(Apologies for spam, I've Cc'ed a few outdated emails in the previous mail) On 12/09/2014 09:15 PM, Sasha Levin wrote: > On 11/18/2014 04:58 PM, Andrew Morton wrote: >>> [ 1026.994816] Modules linked in: >>>> [ 1026.995378] CPU: 7 PID: 7879 Comm: trinity-c374 Not tainted 3.18.0-rc4-next-20141113-sasha-00047-gd1763ce-dirty #1455 >>>> [ 1026.996123] FAULT_INJECTION: forcing a failure. >>>> [ 1026.996123] name failslab, interval 100, probability 30, space 0, times -1 >>>> [ 1026.999050] 0000000000000000 0000000000000000 0000000000b3d300 ffff88061295bbd8 >>>> [ 1027.000676] ffffffff92f71097 0000000000000000 ffffea0000b3d300 ffff88061295bc08 >>>> [ 1027.002020] ffffffff8197ef7a ffffea0000b3d300 ffffffff942dd148 dfffe90000000000 >>>> [ 1027.003359] Call Trace: >>>> [ 1027.003831] dump_stack (lib/dump_stack.c:52) >>>> [ 1027.004725] bad_page (mm/page_alloc.c:338) >>>> [ 1027.005623] free_pages_prepare (mm/page_alloc.c:657 mm/page_alloc.c:763) >>>> [ 1027.006761] free_hot_cold_page (mm/page_alloc.c:1438) >>>> [ 1027.007772] ? __page_cache_release (mm/swap.c:66) >>>> [ 1027.008815] put_page (mm/swap.c:270) >>>> [ 1027.009665] page_cache_pipe_buf_release (fs/splice.c:93) >>>> [ 1027.010888] __splice_from_pipe (fs/splice.c:784 fs/splice.c:886) >>>> [ 1027.011917] ? might_fault (./arch/x86/include/asm/current.h:14 mm/memory.c:3734) >>>> [ 1027.012856] ? pipe_lock (fs/pipe.c:69) >>>> [ 1027.013728] ? write_pipe_buf (fs/splice.c:1534) >>>> [ 1027.014756] vmsplice_to_user (fs/splice.c:1574) >>>> [ 1027.015725] ? rcu_read_lock_held (kernel/rcu/update.c:169) >>>> [ 1027.016757] ? __fget_light (include/linux/fdtable.h:80 fs/file.c:684) >>>> [ 1027.017782] SyS_vmsplice (fs/splice.c:1656 fs/splice.c:1639) >>>> [ 1027.018863] tracesys_phase2 (arch/x86/kernel/entry_64.S:529) >>>> >> So what happened here? Userspace fed some mlocked memory into splice() >> and then, while splice() was running, userspace dropped its reference >> to the memory, leaving splice() with the last reference. Yet somehow, >> that page was still marked as being mlocked. I wouldn't expect the >> kernel to permit userspace to drop its reference to the memory without >> first clearing the mlocked state. >> >> Is it possible to work out from trinity sources what the exact sequence >> was? Which syscalls are being used, for example? > > Phew, this took a long while but I've bisected it (with good confidence) down > to: > > commit a38246260912ba4a0f8b563704a965a7a97cf3c3 > Author: Davidlohr Bueso <dave@xxxxxxxxxxxx> > Date: Wed Dec 3 18:54:27 2014 +1100 > > mm/memory.c: share the i_mmap_rwsem > > The unmap_mapping_range family of functions do the unmapping of user pages > (ultimately via zap_page_range_single) without touching the actual > interval tree, thus share the lock. > > Signed-off-by: Davidlohr Bueso <dbueso@xxxxxxx> > Cc: "Kirill A. Shutemov" <kirill@xxxxxxxxxxxxx> > Acked-by: Hugh Dickins <hughd@xxxxxxxxxx> > Cc: Oleg Nesterov <oleg@xxxxxxxxxx> > Cc: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx> > Cc: Rik van Riel <riel@xxxxxxxxxx> > Cc: Srikar Dronamraju <srikar@xxxxxxxxxxxxxxxxxx> > Acked-by: Mel Gorman <mgorman@xxxxxxx> > Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> > > > Thanks, > Sasha > -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>