Whamcloud - gitweb
LU-6030 ldiskfs: split pdirop patch
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / rhel6.3 / ext4-pdirop.patch
index c6e93c3..32adfc1 100644 (file)
@@ -1,5 +1,7 @@
---- /dev/null  2011-12-14 22:16:16.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/include/linux/htree_lock.h        2011-12-02 17:09:34.000000000 +0800
+Index: linux-2.6.32-504.3.3.el6.x86_64/include/linux/htree_lock.h
+===================================================================
+--- /dev/null
++++ linux-2.6.32-504.3.3.el6.x86_64/include/linux/htree_lock.h
 @@ -0,0 +1,187 @@
 +/*
 + * include/linux/htree_lock.h
 +      ((lck)->lk_nodes[dep].ln_mode == HTREE_LOCK_NL)
 +
 +#endif
---- /dev/null  2011-12-14 22:16:16.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/fs/ext4/htree_lock.c      2011-12-14 22:56:28.000000000 +0800
+Index: linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/htree_lock.c
+===================================================================
+--- /dev/null
++++ linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/htree_lock.c
 @@ -0,0 +1,880 @@
 +/*
 + * fs/ext4/htree_lock.c
 +      kfree(lck);
 +}
 +EXPORT_SYMBOL(htree_lock_free);
---- linux-2.6.32-131.6.1/fs/ext4/ext4.h        2011-10-06 20:10:49.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/fs/ext4/ext4.h    2011-12-08 18:25:00.000000000 +0800
-@@ -28,6 +28,7 @@
+Index: linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/ext4.h
+===================================================================
+--- linux-2.6.32-504.3.3.el6.x86_64.orig/fs/ext4/ext4.h
++++ linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/ext4.h
+@@ -27,6 +27,7 @@
  #include <linux/mutex.h>
  #include <linux/timer.h>
  #include <linux/wait.h>
  #include <linux/blockgroup_lock.h>
  #include <linux/percpu_counter.h>
  #ifdef __KERNEL__
-@@ -1277,6 +1278,7 @@ EXT4_INODE_BIT_FNS(state, state_flags)
- #define EXT4_FEATURE_INCOMPAT_MMP               0x0100
- #define EXT4_FEATURE_INCOMPAT_FLEX_BG         0x0200
- #define EXT4_FEATURE_INCOMPAT_DIRDATA         0x1000
-+#define EXT4_FEATURE_INCOMPAT_LARGEDIR                0x4000
- #define EXT4_FEATURE_COMPAT_SUPP      EXT2_FEATURE_COMPAT_EXT_ATTR
- #define EXT4_FEATURE_INCOMPAT_SUPP    (EXT4_FEATURE_INCOMPAT_FILETYPE| \
-@@ -1286,7 +1288,8 @@ EXT4_INODE_BIT_FNS(state, state_flags)
-                                        EXT4_FEATURE_INCOMPAT_64BIT| \
-                                        EXT4_FEATURE_INCOMPAT_FLEX_BG| \
-                                        EXT4_FEATURE_INCOMPAT_MMP| \
--                                       EXT4_FEATURE_INCOMPAT_DIRDATA)
-+                                       EXT4_FEATURE_INCOMPAT_DIRDATA| \
-+                                       EXT4_FEATURE_INCOMPAT_LARGEDIR)
- #define EXT4_FEATURE_RO_COMPAT_SUPP   (EXT4_FEATURE_RO_COMPAT_SPARSE_SUPER| \
-                                        EXT4_FEATURE_RO_COMPAT_LARGE_FILE| \
-@@ -1536,6 +1539,76 @@ ext4_group_first_block_no(struct super_b
-  */
- #define ERR_BAD_DX_DIR        -75000
+@@ -1625,6 +1626,71 @@ ext4_dir_htree_level(struct super_block
+               EXT4_HTREE_LEVEL : EXT4_HTREE_LEVEL_COMPAT;
+ }
  
-+/* htree levels for ext4 */
-+#define EXT4_HTREE_LEVEL_COMPAT 2
-+#define EXT4_HTREE_LEVEL      3
-+
-+static inline int
-+ext4_dir_htree_level(struct super_block *sb)
-+{
-+      return EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) ?
-+              EXT4_HTREE_LEVEL : EXT4_HTREE_LEVEL_COMPAT;
-+}
-+
 +/* assume name-hash is protected by upper layer */
 +#define EXT4_HTREE_LOCK_HASH  0
 +
 +                          struct inode *dir, unsigned flags);
 +#define ext4_htree_unlock(lck)                  htree_unlock(lck)
 +
++extern struct buffer_head * __ext4_find_entry(struct inode *dir,
++                                      const struct qstr *d_name,
++                                      struct ext4_dir_entry_2 **res_dir,
++                                      struct htree_lock *lck);
++extern int __ext4_add_entry(handle_t *handle, struct dentry *dentry,
++                    struct inode *inode, struct htree_lock *lck);
  void ext4_get_group_no_and_offset(struct super_block *sb, ext4_fsblk_t blocknr,
                        ext4_group_t *blockgrpp, ext4_grpblk_t *offsetp);
  
