The commit is pushed to "branch-rh7-3.10.0-514.10.2.vz7.29.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git after rh7-3.10.0-514.10.2.vz7.29.4 ------> commit 2d47a05314ed0fd03df75c419eeda00fab40ad2d Author: Eric W. Biederman <ebied...@xmission.com> Date: Thu Mar 16 19:21:00 2017 +0400
ms/fs/file.c:fdtable: avoid triggering OOMs from alloc_fdmem This is a backport of upstream (vanilla) commit: commit 96c7a2ff21501691587e1ae969b83cbec8b78e08 ("fs/file.c:fdtable: avoid triggering OOMs from alloc_fdmem") Under certain conditions there might be a lot of alloc_fdmem() invocations with order <= PAGE_ALLOC_COSTLY_ORDER. For example: httpd which is doing a lot of fork() calls. Real-life examples from our customers: [532506.773243] httpd D ffff8803f5fecc20 0 939874 6606 [532506.773257] Call Trace: [532506.773261] [<ffffffff8163ce29>] schedule+0x29/0x70 [532506.773264] [<ffffffff8163a9d5>] schedule_timeout+0x175/0x2d0 [532506.773272] [<ffffffff8108cc90>] ? internal_add_timer+0x70/0x70 [532506.773276] [<ffffffff8163c3ae>] io_schedule_timeout+0xae/0x130 [532506.773280] [<ffffffff8119be85>] wait_iff_congested+0x135/0x150 [532506.773284] [<ffffffff810a86e0>] ? wake_up_atomic_t+0x30/0x30 [532506.773288] [<ffffffff8119071f>] shrink_inactive_list+0x65f/0x6c0 [532506.773292] [<ffffffff81190f55>] shrink_lruvec+0x395/0x800 [532506.773296] [<ffffffff811914af>] shrink_zone+0xef/0x2d0 [532506.773300] [<ffffffff81191a30>] do_try_to_free_pages+0x170/0x530 [532506.773310] [<ffffffff81191ec5>] try_to_free_pages+0xd5/0x160 [532506.773315] [<ffffffff811850ab>] __alloc_pages_nodemask+0x8ab/0xc10 [532506.773320] [<ffffffff811cb2f9>] alloc_pages_current+0xa9/0x170 [532506.773324] [<ffffffff8119f8f8>] kmalloc_order+0x18/0x50 [532506.773327] [<ffffffff8119f956>] kmalloc_order_trace+0x26/0xa0 [532506.773332] [<ffffffff811d8c69>] __kmalloc+0x259/0x270 [532506.773337] [<ffffffff812184d0>] alloc_fdmem+0x20/0x50 [532506.773341] [<ffffffff812185ac>] alloc_fdtable+0x6c/0xe0 [532506.773344] [<ffffffff81218b69>] dup_fd+0x1f9/0x2d0 [532506.773354] [<ffffffff810797cf>] copy_process.part.30+0x87f/0x1510 [532506.773358] [<ffffffff8107a641>] do_fork+0xe1/0x320 [532506.773370] [<ffffffff8107a906>] SyS_clone+0x16/0x20 [532506.773376] [<ffffffff81648299>] stub_clone+0x69/0x90 [532506.773380] [<ffffffff81647f49>] ? system_call_fastpath+0x16/0x1b [513890.005271] httpd D ffff880425db7230 0 811718 6606 [513890.005279] Call Trace: [513890.005282] [<ffffffff8163ce29>] schedule+0x29/0x70 [513890.005284] [<ffffffff8163aa99>] schedule_timeout+0x239/0x2d0 [513890.005292] [<ffffffff8163c3ae>] io_schedule_timeout+0xae/0x130 [513890.005296] [<ffffffff8163c448>] io_schedule+0x18/0x20 [513890.005298] [<ffffffff812c6268>] get_request+0x218/0x780 [513890.005303] [<ffffffff812c8526>] blk_queue_bio+0xc6/0x3a0 [513890.005309] [<ffffffffa0002c59>] ? dm_make_request+0x119/0x170 [dm_mod] [513890.005311] [<ffffffff812c3892>] generic_make_request+0xe2/0x130 [513890.005313] [<ffffffff812c3957>] submit_bio+0x77/0x1c0 [513890.005318] [<ffffffff811bf87e>] __swap_writepage+0x1be/0x260 [513890.005337] [<ffffffff811bf959>] swap_writepage+0x39/0x80 [513890.005340] [<ffffffff8118f68d>] shrink_page_list+0x4ad/0xa80 [513890.005343] [<ffffffff811902bb>] shrink_inactive_list+0x1fb/0x6c0 [513890.005345] [<ffffffff81190f55>] shrink_lruvec+0x395/0x800 [513890.005348] [<ffffffff811914af>] shrink_zone+0xef/0x2d0 [513890.005350] [<ffffffff81191a30>] do_try_to_free_pages+0x170/0x530 [513890.005353] [<ffffffff81191ec5>] try_to_free_pages+0xd5/0x160 [513890.005355] [<ffffffff811850ab>] __alloc_pages_nodemask+0x8ab/0xc10 [513890.005358] [<ffffffff811cb2f9>] alloc_pages_current+0xa9/0x170 [513890.005360] [<ffffffff8119f8f8>] kmalloc_order+0x18/0x50 [513890.005362] [<ffffffff8119f956>] kmalloc_order_trace+0x26/0xa0 [513890.005365] [<ffffffff811d8c69>] __kmalloc+0x259/0x270 [513890.005367] [<ffffffff812184d0>] alloc_fdmem+0x20/0x50 [513890.005369] [<ffffffff812185ac>] alloc_fdtable+0x6c/0xe0 [513890.005371] [<ffffffff81218b69>] dup_fd+0x1f9/0x2d0 [513890.005376] [<ffffffff810797cf>] copy_process.part.30+0x87f/0x1510 [513890.005378] [<ffffffff8107a641>] do_fork+0xe1/0x320 [513890.005380] [<ffffffff8107a906>] SyS_clone+0x16/0x20 [513890.005382] [<ffffffff81648299>] stub_clone+0x69/0x90 We observed that sometimes kswapd cannot handle this which causes many direct reclaim attempts which in turn: 1. Increases iowait time due to congestion_wait 2. Increases number of block reqs per second due to page swapping and writeback 3. May induce OOMs So it's better DO NOT try that hard to allocate contiguous area, and fallback to vmalloc() as soon as possible. ========================================================= Original commit message: fs/file.c:fdtable: avoid triggering OOMs from alloc_fdmem Recently due to a spike in connections per second memcached on 3 separate boxes triggered the OOM killer from accept. At the time the OOM killer was triggered there was 4GB out of 36GB free in zone 1. The problem was that alloc_fdtable was allocating an order 3 page (32KiB) to hold a bitmap, and there was sufficient fragmentation that the largest page available was 8KiB. I find the logic that PAGE_ALLOC_COSTLY_ORDER can't fail pretty dubious but I do agree that order 3 allocations are very likely to succeed. There are always pathologies where order > 0 allocations can fail when there are copious amounts of free memory available. Using the pigeon hole principle it is easy to show that it requires 1 page more than 50% of the pages being free to guarantee an order 1 (8KiB) allocation will succeed, 1 page more than 75% of the pages being free to guarantee an order 2 (16KiB) allocation will succeed and 1 page more than 87.5% of the pages being free to guarantee an order 3 allocate will succeed. A server churning memory with a lot of small requests and replies like memcached is a common case that if anything can will skew the odds against large pages being available. Therefore let's not give external applications a practical way to kill linux server applications, and specify __GFP_NORETRY to the kmalloc in alloc_fdmem. Unless I am misreading the code and by the time the code reaches should_alloc_retry in __alloc_pages_slowpath (where __GFP_NORETRY becomes signification). We have already tried everything reasonable to allocate a page and the only thing left to do is wait. So not waiting and falling back to vmalloc immediately seems like the reasonable thing to do even if there wasn't a chance of triggering the OOM killer. Signed-off-by: "Eric W. Biederman" <ebied...@xmission.com> Cc: Eric Dumazet <eric.duma...@gmail.com> Acked-by: David Rientjes <rient...@google.com> Cc: Cong Wang <cw...@twopensource.com> Cc: <sta...@vger.kernel.org> Signed-off-by: Andrew Morton <a...@linux-foundation.org> Signed-off-by: Linus Torvalds <torva...@linux-foundation.org> Signed-off-by: Anatoly Stepanov <astepa...@cloudlinux.com> Acked-by: Andrey Ryabinin <aryabi...@virtuozzo.com> --- fs/file.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/fs/file.c b/fs/file.c index c1e802c..1db5952 100644 --- a/fs/file.c +++ b/fs/file.c @@ -37,7 +37,7 @@ static void *alloc_fdmem(size_t size) * vmalloc() if the allocation size will be considered "large" by the VM. */ if (size <= (PAGE_SIZE << PAGE_ALLOC_COSTLY_ORDER)) { - void *data = kmalloc(size, GFP_KERNEL_ACCOUNT|__GFP_NOWARN); + void *data = kmalloc(size, GFP_KERNEL_ACCOUNT|__GFP_NOWARN|__GFP_NORETRY); if (data != NULL) return data; } _______________________________________________ Devel mailing list Devel@openvz.org https://lists.openvz.org/mailman/listinfo/devel