ocfs2: Add directory block trailers.
authorMark Fasheh <mfasheh@suse.com>
Thu, 11 Dec 2008 01:36:25 +0000 (17:36 -0800)
committerMark Fasheh <mfasheh@suse.com>
Mon, 5 Jan 2009 16:40:34 +0000 (08:40 -0800)
Future ocfs2 features metaecc and indexed directories need to store a
little bit of data in each dirblock.  For compatibility, we place this
in a trailer at the end of the dirblock.  The trailer plays itself as an
empty dirent, so that if the features are turned off, it can be reused
without requiring a tunefs scan.

This code adds the trailer and validates it when the block is read in.

[ Mark is the original author, but I reinserted this code before his
  dir index work.  -- Joel ]

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
fs/ocfs2/dir.c
fs/ocfs2/ocfs2.h
fs/ocfs2/ocfs2_fs.h

index 45e4e03..1efd0ab 100644 (file)
@@ -83,6 +83,63 @@ static int ocfs2_do_extend_dir(struct super_block *sb,
                               struct ocfs2_alloc_context *meta_ac,
                               struct buffer_head **new_bh);
 
+/*
+ * These are distinct checks because future versions of the file system will
+ * want to have a trailing dirent structure independent of indexing.
+ */
+static int ocfs2_dir_has_trailer(struct inode *dir)
+{
+       if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+               return 0;
+
+       return ocfs2_meta_ecc(OCFS2_SB(dir->i_sb));
+}
+
+static int ocfs2_supports_dir_trailer(struct ocfs2_super *osb)
+{
+       return ocfs2_meta_ecc(osb);
+}
+
+static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
+{
+       return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
+}
+
+#define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
+
+/*
+ * XXX: This is executed once on every dirent. We should consider optimizing
+ * it.
+ */
+static int ocfs2_skip_dir_trailer(struct inode *dir,
+                                 struct ocfs2_dir_entry *de,
+                                 unsigned long offset,
+                                 unsigned long blklen)
+{
+       unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
+
+       if (!ocfs2_dir_has_trailer(dir))
+               return 0;
+
+       if (offset != toff)
+               return 0;
+
+       return 1;
+}
+
+static void ocfs2_init_dir_trailer(struct inode *inode,
+                                  struct buffer_head *bh)
+{
+       struct ocfs2_dir_block_trailer *trailer;
+
+       trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
+       strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
+       trailer->db_compat_rec_len =
+                       cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
+       trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
+       trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
+}
+
 /*
  * bh passed here can be an inode block or a dir data block, depending
  * on the inode inline data flag.
@@ -232,16 +289,60 @@ static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
 {
        int rc = 0;
        struct buffer_head *tmp = *bh;
+       struct ocfs2_dir_block_trailer *trailer;
 
        rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
                                    ocfs2_validate_dir_block);
-       if (rc)
+       if (rc) {
                mlog_errno(rc);
+               goto out;
+       }
+
+       /*
+        * We check the trailer here rather than in
+        * ocfs2_validate_dir_block() because that function doesn't have
+        * the inode to test.
+        */
+       if (!(flags & OCFS2_BH_READAHEAD) &&
+           ocfs2_dir_has_trailer(inode)) {
+               trailer = ocfs2_trailer_from_bh(tmp, inode->i_sb);
+               if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
+                       rc = -EINVAL;
+                       ocfs2_error(inode->i_sb,
+                                   "Invalid dirblock #%llu: "
+                                   "signature = %.*s\n",
+                                   (unsigned long long)tmp->b_blocknr, 7,
+                                   trailer->db_signature);
+                       goto out;
+               }
+               if (le64_to_cpu(trailer->db_blkno) != tmp->b_blocknr) {
+                       rc = -EINVAL;
+                       ocfs2_error(inode->i_sb,
+                                   "Directory block #%llu has an invalid "
+                                   "db_blkno of %llu",
+                                   (unsigned long long)tmp->b_blocknr,
+                                   (unsigned long long)le64_to_cpu(trailer->db_blkno));
+                       goto out;
+               }
+               if (le64_to_cpu(trailer->db_parent_dinode) !=
+                   OCFS2_I(inode)->ip_blkno) {
+                       rc = -EINVAL;
+                       ocfs2_error(inode->i_sb,
+                                   "Directory block #%llu on dinode "
+                                   "#%llu has an invalid parent_dinode "
+                                   "of %llu",
+                                   (unsigned long long)tmp->b_blocknr,
+                                   (unsigned long long)OCFS2_I(inode)->ip_blkno,
+                                   (unsigned long long)le64_to_cpu(trailer->db_blkno));
+                       goto out;
+               }
+       }
 
        /* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
-       if (!rc && !*bh)
+       if (!*bh)
                *bh = tmp;
 
+out:
        return rc ? -EIO : 0;
 }
 
@@ -581,6 +682,16 @@ int __ocfs2_add_entry(handle_t *handle,
                        goto bail;
                }
 
