdatawait(inode->i_mapping);
72 ret = nfs_commit_inode(inode, FLUSH_SYNC);
74 ret = nfs_commit_inode(inode, 0);
77 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
81 void nfs_clear_inode(struct inode *inode)
84 * The following should never happen...
86 BUG_ON(nfs_have_writebacks(inode));
87 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
88 nfs_zap_acl_cache(inode);
89 nfs_access_zap_cache(inode);
93 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
95 int nfs_sync_mapping(struct address_space *mapping)
99 if (mapping->nrpages == 0)
101 unmap_mapping_range(mapping, 0, 0, 0);
102 ret = filemap_write_and_wait(mapping);
105 ret = nfs_wb_all(mapping->host);
111 * Invalidate the local caches
113 static void nfs_zap_caches_locked(struct inode *inode)
115 struct nfs_inode *nfsi = NFS_I(inode);
116 int mode = inode->i_mode;
118 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
120 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
121 nfsi->attrtimeo_timestamp = jiffies;
123 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
124 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
125 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
127 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
130 void nfs_zap_caches(struct inode *inode)
132 spin_lock(&inode->i_lock);
133 nfs_zap_caches_locked(inode);
134 spin_unlock(&inode->i_lock);
137 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
139 if (mapping->nrpages != 0) {
140 spin_lock(&inode->i_lock);
141 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
142 spin_unlock(&inode->i_lock);
146 static void nfs_zap_acl_cache(struct inode *inode)
148 void (*clear_acl_cache)(struct inode *);
150 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
151 if (clear_acl_cache != NULL)
152 clear_acl_cache(inode);
153 spin_lock(&inode->i_lock);
154 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
155 spin_unlock(&inode->i_lock);
158 void nfs_invalidate_atime(struct inode *inode)
160 spin_lock(&inode->i_lock);
161 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
162 spin_unlock(&inode->i_lock);
166 * Invalidate, but do not unhash, the inode.
167 * NB: must be called with inode->i_lock held!
169 static void nfs_invalidate_inode(struct inode *inode)
171 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
172 nfs_zap_caches_locked(inode);
175 struct nfs_find_desc {
177 struct nfs_fattr *fattr;
181 * In NFSv3 we can have 64bit inode numbers. In order to support
182 * this, and re-exported directories (also seen in NFSv2)
183 * we are forced to allow 2 different inodes to have the same
187 nfs_find_actor(struct inode *inode, void *opaque)
189 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
190 struct nfs_fh *fh = desc->fh;
191 struct nfs_fattr *fattr = desc->fattr;
193 if (NFS_FILEID(inode) != fattr->fileid)
195 if (nfs_compare_fh(NFS_FH(inode), fh))
197 if (is_bad_inode(inode) || NFS_STALE(inode))
203 nfs_init_locked(struct inode *inode, void *opaque)
205 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
206 struct nfs_fattr *fattr = desc->fattr;
208 NFS_FILEID(inode) = fattr->fileid;
209 nfs_copy_fh(NFS_FH(inode), desc->fh);
213 /* Don't use READDIRPLUS on directories that we believe are too large */
214 #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
217 * This is our front-end to iget that looks up inodes by file handle
218 * instead of inode number.
221 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
223 struct nfs_find_desc desc = {
227 struct inode *inode = ERR_PTR(-ENOENT);
230 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
234 printk("NFS: Buggy server - nlink == 0!\n");
238 hash = nfs_fattr_to_ino_t(fattr);
240 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
242 inode = ERR_PTR(-ENOMEM);
246 if (inode->i_state & I_NEW) {
247 struct nfs_inode *nfsi = NFS_I(inode);
248 unsigned long now = jiffies;
250 /* We set i_ino for the few things that still rely on it,
254 /* We can't support update_atime(), since the server will reset it */
255 inode->i_flags |= S_NOATIME|S_NOCMTIME;
256 inode->i_mode = fattr->mode;
257 /* Why so? Because we want revalidate for devices/FIFOs, and
258 * that's precisely what we have in nfs_file_inode_operations.
