LU-5626 ldiskfs: update non-htree dotdot in rename

In 2.4+, when renaming a directory, its old dotdot entry will
be removed firstly, then the new dotdot entry is inserted, and
ldiskfs tries to append FID-in-dirent to the new entry.
But the space for dotdot entry may not be enough to hold
the new dotdot with FID-in-dirent, such as an MDT device
restored from file-level backup, or a device upgraded from 1.8.

In that case, for non-HTree directories, the ".." entry
will be written in the next available space in the directory
block.  This is invalid, as the ".." entry must be the
second entry in the block.

The same bug was fixed for HTree directories in LU-2638.
As Fan Yong said then: we do not want to introduce
complex logic to handle directory data moving, instead, in
such case, ignore the FID-in-dirent for the new dotdot entry,
and just insert the new dotdot entry.

There is one known flaw: This patch, like the one for
LU-2638, skips the entire data section rather than just
the FID.  This could cause trouble if something else ever
uses this section with ".." entries.

Signed-off-by: Patrick Farrell <paf@cray.com>
Change-Id: I57fc492e694973f5020191e4e2c79c74c7c4f18c
Reviewed-on: http://review.whamcloud.com/11939
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Tested-by: Jenkins
Reviewed-by: Fan Yong <fan.yong@intel.com>
Reviewed-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Reviewed-by: James Simmons <uja.ornl@gmail.com>
Tested-by: Maloo <hpdd-maloo@intel.com>

diff --git a/ldiskfs/kernel_patches/patches/rhel6.3/ext4-data-in-dirent.patch b/ldiskfs/kernel_patches/patches/rhel6.3/ext4-data-in-dirent.patch
index ae8e51e..5312c9b 100644 (file)
--- a/ldiskfs/kernel_patches/patches/rhel6.3/ext4-data-in-dirent.patch
+++ b/ldiskfs/kernel_patches/patches/rhel6.3/ext4-data-in-dirent.patch
@@ -319,12 +319,15 @@ Index: linux-stage/fs/ext4/ext4.h
                        if (de > to)
                                memmove(to, de, rec_len);
                        to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
-@@ -1335,10 +1345,16 @@ static int add_dirent_to_buf(handle_t *h
+@@ -1334,11 +1344,28 @@ static int add_dirent_to_buf(handle_t *h
+       int             namelen = dentry->d_name.len;
        unsigned int    offset = 0;
        unsigned int    blocksize = dir->i_sb->s_blocksize;
-       unsigned short  reclen;
+-      unsigned short  reclen;
 -      int             nlen, rlen, err;
++      unsigned short  reclen, dotdot_reclen = 0;
 +      int             nlen, rlen, err, dlen = 0;
++      bool            is_dotdot = false, write_short_dotdot = false;
 +      unsigned char   *data;
        char            *top;
  
@@ -334,20 +337,49 @@ Index: linux-stage/fs/ext4/ext4.h
 +      if (data)
 +              dlen = (*data) + 1;
 +
++      is_dotdot = (namelen == 2 &&
++                   memcmp(dentry->d_name.name, "..", 2) == 0);
++
++      /* dotdot entries must be in the second place in a directory block,
++       * so calculate an alternate length without the FID so they can
++       * always be made to fit in the existing slot - LU-5626 */
++      if (is_dotdot)
++              dotdot_reclen = __EXT4_DIR_REC_LEN(namelen);
++
 +      reclen = __EXT4_DIR_REC_LEN(namelen + dlen);
++
        if (!de) {
                de = (struct ext4_dir_entry_2 *)bh->b_data;
                top = bh->b_data + blocksize - reclen;
-@@ -1348,7 +1364,7 @@ static int add_dirent_to_buf(handle_t *h
+@@ -1348,10 +1375,25 @@ static int add_dirent_to_buf(handle_t *h
                                return -EIO;
                        if (ext4_match(namelen, name, de))
                                return -EEXIST;
 -                      nlen = EXT4_DIR_REC_LEN(de->name_len);
 +                      nlen = EXT4_DIR_REC_LEN(de);
                        rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
-                       if ((de->inode? rlen - nlen: rlen) >= reclen)
+-                      if ((de->inode? rlen - nlen: rlen) >= reclen)
++                      /* Check first for enough space for the full entry */
++                      if ((de->inode ? rlen - nlen : rlen) >= reclen)
                                break;
-@@ -1366,7 +1382,7 @@ static int add_dirent_to_buf(handle_t *h
++                      /* Then for dotdot entries, check for the smaller space
++                       * required for just the entry, no FID */
++                      if (is_dotdot) {
++                              if ((de->inode ? rlen - nlen : rlen) >=
++                                  dotdot_reclen) {
++                                      write_short_dotdot = true;
++                                      break;
++                              }
++                              /* The new ".." entry mut be written over the
++                               * previous ".." entry, which is the first
++                               * entry traversed by this scan.  If it doesn't
++                               * fit, something is badly wrong, so -EIO. */
++                              return -EIO;
++                      }
+                       de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
+                       offset += rlen;
+               }
+@@ -1366,7 +1408,7 @@ static int add_dirent_to_buf(handle_t *h
        }
  
