Skip to contentSkip to conversation

dm stripe: Fix bounds

PATCH Alasdair G Kergon 3 participants 11 messages Mar 16, 2006 – Oct 17, 2006 Device mapper

lore
Share & tools
lore

Alasdair G Kergon

author maintainer
#0 lore raw

From: Kevin Corry <[email protected]>

The dm-stripe target currently does not enforce that the size of a stripe device be a multiple of the chunk-size. Under certain conditions, this can lead to I/O requests going off the end of an underlying device. This test-case shows one example.

echo "0 100 linear /dev/hdb1 0" | dmsetup create linear0
echo "0 100 linear /dev/hdb1 100" | dmsetup create linear1 echo "0 200 striped 2 32 /dev/mapper/linear0 0 /dev/mapper/linear1 0" | \
dmsetup create stripe0
dd if=/dev/zero of=/dev/mapper/stripe0 bs=1k

This will produce the output:
dd: writing '/dev/mapper/stripe0': Input/output error 97+0 records in
96+0 records out

And in the kernel log will be:
attempt to access beyond end of device
dm-0: rw=0, want=104, limit=100

The patch below will check that the table size is a multiple of the stripe chunk-size when the table is created, which will prevent the above striped device from being created.

This should not affect tools like LVM or EVMS, since in all the cases I can think of, striped devices are always created with the sizes being a multiple of the chunk-size.

The size of a stripe device must be a multiple of its chunk-size.

 dm-stripe.c |    6 ++++++
 1 file changed, 6 insertions(+)
drivers/md/dm-stripe.c +7 -1
Index: linux-2.6.16-rc5/drivers/md/dm-stripe.c
===================================================================
@@ -103,9 +103,15 @@ static int stripe_ctr(struct dm_target *
 		return -EINVAL;
 	}
 
+	if (((uint32_t)ti->len) & (chunk_size - 1)) {
+		ti->error = "dm-stripe: Target length not divisible by "
+		    "chunk size";
+		return -EINVAL;
+	}
+
 	width = ti->len;
 	if (sector_div(width, stripes)) {
-		ti->error = "dm-stripe: Target length not divisable by "
+		ti->error = "dm-stripe: Target length not divisible by "
 		    "number of stripes";
 		return -EINVAL;
 	}
#1 lore raw

This patch broke my system. I am booting from a hardware fakeraid consisting of a raid-0 between two WD 36 gig 10,000 rpm raptors on a VIA sata raid controller. The dmraid utility previously would correctly configure dm to access the device ( though I did see some of the IO errors you mention in my logs ) and now dmraid fails to configure the dm table because this patch rejects it. My working dmraid table follows:

via_hfciifae: 0 144607678 striped 2 128 8:0 0 8:16 0

I believe the correct thing to do is to special case the last stripe in the volume like such:

0-31 64-67
32-63 68-71

Quoted 52 lines · Alasdair G Kergon

From: Kevin Corry <[email protected]>

The dm-stripe target currently does not enforce that the size of a stripe device be a multiple of the chunk-size. Under certain conditions, this can lead to I/O requests going off the end of an underlying device. This test-case shows one example.

echo "0 100 linear /dev/hdb1 0" | dmsetup create linear0
echo "0 100 linear /dev/hdb1 100" | dmsetup create linear1 echo "0 200 striped 2 32 /dev/mapper/linear0 0 /dev/mapper/linear1 0" | \
dmsetup create stripe0
dd if=/dev/zero of=/dev/mapper/stripe0 bs=1k

This will produce the output:
dd: writing '/dev/mapper/stripe0': Input/output error 97+0 records in
96+0 records out

And in the kernel log will be:
attempt to access beyond end of device
dm-0: rw=0, want=104, limit=100

The patch below will check that the table size is a multiple of the stripe chunk-size when the table is created, which will prevent the above striped device from being created.

This should not affect tools like LVM or EVMS, since in all the cases I can think of, striped devices are always created with the sizes being a multiple of the chunk-size.

The size of a stripe device must be a multiple of its chunk-size.

 dm-stripe.c |    6 ++++++
 1 file changed, 6 insertions(+)

Index: linux-2.6.16-rc5/drivers/md/dm-stripe.c
===================================================================

drivers/md/dm-stripe.c +7 -1
@@ -103,9 +103,15 @@ static int stripe_ctr(struct dm_target *
 		return -EINVAL;
 	}
 
+	if (((uint32_t)ti->len) & (chunk_size - 1)) {
+		ti->error = "dm-stripe: Target length not divisible by "
+		    "chunk size";
+		return -EINVAL;
+	}
+
 	width = ti->len;
 	if (sector_div(width, stripes)) {
-		ti->error = "dm-stripe: Target length not divisable by "
+		ti->error = "dm-stripe: Target length not divisible by "
 		    "number of stripes";
 		return -EINVAL;
 	}
author maintainer
#2 lore raw
Quoted 6 lines · Phillip Susi

now dmraid fails to configure the dm
table because this patch rejects it.

I believe the correct thing to do is to special case the last stripe in
0-31 64-67
32-63 68-71

AFAIK current versions of dmraid handle this correctly - Heinz?

Alasdair

Signature · 1 line
[email protected]
#3 lore raw

Define 'handle it correctly'? The correct thing to do is for dm to accept the values and work properly. dmraid only passes the parameters to dm from the on disk data structure created by the bios.

Quoted 9 lines · Alasdair G Kergon
Quoted 4 lines · Phillip Susi

now dmraid fails to configure the dm table because this patch rejects it.

I believe the correct thing to do is to special case the last stripe in 0-31 64-67
32-63 68-71

AFAIK current versions of dmraid handle this correctly - Heinz?

