ocfs2: Simplify ocfs2_read_block()
[safe/jmp/linux-2.6] / fs / ocfs2 / inode.c
index 42e361f..c5ee9e3 100644 (file)
@@ -28,7 +28,6 @@
 #include <linux/slab.h>
 #include <linux/highmem.h>
 #include <linux/pagemap.h>
-#include <linux/smp_lock.h>
 
 #include <asm/byteorder.h>
 
@@ -50,7 +49,7 @@
 #include "symlink.h"
 #include "sysfile.h"
 #include "uptodate.h"
-#include "vote.h"
+#include "xattr.h"
 
 #include "buffer_head_io.h"
 
@@ -59,8 +58,11 @@ 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);
@@ -89,25 +91,27 @@ void ocfs2_set_inode_flags(struct inode *inode)
                inode->i_flags |= S_DIRSYNC;
 }
 
-struct inode *ocfs2_ilookup_for_vote(struct ocfs2_super *osb,
-                                    u64 blkno,
-                                    int delete_vote)
+/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
+void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
 {
-       struct ocfs2_find_inode_args args;
-
-       /* ocfs2_ilookup_for_vote should *only* be called from the
-        * vote thread */
-       BUG_ON(current != osb->vote_task);
-
-       args.fi_blkno = blkno;
-       args.fi_flags = OCFS2_FI_FLAG_NOWAIT;
-       if (delete_vote)
-               args.fi_flags |= OCFS2_FI_FLAG_DELETE;
-       args.fi_ino = ino_from_blkno(osb->sb, blkno);
-       return ilookup5(osb->sb, args.fi_ino, ocfs2_find_actor, &args);
+       unsigned int flags = oi->vfs_inode.i_flags;
+
+       oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
+                       OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
+       if (flags & S_SYNC)
+               oi->ip_attr |= OCFS2_SYNC_FL;
+       if (flags & S_APPEND)
+               oi->ip_attr |= OCFS2_APPEND_FL;
+       if (flags & S_IMMUTABLE)
+               oi->ip_attr |= OCFS2_IMMUTABLE_FL;
+       if (flags & S_NOATIME)
+               oi->ip_attr |= OCFS2_NOATIME_FL;
+       if (flags & S_DIRSYNC)
+               oi->ip_attr |= OCFS2_DIRSYNC_FL;
 }
 
-struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
+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;
@@ -127,6 +131,7 @@ 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);
@@ -146,7 +151,6 @@ struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, int flags)
        if (is_bad_inode(inode)) {
                iput(inode);
                inode = ERR_PTR(-ESTALE);
-               mlog_errno(PTR_ERR(inode));
                goto bail;
        }
 
@@ -155,8 +159,7 @@ bail:
                mlog(0, "returning inode with number %llu\n",
                     (unsigned long long)OCFS2_I(inode)->ip_blkno);
                mlog_exit_ptr(inode);
-       } else
-               mlog_errno(PTR_ERR(inode));
+       }
 
        return inode;
 }
@@ -184,28 +187,6 @@ static int ocfs2_find_actor(struct inode *inode, void *opaque)
        if (oi->ip_blkno != args->fi_blkno)
                goto bail;
 
-       /* OCFS2_FI_FLAG_NOWAIT is *only* set from
-        * ocfs2_ilookup_for_vote which won't create an inode for one
-        * that isn't found. The vote thread which doesn't want to get
-        * an inode which is in the process of going away - otherwise
-        * the call to __wait_on_freeing_inode in find_inode_fast will
-        * cause it to deadlock on an inode which may be waiting on a
-        * vote (or lock release) in delete_inode */
-       if ((args->fi_flags & OCFS2_FI_FLAG_NOWAIT) &&
-           (inode->i_state & (I_FREEING|I_CLEAR))) {
-               /* As stated above, we're not going to return an
-                * inode.  In the case of a delete vote, the voting
-                * code is going to signal the other node to go
-                * ahead. Mark that state here, so this freeing inode
-                * has the state when it gets to delete_inode. */
-               if (args->fi_flags & OCFS2_FI_FLAG_DELETE) {
-                       spin_lock(&oi->ip_lock);
-                       ocfs2_mark_inode_remotely_deleted(inode);
-                       spin_unlock(&oi->ip_lock);
-               }
-               goto bail;
-       }
-
        ret = 1;
 bail:
        mlog_exit(ret);
@@ -225,6 +206,9 @@ static int ocfs2_init_locked_inode(struct inode *inode, void *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]);
 
        mlog_exit(0);
        return 0;
@@ -236,18 +220,23 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
        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)fe->i_size);
+                  (unsigned long long)le64_to_cpu(fe->i_size));
 
        sb = inode->i_sb;
        osb = OCFS2_SB(sb);
 
