Quoting Eric Blake (eblake@redhat.com):

On 08/13/2013 08:07 AM, Serge Hallyn wrote:

...

Quoting Eric Blake (eblake@redhat.com):

...

Leak reported via Natanael Copa on IRC when used by libvirt:

==623== 5,677,056 bytes in 693 blocks are definitely lost in loss record 3,354 of 3,358 ==623== at 0x4C1138C: malloc (vg_replace_malloc.c:270) ==623== by 0x5B4A9F3: open_memstream (in /lib/libuClibc-0.9.33.2.so) ==623== by 0x5B485AF: vasprintf (in /lib/libuClibc-0.9.33.2.so) ==623== by 0x10BE7C8B: ??? ==623== by 0x10BE7EE2: ??? ==623== by 0x10BE8473: ??? ==623== by 0x10BE8A53: ??? ==623== by 0x109BD755: ??? ==623== by 0x109BE47D: ??? ==623== by 0x53450EF: virConnectListDefinedInterfaces (in /home/ncopa/aports/main/libvirt/src/libvirt-1.1.1/src/.libs/libvirt.so.0.1001.1)

Even without symbols, the lone call to vasprintf in drv_debian was a pretty blatant leak.

• src/drv_debian.c (interface_deps): Free path when done with it.

Thanks. I wonder if this could be responsible for the slow libvirt leak I'm seeing. (Though I doubt it for several reasons..)

I'll aim to push an update to debian and ubuntu this week.

Now pushed.

This does in fact fix the memory leak I was seeing before in libvirt.

I'll try to ship packages today - thanks Eric!