cifs: fix noserverino handling when unix extensions are enabled
[safe/jmp/linux-2.6] / fs / ocfs2 / inode.c
index 00cd5b7..af18988 100644 (file)
@@ -25,9 +25,9 @@
 
 #include <linux/fs.h>
 #include <linux/types.h>
-#include <linux/slab.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
+#include <linux/quotaops.h>
 
 #include <asm/byteorder.h>
 
@@ -37,6 +37,8 @@
 #include "ocfs2.h"
 
 #include "alloc.h"
+#include "dir.h"
+#include "blockcheck.h"
 #include "dlmglue.h"
 #include "extent_map.h"
 #include "file.h"
@@ -49,6 +51,8 @@
 #include "symlink.h"
 #include "sysfile.h"
 #include "uptodate.h"
+#include "xattr.h"
+#include "refcounttree.h"
 
 #include "buffer_head_io.h"
 
@@ -57,8 +61,11 @@ struct ocfs2_find_inode_args
        u64             fi_blkno;
        unsigned long   fi_ino;
        unsigned int    fi_flags;
+       unsigned int    fi_sysfile_type;
 };
 
+static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
+
 static int ocfs2_read_locked_inode(struct inode *inode,
                                   struct ocfs2_find_inode_args *args);
 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
@@ -106,7 +113,19 @@ void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
                oi->ip_attr |= OCFS2_DIRSYNC_FL;
 }
 
-struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
+struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
+{
+       struct ocfs2_find_inode_args args;
+
+       args.fi_blkno = blkno;
+       args.fi_flags = 0;
+       args.fi_ino = ino_from_blkno(sb, blkno);
+       args.fi_sysfile_type = 0;
+
+       return ilookup5(sb, blkno, ocfs2_find_actor, &args);
+}
+struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
+                        int sysfile_type)
 {
        struct inode *inode = NULL;
        struct super_block *sb = osb->sb;
@@ -126,6 +145,7 @@ struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
        args.fi_blkno = blkno;
        args.fi_flags = flags;
        args.fi_ino = ino_from_blkno(sb, blkno);
+       args.fi_sysfile_type = sysfile_type;
 
        inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
                             ocfs2_init_locked_inode, &args);
@@ -195,22 +215,36 @@ bail:
 static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
 {
        struct ocfs2_find_inode_args *args = opaque;
+       static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
+                                    ocfs2_file_ip_alloc_sem_key;
 
        mlog_entry("inode = %p, opaque = %p\n", inode, opaque);
 
        inode->i_ino = args->fi_ino;
        OCFS2_I(inode)->ip_blkno = args->fi_blkno;
+       if (args->fi_sysfile_type != 0)
+               lockdep_set_class(&inode->i_mutex,
+                       &ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
+       if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
+           args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
+           args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
+           args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
+               lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
+                                 &ocfs2_quota_ip_alloc_sem_key);
+       else
+               lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
+                                 &ocfs2_file_ip_alloc_sem_key);
 
        mlog_exit(0);
        return 0;
 }
 
-int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
-                        int create_ino)
+void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
+                         int create_ino)
 {
        struct super_block *sb;
        struct ocfs2_super *osb;
-       int status = -EINVAL;
+       int use_plocks = 1;
 
        mlog_entry("(0x%p, size:%llu)\n", inode,
                   (unsigned long long)le64_to_cpu(fe->i_size));
@@ -218,25 +252,21 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
        sb = inode->i_sb;
        osb = OCFS2_SB(sb);
 
-       /* this means that read_inode cannot create a superblock inode
-        * today.  change if needed. */
-       if (!OCFS2_IS_VALID_DINODE(fe) ||
-           !(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
-               mlog(0, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
-                    "signature = %.*s, flags = 0x%x\n",
-                    inode->i_ino,
-                    (unsigned long long)le64_to_cpu(fe->i_blkno), 7,
-                    fe->i_signature, le32_to_cpu(fe->i_flags));
-               goto bail;
-       }
+       if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
+           ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
+               use_plocks = 0;
+
+       /*
+        * These have all been checked by ocfs2_read_inode_block() or set
+        * by ocfs2_mknod_locked(), so a failure is a code bug.
+        */
+       BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
+                                               cannot create a superblock
+                                               inode today.  change if
+                                               that is needed. */
+       BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
+       BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
 
-       if (le32_to_cpu(fe->i_fs_generation) != osb->fs_generation) {
-               mlog(ML_ERROR, "file entry generation does not match "
-                    "superblock! osb->fs_generation=%x, "
-                    "fe->i_fs_generation=%x\n",
-                    osb->fs_generation, le32_to_cpu(fe->i_fs_generation));
-               goto bail;
-       }
 
        OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
        OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
@@ -268,16 +298,20 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
                     (unsigned long long)OCFS2_I(inode)->ip_blkno,
                     (unsigned long long)le64_to_cpu(fe->i_blkno));
 
-       inode->i_nlink = le16_to_cpu(fe->i_links_count);
+       inode->i_nlink = ocfs2_read_links_count(fe);
 
-       if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL))
+       if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
                OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
+               inode->i_flags |= S_NOQUOTA;
+       }
 
        if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
                OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
                mlog(0, "local alloc inode: i_ino=%lu\n", inode->i_ino);
        } else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
                OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
+       } else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
+               inode->i_flags |= S_NOQUOTA;
        } else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
                mlog(0, "superblock inode: i_ino=%lu\n", inode->i_ino);
                /* we can't actually hit this as read_inode can't
@@ -287,13 +321,19 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 
        switch (inode->i_mode & S_IFMT) {
            case S_IFREG:
-                   inode->i_fop = &ocfs2_fops;
+                   if (use_plocks)
+                           inode->i_fop = &ocfs2_fops;
+                   else
+                           inode->i_fop = &ocfs2_fops_no_plocks;
                    inode->i_op = &ocfs2_file_iops;
                    i_size_write(inode, le64_to_cpu(fe->i_size));
                    break;
            case S_IFDIR:
                    inode->i_op = &ocfs2_dir_iops;
-                   inode->i_fop = &ocfs2_dops;
+                   if (use_plocks)
+                           inode->i_fop = &ocfs2_dops;
+                   else
+                           inode->i_fop = &ocfs2_dops_no_plocks;
                    i_size_write(inode, le64_to_cpu(fe->i_size));
                    break;
            case S_IFLNK:
@@ -334,10 +374,9 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
 
        ocfs2_set_inode_flags(inode);
 
-       status = 0;
-bail:
-       mlog_exit(status);
-       return status;
+       OCFS2_I(inode)->ip_last_used_slot = 0;
+       OCFS2_I(inode)->ip_last_used_group = 0;
+       mlog_exit_void();
 }
 
 static int ocfs2_read_locked_inode(struct inode *inode,
@@ -435,13 +474,19 @@ static int ocfs2_read_locked_inode(struct inode *inode,
        if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
                status = ocfs2_try_open_lock(inode, 0);
                if (status) {
-                       make_bad_inode(inode);  
+                       make_bad_inode(inode);
                        return status;
                }
        }
 
-       status = ocfs2_read_block(osb, args->fi_blkno, &bh, 0,
-                                 can_lock ? inode : NULL);
+       if (can_lock) {
+               status = ocfs2_read_inode_block_full(inode, &bh,
+                                                    OCFS2_BH_IGNORE_CACHE);
+       } else {
+               status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
+               if (!status)
+                       status = ocfs2_validate_inode_block(osb->sb, bh);
+       }
        if (status < 0) {
                mlog_errno(status);
                goto bail;
@@ -449,12 +494,6 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 
        status = -EINVAL;
        fe = (struct ocfs2_dinode *) bh->b_data;
-       if (!OCFS2_IS_VALID_DINODE(fe)) {
-               mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
-                    (unsigned long long)args->fi_blkno, 7,
-                    fe->i_signature);
-               goto bail;
-       }
 
        /*
         * This is a code bug. Right now the caller needs to
@@ -468,10 +507,9 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 
        if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
            S_ISBLK(le16_to_cpu(fe->i_mode)))
-               inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
+               inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 
-       if (ocfs2_populate_inode(inode, fe, 0) < 0)
-               goto bail;
+       ocfs2_populate_inode(inode, fe, 0);
 
        BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 
@@ -514,15 +552,20 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
         * data and fast symlinks.
         */
        if (fe->i_clusters) {
+               if (ocfs2_should_order_data(inode))
+                       ocfs2_begin_ordered_truncate(inode, 0);
+
                handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
                if (IS_ERR(handle)) {
                        status = PTR_ERR(handle);
+                       handle = NULL;
                        mlog_errno(status);
                        goto out;
                }
 
