Merge branch 'merge' of git://git.kernel.org/pub/scm/linux/kernel/git/paulus/powerpc
[safe/jmp/linux-2.6] / fs / nfs / inode.c
index 51bc88b..a4c7cf2 100644 (file)
  *
  */
 
-#include <linux/config.h>
 #include <linux/module.h>
 #include <linux/init.h>
-
+#include <linux/sched.h>
 #include <linux/time.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include "internal.h"
 
 #define NFSDBG_FACILITY                NFSDBG_VFS
-#define NFS_PARANOIA 1
+
+#define NFS_64_BIT_INODE_NUMBERS_ENABLED       1
+
+/* Default is to see 64-bit inode numbers */
+static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
 
 static void nfs_invalidate_inode(struct inode *);
 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
 
 static void nfs_zap_acl_cache(struct inode *);
 
-static kmem_cache_t * nfs_inode_cachep;
+static struct kmem_cache * nfs_inode_cachep;
 
 static inline unsigned long
 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
@@ -64,32 +67,50 @@ nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
        return nfs_fileid_to_ino_t(fattr->fileid);
 }
 
+/**
+ * nfs_compat_user_ino64 - returns the user-visible inode number
+ * @fileid: 64-bit fileid
+ *
+ * This function returns a 32-bit inode number if the boot parameter
+ * nfs.enable_ino64 is zero.
+ */
+u64 nfs_compat_user_ino64(u64 fileid)
+{
+       int ino;
+
+       if (enable_ino64)
+               return fileid;
+       ino = fileid;
+       if (sizeof(ino) < sizeof(fileid))
+               ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
+       return ino;
+}
+
 int nfs_write_inode(struct inode *inode, int sync)
 {
-       int flags = sync ? FLUSH_SYNC : 0;
        int ret;
 
-       ret = nfs_commit_inode(inode, flags);
-       if (ret < 0)
-               return ret;
-       return 0;
+       if (sync) {
+               ret = filemap_fdatawait(inode->i_mapping);
+               if (ret == 0)
+                       ret = nfs_commit_inode(inode, FLUSH_SYNC);
+       } else
+               ret = nfs_commit_inode(inode, 0);
+       if (ret >= 0)
+               return 0;
+       __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+       return ret;
 }
 
 void nfs_clear_inode(struct inode *inode)
 {
-       struct nfs_inode *nfsi = NFS_I(inode);
-       struct rpc_cred *cred;
-
        /*
         * The following should never happen...
         */
        BUG_ON(nfs_have_writebacks(inode));
-       BUG_ON (!list_empty(&nfsi->open_files));
+       BUG_ON(!list_empty(&NFS_I(inode)->open_files));
        nfs_zap_acl_cache(inode);
-       cred = nfsi->cache_access.cred;
-       if (cred)
-               put_rpccred(cred);
-       BUG_ON(atomic_read(&nfsi->data_updates) != 0);
+       nfs_access_zap_cache(inode);
 }
 
 /**
@@ -120,8 +141,8 @@ static void nfs_zap_caches_locked(struct inode *inode)
 
        nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
 
-       NFS_ATTRTIMEO(inode) = NFS_MINATTRTIMEO(inode);
-       NFS_ATTRTIMEO_UPDATE(inode) = jiffies;
+       nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
+       nfsi->attrtimeo_timestamp = jiffies;
 
        memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
        if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
@@ -137,6 +158,15 @@ void nfs_zap_caches(struct inode *inode)
        spin_unlock(&inode->i_lock);
 }
 
+void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
+{
+       if (mapping->nrpages != 0) {
+               spin_lock(&inode->i_lock);
+               NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
+               spin_unlock(&inode->i_lock);
+       }
+}
+
 static void nfs_zap_acl_cache(struct inode *inode)
 {
        void (*clear_acl_cache)(struct inode *);
@@ -149,13 +179,20 @@ static void nfs_zap_acl_cache(struct inode *inode)
        spin_unlock(&inode->i_lock);
 }
 
+void nfs_invalidate_atime(struct inode *inode)
+{
+       spin_lock(&inode->i_lock);
+       NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
+       spin_unlock(&inode->i_lock);
+}
+
 /*
  * Invalidate, but do not unhash, the inode.
  * NB: must be called with inode->i_lock held!
  */
 static void nfs_invalidate_inode(struct inode *inode)
 {
-       set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
+       set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
        nfs_zap_caches_locked(inode);
 }
 
