Skip to contentSkip to conversation

DLM regression in 64-bit 3.7.x Kernel?

Cluster-devel New Jacek Konieczny 2 participants 3 messages Feb 19, 2013 – Feb 25, 2013 Distributed lock manager (DLM)

lore
Share & tools
lore
#0 lore raw

Hi,

I have recently upgraded my development cluster from 3.6.x to 3.7.1 kernel and clvmd stopped working (all locking operation result in 'Invalid argument'). I have traced the problem to this call:

write(8, "\6\0\0\0\0\0\0\0\1\0\0\0\1\1\0\0\5@\0\0\1\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\320\307\351\326\277\177\0\0\20\17\36\331\277\177\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\313\351\326\277\177\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0Hlk5NeaVF0qhDF20RBq61EZaIj5yyUJgWxxs4ntZedP34Rj3E2wadZfVwtOcvpKb", 184) = -1 EINVAL (Invalid argument)

Kernel sources show only a few conditions which can result in the EINVAL error and the one introduced by the 'dlm: check the maximum size of a request from user' patch seemed suspicious to me.

The commit: http://git.kernel.org/?p=linux/kernel/git/teigland/linux-dlm.git;a=commitdiff;h=2b75bc9121e54e22537207b47b71373bcb0be41c

Should a 64-bit kernel really refuse requests using the 64-bit dlm_write_request struct when compiled with CONFIG_COMPAT on?

To verify my suspicions I have compiled the 3.7.9 with a little patch dropping the '(count > sizeof(struct dlm_write_request32) + DLM_RESNAME_MAXLEN)' check, which solved the problem for me.

The patch will follow, though I am still not sure this is the solution.

Greets,
Jacek

#1 lore raw

PATCH 1/1 dlm: Allow 64-bit request when CONFIG_COMPAT is on

Even if kernel is compiled for 32-bit ABI compatibility it still should able to process 64-bit dlm_write_request struct.
The regression was introduced by:

b75bc91 dlm: check the maximum size of a request from user

---
 fs/dlm/user.c | 4 ----
 1 file changed, 4 deletions(-)
fs/dlm/user.c +0 -4
index 7ff4985..d08a156 100644
@@ -503,11 +503,7 @@ static ssize_t device_write(struct file *file, const char __user *buf,
 #endif
 		return -EINVAL;
 
-#ifdef CONFIG_COMPAT
-	if (count > sizeof(struct dlm_write_request32) + DLM_RESNAME_MAXLEN)
-#else
 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
-#endif
 		return -EINVAL;
 
 	kbuf = kzalloc(count + 1, GFP_NOFS);
Signature · 1 line
1.8.0.1
maintainer
#2 lore raw
In reply to Jacek Konieczny
Quoted 14 lines · Jacek Konieczny

Hi,

I have recently upgraded my development cluster from 3.6.x to 3.7.1 kernel and clvmd stopped working (all locking operation result in 'Invalid argument'). I have traced the problem to this call:

write(8, "\6\0\0\0\0\0\0\0\1\0\0\0\1\1\0\0\5@\0\0\1\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\320\307\351\326\277\177\0\0\20\17\36\331\277\177\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\313\351\326\277\177\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0Hlk5NeaVF0qhDF20RBq61EZaIj5yyUJgWxxs4ntZedP34Rj3E2wadZfVwtOcvpKb", 184) = -1 EINVAL (Invalid argument)

Kernel sources show only a few conditions which can result in the EINVAL error and the one introduced by the 'dlm: check the maximum size of a request from user' patch seemed suspicious to me.

The commit: http://git.kernel.org/?p=linux/kernel/git/teigland/linux-dlm.git;a=commitdiff;h=2b75bc9121e54e22537207b47b71373bcb0be41c

Quoted 8 lines

Should a 64-bit kernel really refuse requests using the 64-bit dlm_write_request struct when compiled with CONFIG_COMPAT on?

To verify my suspicions I have compiled the 3.7.9 with a little patch dropping the '(count > sizeof(struct dlm_write_request32) + DLM_RESNAME_MAXLEN)' check, which solved the problem for me.

The patch will follow, though I am still not sure this is the solution.

Hi, the fix is already upstream: d4b0bcf32b946590afd29e202d6a399b84fe6c67 Dave