-               status = ocfs2_journal_access(handle, inode, fe_bh,
-                                             OCFS2_JOURNAL_ACCESS_WRITE);
+               status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+                                                fe_bh,
+                                                OCFS2_JOURNAL_ACCESS_WRITE);
                if (status < 0) {
                        mlog_errno(status);
                        goto out;
@@ -589,30 +632,33 @@ static int ocfs2_remove_inode(struct inode *inode,
                goto bail;
        }
 
-       handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS);
+       handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
+                                  ocfs2_quota_trans_credits(inode->i_sb));
        if (IS_ERR(handle)) {
                status = PTR_ERR(handle);
                mlog_errno(status);
                goto bail_unlock;
        }
 
-       status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
-                                 orphan_dir_bh);
-       if (status < 0) {
-               mlog_errno(status);
-               goto bail_commit;
+       if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+               status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
+                                         orphan_dir_bh);
+               if (status < 0) {
+                       mlog_errno(status);
+                       goto bail_commit;
+               }
        }
 
        /* set the inodes dtime */
-       status = ocfs2_journal_access(handle, inode, di_bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+                                        OCFS2_JOURNAL_ACCESS_WRITE);
        if (status < 0) {
                mlog_errno(status);
                goto bail_commit;
        }
 
        di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
-       le32_and_cpu(&di->i_flags, ~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
+       di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
 
        status = ocfs2_journal_dirty(handle, di_bh);
        if (status < 0) {
@@ -620,7 +666,8 @@ static int ocfs2_remove_inode(struct inode *inode,
                goto bail_commit;
        }
 
-       ocfs2_remove_from_cache(inode, di_bh);
+       ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
+       dquot_free_inode(inode);
 
        status = ocfs2_free_dinode(handle, inode_alloc_inode,
                                   inode_alloc_bh, di);
@@ -639,7 +686,7 @@ bail:
        return status;
 }
 
-/* 
+/*
  * Serialize with orphan dir recovery. If the process doing
  * recovery on this orphan dir does an iget() with the dir
  * i_mutex held, we'll deadlock here. Instead we detect this
@@ -678,38 +725,39 @@ static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
 static int ocfs2_wipe_inode(struct inode *inode,
                            struct buffer_head *di_bh)
 {
-       int status, orphaned_slot;
+       int status, orphaned_slot = -1;
        struct inode *orphan_dir_inode = NULL;
        struct buffer_head *orphan_dir_bh = NULL;
        struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
-       struct ocfs2_dinode *di;
+       struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
 
-       di = (struct ocfs2_dinode *) di_bh->b_data;
-       orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
+       if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+               orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
 
-       status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
-       if (status)
-               return status;
+               status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
+               if (status)
+                       return status;
 
-       orphan_dir_inode = ocfs2_get_system_file_inode(osb,
-                                                      ORPHAN_DIR_SYSTEM_INODE,
-                                                      orphaned_slot);
-       if (!orphan_dir_inode) {
-               status = -EEXIST;
-               mlog_errno(status);
-               goto bail;
-       }
+               orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+                                                              ORPHAN_DIR_SYSTEM_INODE,
+                                                              orphaned_slot);
+               if (!orphan_dir_inode) {
+                       status = -EEXIST;
+                       mlog_errno(status);
+                       goto bail;
+               }
 
-       /* Lock the orphan dir. The lock will be held for the entire
-        * delete_inode operation. We do this now to avoid races with
-        * recovery completion on other nodes. */
-       mutex_lock(&orphan_dir_inode->i_mutex);
-       status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
-       if (status < 0) {
-               mutex_unlock(&orphan_dir_inode->i_mutex);
+               /* Lock the orphan dir. The lock will be held for the entire
+                * delete_inode operation. We do this now to avoid races with
+                * recovery completion on other nodes. */
+               mutex_lock(&orphan_dir_inode->i_mutex);
+               status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+               if (status < 0) {
+                       mutex_unlock(&orphan_dir_inode->i_mutex);
 
-               mlog_errno(status);
-               goto bail;
+                       mlog_errno(status);
+                       goto bail;
+               }
        }
 