@@ -192,7 +229,7 @@ nfs_init_locked(struct inode *inode, void *opaque)
        struct nfs_find_desc    *desc = (struct nfs_find_desc *)opaque;
        struct nfs_fattr        *fattr = desc->fattr;
 
-       NFS_FILEID(inode) = fattr->fileid;
+       set_nfs_fileid(inode, fattr->fileid);
        nfs_copy_fh(NFS_FH(inode), desc->fh);
        return 0;
 }
@@ -232,6 +269,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
 
        if (inode->i_state & I_NEW) {
                struct nfs_inode *nfsi = NFS_I(inode);
+               unsigned long now = jiffies;
 
                /* We set i_ino for the few things that still rely on it,
                 * such as stat(2) */
@@ -243,17 +281,17 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                /* Why so? Because we want revalidate for devices/FIFOs, and
                 * that's precisely what we have in nfs_file_inode_operations.
                 */
-               inode->i_op = NFS_SB(sb)->rpc_ops->file_inode_ops;
+               inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
                if (S_ISREG(inode->i_mode)) {
                        inode->i_fop = &nfs_file_operations;
                        inode->i_data.a_ops = &nfs_file_aops;
                        inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
                } else if (S_ISDIR(inode->i_mode)) {
-                       inode->i_op = NFS_SB(sb)->rpc_ops->dir_inode_ops;
+                       inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
                        inode->i_fop = &nfs_dir_operations;
                        if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
                            && fattr->size <= NFS_LIMIT_READDIRPLUS)
-                               set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
+                               set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
                        /* Deal with crossing mountpoints */
                        if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
                                if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
@@ -261,6 +299,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                                else
                                        inode->i_op = &nfs_mountpoint_inode_operations;
                                inode->i_fop = NULL;
+                               set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
                        }
                } else if (S_ISLNK(inode->i_mode))
                        inode->i_op = &nfs_symlink_inode_operations;
@@ -268,7 +307,8 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                        init_special_inode(inode, inode->i_mode, fattr->rdev);
 
                nfsi->read_cache_jiffies = fattr->time_start;
-               nfsi->last_updated = jiffies;
+               nfsi->last_updated = now;
+               nfsi->cache_change_attribute = now;
                inode->i_atime = fattr->atime;
                inode->i_mtime = fattr->mtime;
                inode->i_ctime = fattr->ctime;
@@ -283,15 +323,13 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                         * report the blocks in 512byte units
                         */
                        inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
-                       inode->i_blksize = inode->i_sb->s_blocksize;
                } else {
                        inode->i_blocks = fattr->du.nfs2.blocks;
-                       inode->i_blksize = fattr->du.nfs2.blocksize;
                }
                nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
-               nfsi->attrtimeo_timestamp = jiffies;
+               nfsi->attrtimeo_timestamp = now;
                memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
-               nfsi->cache_access.cred = NULL;
+               nfsi->access_cache = RB_ROOT;
 
                unlock_new_inode(inode);
        } else
@@ -320,6 +358,10 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
 
        nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
 
+       /* skip mode change if it's just for clearing setuid/setgid */
+       if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
+               attr->ia_valid &= ~ATTR_MODE;
+
        if (attr->ia_valid & ATTR_SIZE) {
                if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
                        attr->ia_valid &= ~ATTR_SIZE;
@@ -331,10 +373,11 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
                return 0;
 
        lock_kernel();
-       nfs_begin_data_update(inode);
        /* Write all dirty data */
-       filemap_write_and_wait(inode->i_mapping);
-       nfs_wb_all(inode);
+       if (S_ISREG(inode->i_mode)) {
+               filemap_write_and_wait(inode->i_mapping);
+               nfs_wb_all(inode);
+       }
        /*
         * Return any delegations if we're going to change ACLs
         */
@@ -343,7 +386,6 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
        error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
        if (error == 0)
                nfs_refresh_inode(inode, &fattr);
-       nfs_end_data_update(inode);
        unlock_kernel();
        return error;
 }
@@ -392,15 +434,11 @@ static int nfs_wait_schedule(void *word)
  */
 static int nfs_wait_on_inode(struct inode *inode)
 {
-       struct rpc_clnt *clnt = NFS_CLIENT(inode);
        struct nfs_inode *nfsi = NFS_I(inode);
-       sigset_t oldmask;
        int error;
 