+               /* We're guaranteed that we should have space, so we
+                * can't possibly have hit the trailer...right? */
+               mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
+                               "Hit dir trailer trying to insert %.*s "
+                               "(namelen %d) into directory %llu.  "
+                               "offset is %lu, trailer offset is %d\n",
+                               namelen, name, namelen,
+                               (unsigned long long)parent_fe_bh->b_blocknr,
+                               offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
+
                if (ocfs2_dirent_would_fit(de, rec_len)) {
                        dir->i_mtime = dir->i_ctime = CURRENT_TIME;
                        retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
@@ -622,6 +733,7 @@ int __ocfs2_add_entry(handle_t *handle,
                        retval = 0;
                        goto bail;
                }
+
                offset += le16_to_cpu(de->rec_len);
                de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
        }
@@ -1059,9 +1171,15 @@ int ocfs2_empty_dir(struct inode *inode)
        return !priv.seen_other;
 }
 
-static void ocfs2_fill_initial_dirents(struct inode *inode,
-                                      struct inode *parent,
-                                      char *start, unsigned int size)
+/*
+ * Fills "." and ".." dirents in a new directory block. Returns dirent for
+ * "..", which might be used during creation of a directory with a trailing
+ * header. It is otherwise safe to ignore the return code.
+ */
+static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
+                                                         struct inode *parent,
+                                                         char *start,
+                                                         unsigned int size)
 {
        struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
 
@@ -1078,6 +1196,8 @@ static void ocfs2_fill_initial_dirents(struct inode *inode,
        de->name_len = 2;
        strcpy(de->name, "..");
        ocfs2_set_de_type(de, S_IFDIR);
+
+       return de;
 }
 
 /*
@@ -1130,10 +1250,15 @@ static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
                                 struct ocfs2_alloc_context *data_ac)
 {
        int status;
+       unsigned int size = osb->sb->s_blocksize;
        struct buffer_head *new_bh = NULL;
+       struct ocfs2_dir_entry *de;
 
        mlog_entry_void();
 
+       if (ocfs2_supports_dir_trailer(osb))
+               size = ocfs2_dir_trailer_blk_off(parent->i_sb);
+
        status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
                                     data_ac, NULL, &new_bh);
        if (status < 0) {
@@ -1151,8 +1276,9 @@ static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
        }
        memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 
-       ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data,
-                                  osb->sb->s_blocksize);
+       de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
+       if (ocfs2_supports_dir_trailer(osb))
+               ocfs2_init_dir_trailer(inode, new_bh);
 
        status = ocfs2_journal_dirty(handle, new_bh);
        if (status < 0) {
@@ -1193,13 +1319,27 @@ int ocfs2_fill_new_dir(struct ocfs2_super *osb,
                                     data_ac);
 }
 
+/*
+ * Expand rec_len of the rightmost dirent in a directory block so that it
+ * contains the end of our valid space for dirents. We do this during
+ * expansion from an inline directory to one with extents. The first dir block
+ * in that case is taken from the inline data portion of the inode block.
+ *
+ * We add the dir trailer if this filesystem wants it.
+ */
 static void ocfs2_expand_last_dirent(char *start, unsigned int old_size,
-                                    unsigned int new_size)
+                                    struct super_block *sb)
 {
        struct ocfs2_dir_entry *de;
        struct ocfs2_dir_entry *prev_de;
        char *de_buf, *limit;
-       unsigned int bytes = new_size - old_size;
+       unsigned int new_size = sb->s_blocksize;
+       unsigned int bytes;
+
+       if (ocfs2_supports_dir_trailer(OCFS2_SB(sb)))
+               new_size = ocfs2_dir_trailer_blk_off(sb);
+
+       bytes = new_size - old_size;
 
        limit = start + old_size;
        de_buf = start;
@@ -1316,8 +1456,9 @@ static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
        memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
        memset(dirdata_bh->b_data + i_size_read(dir), 0,
               sb->s_blocksize - i_size_read(dir));
-       ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir),
-                                sb->s_blocksize);
+       ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), sb);
+       if (ocfs2_supports_dir_trailer(osb))
+               ocfs2_init_dir_trailer(dir, dirdata_bh);
 
        ret = ocfs2_journal_dirty(handle, dirdata_bh);
        if (ret) {
@@ -1604,9 +1745,15 @@ do_extend:
                goto bail;
        }
        memset(new_bh->b_data, 0, sb->s_blocksize);
+
        de = (struct ocfs2_dir_entry *) new_bh->b_data;
        de->inode = 0;
-       de->rec_len = cpu_to_le16(sb->s_blocksize);
+       if (ocfs2_dir_has_trailer(dir)) {
+               de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
+               ocfs2_init_dir_trailer(dir, new_bh);
+       } else {
+               de->rec_len = cpu_to_le16(sb->s_blocksize);
+       }
        status = ocfs2_journal_dirty(handle, new_bh);
        if (status < 0) {
                mlog_errno(status);
@@ -1648,11 +1795,21 @@ static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
                                   unsigned int *blocks_wanted)
 {
        int ret;
+       struct super_block *sb = dir->i_sb;
        struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
        struct ocfs2_dir_entry *de, *last_de = NULL;
        char *de_buf, *limit;
        unsigned long offset = 0;
-       unsigned int rec_len, new_rec_len;
+       unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
+
+       /*
+        * This calculates how many free bytes we'd have in block zero, should
+        * this function force expansion to an extent tree.
+        */
+       if (ocfs2_supports_dir_trailer(OCFS2_SB(sb)))
+               free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
+       else
+               free_space = dir->i_sb->s_blocksize - i_size_read(dir);
 