        /* By now the buffer is marked for journaling */
@@ -356,11 +388,12 @@ Index: linux-stage/fs/ext4/ext4.h
        rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
        if (de->inode) {
                struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
-@@ -1382,6 +1398,12 @@ static int add_dirent_to_buf(handle_t *h
+@@ -1382,6 +1424,13 @@ static int add_dirent_to_buf(handle_t *h
                de->inode = 0;
        de->name_len = namelen;
        memcpy(de->name, name, namelen);
-+      if (data) {
++      /* If we're writing the short form of "dotdot", don't add the data section */
++      if (data && !write_short_dotdot) {
 +              de->name[namelen] = 0;
 +              memcpy(&de->name[namelen + 1], data, *(char *) data);
 +              de->file_type |= EXT4_DIRENT_LUFID;
@@ -369,7 +402,7 @@ Index: linux-stage/fs/ext4/ext4.h
        /*
         * XXX shouldn't update any times until successful
         * completion of syscall, but too many callers depend
-@@ -1480,7 +1502,8 @@ static int make_indexed_dir(handle_t *ha
+@@ -1480,7 +1529,8 @@ static int make_indexed_dir(handle_t *ha
  
        dx_set_block(entries, 1);
        dx_set_count(entries, 1);
@@ -379,7 +412,7 @@ Index: linux-stage/fs/ext4/ext4.h
  
        /* Initialize as for dx_probe */
        hinfo.hash_version = dx_info->hash_version;
-@@ -1523,6 +1546,8 @@ static int ext4_update_dotdot(handle_t *
+@@ -1523,6 +1573,8 @@ static int ext4_update_dotdot(handle_t *
        struct buffer_head * dir_block;
        struct ext4_dir_entry_2 * de;
        int len, journal = 0, err = 0;
@@ -388,7 +421,7 @@ Index: linux-stage/fs/ext4/ext4.h
  
        if (IS_ERR(handle))
                return PTR_ERR(handle);
-@@ -1538,19 +1563,24 @@ static int ext4_update_dotdot(handle_t *
+@@ -1538,19 +1590,24 @@ static int ext4_update_dotdot(handle_t *
        /* the first item must be "." */
        assert(de->name_len == 1 && de->name[0] == '.');
        len = le16_to_cpu(de->rec_len);
@@ -418,7 +451,7 @@ Index: linux-stage/fs/ext4/ext4.h
        de = (struct ext4_dir_entry_2 *)
                        ((char *) de + le16_to_cpu(de->rec_len));
        if (!journal) {
-@@ -1564,10 +1594,15 @@ static int ext4_update_dotdot(handle_t *
+@@ -1564,10 +1621,15 @@ static int ext4_update_dotdot(handle_t *
        if (len > 0)
                de->rec_len = cpu_to_le16(len);
        else
@@ -436,7 +469,7 @@ Index: linux-stage/fs/ext4/ext4.h
  
  out_journal:
        if (journal) {
-@@ -1989,12 +2024,13 @@ retry:
+@@ -1991,12 +2053,13 @@ retry:
  /* Initialize @inode as a subdirectory of @dir, and add the
   * "." and ".." entries into the first directory block. */
  int ext4_add_dot_dotdot(handle_t *handle, struct inode * dir,
@@ -452,7 +485,7 @@ Index: linux-stage/fs/ext4/ext4.h
  
        if (IS_ERR(handle))
                return PTR_ERR(handle);
-@@ -2015,17 +2051,32 @@ int ext4_add_dot_dotdot(handle_t *handle
+@@ -2017,17 +2080,32 @@ int ext4_add_dot_dotdot(handle_t *handle
        de = (struct ext4_dir_entry_2 *) dir_block->b_data;
        de->inode = cpu_to_le32(inode->i_ino);
        de->name_len = 1;
@@ -488,7 +521,7 @@ Index: linux-stage/fs/ext4/ext4.h
        inode->i_nlink = 2;
        BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
        err = ext4_handle_dirty_metadata(handle, inode, dir_block);
-@@ -2064,7 +2115,7 @@ retry:
+@@ -2066,7 +2144,7 @@ retry:
        if (IS_ERR(inode))
                goto out_stop;
  
@@ -497,7 +530,7 @@ Index: linux-stage/fs/ext4/ext4.h
        if (err)
                goto out_clear_inode;
  
-@@ -2103,7 +2154,7 @@ static int empty_dir(struct inode *inode
+@@ -2105,7 +2183,7 @@ static int empty_dir(struct inode *inode
        int err = 0;
  
        sb = inode->i_sb;