2022-10-17 07:10:27

by Gaosheng Cui

Subject: [PATCH] copy_process(): fix a memleak in copy_process()

If CLONE_PIDFD is set in clone_flags, pidfile will hold the reference
count of pid by getpid(pid), In the error path bad_fork_put_pidfd, the
reference of pid needs to be released, otherwise there will be a
memleak issue, fix it.

unreferenced object 0xffff888164aed400 (size 224):
comm "sh", pid 75274, jiffies 4295717290 (age 2955.536s)
hex dump (first 32 bytes):
01 00 00 00 00 00 00 00 00 00 00 00 ad 4e ad de .............N..
ff ff ff ff 00 00 00 00 ff ff ff ff ff ff ff ff ................
backtrace:
[<00000000bcb9eebb>] kmem_cache_alloc+0x16a/0x7f0
[<00000000340cf9ad>] alloc_pid+0xc5/0xce0
[<000000002387362c>] copy_process+0x29ef/0x6c90
[<00000000bf7d7efc>] kernel_clone+0xd9/0xc70
[<0000000047b1a04f>] __do_sys_clone+0xe1/0x120
[<0000000000f1aa25>] __x64_sys_clone+0xc3/0x150
[<00000000250a19f1>] do_syscall_64+0x5c/0x90
[<000000007e0ac417>] entry_SYSCALL_64_after_hwframe+0x63/0xcd

Fixes: 6fd2fe494b17 ("copy_process(): don't use ksys_close() on cleanups")
Signed-off-by: Gaosheng Cui <[email protected]>
---
kernel/fork.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/kernel/fork.c b/kernel/fork.c
index 08969f5aa38d..8706c06be8af 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -2499,6 +2499,7 @@ static __latent_entropy struct task_struct *copy_process(
cgroup_cancel_fork(p, args);
bad_fork_put_pidfd:
if (clone_flags & CLONE_PIDFD) {
+ put_pid(pid);
fput(pidfile);
put_unused_fd(pidfd);
}
--
2.25.1

2022-10-17 07:29:19

by Christian Brauner

Subject: Re: [PATCH] copy_process(): fix a memleak in copy_process()

On Mon, Oct 17, 2022 at 02:34:06PM +0800, Gaosheng Cui wrote:
> If CLONE_PIDFD is set in clone_flags, pidfile will hold the reference
> count of pid by getpid(pid), In the error path bad_fork_put_pidfd, the
> reference of pid needs to be released, otherwise there will be a
> memleak issue, fix it.
>
> unreferenced object 0xffff888164aed400 (size 224):
> comm "sh", pid 75274, jiffies 4295717290 (age 2955.536s)
> hex dump (first 32 bytes):
> 01 00 00 00 00 00 00 00 00 00 00 00 ad 4e ad de .............N..
> ff ff ff ff 00 00 00 00 ff ff ff ff ff ff ff ff ................
> backtrace:
> [<00000000bcb9eebb>] kmem_cache_alloc+0x16a/0x7f0
> [<00000000340cf9ad>] alloc_pid+0xc5/0xce0
> [<000000002387362c>] copy_process+0x29ef/0x6c90
> [<00000000bf7d7efc>] kernel_clone+0xd9/0xc70
> [<0000000047b1a04f>] __do_sys_clone+0xe1/0x120
> [<0000000000f1aa25>] __x64_sys_clone+0xc3/0x150
> [<00000000250a19f1>] do_syscall_64+0x5c/0x90
> [<000000007e0ac417>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
> Fixes: 6fd2fe494b17 ("copy_process(): don't use ksys_close() on cleanups")
> Signed-off-by: Gaosheng Cui <[email protected]>
> ---
> kernel/fork.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/kernel/fork.c b/kernel/fork.c
> index 08969f5aa38d..8706c06be8af 100644
> --- a/kernel/fork.c
> +++ b/kernel/fork.c
> @@ -2499,6 +2499,7 @@ static __latent_entropy struct task_struct *copy_process(
> cgroup_cancel_fork(p, args);
> bad_fork_put_pidfd:
> if (clone_flags & CLONE_PIDFD) {
> + put_pid(pid);

This will be released during __fput() when ->release() ==
pidfd_release() is called so calling put_pid() here will result in UAF
later on. What syzbot instance does this report come from?

2022-10-18 02:48:32

by Gaosheng Cui

Subject: Re: [PATCH] copy_process(): fix a memleak in copy_process()

> This will be released during __fput() when ->release() ==
> pidfd_release() is called so calling put_pid() here will result in UAF
> later on. What syzbot instance does this report come from?
> .

Thanks for taking time to review the patch, kmemleak report this when
I'm doing some kernel tests, unfortunately, I didn't realize that
pidfd_release would release pid. Perhaps there is another reason for this
issue, but I'm not sure now, I will try to prove it.

By the way, do you have any ideas about it? such as "->release() == NULL"?

Thanks very much!

On 2022/10/17 15:18, Christian Brauner wrote:
> On Mon, Oct 17, 2022 at 02:34:06PM +0800, Gaosheng Cui wrote:
>> If CLONE_PIDFD is set in clone_flags, pidfile will hold the reference
>> count of pid by getpid(pid), In the error path bad_fork_put_pidfd, the
>> reference of pid needs to be released, otherwise there will be a
>> memleak issue, fix it.
>>
>> unreferenced object 0xffff888164aed400 (size 224):
>> comm "sh", pid 75274, jiffies 4295717290 (age 2955.536s)
>> hex dump (first 32 bytes):
>> 01 00 00 00 00 00 00 00 00 00 00 00 ad 4e ad de .............N..
>> ff ff ff ff 00 00 00 00 ff ff ff ff ff ff ff ff ................
>> backtrace:
>> [<00000000bcb9eebb>] kmem_cache_alloc+0x16a/0x7f0
>> [<00000000340cf9ad>] alloc_pid+0xc5/0xce0
>> [<000000002387362c>] copy_process+0x29ef/0x6c90
>> [<00000000bf7d7efc>] kernel_clone+0xd9/0xc70
>> [<0000000047b1a04f>] __do_sys_clone+0xe1/0x120
>> [<0000000000f1aa25>] __x64_sys_clone+0xc3/0x150
>> [<00000000250a19f1>] do_syscall_64+0x5c/0x90
>> [<000000007e0ac417>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>
>> Fixes: 6fd2fe494b17 ("copy_process(): don't use ksys_close() on cleanups")
>> Signed-off-by: Gaosheng Cui <[email protected]>
>> ---
>> kernel/fork.c | 1 +
>> 1 file changed, 1 insertion(+)
>>
>> diff --git a/kernel/fork.c b/kernel/fork.c
>> index 08969f5aa38d..8706c06be8af 100644
>> --- a/kernel/fork.c
>> +++ b/kernel/fork.c
>> @@ -2499,6 +2499,7 @@ static __latent_entropy struct task_struct *copy_process(
>> cgroup_cancel_fork(p, args);
>> bad_fork_put_pidfd:
>> if (clone_flags & CLONE_PIDFD) {
>> + put_pid(pid);
> This will be released during __fput() when ->release() ==
> pidfd_release() is called so calling put_pid() here will result in UAF
> later on. What syzbot instance does this report come from?
> .