        /* we do this while holding the orphan dir lock because we
@@ -722,12 +770,37 @@ static int ocfs2_wipe_inode(struct inode *inode,
                goto bail_unlock_dir;
        }
 
+       /* Remove any dir index tree */
+       if (S_ISDIR(inode->i_mode)) {
+               status = ocfs2_dx_dir_truncate(inode, di_bh);
+               if (status) {
+                       mlog_errno(status);
+                       goto bail_unlock_dir;
+               }
+       }
+
+       /*Free extended attribute resources associated with this inode.*/
+       status = ocfs2_xattr_remove(inode, di_bh);
+       if (status < 0) {
+               mlog_errno(status);
+               goto bail_unlock_dir;
+       }
+
+       status = ocfs2_remove_refcount_tree(inode, di_bh);
+       if (status < 0) {
+               mlog_errno(status);
+               goto bail_unlock_dir;
+       }
+
        status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
                                    orphan_dir_bh);
        if (status < 0)
                mlog_errno(status);
 
 bail_unlock_dir:
+       if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
+               return status;
+
        ocfs2_inode_unlock(orphan_dir_inode, 1);
        mutex_unlock(&orphan_dir_inode->i_mutex);
        brelse(orphan_dir_bh);
@@ -823,7 +896,23 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
 
        /* Do some basic inode verification... */
        di = (struct ocfs2_dinode *) di_bh->b_data;
-       if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL))) {
+       if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
+           !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+               /*
+                * Inodes in the orphan dir must have ORPHANED_FL.  The only
+                * inodes that come back out of the orphan dir are reflink
+                * targets. A reflink target may be moved out of the orphan
+                * dir between the time we scan the directory and the time we
+                * process it. This would lead to HAS_REFCOUNT_FL being set but
+                * ORPHANED_FL not.
+                */
+               if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
+                       mlog(0, "Reflinked inode %llu is no longer orphaned.  "
+                            "it shouldn't be deleted\n",
+                            (unsigned long long)oi->ip_blkno);
+                       goto bail;
+               }
+
                /* for lack of a better error? */
                status = -EEXIST;
                mlog(ML_ERROR,
@@ -896,11 +985,16 @@ void ocfs2_delete_inode(struct inode *inode)
 
        mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
 
-       if (is_bad_inode(inode)) {
+       /* When we fail in read_inode() we mark inode as bad. The second test
+        * catches the case when inode allocation fails before allocating
+        * a block for inode. */
+       if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno) {
                mlog(0, "Skipping delete of bad inode\n");
                goto bail;
        }
 
+       dquot_initialize(inode);
+
        if (!ocfs2_inode_is_valid_to_delete(inode)) {
                /* It's probably not necessary to truncate_inode_pages
                 * here but we do it for safety anyway (it will most
@@ -921,6 +1015,17 @@ void ocfs2_delete_inode(struct inode *inode)
                goto bail;
        }
 
+       /*
+        * Synchronize us against ocfs2_get_dentry. We take this in
+        * shared mode so that all nodes can still concurrently
+        * process deletes.
+        */
+       status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
+       if (status < 0) {
+               mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
+               ocfs2_cleanup_delete_inode(inode, 0);
+               goto bail_unblock;
+       }
        /* Lock down the inode. This gives us an up to date view of
         * it's metadata (for verification), and allows us to
         * serialize delete_inode on multiple nodes.
@@ -934,7 +1039,7 @@ void ocfs2_delete_inode(struct inode *inode)
                if (status != -ENOENT)
                        mlog_errno(status);
                ocfs2_cleanup_delete_inode(inode, 0);
-               goto bail_unblock;
+               goto bail_unlock_nfs_sync;
        }
 
        /* Query the cluster. This will be the final decision made
@@ -977,6 +1082,10 @@ void ocfs2_delete_inode(struct inode *inode)
 bail_unlock_inode:
        ocfs2_inode_unlock(inode, 1);
        brelse(di_bh);
+
+bail_unlock_nfs_sync:
+       ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
+
 bail_unblock:
        status = sigprocmask(SIG_SETMASK, &oldset, NULL);
        if (status < 0)
@@ -1002,6 +1111,8 @@ void ocfs2_clear_inode(struct inode *inode)
        mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
                        "Inode=%lu\n", inode->i_ino);
 