+       if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
+           ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
+               use_plocks = 0;
+
        /* 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(ML_ERROR, "Invalid dinode: i_ino=%lu, i_blkno=%llu, "
+               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,
@@ -263,6 +252,10 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
                goto bail;
        }
 
+       OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+       OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+       OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
+
        inode->i_version = 1;
        inode->i_generation = le32_to_cpu(fe->i_generation);
        inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
@@ -274,8 +267,7 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
        if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
                inode->i_blocks = 0;
        else
-               inode->i_blocks =
-                       ocfs2_align_bytes_to_sectors(le64_to_cpu(fe->i_size));
+               inode->i_blocks = ocfs2_inode_sector_count(inode);
        inode->i_mapping->a_ops = &ocfs2_aops;
        inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
        inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
@@ -288,11 +280,7 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
                mlog(ML_ERROR,
                     "ip_blkno %llu != i_blkno %llu!\n",
                     (unsigned long long)OCFS2_I(inode)->ip_blkno,
-                    (unsigned long long)fe->i_blkno);
-
-       OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
-       OCFS2_I(inode)->ip_orphaned_slot = OCFS2_INVALID_SLOT;
-       OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+                    (unsigned long long)le64_to_cpu(fe->i_blkno));
 
        inode->i_nlink = le16_to_cpu(fe->i_links_count);
 
@@ -313,13 +301,19 @@ 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:
@@ -345,20 +339,19 @@ int ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
                 * the generation argument to
                 * ocfs2_inode_lock_res_init() will have to change.
                 */
-               BUG_ON(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL));
+               BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
 
-               ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
+               ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
                                          OCFS2_LOCK_TYPE_META, 0, inode);
+
+               ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+                                         OCFS2_LOCK_TYPE_OPEN, 0, inode);
        }
 
        ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
                                  OCFS2_LOCK_TYPE_RW, inode->i_generation,
                                  inode);
 
-       ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_data_lockres,
-                                 OCFS2_LOCK_TYPE_DATA, inode->i_generation,
-                                 inode);
-
        ocfs2_set_inode_flags(inode);
 
        status = 0;
@@ -423,7 +416,8 @@ static int ocfs2_read_locked_inode(struct inode *inode,
         * cluster lock before trusting anything anyway.
         */
        can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
-               && !(args->fi_flags & OCFS2_FI_FLAG_NOLOCK);
+               && !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
+               && !ocfs2_mount_local(osb);
 
        /*
         * To maintain backwards compatibility with older versions of
@@ -435,21 +429,41 @@ static int ocfs2_read_locked_inode(struct inode *inode,
        if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
                generation = osb->fs_generation;
 
-       ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_meta_lockres,
+       ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
                                  OCFS2_LOCK_TYPE_META,
                                  generation, inode);
 
+       ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+                                 OCFS2_LOCK_TYPE_OPEN,
+                                 0, inode);
+
        if (can_lock) {
-               status = ocfs2_meta_lock(inode, NULL, 0);
+               status = ocfs2_open_lock(inode);
                if (status) {
                        make_bad_inode(inode);
                        mlog_errno(status);
                        return status;
                }
+               status = ocfs2_inode_lock(inode, NULL, 0);
+               if (status) {
+                       make_bad_inode(inode);
+                       mlog_errno(status);
+                       return status;
+               }
+       }
+
+       if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
+               status = ocfs2_try_open_lock(inode, 0);
+               if (status) {
+                       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_blocks(inode, args->fi_blkno, 1, &bh, 0);
+       else
+               status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
        if (status < 0) {
                mlog_errno(status);
                goto bail;
@@ -458,8 +472,9 @@ 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(ML_ERROR, "Invalid dinode #%llu: signature = %.*s\n",
-                    (unsigned long long)fe->i_blkno, 7, fe->i_signature);
+               mlog(0, "Invalid dinode #%llu: signature = %.*s\n",
+                    (unsigned long long)args->fi_blkno, 7,
+                    fe->i_signature);
                goto bail;
        }
 