-@@ -1769,14 +1842,16 @@ extern int ext4_htree_fill_tree(struct f
- extern struct inode *ext4_create_inode(handle_t *handle,
-                                      struct inode * dir, int mode);
- extern int ext4_add_entry(handle_t *handle, struct dentry *dentry,
--                        struct inode *inode);
-+                        struct inode *inode, struct htree_lock *lck);
- extern int ext4_delete_entry(handle_t *handle, struct inode * dir,
-                            struct ext4_dir_entry_2 * de_del,
-                            struct buffer_head * bh);
- extern struct buffer_head * ext4_find_entry(struct inode *dir,
-                                           const struct qstr *d_name,
--                                          struct ext4_dir_entry_2 ** res_dir);
--#define ll_ext4_find_entry(inode, dentry, res_dir) ext4_find_entry(inode, &(dentry)->d_name, res_dir)
-+                                          struct ext4_dir_entry_2 **res_dir,
-+                                          struct htree_lock *lck);
-+#define ll_ext4_find_entry(inode, dentry, res_dir, lck) \
-+      ext4_find_entry(inode, &(dentry)->d_name, res_dir, lck)
- extern int ext4_add_dot_dotdot(handle_t *handle, struct inode *dir,
-                              struct inode *inode, const void *, const void *);
- extern struct buffer_head *ext4_append(handle_t *handle,
-@@ -1893,13 +1968,15 @@ static inline void ext4_r_blocks_count_s
-       es->s_r_blocks_count_hi = cpu_to_le32(blk >> 32);
- }
--static inline loff_t ext4_isize(struct ext4_inode *raw_inode)
-+static inline loff_t ext4_isize(struct super_block *sb,
-+                              struct ext4_inode *raw_inode)
- {
--      if (S_ISREG(le16_to_cpu(raw_inode->i_mode)))
-+      if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_LARGEDIR) ||
-+          S_ISREG(le16_to_cpu(raw_inode->i_mode)))
-               return ((loff_t)le32_to_cpu(raw_inode->i_size_high) << 32) |
-                       le32_to_cpu(raw_inode->i_size_lo);
--      else
--              return (loff_t) le32_to_cpu(raw_inode->i_size_lo);
-+
-+      return (loff_t) le32_to_cpu(raw_inode->i_size_lo);
- }
- static inline void ext4_isize_set(struct ext4_inode *raw_inode, loff_t i_size)
---- linux-2.6.32-131.6.1/fs/ext4/namei.c       2011-10-06 20:10:49.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/fs/ext4/namei.c   2011-12-14 22:55:28.000000000 +0800
+Index: linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/namei.c
+===================================================================
+--- linux-2.6.32-504.3.3.el6.x86_64.orig/fs/ext4/namei.c
++++ linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/namei.c
 @@ -176,7 +176,7 @@ static struct dx_frame *dx_probe(const s
                                 struct inode *dir,
                                 struct dx_hash_info *hinfo,
  
  /*
   * p is at least 6 bytes before the end of page
-@@ -225,7 +225,7 @@ struct dx_root_info * dx_get_dx_info(str
- static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
- {
--      return le32_to_cpu(entry->block) & 0x00ffffff;
-+      return le32_to_cpu(entry->block) & 0x0fffffff;
- }
- static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
-@@ -368,6 +368,223 @@ struct stats dx_show_entries(struct dx_h
+@@ -368,6 +368,225 @@ struct stats dx_show_entries(struct dx_h
  }
  #endif /* DX_DEBUG */
  
