Re: [PATCH] slab: fix a crash by reading /proc/slab_allocators

[Qian Cai <cai@lca.pw>]

On Sun, 2019-04-07 at 19:35 -1000, Linus Torvalds wrote:
> On Sat, Apr 6, 2019 at 12:59 PM Qian Cai <cai@lca.pw> wrote:
> > 
> > The commit 510ded33e075 ("slab: implement slab_root_caches list")
> > changes the name of the list node within "struct kmem_cache" from
> > "list" to "root_caches_node", but leaks_show() still use the "list"
> > which causes a crash when reading /proc/slab_allocators.
> 
> The patch does seem to be correct, and I have applied it.
> 
> However, it does strike me that apparently this wasn't caught for two
> years. Which makes me wonder whether we should (once again) discuss
> just removing SLAB entirely, or at least removing the
> /proc/slab_allocators file. Apparently it has never been used in the
> last two years. At some point a "this can't have worked if  anybody
> ever tried to use it" situation means that the code should likely be
> excised.
> 
> Qian, how did you end up noticing and debugging this?

There are some nice texts for CONFIG_SLAB Kconfig written in 2007,

"The regular slab allocator that is established and known to work well in all
environments."

"tricked" me into enabling it in a debug kernel for running testing where LTP
proc01 test case (read all files in procfs) would usually trigger the crash
(Sometimes, "cat /proc/slab_allocators" would just end up printing nothing).

Normally, all those debug kernels would use CONFIG_KASAN which would set
CONFIG_DEBUG_SLAB=n. However, there is no KASAN for powerpc yet, so it selects
CONFIG_DEBUG_SLAB=y there, and then the testing found the issue.