@@ -477,11 +492,8 @@ static int ocfs2_read_locked_inode(struct inode *inode,
            S_ISBLK(le16_to_cpu(fe->i_mode)))
                inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
 
-       if (ocfs2_populate_inode(inode, fe, 0) < 0) {
-               mlog(ML_ERROR, "populate failed! i_blkno=%llu, i_ino=%lu\n",
-                    (unsigned long long)fe->i_blkno, inode->i_ino);
+       if (ocfs2_populate_inode(inode, fe, 0) < 0)
                goto bail;
-       }
 
        BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
 
@@ -489,7 +501,7 @@ static int ocfs2_read_locked_inode(struct inode *inode,
 
 bail:
        if (can_lock)
-               ocfs2_meta_unlock(inode, 0);
+               ocfs2_inode_unlock(inode, 0);
 
        if (status < 0)
                make_bad_inode(inode);
@@ -511,50 +523,63 @@ static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
                                     struct buffer_head *fe_bh)
 {
        int status = 0;
-       handle_t *handle = NULL;
        struct ocfs2_truncate_context *tc = NULL;
        struct ocfs2_dinode *fe;
+       handle_t *handle = NULL;
 
        mlog_entry_void();
 
        fe = (struct ocfs2_dinode *) fe_bh->b_data;
 
-       /* zero allocation, zero truncate :) */
-       if (!fe->i_clusters)
-               goto bail;
+       /*
+        * This check will also skip truncate of inodes with inline
+        * 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 bail;
-       }
+               handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+               if (IS_ERR(handle)) {
+                       status = PTR_ERR(handle);
+                       mlog_errno(status);
+                       goto out;
+               }
 
-       status = ocfs2_set_inode_size(handle, inode, fe_bh, 0ULL);
-       if (status < 0) {
-               mlog_errno(status);
-               goto bail;
-       }
+               status = ocfs2_journal_access(handle, inode, fe_bh,
+                                             OCFS2_JOURNAL_ACCESS_WRITE);
+               if (status < 0) {
+                       mlog_errno(status);
+                       goto out;
+               }
 
-       ocfs2_commit_trans(osb, handle);
-       handle = NULL;
+               i_size_write(inode, 0);
 
-       status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
-       if (status < 0) {
-               mlog_errno(status);
-               goto bail;
-       }
+               status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+               if (status < 0) {
+                       mlog_errno(status);
+                       goto out;
+               }
 
-       status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
-       if (status < 0) {
-               mlog_errno(status);
-               goto bail;
+               ocfs2_commit_trans(osb, handle);
+               handle = NULL;
+
+               status = ocfs2_prepare_truncate(osb, inode, fe_bh, &tc);
+               if (status < 0) {
+                       mlog_errno(status);
+                       goto out;
+               }
+
+               status = ocfs2_commit_truncate(osb, inode, fe_bh, tc);
+               if (status < 0) {
+                       mlog_errno(status);
+                       goto out;
+               }
        }
-bail:
+
+out:
        if (handle)
                ocfs2_commit_trans(osb, handle);
-
        mlog_exit(status);
        return status;
 }
@@ -581,7 +606,7 @@ static int ocfs2_remove_inode(struct inode *inode,
        }
 
        mutex_lock(&inode_alloc_inode->i_mutex);
-       status = ocfs2_meta_lock(inode_alloc_inode, &inode_alloc_bh, 1);
+       status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
        if (status < 0) {
                mutex_unlock(&inode_alloc_inode->i_mutex);
 
@@ -612,7 +637,7 @@ static int ocfs2_remove_inode(struct inode *inode,
        }
 
        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) {
@@ -630,7 +655,7 @@ static int ocfs2_remove_inode(struct inode *inode,
 bail_commit:
        ocfs2_commit_trans(osb, handle);
 bail_unlock:
-       ocfs2_meta_unlock(inode_alloc_inode, 1);
+       ocfs2_inode_unlock(inode_alloc_inode, 1);
        mutex_unlock(&inode_alloc_inode->i_mutex);
        brelse(inode_alloc_bh);
 bail:
@@ -682,10 +707,10 @@ static int ocfs2_wipe_inode(struct inode *inode,
        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;
 