 +      struct dx_entry         *ld_at;    /* position of leaf dx_entry */
 +};
 +
-+#define ext4_htree_lock_data(l)       ((struct ext4_dir_lock_data *)(l)->lk_private)
++#define ext4_htree_lock_data(l) ((struct ext4_dir_lock_data *)(l)->lk_private)
++#define ext4_find_entry(dir, name, dirent) __ext4_find_entry(dir, name, dirent, NULL)
++#define ext4_add_entry(handle, dentry, inode) __ext4_add_entry(handle, dentry, inode, NULL)
 +
 +/* NB: ext4_lblk_t is 32 bits so we use high bits to identify invalid blk */
 +#define EXT4_HTREE_NODE_CHANGED       (0xcafeULL << 32)
  /*
   * Probe for a directory leaf block to search.
   *
-@@ -379,16 +596,17 @@ struct stats dx_show_entries(struct dx_h
+@@ -379,10 +598,11 @@ struct stats dx_show_entries(struct dx_h
   */
  static struct dx_frame *
  dx_probe(const struct qstr *d_name, struct inode *dir,
        struct dx_root_info * info;
        struct buffer_head *bh;
        struct dx_frame *frame = frame_in;
-       u32 hash;
--      frame->bh = NULL;
-+      memset(frame_in, 0, EXT4_HTREE_LEVEL * sizeof(frame_in[0]));
-       if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
-               goto fail;
-@@ -418,9 +636,16 @@ dx_probe(const struct qstr *d_name, stru
-               goto fail;
-       }
--      if ((indirect = info->indirect_levels) > 1) {
--              ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
--                           info->indirect_levels);
-+      indirect = info->indirect_levels;
-+      if (indirect >= ext4_dir_htree_level(dir->i_sb)) {
-+              ext4_warning(dir->i_sb,
-+                           "Directory (ino: %lu) htree depth %#06x exceed "
-+                           "supported value", dir->i_ino,
-+                           ext4_dir_htree_level(dir->i_sb));
-+              if (ext4_dir_htree_level(dir->i_sb) < EXT4_HTREE_LEVEL) {
-+                      ext4_warning(dir->i_sb, "Enable large directory "
-+                                              "feature to access it");
-+              }
-               brelse(bh);
-               *err = ERR_BAD_DX_DIR;
-               goto fail;
-@@ -440,8 +665,15 @@ dx_probe(const struct qstr *d_name, stru
+@@ -447,8 +667,15 @@ dx_probe(const struct qstr *d_name, stru
        dxtrace(printk("Look up %x", hash));
        while (1)
        {
                        ext4_warning(dir->i_sb,
                                     "dx entry: no count or count > limit");
                        brelse(bh);
-@@ -482,9 +714,73 @@ dx_probe(const struct qstr *d_name, stru
+@@ -489,9 +716,73 @@ dx_probe(const struct qstr *d_name, stru
                frame->bh = bh;
                frame->entries = entries;
                frame->at = at;
                at = entries = ((struct dx_node *) bh->b_data)->entries;
                if (dx_get_limit(entries) != dx_node_limit (dir)) {
                        ext4_warning(dir->i_sb,
-@@ -512,13 +808,18 @@ fail:
- static void dx_release (struct dx_frame *frames)
- {
-       struct dx_root_info *info;
-+      int i;
-+
-       if (frames[0].bh == NULL)
-               return;
-       info = dx_get_dx_info((struct ext4_dir_entry_2*)frames[0].bh->b_data);
--      if (info->indirect_levels)
--              brelse(frames[1].bh);
--      brelse(frames[0].bh);
-+      for (i = 0; i <= info->indirect_levels; i++) {
-+              if (frames[i].bh == NULL)
-+                      break;
-+              brelse(frames[i].bh);
-+              frames[i].bh = NULL;
-+      }
- }
- /*
-@@ -541,7 +842,7 @@ static void dx_release (struct dx_frame 
+@@ -553,7 +844,7 @@ static void dx_release (struct dx_frame
  static int ext4_htree_next_block(struct inode *dir, __u32 hash,
                                 struct dx_frame *frame,
                                 struct dx_frame *frames,
  {
        struct dx_frame *p;
        struct buffer_head *bh;
-@@ -556,12 +857,22 @@ static int ext4_htree_next_block(struct 
+@@ -568,12 +859,22 @@ static int ext4_htree_next_block(struct
         * this loop, num_frames indicates the number of interior
         * nodes need to be read.
         */
                p--;
        }
  
-@@ -584,6 +895,13 @@ static int ext4_htree_next_block(struct 
+@@ -596,6 +897,13 @@ static int ext4_htree_next_block(struct
         * block so no check is necessary
         */
        while (num_frames--) {
                if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
                                      0, &err)))
                        return err; /* Failure */
-@@ -592,6 +910,7 @@ static int ext4_htree_next_block(struct 
+@@ -604,6 +912,7 @@ static int ext4_htree_next_block(struct
                p->bh = bh;
                p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
        }
        return 1;
  }
  
