vfs: trivial __d_lookup_rcu() cleanups
[linux-2.6.git] / fs / nfs / inode.c
index ec7a8f9..f649fba 100644 (file)
@@ -37,6 +37,8 @@
 #include <linux/inet.h>
 #include <linux/nfs_xdr.h>
 #include <linux/slab.h>
+#include <linux/compat.h>
+#include <linux/freezer.h>
 
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include "internal.h"
 #include "fscache.h"
 #include "dns_resolve.h"
+#include "pnfs.h"
 
 #define NFSDBG_FACILITY                NFSDBG_VFS
 
 #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 bool 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 *);
@@ -75,7 +78,7 @@ int nfs_wait_bit_killable(void *word)
 {
        if (fatal_signal_pending(current))
                return -ERESTARTSYS;
-       schedule();
+       freezable_schedule();
        return 0;
 }
 
@@ -88,7 +91,11 @@ int nfs_wait_bit_killable(void *word)
  */
 u64 nfs_compat_user_ino64(u64 fileid)
 {
-       int ino;
+#ifdef CONFIG_COMPAT
+       compat_ulong_t ino;
+#else  
+       unsigned long ino;
+#endif
 
        if (enable_ino64)
                return fileid;
@@ -98,7 +105,7 @@ u64 nfs_compat_user_ino64(u64 fileid)
        return ino;
 }
 
-void nfs_clear_inode(struct inode *inode)
+static void nfs_clear_inode(struct inode *inode)
 {
        /*
         * The following should never happen...
@@ -110,6 +117,13 @@ void nfs_clear_inode(struct inode *inode)
        nfs_fscache_release_inode_cookie(inode);
 }
 
+void nfs_evict_inode(struct inode *inode)
+{
+       truncate_inode_pages(&inode->i_data, 0);
+       end_writeback(inode);
+       nfs_clear_inode(inode);
+}
+
 /**
  * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
  */
@@ -227,9 +241,6 @@ nfs_init_locked(struct inode *inode, void *opaque)
        return 0;
 }
 
-/* Don't use READDIRPLUS on directories that we believe are too large */
-#define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
-
 /*
  * This is our front-end to iget that looks up inodes by file handle
  * instead of inode number.
@@ -244,7 +255,10 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
        struct inode *inode = ERR_PTR(-ENOENT);
        unsigned long hash;
 
-       if ((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0)
+       nfs_attr_check_mountpoint(sb, fattr);
+
+       if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
+           !nfs_attr_use_mounted_on_fileid(fattr))
                goto out_no_inode;
        if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
                goto out_no_inode;
@@ -278,24 +292,24 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                 */
                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_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
                        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)->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)
+                       inode->i_data.a_ops = &nfs_dir_aops;
+                       if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS))
                                set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
                        /* Deal with crossing mountpoints */
-                       if ((fattr->valid & NFS_ATTR_FATTR_FSID)
-                                       && !nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
+                       if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
+                                       fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
                                if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
                                        inode->i_op = &nfs_referral_inode_operations;
                                else
                                        inode->i_op = &nfs_mountpoint_inode_operations;
                                inode->i_fop = NULL;
-                               set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
+                               inode->i_flags |= S_AUTOMOUNT;
                        }
                } else if (S_ISLNK(inode->i_mode))
                        inode->i_op = &nfs_symlink_inode_operations;
@@ -305,9 +319,9 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                memset(&inode->i_atime, 0, sizeof(inode->i_atime));
                memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
                memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
-               nfsi->change_attr = 0;
+               inode->i_version = 0;
                inode->i_size = 0;
-               inode->i_nlink = 0;
+               clear_nlink(inode);
                inode->i_uid = -2;
                inode->i_gid = -2;
                inode->i_blocks = 0;
@@ -331,7 +345,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                                | NFS_INO_INVALID_ACCESS
                                | NFS_INO_INVALID_ACL;
                if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
-                       nfsi->change_attr = fattr->change_attr;
+                       inode->i_version = fattr->change_attr;
                else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
                        nfsi->cache_validity |= NFS_INO_INVALID_ATTR
                                | NFS_INO_INVALID_DATA;
@@ -342,7 +356,7 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
                                | NFS_INO_INVALID_DATA
                                | NFS_INO_REVAL_PAGECACHE;
                if (fattr->valid & NFS_ATTR_FATTR_NLINK)
-                       inode->i_nlink = fattr->nlink;
+                       set_nlink(inode, fattr->nlink);
                else if (nfs_server_capable(inode, NFS_CAP_NLINK))
                        nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
                if (fattr->valid & NFS_ATTR_FATTR_OWNER)
@@ -413,10 +427,8 @@ nfs_setattr(struct dentry *dentry, struct iattr *attr)
                return 0;
 
