From 9c56751271e7a917783fb57ec49fe8382e0dc867 Mon Sep 17 00:00:00 2001 From: David Rientjes Date: Wed, 16 Oct 2013 13:46:43 -0700 Subject: mm, memcg: protect mem_cgroup_read_events for cpu hotplug for_each_online_cpu() needs the protection of {get,put}_online_cpus() so cpu_online_mask doesn't change during the iteration. cpu_hotplug.lock is held while a cpu is going down, it's a coarse lock that is used kernel-wide to synchronize cpu hotplug activity. Memcg has a cpu hotplug notifier, called while there may not be any cpu hotplug refcounts, which drains per-cpu event counts to memcg->nocpu_base.events to maintain a cumulative event count as cpus disappear. Without get_online_cpus() in mem_cgroup_read_events(), it's possible to account for the event count on a dying cpu twice, and this value may be significantly large. In fact, all memcg->pcp_counter_lock use should be nested by {get,put}_online_cpus(). This fixes that issue and ensures the reported statistics are not vastly over-reported during cpu hotplug. Signed-off-by: David Rientjes Cc: Johannes Weiner Cc: Michal Hocko Cc: KAMEZAWA Hiroyuki Acked-by: KOSAKI Motohiro Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/memcontrol.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'mm/memcontrol.c') diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 1c52ddbc839b..5335b2b6be77 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -866,6 +866,7 @@ static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, unsigned long val = 0; int cpu; + get_online_cpus(); for_each_online_cpu(cpu) val += per_cpu(memcg->stat->events[idx], cpu); #ifdef CONFIG_HOTPLUG_CPU @@ -873,6 +874,7 @@ static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, val += memcg->nocpu_base.events[idx]; spin_unlock(&memcg->pcp_counter_lock); #endif + put_online_cpus(); return val; } -- cgit v1.2.3 From 4942642080ea82d99ab5b653abb9a12b7ba31f4a Mon Sep 17 00:00:00 2001 From: Johannes Weiner Date: Wed, 16 Oct 2013 13:46:59 -0700 Subject: mm: memcg: handle non-error OOM situations more gracefully Commit 3812c8c8f395 ("mm: memcg: do not trap chargers with full callstack on OOM") assumed that only a few places that can trigger a memcg OOM situation do not return VM_FAULT_OOM, like optional page cache readahead. But there are many more and it's impractical to annotate them all. First of all, we don't want to invoke the OOM killer when the failed allocation is gracefully handled, so defer the actual kill to the end of the fault handling as well. This simplifies the code quite a bit for added bonus. Second, since a failed allocation might not be the abrupt end of the fault, the memcg OOM handler needs to be re-entrant until the fault finishes for subsequent allocation attempts. If an allocation is attempted after the task already OOMed, allow it to bypass the limit so that it can quickly finish the fault and invoke the OOM killer. Reported-by: azurIt Signed-off-by: Johannes Weiner Cc: Michal Hocko Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/memcontrol.c | 139 +++++++++++++++++++++----------------------------------- 1 file changed, 51 insertions(+), 88 deletions(-) (limited to 'mm/memcontrol.c') diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 5335b2b6be77..65fc6a449841 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -2161,110 +2161,59 @@ static void memcg_oom_recover(struct mem_cgroup *memcg) memcg_wakeup_oom(memcg); } -/* - * try to call OOM killer - */ static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) { - bool locked; - int wakeups; - if (!current->memcg_oom.may_oom) return; - - current->memcg_oom.in_memcg_oom = 1; - /* - * As with any blocking lock, a contender needs to start - * listening for wakeups before attempting the trylock, - * otherwise it can miss the wakeup from the unlock and sleep - * indefinitely. This is just open-coded because our locking - * is so particular to memcg hierarchies. + * We are in the middle of the charge context here, so we + * don't want to block when potentially sitting on a callstack + * that holds all kinds of filesystem and mm locks. + * + * Also, the caller may handle a failed allocation gracefully + * (like optional page cache readahead) and so an OOM killer + * invocation might not even be necessary. + * + * That's why we don't do anything here except remember the + * OOM context and then deal with it at the end of the page + * fault when the stack is unwound, the locks are released, + * and when we know whether the fault was overall successful. */ - wakeups = atomic_read(&memcg->oom_wakeups); - mem_cgroup_mark_under_oom(memcg); - - locked = mem_cgroup_oom_trylock(memcg); - - if (locked) - mem_cgroup_oom_notify(memcg); - - if (locked && !memcg->oom_kill_disable) { - mem_cgroup_unmark_under_oom(memcg); - mem_cgroup_out_of_memory(memcg, mask, order); - mem_cgroup_oom_unlock(memcg); - /* - * There is no guarantee that an OOM-lock contender - * sees the wakeups triggered by the OOM kill - * uncharges. Wake any sleepers explicitely. - */ - memcg_oom_recover(memcg); - } else { - /* - * A system call can just return -ENOMEM, but if this - * is a page fault and somebody else is handling the - * OOM already, we need to sleep on the OOM waitqueue - * for this memcg until the situation is resolved. - * Which can take some time because it might be - * handled by a userspace task. - * - * However, this is the charge context, which means - * that we may sit on a large call stack and hold - * various filesystem locks, the mmap_sem etc. and we - * don't want the OOM handler to deadlock on them - * while we sit here and wait. Store the current OOM - * context in the task_struct, then return -ENOMEM. - * At the end of the page fault handler, with the - * stack unwound, pagefault_out_of_memory() will check - * back with us by calling - * mem_cgroup_oom_synchronize(), possibly putting the - * task to sleep. - */ - current->memcg_oom.oom_locked = locked; - current->memcg_oom.wakeups = wakeups; - css_get(&memcg->css); - current->memcg_oom.wait_on_memcg = memcg; - } + css_get(&memcg->css); + current->memcg_oom.memcg = memcg; + current->memcg_oom.gfp_mask = mask; + current->memcg_oom.order = order; } /** * mem_cgroup_oom_synchronize - complete memcg OOM handling + * @handle: actually kill/wait or just clean up the OOM state * - * This has to be called at the end of a page fault if the the memcg - * OOM handler was enabled and the fault is returning %VM_FAULT_OOM. + * This has to be called at the end of a page fault if the memcg OOM + * handler was enabled. * - * Memcg supports userspace OOM handling, so failed allocations must + * Memcg supports userspace OOM handling where failed allocations must * sleep on a waitqueue until the userspace task resolves the * situation. Sleeping directly in the charge context with all kinds * of locks held is not a good idea, instead we remember an OOM state * in the task and mem_cgroup_oom_synchronize() has to be called at - * the end of the page fault to put the task to sleep and clean up the - * OOM state. + * the end of the page fault to complete the OOM handling. * * Returns %true if an ongoing memcg OOM situation was detected and - * finalized, %false otherwise. + * completed, %false otherwise. */ -bool mem_cgroup_oom_synchronize(void) +bool mem_cgroup_oom_synchronize(bool handle) { + struct mem_cgroup *memcg = current->memcg_oom.memcg; struct oom_wait_info owait; - struct mem_cgroup *memcg; + bool locked; /* OOM is global, do not handle */ - if (!current->memcg_oom.in_memcg_oom) - return false; - - /* - * We invoked the OOM killer but there is a chance that a kill - * did not free up any charges. Everybody else might already - * be sleeping, so restart the fault and keep the rampage - * going until some charges are released. - */ - memcg = current->memcg_oom.wait_on_memcg; if (!memcg) - goto out; + return false; - if (test_thread_flag(TIF_MEMDIE) || fatal_signal_pending(current)) - goto out_memcg; + if (!handle) + goto cleanup; owait.memcg = memcg; owait.wait.flags = 0; @@ -2273,13 +2222,25 @@ bool mem_cgroup_oom_synchronize(void) INIT_LIST_HEAD(&owait.wait.task_list); prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); - /* Only sleep if we didn't miss any wakeups since OOM */ - if (atomic_read(&memcg->oom_wakeups) == current->memcg_oom.wakeups) + mem_cgroup_mark_under_oom(memcg); + + locked = mem_cgroup_oom_trylock(memcg); + + if (locked) + mem_cgroup_oom_notify(memcg); + + if (locked && !memcg->oom_kill_disable) { + mem_cgroup_unmark_under_oom(memcg); + finish_wait(&memcg_oom_waitq, &owait.wait); + mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, + current->memcg_oom.order); + } else { schedule(); - finish_wait(&memcg_oom_waitq, &owait.wait); -out_memcg: - mem_cgroup_unmark_under_oom(memcg); - if (current->memcg_oom.oom_locked) { + mem_cgroup_unmark_under_oom(memcg); + finish_wait(&memcg_oom_waitq, &owait.wait); + } + + if (locked) { mem_cgroup_oom_unlock(memcg); /* * There is no guarantee that an OOM-lock contender @@ -2288,10 +2249,9 @@ out_memcg: */ memcg_oom_recover(memcg); } +cleanup: + current->memcg_oom.memcg = NULL; css_put(&memcg->css); - current->memcg_oom.wait_on_memcg = NULL; -out: - current->memcg_oom.in_memcg_oom = 0; return true; } @@ -2705,6 +2665,9 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm, || fatal_signal_pending(current))) goto bypass; + if (unlikely(task_in_memcg_oom(current))) + goto bypass; + /* * We always charge the cgroup the mm_struct belongs to. * The mm_struct's mem_cgroup changes on task migration if the -- cgit v1.2.3 From 84235de394d9775bfaa7fa9762a59d91fef0c1fc Mon Sep 17 00:00:00 2001 From: Johannes Weiner Date: Wed, 16 Oct 2013 13:47:00 -0700 Subject: fs: buffer: move allocation failure loop into the allocator Buffer allocation has a very crude indefinite loop around waking the flusher threads and performing global NOFS direct reclaim because it can not handle allocation failures. The most immediate problem with this is that the allocation may fail due to a memory cgroup limit, where flushers + direct reclaim might not make any progress towards resolving the situation at all. Because unlike the global case, a memory cgroup may not have any cache at all, only anonymous pages but no swap. This situation will lead to a reclaim livelock with insane IO from waking the flushers and thrashing unrelated filesystem cache in a tight loop. Use __GFP_NOFAIL allocations for buffers for now. This makes sure that any looping happens in the page allocator, which knows how to orchestrate kswapd, direct reclaim, and the flushers sensibly. It also allows memory cgroups to detect allocations that can't handle failure and will allow them to ultimately bypass the limit if reclaim can not make progress. Reported-by: azurIt Signed-off-by: Johannes Weiner Cc: Michal Hocko Cc: Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- mm/memcontrol.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'mm/memcontrol.c') diff --git a/mm/memcontrol.c b/mm/memcontrol.c index 65fc6a449841..34d3ca9572d6 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -2766,6 +2766,8 @@ done: return 0; nomem: *ptr = NULL; + if (gfp_mask & __GFP_NOFAIL) + return 0; return -ENOMEM; bypass: *ptr = root_mem_cgroup; -- cgit v1.2.3