-       rpc_clnt_sigmask(clnt, &oldmask);
        error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
-                                       nfs_wait_schedule, TASK_INTERRUPTIBLE);
-       rpc_clnt_sigunmask(clnt, &oldmask);
+                                       nfs_wait_schedule, TASK_KILLABLE);
 
        return error;
 }
@@ -420,8 +458,18 @@ int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
        int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
        int err;
 
-       /* Flush out writes to the server in order to update c/mtime */
-       nfs_sync_inode_wait(inode, 0, 0, FLUSH_NOCOMMIT);
+       /*
+        * Flush out writes to the server in order to update c/mtime.
+        *
+        * Hold the i_mutex to suspend application writes temporarily;
+        * this prevents long-running writing applications from blocking
+        * nfs_wb_nocommit.
+        */
+       if (S_ISREG(inode->i_mode)) {
+               mutex_lock(&inode->i_mutex);
+               nfs_wb_nocommit(inode);
+               mutex_unlock(&inode->i_mutex);
+       }
 
        /*
         * We may force a getattr if the user cares about atime.
@@ -440,8 +488,10 @@ int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
                err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
        else
                err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
-       if (!err)
+       if (!err) {
                generic_fillattr(inode, stat);
+               stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
+       }
        return err;
 }
 
@@ -449,16 +499,16 @@ static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, str
 {
        struct nfs_open_context *ctx;
 
-       ctx = (struct nfs_open_context *)kmalloc(sizeof(*ctx), GFP_KERNEL);
+       ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
        if (ctx != NULL) {
-               atomic_set(&ctx->count, 1);
-               ctx->dentry = dget(dentry);
-               ctx->vfsmnt = mntget(mnt);
+               ctx->path.dentry = dget(dentry);
+               ctx->path.mnt = mntget(mnt);
                ctx->cred = get_rpccred(cred);
                ctx->state = NULL;
                ctx->lockowner = current->files;
                ctx->error = 0;
                ctx->dir_cookie = 0;
+               atomic_set(&ctx->count, 1);
        }
        return ctx;
 }
@@ -470,23 +520,35 @@ struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
        return ctx;
 }
 
-void put_nfs_open_context(struct nfs_open_context *ctx)
+static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
 {
-       if (atomic_dec_and_test(&ctx->count)) {
-               if (!list_empty(&ctx->list)) {
-                       struct inode *inode = ctx->dentry->d_inode;
-                       spin_lock(&inode->i_lock);
-                       list_del(&ctx->list);
-                       spin_unlock(&inode->i_lock);
-               }
-               if (ctx->state != NULL)
-                       nfs4_close_state(ctx->state, ctx->mode);
-               if (ctx->cred != NULL)
-                       put_rpccred(ctx->cred);
-               dput(ctx->dentry);
-               mntput(ctx->vfsmnt);
-               kfree(ctx);
+       struct inode *inode = ctx->path.dentry->d_inode;
+
+       if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
+               return;
+       list_del(&ctx->list);
+       spin_unlock(&inode->i_lock);
+       if (ctx->state != NULL) {
+               if (wait)
+                       nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
+               else
+                       nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
        }
+       if (ctx->cred != NULL)
+               put_rpccred(ctx->cred);
+       dput(ctx->path.dentry);
+       mntput(ctx->path.mnt);
+       kfree(ctx);
+}
+
+void put_nfs_open_context(struct nfs_open_context *ctx)
+{
+       __put_nfs_open_context(ctx, 0);
+}
+
+static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
+{
+       __put_nfs_open_context(ctx, 1);
 }
 
 /*
@@ -495,7 +557,7 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
  */
 static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
+       struct inode *inode = filp->f_path.dentry->d_inode;
        struct nfs_inode *nfsi = NFS_I(inode);
 
        filp->private_data = get_nfs_open_context(ctx);
@@ -527,15 +589,15 @@ struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_c
 
 static void nfs_file_clear_open_context(struct file *filp)
 {
-       struct inode *inode = filp->f_dentry->d_inode;
-       struct nfs_open_context *ctx = (struct nfs_open_context *)filp->private_data;
+       struct inode *inode = filp->f_path.dentry->d_inode;
+       struct nfs_open_context *ctx = nfs_file_open_context(filp);
 
        if (ctx) {
                filp->private_data = NULL;
                spin_lock(&inode->i_lock);
                list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
                spin_unlock(&inode->i_lock);
-               put_nfs_open_context(ctx);
+               put_nfs_open_context_sync(ctx);
        }
 }
 