-@@ -661,7 +980,7 @@ int ext4_htree_fill_tree(struct file *di
- {
-       struct dx_hash_info hinfo;
-       struct ext4_dir_entry_2 *de;
--      struct dx_frame frames[2], *frame;
-+      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
-       struct inode *dir;
-       ext4_lblk_t block;
-       int count = 0;
-@@ -684,10 +1003,10 @@ int ext4_htree_fill_tree(struct file *di
+@@ -696,10 +1005,10 @@ int ext4_htree_fill_tree(struct file *di
        }
        hinfo.hash = start_hash;
        hinfo.minor_hash = 0;
        /* Add '.' and '..' from the htree header */
        if (!start_hash && !start_minor_hash) {
                de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
-@@ -714,7 +1033,7 @@ int ext4_htree_fill_tree(struct file *di
+@@ -726,7 +1035,7 @@ int ext4_htree_fill_tree(struct file *di
                count += ret;
                hashval = ~0;
                ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
                *next_hash = hashval;
                if (ret < 0) {
                        err = ret;
-@@ -814,9 +1133,17 @@ static void dx_insert_block(struct dx_fr
+@@ -826,9 +1135,17 @@ static void dx_insert_block(struct dx_fr
  
  static void ext4_update_dx_flag(struct inode *inode)
  {
  }
  
  /*
-@@ -889,8 +1216,9 @@ static inline int search_dirblock(struct
+@@ -900,9 +1217,10 @@ static inline int search_dirblock(struct
+  * The returned buffer_head has ->b_count elevated.  The caller is expected
   * to brelse() it when appropriate.
   */
- struct buffer_head * ext4_find_entry(struct inode *dir,
--                                    const struct qstr *d_name,
--                                    struct ext4_dir_entry_2 ** res_dir)
-+                                   const struct qstr *d_name,
-+                                   struct ext4_dir_entry_2 **res_dir,
-+                                   struct htree_lock *lck)
+-static struct buffer_head * ext4_find_entry (struct inode *dir,
++struct buffer_head * __ext4_find_entry(struct inode *dir,
+                                       const struct qstr *d_name,
+-                                      struct ext4_dir_entry_2 ** res_dir)
++                                      struct ext4_dir_entry_2 **res_dir,
++                                      struct htree_lock *lck)
  {
        struct super_block *sb;
        struct buffer_head *bh_use[NAMEI_RA_SIZE];
-@@ -911,7 +1239,7 @@ struct buffer_head * ext4_find_entry(str
+@@ -923,7 +1241,7 @@ static struct buffer_head * ext4_find_en
        if (namelen > EXT4_NAME_LEN)
                return NULL;
        if (is_dx(dir)) {
                /*
                 * On success, or if the error was file not found,
                 * return.  Otherwise, fall back to doing a search the
-@@ -921,6 +1249,7 @@ struct buffer_head * ext4_find_entry(str
+@@ -933,6 +1251,7 @@ static struct buffer_head * ext4_find_en
                        return bh;
                dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
                               "falling back\n"));
        }
        nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
        start = EXT4_I(dir)->i_dir_start_lookup;
-@@ -998,13 +1327,15 @@ cleanup_and_exit:
+@@ -1008,9 +1327,12 @@ cleanup_and_exit:
+               brelse(bh_use[ra_ptr]);
+       return ret;
  }
- EXPORT_SYMBOL(ext4_find_entry);
++EXPORT_SYMBOL(__ext4_find_entry);
  
 -static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
 -                     struct ext4_dir_entry_2 **res_dir, int *err)
  {
        struct super_block * sb;
        struct dx_hash_info     hinfo;
-       u32 hash;
--      struct dx_frame frames[2], *frame;
-+      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
-       struct ext4_dir_entry_2 *de, *top;
-       struct buffer_head *bh;
-       ext4_lblk_t block;
-@@ -1015,13 +1346,16 @@ static struct buffer_head * ext4_dx_find
+@@ -1026,13 +1348,16 @@ static struct buffer_head * ext4_dx_find
        sb = dir->i_sb;
        /* NFS may look up ".." - look at dx_root directory block */
        if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
        }
        hash = hinfo.hash;
        do {
-@@ -1050,7 +1384,7 @@ static struct buffer_head * ext4_dx_find
+@@ -1061,7 +1386,7 @@ static struct buffer_head * ext4_dx_find
                brelse(bh);
                /* Check to see if we should continue to search */
                retval = ext4_htree_next_block(dir, hash, frame,
                if (retval < 0) {
                        ext4_warning(sb,
                             "error reading index page in directory #%lu",
-@@ -1076,7 +1410,7 @@ static struct dentry *ext4_lookup(struct
-       if (dentry->d_name.len > EXT4_NAME_LEN)
-               return ERR_PTR(-ENAMETOOLONG);
--      bh = ext4_find_entry(dir, &dentry->d_name, &de);
-+      bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
-       inode = NULL;
-       if (bh) {
-               __u32 ino = le32_to_cpu(de->inode);
-@@ -1144,7 +1478,7 @@ struct dentry *ext4_get_parent(struct de
-       struct ext4_dir_entry_2 * de;
-       struct buffer_head *bh;
--      bh = ext4_find_entry(child->d_inode, &dotdot, &de);
-+      bh = ext4_find_entry(child->d_inode, &dotdot, &de, NULL);
-       inode = NULL;
-       if (!bh)
-               return ERR_PTR(-ENOENT);
-@@ -1233,8 +1567,9 @@ static struct ext4_dir_entry_2* dx_pack_
+@@ -1244,8 +1569,9 @@ static struct ext4_dir_entry_2* dx_pack_
   * Returns pointer to de in block into which the new entry will be inserted.
   */
  static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
  {
        unsigned blocksize = dir->i_sb->s_blocksize;
        unsigned count, continued;
-@@ -1291,7 +1626,14 @@ static struct ext4_dir_entry_2 *do_split
+@@ -1302,7 +1628,14 @@ static struct ext4_dir_entry_2 *do_split
                                        hash2, split, count-split));
  
        /* Fancy dance to stay within two buffers */
        de = dx_pack_dirents(data1, blocksize);
        de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
                                           blocksize);
-@@ -1300,13 +1642,21 @@ static struct ext4_dir_entry_2 *do_split
+@@ -1311,13 +1644,21 @@ static struct ext4_dir_entry_2 *do_split
        dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
        dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
  
        err = ext4_handle_dirty_metadata(handle, dir, bh2);
        if (err)
                goto journal_error;
-@@ -1418,7 +1768,7 @@ static int add_dirent_to_buf(handle_t *h
-       if (!IS_NOCMTIME(dir))
-               dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
-       ext4_update_dx_flag(dir);
--      dir->i_version++;
-+      inode_inc_iversion(dir);
-       ext4_mark_inode_dirty(handle, dir);
-       BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
-       err = ext4_handle_dirty_metadata(handle, dir, bh);
-@@ -1438,7 +1788,7 @@ static int make_indexed_dir(handle_t *ha
-       const char      *name = dentry->d_name.name;
-       int             namelen = dentry->d_name.len;
-       struct buffer_head *bh2;
--      struct dx_frame frames[2], *frame;
-+      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
-       struct dx_entry *entries;
-       struct ext4_dir_entry_2 *de, *de2, *dot_de, *dotdot_de;
-       char            *data1, *top;
-@@ -1517,7 +1867,7 @@ static int make_indexed_dir(handle_t *ha
-       ext4_handle_dirty_metadata(handle, dir, frame->bh);
-       ext4_handle_dirty_metadata(handle, dir, bh);
-
+@@ -1558,7 +1899,7 @@ static int make_indexed_dir(handle_t *ha
+       ext4_handle_dirty_metadata(handle, dir, frame->bh);
+       ext4_handle_dirty_metadata(handle, dir, bh);
 -      de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
 +      de = do_split(handle,dir, &bh, frames, frame, &hinfo, NULL, &retval);
-       if (!de) {
-               /*
-                * Even if the block split failed, we have to properly write
-@@ -1616,7 +1966,7 @@ out:
+       if (!de) {
+               /*
+                * Even if the block split failed, we have to properly write
+@@ -1664,8 +2005,8 @@ out:
+  * may not sleep between calling this and putting something into
   * the entry, as someone else might have used it while you slept.
   */
- int ext4_add_entry(handle_t *handle, struct dentry *dentry,
--                 struct inode *inode)
-+                 struct inode *inode, struct htree_lock *lck)
+-static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
+-                        struct inode *inode)
++int __ext4_add_entry(handle_t *handle, struct dentry *dentry,
++                    struct inode *inode, struct htree_lock *lck)
  {
        struct inode *dir = dentry->d_parent->d_inode;
        struct buffer_head *bh;
-@@ -1635,9 +1985,10 @@ int ext4_add_entry(handle_t *handle, str
+@@ -1684,9 +2025,10 @@ static int ext4_add_entry(handle_t *hand
                if (dentry->d_name.len == 2 &&
                    memcmp(dentry->d_name.name, "..", 2) == 0)
                        return ext4_update_dotdot(handle, dentry, inode);
                ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
                dx_fallback++;
                ext4_mark_inode_dirty(handle, dir);
-@@ -1674,18 +2025,21 @@ EXPORT_SYMBOL(ext4_add_entry);
+@@ -1717,12 +2059,13 @@ static int ext4_add_entry(handle_t *hand
+       brelse(bh);
+       return retval;
+ }
++EXPORT_SYMBOL(__ext4_add_entry);
+ /*
   * Returns 0 for success, or a negative error value
   */
  static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
 -                           struct inode *inode)
 +                           struct inode *inode, struct htree_lock *lck)
  {
--      struct dx_frame frames[2], *frame;
-+      struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
+       struct dx_frame frames[EXT4_HTREE_LEVEL], *frame;
        struct dx_entry *entries, *at;
-       struct dx_hash_info hinfo;
-       struct buffer_head *bh;
-       struct inode *dir = dentry->d_parent->d_inode;
-       struct super_block *sb = dir->i_sb;
-       struct ext4_dir_entry_2 *de;
-+      int restart;
-       int err;
+@@ -1736,7 +2079,7 @@ static int ext4_dx_add_entry(handle_t *h
  