        /* Write all dirty data */
-       if (S_ISREG(inode->i_mode)) {
-               filemap_write_and_wait(inode->i_mapping);
+       if (S_ISREG(inode->i_mode))
                nfs_wb_all(inode);
-       }
 
        fattr = nfs_alloc_fattr();
        if (fattr == NULL)
@@ -556,7 +568,7 @@ static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context
 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
 {
        struct nfs_lock_context *res, *new = NULL;
-       struct inode *inode = ctx->path.dentry->d_inode;
+       struct inode *inode = ctx->dentry->d_inode;
 
        spin_lock(&inode->i_lock);
        res = __nfs_find_lock_context(ctx);
@@ -583,7 +595,7 @@ struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
 {
        struct nfs_open_context *ctx = l_ctx->open_context;
-       struct inode *inode = ctx->path.dentry->d_inode;
+       struct inode *inode = ctx->dentry->d_inode;
 
        if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
                return;
@@ -609,7 +621,7 @@ void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
                return;
        if (!is_sync)
                return;
-       inode = ctx->path.dentry->d_inode;
+       inode = ctx->dentry->d_inode;
        if (!list_empty(&NFS_I(inode)->open_files))
                return;
        server = NFS_SERVER(inode);
@@ -618,22 +630,28 @@ void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
        nfs_revalidate_inode(server, inode);
 }
 
-static struct nfs_open_context *alloc_nfs_open_context(struct path *path, struct rpc_cred *cred)
+struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
 {
        struct nfs_open_context *ctx;
+       struct rpc_cred *cred = rpc_lookup_cred();
+       if (IS_ERR(cred))
+               return ERR_CAST(cred);
 
        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
-       if (ctx != NULL) {
-               ctx->path = *path;
-               path_get(&ctx->path);
-               ctx->cred = get_rpccred(cred);
-               ctx->state = NULL;
-               ctx->flags = 0;
-               ctx->error = 0;
-               ctx->dir_cookie = 0;
-               nfs_init_lock_context(&ctx->lock_context);
-               ctx->lock_context.open_context = ctx;
+       if (!ctx) {
+               put_rpccred(cred);
+               return ERR_PTR(-ENOMEM);
        }
+       nfs_sb_active(dentry->d_sb);
+       ctx->dentry = dget(dentry);
+       ctx->cred = cred;
+       ctx->state = NULL;
+       ctx->mode = f_mode;
+       ctx->flags = 0;
+       ctx->error = 0;
+       nfs_init_lock_context(&ctx->lock_context);
+       ctx->lock_context.open_context = ctx;
+       INIT_LIST_HEAD(&ctx->list);
        return ctx;
 }
 
@@ -646,16 +664,22 @@ struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
 
 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
 {
-       struct inode *inode = ctx->path.dentry->d_inode;
+       struct inode *inode = ctx->dentry->d_inode;
+       struct super_block *sb = ctx->dentry->d_sb;
 
-       if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
+       if (!list_empty(&ctx->list)) {
+               if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
+                       return;
+               list_del(&ctx->list);
+               spin_unlock(&inode->i_lock);
+       } else if (!atomic_dec_and_test(&ctx->lock_context.count))
                return;
-       list_del(&ctx->list);
-       spin_unlock(&inode->i_lock);
-       NFS_PROTO(inode)->close_context(ctx, is_sync);
+       if (inode != NULL)
+               NFS_PROTO(inode)->close_context(ctx, is_sync);
        if (ctx->cred != NULL)
                put_rpccred(ctx->cred);
-       path_put(&ctx->path);
+       dput(ctx->dentry);
+       nfs_sb_deactive(sb);
        kfree(ctx);
 }
 
@@ -668,7 +692,7 @@ void put_nfs_open_context(struct nfs_open_context *ctx)
  * Ensure that mmap has a recent RPC credential for use when writing out
  * shared pages
  */
-static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
+void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
 {
        struct inode *inode = filp->f_path.dentry->d_inode;
        struct nfs_inode *nfsi = NFS_I(inode);
@@ -720,16 +744,10 @@ static void nfs_file_clear_open_context(struct file *filp)
 int nfs_open(struct inode *inode, struct file *filp)
 {
        struct nfs_open_context *ctx;
-       struct rpc_cred *cred;
 
-       cred = rpc_lookup_cred();
-       if (IS_ERR(cred))
-               return PTR_ERR(cred);
-       ctx = alloc_nfs_open_context(&filp->f_path, cred);
-       put_rpccred(cred);
-       if (ctx == NULL)
-               return -ENOMEM;
-       ctx->mode = filp->f_mode;
+       ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
+       if (IS_ERR(ctx))
+               return PTR_ERR(ctx);
        nfs_file_set_open_context(filp, ctx);
        put_nfs_open_context(ctx);
        nfs_fscache_set_inode_cookie(inode, filp);
@@ -873,35 +891,43 @@ out:
        return ret;
 }
 
-static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
+static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
        struct nfs_inode *nfsi = NFS_I(inode);
+       unsigned long ret = 0;
 
        if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
                        && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
-                       && nfsi->change_attr == fattr->pre_change_attr) {
-               nfsi->change_attr = fattr->change_attr;
+                       && inode->i_version == fattr->pre_change_attr) {
+               inode->i_version = fattr->change_attr;
                if (S_ISDIR(inode->i_mode))
                        nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+               ret |= NFS_INO_INVALID_ATTR;
        }
        /* If we have atomic WCC data, we may update some attributes */
        if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
                        && (fattr->valid & NFS_ATTR_FATTR_CTIME)
-                       && timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
-                       memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
+                       && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
+               memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
+               ret |= NFS_INO_INVALID_ATTR;
+       }
 
        if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
                        && (fattr->valid & NFS_ATTR_FATTR_MTIME)
                        && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
-                       memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
-                       if (S_ISDIR(inode->i_mode))
-                               nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+               memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
+               if (S_ISDIR(inode->i_mode))
+                       nfsi->cache_validity |= NFS_INO_INVALID_DATA;
+               ret |= NFS_INO_INVALID_ATTR;
        }
        if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
                        && (fattr->valid & NFS_ATTR_FATTR_SIZE)
                        && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
-                       && nfsi->npages == 0)
-                       i_size_write(inode, nfs_size_to_loff_t(fattr->size));
+                       && nfsi->npages == 0) {
+               i_size_write(inode, nfs_size_to_loff_t(fattr->size));
+               ret |= NFS_INO_INVALID_ATTR;
+       }
+       return ret;
 }
 
 /**
@@ -927,7 +953,7 @@ static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fat
                return -EIO;
 
        if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
-                       nfsi->change_attr != fattr->change_attr)
+                       inode->i_version != fattr->change_attr)
                invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
 
        /* Verify a few of the more important attributes */
@@ -994,6 +1020,8 @@ void nfs_fattr_init(struct nfs_fattr *fattr)
        fattr->valid = 0;
        fattr->time_start = jiffies;
        fattr->gencount = nfs_inc_attr_generation_counter();
+       fattr->owner_name = NULL;
+       fattr->group_name = NULL;
 }
 
 struct nfs_fattr *nfs_alloc_fattr(void)
@@ -1138,7 +1166,7 @@ int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fa
        }
        if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
                        (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
-               fattr->pre_change_attr = NFS_I(inode)->change_attr;
+               fattr->pre_change_attr = inode->i_version;
                fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
        }
        if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
@@ -1200,7 +1228,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
        /* Update the fsid? */
        if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
                        !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
-                       !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
+                       !IS_AUTOMOUNT(inode))
                server->fsid = fattr->fsid;
 
        /*
@@ -1215,17 +1243,17 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                        | NFS_INO_REVAL_PAGECACHE);
 
        /* Do atomic weak cache consistency updates */
-       nfs_wcc_update_inode(inode, fattr);
+       invalid |= nfs_wcc_update_inode(inode, fattr);
 
