mm/thp: unmap_mapping_page() to fix THP truncate_cleanup_page()
authorHugh Dickins <hughd@google.com>
Wed, 16 Jun 2021 01:24:03 +0000 (18:24 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 16 Jun 2021 16:24:42 +0000 (09:24 -0700)
commit22061a1ffabdb9c3385de159c5db7aac3a4df1cc
tree5b7d0f2d9bc7b332e52c7707edd077e585ce52c8
parent31657170deaf1d8d2f6a1955fbc6fa9d228be036
mm/thp: unmap_mapping_page() to fix THP truncate_cleanup_page()

There is a race between THP unmapping and truncation, when truncate sees
pmd_none() and skips the entry, after munmap's zap_huge_pmd() cleared
it, but before its page_remove_rmap() gets to decrement
compound_mapcount: generating false "BUG: Bad page cache" reports that
the page is still mapped when deleted.  This commit fixes that, but not
in the way I hoped.

The first attempt used try_to_unmap(page, TTU_SYNC|TTU_IGNORE_MLOCK)
instead of unmap_mapping_range() in truncate_cleanup_page(): it has
often been an annoyance that we usually call unmap_mapping_range() with
no pages locked, but there apply it to a single locked page.
try_to_unmap() looks more suitable for a single locked page.

However, try_to_unmap_one() contains a VM_BUG_ON_PAGE(!pvmw.pte,page):
it is used to insert THP migration entries, but not used to unmap THPs.
Copy zap_huge_pmd() and add THP handling now? Perhaps, but their TLB
needs are different, I'm too ignorant of the DAX cases, and couldn't
decide how far to go for anon+swap.  Set that aside.

The second attempt took a different tack: make no change in truncate.c,
but modify zap_huge_pmd() to insert an invalidated huge pmd instead of
clearing it initially, then pmd_clear() between page_remove_rmap() and
unlocking at the end.  Nice.  But powerpc blows that approach out of the
water, with its serialize_against_pte_lookup(), and interesting pgtable
usage.  It would need serious help to get working on powerpc (with a
minor optimization issue on s390 too).  Set that aside.

Just add an "if (page_mapped(page)) synchronize_rcu();" or other such
delay, after unmapping in truncate_cleanup_page()? Perhaps, but though
that's likely to reduce or eliminate the number of incidents, it would
give less assurance of whether we had identified the problem correctly.

This successful iteration introduces "unmap_mapping_page(page)" instead
of try_to_unmap(), and goes the usual unmap_mapping_range_tree() route,
with an addition to details.  Then zap_pmd_range() watches for this
case, and does spin_unlock(pmd_lock) if so - just like
page_vma_mapped_walk() now does in the PVMW_SYNC case.  Not pretty, but
safe.

Note that unmap_mapping_page() is doing a VM_BUG_ON(!PageLocked) to
assert its interface; but currently that's only used to make sure that
page->mapping is stable, and zap_pmd_range() doesn't care if the page is
locked or not.  Along these lines, in invalidate_inode_pages2_range()
move the initial unmap_mapping_range() out from under page lock, before
then calling unmap_mapping_page() under page lock if still mapped.

Link: https://lkml.kernel.org/r/a2a4a148-cdd8-942c-4ef8-51b77f643dbe@google.com
Fixes: fc127da085c2 ("truncate: handle file thp")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/mm.h
mm/memory.c
mm/truncate.c