Patch 1/2 had been sent about weeks ago, the original link here: https://patchwork.kernel.org/project/linux-mm/patch/20211011022241.97072-2-r... It seems to be ignored, and send out again here.
Patch 2/2 is sent out first time. And Patch 1/2 and 2/2 both to fix different bugs for commit eb6ecbed0aa2 ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs").
In addition, I find the stable version (I had check 5.14) has contained above commit (eb6ecbed0aa2), so I will add "cc stable mail list" when sending Patches.
Thanks!
Rongwei Wang (2): mm, thp: lock filemap when truncating page cache mm, thp: fix incorrect unmap behavior for private pages
fs/open.c | 16 ++++++++++++++-- 1 file changed, 14 insertions(+), 2 deletions(-)
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=--) pc : truncate_inode_page+0x64/0x70 lr : truncate_inode_page+0x64/0x70 sp : ffff80001b60b900 x29: ffff80001b60b900 x28: 00000000000007ff x27: ffff80001b60b9a0 x26: 0000000000000000 x25: 000000000000000f x24: ffff80001b60b9a0 x23: ffff80001b60ba18 x22: ffff0001e0999ea8 x21: ffff0000c21db300 x20: ffffffffffffffff x19: fffffe001310ffc0 x18: 0000000000000020 x17: 0000000000000000 x16: 0000000000000000 x15: ffff0000c21db960 x14: 3030306666666620 x13: 6666666666666666 x12: 3130303030303030 x11: ffff8000117b69b8 x10: 00000000ffff8000 x9 : ffff80001012690c x8 : 0000000000000000 x7 : ffff8000114f69b8 x6 : 0000000000017ffd x5 : ffff0007fffbcbc8 x4 : ffff80001b60b5c0 x3 : 0000000000000001 x2 : 0000000000000000 x1 : 0000000000000000 x0 : 0000000000000000 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) ---[ end trace f70cdb42cb7c2d42 ]--- Kernel panic - not syncing: Oops - BUG: Fatal exception
This patch mainly to lock filemap when one enter truncate_pagecache(), avoiding truncating the same page cache concurrently.
Fixes: eb6ecbed0aa2 ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs") Cc: stable@vger.kernel.org Suggested-by: Matthew Wilcox (Oracle) willy@infradead.org Tested-by: Song Liu song@kernel.org Signed-off-by: Xu Yu xuyu@linux.alibaba.com Signed-off-by: Rongwei Wang rongwei.wang@linux.alibaba.com --- fs/open.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/fs/open.c b/fs/open.c index a7f6cab81267..e73bf88e5060 100644 --- a/fs/open.c +++ b/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;
When truncating pagecache on file THP, the private pages of a process should not be unmapped mapping. This incorrect behavior on a dynamic shared libraries which will cause related processes to happen core dump.
A simple test for a DSO (Prerequisite is the DSO mapped in file THP):
int main(int argc, char *argv[]) { int fd;
fd = open(argv[1], O_WRONLY); if (fd < 0) { perror("open"); }
close(fd); return 0; }
The test only to open a target DSO, and do nothing. But this operation will lead one or more process to happen core dump. This patch mainly to fix this bug.
Fixes: eb6ecbed0aa2 ("mm, thp: relax the VM_DENYWRITE constraint on file-backed THPs") Cc: stable@vger.kernel.org Tested-by: Xu Yu xuyu@linux.alibaba.com Signed-off-by: Rongwei Wang rongwei.wang@linux.alibaba.com --- fs/open.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-)
diff --git a/fs/open.c b/fs/open.c index e73bf88e5060..f732fb94600c 100644 --- a/fs/open.c +++ b/fs/open.c @@ -857,8 +857,17 @@ static int do_dentry_open(struct file *f, */ smp_mb(); if (filemap_nr_thps(inode->i_mapping)) { + struct address_space *mapping = inode->i_mapping; + filemap_invalidate_lock(inode->i_mapping); - truncate_pagecache(inode, 0); + /* + * unmap_mapping_range just need to be called once + * here, because the private pages is not need to be + * unmapped mapping (e.g. data segment of dynamic + * shared libraries here). + */ + unmap_mapping_range(mapping, 0, 0, 0); + truncate_inode_pages(mapping, 0); filemap_invalidate_unlock(inode->i_mapping); } }
linux-stable-mirror@lists.linaro.org
-
Rongwei Wang