xfs_repair fixes for fragmented multiblock v2 dirs

PATCH In review Eric Sandeen 3 participants 10 messages May 21, 2012 – Jun 1, 2012

lore

Share & tools

lore

Eric Sandeen

author

2012-05-21 22:07 UTC #0 lore raw

cover xfs_repair fixes for fragmented multiblock v2 dirs

3 fixes for xfs_repair on fragmented multiblock v2 dirs.

xfs_repair: Fix fragmented multiblock dir2 handling in blkmap_getn()
xfs_repair: Document & robustify blkmap_next_off()
xfs_repair: handle fragmented multiblock dir2 in process_leaf_node_dir2()

xfs_check seems to handle things just fine, interstingly enough.

-Eric

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

author

2012-05-21 22:10 UTC #1 lore raw

PATCH 1/3 xfs_repair: Fix fragmented multiblock dir2 handling in blkmap_getn()

blkmap_getn() contains a loop which populates an array of extents with mapping information for a dir2 "block," which may consist of multiple filesystem blocks.

As written, the loop re-allocates the array for each new extent, leaking the previously allocated memory and worse, losing the previously filled-in extent information.

Fix this by only allocating the array once, for the maximum possible number of extents - the number of fs blocks in the dir block.

Signed-off-by Eric Sandeen <[email protected]>

---

repair/bmap.c +2 -1

index 2f1c307..c43ca7f 100644
@@ -168,7 +168,8 @@ blkmap_getn(
 		/*
 		 * rare case - multiple extents for a single dir block
 		 */
-		bmp = malloc(nb * sizeof(bmap_ext_t));
+		if (!bmp)
+			bmp = malloc(nb * sizeof(bmap_ext_t));
 		if (!bmp)
 			do_error(_("blkmap_getn malloc failed (%" PRIu64 " bytes)\n"),
 						nb * sizeof(bmap_ext_t));

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Ben Myers

Reviewed-by on v1 1/3

2012-05-31 20:49 UTC #2 lore raw

Re: [PATCH 1/3] xfs_repair: Fix fragmented multiblock dir2 handling in blkmap_getn()

Hey Eric,

Quoted 20 lines · Eric Sandeen

blkmap_getn() contains a loop which populates an array of extents with mapping information for a dir2 "block," which may consist of multiple filesystem blocks.
As written, the loop re-allocates the array for each new extent, leaking the previously allocated memory and worse, losing the previously filled-in extent information.
Fix this by only allocating the array once, for the maximum possible number of extents - the number of fs blocks in the dir block.
Signed-off-by Eric Sandeen <[email protected]>
---
repair/bmap.c +0 -0
index 2f1c307..c43ca7f 100644

repair/bmap.c blkmap_getn( · line 172

Quoted 6 lines

repair/bmap.c +2 -1

@@ -168,7 +168,8 @@ blkmap_getn(
 		/*
 		 * rare case - multiple extents for a single dir block
 		 */
-		bmp = malloc(nb * sizeof(bmap_ext_t));
+		if (!bmp)
+			bmp = malloc(nb * sizeof(bmap_ext_t));

Cool. That's an area where we don't have much test coverage. I don't see a single '-n size' in xfstests.

This looks good to me. FWICS all the callers free it later.

Reviewed-by Ben Myers <[email protected]>

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2012-05-22 01:26 UTC #3 lore raw

In reply to Eric Sandeen

Quoted 9 lines · Eric Sandeen

3 fixes for xfs_repair on fragmented multiblock v2 dirs.
xfs_repair: Fix fragmented multiblock dir2 handling in blkmap_getn()
xfs_repair: Document & robustify blkmap_next_off()
xfs_repair: handle fragmented multiblock dir2 in process_leaf_node_dir2()
xfs_check seems to handle things just fine, interstingly enough.
-Eric

Grr.

Well, I did send the other two.

but it seems to be eaten/stuck/constipated. Again.

Grr.

