Re: CVE-2024-41001: io_uring/sqpoll: work around a potential audit memory leak

Re: CVE-2024-41001: io_uring/sqpoll: work around a potential audit memory leak

[Wang Zhaolong]

Hello,

I was confused when reviewing the fix for CVE-2024-41001.
To better understand the issue and the proposed solution, I would
greatly appreciate your help in clarifying the following points:

1. What was the original patch that introduced this issue (any Fixes tag)?
2. Is the leaking variable member the "context->sockaddr"?
3. Could you shed some light on how the reference to the leaked memory is
    lost during the transition from the prep phase to the issue phase?
4. The fix introduces a NOP operation "before the SQPOLL does anything."
    How does this addition of a NOP operation prevent the memory leak from
    occurring?

Thank you in advance for taking the time to address my questions. Your
insights will help me better understand this fix.

Best regards,
Wang Zhaolong

> Description
> ===========
> 
> In the Linux kernel, the following vulnerability has been resolved:
> 
> io_uring/sqpoll: work around a potential audit memory leak
> 
> kmemleak complains that there's a memory leak related to connect
> handling:
> 
> unreferenced object 0xffff0001093bdf00 (size 128):
> comm "iou-sqp-455", pid 457, jiffies 4294894164
> hex dump (first 32 bytes):
> 02 00 fa ea 7f 00 00 01 00 00 00 00 00 00 00 00  ................
> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
> backtrace (crc 2e481b1a):
> [<00000000c0a26af4>] kmemleak_alloc+0x30/0x38
> [<000000009c30bb45>] kmalloc_trace+0x228/0x358
> [<000000009da9d39f>] __audit_sockaddr+0xd0/0x138
> [<0000000089a93e34>] move_addr_to_kernel+0x1a0/0x1f8
> [<000000000b4e80e6>] io_connect_prep+0x1ec/0x2d4
> [<00000000abfbcd99>] io_submit_sqes+0x588/0x1e48
> [<00000000e7c25e07>] io_sq_thread+0x8a4/0x10e4
> [<00000000d999b491>] ret_from_fork+0x10/0x20
> 
> which can can happen if:
> 
> 1) The command type does something on the prep side that triggers an
>     audit call.
> 2) The thread hasn't done any operations before this that triggered
>     an audit call inside ->issue(), where we have audit_uring_entry()
>     and audit_uring_exit().
> 
> Work around this by issuing a blanket NOP operation before the SQPOLL
> does anything.
> 
> The Linux kernel CVE team has assigned CVE-2024-41001 to this issue.

Re: CVE-2024-41001: io_uring/sqpoll: work around a potential audit memory leak

[Wang Zhaolong]

Hello,

I think a possible reason for the leak scenario is:

When `audit_context->dummy` is 0. __audit_sockaddr() allocates sockaddr.

In the below process, audit_reset_context() return early. ctx->sockaddr
is not released.

   io_issue_sqe
     audit_uring_entry
       __audit_uring_entry
         ctx->dummy -- set dummy as non-zero
     def->issue()
     audit_uring_exit
       __audit_uring_exit
         audit_reset_context

static void audit_reset_context(struct audit_context *ctx)
{
     ......
     /* if ctx is non-null, reset the "ctx->context" regardless */
     ctx->context = AUDIT_CTX_UNUSED;
     if (ctx->dummy)
         return;

     ......
     kfree(ctx->sockaddr);
     ......
}

The `audit_uring_entry(IORING_OP_NOP);` statement initializes the 'dummy' once at the
beginning to ensure that ctx->sockaddr is allocated and deallocated in pairs later
in the process.

According to the above analysis, I think the fixes tag should be
5bd2182d58e9 ("audit,io_uring,io-wq: add some basic audit support to io_uring")
Is my understanding correct?

I look forward to hearing back.

