On Thu, Nov 12, 2020 at 7:25 AM Kuan-Ying Lee
<Kuan-Ying.Lee@xxxxxxxxxxxx> wrote:
>
> We hit this issue in our internal test.
> When enabling generic kasan, a kfree()'d object is put into per-cpu
> quarantine first. If the cpu goes offline, object still remains in
> the per-cpu quarantine. If we call kmem_cache_destroy() now, slub
> will report "Objects remaining" error.
>
> [ 74.982625] =============================================================================
> [ 74.983380] BUG test_module_slab (Not tainted): Objects remaining in test_module_slab on __kmem_cache_shutdown()
> [ 74.984145] -----------------------------------------------------------------------------
> [ 74.984145]
> [ 74.984883] Disabling lock debugging due to kernel taint
> [ 74.985561] INFO: Slab 0x(____ptrval____) objects=34 used=1 fp=0x(____ptrval____) flags=0x2ffff00000010200
> [ 74.986638] CPU: 3 PID: 176 Comm: cat Tainted: G B 5.10.0-rc1-00007-g4525c8781ec0-dirty #10
> [ 74.987262] Hardware name: linux,dummy-virt (DT)
> [ 74.987606] Call trace:
> [ 74.987924] dump_backtrace+0x0/0x2b0
> [ 74.988296] show_stack+0x18/0x68
> [ 74.988698] dump_stack+0xfc/0x168
> [ 74.989030] slab_err+0xac/0xd4
> [ 74.989346] __kmem_cache_shutdown+0x1e4/0x3c8
> [ 74.989779] kmem_cache_destroy+0x68/0x130
> [ 74.990176] test_version_show+0x84/0xf0
> [ 74.990679] module_attr_show+0x40/0x60
> [ 74.991218] sysfs_kf_seq_show+0x128/0x1c0
> [ 74.991656] kernfs_seq_show+0xa0/0xb8
> [ 74.992059] seq_read+0x1f0/0x7e8
> [ 74.992415] kernfs_fop_read+0x70/0x338
> [ 74.993051] vfs_read+0xe4/0x250
> [ 74.993498] ksys_read+0xc8/0x180
> [ 74.993825] __arm64_sys_read+0x44/0x58
> [ 74.994203] el0_svc_common.constprop.0+0xac/0x228
> [ 74.994708] do_el0_svc+0x38/0xa0
> [ 74.995088] el0_sync_handler+0x170/0x178
> [ 74.995497] el0_sync+0x174/0x180
> [ 74.996050] INFO: Object 0x(____ptrval____) @offset=15848
> [ 74.996752] INFO: Allocated in test_version_show+0x98/0xf0 age=8188 cpu=6 pid=172
> [ 75.000802] stack_trace_save+0x9c/0xd0
> [ 75.002420] set_track+0x64/0xf0
> [ 75.002770] alloc_debug_processing+0x104/0x1a0
> [ 75.003171] ___slab_alloc+0x628/0x648
> [ 75.004213] __slab_alloc.isra.0+0x2c/0x58
> [ 75.004757] kmem_cache_alloc+0x560/0x588
> [ 75.005376] test_version_show+0x98/0xf0
> [ 75.005756] module_attr_show+0x40/0x60
> [ 75.007035] sysfs_kf_seq_show+0x128/0x1c0
> [ 75.007433] kernfs_seq_show+0xa0/0xb8
> [ 75.007800] seq_read+0x1f0/0x7e8
> [ 75.008128] kernfs_fop_read+0x70/0x338
> [ 75.008507] vfs_read+0xe4/0x250
> [ 75.008990] ksys_read+0xc8/0x180
> [ 75.009462] __arm64_sys_read+0x44/0x58
> [ 75.010085] el0_svc_common.constprop.0+0xac/0x228
> [ 75.011006] kmem_cache_destroy test_module_slab: Slab cache still has objects
>
> Register a cpu hotplug function to remove all objects in the offline
> per-cpu quarantine when cpu is going offline. Set a per-cpu variable
> to indicate this cpu is offline.
>
> Signed-off-by: Kuan-Ying Lee <Kuan-Ying.Lee@xxxxxxxxxxxx>
> ---
> mm/kasan/quarantine.c | 59 +++++++++++++++++++++++++++++++++++++++++--
> 1 file changed, 57 insertions(+), 2 deletions(-)
>
> diff --git a/mm/kasan/quarantine.c b/mm/kasan/quarantine.c
> index 4c5375810449..67fb91ae2bd0 100644
> --- a/mm/kasan/quarantine.c
> +++ b/mm/kasan/quarantine.c
> @@ -29,6 +29,7 @@
> #include <linux/srcu.h>
> #include <linux/string.h>
> #include <linux/types.h>
> +#include <linux/cpuhotplug.h>
>
> #include "../slab.h"
> #include "kasan.h"
> @@ -97,6 +98,7 @@ static void qlist_move_all(struct qlist_head *from, struct qlist_head *to)
> * guarded by quarantine_lock.