Alasdair

author maintainer
#4 lore raw
Quoted 2 lines · Phillip Susi

The correct thing to do is for dm to
accept the values and work properly.

But there's ambiguity: should dm truncate just the last stripe(s) or all stripes equally, or use some other combination of truncation? The answer depends on the circumstances, and so it is for userspace, which should know which of those is required, to resolve the matter by supplying appropriate chunk sizes.

Alasdair

Signature · 1 line
[email protected]
#5 lore raw

So you are saying that dmraid should build 3 tables: 1 for the bulk of the array, 1 for only the last stripe, and 1 linear to connect them?

I'm not sure where the ambiguity is. Obviously all the stripes prior to the last should behave normally, the only problem comes from the last stripe. How else could you map the last stripe other than laying down x sectors onto y drives as x / y sectors on each drive in sequence?

Quoted 7 lines · Alasdair G Kergon

But there's ambiguity: should dm truncate just the last stripe(s) or all stripes equally, or use some other combination of truncation? The answer depends on the circumstances, and so it is for userspace, which should know which of those is required, to resolve the matter by supplying appropriate chunk sizes.

Alasdair

author maintainer
#6 lore raw
Quoted 2 lines · Phillip Susi

So you are saying that dmraid should build 3 tables: 1 for the bulk of the array, 1 for only the last stripe, and 1 linear to connect them?

No. 1 table. 2 consecutive targets with different stripe sizes, if that's how the data is actually laid out.

Quoted 3 lines

the only problem comes from the last
stripe. How else could you map the last stripe other than laying down x sectors onto y drives as x / y sectors on each drive in sequence?

Depends whether or not you give precedence to the stripe size. The underlying device might be much larger - dm doesn't know or care - and the intention of userspace might have been to truncate a larger striped device part-way through one of the stripes - an equally reasonable thing to do.

Alasdair

Signature · 1 line
[email protected]
#7 lore raw
Quoted 6 lines · Alasdair G Kergon
Quoted 1 line · Phillip Susi

So you are saying that dmraid should build 3 tables: 1 for the bulk of the array, 1 for only the last stripe, and 1 linear to connect them?

No. 1 table. 2 consecutive targets with different stripe sizes, if that's how the data is actually laid out.

One stripe table can only contain one stripe size, so to have two would require two tables, and a third table to tie them back together.

Quoted 7 lines
Quoted 1 line

the only problem comes from the last stripe. How else could you map the last stripe other than laying down x sectors onto y drives as x / y sectors on each drive in sequence?

Depends whether or not you give precedence to the stripe size. The underlying device might be much larger - dm doesn't know or care - and the intention of userspace might have been to truncate a larger striped device part-way through one of the stripes - an equally reasonable thing to do.

The entire idea of a stripe is that you are using multiple identical drives ( or partitions ), so it doesn't make any sense to be able to truncate one of the drives. In any case, this is not something you can do now, so the fact that you could not do it then either does not seem to be a good argument against allowing partial tails.

author maintainer
#8 lore raw
Quoted 2 lines · Phillip Susi

One stripe table can only contain one stripe size, so to have two would require two tables, and a third table to tie them back together.

One device-mapper table can contain two concatenated stripe targets.

http://people.redhat.com/agk/talks/FOSDEM_2005/text6.html

Quoted 3 lines

The entire idea of a stripe is that you are using multiple identical drives ( or partitions ), so it doesn't make any sense to be able to truncate one of the drives.

dmraid is not the only user of device-mapper striping. Userspace volume managers may want to use all sorts of odd layouts quite legitimately.

Quoted 2 lines

In any case, this is not something you can
do now,

[Actually that's what the code did before this patch:-(]

Quoted 2 lines

so the fact that you could not do it then either does not seem to be a good argument against allowing partial tails.

The arguments are (1) to avoid the ambiguity I've discussed and (2) to avoid the additional complexity the in-kernel striped target would require (calculating a stripe size for the end of the device to override the one supplied), when it's so simple for userspace to specify exactly what it requires by using two targets.

Alasdair

Signature · 1 line
[email protected]
#9 lore raw
In reply to Alasdair G Kergon
Quoted 9 lines · Alasdair G Kergon
Quoted 6 lines · Phillip Susi

now dmraid fails to configure the dm
table because this patch rejects it.

I believe the correct thing to do is to special case the last stripe in
0-31 64-67
32-63 68-71

AFAIK current versions of dmraid handle this correctly - Heinz?

Yes, that's correct.

Quoted 4 lines

Alasdair

Signature · 1 line
[email protected]
Signature · 15 lines
Regards,
Heinz    -- The LVM Guy --

*** Software bugs are stupid.
    Nevertheless it needs not so stupid people to solve them ***

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Red Hat GmbH
Consulting Development Engineer                   Am Sonnenhang 11
Storage Development                               56242 Marienrachdorf
                                                  Germany
[email protected]                            PHONE +49  171 7803392
                                                  FAX   +49 2626 924446
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
#10 lore raw

Turns out my raid volume's size according to the metadata on disk IS an even multiple of the stripe factor, so it looks like this is a bug in dmraid, and I have taken the issue to the ataraid mailing list.

Quoted 15 lines · Heinz Mauelshagen
Quoted 7 lines · Alasdair G Kergon
Quoted 4 lines · Phillip Susi

now dmraid fails to configure the dm table because this patch rejects it.

I believe the correct thing to do is to special case the last stripe in 0-31 64-67
32-63 68-71

AFAIK current versions of dmraid handle this correctly - Heinz?

Yes, that's correct.

Quoted 3 lines

Alasdair

Signature · 1 line
[email protected]