260 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
261 if (S_ISREG(inode->i_mode)) {
262 inode->i_fop = &nfs_file_operations;
263 inode->i_data.a_ops = &nfs_file_aops;
264 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
265 } else if (S_ISDIR(inode->i_mode)) {
266 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
267 inode->i_fop = &nfs_dir_operations;
268 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
269 && fattr->size <= NFS_LIMIT_READDIRPLUS)
270 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
271 /* Deal with crossing mountpoints */
272 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
273 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
274 inode->i_op = &nfs_referral_inode_operations;
276 inode->i_op = &nfs_mountpoint_inode_operations;
279 } else if (S_ISLNK(inode->i_mode))
280 inode->i_op = &nfs_symlink_inode_operations;
282 init_special_inode(inode, inode->i_mode, fattr->rdev);
284 nfsi->read_cache_jiffies = fattr->time_start;
285 nfsi->last_updated = now;
286 nfsi->cache_change_attribute = now;
287 inode->i_atime = fattr->atime;
288 inode->i_mtime = fattr->mtime;
289 inode->i_ctime = fattr->ctime;
290 if (fattr->valid & NFS_ATTR_FATTR_V4)
291 nfsi->change_attr = fattr->change_attr;
292 inode->i_size = nfs_size_to_loff_t(fattr->size);
293 inode->i_nlink = fattr->nlink;
294 inode->i_uid = fattr->uid;
295 inode->i_gid = fattr->gid;
296 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
298 * report the blocks in 512byte units
300 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
302 inode->i_blocks = fattr->du.nfs2.blocks;
304 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
305 nfsi->attrtimeo_timestamp = now;
306 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
307 nfsi->access_cache = RB_ROOT;
309 unlock_new_inode(inode);
311 nfs_refresh_inode(inode, fattr);
312 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
314 (long long)NFS_FILEID(inode),
315 atomic_read(&inode->i_count));
321 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
325 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET)
328 nfs_setattr(struct dentry *dentry, struct iattr *attr)
330 struct inode *inode = dentry->d_inode;
331 struct nfs_fattr fattr;
334 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
336 if (attr->ia_valid & ATTR_SIZE) {
337 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
338 attr->ia_valid &= ~ATTR_SIZE;
341 /* Optimization: if the end result is no change, don't RPC */
342 attr->ia_valid &= NFS_VALID_ATTRS;
343 if (attr->ia_valid == 0)
347 /* Write all dirty data */
348 if (S_ISREG(inode->i_mode)) {
349 filemap_write_and_wait(inode->i_mapping);
353 * Return any delegations if we're going to change ACLs
355 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
356 nfs_inode_return_delegation(inode);
357 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
359 nfs_refresh_inode(inode, &fattr);
365 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
366 * @inode: pointer to struct inode
367 * @attr: pointer to struct iattr
369 * Note: we do this in the *proc.c in order to ensure that
370 * it works for things like exclusive creates too.
372 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
374 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
375 if ((attr->ia_valid & ATTR_MODE) != 0) {
376 int mode = attr->ia_mode & S_IALLUGO;
377 mode |= inode->i_mode & ~S_IALLUGO;
378 inode->i_mode = mode;
380 if ((attr->ia_valid & ATTR_UID) != 0)
381 inode->i_uid = attr->ia_uid;
382 if ((attr->ia_valid & ATTR_GID) != 0)
383 inode->i_gid = attr->ia_gid;
384 spin_lock(&inode->i_lock);
385 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
386 spin_unlock(&inode->i_lock);
388 if ((attr->ia_valid & ATTR_SIZE) != 0) {
389 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
390 inode->i_size = attr->ia_size;
391 vmtruncate(inode, attr->ia_size);
395 static int nfs_wait_schedule(void *word)
397 if (signal_pending(current))
404 * Wait for the inode to get unlocked.