+ again:
+       restart = 0;
 -      frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
-+again:
-+      restart = 0;
 +      frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, lck, &err);
        if (!frame)
                return err;
        entries = frame->entries;
-@@ -1694,33 +2048,53 @@ static int ext4_dx_add_entry(handle_t *h
-       if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
-               goto cleanup;
--      BUFFER_TRACE(bh, "get_write_access");
--      err = ext4_journal_get_write_access(handle, bh);
--      if (err)
--              goto journal_error;
--
-       err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
-       if (err != -ENOSPC)
-               goto cleanup;
-+      err = 0;
-       /* Block full, should compress but for now just split */
-       dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
-                      dx_get_count(entries), dx_get_limit(entries)));
-       /* Need to split index? */
-       if (dx_get_count(entries) == dx_get_limit(entries)) {
-               ext4_lblk_t newblock;
--              unsigned icount = dx_get_count(entries);
--              int levels = frame - frames;
-+              int levels = frame - frames + 1;
-+              unsigned icount;
-+              int add_level = 1;
-               struct dx_entry *entries2;
+@@ -1763,6 +2106,11 @@ again:
                struct dx_node *node2;
                struct buffer_head *bh2;
  
--              if (levels && (dx_get_count(frames->entries) ==
--                             dx_get_limit(frames->entries))) {
--                      ext4_warning(sb, "Directory index full!");
 +              if (!ext4_htree_safe_locked(lck)) { /* retry with EX lock */
 +                      ext4_htree_safe_relock(lck);
 +                      restart = 1;
 +                      goto cleanup;
 +              }
-+              while (frame > frames) {
-+                      if (dx_get_count((frame - 1)->entries) <
-+                          dx_get_limit((frame - 1)->entries)) {
-+                              add_level = 0;
-+                              break;
-+                      }
-+                      frame--; /* split higher index block */
-+                      at = frame->at;
-+                      entries = frame->entries;
-+                      restart = 1;
-+              }
-+              if (add_level && levels == ext4_dir_htree_level(sb)) {
-+                      ext4_warning(sb, "Directory (ino: %lu) index full, "
-+                                       "reach max htree level :%d",
-+                                       dir->i_ino, levels);
-+                      if (ext4_dir_htree_level(sb) < EXT4_HTREE_LEVEL) {
-+                              ext4_warning(sb, "Large directory feature is"
-+                                               "not enabled on this "
-+                                               "filesystem");
-+                      }
-                       err = -ENOSPC;
+               while (frame > frames) {
+                       if (dx_get_count((frame - 1)->entries) <
+                           dx_get_limit((frame - 1)->entries)) {
+@@ -1860,16 +2208,43 @@ again:
+                       restart = 1;
                        goto cleanup;
                }
-+              icount = dx_get_count(entries);
-               bh2 = ext4_append (handle, dir, &newblock, &err);
-               if (!(bh2))
-                       goto cleanup;
-@@ -1733,7 +2107,7 @@ static int ext4_dx_add_entry(handle_t *h
-               err = ext4_journal_get_write_access(handle, frame->bh);
-               if (err)
-                       goto journal_error;
--              if (levels) {
-+              if (!add_level) {
-                       unsigned icount1 = icount/2, icount2 = icount - icount1;
-                       unsigned hash2 = dx_get_hash(entries + icount1);
-                       dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
-@@ -1741,7 +2115,7 @@ static int ext4_dx_add_entry(handle_t *h
-                       BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
-                       err = ext4_journal_get_write_access(handle,
--                                                           frames[0].bh);
-+                                                          (frame - 1)->bh);
-                       if (err)
-                               goto journal_error;
-@@ -1757,18 +2131,24 @@ static int ext4_dx_add_entry(handle_t *h
-                               frame->entries = entries = entries2;
-                               swap(frame->bh, bh2);
-                       }
--                      dx_insert_block(frames + 0, hash2, newblock);
--                      dxtrace(dx_show_index("node", frames[1].entries));
-+                      dx_insert_block((frame - 1), hash2, newblock);
-+                      dxtrace(dx_show_index("node", frame->entries));
-                       dxtrace(dx_show_index("node",
-                              ((struct dx_node *) bh2->b_data)->entries));
-                       err = ext4_handle_dirty_metadata(handle, inode, bh2);
-                       if (err)
-                               goto journal_error;
-                       brelse (bh2);
-+                      ext4_handle_dirty_metadata(handle, inode,
-+                                                 (frame - 1)->bh);
-+                      if (restart) {
-+                              ext4_handle_dirty_metadata(handle, inode,
-+                                                         frame->bh);
-+                              goto cleanup;
-+                      }
-               } else {
-                       struct dx_root_info * info;
--                      dxtrace(printk(KERN_DEBUG
--                                     "Creating second level index...\n"));
-+
-                       memcpy((char *) entries2, (char *) entries,
-                              icount * sizeof(struct dx_entry));
-                       dx_set_limit(entries2, dx_node_limit(dir));