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2012-05-22 13:41 UTC #4 lore raw

PATCH RESEND 2/3 xfs_repair fixes for fragmented multiblock v2 dirs

In reply to Eric Sandeen

blkmap_next_off() was cryptic (to me), so document what it does. Also catch cases when the passed in extent index 't' is beyond the number of extents in the blkmap, so that:

        ext = blkmap->exts + *t;

doesn't walk off the end of the array into garbage.

Signed-off-by Eric Sandeen <[email protected]>

---

Resending, last one never made it through the mailing list gauntlet...

repair/bmap.c +22 -3

index c43ca7f..85d66dc 100644
@@ -206,8 +206,25 @@ blkmap_last_off(
 	return ext->startoff + ext->blockcount;
 }
 
-/*
- * Return the next offset in a block map.
+/**
+ * blkmap_next_off - Return next logical block offset in a block map.
+ * @blkmap:	blockmap to use
+ * @o:		current file logical block number
+ * @t:		current extent index into blockmap (in/out)
+ *
+ * Given a logical block offset in a file, return the next mapped logical offset
+ * The map index "t" tracks the current extent number in the block map, and
+ * is updated automatically if the returned offset resides within the next
+ * mapped extent.
+ *
+ * If the blockmap contains no extents, or no more logical offsets are mapped,
+ * or the extent index exceeds the number of extents in the map,
+ * return NULLDFILOFF.
+ *
+ * If offset o is beyond extent index t, the first offset in the next extent
+ * after extent t will be returned.
+ *
+ * Intended to be called starting with offset 0, index 0, and iterated.
  */
 xfs_dfiloff_t
 blkmap_next_off(

@@ -223,10 +240,12 @@ blkmap_next_off(
 		*t = 0;
 		return blkmap->exts[0].startoff;
 	}
+	if (*t >= blkmap->nexts)
+		return NULLDFILOFF;
 	ext = blkmap->exts + *t;
 	if (o < ext->startoff + ext->blockcount - 1)
 		return o + 1;
-	if (*t >= blkmap->nexts - 1)
+	if (*t == blkmap->nexts - 1)
 		return NULLDFILOFF;
 	(*t)++;
 	return ext[1].startoff;

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

2012-05-22 13:42 UTC #5 lore raw

PATCH RESEND 3/3 xfs_repair: handle fragmented multiblock dir2 in process_leaf_node_dir2()

In reply to Eric Sandeen

process_leaf_node_dir2() had the following loop:

	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
		...
		ndbno = dbno + mp->m_dirblkfsbs - 1;
		...
	}

which does not account for fragmented multiblock dir2.

ndbno was blindly being advanced by m_dirblkfsbs, but then blkmap_next_off() would return the logical block of the next mapped extent in blkmap, which may be within the current (fragmented) dir2 multi-block, not the next multi-block, because the extent index t hadn't been advanced.

Fix this by calling blkmap_next_off() until ndbno has advanced into the next multiblock dir2 block, thereby keeping the extent index t straight while properly advancing ndbno.

Signed-off-by Eric Sandeen <[email protected]>

---

Resending, first one didn't make it through the mailing list gauntlet

repair/dir2.c +5 -1

index f9562d7..7a614a8 100644
@@ -2003,7 +2003,11 @@ process_leaf_node_dir2(
 	ndbno = NULLDFILOFF;
 	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
 		nex = blkmap_getn(blkmap, dbno, mp->m_dirblkfsbs, &bmp, &lbmp);
-		ndbno = dbno + mp->m_dirblkfsbs - 1;
+		/* Advance through map to last dfs block in this dir block */
+		ndbno = dbno;
+		while (ndbno < dbno + mp->m_dirblkfsbs - 1) {
+			ndbno = blkmap_next_off(blkmap, ndbno, &t);
+		}
 		if (nex == 0) {
 			do_warn(
 _("block %" PRIu64 " for directory inode %" PRIu64 " is missing\n"),

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

author

2012-05-22 21:05 UTC #6 lore raw

PATCH 2/3 xfs_repair: Document & robustify blkmap_next_off()

In reply to Eric Sandeen

blkmap_next_off() was cryptic (to me), so document what it does. Also catch cases when the passed in extent index 't' is beyond the number of extents in the blkmap, so that:

        ext = blkmap->exts + *t;

doesn't walk off the end of the array into garbage.