@@ -550,7 +612,7 @@ int nfs_open(struct inode *inode, struct file *filp)
        cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
        if (IS_ERR(cred))
                return PTR_ERR(cred);
-       ctx = alloc_nfs_open_context(filp->f_vfsmnt, filp->f_dentry, cred);
+       ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
        put_rpccred(cred);
        if (ctx == NULL)
                return -ENOMEM;
@@ -582,7 +644,7 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
 
        nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
        lock_kernel();
-       if (!inode || is_bad_inode(inode))
+       if (is_bad_inode(inode))
                goto out_nowait;
        if (NFS_STALE(inode))
                goto out_nowait;
@@ -590,16 +652,10 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
        status = nfs_wait_on_inode(inode);
        if (status < 0)
                goto out;
-       if (NFS_STALE(inode)) {
-               status = -ESTALE;
-               /* Do we trust the cached ESTALE? */
-               if (NFS_ATTRTIMEO(inode) != 0) {
-                       if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME)) {
-                               /* no */
-                       } else
-                               goto out;
-               }
-       }
+
+       status = -ESTALE;
+       if (NFS_STALE(inode))
+               goto out;
 
        status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
        if (status != 0) {
@@ -609,7 +665,7 @@ __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
                if (status == -ESTALE) {
                        nfs_zap_caches(inode);
                        if (!S_ISDIR(inode->i_mode))
-                               set_bit(NFS_INO_STALE, &NFS_FLAGS(inode));
+                               set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
                }
                goto out;
        }
@@ -646,7 +702,7 @@ int nfs_attribute_timeout(struct inode *inode)
 
        if (nfs_have_delegation(inode, FMODE_READ))
                return 0;
-       return time_after(jiffies, nfsi->read_cache_jiffies+nfsi->attrtimeo);
+       return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
 }
 
 /**
@@ -664,95 +720,109 @@ int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
        return __nfs_revalidate_inode(server, inode);
 }
 
+static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
+{
+       struct nfs_inode *nfsi = NFS_I(inode);
+       
+       if (mapping->nrpages != 0) {
+               int ret = invalidate_inode_pages2(mapping);
+               if (ret < 0)
+                       return ret;
+       }
+       spin_lock(&inode->i_lock);
+       nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
+       if (S_ISDIR(inode->i_mode))
+               memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
+       spin_unlock(&inode->i_lock);
+       nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
+       dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
+                       inode->i_sb->s_id, (long long)NFS_FILEID(inode));
+       return 0;
+}
+
+static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
+{
+       int ret = 0;
+
+       mutex_lock(&inode->i_mutex);
+       if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
+               ret = nfs_sync_mapping(mapping);
+               if (ret == 0)
+                       ret = nfs_invalidate_mapping_nolock(inode, mapping);
+       }
+       mutex_unlock(&inode->i_mutex);
+       return ret;
+}
+
 /**
- * nfs_revalidate_mapping - Revalidate the pagecache
+ * nfs_revalidate_mapping_nolock - Revalidate the pagecache
  * @inode - pointer to host inode
  * @mapping - pointer to mapping
  */
-int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
+int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
        int ret = 0;
 
-       if (NFS_STALE(inode))
-               ret = -ESTALE;
        if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
-                       || nfs_attribute_timeout(inode))
+                       || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
                ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
-
-       if (nfsi->cache_validity & NFS_INO_INVALID_DATA) {
-               nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
-               if (S_ISREG(inode->i_mode))
-                       nfs_sync_mapping(mapping);
-               invalidate_inode_pages2(mapping);
-
-               spin_lock(&inode->i_lock);
-               nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
-               if (S_ISDIR(inode->i_mode)) {
-                       memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
-                       /* This ensures we revalidate child dentries */
-                       nfsi->cache_change_attribute = jiffies;
-               }
-               spin_unlock(&inode->i_lock);
-
-               dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
-                               inode->i_sb->s_id,
-                               (long long)NFS_FILEID(inode));
+               if (ret < 0)
+                       goto out;
        }
+       if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
+               ret = nfs_invalidate_mapping_nolock(inode, mapping);
+out:
        return ret;
 }
 
