Message ID | 20200421163455.2177998-1-daniel.m.jordan@oracle.com (mailing list archive) |
---|---|
State | Accepted |
Delegated to: | Herbert Xu |
Headers | show |
Series | padata: add separate cpuhp node for CPUHP_PADATA_DEAD | expand |
Hi [This is an automated email] This commit has been processed because it contains a "Fixes:" tag fixing commit: 894c9ef9780c ("padata: validate cpumask without removed CPU during offline"). The bot has tested the following trees: v5.6.5, v5.4.33. v5.6.5: Build OK! v5.4.33: Failed to apply! Possible dependencies: bfcdcef8c8e3 ("padata: update documentation") NOTE: The patch will not be queued to stable trees until it is upstream. How should we proceed with this patch?
On Wed, Apr 22, 2020 at 01:27:02PM +0000, Sasha Levin wrote: > Hi Hi! > v5.4.33: Failed to apply! Possible dependencies: > bfcdcef8c8e3 ("padata: update documentation") Yes, it's a trivial conflict in the header comment for padata_instance. > How should we proceed with this patch? If mainline is ok with the change, I will send a 5.4-only version with the resolution. By the way, I should not have cc'd stable on the email, it was picked up from the Cc: line in the patch. Daniel
On Tue, Apr 21, 2020 at 12:34:55PM -0400, Daniel Jordan wrote: > Removing the pcrypt module triggers this: > > general protection fault, probably for non-canonical > address 0xdead000000000122 > CPU: 5 PID: 264 Comm: modprobe Not tainted 5.6.0+ #2 > Hardware name: QEMU Standard PC > RIP: 0010:__cpuhp_state_remove_instance+0xcc/0x120 > Call Trace: > padata_sysfs_release+0x74/0xce > kobject_put+0x81/0xd0 > padata_free+0x12/0x20 > pcrypt_exit+0x43/0x8ee [pcrypt] > > padata instances wrongly use the same hlist node for the online and dead > states, so __padata_free()'s second cpuhp remove call chokes on the node > that the first poisoned. > > cpuhp multi-instance callbacks only walk forward in cpuhp_step->list and > the same node is linked in both the online and dead lists, so the list > corruption that results from padata_alloc() adding the node to a second > list without removing it from the first doesn't cause problems as long > as no instances are freed. > > Avoid the issue by giving each state its own node. > > Fixes: 894c9ef9780c ("padata: validate cpumask without removed CPU during offline") > Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> > Cc: Herbert Xu <herbert@gondor.apana.org.au> > Cc: Steffen Klassert <steffen.klassert@secunet.com> > Cc: linux-crypto@vger.kernel.org > Cc: linux-kernel@vger.kernel.org > Cc: stable@vger.kernel.org # v5.4+ > --- > include/linux/padata.h | 6 ++++-- > kernel/padata.c | 14 ++++++++------ > 2 files changed, 12 insertions(+), 8 deletions(-) Patch applied. Thanks.
diff --git a/include/linux/padata.h b/include/linux/padata.h index a0d8b41850b2..693cae9bfe66 100644 --- a/include/linux/padata.h +++ b/include/linux/padata.h @@ -139,7 +139,8 @@ struct padata_shell { /** * struct padata_instance - The overall control structure. * - * @node: Used by CPU hotplug. + * @cpu_online_node: Linkage for CPU online callback. + * @cpu_dead_node: Linkage for CPU offline callback. * @parallel_wq: The workqueue used for parallel work. * @serial_wq: The workqueue used for serial work. * @pslist: List of padata_shell objects attached to this instance. @@ -150,7 +151,8 @@ struct padata_shell { * @flags: padata flags. */ struct padata_instance { - struct hlist_node node; + struct hlist_node cpu_online_node; + struct hlist_node cpu_dead_node; struct workqueue_struct *parallel_wq; struct workqueue_struct *serial_wq; struct list_head pslist; diff --git a/kernel/padata.c b/kernel/padata.c index a6afa12fb75e..aae789896616 100644 --- a/kernel/padata.c +++ b/kernel/padata.c @@ -703,7 +703,7 @@ static int padata_cpu_online(unsigned int cpu, struct hlist_node *node) struct padata_instance *pinst; int ret; - pinst = hlist_entry_safe(node, struct padata_instance, node); + pinst = hlist_entry_safe(node, struct padata_instance, cpu_online_node); if (!pinst_has_cpu(pinst, cpu)) return 0; @@ -718,7 +718,7 @@ static int padata_cpu_dead(unsigned int cpu, struct hlist_node *node) struct padata_instance *pinst; int ret; - pinst = hlist_entry_safe(node, struct padata_instance, node); + pinst = hlist_entry_safe(node, struct padata_instance, cpu_dead_node); if (!pinst_has_cpu(pinst, cpu)) return 0; @@ -734,8 +734,9 @@ static enum cpuhp_state hp_online; static void __padata_free(struct padata_instance *pinst) { #ifdef CONFIG_HOTPLUG_CPU - cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD, &pinst->node); - cpuhp_state_remove_instance_nocalls(hp_online, &pinst->node); + cpuhp_state_remove_instance_nocalls(CPUHP_PADATA_DEAD, + &pinst->cpu_dead_node); + cpuhp_state_remove_instance_nocalls(hp_online, &pinst->cpu_online_node); #endif WARN_ON(!list_empty(&pinst->pslist)); @@ -939,9 +940,10 @@ static struct padata_instance *padata_alloc(const char *name, mutex_init(&pinst->lock); #ifdef CONFIG_HOTPLUG_CPU - cpuhp_state_add_instance_nocalls_cpuslocked(hp_online, &pinst->node); + cpuhp_state_add_instance_nocalls_cpuslocked(hp_online, + &pinst->cpu_online_node); cpuhp_state_add_instance_nocalls_cpuslocked(CPUHP_PADATA_DEAD, - &pinst->node); + &pinst->cpu_dead_node); #endif put_online_cpus();
Removing the pcrypt module triggers this: general protection fault, probably for non-canonical address 0xdead000000000122 CPU: 5 PID: 264 Comm: modprobe Not tainted 5.6.0+ #2 Hardware name: QEMU Standard PC RIP: 0010:__cpuhp_state_remove_instance+0xcc/0x120 Call Trace: padata_sysfs_release+0x74/0xce kobject_put+0x81/0xd0 padata_free+0x12/0x20 pcrypt_exit+0x43/0x8ee [pcrypt] padata instances wrongly use the same hlist node for the online and dead states, so __padata_free()'s second cpuhp remove call chokes on the node that the first poisoned. cpuhp multi-instance callbacks only walk forward in cpuhp_step->list and the same node is linked in both the online and dead lists, so the list corruption that results from padata_alloc() adding the node to a second list without removing it from the first doesn't cause problems as long as no instances are freed. Avoid the issue by giving each state its own node. Fixes: 894c9ef9780c ("padata: validate cpumask without removed CPU during offline") Signed-off-by: Daniel Jordan <daniel.m.jordan@oracle.com> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Steffen Klassert <steffen.klassert@secunet.com> Cc: linux-crypto@vger.kernel.org Cc: linux-kernel@vger.kernel.org Cc: stable@vger.kernel.org # v5.4+ --- include/linux/padata.h | 6 ++++-- kernel/padata.c | 14 ++++++++------ 2 files changed, 12 insertions(+), 8 deletions(-) base-commit: ae83d0b416db002fe95601e7f97f64b59514d936