Signed-off-by Eric Sandeen <[email protected]>

---

repair/bmap.c +22 -3

index c43ca7f..85d66dc 100644
@@ -206,8 +206,25 @@ blkmap_last_off(
 	return ext->startoff + ext->blockcount;
 }
 
-/*
- * Return the next offset in a block map.
+/**
+ * blkmap_next_off - Return next logical block offset in a block map.
+ * @blkmap:	blockmap to use
+ * @o:		current file logical block number
+ * @t:		current extent index into blockmap (in/out)
+ *
+ * Given a logical block offset in a file, return the next mapped logical offset
+ * The map index "t" tracks the current extent number in the block map, and
+ * is updated automatically if the returned offset resides within the next
+ * mapped extent.
+ *
+ * If the blockmap contains no extents, or no more logical offsets are mapped,
+ * or the extent index exceeds the number of extents in the map,
+ * return NULLDFILOFF.
+ *
+ * If offset o is beyond extent index t, the first offset in the next extent
+ * after extent t will be returned.
+ *
+ * Intended to be called starting with offset 0, index 0, and iterated.
  */
 xfs_dfiloff_t
 blkmap_next_off(

@@ -223,10 +240,12 @@ blkmap_next_off(
 		*t = 0;
 		return blkmap->exts[0].startoff;
 	}
+	if (*t >= blkmap->nexts)
+		return NULLDFILOFF;
 	ext = blkmap->exts + *t;
 	if (o < ext->startoff + ext->blockcount - 1)
 		return o + 1;
-	if (*t >= blkmap->nexts - 1)
+	if (*t == blkmap->nexts - 1)
 		return NULLDFILOFF;
 	(*t)++;
 	return ext[1].startoff;

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Ben Myers

Reviewed-by on v1 2/3

2012-05-31 21:21 UTC #7 lore raw

Re: [PATCH 2/3] xfs_repair: Document & robustify blkmap_next_off()

Quoted 47 lines · Eric Sandeen

blkmap_next_off() was cryptic (to me), so document what it does. Also catch cases when the passed in extent index 't' is beyond the number of extents in the blkmap, so that:

        ext = blkmap->exts + *t;

doesn't walk off the end of the array into garbage.

Signed-off-by Eric Sandeen <[email protected]>

---

repair/bmap.c +19 -2

index c43ca7f..85d66dc 100644
@@ -206,8 +206,25 @@ blkmap_last_off(
 	return ext->startoff + ext->blockcount;
 }
 
-/*
- * Return the next offset in a block map.
+/**
+ * blkmap_next_off - Return next logical block offset in a block map.
+ * @blkmap:	blockmap to use
+ * @o:		current file logical block number
+ * @t:		current extent index into blockmap (in/out)
+ *
+ * Given a logical block offset in a file, return the next mapped logical offset
+ * The map index "t" tracks the current extent number in the block map, and
+ * is updated automatically if the returned offset resides within the next
+ * mapped extent.
+ *
+ * If the blockmap contains no extents, or no more logical offsets are mapped,
+ * or the extent index exceeds the number of extents in the map,
+ * return NULLDFILOFF.
+ *
+ * If offset o is beyond extent index t, the first offset in the next extent
+ * after extent t will be returned.
+ *
+ * Intended to be called starting with offset 0, index 0, and iterated.
  */
 xfs_dfiloff_t
 blkmap_next_off(

@@ -223,10 +240,12 @@ blkmap_next_off(
 		*t = 0;
 		return blkmap->exts[0].startoff;
 	}

repair/bmap.c blkmap_next_off( · line 251

Quoted 10 lines

repair/bmap.c +3 -1

+	if (*t >= blkmap->nexts)
+		return NULLDFILOFF;
 	ext = blkmap->exts + *t;
 	if (o < ext->startoff + ext->blockcount - 1)
 		return o + 1;
-	if (*t >= blkmap->nexts - 1)
+	if (*t == blkmap->nexts - 1)
 		return NULLDFILOFF;
 	(*t)++;
 	return ext[1].startoff;

Ok, so you have 50 extents, and when *t == 49 you don't want to go looking at ext[50].startoff. Looks good to me.