 /**
- * nfs_begin_data_update
- * @inode - pointer to inode
- * Declare that a set of operations will update file data on the server
- */
-void nfs_begin_data_update(struct inode *inode)
-{
-       atomic_inc(&NFS_I(inode)->data_updates);
-}
-
-/**
- * nfs_end_data_update
- * @inode - pointer to inode
- * Declare end of the operations that will update file data
- * This will mark the inode as immediately needing revalidation
- * of its attribute cache.
+ * nfs_revalidate_mapping - Revalidate the pagecache
+ * @inode - pointer to host inode
+ * @mapping - pointer to mapping
+ *
+ * This version of the function will take the inode->i_mutex and attempt to
+ * flush out all dirty data if it needs to invalidate the page cache.
  */
-void nfs_end_data_update(struct inode *inode)
+int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
+       int ret = 0;
 
-       if (!nfs_have_delegation(inode, FMODE_READ)) {
-               /* Directories and symlinks: invalidate page cache */
-               if (S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)) {
-                       spin_lock(&inode->i_lock);
-                       nfsi->cache_validity |= NFS_INO_INVALID_DATA;
-                       spin_unlock(&inode->i_lock);
-               }
+       if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
+                       || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
+               ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
+               if (ret < 0)
+                       goto out;
        }
-       nfsi->cache_change_attribute = jiffies;
-       atomic_dec(&nfsi->data_updates);
+       if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
+               ret = nfs_invalidate_mapping(inode, mapping);
+out:
+       return ret;
 }
 
 static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
 
+       if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
+                       nfsi->change_attr == fattr->pre_change_attr) {
+               nfsi->change_attr = fattr->change_attr;
+               if (S_ISDIR(inode->i_mode))
+                       nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+       }
        /* If we have atomic WCC data, we may update some attributes */
        if ((fattr->valid & NFS_ATTR_WCC) != 0) {
-               if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
+               if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
                        memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
-                       nfsi->cache_change_attribute = jiffies;
-               }
                if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
                        memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-                       nfsi->cache_change_attribute = jiffies;
-               }
-               if (inode->i_size == fattr->pre_size && nfsi->npages == 0) {
-                       inode->i_size = fattr->size;
-                       nfsi->cache_change_attribute = jiffies;
+                       if (S_ISDIR(inode->i_mode))
+                               nfsi->cache_validity |= NFS_INO_INVALID_DATA;
                }
+               if (inode->i_size == nfs_size_to_loff_t(fattr->pre_size) &&
+                   nfsi->npages == 0)
+                       inode->i_size = nfs_size_to_loff_t(fattr->size);
        }
 }
 
@@ -769,7 +839,7 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
 {
        struct nfs_inode *nfsi = NFS_I(inode);
        loff_t cur_size, new_isize;
-       int data_unstable;
+       unsigned long invalid = 0;
 
 
        /* Has the inode gone and changed behind our back? */
@@ -778,37 +848,41 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
                return -EIO;
        }
 
-       /* Are we in the process of updating data on the server? */
-       data_unstable = nfs_caches_unstable(inode);
-
        /* Do atomic weak cache consistency updates */
        nfs_wcc_update_inode(inode, fattr);
 
        if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
                        nfsi->change_attr != fattr->change_attr)
-               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 
        /* Verify a few of the more important attributes */
        if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
-               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 
        cur_size = i_size_read(inode);
        new_isize = nfs_size_to_loff_t(fattr->size);
        if (cur_size != new_isize && nfsi->npages == 0)
-               nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 
        /* Have any file permissions changed? */
        if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
                        || inode->i_uid != fattr->uid
                        || inode->i_gid != fattr->gid)
-               nfsi->cache_validity |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
+               invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
 
        /* Has the link count changed? */
        if (inode->i_nlink != fattr->nlink)
-               nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
+               invalid |= NFS_INO_INVALID_ATTR;
 
        if (!timespec_equal(&inode->i_atime, &fattr->atime))
-               nfsi->cache_validity |= NFS_INO_INVALID_ATIME;
+               invalid |= NFS_INO_INVALID_ATIME;
+
+       if (invalid != 0)
+               nfsi->cache_validity |= invalid;
+       else
+               nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
+                               | NFS_INO_INVALID_ATIME
+                               | NFS_INO_REVAL_PAGECACHE);
 