+       dquot_drop(inode);
+
        /* To preven remote deletes we hold open lock before, now it
         * is time to unlock PR and EX open locks. */
        ocfs2_open_unlock(inode);
@@ -1035,13 +1146,14 @@ void ocfs2_clear_inode(struct inode *inode)
        ocfs2_lock_res_free(&oi->ip_inode_lockres);
        ocfs2_lock_res_free(&oi->ip_open_lockres);
 
-       ocfs2_metadata_cache_purge(inode);
+       ocfs2_metadata_cache_exit(INODE_CACHE(inode));
 
-       mlog_bug_on_msg(oi->ip_metadata_cache.ci_num_cached,
+       mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
                        "Clear inode of %llu, inode has %u cache items\n",
-                       (unsigned long long)oi->ip_blkno, oi->ip_metadata_cache.ci_num_cached);
+                       (unsigned long long)oi->ip_blkno,
+                       INODE_CACHE(inode)->ci_num_cached);
 
-       mlog_bug_on_msg(!(oi->ip_flags & OCFS2_INODE_CACHE_INLINE),
+       mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
                        "Clear inode of %llu, inode has a bad flag\n",
                        (unsigned long long)oi->ip_blkno);
 
@@ -1068,12 +1180,18 @@ void ocfs2_clear_inode(struct inode *inode)
                        (unsigned long long)oi->ip_blkno, oi->ip_open_count);
 
        /* Clear all other flags. */
-       oi->ip_flags = OCFS2_INODE_CACHE_INLINE;
-       oi->ip_created_trans = 0;
-       oi->ip_last_trans = 0;
+       oi->ip_flags = 0;
        oi->ip_dir_start_lookup = 0;
        oi->ip_blkno = 0ULL;
 
+       /*
+        * ip_jinode is used to track txns against this inode. We ensure that
+        * the journal is flushed before journal shutdown. Thus it is safe to
+        * have inodes get cleaned up after journal shutdown.
+        */
+       jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
+                                      &oi->ip_jinode);
+
 bail:
        mlog_exit_void();
 }
@@ -1099,58 +1217,6 @@ void ocfs2_drop_inode(struct inode *inode)
 }
 
 /*
- * TODO: this should probably be merged into ocfs2_get_block
- *
- * However, you now need to pay attention to the cont_prepare_write()
- * stuff in ocfs2_get_block (that is, ocfs2_get_block pretty much
- * expects never to extend).
- */
-struct buffer_head *ocfs2_bread(struct inode *inode,
-                               int block, int *err, int reada)
-{
-       struct buffer_head *bh = NULL;
-       int tmperr;
-       u64 p_blkno;
-       int readflags = OCFS2_BH_CACHED;
-
-       if (reada)
-               readflags |= OCFS2_BH_READAHEAD;
-
-       if (((u64)block << inode->i_sb->s_blocksize_bits) >=
-           i_size_read(inode)) {
-               BUG_ON(!reada);
-               return NULL;
-       }
-
-       down_read(&OCFS2_I(inode)->ip_alloc_sem);
-       tmperr = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
-                                            NULL);
-       up_read(&OCFS2_I(inode)->ip_alloc_sem);
-       if (tmperr < 0) {
-               mlog_errno(tmperr);
-               goto fail;
-       }
-
-       tmperr = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno, &bh,
-                                 readflags, inode);
-       if (tmperr < 0)
-               goto fail;
-
-       tmperr = 0;
-
-       *err = 0;
-       return bh;
-
-fail:
-       if (bh) {
-               brelse(bh);
-               bh = NULL;
-       }
-       *err = -EIO;
-       return NULL;
-}
-
-/*
  * This is called from our getattr.
  */
 int ocfs2_inode_revalidate(struct dentry *dentry)
@@ -1206,8 +1272,8 @@ int ocfs2_mark_inode_dirty(handle_t *handle,
        mlog_entry("(inode %llu)\n",
                   (unsigned long long)OCFS2_I(inode)->ip_blkno);
 