        de_buf = di->id2.i_data.id_data;
        limit = de_buf + i_size_read(dir);
@@ -1669,6 +1826,11 @@ static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
                        ret = -EEXIST;
                        goto out;
                }
+               /*
+                * No need to check for a trailing dirent record here as
+                * they're not used for inline dirs.
+                */
+
                if (ocfs2_dirent_would_fit(de, rec_len)) {
                        /* Ok, we found a spot. Return this bh and let
                         * the caller actually fill it in. */
@@ -1689,7 +1851,7 @@ static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
         * dirent can be found.
         */
        *blocks_wanted = 1;
-       new_rec_len = le16_to_cpu(last_de->rec_len) + (dir->i_sb->s_blocksize - i_size_read(dir));
+       new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
        if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
                *blocks_wanted = 2;
 
@@ -1707,6 +1869,7 @@ static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
        struct ocfs2_dir_entry *de;
        struct super_block *sb = dir->i_sb;
        int status;
+       int blocksize = dir->i_sb->s_blocksize;
 
        status = ocfs2_read_dir_block(dir, 0, &bh, 0);
        if (status) {
@@ -1748,6 +1911,11 @@ static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
                        status = -EEXIST;
                        goto bail;
                }
+
+               if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
+                                          blocksize))
+                       goto next;
+
                if (ocfs2_dirent_would_fit(de, rec_len)) {
                        /* Ok, we found a spot. Return this bh and let
                         * the caller actually fill it in. */
@@ -1756,6 +1924,7 @@ static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
                        status = 0;
                        goto bail;
                }
+next:
                offset += le16_to_cpu(de->rec_len);
                de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
        }
index bad87d0..ad5c24a 100644 (file)
@@ -470,6 +470,9 @@ static inline int ocfs2_uses_extended_slot_map(struct ocfs2_super *osb)
 #define OCFS2_IS_VALID_XATTR_BLOCK(ptr)                                        \
        (!strcmp((ptr)->xb_signature, OCFS2_XATTR_BLOCK_SIGNATURE))
 
+#define OCFS2_IS_VALID_DIR_TRAILER(ptr)                                        \
+       (!strcmp((ptr)->db_signature, OCFS2_DIR_TRAILER_SIGNATURE))
+
 static inline unsigned long ino_from_blkno(struct super_block *sb,
                                           u64 blkno)
 {
index 290fa26..af0013b 100644 (file)
@@ -65,6 +65,7 @@
 #define OCFS2_EXTENT_BLOCK_SIGNATURE   "EXBLK01"
 #define OCFS2_GROUP_DESC_SIGNATURE      "GROUP01"
 #define OCFS2_XATTR_BLOCK_SIGNATURE    "XATTR01"
+#define OCFS2_DIR_TRAILER_SIGNATURE    "DIRTRL1"
 
 /* Compatibility flags */
 #define OCFS2_HAS_COMPAT_FEATURE(sb,mask)                      \
@@ -751,6 +752,34 @@ struct ocfs2_dir_entry {
 /* Actual on-disk length specified by rec_len */
 } __attribute__ ((packed));
 
+/*
+ * Per-block record for the unindexed directory btree. This is carefully
+ * crafted so that the rec_len and name_len records of an ocfs2_dir_entry are
+ * mirrored. That way, the directory manipulation code needs a minimal amount
+ * of update.
+ *
+ * NOTE: Keep this structure aligned to a multiple of 4 bytes.
+ */
+struct ocfs2_dir_block_trailer {
+/*00*/ __le64          db_compat_inode;        /* Always zero. Was inode */
+
+       __le16          db_compat_rec_len;      /* Backwards compatible with
+                                                * ocfs2_dir_entry. */
+       __u8            db_compat_name_len;     /* Always zero. Was name_len */
+       __u8            db_reserved0;
+       __le16          db_reserved1;
+       __le16          db_free_rec_len;        /* Size of largest empty hole
+                                                * in this block. (unused) */
+/*10*/ __u8            db_signature[8];        /* Signature for verification */
+       __le64          db_reserved2;
+       __le64          db_free_next;           /* Next block in list (unused) */
+/*20*/ __le64          db_blkno;               /* Offset on disk, in blocks */
+       __le64          db_parent_dinode;       /* dinode which owns me, in
+                                                  blocks */
+/*30*/ __le64          db_check;               /* Error checking */
+/*40*/
+};
+
 /*
  * On disk allocator group structure for OCFS2
  */