[OmniOS-discuss] ILB memory leak?

[Al Slater]

On 06/11/2015 18:33, Bob Friesenhahn wrote:
> Tracing mmap() calls on the program while is is running might reveal
> something.

Tracing mmap calls with dtrace gives me the following stack, called 
roughly the same number of times as there are doubled-in-size anon 
mappings in the mmap output.

               libc.so.1`mmap
               libc.so.1`lmalloc+0x168
               libc.so.1`posix_spawn_file_actions_addclose+0x2f
               ilbd`ilbd_run_probe+0x19b
               ilbd`ilbd_hc_probe_timer+0x10

Looking at the source for lmalloc (1), it implements the allocation 
attributes we are seeing.

Now, looking at posix_spawn_file_actions_addclose (2), this allocates a 
file_attr_t which is added to the file_actions list (which was stack 
allocated as fd_actions in ilbd_run_probe (3) and inititialized with 
posix_spawn_file_actions_init.

After fd_actions is passed to posix_spawn, I can't see anywhere calling 
posix_spawn_file_actions_destroy to clean up the list.

It seems to me that ilbd_run_probe just needs to call 
posix_spawn_file_actions_destroy appropriately.


(1) /usr/src/lib/libc/port/threads/alloc.c
(2) /usr/src/lib/libc/port/threads/spawn.c
(3) /usr/src/cmd/cmd-inet/usr.lib/ilbd/ilbd_hc.c

-- 
Al Slater