        /* More cache consistency checks */
        if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
-               if (nfsi->change_attr != fattr->change_attr) {
+               if (inode->i_version != 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);
-                       nfsi->change_attr = fattr->change_attr;
+                       inode->i_version = fattr->change_attr;
                }
        } else if (server->caps & NFS_CAP_CHANGE_ATTR)
                invalid |= save_cache_validity;
@@ -1272,12 +1300,17 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                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) {
+                       if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
+                            new_isize > cur_isize) {
                                i_size_write(inode, new_isize);
                                invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
                        }
-                       dprintk("NFS: isize change on server for file %s/%ld\n",
-                                       inode->i_sb->s_id, inode->i_ino);
+                       dprintk("NFS: isize change on server for file %s/%ld "
+                                       "(%Ld to %Ld)\n",
+                                       inode->i_sb->s_id,
+                                       inode->i_ino,
+                                       (long long)cur_isize,
+                                       (long long)new_isize);
                }
        } else
                invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
@@ -1331,7 +1364,7 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
                        invalid |= NFS_INO_INVALID_ATTR;
                        if (S_ISDIR(inode->i_mode))
                                invalid |= NFS_INO_INVALID_DATA;
-                       inode->i_nlink = fattr->nlink;
+                       set_nlink(inode, fattr->nlink);
                }
        } else if (server->caps & NFS_CAP_NLINK)
                invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
@@ -1400,8 +1433,12 @@ static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
  * to open() calls that passed nfs_atomic_lookup, but failed to call
  * nfs_open().
  */
-void nfs4_clear_inode(struct inode *inode)
+void nfs4_evict_inode(struct inode *inode)
 {
+       truncate_inode_pages(&inode->i_data, 0);
+       end_writeback(inode);
+       pnfs_return_layout(inode);
+       pnfs_destroy_layout(NFS_I(inode));
        /* If we are holding a delegation, return it! */
        nfs_inode_return_delegation_noreclaim(inode);
        /* First call standard NFS clear_inode() code */
@@ -1427,11 +1464,17 @@ struct inode *nfs_alloc_inode(struct super_block *sb)
        return &nfsi->vfs_inode;
 }
 
-void nfs_destroy_inode(struct inode *inode)
+static void nfs_i_callback(struct rcu_head *head)
 {
+       struct inode *inode = container_of(head, struct inode, i_rcu);
        kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
 }
 
+void nfs_destroy_inode(struct inode *inode)
+{
+       call_rcu(&inode->i_rcu, nfs_i_callback);
+}
+
 static inline void nfs4_init_once(struct nfs_inode *nfsi)
 {
 #ifdef CONFIG_NFS_V4
@@ -1439,6 +1482,8 @@ static inline void nfs4_init_once(struct nfs_inode *nfsi)
        nfsi->delegation = NULL;
        nfsi->delegation_state = 0;
        init_rwsem(&nfsi->rwsem);
+       nfsi->layout = NULL;
+       atomic_set(&nfsi->commits_outstanding, 0);
 #endif
 }
 
@@ -1486,7 +1531,7 @@ static int nfsiod_start(void)
 {
        struct workqueue_struct *wq;
        dprintk("RPC:       creating workqueue nfsiod\n");
-       wq = create_singlethread_workqueue("nfsiod");
+       wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
        if (wq == NULL)
                return -ENOMEM;
        nfsiod_workqueue = wq;
@@ -1514,6 +1559,10 @@ static int __init init_nfs_fs(void)
 {
        int err;
 
+       err = nfs_idmap_init();
+       if (err < 0)
+               goto out9;
+
        err = nfs_dns_resolver_init();
        if (err < 0)
                goto out8;
@@ -1578,6 +1627,8 @@ out6:
 out7:
        nfs_dns_resolver_destroy();
 out8:
+       nfs_idmap_quit();
+out9:
        return err;
 }
 
@@ -1590,9 +1641,11 @@ static void __exit exit_nfs_fs(void)
        nfs_destroy_nfspagecache();
        nfs_fscache_unregister();
        nfs_dns_resolver_destroy();
+       nfs_idmap_quit();
 #ifdef CONFIG_PROC_FS
        rpc_proc_unregister("nfs");
 #endif
+       nfs_cleanup_cb_ident_idr();
        unregister_nfs_fs();
        nfs_fs_proc_exit();
        nfsiod_stop();