Re: KMSAN: kernel-infoleak in copy_page_to_iter (2)

From: Alexander Potapenko
Date: Wed Jan 02 2019 - 05:10:08 EST

On Wed, Dec 19, 2018 at 2:23 PM Jens Axboe <axboe@xxxxxxxxx> wrote:
>
> On 12/19/18 3:23 AM, Alexander Potapenko wrote:
> > On Thu, Sep 13, 2018 at 11:23 AM Alexander Potapenko <glider@xxxxxxxxxx> wrote:
> >>
> >> On Thu, Sep 13, 2018 at 11:18 AM syzbot
> >> <syzbot+2dcfeaf8cb49b05e8f1a@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote:
> >>>
> >>> Hello,
> >>>
> >>> syzbot found the following crash on:
> >>>
> >>> HEAD commit: 123906095e30 kmsan: introduce kmsan_interrupt_enter()/kmsa..
> >>> git tree: https://github.com/google/kmsan.git/master
> >>> console output: https://syzkaller.appspot.com/x/log.txt?x=1249fcb8400000
> >>> kernel config: https://syzkaller.appspot.com/x/.config?x=848e40757852af3e
> >>> dashboard link: https://syzkaller.appspot.com/bug?extid=2dcfeaf8cb49b05e8f1a
> >>> compiler: clang version 7.0.0 (trunk 334104)
> >>> syz repro: https://syzkaller.appspot.com/x/repro.syz?x=116ef050400000
> >>> C reproducer: https://syzkaller.appspot.com/x/repro.c?x=122870ff800000
> >>>
> >>> IMPORTANT: if you fix the bug, please add the following tag to the commit:
> >>> Reported-by: syzbot+2dcfeaf8cb49b05e8f1a@xxxxxxxxxxxxxxxxxxxxxxxxx
> >>>
> >>> random: sshd: uninitialized urandom read (32 bytes read)
> >>> ==================================================================
> >>> BUG: KMSAN: kernel-infoleak in copyout lib/iov_iter.c:140 [inline]
> >>> BUG: KMSAN: kernel-infoleak in copy_page_to_iter_iovec lib/iov_iter.c:212
> >>> [inline]
> >>> BUG: KMSAN: kernel-infoleak in copy_page_to_iter+0x754/0x1b70
> >>> lib/iov_iter.c:716
> >>> CPU: 0 PID: 4516 Comm: blkid Not tainted 4.17.0+ #9
> >>> Hardware name: Google Google Compute Engine/Google Compute Engine, BIOS
> >>> Google 01/01/2011
> >>> Call Trace:
> >>> __dump_stack lib/dump_stack.c:77 [inline]
> >>> dump_stack+0x185/0x1d0 lib/dump_stack.c:113
> >>> kmsan_report+0x188/0x2a0 mm/kmsan/kmsan.c:1125
> >>> kmsan_internal_check_memory+0x17e/0x1f0 mm/kmsan/kmsan.c:1238
> >>> kmsan_copy_to_user+0x7a/0x160 mm/kmsan/kmsan.c:1261
> >>> copyout lib/iov_iter.c:140 [inline]
> >>> copy_page_to_iter_iovec lib/iov_iter.c:212 [inline]
> >>> copy_page_to_iter+0x754/0x1b70 lib/iov_iter.c:716
> >>> generic_file_buffered_read mm/filemap.c:2185 [inline]
> >>> generic_file_read_iter+0x2ef8/0x44d0 mm/filemap.c:2362
> >>> blkdev_read_iter+0x20d/0x280 fs/block_dev.c:1930
> >>> call_read_iter include/linux/fs.h:1778 [inline]
> >>> new_sync_read fs/read_write.c:406 [inline]
> >>> __vfs_read+0x775/0x9d0 fs/read_write.c:418
> >>> vfs_read+0x36c/0x6b0 fs/read_write.c:452
> >>> ksys_read fs/read_write.c:578 [inline]
> >>> __do_sys_read fs/read_write.c:588 [inline]
> >>> __se_sys_read fs/read_write.c:586 [inline]
> >>> __x64_sys_read+0x1bf/0x3e0 fs/read_write.c:586
> >>> do_syscall_64+0x15b/0x230 arch/x86/entry/common.c:287
> >>> entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >>> RIP: 0033:0x7fdeff68f310
> >>> RSP: 002b:00007ffe999660b8 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