406 static int nfs_wait_on_inode(struct inode *inode)
408 struct rpc_clnt *clnt = NFS_CLIENT(inode);
409 struct nfs_inode *nfsi = NFS_I(inode);
413 rpc_clnt_sigmask(clnt, &oldmask);
414 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
415 nfs_wait_schedule, TASK_INTERRUPTIBLE);
416 rpc_clnt_sigunmask(clnt, &oldmask);
421 static void nfs_wake_up_inode(struct inode *inode)
423 struct nfs_inode *nfsi = NFS_I(inode);
425 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
426 smp_mb__after_clear_bit();
427 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
430 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
432 struct inode *inode = dentry->d_inode;
433 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
436 /* Flush out writes to the server in order to update c/mtime */
437 if (S_ISREG(inode->i_mode))
438 nfs_wb_nocommit(inode);
441 * We may force a getattr if the user cares about atime.
443 * Note that we only have to check the vfsmount flags here:
444 * - NFS always sets S_NOATIME by so checking it would give a
446 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
447 * no point in checking those.
449 if ((mnt->mnt_flags & MNT_NOATIME) ||
450 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
454 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
456 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
458 generic_fillattr(inode, stat);
459 stat->ino = NFS_FILEID(inode);
464 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
466 struct nfs_open_context *ctx;
468 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
470 ctx->path.dentry = dget(dentry);
471 ctx->path.mnt = mntget(mnt);
472 ctx->cred = get_rpccred(cred);
474 ctx->lockowner = current->files;
477 atomic_set(&ctx->count, 1);
482 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
485 atomic_inc(&ctx->count);
489 void put_nfs_open_context(struct nfs_open_context *ctx)
491 struct inode *inode = ctx->path.dentry->d_inode;
493 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
495 list_del(&ctx->list);
496 spin_unlock(&inode->i_lock);
497 if (ctx->state != NULL)
498 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
499 if (ctx->cred != NULL)
500 put_rpccred(ctx->cred);
501 dput(ctx->path.dentry);
502 mntput(ctx->path.mnt);
507 * Ensure that mmap has a recent RPC credential for use when writing out
510 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
512 struct inode *inode = filp->f_path.dentry->d_inode;
513 struct nfs_inode *nfsi = NFS_I(inode);
515 filp->private_data = get_nfs_open_context(ctx);
516 spin_lock(&inode->i_lock);
517 list_add(&ctx->list, &nfsi->open_files);
518 spin_unlock(&inode->i_lock);
522 * Given an inode, search for an open context with the desired characteristics
524 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
526 struct nfs_inode *nfsi = NFS_I(inode);
527 struct nfs_open_context *pos, *ctx = NULL;
529 spin_lock(&inode->i_lock);
530 list_for_each_entry(pos, &nfsi->open_files, list) {
531 if (cred != NULL && pos->cred != cred)
533 if ((pos->mode & mode) == mode) {
534 ctx = get_nfs_open_context(pos);
538 spin_unlock(&inode->i_lock);
542 static void nfs_file_clear_open_context(struct file *filp)
544 struct inode *inode = filp->f_path.dentry->d_inode;
545 struct nfs_open_context *ctx = nfs_file_open_context(filp);
548 filp->private_data = NULL;
549 spin_lock(&inode->i_lock);
550 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
551 spin_unlock(&inode->i_lock);
552 put_nfs_open_context(ctx);
557 * These allocate and release file read/write context information.
559 int nfs_open(struct inode *inode, struct file *filp)
561 struct nfs_open_context *ctx;
562 struct rpc_cred *cred;
564 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
566 return PTR_ERR(cred);
567 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
571 ctx->mode = filp->f_mode;
572 nfs_file_set_open_context(filp, ctx);
573 put_nfs_open_context(ctx);
577 int nfs_release(struct inode *inode, struct file *filp)
579 nfs_file_clear_open_context(filp);
584 * This function is called whenever some part of NFS notices that
585 * the cached attributes have to be refreshed.
588 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
590 int status = -ESTALE;
591 struct nfs_fattr fattr;
592 struct nfs_inode *nfsi = NFS_I(inode);
594 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
595 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
597 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
599 if (is_bad_inode(inode))
601 if (NFS_STALE(inode))
604 status = nfs_wait_on_inode(inode);
609 if (NFS_STALE(inode))
612 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
614 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
616 (long long)NFS_FILEID(inode), status);
617 if (status == -ESTALE) {
618 nfs_zap_caches(inode);
619 if (!S_ISDIR(inode->i_mode))
620 set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
625 spin_lock(&inode->i_lock);
626 status = nfs_update_inode(inode, &fattr);
628 spin_unlock(&inode->i_lock);
629 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
631 (long long)NFS_FILEID(inode), status);
634 spin_unlock(&inode->i_lock);
636 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
637 nfs_zap_acl_cache(inode);
639 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
641 (long long)NFS_FILEID(inode));
644 nfs_wake_up_inode(inode);
651 int nfs_attribute_timeout(struct inode *inode)
653 struct nfs_inode *nfsi = NFS_I(inode);
655 if (nfs_have_delegation(inode, FMODE_READ))
657 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
661 * nfs_revalidate_inode - Revalidate the inode attributes
662 * @server - pointer to nfs_server struct
663 * @inode - pointer to inode struct
665 * Updates inode attribute information by retrieving the data from the server.
667 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
669 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
670 && !nfs_attribute_timeout(inode))
671 return NFS_STALE(inode) ? -ESTALE : 0;
672 return __nfs_revalidate_inode(server, inode);
675 static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
677 struct nfs_inode *nfsi = NFS_I(inode);
679 if (mapping->nrpages != 0) {
680 int ret = invalidate_inode_pages2(mapping);
684 spin_lock(&inode->i_lock);
685 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
686 if (S_ISDIR(inode->i_mode))
687 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
688 spin_unlock(&inode->i_lock);
689 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
690 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
691 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
695 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
699 mutex_lock(&inode->i_mutex);
700 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
701 ret = nfs_sync_mapping(mapping);
703 ret = nfs_invalidate_mapping_nolock(inode, mapping);
705 mutex_unlock(&inode->i_mutex);
710 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
711 * @inode - pointer to host inode
712 * @mapping - pointer to mapping
714 int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
716 struct nfs_inode *nfsi = NFS_I(inode);
719 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
720 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
721 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
725 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
726 ret = nfs_invalidate_mapping_nolock(inode, mapping);
732 * nfs_revalidate_mapping - Revalidate the pagecache
733 * @inode - pointer to host inode
734 * @mapping - pointer to mapping
736 * This version of the function will take the inode->i_mutex and attempt to
737 * flush out all dirty data if it needs to invalidate the page cache.
739 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
741 struct nfs_inode *nfsi = NFS_I(inode);
744 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
745 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
746 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
750 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
751 ret = nfs_invalidate_mapping(inode, mapping);
756 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
758 struct nfs_inode *nfsi = NFS_I(inode);
760 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
761 nfsi->change_attr == fattr->pre_change_attr) {
762 nfsi->change_attr = fattr->change_attr;
763 if (S_ISDIR(inode->i_mode))
764 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
766 /* If we have atomic WCC data, we may update some attributes */
767 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
768 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
769 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
770 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
771 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
772 if (S_ISDIR(inode->i_mode))
773 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
775 if (inode->i_size == fattr->pre_size && nfsi->npages == 0)
776 inode->i_size = fattr->size;
781 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
782 * @inode - pointer to inode
783 * @fattr - updated attributes
785 * Verifies the attribute cache. If we have just changed the attributes,
786 * so that fattr carries weak cache consistency data, then it may
787 * also update the ctime/mtime/change_attribute.
789 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
791 struct nfs_inode *nfsi = NFS_I(inode);
792 loff_t cur_size, new_isize;
793 unsigned long invalid = 0;
796 /* Has the inode gone and changed behind our back? */
797 if (nfsi->fileid != fattr->fileid
798 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
802 /* Do atomic weak cache consistency updates */
803 nfs_wcc_update_inode(inode, fattr);
805 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
806 nfsi->change_attr != fattr->change_attr)
807 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
809 /* Verify a few of the more important attributes */
810 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
811 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
813 cur_size = i_size_read(inode);
814 new_isize = nfs_size_to_loff_t(fattr->size);
815 if (cur_size != new_isize && nfsi->npages == 0)
816 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
818 /* Have any file permissions changed? */
819 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
820 || inode->i_uid != fattr->uid
821 || inode->i_gid != fattr->gid)
822 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
824 /* Has the link count changed? */
825 if (inode->i_nlink != fattr->nlink)
826 invalid |= NFS_INO_INVALID_ATTR;
828 if (!timespec_equal(&inode->i_atime, &fattr->atime))
829 invalid |= NFS_INO_INVALID_ATIME;
832 nfsi->cache_validity |= invalid;
834 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
835 | NFS_INO_INVALID_ATIME
836 | NFS_INO_REVAL_PAGECACHE);
838 nfsi->read_cache_jiffies = fattr->time_start;
843 * nfs_refresh_inode - try to update the inode attribute cache
844 * @inode - pointer to inode
845 * @fattr - updated attributes
847 * Check that an RPC call that returned attributes has not overlapped with
848 * other recent updates of the inode metadata, then decide whether it is
849 * safe to do a full update of the inode attributes, or whether just to
850 * call nfs_check_inode_attributes.
852 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
854 struct nfs_inode *nfsi = NFS_I(inode);
857 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
859 spin_lock(&inode->i_lock);
860 if (time_after(fattr->time_start, nfsi->last_updated))
861 status = nfs_update_inode(inode, fattr);
863 status = nfs_check_inode_attributes(inode, fattr);
865 spin_unlock(&inode->i_lock);
870 * nfs_post_op_update_inode - try to update the inode attribute cache
871 * @inode - pointer to inode
872 * @fattr - updated attributes
874 * After an operation that has changed the inode metadata, mark the
875 * attribute cache as being invalid, then try to update it.
877 * NB: if the server didn't return any post op attributes, this
878 * function will force the retrieval of attributes before the next
879 * NFS request. Thus it should be used only for operations that
880 * are expected to change one or more attributes, to avoid
881 * unnecessary NFS requests and trips through nfs_update_inode().
883 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
885 struct nfs_inode *nfsi = NFS_I(inode);
887 spin_lock(&inode->i_lock);
888 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
889 if (S_ISDIR(inode->i_mode))
890 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
891 spin_unlock(&inode->i_lock);
892 return nfs_refresh_inode(inode, fattr);
896 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
897 * @inode - pointer to inode
898 * @fattr - updated attributes
900 * After an operation that has changed the inode metadata, mark the
901 * attribute cache as being invalid, then try to update it. Fake up
902 * weak cache consistency data, if none exist.
904 * This function is mainly designed to be used by the ->write_done() functions.
906 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
908 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
909 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
910 fattr->pre_change_attr = NFS_I(inode)->change_attr;
911 fattr->valid |= NFS_ATTR_WCC_V4;
913 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
914 (fattr->valid & NFS_ATTR_WCC) == 0) {
915 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
916 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
917 fattr->pre_size = inode->i_size;
918 fattr->valid |= NFS_ATTR_WCC;
920 return nfs_post_op_update_inode(inode, fattr);
924 * Many nfs protocol calls return the new file attributes after
925 * an operation. Here we update the inode to reflect the state
926 * of the server's inode.
928 * This is a bit tricky because we have to make sure all dirty pages
929 * have been sent off to the server before calling invalidate_inode_pages.
930 * To make sure no other process adds more write requests while we try
931 * our best to flush them, we make them sleep during the attribute refresh.
933 * A very similar scenario holds for the dir cache.
935 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
937 struct nfs_server *server;
938 struct nfs_inode *nfsi = NFS_I(inode);
939 loff_t cur_isize, new_isize;
940 unsigned long invalid = 0;
941 unsigned long now = jiffies;
943 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
944 __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
945 atomic_read(&inode->i_count), fattr->valid);
947 if (nfsi->fileid != fattr->fileid)
951 * Make sure the inode's type hasn't changed.
953 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
956 server = NFS_SERVER(inode);
957 /* Update the fsid? */
958 if (S_ISDIR(inode->i_mode)
959 && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
960 server->fsid = fattr->fsid;
963 * Update the read time so we don't revalidate too often.
965 nfsi->read_cache_jiffies = fattr->time_start;
967 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
968 | NFS_INO_REVAL_PAGECACHE);
970 /* Do atomic weak cache consistency updates */
971 nfs_wcc_update_inode(inode, fattr);
973 /* More cache consistency checks */
974 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
975 /* NFSv2/v3: Check if the mtime agrees */
976 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
977 dprintk("NFS: mtime change on server for file %s/%ld\n",
978 inode->i_sb->s_id, inode->i_ino);
979 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
980 nfsi->cache_change_attribute = now;
982 /* If ctime has changed we should definitely clear access+acl caches */
983 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
984 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
985 } else if (nfsi->change_attr != fattr->change_attr) {
986 dprintk("NFS: change_attr change on server for file %s/%ld\n",
987 inode->i_sb->s_id, inode->i_ino);
988 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
989 nfsi->cache_change_attribute = now;
992 /* Check if our cached file size is stale */
993 new_isize = nfs_size_to_loff_t(fattr->size);
994 cur_isize = i_size_read(inode);
995 if (new_isize != cur_isize) {
996 /* Do we perhaps have any outstanding writes, or has
997 * the file grown beyond our last write? */
998 if (nfsi->npages == 0 || new_isize > cur_isize) {
999 inode->i_size = new_isize;
1000 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1002 dprintk("NFS: isize change on server for file %s/%ld\n",
1003 inode->i_sb->s_id, inode->i_ino);
1007 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1008 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1009 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1010 nfsi->change_attr = fattr->change_attr;
1012 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1013 inode->i_uid != fattr->uid ||
1014 inode->i_gid != fattr->gid)
1015 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1017 inode->i_mode = fattr->mode;
1018 inode->i_nlink = fattr->nlink;
1019 inode->i_uid = fattr->uid;
1020 inode->i_gid = fattr->gid;
1022 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1024 * report the blocks in 512byte units
1026 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1028 inode->i_blocks = fattr->du.nfs2.blocks;
1031 /* Update attrtimeo value if we're out of the unstable period */
1032 if (invalid & NFS_INO_INVALID_ATTR) {
1033 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1034 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1035 nfsi->attrtimeo_timestamp = now;
1036 nfsi->last_updated = now;
1038 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1039 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1040 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1041 nfsi->attrtimeo_timestamp = now;
1044 * Avoid jiffy wraparound issues with nfsi->last_updated
1046 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1047 nfsi->last_updated = nfsi->read_cache_jiffies;
1049 invalid &= ~NFS_INO_INVALID_ATTR;
1050 /* Don't invalidate the data if we were to blame */
1051 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1052 || S_ISLNK(inode->i_mode)))
1053 invalid &= ~NFS_INO_INVALID_DATA;
1054 if (!nfs_have_delegation(inode, FMODE_READ) ||
1055 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
1056 nfsi->cache_validity |= invalid;
1057 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1062 * Big trouble! The inode has become a different object.
