kernel BUG at fs/direct-io.c:211! in next-20160930

kernel BUG at fs/direct-io.c:211! in next-20160930

[Joseph Bisch]

While fuzzing next-20160930 with syzkaller I encountered the following:

kernel BUG at fs/direct-io.c:211!
invalid opcode: 0000 [#1] SMP
Dumping ftrace buffer:
   (ftrace buffer empty)
Modules linked in:
CPU: 1 PID: 2488 Comm: syz-executor Not tainted 4.8.0-rc8-next-20160930 #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Debian-1.8.2-1 04/01/2014
task: ffff880436b9d440 task.stack: ffffc90005d70000
RIP: 0010:[<ffffffff812433e1>]  [<     inline     >] dio_get_page fs/direct-io.c:211
RIP: 0010:[<ffffffff812433e1>]  [<     inline     >] do_direct_IO fs/direct-io.c:930
RIP: 0010:[<ffffffff812433e1>]  [<ffffffff812433e1>] do_blockdev_direct_IO+0x2481/0x3b70 fs/direct-io.c:1270
RSP: 0018:ffffc90005d73930  EFLAGS: 00010283
RAX: ffffffff812433e1 RBX: 0000000000000000 RCX: ffffc9000b157000
RDX: 00000000000009ba RSI: 0000000000000000 RDI: ffffc90005d73c90
RBP: ffffc90005d73b48 R08: ffffc90005d73b10 R09: 0000000000000000
R10: 0000000000000001 R11: 0000000000000001 R12: 0000000000000000
R13: 000000000007ffff R14: dead000000000100 R15: ffff880234528040
FS:  00007f4b22781700(0000) GS:ffff880237d00000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00000000006d0000 CR3: 00000004359dd000 CR4: 00000000000006e0
Stack:
 ffffc90005d73968 000000000000000c ffff880436b9d440 ffff880233d29000
 ffff880436b9d440 0000000000001000 ffff880436b9d440 0000000000001000
 00001000000044e3 000000000001e211 ffff880436b9d440 fffffffffffff000
Call Trace:
 [<ffffffff81244b39>] __blockdev_direct_IO+0x69/0x80 fs/direct-io.c:1356
 [<     inline     >] ext4_direct_IO_read fs/ext4/inode.c:3553
 [<ffffffff812a7906>] ext4_direct_IO+0x366/0x8f0 fs/ext4/inode.c:3588
 [<ffffffff8117d786>] generic_file_read_iter+0x946/0xa90 mm/filemap.c:1922
 [<ffffffff81231a6b>] generic_file_splice_read+0xeb/0x1f0 fs/splice.c:313
 [<ffffffff812311d5>] do_splice_to+0x95/0xc0 fs/splice.c:908
 [<ffffffff812312c5>] splice_direct_to_actor+0xc5/0x280 fs/splice.c:980
 [<ffffffff81231530>] do_splice_direct+0xb0/0xf0 fs/splice.c:1089
 [<ffffffff811f4043>] do_sendfile+0x213/0x440 fs/read_write.c:1372
 [<     inline     >] SYSC_sendfile64 fs/read_write.c:1427
 [<ffffffff811f504e>] SyS_sendfile64+0x6e/0xd0 fs/read_write.c:1419
 [<ffffffff81bc7da0>] entry_SYSCALL_64_fastpath+0x13/0x94
Code: 00 48 8d 43 ff 31 db 25 ff 0f 00 00 48 83 c0 01 48 89 84 24 e8 01 00 00 e8 dd 81 ee ff 41 39 dc 0f 85 52 df ff ff e8 cf 81 ee ff <0f> 0b e8 c8 81 ee ff be 0f 00 00 00 48 c7 c7 6c 07 da 81 e8 e7 
RIP  [<     inline     >] dio_get_page fs/direct-io.c:211
RIP  [<     inline     >] do_direct_IO fs/direct-io.c:930
RIP  [<ffffffff812433e1>] do_blockdev_direct_IO+0x2481/0x3b70 fs/direct-io.c:1270
 RSP <ffffc90005d73930>
---[ end trace 8d37bff5680e79fa ]---
Kernel panic - not syncing: Fatal exception
Dumping ftrace buffer:
   (ftrace buffer empty)
Kernel Offset: disabled

The following program was generated by syz-repro and reproduces the crash:

http://pastebin.com/vE6cXzEg

-- 
Joe Bisch
HPE Linux, Hewlett Packard Enterprise

Re: kernel BUG at fs/direct-io.c:211! in next-20160930

[Al Viro]

On Wed, Oct 05, 2016 at 03:03:46PM -0600, Joseph Bisch wrote:
> While fuzzing next-20160930 with syzkaller I encountered the following:
> 
> kernel BUG at fs/direct-io.c:211!
> invalid opcode: 0000 [#1] SMP

Should've been fixed in current -next - it's handling of iov_iter_get_pages()
n ITER_PIPE iterators when they get full; the things to watch for are
a) __pipe_get_pages() containing
        size_t n = push_pipe(i, maxsize, &idx, start);
        if (!n)
                return -EFAULT;
and
b) "consistent treatment of EFAULT on O_DIRECT read/write" applied in
fs/direct-io.c