Skip to contentSkip to conversation

xfs_repair: fix max block offset test

PATCH In review Eric Sandeen 3 participants 7 messages Dec 11, 2014 – Dec 19, 2014

lore
Share & tools
lore

Eric Sandeen

author
#0 lore raw

PATCH 1/1 xfs_repair: fix max block offset test

Eryu pointed out that in fstest xfs/071, we find corruption reported at the end. This test attempts to do IO at the maximum possible offsets, and repair yields:

inode 1027 - extent offset too large - start 70, count 1, offset 2251799813685247 correcting nextents for inode 1027
bad data fork in inode 1027
would have cleared inode 1027

Repair is complaining that an extent *starts* at the maximum block, but AFAICT, starting there is just fine, as long as we also end there. i.e. a one-block extent at the limit is just fine.

So change the xfs_repair test to allow this situation.

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

---

repair/dinode.c +3 -1
index 38a6562..ca57a61 100644
@@ -667,7 +667,9 @@ _("inode %" PRIu64 " - bad extent overflows - start %" PRIu64 ", "
 					irec.br_startoff);
 				goto done;
 		}
-		if (irec.br_startoff >= fs_max_file_offset)  {
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
 			do_warn(
 _("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
   "count %" PRIu64 ", offset %" PRIu64 "\n"),

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

#1 lore raw
Quoted 14 lines · Eric Sandeen

Eryu pointed out that in fstest xfs/071, we find corruption reported at the end. This test attempts to do IO at the maximum possible offsets, and repair yields:

inode 1027 - extent offset too large - start 70, count 1, offset 2251799813685247 correcting nextents for inode 1027
bad data fork in inode 1027
would have cleared inode 1027

Repair is complaining that an extent *starts* at the maximum block, but AFAICT, starting there is just fine, as long as we also end there. i.e. a one-block extent at the limit is just fine.

Quoted 5 lines

So change the xfs_repair test to allow this situation.

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

---

Thought this looked familiar:

http://oss.sgi.com/archives/xfs/2014-09/msg00524.html

Either one is fine with me. If we use the fix below, I think we should update the error message since it technically refers to the extent offset and we slightly tweak the meaning of the failure.

Brian

Quoted 17 lines
repair/dinode.c +3 -1
index 38a6562..ca57a61 100644
@@ -667,7 +667,9 @@ _("inode %" PRIu64 " - bad extent overflows - start %" PRIu64 ", "
 					irec.br_startoff);
 				goto done;
 		}
-		if (irec.br_startoff >= fs_max_file_offset)  {
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
 			do_warn(
 _("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
   "count %" PRIu64 ", offset %" PRIu64 "\n"),
Quoted 4 lines

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

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

author
#2 lore raw
Quoted 3 lines · Brian Foster

Thought this looked familiar:

http://oss.sgi.com/archives/xfs/2014-09/msg00524.html

oh, whoops. let me go read that, sorry!

-Eric

Quoted 23 lines

Either one is fine with me. If we use the fix below, I think we should update the error message since it technically refers to the extent offset and we slightly tweak the meaning of the failure.

Brian

Quoted 17 lines
repair/dinode.c +3 -1
index 38a6562..ca57a61 100644
@@ -667,7 +667,9 @@ _("inode %" PRIu64 " - bad extent overflows - start %" PRIu64 ", "
 					irec.br_startoff);
 				goto done;
 		}
-		if (irec.br_startoff >= fs_max_file_offset)  {
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
 			do_warn(
 _("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
   "count %" PRIu64 ", offset %" PRIu64 "\n"),
Quoted 4 lines
Quoted 4 lines

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

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

#3 lore raw
In reply to Brian Foster

On 12/18/14 9:18 AM, Brian Foster wrote:

hought this looked familiar:

Quoted 8 lines

http://oss.sgi.com/archives/xfs/2014-09/msg00524.html

Either one is fine with me. If we use the fix below, I think we should update the error message since it technically refers to the extent offset and we slightly tweak the meaning of the failure.

Brian

Ok, so here's what I wonder about with your patch.

+ if (irec.br_startoff > fs_max_file_offset) {

What if the extent starting logical offset is *at* fs_max_file_offset (which is sadly in blocks, not bytes, for extra confusion points), but the extent is 1GB in length?

Isn't this a problem because we've now got files in the block which are beyond fs_max_file_offset, and probably also beyond sb_maxbytes?

I'm not sure I follow why you feel the message needs to change with my version:

+ /* Ensure this extent does not extend beyond the max offset */
+ if (irec.br_startoff + irec.br_blockcount - 1 >

+							fs_max_file_offset) {
                        do_warn(

_("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
"count %" PRIu64 ", offset %" PRIu64 "\n"),
ino, irec.br_startblock, irec.br_blockcount, irec.br_startoff);

If any (block) offset in the file is beyond the max, then we say the offset is too large, and give it's physical block, its length, and its starting offset (granted, the message is a bit *weird* - br_startblock really doesn't matter here, but I don't think I've fundamentally changed the meaning of the test).

However, I've come to learn that you are almost always right about subtleties like this, so perhaps you can explain again to me, using smaller words. :)

-Eric

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

#4 lore raw
Quoted 14 lines · Eric Sandeen

On 12/18/14 9:18 AM, Brian Foster wrote:

hought this looked familiar:

Quoted 8 lines

http://oss.sgi.com/archives/xfs/2014-09/msg00524.html

Either one is fine with me. If we use the fix below, I think we should update the error message since it technically refers to the extent offset and we slightly tweak the meaning of the failure.

