Patchwork Memory leak on UBI volume truncating

login
register
mail settings
Submitter Artem Bityutskiy
Date Jan. 17, 2010, 10:37 a.m.
Message ID <1263724633.8276.135.camel@localhost.localdomain>
Download mbox | patch
Permalink /patch/43020/
State New
Headers show

Comments

Artem Bityutskiy - Jan. 17, 2010, 10:37 a.m.
Hi,

On Wed, 2010-01-13 at 15:28 +0100, Marek Skuczynski wrote:
> Hello,
>   I have prepare a simple volume update stress test (see test code below).
>   This test has been run for kernel 2.6.23 with some updates from 2.6.28,
>   and always after a few minutes the OOM killer was launched.
> 
>   What i found it that each an UBI volume truncate operation with
> ubiupdatevol tool
>   causes memory leak. I think this happens because:
> 
>   - ubi_start_update()  param "bytes" is equal 0
> 
>   - vol->updating flag is re-set to 0
> 
>   - vol->upd_buf is allocated regardless of vol->updating flag,
>      but not released on device close by vol_cdev_release()
> 
>   I never run the test on a newer kernel version, so I cannot confirm
> that this problem still exists.
> 
>   Please confirm, whether my findings are correct or not, thanks.

thanks for this finding. Looks like your analysis is right. Does this
simple patch help?

Patch

diff --git a/drivers/mtd/ubi/upd.c b/drivers/mtd/ubi/upd.c
index c1d7b88..9e3cd34 100644
--- a/drivers/mtd/ubi/upd.c
+++ b/drivers/mtd/ubi/upd.c
@@ -155,12 +155,12 @@  int ubi_start_update(struct ubi_device *ubi, struct ubi_volume *vol,
                if (err)
                        return err;
                vol->updating = 0;
+       } else {
+               vol->upd_buf = vmalloc(ubi->leb_size);
+               if (!vol->upd_buf)
+                       return -ENOMEM;
        }

-       vol->upd_buf = vmalloc(ubi->leb_size);
-       if (!vol->upd_buf)
-               return -ENOMEM;
-
        vol->upd_ebs = div_u64(bytes + vol->usable_leb_size - 1,
                               vol->usable_leb_size);
        vol->upd_bytes = bytes;