-       /* We've already voted on this so it should be readonly - no
-        * spinlock needed. */
-       orphaned_slot = OCFS2_I(inode)->ip_orphaned_slot;
+       di = (struct ocfs2_dinode *) di_bh->b_data;
+       orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
 
        status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
        if (status)
@@ -704,7 +729,7 @@ static int ocfs2_wipe_inode(struct inode *inode,
         * 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_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+       status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
        if (status < 0) {
                mutex_unlock(&orphan_dir_inode->i_mutex);
 
@@ -713,8 +738,8 @@ static int ocfs2_wipe_inode(struct inode *inode,
        }
 
        /* we do this while holding the orphan dir lock because we
-        * don't want recovery being run from another node to vote for
-        * an inode delete on us -- this will result in two nodes
+        * don't want recovery being run from another node to try an
+        * inode delete underneath us -- this will result in two nodes
         * truncating the same file! */
        status = ocfs2_truncate_for_delete(osb, inode, di_bh);
        if (status < 0) {
@@ -722,13 +747,20 @@ static int ocfs2_wipe_inode(struct inode *inode,
                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_inode(inode, di_bh, orphan_dir_inode,
                                    orphan_dir_bh);
        if (status < 0)
                mlog_errno(status);
 
 bail_unlock_dir:
-       ocfs2_meta_unlock(orphan_dir_inode, 1);
+       ocfs2_inode_unlock(orphan_dir_inode, 1);
        mutex_unlock(&orphan_dir_inode->i_mutex);
        brelse(orphan_dir_bh);
 bail:
@@ -739,7 +771,7 @@ bail:
 }
 
 /* There is a series of simple checks that should be done before a
- * vote is even considered. Encapsulate those in this function. */
+ * trylock is even considered. Encapsulate those in this function. */
 static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
 {
        int ret = 0;
@@ -753,14 +785,14 @@ static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
                goto bail;
        }
 
-       /* If we're coming from process_vote we can't go into our own
+       /* If we're coming from downconvert_thread we can't go into our own
         * voting [hello, deadlock city!], so unforuntately we just
         * have to skip deleting this guy. That's OK though because
         * the node who's doing the actual deleting should handle it
         * anyway. */
-       if (current == osb->vote_task) {
+       if (current == osb->dc_task) {
                mlog(0, "Skipping delete of %lu because we're currently "
-                    "in process_vote\n", inode->i_ino);
+                    "in downconvert\n", inode->i_ino);
                goto bail;
        }
 
@@ -774,10 +806,9 @@ static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
                goto bail_unlock;
        }
 
-       /* If we have voted "yes" on the wipe of this inode for
-        * another node, it will be marked here so we can safely skip
-        * it. Recovery will cleanup any inodes we might inadvertantly
-        * skip here. */
+       /* If we have allowd wipe of this inode for another node, it
+        * will be marked here so we can safely skip it. Recovery will
+        * cleanup any inodes we might inadvertantly skip here. */
        if (oi->ip_flags & OCFS2_INODE_SKIP_DELETE) {
                mlog(0, "Skipping delete of %lu because another node "
                     "has done this for us.\n", inode->i_ino);
@@ -831,8 +862,8 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
                     "Inode %llu (on-disk %llu) not orphaned! "
                     "Disk flags  0x%x, inode flags 0x%x\n",
                     (unsigned long long)oi->ip_blkno,
-                    (unsigned long long)di->i_blkno, di->i_flags,
-                    oi->ip_flags);
+                    (unsigned long long)le64_to_cpu(di->i_blkno),
+                    le32_to_cpu(di->i_flags), oi->ip_flags);
                goto bail;
        }
 
@@ -843,13 +874,22 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
                goto bail;
        }
 