-@@ -1778,32 +2158,60 @@ static int ext4_dx_add_entry(handle_t *h
-                       dx_set_block(entries + 0, newblock);
-                       info = dx_get_dx_info((struct ext4_dir_entry_2*)
-                                       frames[0].bh->b_data);
--                      info->indirect_levels = 1;
-+                      info->indirect_levels += 1;
-+                      dxtrace(printk(KERN_DEBUG
-+                                     "Creating %d level index...\n",
-+                                     info->indirect_levels));
-+                      ext4_handle_dirty_metadata(handle, inode, frame->bh);
-+                      ext4_handle_dirty_metadata(handle, inode, bh2);
-+                      brelse(bh2);
-+                      restart = 1;
-+                      goto cleanup;
-+              }
 +      } else if (!ext4_htree_dx_locked(lck)) {
 +              struct ext4_dir_lock_data *ld = ext4_htree_lock_data(lck);
--                      /* Add new access path frame */
--                      frame = frames + 1;
--                      frame->at = at = at - entries + entries2;
--                      frame->entries = entries = entries2;
--                      frame->bh = bh2;
--                      err = ext4_journal_get_write_access(handle,
--                                                           frame->bh);
--                      if (err)
--                              goto journal_error;
++
 +              /* not well protected, require DX lock */
 +              ext4_htree_dx_need_lock(lck);
 +              at = frame > frames ? (frame - 1)->at : NULL;
 +                  (ld->ld_count != dx_get_count(entries))) {
 +                      restart = 1;
 +                      goto cleanup;
-               }
--              ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
++              }
 +              /* OK, I've got DX lock and nothing changed */
 +              frame->at = ld->ld_at;
        }
        if (bh)
                brelse(bh);
        dx_release(frames);
-+      /* @restart is true means htree-path has been changed, we need to
-+       * repeat dx_probe() to find out valid htree-path */
-+      if (restart && err == 0)
-+              goto again;
-       return err;
- }
-@@ -1838,7 +2246,7 @@ int ext4_delete_entry(handle_t *handle,
-                                       blocksize);
-                       else
-                               de->inode = 0;
--                      dir->i_version++;
-+                      inode_inc_iversion(dir);
-                       BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
-                       ext4_handle_dirty_metadata(handle, dir, bh);
-                       return 0;
-@@ -1882,7 +2290,7 @@ static void ext4_dec_count(handle_t *han
- static int ext4_add_nondir(handle_t *handle,
-               struct dentry *dentry, struct inode *inode)
- {
--      int err = ext4_add_entry(handle, dentry, inode);
-+      int err = ext4_add_entry(handle, dentry, inode, NULL);
-       if (!err) {
-               ext4_mark_inode_dirty(handle, inode);
-               d_instantiate(dentry, inode);
-@@ -2112,7 +2520,7 @@ retry:
-               goto out_stop;
-       }
--      err = ext4_add_entry(handle, dentry, inode);
-+      err = ext4_add_entry(handle, dentry, inode, NULL);
-       if (err)
-               goto out_clear_inode;
-       ext4_inc_count(handle, dir);
-@@ -2381,7 +2789,7 @@ static int ext4_rmdir(struct inode *dir,
-               return PTR_ERR(handle);
-       retval = -ENOENT;
--      bh = ext4_find_entry(dir, &dentry->d_name, &de);
-+      bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
-       if (!bh)
-               goto end_rmdir;
-@@ -2443,7 +2851,7 @@ static int ext4_unlink(struct inode *dir
-               ext4_handle_sync(handle);
-       retval = -ENOENT;
--      bh = ext4_find_entry(dir, &dentry->d_name, &de);
-+      bh = ext4_find_entry(dir, &dentry->d_name, &de, NULL);
-       if (!bh)
-               goto end_unlink;
-@@ -2567,7 +2975,7 @@ retry:
-       ext4_inc_count(handle, inode);
-       atomic_inc(&inode->i_count);
--      err = ext4_add_entry(handle, dentry, inode);
-+      err = ext4_add_entry(handle, dentry, inode, NULL);
-       if (!err) {
-               ext4_mark_inode_dirty(handle, inode);
-               d_instantiate(dentry, inode);
-@@ -2612,7 +3020,7 @@ static int ext4_rename(struct inode *old
-       if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
-               ext4_handle_sync(handle);
--      old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
-+      old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de, NULL);
-       /*
-        *  Check for inode number is _not_ due to possible IO errors.
-        *  We might rmdir the source, keep it as pwd of some process
-@@ -2625,7 +3033,7 @@ static int ext4_rename(struct inode *old
-               goto end_rename;
-       new_inode = new_dentry->d_inode;
--      new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
-+      new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de, NULL);
-       if (new_bh) {
-               if (!new_inode) {
-                       brelse(new_bh);
-@@ -2651,7 +3059,7 @@ static int ext4_rename(struct inode *old
-                       goto end_rename;
-       }
-       if (!new_bh) {
--              retval = ext4_add_entry(handle, new_dentry, old_inode);
-+              retval = ext4_add_entry(handle, new_dentry, old_inode, NULL);
-               if (retval)
-                       goto end_rename;
-       } else {
-@@ -2693,7 +3101,8 @@ static int ext4_rename(struct inode *old
-               struct buffer_head *old_bh2;
-               struct ext4_dir_entry_2 *old_de2;
+Index: linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/Makefile
+===================================================================
+--- linux-2.6.32-504.3.3.el6.x86_64.orig/fs/ext4/Makefile
++++ linux-2.6.32-504.3.3.el6.x86_64/fs/ext4/Makefile
+@@ -6,6 +6,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
  