-       status = ocfs2_journal_access(handle, inode, bh,
-                                     OCFS2_JOURNAL_ACCESS_WRITE);
+       status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+                                        OCFS2_JOURNAL_ACCESS_WRITE);
        if (status < 0) {
                mlog_errno(status);
                goto leave;
@@ -1221,7 +1287,7 @@ int ocfs2_mark_inode_dirty(handle_t *handle,
        spin_unlock(&OCFS2_I(inode)->ip_lock);
 
        fe->i_size = cpu_to_le64(i_size_read(inode));
-       fe->i_links_count = cpu_to_le16(inode->i_nlink);
+       ocfs2_set_links_count(fe, inode->i_nlink);
        fe->i_uid = cpu_to_le32(inode->i_uid);
        fe->i_gid = cpu_to_le32(inode->i_gid);
        fe->i_mode = cpu_to_le16(inode->i_mode);
@@ -1258,7 +1324,7 @@ void ocfs2_refresh_inode(struct inode *inode,
        OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
        ocfs2_set_inode_flags(inode);
        i_size_write(inode, le64_to_cpu(fe->i_size));
-       inode->i_nlink = le16_to_cpu(fe->i_links_count);
+       inode->i_nlink = ocfs2_read_links_count(fe);
        inode->i_uid = le32_to_cpu(fe->i_uid);
        inode->i_gid = le32_to_cpu(fe->i_gid);
        inode->i_mode = le16_to_cpu(fe->i_mode);
@@ -1275,3 +1341,142 @@ void ocfs2_refresh_inode(struct inode *inode,
 
        spin_unlock(&OCFS2_I(inode)->ip_lock);
 }
+
+int ocfs2_validate_inode_block(struct super_block *sb,
+                              struct buffer_head *bh)
+{
+       int rc;
+       struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
+
+       mlog(0, "Validating dinode %llu\n",
+            (unsigned long long)bh->b_blocknr);
+
+       BUG_ON(!buffer_uptodate(bh));
+
+       /*
+        * If the ecc fails, we return the error but otherwise
+        * leave the filesystem running.  We know any error is
+        * local to this block.
+        */
+       rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
+       if (rc) {
+               mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
+                    (unsigned long long)bh->b_blocknr);
+               goto bail;
+       }
+
+       /*
+        * Errors after here are fatal.
+        */
+
+       rc = -EINVAL;
+
+       if (!OCFS2_IS_VALID_DINODE(di)) {
+               ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
+                           (unsigned long long)bh->b_blocknr, 7,
+                           di->i_signature);
+               goto bail;
+       }
+
+       if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
+               ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
+                           (unsigned long long)bh->b_blocknr,
+                           (unsigned long long)le64_to_cpu(di->i_blkno));
+               goto bail;
+       }
+
+       if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
+               ocfs2_error(sb,
+                           "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
+                           (unsigned long long)bh->b_blocknr);
+               goto bail;
+       }
+
+       if (le32_to_cpu(di->i_fs_generation) !=
+           OCFS2_SB(sb)->fs_generation) {
+               ocfs2_error(sb,
+                           "Invalid dinode #%llu: fs_generation is %u\n",
+                           (unsigned long long)bh->b_blocknr,
+                           le32_to_cpu(di->i_fs_generation));
+               goto bail;
+       }
+
+       rc = 0;
+
+bail:
+       return rc;
+}
+
+int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
+                               int flags)
+{
+       int rc;
+       struct buffer_head *tmp = *bh;
+
+       rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
+                              1, &tmp, flags, ocfs2_validate_inode_block);
+
+       /* If ocfs2_read_blocks() got us a new bh, pass it up. */
+       if (!rc && !*bh)
+               *bh = tmp;
+
+       return rc;
+}
+
+int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
+{
+       return ocfs2_read_inode_block_full(inode, bh, 0);
+}
+
+
+static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       return oi->ip_blkno;
+}
+
+static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       return oi->vfs_inode.i_sb;
+}
+
+static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       spin_lock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       spin_unlock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       mutex_lock(&oi->ip_io_mutex);
+}
+
+static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+       struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+       mutex_unlock(&oi->ip_io_mutex);
+}
+
+const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
+       .co_owner               = ocfs2_inode_cache_owner,
+       .co_get_super           = ocfs2_inode_cache_get_super,
+       .co_cache_lock          = ocfs2_inode_cache_lock,
+       .co_cache_unlock        = ocfs2_inode_cache_unlock,
+       .co_io_lock             = ocfs2_inode_cache_io_lock,
+       .co_io_unlock           = ocfs2_inode_cache_io_unlock,
+};
+