[PATCH] Add a check avoid segfault.

[PATCH] Add a check avoid segfault.

[majianpeng]

If new_size was zero or realloc failed, it would be segment fault.
So add a check.

Signed-off-by: Jianpeng Ma <majianpeng@gmail.com>
---
 filesetup.c |    5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/filesetup.c b/filesetup.c
index b1ca921..324e9a6 100644
--- a/filesetup.c
+++ b/filesetup.c
@@ -1077,8 +1077,11 @@ int add_file(struct thread_data *td, const char *fname)
 		int new_size = td->o.nr_files + 1;
 
 		dprint(FD_FILE, "resize file array to %d files\n", new_size);
-
 		td->files = realloc(td->files, new_size * sizeof(f));
+		if (td->files == NULL) {
+			log_err("fio: realloc OOM\n");
+			assert(0);
+		}
 		td->files_size = new_size;
 	}
 	td->files[cur_files] = f;
-- 
1.7.9.5

Re: [PATCH] Add a check avoid segfault.

[Jens Axboe]

On 2013-01-10 12:36, majianpeng wrote:
> If new_size was zero or realloc failed, it would be segment fault.
> So add a check.

It's a drop in the bucket, fio basically does not alloc failure
checking. It'd be nice to improve on that, but you'd need a lot more
than the below to get near fixing that.

-- 
Jens Axboe

Re: [PATCH] Add a check avoid segfault.

[majianpeng]

>On 2013-01-10 12:36, majianpeng wrote:
>> If new_size was zero or realloc failed, it would be segment fault.
>> So add a check.
>
>It's a drop in the bucket, fio basically does not alloc failure
>checking. It'd be nice to improve on that, but you'd need a lot more
>than the below to get near fixing that.
>
>-- 
>Jens Axboe
>
yes, why did i sent this patch?Because i set nrfiles=-1,fio would be segfault.
I viewed the code about judging nrfiles, the code did not check this condition.
Anthor case can cause this bug if nrfiles=0xffffffff.

There are many parameters like nrfiles.So they should do same check.
Maybe add those check is better rather than check alloc fail or success?

Jianpeng
Thanks!

Re: [PATCH] Add a check avoid segfault.

[Jens Axboe]

On 2013-01-11 01:39, majianpeng wrote:
> 
>> On 2013-01-10 12:36, majianpeng wrote:
>>> If new_size was zero or realloc failed, it would be segment fault.
>>> So add a check.
>>
>> It's a drop in the bucket, fio basically does not alloc failure
>> checking. It'd be nice to improve on that, but you'd need a lot more
>> than the below to get near fixing that.
>>
>> -- 
>> Jens Axboe
>>
> yes, why did i sent this patch?Because i set nrfiles=-1,fio would be segfault.
> I viewed the code about judging nrfiles, the code did not check this condition.
> Anthor case can cause this bug if nrfiles=0xffffffff.
> 
> There are many parameters like nrfiles.So they should do same check.
> Maybe add those check is better rather than check alloc fail or success?

OK, you should have mentioned that. Segfaults that are easily caused by
invalid parameters being specified should of course be fixed.

-- 
Jens Axboe