| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <f796c459-cfa0-7643-acd5-ae5638e213dc@alu.unizg.hr>
Date: Tue, 14 Feb 2023 21:54:11 +0100
From: Mirsad Goran Todorovac <mirsad.todorovac@....unizg.hr>
To: Andy Shevchenko <andriy.shevchenko@...el.com>
Cc: linux-gpio@...r.kernel.org,
Linus Walleij <linus.walleij@...aro.org>,
Bartosz Golaszewski <brgl@...ev.pl>,
linux-kernel@...r.kernel.org,
Thorsten Leemhuis <regressions@...mhuis.info>
Subject: Re: INFO: BISECTED: memory leak in gpio device in 6.2-rc6
On 12. 02. 2023. 15:19, Andy Shevchenko wrote:
> On Wed, Feb 08, 2023 at 08:55:24PM +0100, Mirsad Goran Todorovac wrote:
>> On 31. 01. 2023. 10:36, Mirsad Goran Todorovac wrote:
>>> I came across this memory leak apparently in the GPIO device driver.
>>> It is still present in 6.2-rc6 release candidate kernel (just ran kselftest).
>>>
>>> This is a vanilla Torvalds tree kernel with MGLRU and KMEMLEAK (obviously)
>>> enabled.
>>>
>>> If you think this bug is significant, I can attempt the bug bisect in the
>>> environment that triggered it (Lenovo LENOVO_MT_10TX_BU_Lenovo_FM_V530S-07ICB)
>>> with BIOS M22KT49A from 11/10/2022 and AlmaLinux 8.7.
>>>
>>> Here is the /sys/kernel/debug/kmemleak output:
>>>
>>> unreferenced object 0xffff9e67ad71f160 (size 32):
>>> comm "gpio-sim.sh", pid 208926, jiffies 4372229685 (age 2101.564s)
>>> hex dump (first 32 bytes):
>>> 67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
>>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
>>> backtrace:
>>> [<0000000098bf3d1b>] slab_post_alloc_hook+0x91/0x320
>>> [<00000000da3205c5>] __kmem_cache_alloc_node+0x1bf/0x2b0
>>> [<00000000aa51a58a>] __kmalloc_node_track_caller+0x55/0x140
>>> [<00000000bd682ecc>] kvasprintf+0x6b/0xd0
>>> [<00000000a3431d55>] kasprintf+0x4e/0x70
>>> [<00000000f52d2629>] gpio_sim_device_config_live_store+0x401/0x59d [gpio_sim]
>>> [<00000000673fc6df>] configfs_write_iter+0xcc/0x130
>>> [<000000001d5d0829>] vfs_write+0x2b4/0x3d0
>>> [<00000000d2336251>] ksys_write+0x61/0xe0
>>> [<00000000f7015bb1>] __x64_sys_write+0x1a/0x20
>>> [<000000008ac743d2>] do_syscall_64+0x58/0x80
>>> [<000000004d7b7d50>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>> [root@...mtodorov marvin]#
>>
>> The new development on the bug is that it probably requires some superuser privileges
>> or some capability to be exploited, for it requires access to configfs,
>> but it was reproduced on different hardware as well.
>>
>> The minimum reproducing script is attached, with its output log.
>>
>> From the testing "age" of the unreferenced object I assume that it is allocated earlier
>> in a part of script I am unable to locate or specify, but orphaned at the end of the script.
>>
>> root@...me/user/kernel_bugs/gpio-sim# time ./gpio-reproduce-min.sh
>> 2.14. Lines can be hogged
>> Scanning stage 2.14.7 ... done.
>> Sleeping 60 seconds ... done.
>> Stage 2.14.7 clean.
>> Rescanning stage 2.14.7 ... done.
>> Sleeping 60 seconds ... done.