Best regards,
Wang Zhaolong

> Hello,
> 
> I was confused when reviewing the fix for CVE-2024-41001.
> To better understand the issue and the proposed solution, I would
> greatly appreciate your help in clarifying the following points:
> 
> 1. What was the original patch that introduced this issue (any Fixes tag)?
> 2. Is the leaking variable member the "context->sockaddr"?
> 3. Could you shed some light on how the reference to the leaked memory is
>     lost during the transition from the prep phase to the issue phase?
> 4. The fix introduces a NOP operation "before the SQPOLL does anything."
>     How does this addition of a NOP operation prevent the memory leak from
>     occurring?
> 
> Thank you in advance for taking the time to address my questions. Your
> insights will help me better understand this fix.
> 
> Best regards,
> Wang Zhaolong
> 
>> Description
>> ===========
>>
>> In the Linux kernel, the following vulnerability has been resolved:
>>
>> io_uring/sqpoll: work around a potential audit memory leak
>>
>> kmemleak complains that there's a memory leak related to connect
>> handling:
>>
>> unreferenced object 0xffff0001093bdf00 (size 128):
>> comm "iou-sqp-455", pid 457, jiffies 4294894164
>> hex dump (first 32 bytes):
>> 02 00 fa ea 7f 00 00 01 00 00 00 00 00 00 00 00  ................
>> 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
>> backtrace (crc 2e481b1a):
>> [<00000000c0a26af4>] kmemleak_alloc+0x30/0x38
>> [<000000009c30bb45>] kmalloc_trace+0x228/0x358
>> [<000000009da9d39f>] __audit_sockaddr+0xd0/0x138
>> [<0000000089a93e34>] move_addr_to_kernel+0x1a0/0x1f8
>> [<000000000b4e80e6>] io_connect_prep+0x1ec/0x2d4
>> [<00000000abfbcd99>] io_submit_sqes+0x588/0x1e48
>> [<00000000e7c25e07>] io_sq_thread+0x8a4/0x10e4
>> [<00000000d999b491>] ret_from_fork+0x10/0x20
>>
>> which can can happen if:
>>
>> 1) The command type does something on the prep side that triggers an
>>     audit call.
>> 2) The thread hasn't done any operations before this that triggered
>>     an audit call inside ->issue(), where we have audit_uring_entry()
>>     and audit_uring_exit().
>>
>> Work around this by issuing a blanket NOP operation before the SQPOLL
>> does anything.
>>
>> The Linux kernel CVE team has assigned CVE-2024-41001 to this issue.
> 

Re: CVE-2024-41001: io_uring/sqpoll: work around a potential audit memory leak

[Jens Axboe]

On 7/18/24 8:41 AM, Wang Zhaolong wrote:
> Hello,
> 
> I think a possible reason for the leak scenario is:
> 
> When `audit_context->dummy` is 0. __audit_sockaddr() allocates sockaddr.
> 
> In the below process, audit_reset_context() return early. ctx->sockaddr
> is not released.
> 
>   io_issue_sqe
>     audit_uring_entry
>       __audit_uring_entry
>         ctx->dummy -- set dummy as non-zero
>     def->issue()
>     audit_uring_exit
>       __audit_uring_exit
>         audit_reset_context
> 
> static void audit_reset_context(struct audit_context *ctx)
> {
>     ......
>     /* if ctx is non-null, reset the "ctx->context" regardless */
>     ctx->context = AUDIT_CTX_UNUSED;
>     if (ctx->dummy)
>         return;
> 
>     ......
>     kfree(ctx->sockaddr);
>     ......
> }
> 
> The `audit_uring_entry(IORING_OP_NOP);` statement initializes the 'dummy' once at the
> beginning to ensure that ctx->sockaddr is allocated and deallocated in pairs later
> in the process.
> 
> According to the above analysis, I think the fixes tag should be
> 5bd2182d58e9 ("audit,io_uring,io-wq: add some basic audit support to io_uring")

It was introduced with the changes to the above commit, where you could
end up calling prep (which does the move_addr_to_kernel()) before audit
was ready for it. This is the call trace shown in the commit as well.
Which I _think_ is:

Fixes: f482aa986527 ("audit, io_uring, io-wq: Fix memory leak in io_sq_thread() and io_wqe_worker()")

but I'd have to double check. In any case, it's a leak on the audit
side.

-- 
Jens Axboe