Whamcloud - gitweb
e2fsck: detect invalid extents at the end of an extent-block
authorDavid Jeffery <djeffery@redhat.com>
Fri, 7 Jun 2013 00:04:33 +0000 (20:04 -0400)
committerTheodore Ts'o <tytso@mit.edu>
Fri, 7 Jun 2013 02:53:31 +0000 (22:53 -0400)
e2fsck does not detect extents which are outside their location in the
extent tree.  This can result in a bad extent at the end of an extent-block
not being detected.

From a part of a dump_extents output:

 1/ 2  37/ 68 143960 - 146679 123826181               2720
 2/ 2   1/  2 143960 - 146679 123785816 - 123788535   2720
 2/ 2   2/  2 146680 - 147583 123788536 - 123789439    904 Uninit <-bad extent
 1/ 2  38/ 68 146680 - 149391 123826182               2712
 2/ 2   1/  2 146680 - 147583     18486 -     19389    904
 2/ 2   2/  2 147584 - 149391 123789440 - 123791247   1808

e2fsck does not detect this bad extent which both overlaps another, valid
extent, and is invalid by being beyond the end of the extent above it in
the tree.

This patch modifies e2fsck to detect this invalid extent and remove it.

Signed-off-by: David Jeffery <djeffery@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
e2fsck/pass1.c
e2fsck/problem.c
e2fsck/problem.h

index de00638..af9afe3 100644 (file)
@@ -1760,11 +1760,11 @@ void e2fsck_clear_inode(e2fsck_t ctx, ext2_ino_t ino,
 
 static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
                             struct process_block_struct *pb,
-                            blk64_t start_block,
+                            blk64_t start_block, blk64_t end_block,
                             ext2_extent_handle_t ehandle)
 {
        struct ext2fs_extent    extent;
-       blk64_t                 blk;
+       blk64_t                 blk, last_lblk;
        e2_blkcnt_t             blockcnt;
        unsigned int            i;
        int                     is_dir, is_leaf;
@@ -1780,6 +1780,7 @@ static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
        while (!pctx->errcode && info.num_entries-- > 0) {
                is_leaf = extent.e_flags & EXT2_EXTENT_FLAGS_LEAF;
                is_dir = LINUX_S_ISDIR(pctx->inode->i_mode);
+               last_lblk = extent.e_lblk + extent.e_len - 1;
 
                problem = 0;
                if (extent.e_pblk == 0 ||
@@ -1788,6 +1789,8 @@ static void scan_extent_node(e2fsck_t ctx, struct problem_context *pctx,
                        problem = PR_1_EXTENT_BAD_START_BLK;
                else if (extent.e_lblk < start_block)
                        problem = PR_1_OUT_OF_ORDER_EXTENTS;
+               else if (end_block && last_lblk > end_block)
+                       problem = PR_1_EXTENT_END_OUT_OF_BOUNDS;
                else if (is_leaf && extent.e_len == 0)
                        problem = PR_1_EXTENT_LENGTH_ZERO;
                else if (is_leaf &&
@@ -1822,10 +1825,9 @@ report_problem:
                }
 
                if (!is_leaf) {
-                       blk64_t lblk;
+                       blk64_t lblk = extent.e_lblk;
 
                        blk = extent.e_pblk;
-                       lblk = extent.e_lblk;
                        pctx->errcode = ext2fs_extent_get(ehandle,
                                                  EXT2_EXTENT_DOWN, &extent);
                        if (pctx->errcode) {
@@ -1847,7 +1849,8 @@ report_problem:
                                if (fix_problem(ctx, problem, pctx))
                                        ext2fs_extent_fix_parents(ehandle);
                        }
-                       scan_extent_node(ctx, pctx, pb, extent.e_lblk, ehandle);
+                       scan_extent_node(ctx, pctx, pb, extent.e_lblk,
+                                        last_lblk, ehandle);
                        if (pctx->errcode)
                                return;
                        pctx->errcode = ext2fs_extent_get(ehandle,
@@ -1928,10 +1931,10 @@ report_problem:
                if (is_dir && extent.e_len > 0)
                        pb->last_db_block = blockcnt - 1;
                pb->previous_block = extent.e_pblk + extent.e_len - 1;
-               start_block = pb->last_block = extent.e_lblk + extent.e_len - 1;
+               start_block = pb->last_block = last_lblk;
                if (is_leaf && !is_dir &&
                    !(extent.e_flags & EXT2_EXTENT_FLAGS_UNINIT))
-                       pb->last_init_lblock = extent.e_lblk + extent.e_len - 1;
+                       pb->last_init_lblock = last_lblk;
        next:
                pctx->errcode = ext2fs_extent_get(ehandle,
                                                  EXT2_EXTENT_NEXT_SIB,
@@ -1967,7 +1970,7 @@ static void check_blocks_extents(e2fsck_t ctx, struct problem_context *pctx,
                ctx->extent_depth_count[info.max_depth]++;
        }
 
-       scan_extent_node(ctx, pctx, pb, 0, ehandle);
+       scan_extent_node(ctx, pctx, pb, 0, 0, ehandle);
        if (pctx->errcode &&
            fix_problem(ctx, PR_1_EXTENT_ITERATE_FAILURE, pctx)) {
                pb->num_blocks = 0;
index 05ef626..6d03765 100644 (file)
@@ -954,6 +954,12 @@ static struct e2fsck_problem problem_table[] = {
             "Logical start %b does not match logical start %c at next level.  "),
          PROMPT_FIX, 0 },
 
+       /* Extent end is out of bounds for the tree */
+       { PR_1_EXTENT_END_OUT_OF_BOUNDS,
+         N_("@i %i, end of extent exceeds allowed value\n\t(logical @b %c, physical @b %b, len %N)\n"),
+         PROMPT_CLEAR, 0 },
+
+
        /* Pass 1b errors */
 
        /* Pass 1B: Rescan for duplicate/bad blocks */
index aed524d..b578678 100644 (file)
@@ -561,6 +561,7 @@ struct problem_context {
 /* Index start doesn't match start of next extent down */
 #define PR_1_EXTENT_INDEX_START_INVALID        0x01006D
 
+#define PR_1_EXTENT_END_OUT_OF_BOUNDS  0x01006E
 /*
  * Pass 1b errors
  */