> >>> RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fdeff68f310
> >>> RDX: 0000000000000100 RSI: 0000000001e78df8 RDI: 0000000000000003
> >>> RBP: 0000000001e78dd0 R08: 0000000000000028 R09: 0000000001680000
> >>> R10: 0000000000000000 R11: 0000000000000246 R12: 0000000001e78030
> >>> R13: 0000000000000100 R14: 0000000001e78080 R15: 0000000001e78de8
> >>>
> >>> Uninit was created at:
> >>> kmsan_save_stack_with_flags mm/kmsan/kmsan.c:282 [inline]
> >>> kmsan_alloc_meta_for_pages+0x161/0x3a0 mm/kmsan/kmsan.c:819
> >>> kmsan_alloc_page+0x82/0xe0 mm/kmsan/kmsan.c:889
> >>> __alloc_pages_nodemask+0xf7b/0x5cc0 mm/page_alloc.c:4402
> >>> alloc_pages_current+0x6b1/0x970 mm/mempolicy.c:2093
> >>> alloc_pages include/linux/gfp.h:494 [inline]
> >>> __page_cache_alloc+0x95/0x320 mm/filemap.c:946
> >>> pagecache_get_page+0x52b/0x1450 mm/filemap.c:1577
> >>> grab_cache_page_write_begin+0x10d/0x190 mm/filemap.c:3089
> >>> block_write_begin+0xf9/0x3a0 fs/buffer.c:2068
> >>> blkdev_write_begin+0xf5/0x110 fs/block_dev.c:584
> >>> generic_perform_write+0x438/0x9d0 mm/filemap.c:3139
> >>> __generic_file_write_iter+0x43b/0xa10 mm/filemap.c:3264
> >>> blkdev_write_iter+0x3a8/0x5f0 fs/block_dev.c:1910
> >>> do_iter_readv_writev+0x81c/0xa20 include/linux/fs.h:1778
> >>> do_iter_write+0x30d/0xd50 fs/read_write.c:959
> >>> vfs_writev fs/read_write.c:1004 [inline]
> >>> do_writev+0x3be/0x820 fs/read_write.c:1039
> >>> __do_sys_writev fs/read_write.c:1112 [inline]
> >>> __se_sys_writev fs/read_write.c:1109 [inline]
> >>> __x64_sys_writev+0xe1/0x120 fs/read_write.c:1109
> >>> do_syscall_64+0x15b/0x230 arch/x86/entry/common.c:287
> >>> entry_SYSCALL_64_after_hwframe+0x44/0xa9
> >>>
> >>> Bytes 4-255 of 256 are uninitialized
> >>> Memory access starts at ffff8801b9903000
> >>> ==================================================================
> >> This particular report was caused by the repro program writing a byte
> >> to /dev/nullb0 and /sbin/blkid reading from that device in the
> >> background.
> >> But it turns out that simply running `cat /dev/nullb0` already prints
> >> uninitialized kernel memory.
> >> Is this the intended behavior of the null block driver?
> > A friendly ping, this bug is still reproducible on syzbot.
>
> Does this fix it?
There must be something wrong with my tool, as it stops reporting this
bug when I apply your patch.
However when I run `cat /dev/nullb0 | strings` and wait long enough I
start seeing meaningful strings (file names, env dumps etc.)
I suspect this is still unexpected, right?
>
> diff --git a/drivers/block/null_blk_main.c b/drivers/block/null_blk_main.c
> index 62c9654b9ce8..08808c572692 100644
> --- a/drivers/block/null_blk_main.c
> +++ b/drivers/block/null_blk_main.c
> @@ -655,7 +655,7 @@ static struct nullb_page *null_alloc_page(gfp_t gfp_flags)
> if (!t_page)
> goto out;
>
> - t_page->page = alloc_pages(gfp_flags, 0);
> + t_page->page = alloc_pages(gfp_flags | __GFP_ZERO, 0);
> if (!t_page->page)
> goto out_freepage;
>
>
> --
> Jens Axboe
>


--
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-StraÃe, 33
80636 MÃnchen

GeschÃftsfÃhrer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg