Skip to contentSkip to conversation

Double free vulnerability in do_rename_gpt_parts

DISCUSSION Jordy 3 participants 5 messages Jan 17, 2020

lore
Share & tools
lore

Jordy

author
#0 lore raw

Hello U-Boot lists!

I think I found a double free bug in U-Boot, in /cmp/gpt.c in the function do_rename_gpt_parts().

On line 702 the partition_list is being free'd if ret is smaller than 0.
If the return value is not -ENOMEM it will go to the out: label and free the partition_list again.

Double freeing may result in a write-what-where condition, allowing an attacker to execute arbitrary code.

My advice would be to not free the parition_list on line 702 as nothing is being done with it there afterwards anyway and leave your clean_up in the out: label :)

Kind Regards,

Jordy Zomer

#1 lore raw

+ Some contributors of this file

Quoted 7 lines · Jordy

Hello U-Boot lists!

I think I found a double free bug in U-Boot, in /cmp/gpt.c in the function do_rename_gpt_parts().

On line 702 the partition_list is being free'd if ret is smaller than 0.
If the return value is not -ENOMEM it will go to the out: label and free the partition_list again.

Reading the code, I can confirm that. Funny enough, the code in question was introduced by commit 18030d04 ("GPT: fix memory leaks identified by Coverity"). Although I think Coverity should have detected the resulting double-free...

However, I'm not sure of the fix: the code just continues for -ENOMEM and then goes on using partitions_list at line 757...

Regards,
Simon

Quoted 8 lines

Double freeing may result in a write-what-where condition, allowing an attacker to execute arbitrary code.

My advice would be to not free the parition_list on line 702 as nothing is being done with it there afterwards anyway and leave your clean_up in the out: label :)

Kind Regards,

Jordy Zomer

#2 lore raw
Quoted 11 lines · Simon Goldschmidt

+ Some contributors of this file

Quoted 7 lines · Jordy

Hello U-Boot lists!

I think I found a double free bug in U-Boot, in /cmp/gpt.c in the function do_rename_gpt_parts().

On line 702 the partition_list is being free'd if ret is smaller than 0.
If the return value is not -ENOMEM it will go to the out: label and free the partition_list again.

Quoted 6 lines

Reading the code, I can confirm that. Funny enough, the code in question was introduced by commit 18030d04 ("GPT: fix memory leaks identified by Coverity"). Although I think Coverity should have detected the resulting double-free...

However, I'm not sure of the fix: the code just continues for -ENOMEM and then goes on using partitions_list at line 757...

So, Coverity later did complain about that change (but not immediately, funny enough). I posted http://patchwork.ozlabs.org/patch/1192036/ and was hoping for a review on it as it's complex enough I'd like to avoid adding a 3rd round of issues there. Thanks!

Signature · 8 lines
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20200117/a3bd537a/attachment.sig>
#3 lore raw

Am 17.01.2020 um 17:31 schrieb Tom Rini:

Quoted 16 lines · Simon Goldschmidt
Quoted 15 lines

+ Some contributors of this file

Quoted 7 lines · Jordy

Hello U-Boot lists!

I think I found a double free bug in U-Boot, in /cmp/gpt.c in the function do_rename_gpt_parts().

On line 702 the partition_list is being free'd if ret is smaller than 0.
If the return value is not -ENOMEM it will go to the out: label and free the partition_list again.

Reading the code, I can confirm that. Funny enough, the code in question was introduced by commit 18030d04 ("GPT: fix memory leaks identified by Coverity"). Although I think Coverity should have detected the resulting double-free...

Quoted 8 lines
Quoted 2 lines

However, I'm not sure of the fix: the code just continues for -ENOMEM and then goes on using partitions_list at line 757...

So, Coverity later did complain about that change (but not immediately, funny enough). I posted http://patchwork.ozlabs.org/patch/1192036/ and was hoping for a review on it as it's complex enough I'd like to avoid adding a 3rd round of issues there. Thanks!

Ah, yes. I've just responded there.

Regards,
Simon

author
#4 lore raw
In reply to Tom Rini

Hey Tom,

The patch looks good to me, I do agree with simon that you'd have to initialize the pointers to 0, because uninitialized pointers could contain garbage, unless it's a global variable (those are automagically initialized to 0).

Btw I really appreciate the fast and friendly responses, I've seen that different at other software packages thumbs-up for that!

Best Regards,

Jordy

Quoted 29 lines · Tom Rini
Quoted 11 lines · Simon Goldschmidt

+ Some contributors of this file

Quoted 7 lines · Jordy

Hello U-Boot lists!

I think I found a double free bug in U-Boot, in /cmp/gpt.c in the function do_rename_gpt_parts().

On line 702 the partition_list is being free'd if ret is smaller than 0.
If the return value is not -ENOMEM it will go to the out: label and free the partition_list again.

Quoted 6 lines

Reading the code, I can confirm that. Funny enough, the code in question was introduced by commit 18030d04 ("GPT: fix memory leaks identified by Coverity"). Although I think Coverity should have detected the resulting double-free...

However, I'm not sure of the fix: the code just continues for -ENOMEM and then goes on using partitions_list at line 757...

So, Coverity later did complain about that change (but not immediately, funny enough). I posted http://patchwork.ozlabs.org/patch/1192036/ and was hoping for a review on it as it's complex enough I'd like to avoid adding a 3rd round of issues there. Thanks!

Signature · 1 line
Tom