> */
Hi Kuan-Ying,
Thanks for fixing this.
> static DEFINE_PER_CPU(struct qlist_head, cpu_quarantine);
> +static DEFINE_PER_CPU(int, cpu_quarantine_offline);
I think cpu_quarantine_offline is better be part of cpu_quarantine
because it logically is and we already obtain a pointer to
cpu_quarantine in quarantine_put, so it will also make the code a bit
shorter.
> /* Round-robin FIFO array of batches. */
> static struct qlist_head global_quarantine[QUARANTINE_BATCHES];
> @@ -176,6 +178,8 @@ void quarantine_put(struct kasan_free_meta *info, struct kmem_cache *cache)
> unsigned long flags;
> struct qlist_head *q;
> struct qlist_head temp = QLIST_INIT;
> + int *offline;
> + struct qlist_head q_offline = QLIST_INIT;
>
> /*
> * Note: irq must be disabled until after we move the batch to the
> @@ -187,8 +191,16 @@ void quarantine_put(struct kasan_free_meta *info, struct kmem_cache *cache)
> */
> local_irq_save(flags);
>
> - q = this_cpu_ptr(&cpu_quarantine);
> - qlist_put(q, &info->quarantine_link, cache->size);
> + offline = this_cpu_ptr(&cpu_quarantine_offline);
> + if (*offline == 0) {
> + q = this_cpu_ptr(&cpu_quarantine);
> + qlist_put(q, &info->quarantine_link, cache->size);
> + } else {
> + qlist_put(&q_offline, &info->quarantine_link, cache->size);
> + qlist_free_all(&q_offline, cache);
This looks like a convoluted way to call qlink_free. I think it will
be better to call qlink_free directly here.
And why do we need this? Because CPU shutdown code can still free some
objects afterwards?
> + local_irq_restore(flags);
> + return;
You add both if/else and early return, this looks like unnecessary
code complication. It would be simpler with:
if (*offline) {
qlink_free(...);
return;
}
... all current per-cpu local ...
> + }
> if (unlikely(q->bytes > QUARANTINE_PERCPU_SIZE)) {
> qlist_move_all(q, &temp);
>
> @@ -328,3 +340,46 @@ void quarantine_remove_cache(struct kmem_cache *cache)
>
> synchronize_srcu(&remove_cache_srcu);
> }
> +
> +static int kasan_cpu_online(unsigned int cpu)
> +{
> + int *offline;
> + unsigned long flags;
> +
> + local_irq_save(flags);
I assume this local_irq_save/restore is to prevent some warnings from
this_cpu_ptr.
But CPU online/offline callbacks should run without preemption already
(preempting/rescheduling on other CPUs does not make sense for them,
right?), so I would assume that is already at least preemption
disabled or something. Is there this_cpu_ptr variant that won't
produce warnings on its own in cpu online/offline callbacks?
This whole function could be a 1-liner:
this_cpu_ptr(&cpu_quarantine)->offline = true;
So I am trying to understand if we could avoid all this unnecessary danse.
> + offline = this_cpu_ptr(&cpu_quarantine_offline);
> + *offline = 0;
> + local_irq_restore(flags);
> + return 0;
> +}
> +
> +static int kasan_cpu_offline(unsigned int cpu)
> +{
> + struct kmem_cache *s;
> + int *offline;
> + unsigned long flags;
> +
> + local_irq_save(flags);
> + offline = this_cpu_ptr(&cpu_quarantine_offline);
> + *offline = 1;
> + local_irq_restore(flags);
> +
> + mutex_lock(&slab_mutex);
> + list_for_each_entry(s, &slab_caches, list) {
> + per_cpu_remove_cache(s);
> + }
> + mutex_unlock(&slab_mutex);
We just want to drop the whole per-cpu cache at once, right? I would
assume there should be a simpler way to do this all at once, rather
than doing this per-slab.
> + return 0;
> +}
> +
> +static int __init kasan_cpu_offline_quarantine_init(void)
> +{
> + int ret = 0;
> +
> + ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "mm/kasan:online",
> + kasan_cpu_online, kasan_cpu_offline);
> + if (ret)
> + pr_err("kasan offline cpu quarantine register failed [%d]\n", ret);
> + return ret;
> +}
> +late_initcall(kasan_cpu_offline_quarantine_init);
> --
> 2.18.0
>
> --
> You received this message because you are subscribed to the Google Groups "kasan-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to kasan-dev+unsubscribe@xxxxxxxxxxxxxxxx.
> To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/1605162252-23886-2-git-send-email-Kuan-Ying.Lee%40mediatek.com.
