Skip to contentSkip to conversation

dmsetup table output changed from 2.6.18 to 2.6.19-rc3 and breaks yaird.

BUG Stefan Schmidt 5 participants 7 messages Oct 30, 2006 – Oct 31, 2006 Device mapper

lore
Share & tools
lore
#0 lore raw

Hello.

Today the build of an initrd from a 2.6.19-rc3 kernel failed.

fairlight:/boot# yaird -o /tmp/foo.initrd 2.6.19-rc3-fairlight
yaird error: Could not read output for /sbin/modprobe -v -n --show-depends --set-version 2.6.19-rc3-fairlight cbc(aes) (fatal)

dmsetup table has changed the report for luks cipher.

dmsetup table on 2.6.18 reports: aes-cbc-essiv:sha256

dmsetup table on 2.6.19-rc3 reports: cbc(aes)-cbc-essiv:sha256

The problem seems to be on the kernel side here. Herbert Xu changed the output with d1806f6a97a536b043fe50e6d8a25b061755cf50

http://kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d1806f6a97a536b043fe50e6d8a25b061755cf50

The question is if this change was intentional and yaird should be fixed, or it's a kernel API breakage.

regards
Stefan Schmidt

#1 lore raw
Quoted 9 lines · Stefan Schmidt

dmsetup table on 2.6.18 reports: aes-cbc-essiv:sha256 dmsetup table on 2.6.19-rc3 reports: cbc(aes)-cbc-essiv:sha256

The problem seems to be on the kernel side here. Herbert Xu changed the output with d1806f6a97a536b043fe50e6d8a25b061755cf50
http://kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=d1806f6a97a536b043fe50e6d8a25b061755cf50

The question is if this change was intentional and yaird should be fixed, or it's a kernel API breakage.

It cannot have been intentional as there was no mention of the change to the userspace interface in the git changelog (and the interface version number was not changed).

A new patch is needed to revert the part of the patch that changed the userspace interface.

Please don't forget to copy in the appropriate maintainers when you send messages like this one:
http://marc.theaimsgroup.com/?l=linux-netdev&m=115547174417490&w=2
so they can provide acks:-)

Alasdair

Signature · 1 line
[email protected]
#2 lore raw
Quoted 5 lines · Alasdair G Kergon

It cannot have been intentional as there was no mention of the change to the userspace interface in the git changelog (and the interface version number was not changed).

Quoted 7 lines

A new patch is needed to revert the part of the patch that changed the userspace interface.

Please don't forget to copy in the appropriate maintainers when you send messages like this one:
http://marc.theaimsgroup.com/?l=linux-netdev&m=115547174417490&w=2
so they can provide acks:-)

Yeah.

Herbert, the breakage _seems_ to be due to the STATUSTYPE_TABLE case change:

- cipher = crypto_tfm_alg_name(cc->tfm);
+ cipher = crypto_blkcipher_name(cc->tfm);

which effectively changes "aes" into "cbc(aes)", which is wrong, since we show the chainmode separately.

Please, somebody who knows this area, send me a fix,

Linus

----
(maybe something like this trivial one? Totally untested, but it would seem to be the sane approach)

drivers/md/dm-crypt.c +1 -2
index a625576..645e3ce 100644
@@ -925,8 +925,7 @@ static int crypt_status(struct dm_target
 		break;
 
 	case STATUSTYPE_TABLE:
-		cipher = crypto_blkcipher_name(cc->tfm);
-
+		cipher = cc->cipher;
 		chainmode = cc->chainmode;
 
 		if (cc->iv_mode)
#3 lore raw
Quoted 10 lines · Linus Torvalds

(maybe something like this trivial one? Totally untested, but it would seem to be the sane approach)

drivers/md/dm-crypt.c +0 -0
index a625576..645e3ce 100644

drivers/md/dm-crypt.c static int crypt_status(struct dm_target · line 931

Quoted 9 lines
drivers/md/dm-crypt.c +1 -2
@@ -925,8 +925,7 @@ static int crypt_status(struct dm_target
 		break;
 
 	case STATUSTYPE_TABLE:
-		cipher = crypto_blkcipher_name(cc->tfm);
-
+		cipher = cc->cipher;
 		chainmode = cc->chainmode;
 
 		if (cc->iv_mode)

Looks correct.

