On Fri, Aug 04, 2023 at 03:38:07PM +0200, Michal Privoznik wrote:

When spawning a new container (via clone()) we allocate stack for lxcContainerChild(). So far, we allocate 4 pages for the stack and this used to be enough until we started rewriting everything to glib. With glib we switched to g_strerror() which localizes errno strings and thus increases stack usage, while the previously used strerror_r() was more compact.

We're allocating the stack using g_new0, so when we overflowed the stack we started scribbling over other allocations which is horrible to diagnose.

Fortunately, the solution is easy - just increase how much stack the child can use (16 pages ought to be enough for anybody).

I wonder if we're better off switching to mmap(), allocating 17 pages,and then using mprotect() to remove read+write perms from first and/or last page, so that any future overflow will generate SIGBUS immediately.

Resolves: https://gitlab.com/libvirt/libvirt/-/issues/511 Signed-off-by: Michal Privoznik <mprivozn@redhat.com> --- src/lxc/lxc_container.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/lxc/lxc_container.c b/src/lxc/lxc_container.c index 63cf283285..f741a754ce 100644 --- a/src/lxc/lxc_container.c +++ b/src/lxc/lxc_container.c @@ -2132,7 +2132,7 @@ int lxcContainerStart(virDomainDef *def, { pid_t pid; int cflags; - int stacksize = getpagesize() * 4; + int stacksize = getpagesize() * 16; g_autofree char *stack = NULL; char *stacktop; lxc_child_argv_t args = { -- 2.41.0

With regards, Daniel [1] first or last - arches differ on whether stack grows up vs down IIRC -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|