Brian

Ok, so here's what I wonder about with your patch.

Quoted 9 lines

+ if (irec.br_startoff > fs_max_file_offset) {

What if the extent starting logical offset is *at* fs_max_file_offset (which is sadly in blocks, not bytes, for extra confusion points), but the extent is 1GB in length?

Isn't this a problem because we've now got files in the block which are beyond fs_max_file_offset, and probably also beyond sb_maxbytes?

Yeah... at the time I simply viewed this as outside the scope of this particular check due to the logic and wording. It makes sense to widen the logic to catch anything that starts before yet extends past max offset, assuming we aren't actually checking that anywhere else.

Quoted 12 lines

I'm not sure I follow why you feel the message needs to change with my version:

repair/dinode.c +3 -0
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
                        do_warn(

_("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
"count %" PRIu64 ", offset %" PRIu64 "\n"),
ino, irec.br_startblock, irec.br_blockcount, irec.br_startoff);

Quoted 4 lines

If any (block) offset in the file is beyond the max, then we say the offset is too large, and give it's physical block, its length, and its starting offset (granted, the message is a bit *weird* - br_startblock really doesn't matter here, but I don't think I've fundamentally changed the meaning of the test).

As discussed on irc, I read "extent offset" as the offset of the extent in the file, so it seemed a bit confusing if br_startoff isn't necessarily the problem.

Brian

Quoted 10 lines

However, I've come to learn that you are almost always right about subtleties like this, so perhaps you can explain again to me, using smaller words. :)

-Eric

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

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

#5 lore raw

PATCH v2 1/1 xfs_repair: fix max block offset test

In reply to Eric Sandeen

Eryu pointed out that in fstest xfs/071, we find corruption reported at the end. This test attempts to do IO at the maximum possible offsets, and repair yields:

inode 1027 - extent offset too large - start 70, count 1, offset 2251799813685247 correcting nextents for inode 1027
bad data fork in inode 1027
would have cleared inode 1027

Repair is complaining that an extent *starts* at the maximum block, but AFAICT, starting there is just fine, as long as we also end there. i.e. a one-block extent at the limit is just fine.

So change the xfs_repair test to allow this situation.

Also, the warning text is a bit unclear, mixing in the physical block w/ the logical block... rearrange that a little to make it obvious.

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

---

V2: Update the warning text

repair/dinode.c +7 -5
index 38a6562..59824ec 100644
@@ -667,12 +667,14 @@ _("inode %" PRIu64 " - bad extent overflows - start %" PRIu64 ", "
 					irec.br_startoff);
 				goto done;
 		}
-		if (irec.br_startoff >= fs_max_file_offset)  {
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
 			do_warn(
-_("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
-  "count %" PRIu64 ", offset %" PRIu64 "\n"),
-				ino, irec.br_startblock, irec.br_blockcount,
-				irec.br_startoff);
+_("inode %" PRIu64 " - extent exceeds max offset - start %" PRIu64 ", "
+  "count %" PRIu64 ", physical block %" PRIu64 "\n"),
+				ino, irec.br_startoff, irec.br_blockcount,
+				irec.br_startblock);
 			goto done;
 		}

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

Reviewed-by on v2 1/1
#6 lore raw
Quoted 16 lines · Eric Sandeen

Eryu pointed out that in fstest xfs/071, we find corruption reported at the end. This test attempts to do IO at the maximum possible offsets, and repair yields:

inode 1027 - extent offset too large - start 70, count 1, offset 2251799813685247 correcting nextents for inode 1027
bad data fork in inode 1027
would have cleared inode 1027

Repair is complaining that an extent *starts* at the maximum block, but AFAICT, starting there is just fine, as long as we also end there. i.e. a one-block extent at the limit is just fine.

So change the xfs_repair test to allow this situation.

Quoted 7 lines

Also, the warning text is a bit unclear, mixing in the physical block w/ the logical block... rearrange that a little to make it obvious.

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

---

Looks good to me. Thanks for the fixup. :)

Reviewed-by Brian Foster <[email protected]>
Quoted 29 lines

V2: Update the warning text

repair/dinode.c +7 -5
index 38a6562..59824ec 100644
@@ -667,12 +667,14 @@ _("inode %" PRIu64 " - bad extent overflows - start %" PRIu64 ", "
 					irec.br_startoff);
 				goto done;
 		}
-		if (irec.br_startoff >= fs_max_file_offset)  {
+		/* Ensure this extent does not extend beyond the max offset */
+		if (irec.br_startoff + irec.br_blockcount - 1 >
+							fs_max_file_offset) {
 			do_warn(
-_("inode %" PRIu64 " - extent offset too large - start %" PRIu64 ", "
-  "count %" PRIu64 ", offset %" PRIu64 "\n"),
-				ino, irec.br_startblock, irec.br_blockcount,
-				irec.br_startoff);
+_("inode %" PRIu64 " - extent exceeds max offset - start %" PRIu64 ", "
+  "count %" PRIu64 ", physical block %" PRIu64 "\n"),
+				ino, irec.br_startoff, irec.br_blockcount,
+				irec.br_startblock);
 			goto done;
 		}
Quoted 4 lines

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

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