The point of STATUSTYPE_TABLE is to return (readable) output to userspace in a format that the crypt_ctr() function would accept back in.

So crypt_ctr() now stores a private copy of cipher and chainmode for crypt_status() to regurgitate when requested.

Alasdair

Signature · 1 line
[email protected]
#4 lore raw
In reply to Linus Torvalds

Am Montag, den 30.10.2006, 11:00 -0800 schrieb Linus Torvalds:

Quoted 22 lines · Alasdair G Kergon
Quoted 5 lines

It cannot have been intentional as there was no mention of the change to the userspace interface in the git changelog (and the interface version number was not changed).

Quoted 7 lines

A new patch is needed to revert the part of the patch that changed the userspace interface.

Please don't forget to copy in the appropriate maintainers when you send messages like this one:
http://marc.theaimsgroup.com/?l=linux-netdev&m=115547174417490&w=2
so they can provide acks:-)

Yeah.

Herbert, the breakage _seems_ to be due to the STATUSTYPE_TABLE case change:

(quoted diff) +1 -1
-		cipher = crypto_tfm_alg_name(cc->tfm);
+		cipher = crypto_blkcipher_name(cc->tfm);
Quoted 7 lines

which effectively changes "aes" into "cbc(aes)", which is wrong, since we show the chainmode separately.

Please, somebody who knows this area, send me a fix,

(maybe something like this trivial one? Totally untested, but it would seem to be the sane approach)

Yes, this works just fine. It can also be cleaned up a little further as the temporary variables are unnecessary at that point.

----

Fix dm-crypt after the block cipher API changes to correctly return the backwards compatible cipher-chainmode[-ivmode] format for "dmsetup table".

drivers/md/dm-crypt.c +3 -8
diff linux-2.6.19-rc3.orig/drivers/md/dm-crypt.c linux-2.6.19-rc3/drivers/md/dm-crypt.c
@@ -915,8 +915,6 @@ static int crypt_status(struct dm_target
 			char *result, unsigned int maxlen)
 {
 	struct crypt_config *cc = (struct crypt_config *) ti->private;
-	const char *cipher;
-	const char *chainmode = NULL;
 	unsigned int sz = 0;
 
 	switch (type) {
@@ -925,14 +923,11 @@ static int crypt_status(struct dm_target
 		break;
 
 	case STATUSTYPE_TABLE:
-		cipher = crypto_blkcipher_name(cc->tfm);
-
-		chainmode = cc->chainmode;
-
 		if (cc->iv_mode)
-			DMEMIT("%s-%s-%s ", cipher, chainmode, cc->iv_mode);
+			DMEMIT("%s-%s-%s ", cc->cipher, cc->chainmode,
+			       cc->iv_mode);
 		else
-			DMEMIT("%s-%s ", cipher, chainmode);
+			DMEMIT("%s-%s ", cc->cipher, cc->chainmode);
 
 		if (cc->key_size > 0) {
 			if ((maxlen - sz) < ((cc->key_size << 1) + 1))
Acked-by
#5 lore raw
Quoted 9 lines · Christophe Saout
Quoted 9 lines

Herbert, the breakage _seems_ to be due to the STATUSTYPE_TABLE case change:

(quoted diff) +1 -1
-		cipher = crypto_tfm_alg_name(cc->tfm);
+		cipher = crypto_blkcipher_name(cc->tfm);

which effectively changes "aes" into "cbc(aes)", which is wrong, since we show the chainmode separately.

Yes that's wrong.

However, the bigger problem is that dmcrypt's algorithm specification does not allow all algorithms to be specified. In particular, algorithms with dashes in their names cannot be represented in an unambiguous way.

The separation of chain modes and algorithm names is in fact unnecessary and only complicates matters.

For now it's no big deal since the algorithms people actually use can be represented. But it would be nice to get this fixed.

Quoted 7 lines

----

Fix dm-crypt after the block cipher API changes to correctly return the backwards compatible cipher-chainmode[-ivmode] format for "dmsetup table".

Looks good to me.

Thanks,

Signature · 4 lines
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#6 lore raw
In reply to Linus Torvalds

Hello.

Quoted 3 lines · Linus Torvalds

(maybe something like this trivial one? Totally untested, but it would seem to be the sane approach)

I just tested it. Works fine for me.

regards
Stefan Schmidt