        nfsi->read_cache_jiffies = fattr->time_start;
        return 0;
@@ -848,21 +922,51 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
  *
  * After an operation that has changed the inode metadata, mark the
  * attribute cache as being invalid, then try to update it.
+ *
+ * NB: if the server didn't return any post op attributes, this
+ * function will force the retrieval of attributes before the next
+ * NFS request.  Thus it should be used only for operations that
+ * are expected to change one or more attributes, to avoid
+ * unnecessary NFS requests and trips through nfs_update_inode().
  */
 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
-       int status = 0;
 
        spin_lock(&inode->i_lock);
-       if (unlikely((fattr->valid & NFS_ATTR_FATTR) == 0)) {
-               nfsi->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
-               goto out;
-       }
-       status = nfs_update_inode(inode, fattr);
-out:
+       nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
+       if (S_ISDIR(inode->i_mode))
+               nfsi->cache_validity |= NFS_INO_INVALID_DATA;
        spin_unlock(&inode->i_lock);
-       return status;
+       return nfs_refresh_inode(inode, fattr);
+}
+
+/**
+ * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
+ * @inode - pointer to inode
+ * @fattr - updated attributes
+ *
+ * After an operation that has changed the inode metadata, mark the
+ * attribute cache as being invalid, then try to update it. Fake up
+ * weak cache consistency data, if none exist.
+ *
+ * This function is mainly designed to be used by the ->write_done() functions.
+ */
+int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
+{
+       if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
+                       (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
+               fattr->pre_change_attr = NFS_I(inode)->change_attr;
+               fattr->valid |= NFS_ATTR_WCC_V4;
+       }
+       if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
+                       (fattr->valid & NFS_ATTR_WCC) == 0) {
+               memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
+               memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
+               fattr->pre_size = inode->i_size;
+               fattr->valid |= NFS_ATTR_WCC;
+       }
+       return nfs_post_op_update_inode(inode, fattr);
 }
 
 /*
@@ -882,8 +986,8 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
        struct nfs_server *server;
        struct nfs_inode *nfsi = NFS_I(inode);
        loff_t cur_isize, new_isize;
-       unsigned int    invalid = 0;
-       int data_stable;
+       unsigned long invalid = 0;
+       unsigned long now = jiffies;
 
        dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
                        __FUNCTION__, inode->i_sb->s_id, inode->i_ino,
@@ -899,62 +1003,63 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                goto out_changed;
 
        server = NFS_SERVER(inode);
-       /* Update the fsid if and only if this is the root directory */
-       if (inode == inode->i_sb->s_root->d_inode
-                       && !nfs_fsid_equal(&server->fsid, &fattr->fsid))
+       /* Update the fsid? */
+       if (S_ISDIR(inode->i_mode) &&
+                       !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
+                       !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
                server->fsid = fattr->fsid;
 
        /*
         * Update the read time so we don't revalidate too often.
         */
        nfsi->read_cache_jiffies = fattr->time_start;
-       nfsi->last_updated = jiffies;
 
-       /* Are we racing with known updates of the metadata on the server? */
-       data_stable = nfs_verify_change_attribute(inode, fattr->time_start);
-       if (data_stable)
-               nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE|NFS_INO_INVALID_ATIME);
+       nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
+                       | NFS_INO_REVAL_PAGECACHE);
 
        /* Do atomic weak cache consistency updates */
        nfs_wcc_update_inode(inode, fattr);
 
+       /* More cache consistency checks */
+       if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
+               /* NFSv2/v3: Check if the mtime agrees */
+               if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
+                       dprintk("NFS: mtime change on server for file %s/%ld\n",
+                                       inode->i_sb->s_id, inode->i_ino);
+                       invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
+                       if (S_ISDIR(inode->i_mode))
+                               nfs_force_lookup_revalidate(inode);
+               }
+               /* If ctime has changed we should definitely clear access+acl caches */
+               if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
+                       invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+       } else if (nfsi->change_attr != fattr->change_attr) {
+               dprintk("NFS: change_attr change on server for file %s/%ld\n",
+                               inode->i_sb->s_id, inode->i_ino);
+               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
+               if (S_ISDIR(inode->i_mode))
+                       nfs_force_lookup_revalidate(inode);
+       }
+
        /* Check if our cached file size is stale */
        new_isize = nfs_size_to_loff_t(fattr->size);
        cur_isize = i_size_read(inode);
        if (new_isize != cur_isize) {
-               /* Do we perhaps have any outstanding writes? */
-               if (nfsi->npages == 0) {
-                       /* No, but did we race with nfs_end_data_update()? */
-                       if (data_stable) {
-                               inode->i_size = new_isize;
-                               invalid |= NFS_INO_INVALID_DATA;
-                       }
-                       invalid |= NFS_INO_INVALID_ATTR;
-               } else if (new_isize > cur_isize) {
+               /* Do we perhaps have any outstanding writes, or has
+                * the file grown beyond our last write? */
+               if (nfsi->npages == 0 || new_isize > cur_isize) {
                        inode->i_size = new_isize;
                        invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
                }
-               nfsi->cache_change_attribute = jiffies;
                dprintk("NFS: isize change on server for file %s/%ld\n",
                                inode->i_sb->s_id, inode->i_ino);
        }
 
-       /* Check if the mtime agrees */
-       if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
-               memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-               dprintk("NFS: mtime change on server for file %s/%ld\n",
-                               inode->i_sb->s_id, inode->i_ino);
-               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
-               nfsi->cache_change_attribute = jiffies;
-       }
 
-       /* If ctime has changed we should definitely clear access+acl caches */
-       if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) {
-               invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
-               memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
-               nfsi->cache_change_attribute = jiffies;
-       }
+       memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
+       memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
        memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
+       nfsi->change_attr = fattr->change_attr;
 
        if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
            inode->i_uid != fattr->uid ||
@@ -971,49 +1076,45 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                 * report the blocks in 512byte units
                 */
                inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
-               inode->i_blksize = inode->i_sb->s_blocksize;
        } else {
                inode->i_blocks = fattr->du.nfs2.blocks;
-               inode->i_blksize = fattr->du.nfs2.blocksize;
        }
 
-       if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
-                       nfsi->change_attr != fattr->change_attr) {
-               dprintk("NFS: change_attr change on server for file %s/%ld\n",
-                               inode->i_sb->s_id, inode->i_ino);
-               nfsi->change_attr = fattr->change_attr;
-               invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
-               nfsi->cache_change_attribute = jiffies;
-       }
-
        /* Update attrtimeo value if we're out of the unstable period */
        if (invalid & NFS_INO_INVALID_ATTR) {
                nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
                nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
-               nfsi->attrtimeo_timestamp = jiffies;
-       } else if (time_after(jiffies, nfsi->attrtimeo_timestamp+nfsi->attrtimeo)) {
-               if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
-                       nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
-               nfsi->attrtimeo_timestamp = jiffies;
+               nfsi->attrtimeo_timestamp = now;
+               nfsi->last_updated = now;
+       } else {
+               if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
+                       if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
+                               nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
+                       nfsi->attrtimeo_timestamp = now;
+               }
+               /*
+                * Avoid jiffy wraparound issues with nfsi->last_updated
+                */
+               if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
+                       nfsi->last_updated = nfsi->read_cache_jiffies;
        }
+       invalid &= ~NFS_INO_INVALID_ATTR;
        /* Don't invalidate the data if we were to blame */
        if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
                                || S_ISLNK(inode->i_mode)))
                invalid &= ~NFS_INO_INVALID_DATA;
-       if (data_stable)
-               invalid &= ~(NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ATIME|NFS_INO_REVAL_PAGECACHE);
-       if (!nfs_have_delegation(inode, FMODE_READ))
+       if (!nfs_have_delegation(inode, FMODE_READ) ||
+                       (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
                nfsi->cache_validity |= invalid;
+       nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
 
        return 0;
  out_changed:
        /*
         * Big trouble! The inode has become a different object.
         */
-#ifdef NFS_PARANOIA
        printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
                        __FUNCTION__, inode->i_ino, inode->i_mode, fattr->mode);
-#endif
  out_err:
        /*
         * No need to worry about unhashing the dentry, as the
@@ -1026,7 +1127,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
  out_fileid:
        printk(KERN_ERR "NFS: server %s error: fileid changed\n"
                "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
-               NFS_SERVER(inode)->hostname, inode->i_sb->s_id,
+               NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
                (long long)nfsi->fileid, (long long)fattr->fileid);
        goto out_err;
 }
@@ -1041,39 +1142,21 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
  */
 void nfs4_clear_inode(struct inode *inode)
 {
-       struct nfs_inode *nfsi = NFS_I(inode);
-
        /* If we are holding a delegation, return it! */
-       nfs_inode_return_delegation(inode);
+       nfs_inode_return_delegation_noreclaim(inode);
        /* First call standard NFS clear_inode() code */
        nfs_clear_inode(inode);
-       /* Now clear out any remaining state */
-       while (!list_empty(&nfsi->open_states)) {
-               struct nfs4_state *state;
-               
-               state = list_entry(nfsi->open_states.next,
-                               struct nfs4_state,
-                               inode_states);
-               dprintk("%s(%s/%Ld): found unclaimed NFSv4 state %p\n",
-                               __FUNCTION__,
-                               inode->i_sb->s_id,
-                               (long long)NFS_FILEID(inode),
-                               state);
-               BUG_ON(atomic_read(&state->count) != 1);
-               nfs4_close_state(state, state->state);
-       }
 }
 #endif
 