Reviewed-by Ben Myers <[email protected]>

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Eric Sandeen

author

2012-05-22 21:05 UTC #8 lore raw

PATCH 3/3 xfs_repair: handle fragmented multiblock dir2 in process_leaf_node_dir2()

In reply to Eric Sandeen

process_leaf_node_dir2() had the following loop:

	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
		...
		ndbno = dbno + mp->m_dirblkfsbs - 1;
		...
	}

which does not account for fragmented multiblock dir2.

Fix this by calling blkmap_next_off() until ndbno has advanced into the next multiblock dir2 block, thereby keeping the extent index t straight while properly advancing ndbno.

Signed-off-by Eric Sandeen <[email protected]>

---

repair/dir2.c +5 -1

index f9562d7..7a614a8 100644
@@ -2003,7 +2003,11 @@ process_leaf_node_dir2(
 	ndbno = NULLDFILOFF;
 	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
 		nex = blkmap_getn(blkmap, dbno, mp->m_dirblkfsbs, &bmp, &lbmp);
-		ndbno = dbno + mp->m_dirblkfsbs - 1;
+		/* Advance through map to last dfs block in this dir block */
+		ndbno = dbno;
+		while (ndbno < dbno + mp->m_dirblkfsbs - 1) {
+			ndbno = blkmap_next_off(blkmap, ndbno, &t);
+		}
 		if (nex == 0) {
 			do_warn(
 _("block %" PRIu64 " for directory inode %" PRIu64 " is missing\n"),

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs

Ben Myers

Reviewed-by on v1 3/3

2012-06-01 20:19 UTC #9 lore raw

Re: [PATCH 3/3] xfs_repair: handle fragmented multiblock dir2 in process_leaf_node_dir2()

Quoted 29 lines · Eric Sandeen

process_leaf_node_dir2() had the following loop:
	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
		...
		ndbno = dbno + mp->m_dirblkfsbs - 1;
		...
	}
which does not account for fragmented multiblock dir2.
ndbno was blindly being advanced by m_dirblkfsbs, but then blkmap_next_off() would return the logical block of the next mapped extent in blkmap, which may be within the current (fragmented) dir2 multi-block, not the next multi-block, because the extent index t hadn't been advanced.
Fix this by calling blkmap_next_off() until ndbno has advanced into the next multiblock dir2 block, thereby keeping the extent index t straight while properly advancing ndbno.
Signed-off-by Eric Sandeen <[email protected]>
---
repair/dir2.c +0 -0
index f9562d7..7a614a8 100644

repair/dir2.c process_leaf_node_dir2( · line 2010

Quoted 9 lines

repair/dir2.c +5 -1

@@ -2003,7 +2003,11 @@ process_leaf_node_dir2(
 	ndbno = NULLDFILOFF;
 	while ((dbno = blkmap_next_off(blkmap, ndbno, &t)) < mp->m_dirleafblk) {
 		nex = blkmap_getn(blkmap, dbno, mp->m_dirblkfsbs, &bmp, &lbmp);
-		ndbno = dbno + mp->m_dirblkfsbs - 1;
+		/* Advance through map to last dfs block in this dir block */
+		ndbno = dbno;
+		while (ndbno < dbno + mp->m_dirblkfsbs - 1) {
+			ndbno = blkmap_next_off(blkmap, ndbno, &t);
+		}

Yep, looks good to me. An important thing to notice is that blkmap_getn handles all extents in a dir2 multiblock...

Reviewed-by Ben Myers <[email protected]>

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs