mm, thp: lock filemap when truncating page cache

Patch series "fix two bugs for file THP".

This patch (of 2):

Transparent huge page has supported read-only non-shmem files.  The
file- backed THP is collapsed by khugepaged and truncated when written
(for shared libraries).

However, there is a race when multiple writers truncate the same page
cache concurrently.

In that case, subpage(s) of file THP can be revealed by find_get_entry
in truncate_inode_pages_range, which will trigger PageTail BUG_ON in
truncate_inode_page, as follows:

    page:000000009e420ff2 refcount:1 mapcount:0 mapping:0000000000000000 index:0x7ff pfn:0x50c3ff
    head:0000000075ff816d order:9 compound_mapcount:0 compound_pincount:0
    flags: 0x37fffe0000010815(locked|uptodate|lru|arch_1|head)
    raw: 37fffe0000000000 fffffe0013108001 dead000000000122 dead000000000400
    raw: 0000000000000001 0000000000000000 00000000ffffffff 0000000000000000
    head: 37fffe0000010815 fffffe001066bd48 ffff000404183c20 0000000000000000
    head: 0000000000000600 0000000000000000 00000001ffffffff ffff000c0345a000
    page dumped because: VM_BUG_ON_PAGE(PageTail(page))
    ------------[ cut here ]------------
    kernel BUG at mm/truncate.c:213!
    Internal error: Oops - BUG: 0 [#1] SMP
    Modules linked in: xfs(E) libcrc32c(E) rfkill(E) ...
    CPU: 14 PID: 11394 Comm: check_madvise_d Kdump: ...
    Hardware name: ECS, BIOS 0.0.0 02/06/2015
    pstate: 60400005 (nZCv daif +PAN -UAO -TCO BTYPE=--)
    Call trace:
     truncate_inode_page+0x64/0x70
     truncate_inode_pages_range+0x550/0x7e4
     truncate_pagecache+0x58/0x80
     do_dentry_open+0x1e4/0x3c0
     vfs_open+0x38/0x44
     do_open+0x1f0/0x310
     path_openat+0x114/0x1dc
     do_filp_open+0x84/0x134
     do_sys_openat2+0xbc/0x164
     __arm64_sys_openat+0x74/0xc0
     el0_svc_common.constprop.0+0x88/0x220
     do_el0_svc+0x30/0xa0
     el0_svc+0x20/0x30
     el0_sync_handler+0x1a4/0x1b0
     el0_sync+0x180/0x1c0
    Code: aa0103e0 900061e1 910ec021 9400d300 (d4210000)

This patch mainly to lock filemap when one enter truncate_pagecache(),
avoiding truncating the same page cache concurrently.

Link: https://lkml.kernel.org/r/20211025092134.18562-1-rongwei.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/20211025092134.18562-2-rongwei.wang@linux.alibaba.com
Fixes: eb6ecbed ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs")
Signed-off-by: Xu Yu <xuyu@linux.alibaba.com>
Signed-off-by: Rongwei Wang <rongwei.wang@linux.alibaba.com>
Suggested-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Tested-by: Song Liu <song@kernel.org>
Cc: Collin Fijalkovich <cfijalkovich@google.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: William Kucharski <william.kucharski@oracle.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

fs/open.c

@@ -856,8 +856,11 @@ static int do_dentry_open(struct file *f,
 		 * of THPs into the page cache will fail.
 		 */
 		smp_mb();
-		if (filemap_nr_thps(inode->i_mapping))
+		if (filemap_nr_thps(inode->i_mapping)) {
+			filemap_invalidate_lock(inode->i_mapping);
 			truncate_pagecache(inode, 0);
+			filemap_invalidate_unlock(inode->i_mapping);
+		}
 	}
 
 	return 0;