1064 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
1065 __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
1068 * No need to worry about unhashing the dentry, as the
1069 * lookup validation will know that the inode is bad.
1070 * (But we fall through to invalidate the caches.)
1072 nfs_invalidate_inode(inode);
1076 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1077 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1078 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
1079 (long long)nfsi->fileid, (long long)fattr->fileid);
1084 #ifdef CONFIG_NFS_V4
1087 * Clean out any remaining NFSv4 state that might be left over due
1088 * to open() calls that passed nfs_atomic_lookup, but failed to call
1091 void nfs4_clear_inode(struct inode *inode)
1093 /* If we are holding a delegation, return it! */
1094 nfs_inode_return_delegation(inode);
1095 /* First call standard NFS clear_inode() code */
1096 nfs_clear_inode(inode);
1100 struct inode *nfs_alloc_inode(struct super_block *sb)
1102 struct nfs_inode *nfsi;
1103 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1107 nfsi->cache_validity = 0UL;
1108 #ifdef CONFIG_NFS_V3_ACL
1109 nfsi->acl_access = ERR_PTR(-EAGAIN);
1110 nfsi->acl_default = ERR_PTR(-EAGAIN);
1112 #ifdef CONFIG_NFS_V4
1113 nfsi->nfs4_acl = NULL;
1114 #endif /* CONFIG_NFS_V4 */
1115 return &nfsi->vfs_inode;
1118 void nfs_destroy_inode(struct inode *inode)
1120 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1123 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1125 #ifdef CONFIG_NFS_V4
1126 INIT_LIST_HEAD(&nfsi->open_states);
1127 nfsi->delegation = NULL;
1128 nfsi->delegation_state = 0;
1129 init_rwsem(&nfsi->rwsem);
1133 static void init_once(void * foo, struct kmem_cache * cachep, unsigned long flags)
1135 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1137 inode_init_once(&nfsi->vfs_inode);
1138 INIT_LIST_HEAD(&nfsi->open_files);
1139 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1140 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1141 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
1144 nfs4_init_once(nfsi);
1147 static int __init nfs_init_inodecache(void)
1149 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1150 sizeof(struct nfs_inode),
1151 0, (SLAB_RECLAIM_ACCOUNT|
1154 if (nfs_inode_cachep == NULL)
1160 static void nfs_destroy_inodecache(void)
1162 kmem_cache_destroy(nfs_inode_cachep);
1168 static int __init init_nfs_fs(void)
1172 err = nfs_fs_proc_init();
1176 err = nfs_init_nfspagecache();
1180 err = nfs_init_inodecache();
1184 err = nfs_init_readpagecache();
1188 err = nfs_init_writepagecache();
1192 err = nfs_init_directcache();
1196 #ifdef CONFIG_PROC_FS
1197 rpc_proc_register(&nfs_rpcstat);
1199 if ((err = register_nfs_fs()) != 0)
1203 #ifdef CONFIG_PROC_FS
1204 rpc_proc_unregister("nfs");
1206 nfs_destroy_directcache();
1208 nfs_destroy_writepagecache();
1210 nfs_destroy_readpagecache();
1212 nfs_destroy_inodecache();
1214 nfs_destroy_nfspagecache();
1221 static void __exit exit_nfs_fs(void)
1223 nfs_destroy_directcache();
1224 nfs_destroy_writepagecache();
1225 nfs_destroy_readpagecache();
1226 nfs_destroy_inodecache();
1227 nfs_destroy_nfspagecache();
1228 #ifdef CONFIG_PROC_FS
1229 rpc_proc_unregister("nfs");
1231 unregister_nfs_fs();
1235 /* Not quite true; I just maintain it */
1236 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1237 MODULE_LICENSE("GPL");
1239 module_init(init_nfs_fs)
1240 module_exit(exit_nfs_fs)