>> unreferenced object 0xffff9593b9d16bc0 (size 32):
>> comm "gpio-reproduce-", pid 7594, jiffies 4295865460 (age 136.184s)
>> hex dump (first 32 bytes):
>> 67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
>> backtrace:
>> [<00000000fe76444b>] __kmem_cache_alloc_node+0x380/0x4e0
>> [<0000000099f63f55>] __kmalloc_node_track_caller+0x55/0x140
>> [<00000000c4efe87f>] kvasprintf+0x6b/0xd0
>> [<000000000c0f91cd>] kasprintf+0x4e/0x70
>> [<000000003434d9b5>] gpio_sim_device_config_live_store+0x401/0x59d [gpio_sim]
>> [<0000000052ce6759>] configfs_write_iter+0xcc/0x130
>> [<0000000006087fd2>] vfs_write+0x2b4/0x3d0
>> [<000000008a17e041>] ksys_write+0x61/0xe0
>> [<000000007bded8ea>] __x64_sys_write+0x1a/0x20
>> [<00000000e1220148>] do_syscall_64+0x58/0x80
>> [<0000000006093069>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>
>> real 2m16.398s
>> user 0m0.023s
>> sys 0m16.360s
>> root@...me/user/kernel_bugs/gpio-sim#
>>
>> The effect is cummulative:
>>
>> root@...vin-IdeaPad-3-15ITL6:/home/user/kernel_bugs/gpio-sim# cat /sys/kernel/debug/kmemleak
>> unreferenced object 0xffff9593b9d16bc0 (size 32):
>> comm "gpio-reproduce-", pid 7594, jiffies 4295865460 (age 520.296s)
>> hex dump (first 32 bytes):
>> 67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
>> backtrace:
>> [<00000000fe76444b>] __kmem_cache_alloc_node+0x380/0x4e0
>> [<0000000099f63f55>] __kmalloc_node_track_caller+0x55/0x140
>> [<00000000c4efe87f>] kvasprintf+0x6b/0xd0
>> [<000000000c0f91cd>] kasprintf+0x4e/0x70
>> [<000000003434d9b5>] gpio_sim_device_config_live_store+0x401/0x59d [gpio_sim]
>> [<0000000052ce6759>] configfs_write_iter+0xcc/0x130
>> [<0000000006087fd2>] vfs_write+0x2b4/0x3d0
>> [<000000008a17e041>] ksys_write+0x61/0xe0
>> [<000000007bded8ea>] __x64_sys_write+0x1a/0x20
>> [<00000000e1220148>] do_syscall_64+0x58/0x80
>> [<0000000006093069>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>> unreferenced object 0xffff95938918fb40 (size 32):
>> comm "gpio-reproduce-", pid 7675, jiffies 4295954327 (age 164.832s)
>> hex dump (first 32 bytes):
>> 67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
>> backtrace:
>> [<00000000fe76444b>] __kmem_cache_alloc_node+0x380/0x4e0
>> [<0000000099f63f55>] __kmalloc_node_track_caller+0x55/0x140
>> [<00000000c4efe87f>] kvasprintf+0x6b/0xd0
>> [<000000000c0f91cd>] kasprintf+0x4e/0x70
>> [<000000003434d9b5>] gpio_sim_device_config_live_store+0x401/0x59d [gpio_sim]
>> [<0000000052ce6759>] configfs_write_iter+0xcc/0x130
>> [<0000000006087fd2>] vfs_write+0x2b4/0x3d0
>> [<000000008a17e041>] ksys_write+0x61/0xe0
>> [<000000007bded8ea>] __x64_sys_write+0x1a/0x20
>> [<00000000e1220148>] do_syscall_64+0x58/0x80
>> [<0000000006093069>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>> unreferenced object 0xffff9594a3cf1820 (size 32):
>> comm "gpio-reproduce-", pid 7721, jiffies 4295976853 (age 74.728s)
>> hex dump (first 32 bytes):
>> 67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
>> backtrace:
>> [<00000000fe76444b>] __kmem_cache_alloc_node+0x380/0x4e0
>> [<0000000099f63f55>] __kmalloc_node_track_caller+0x55/0x140
>> [<00000000c4efe87f>] kvasprintf+0x6b/0xd0
>> [<000000000c0f91cd>] kasprintf+0x4e/0x70
>> [<000000003434d9b5>] gpio_sim_device_config_live_store+0x401/0x59d [gpio_sim]
>> [<0000000052ce6759>] configfs_write_iter+0xcc/0x130
>> [<0000000006087fd2>] vfs_write+0x2b4/0x3d0
>> [<000000008a17e041>] ksys_write+0x61/0xe0
>> [<000000007bded8ea>] __x64_sys_write+0x1a/0x20
>> [<00000000e1220148>] do_syscall_64+0x58/0x80
>> [<0000000006093069>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>> root@...vin-IdeaPad-3-15ITL6:/home/user/kernel_bugs/gpio-sim#
>>
>> With sufficient privileges (or possibly even without them), an exploit can be devised
>> to automate allocation of orphaned objects, at a rate of a couple per second (PoC attached).
