Re: [PATCHv12 32/37] thp: reintroduce split_huge_page()

[Kirill A. Shutemov]

On Wed, Nov 18, 2015 at 11:24:28AM -0500, Sasha Levin wrote:
> On 10/06/2015 11:23 AM, Kirill A. Shutemov wrote:
> > This patch adds implementation of split_huge_page() for new
> > refcountings.
> > 
> > Unlike previous implementation, new split_huge_page() can fail if
> > somebody holds GUP pin on the page. It also means that pin on page
> > would prevent it from bening split under you. It makes situation in
> > many places much cleaner.
> > 
> > The basic scheme of split_huge_page():
> > 
> >   - Check that sum of mapcounts of all subpage is equal to page_count()
> >     plus one (caller pin). Foll off with -EBUSY. This way we can avoid
> >     useless PMD-splits.
> > 
> >   - Freeze the page counters by splitting all PMD and setup migration
> >     PTEs.
> > 
> >   - Re-check sum of mapcounts against page_count(). Page's counts are
> >     stable now. -EBUSY if page is pinned.
> > 
> >   - Split compound page.
> > 
> >   - Unfreeze the page by removing migration entries.
> > 
> > Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx>
> > Tested-by: Sasha Levin <sasha.levin@xxxxxxxxxx>
> > Tested-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxxxxxxx>
> > Acked-by: Jerome Marchand <jmarchan@xxxxxxxxxx>
> 
> Hey Kirill,
> 
> I saw the following while fuzzing:
> 
> [ 3400.024040] ==================================================================
> [ 3400.024040] BUG: KASAN: slab-out-of-bounds in unfreeze_page+0x706/0xbf0 at addr ffff880670dbc0c0
> [ 3400.024040] Read of size 8 by task run_vmtests/10752
> [ 3400.024040] =============================================================================
> [ 3400.024040] BUG vm_area_struct (Not tainted): kasan: bad access detected
> [ 3400.024040] -----------------------------------------------------------------------------
> [ 3400.024040]
> [ 3400.024040] Disabling lock debugging due to kernel taint
> [ 3400.024040] INFO: Allocated in copy_process+0x2d6d/0x5b00 age=18600 cpu=28 pid=9566
> [ 3400.024040]  ___slab_alloc+0x434/0x5b0
> [ 3400.024040]  __slab_alloc.isra.37+0x79/0xd0
> [ 3400.024040]  kmem_cache_alloc+0x103/0x330
> [ 3400.024040]  copy_process+0x2d6d/0x5b00
> [ 3400.024040]  _do_fork+0x180/0xbb0
> [ 3400.024040]  SyS_clone+0x3c/0x50
> [ 3400.024040]  tracesys_phase2+0x88/0x8d
> [ 3400.024040] INFO: Freed in remove_vma+0x170/0x180 age=18613 cpu=10 pid=21787
> [ 3400.024040]  __slab_free+0x64/0x260
> [ 3400.024040]  kmem_cache_free+0x1e1/0x3b0
> [ 3400.024040]  remove_vma+0x170/0x180
> [ 3400.024040]  exit_mmap+0x30a/0x3c0
> [ 3400.024040]  mmput+0x98/0x240
> [ 3400.024040]  do_exit+0xbe5/0x2830
> [ 3400.024040]  do_group_exit+0x1b5/0x300
> [ 3400.024040]  SyS_exit_group+0x22/0x30
> [ 3400.024040]  tracesys_phase2+0x88/0x8d
> [ 3400.024040] INFO: Slab 0xffffea0019c36f00 objects=33 used=33 fp=0x          (null) flags=0x12fffff80004080
> [ 3400.024040] INFO: Object 0xffff880670dbc000 @offset=0 fp=0x00007f6bff4e7000
> [ 3400.024040]
> [ 3400.024040] Object ffff880670dbc000: 00 70 4e ff 6b 7f 00 00 00 90 80 ff 6b 7f 00 00  .pN.k.......k...
> [ 3400.024040] Object ffff880670dbc010: f0 e0 db 70 06 88 ff ff e0 03 80 02 18 88 ff ff  ...p............
> [ 3400.024040] Object ffff880670dbc020: 01 04 80 02 18 88 ff ff 00 00 00 00 00 00 00 00  ................
> [ 3400.024040] Object ffff880670dbc030: 00 00 00 00 00 00 00 00 00 70 ce 7e 3f 7f 00 00  .........p.~?...
> [ 3400.024040] Object ffff880670dbc040: 00 f0 83 a6 06 88 ff ff 25 00 00 00 00 00 00 80  ........%.......
> [ 3400.024040] Object ffff880670dbc050: 73 00 10 08 00 00 00 00 00 00 00 00 00 00 00 00  s...............
> [ 3400.024040] Object ffff880670dbc060: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
