Re: [PATCH V4] scsi_debugfs: fix crash in scsi_show_rq()

[Ming Lei <ming.lei@redhat.com>]

Hi James,

On Mon, Nov 13, 2017 at 10:55:52AM -0800, James Bottomley wrote:
> On Sat, 2017-11-11 at 10:43 +0800, Ming Lei wrote:
> > On Fri, Nov 10, 2017 at 08:51:58AM -0800, James Bottomley wrote:
> > > 
> > > On Fri, 2017-11-10 at 17:01 +0800, Ming Lei wrote:
> > > > 
> > > > cmd->cmnd can be allocated/freed dynamically in case of
> > > > T10_PI_TYPE2_PROTECTION, so we should check it in scsi_show_rq()
> > > > because this request may have been freed already here, and cmd-
> > > > >cmnd has been set as null.
> > > > 
> > > > We choose to accept read-after-free and dump request data as far
> > > > as possible.
> > > > 
> > > > This patch fixs the following kernel crash when dumping request
> > > > via block's debugfs interface:
> > > > 
> > > > [��252.962045] BUG: unable to handle kernel NULL pointer
> > > > dereference
> > > > at�����������(null)
> > > > [��252.963007] IP: scsi_format_opcode_name+0x1a/0x1c0
> > > > [��252.963007] PGD 25e75a067 P4D 25e75a067 PUD 25e75b067 PMD 0
> > > > [��252.963007] Oops: 0000 [#1] PREEMPT SMP
> > > > [��252.963007] Dumping ftrace buffer:
> > > > [��252.963007]����(ftrace buffer empty)
> > > > [��252.963007] Modules linked in: scsi_debug ebtable_filter
> > > > ebtables
> > > > ip6table_filter ip6_tables xt_CHECKSUM iptable_mangle
> > > > ipt_MASQUERADE
> > > > nf_nat_masquerade_ipv4 iptable_nat nf_conntrack_ipv4
> > > > nf_defrag_ipv4
> > > > nf_nat_ipv4 nf_nat nf_conntrack libcrc32c bridge stp llc
> > > > iptable_filter fuse ip_tables sd_mod sg mptsas mptscsih mptbase
> > > > crc32c_intel ahci libahci nvme serio_raw scsi_transport_sas
> > > > libata
> > > > lpc_ich nvme_core virtio_scsi binfmt_misc dm_mod iscsi_tcp
> > > > libiscsi_tcp libiscsi scsi_transport_iscsi null_blk configs
> > > > [��252.963007] CPU: 1 PID: 1881 Comm: cat Not tainted 4.14.0-
> > > > rc2.blk_mq_io_hang+ #516
> > > > [��252.963007] Hardware name: QEMU Standard PC (Q35 + ICH9,
> > > > 2009),
> > > > BIOS 1.9.3-1.fc25 04/01/2014
> > > > [��252.963007] task: ffff88025e6f6000 task.stack:
> > > > ffffc90001bd0000
> > > > [��252.963007] RIP: 0010:scsi_format_opcode_name+0x1a/0x1c0
> > > > [��252.963007] RSP: 0018:ffffc90001bd3c50 EFLAGS: 00010286
> > > > [��252.963007] RAX: 00000000ffff4843 RBX: 0000000000000050 RCX:
> > > > 0000000000000000
> > > > [��252.963007] RDX: 0000000000000000 RSI: 0000000000000050 RDI:
> > > > ffffc90001bd3cd8
> > > > [��252.963007] RBP: ffffc90001bd3c88 R08: 0000000000001000 R09:
> > > > 0000000000000000
> > > > [��252.963007] R10: ffff880275134000 R11: ffff88027513406c R12:
> > > > 0000000000000050
> > > > [��252.963007] R13: ffffc90001bd3cd8 R14: 0000000000000000 R15:
> > > > 0000000000000000
> > > > [��252.963007] FS:��00007f4d11762700(0000)
> > > > GS:ffff88027fc40000(0000)
> > > > knlGS:0000000000000000
> > > > [��252.963007] CS:��0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > [��252.963007] CR2: 0000000000000000 CR3: 000000025e789003 CR4:
> > > > 00000000003606e0
> > > > [��252.963007] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
> > > > 0000000000000000
> > > > [��252.963007] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