 struct inode *nfs_alloc_inode(struct super_block *sb)
 {
        struct nfs_inode *nfsi;
-       nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, SLAB_KERNEL);
+       nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
        if (!nfsi)
                return NULL;
        nfsi->flags = 0UL;
        nfsi->cache_validity = 0UL;
-       nfsi->cache_change_attribute = jiffies;
 #ifdef CONFIG_NFS_V3_ACL
        nfsi->acl_access = ERR_PTR(-EAGAIN);
        nfsi->acl_default = ERR_PTR(-EAGAIN);
@@ -1099,43 +1182,39 @@ static inline void nfs4_init_once(struct nfs_inode *nfsi)
 #endif
 }
 
-static void init_once(void * foo, kmem_cache_t * cachep, unsigned long flags)
+static void init_once(struct kmem_cache * cachep, void *foo)
 {
        struct nfs_inode *nfsi = (struct nfs_inode *) foo;
 
-       if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
-           SLAB_CTOR_CONSTRUCTOR) {
-               inode_init_once(&nfsi->vfs_inode);
-               spin_lock_init(&nfsi->req_lock);
-               INIT_LIST_HEAD(&nfsi->dirty);
-               INIT_LIST_HEAD(&nfsi->commit);
-               INIT_LIST_HEAD(&nfsi->open_files);
-               INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
-               atomic_set(&nfsi->data_updates, 0);
-               nfsi->ndirty = 0;
-               nfsi->ncommit = 0;
-               nfsi->npages = 0;
-               nfs4_init_once(nfsi);
-       }
+       inode_init_once(&nfsi->vfs_inode);
+       INIT_LIST_HEAD(&nfsi->open_files);
+       INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
+       INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
+       INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
+       nfsi->ncommit = 0;
+       nfsi->npages = 0;
+       atomic_set(&nfsi->silly_count, 1);
+       INIT_HLIST_HEAD(&nfsi->silly_list);
+       init_waitqueue_head(&nfsi->waitqueue);
+       nfs4_init_once(nfsi);
 }
+
 static int __init nfs_init_inodecache(void)
 {
        nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
                                             sizeof(struct nfs_inode),
                                             0, (SLAB_RECLAIM_ACCOUNT|
                                                SLAB_MEM_SPREAD),
-                                            init_once, NULL);
+                                            init_once);
        if (nfs_inode_cachep == NULL)
                return -ENOMEM;
 
        return 0;
 }
 
-static void __exit nfs_destroy_inodecache(void)
+static void nfs_destroy_inodecache(void)
 {
-       if (kmem_cache_destroy(nfs_inode_cachep))
-               printk(KERN_INFO "nfs_inode_cache: not all structures were freed\n");
+       kmem_cache_destroy(nfs_inode_cachep);
 }
 
 /*
@@ -1145,6 +1224,10 @@ static int __init init_nfs_fs(void)
 {
        int err;
 
+       err = nfs_fs_proc_init();
+       if (err)
+               goto out5;
+
        err = nfs_init_nfspagecache();
        if (err)
                goto out4;
@@ -1185,6 +1268,8 @@ out2:
 out3:
        nfs_destroy_nfspagecache();
 out4:
+       nfs_fs_proc_exit();
+out5:
        return err;
 }
 
@@ -1199,11 +1284,13 @@ static void __exit exit_nfs_fs(void)
        rpc_proc_unregister("nfs");
 #endif
        unregister_nfs_fs();
+       nfs_fs_proc_exit();
 }
 
 /* Not quite true; I just maintain it */
 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
 MODULE_LICENSE("GPL");
+module_param(enable_ino64, bool, 0644);
 
 module_init(init_nfs_fs)
 module_exit(exit_nfs_fs)