-       status = ocfs2_request_delete_vote(inode);
-       /* -EBUSY means that other nodes are still using the
-        * inode. We're done here though, so avoid doing anything on
-        * disk and let them worry about deleting it. */
-       if (status == -EBUSY) {
+       /*
+        * This is how ocfs2 determines whether an inode is still live
+        * within the cluster. Every node takes a shared read lock on
+        * the inode open lock in ocfs2_read_locked_inode(). When we
+        * get to ->delete_inode(), each node tries to convert it's
+        * lock to an exclusive. Trylocks are serialized by the inode
+        * meta data lock. If the upconvert suceeds, we know the inode
+        * is no longer live and can be deleted.
+        *
+        * Though we call this with the meta data lock held, the
+        * trylock keeps us from ABBA deadlock.
+        */
+       status = ocfs2_try_open_lock(inode, 1);
+       if (status == -EAGAIN) {
                status = 0;
-               mlog(0, "Skipping delete of %llu because it is in use on"
+               mlog(0, "Skipping delete of %llu because it is in use on "
                     "other nodes\n", (unsigned long long)oi->ip_blkno);
                goto bail;
        }
@@ -858,21 +898,10 @@ static int ocfs2_query_inode_wipe(struct inode *inode,
                goto bail;
        }
 
-       spin_lock(&oi->ip_lock);
-       if (oi->ip_orphaned_slot == OCFS2_INVALID_SLOT) {
-               /* Nobody knew which slot this inode was orphaned
-                * into. This may happen during node death and
-                * recovery knows how to clean it up so we can safely
-                * ignore this inode for now on. */
-               mlog(0, "Nobody knew where inode %llu was orphaned!\n",
-                    (unsigned long long)oi->ip_blkno);
-       } else {
-               *wipe = 1;
-
-               mlog(0, "Inode %llu is ok to wipe from orphan dir %d\n",
-                    (unsigned long long)oi->ip_blkno, oi->ip_orphaned_slot);
-       }
-       spin_unlock(&oi->ip_lock);
+       *wipe = 1;
+       mlog(0, "Inode %llu is ok to wipe from orphan dir %u\n",
+            (unsigned long long)oi->ip_blkno,
+            le16_to_cpu(di->i_orphaned_slot));
 
 bail:
        return status;
@@ -926,13 +955,13 @@ void ocfs2_delete_inode(struct inode *inode)
 
        /* 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 votes. 
+        * serialize delete_inode on multiple nodes.
         *
         * Even though we might be doing a truncate, we don't take the
         * allocation lock here as it won't be needed - nobody will
         * have the file open.
         */
-       status = ocfs2_meta_lock(inode, &di_bh, 1);
+       status = ocfs2_inode_lock(inode, &di_bh, 1);
        if (status < 0) {
                if (status != -ENOENT)
                        mlog_errno(status);
@@ -944,15 +973,15 @@ void ocfs2_delete_inode(struct inode *inode)
         * before we go ahead and wipe the inode. */
        status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
        if (!wipe || status < 0) {
-               /* Error and inode busy vote both mean we won't be
+               /* Error and remote inode busy both mean we won't be
                 * removing the inode, so they take almost the same
                 * path. */
                if (status < 0)
                        mlog_errno(status);
 
-               /* Someone in the cluster has voted to not wipe this
-                * inode, or it was never completely orphaned. Write
-                * out the pages and exit now. */
+               /* Someone in the cluster has disallowed a wipe of
+                * this inode, or it was never completely
+                * orphaned. Write out the pages and exit now. */
                ocfs2_cleanup_delete_inode(inode, 1);
                goto bail_unlock_inode;
        }
@@ -978,7 +1007,7 @@ void ocfs2_delete_inode(struct inode *inode)
        OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
 
 bail_unlock_inode:
-       ocfs2_meta_unlock(inode, 1);
+       ocfs2_inode_unlock(inode, 1);
        brelse(di_bh);
 bail_unblock:
        status = sigprocmask(SIG_SETMASK, &oldset, NULL);
@@ -1005,11 +1034,15 @@ void ocfs2_clear_inode(struct inode *inode)
        mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
                        "Inode=%lu\n", inode->i_ino);
 
+       /* To preven remote deletes we hold open lock before, now it
+        * is time to unlock PR and EX open locks. */
+       ocfs2_open_unlock(inode);
+
        /* Do these before all the other work so that we don't bounce
-        * the vote thread while waiting to destroy the locks. */
+        * the downconvert thread while waiting to destroy the locks. */
        ocfs2_mark_lockres_freeing(&oi->ip_rw_lockres);
-       ocfs2_mark_lockres_freeing(&oi->ip_meta_lockres);
-       ocfs2_mark_lockres_freeing(&oi->ip_data_lockres);
+       ocfs2_mark_lockres_freeing(&oi->ip_inode_lockres);
+       ocfs2_mark_lockres_freeing(&oi->ip_open_lockres);
 