>
> Looking into the code I found one inconsistency. It might be that that brings
> an issue, dunno. Can you try the patch below if it helps (but TBH I'm a bit
> sceptical)?
>
>>>From 499cfb52aa7de67a8bbb56ce183d9528b2376db0 Mon Sep 17 00:00:00 2001
> From: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>
> Date: Sun, 12 Feb 2023 16:15:00 +0200
> Subject: [PATCH 1/1] gpio: sim: Deactivate device in reversed order
>
> Run the steps to deactivate device in the reserved order to what
> it has been done in gpio_sim_device_activate_unlocked().
>
> Signed-off-by: Andy Shevchenko <andriy.shevchenko@...ux.intel.com>
> ---
> drivers/gpio/gpio-sim.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/gpio/gpio-sim.c b/drivers/gpio/gpio-sim.c
> index a51b5ea38ad5..b0111d18808c 100644
> --- a/drivers/gpio/gpio-sim.c
> +++ b/drivers/gpio/gpio-sim.c
> @@ -953,9 +953,9 @@ static void gpio_sim_device_deactivate_unlocked(struct gpio_sim_device *dev)
>
> swnode = dev_fwnode(&dev->pdev->dev);
> platform_device_unregister(dev->pdev);
> + gpio_sim_remove_hogs(dev);
> gpio_sim_remove_swnode_recursive(swnode);
> dev->pdev = NULL;
> - gpio_sim_remove_hogs(dev);
> }
>
> static ssize_t
P.S.
Hi, Andy,
Lucky or unlucky, bisect was rather quick: the gpio-sim module seems to appear in
the 5.17-rc1 kernel, and it seems that this first version is already "bad", that is,
leaking memory.
I have managed to automate the exploit to create cca. 330,000 leaks, but the kernel
did not crash yet, though much of the memory was used. But the leak is rather small,
32 bytes + overhead.
Here is the result of the run of tools/testing/selftests/gpio/gpio-sim.sh:
root@...vin-IdeaPad-3-15ITL6:/home/marvin/linux/kernel/linux_torvalds/tools/testing/selftests/gpio# cat /sys/kernel/debug/kmemleak
unreferenced object 0xffff9ed2ca929f40 (size 32):
comm "gpio-sim.sh", pid 15074, jiffies 4294959639 (age 25.848s)
hex dump (first 32 bytes):
67 70 69 6f 2d 73 69 6d 2e 30 2d 6e 6f 64 65 30 gpio-sim.0-node0
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<000000008ae1eb47>] slab_post_alloc_hook+0x80/0x2a0
[<000000000d18725c>] __kmalloc_track_caller+0x118/0x290
[<0000000031cbee97>] kvasprintf+0x66/0xd0
[<00000000f315aa2a>] kasprintf+0x4e/0x70
[<00000000a996121f>] gpio_sim_device_config_live_store+0x372/0x578 [gpio_sim]
[<00000000fd890a9e>] configfs_write_iter+0xc8/0x130
[<00000000e3ef7485>] new_sync_write+0x117/0x1b0
[<000000006dc1f988>] vfs_write+0x1aa/0x290
[<0000000027668ff9>] ksys_write+0x61/0xe0
[<00000000ac7bd963>] __x64_sys_write+0x1a/0x20
[<00000000ed9d842c>] do_syscall_64+0x58/0x80
[<00000000c81cc832>] entry_SYSCALL_64_after_hwframe+0x44/0xae
root@...vin-IdeaPad-3-15ITL6:/home/marvin/linux/kernel/linux_torvalds/tools/testing/selftests/gpio# uname -rms
Linux 5.17.0-rc1-mglru-kmlk-andy x86_64
root@...vin-IdeaPad-3-15ITL6:/home/marvin/linux/kernel/linux_torvalds/tools/testing/selftests/gpio#
Basically, this is the similar output to the one with 6.2-rc6 leak.
I hope this can help narrow down the culprit code.
Regards,
Mirsad
--
Mirsad Goran Todorovac
Sistem inženjer
Grafički fakultet | Akademija likovnih umjetnosti
Sveučilište u Zagrebu
System engineer
Faculty of Graphic Arts | Academy of Fine Arts
University of Zagreb, Republic of Croatia
The European Union
Powered by blists - more mailing lists