Message-ID: <55249D3D.7020706@huawei.com>
Date: Wed, 8 Apr 2015 11:15:09 +0800
From: He Kuang <hekuang@huawei.com>
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:31.0) Gecko/20100101 Thunderbird/31.3.0
MIME-Version: 1.0
To: Arnaldo Carvalho de Melo <acme@kernel.org>
CC: <a.p.zijlstra@chello.nl>, <mingo@redhat.com>, <jolsa@kernel.org>,
        <wangnan0@huawei.com>, <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 2/2] perf trace: Fix segmentfault on perf trace
References: <1428399071-7141-1-git-send-email-hekuang@huawei.com> <1428399071-7141-2-git-send-email-hekuang@huawei.com> <20150407123633.GG11983@kernel.org>
In-Reply-To: <20150407123633.GG11983@kernel.org>
Content-Type: text/plain; charset="windows-1252"; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3888
Lines: 112

Hi, Arnaldo
On 2015/4/7 20:36, Arnaldo Carvalho de Melo wrote:
> Em Tue, Apr 07, 2015 at 05:31:11PM +0800, He Kuang escreveu:
>> After perf_evlist__filter_pollfd() filters out fds and releases
>> perf_mmap by using perf_evlist__mmap_put(), refcnt of perf_mmap hits 1
>> then perf_evlist__mmap_consume() will do the final unmap. In this
>> condition, perf_evlist__mmap_read() will crash by referencing invalid
>> mmap. Put refcnt check before use.
>>
>> Can be reproduced as following:
> After applying 1/2 in this series and trying to reproduce I couldn't, it
> works, looking at the code...
>
> Let me get my head around this, idea was that after all fds associated
> with a mmap would be closed, i.e. the perf_mmap->refcnt hits zero, then
> we would have to drain whatever was left in the mmap, but looking again
> that doesn't look like that is what is doing, becaue in filter_pollfd we
> will munmap it before being able to "drain" it, as all mmaps were
> closed, thus filter_pollfd returned zero...

In function __perf_evlist__mmap(), refcnt is initialized to 2, see commit:
   823969860329 ("perf evlist: Refcount mmaps")

After filter_pollfd,  perf_mmap->refcnt is 1 not 0.

   perf_evlist__filter_pollfd()                     -- refcnt=1
   draining = true
   if (perf_evlist__mmap_read() != NULL)
           perf_evlist__mmap_consume()    -- unmap, refcnt = 0
           perf_evlist__mmap_read()            -- segfault
else
exit

I noticed that this issue also exists in builtin-record.c, but it
checks before mmap_read():

if (rec->evlist->mmap[i].base) {
         if (record__mmap_read(rec, i, draining) != 0) {

So we can either do the check outside
builtin-trace.c:perf_evlist__mmap_read() like what
builtin-record.c do or inside. What's your opinion?
>
> Reading on, thanks for the patch!
>
> - Arnaldo
>
>   
>>    $ perf trace --duration 1.0 ls
>>      ...
>>      perf: Segmentation fault
>>      Obtained 14 stack frames.
>>      ./perf(dump_stack+0x2e) [0x503c2d]
>>      ./perf(sighandler_dump_stack+0x2e)
>>      [0x503d0c]
>>      /lib64/libc.so.6(+0x34df0) [0x7f5fd9a4adf0]
>>      ./perf() [0x4a8fda]
>>      ./perf(perf_evlist__mmap_read+0x56)
>>      [0x4aae93]
>>      ./perf() [0x470b28]
>>      ./perf(cmd_trace+0xada) [0x4727bd]
>>      ./perf() [0x49c4f4]
>>      ./perf() [0x49c74d]
>>      ./perf() [0x49c899]
>>      ./perf(main+0x23b)
>>      [0x49cbfa]
>>      /lib64/libc.so.6(__libc_start_main+0xf5)
>>      [0x7f5fd9a377b5]
>>      ./perf() [0x434ea5]
>>      [(nil)]
>>
>> Signed-off-by: He Kuang <hekuang@huawei.com>
>> ---
>>   tools/perf/util/evlist.c | 13 ++++++++++---
>>   1 file changed, 10 insertions(+), 3 deletions(-)
>>
>> diff --git a/tools/perf/util/evlist.c b/tools/perf/util/evlist.c
>> index 76ef7ee..9d36433 100644
>> --- a/tools/perf/util/evlist.c
>> +++ b/tools/perf/util/evlist.c
>> @@ -634,11 +634,18 @@ static struct perf_evsel *perf_evlist__event2evsel(struct perf_evlist *evlist,
>>   union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx)
>>   {
>>   	struct perf_mmap *md = &evlist->mmap[idx];
>> -	unsigned int head = perf_mmap__read_head(md);
>> -	unsigned int old = md->prev;
>> -	unsigned char *data = md->base + page_size;
>> +	unsigned int head;
>> +	unsigned int old;
>> +	unsigned char *data;
>>   	union perf_event *event = NULL;
>>   
>> +	if (md == NULL || md->refcnt == 0)
>> +		return NULL;
>> +
>> +	head = perf_mmap__read_head(md);
>> +	old = md->prev;
>> +	data = md->base + page_size;
>> +
>>   	if (evlist->overwrite) {
>>   		/*
>>   		 * If we're further behind than half the buffer, there's a chance
>> -- 
>> 2.3.3.220.g9ab698f
>


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/