--              old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
-+              old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name,
-+                                        &old_de2, NULL);
-               if (old_bh2) {
-                       retval = ext4_delete_entry(handle, old_dir,
-                                                  old_de2, old_bh2);
---- linux-2.6.32-131.6.1/fs/ext4/inode.c       2011-10-06 20:10:49.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/fs/ext4/inode.c   2011-12-01 22:02:11.000000000 +0800
-@@ -5112,7 +5112,7 @@ struct inode *ext4_iget(struct super_blo
-       if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT))
-               ei->i_file_acl |=
-                       ((__u64)le16_to_cpu(raw_inode->i_file_acl_high)) << 32;
--      inode->i_size = ext4_isize(raw_inode);
-+      inode->i_size = ext4_isize(sb, raw_inode);
-       ei->i_disksize = inode->i_size;
- #ifdef CONFIG_QUOTA
-       ei->i_reserved_quota = 0;
---- linux-2.6.32-131.6.1/fs/ext4/Makefile      2011-10-06 20:10:49.000000000 +0800
-+++ linux-2.6.32-131.6.1-pdo/fs/ext4/Makefile  2011-10-06 12:21:30.000000000 +0800
-@@ -7,7 +7,7 @@ obj-$(CONFIG_EXT4_FS) += ext4.o
  ext4-y        := balloc.o bitmap.o dir.o file.o fsync.o ialloc.o inode.o \
                ioctl.o namei.o super.o symlink.o hash.o resize.o extents.o \
++              htree_lock.o \
                ext4_jbd2.o migrate.o mballoc.o block_validity.o move_extent.o \
--              mmp.o
-+              htree_lock.o mmp.o
+               mmp.o
  
- ext4-$(CONFIG_EXT4_FS_XATTR)          += xattr.o xattr_user.o xattr_trusted.o
- ext4-$(CONFIG_EXT4_FS_POSIX_ACL)      += acl.o