> [ 3400.024040] Object ffff880670dbc070: 00 00 00 00 00 00 00 00 90 dc cf a4 06 88 ff ff  ................
> [ 3400.024040] Object ffff880670dbc080: 10 d8 cf a4 06 88 ff ff b8 d1 ee 24 08 88 ff ff  ...........$....
> [ 3400.024040] Object ffff880670dbc090: 00 00 00 00 00 00 00 00 e7 f4 bf f6 07 00 00 00  ................
> [ 3400.024040] Object ffff880670dbc0a0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
> [ 3400.024040] Object ffff880670dbc0b0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00  ................
> [ 3400.024040] CPU: 20 PID: 10752 Comm: run_vmtests Tainted: G    B           4.3.0-next-20151115-sasha-00042-g0f5ce29 #2641
> [ 3400.024040]  0000000000000014 0000000076bf224e ffff880ef1b7ee70 ffffffffabe427db
> [ 3400.024040]  ffff880182342000 ffff880670dbc000 ffff880670dbc000 ffff880ef1b7eea0
> [ 3400.024040]  ffffffffaa792a7a ffff880182342000 ffffea0019c36f00 ffff880670dbc000
> [ 3400.024040] Call Trace:
> [ 3400.024040] dump_stack (lib/dump_stack.c:52)
> [ 3400.024040] print_trailer (mm/slub.c:655)
> [ 3400.024040] object_err (mm/slub.c:662)
> [ 3400.024040] kasan_report_error (mm/kasan/report.c:138 mm/kasan/report.c:236)
> [ 3400.024040] __asan_report_load8_noabort (mm/kasan/report.c:280)
> [ 3400.024040] unfreeze_page (mm/huge_memory.c:3062 mm/huge_memory.c:3099)
> [ 3400.024040] split_huge_page_to_list (include/linux/compiler.h:218 mm/huge_memory.c:3208 mm/huge_memory.c:3291)
> [ 3400.024040] deferred_split_scan (mm/huge_memory.c:3378)
> [ 3400.024040] shrink_slab (mm/vmscan.c:352 mm/vmscan.c:444)
> [ 3400.024040] shrink_zone (mm/vmscan.c:2444)
> [ 3400.024040] do_try_to_free_pages (mm/vmscan.c:2595 mm/vmscan.c:2645)
> [ 3400.024040] try_to_free_pages (mm/vmscan.c:2853)
> [ 3400.024040] __alloc_pages_nodemask (mm/page_alloc.c:2864 mm/page_alloc.c:2882 mm/page_alloc.c:3150 mm/page_alloc.c:3261)
> [ 3400.024040] alloc_fresh_huge_page (include/linux/gfp.h:415 include/linux/gfp.h:428 mm/hugetlb.c:1330 mm/hugetlb.c:1348)
> [ 3400.024040] __nr_hugepages_store_common (include/linux/spinlock.h:302 mm/hugetlb.c:2164 mm/hugetlb.c:2279)
> [ 3400.024040] hugetlb_sysctl_handler_common (mm/hugetlb.c:2784)
> [ 3400.024040] hugetlb_sysctl_handler (mm/hugetlb.c:2796)
> [ 3400.024040] proc_sys_call_handler (fs/proc/proc_sysctl.c:543)
> [ 3400.024040] proc_sys_write (fs/proc/proc_sysctl.c:562)
> [ 3400.024040] __vfs_write (fs/read_write.c:489)
> [ 3400.024040] vfs_write (fs/read_write.c:538)
> [ 3400.024040] SyS_write (fs/read_write.c:585 fs/read_write.c:577)
> [ 3400.024040] tracesys_phase2 (arch/x86/entry/entry_64.S:273)
> [ 3400.024040] Memory state around the buggy address:
> [ 3400.024040]  ffff880670dbbf80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> [ 3400.024040]  ffff880670dbc000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
> [ 3400.024040] >ffff880670dbc080: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
> [ 3400.024040]                                            ^
> [ 3400.024040]  ffff880670dbc100: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
> [ 3400.024040]  ffff880670dbc180: fc fc fc fc fc fc fc fc fc fc fc fc fc fc 00 00
> [ 3400.024040] ==================================================================

Hm. This looks like THP leak. I fixed one with this patch:

http://lkml.kernel.org/g/1447236567-68751-1-git-send-email-kirill.shutemov@xxxxxxxxxxxxxxx

It's in -mm tree, but there wasn't any releases since it's applied. It's
not in -next for this reason.

There's one more patch with the same status:

http://lkml.kernel.org/g/1447236557-68682-1-git-send-email-kirill.shutemov@xxxxxxxxxxxxxxx

There's also one patch I've asked Minchan Kim to test. I'm not yet sure
it's correct:

http://lkml.kernel.org/g/20151117093213.GA16243@xxxxxxxxxxxxxxxxxx

-- 
 Kirill A. Shutemov
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/