        /* We very well may get a clear_inode before all an inodes
         * metadata has hit disk. Of course, we can't drop any cluster
@@ -1024,16 +1057,15 @@ void ocfs2_clear_inode(struct inode *inode)
                        "Clear inode of %llu, inode has io markers\n",
                        (unsigned long long)oi->ip_blkno);
 
-       ocfs2_extent_map_drop(inode, 0);
-       ocfs2_extent_map_init(inode);
+       ocfs2_extent_map_trunc(inode, 0);
 
        status = ocfs2_drop_inode_locks(inode);
        if (status < 0)
                mlog_errno(status);
 
        ocfs2_lock_res_free(&oi->ip_rw_lockres);
-       ocfs2_lock_res_free(&oi->ip_meta_lockres);
-       ocfs2_lock_res_free(&oi->ip_data_lockres);
+       ocfs2_lock_res_free(&oi->ip_inode_lockres);
+       ocfs2_lock_res_free(&oi->ip_open_lockres);
 
        ocfs2_metadata_cache_purge(inode);
 
@@ -1073,6 +1105,8 @@ void ocfs2_clear_inode(struct inode *inode)
        oi->ip_last_trans = 0;
        oi->ip_dir_start_lookup = 0;
        oi->ip_blkno = 0ULL;
+       jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
+                                      &oi->ip_jinode);
 
 bail:
        mlog_exit_void();
@@ -1090,9 +1124,6 @@ void ocfs2_drop_inode(struct inode *inode)
        mlog(0, "Drop inode %llu, nlink = %u, ip_flags = 0x%x\n",
             (unsigned long long)oi->ip_blkno, inode->i_nlink, oi->ip_flags);
 
-       /* Testing ip_orphaned_slot here wouldn't work because we may
-        * not have gotten a delete_inode vote from any other nodes
-        * yet. */
        if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
                generic_delete_inode(inode);
        else
@@ -1125,15 +1156,16 @@ struct buffer_head *ocfs2_bread(struct inode *inode,
                return NULL;
        }
 
-       tmperr = ocfs2_extent_map_get_blocks(inode, block, 1,
-                                            &p_blkno, 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);
+       tmperr = ocfs2_read_blocks(inode, p_blkno, 1, &bh, readflags);
        if (tmperr < 0)
                goto fail;
 
@@ -1143,10 +1175,9 @@ struct buffer_head *ocfs2_bread(struct inode *inode,
        return bh;
 
 fail:
-       if (bh) {
-               brelse(bh);
-               bh = NULL;
-       }
+       brelse(bh);
+       bh = NULL;
+
        *err = -EIO;
        return NULL;
 }
@@ -1177,15 +1208,15 @@ int ocfs2_inode_revalidate(struct dentry *dentry)
        }
        spin_unlock(&OCFS2_I(inode)->ip_lock);
 
-       /* Let ocfs2_meta_lock do the work of updating our struct
+       /* Let ocfs2_inode_lock do the work of updating our struct
         * inode for us. */
-       status = ocfs2_meta_lock(inode, NULL, 0);
+       status = ocfs2_inode_lock(inode, NULL, 0);
        if (status < 0) {
                if (status != -ENOENT)
                        mlog_errno(status);
                goto bail;
        }
-       ocfs2_meta_unlock(inode, 0);
+       ocfs2_inode_unlock(inode, 0);
 bail:
        mlog_exit(status);
 
@@ -1216,7 +1247,9 @@ int ocfs2_mark_inode_dirty(handle_t *handle,
 
        spin_lock(&OCFS2_I(inode)->ip_lock);
        fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
+       ocfs2_get_inode_flags(OCFS2_I(inode));
        fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
+       fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
        spin_unlock(&OCFS2_I(inode)->ip_lock);
 
        fe->i_size = cpu_to_le64(i_size_read(inode));
@@ -1254,6 +1287,7 @@ void ocfs2_refresh_inode(struct inode *inode,
 
        OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
        OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+       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);
@@ -1263,7 +1297,7 @@ void ocfs2_refresh_inode(struct inode *inode,
        if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
                inode->i_blocks = 0;
        else
-               inode->i_blocks = ocfs2_align_bytes_to_sectors(i_size_read(inode));
+               inode->i_blocks = ocfs2_inode_sector_count(inode);
        inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
        inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
        inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);