Re: [PATCH bpf v3] bpf: Fix memory leaks in __check_func_call

[wangyufen <wangyufen@huawei.com>]

在 2022/11/8 8:46, Martin KaFai Lau 写道:
> On 11/3/22 2:42 AM, Wang Yufen wrote:
>> kmemleak reports this issue:
>>
>> unreferenced object 0xffff88817139d000 (size 2048):
>>    comm "test_progs", pid 33246, jiffies 4307381979 (age 45851.820s)
>>    hex dump (first 32 bytes):
>>      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 00 00 00 00 00 00 00 ................
>>    backtrace:
>>      [<0000000045f075f0>] kmalloc_trace+0x27/0xa0
>>      [<0000000098b7c90a>] __check_func_call+0x316/0x1230
>>      [<00000000b4c3c403>] check_helper_call+0x172e/0x4700
>>      [<00000000aa3875b7>] do_check+0x21d8/0x45e0
>>      [<000000001147357b>] do_check_common+0x767/0xaf0
>>      [<00000000b5a595b4>] bpf_check+0x43e3/0x5bc0
>>      [<0000000011e391b1>] bpf_prog_load+0xf26/0x1940
>>      [<0000000007f765c0>] __sys_bpf+0xd2c/0x3650
>>      [<00000000839815d6>] __x64_sys_bpf+0x75/0xc0
>>      [<00000000946ee250>] do_syscall_64+0x3b/0x90
>>      [<0000000000506b7f>] entry_SYSCALL_64_after_hwframe+0x63/0xcd
>>
>> The root case here is: In function prepare_func_exit(), the callee is
>> not released in the abnormal scenario after "state->curframe--;". To
>> fix, move "state->curframe--;" to the very bottom of the function,
>> right when we free callee and reset frame[] pointer to NULL, as Andrii
>> suggested.
>>
>> In addition, function __check_func_call() has a similar problem. In
>> the abnormal scenario before "state->curframe++;", the callee is alse
>> not released.
>>
>> Fixes: 69c087ba6225 ("bpf: Add bpf_for_each_map_elem() helper")
>> Fixes: fd978bf7fd31 ("bpf: Add reference tracking to verifier")
>> Signed-off-by: Wang Yufen <wangyufen@huawei.com>
>> ---
>>   kernel/bpf/verifier.c | 13 +++++++++----
>>   1 file changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
>> index 7f0a9f6..eff7a5a 100644
>> --- a/kernel/bpf/verifier.c
>> +++ b/kernel/bpf/verifier.c
>> @@ -6736,11 +6736,11 @@ static int __check_func_call(struct 
>> bpf_verifier_env *env, struct bpf_insn *insn
>>       /* Transfer references to the callee */
>>       err = copy_reference_state(callee, caller);
>>       if (err)
>> -        return err;
>> +        goto err_out;
>>         err = set_callee_state_cb(env, caller, callee, *insn_idx);
>>       if (err)
>> -        return err;
>> +        goto err_out;
>>         clear_caller_saved_regs(env, caller->regs);
>>   @@ -6757,6 +6757,11 @@ static int __check_func_call(struct 
>> bpf_verifier_env *env, struct bpf_insn *insn
>>           print_verifier_state(env, callee, true);
>>       }
>>       return 0;
>> +
>> +err_out:
>> +    kfree(callee);
>
> Is it sure that free_func_state() is not needed ?

No, free_func_state() is needed. Sorry, I didn't notice that memory is 
alloced for callee->refs in

copy_reference_state()-> copy_array() -> kmalloc_track_caller().  will 
change in v4

Thanks.

Wang

>
>> +    state->frame[state->curframe + 1] = NULL;
>> +    return err;
>>   }
>>     int map_set_for_each_callback_args(struct bpf_verifier_env *env,
>> @@ -6970,8 +6975,7 @@ static int prepare_func_exit(struct 
>> bpf_verifier_env *env, int *insn_idx)
>>           return -EINVAL;
>>       }
>>   -    state->curframe--;
>> -    caller = state->frame[state->curframe];
>> +    caller = state->frame[state->curframe - 1];
>>       if (callee->in_callback_fn) {
>>           /* enforce R0 return value range [0, 1]. */
>>           struct tnum range = callee->callback_ret_range;
>> @@ -7001,6 +7005,7 @@ static int prepare_func_exit(struct 
>> bpf_verifier_env *env, int *insn_idx)
>>               return err;
>>       }
>>   +    state->curframe--;
>
> nit. state->curframe is always pointing to callee upto this point?  
> Instead of doing another +1 dance in the latter 
> 'state->frame[state->curframe + 1] = NULL;', how about do it later like:
>
>     /* clear everything in the callee */
>         free_func_state(callee);
>     state->frame[state->curframe--] = NULL;
>
>
> It shouldn't affect the earlier print_verifier_state() which 
> explicitly takes callee and caller as its arg, right?

Yes, state->curframe not affect print_verifier_state(),  doing 
"state->curframe--"  at the end of the function is better, also will 
change in v4.  Thanks!

>
>>       *insn_idx = callee->callsite + 1;
>>       if (env->log.level & BPF_LOG_LEVEL) {
>>           verbose(env, "returning from callee:\n");
>