> > > > 0000000000000400
> > > > [��252.963007] Call Trace:
> > > > [��252.963007]��__scsi_format_command+0x27/0xc0
> > > > [��252.963007]��scsi_show_rq+0x5c/0xc0
> > > > [��252.963007]��? seq_printf+0x4e/0x70
> > > > [��252.963007]��? blk_flags_show+0x5b/0xf0
> > > > [��252.963007]��__blk_mq_debugfs_rq_show+0x116/0x130
> > > > [��252.963007]��blk_mq_debugfs_rq_show+0xe/0x10
> > > > [��252.963007]��seq_read+0xfe/0x3b0
> > > > [��252.963007]��? __handle_mm_fault+0x631/0x1150
> > > > [��252.963007]��full_proxy_read+0x54/0x90
> > > > [��252.963007]��__vfs_read+0x37/0x160
> > > > [��252.963007]��? security_file_permission+0x9b/0xc0
> > > > [��252.963007]��vfs_read+0x96/0x130
> > > > [��252.963007]��SyS_read+0x55/0xc0
> > > > [��252.963007]��entry_SYSCALL_64_fastpath+0x1a/0xa5
> > > > [��252.963007] RIP: 0033:0x7f4d1127e9b0
> > > > [��252.963007] RSP: 002b:00007ffd27082568 EFLAGS: 00000246
> > > > ORIG_RAX:
> > > > 0000000000000000
> > > > [��252.963007] RAX: ffffffffffffffda RBX: 00007f4d1154bb20 RCX:
> > > > 00007f4d1127e9b0
> > > > [��252.963007] RDX: 0000000000020000 RSI: 00007f4d115a7000 RDI:
> > > > 0000000000000003
> > > > [��252.963007] RBP: 0000000000021010 R08: ffffffffffffffff R09:
> > > > 0000000000000000
> > > > [��252.963007] R10: 000000000000037b R11: 0000000000000246 R12:
> > > > 0000000000022000
> > > > [��252.963007] R13: 00007f4d1154bb78 R14: 0000000000001000 R15:
> > > > 0000000000020000
> > > > [��252.963007] Code: c6 e8 1b ca 24 00 eb 8c e8 74 2c ae ff 0f 1f
> > > > 40
> > > > 00 0f 1f 44 00 00 55 48 89 e5 41 56 41 55 41 54 53 49 89 fd 49 89
> > > > f4
> > > > 48 83 ec 18 <44> 0f b6 32 48 c7 45 c8 00 00 00 00 65 48 8b 04 25
> > > > 28
> > > > 00 00 00
> > > > [��252.963007] RIP: scsi_format_opcode_name+0x1a/0x1c0 RSP:
> > > > ffffc90001bd3c50
> > > > [��252.963007] CR2: 0000000000000000
> > > > [��252.963007] ---[ end trace 83c5bddfbaa6573c ]---
> > > > [��252.963007] Kernel panic - not syncing: Fatal exception
> > > > [��252.963007] Dumping ftrace buffer:
> > > > [��252.963007]����(ftrace buffer empty)
> > > > [��252.963007] Kernel Offset: disabled
> > > > [��252.963007] ---[ end Kernel panic - not syncing: Fatal
> > > > exception
> > > > 
> > > > Fixes: 0eebd005dd07(scsi: Implement blk_mq_ops.show_rq())
> > > > Cc: Bart Van Assche <bart.vanassche@sandisk.com>
> > > > Cc: Omar Sandoval <osandov@fb.com>
> > > > Cc: Martin K. Petersen <martin.petersen@oracle.com>
> > > > Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
> > > > Cc: Hannes Reinecke <hare@suse.com>
> > > > Signed-off-by: Ming Lei <ming.lei@redhat.com>
> > > > ---
> > > > V2:
> > > > ���- fix typo
> > > > V3:
> > > > ���- prefer to dump data and accept read-after-free
> > > > ���- add some comment
> > > > V4:
> > > > ���- read the two variable into local variable first, for
> > > > avoiding
> > > > 	free between check and passing to __scsi_format_command().
> > > > 	
> > > > 
> > > > �drivers/scsi/scsi_debugfs.c | 14 +++++++++++++-
> > > > �1 file changed, 13 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/scsi/scsi_debugfs.c
> > > > b/drivers/scsi/scsi_debugfs.c
> > > > index 5e9755008aed..f8437c456458 100644
> > > > --- a/drivers/scsi/scsi_debugfs.c
> > > > +++ b/drivers/scsi/scsi_debugfs.c
> > > > @@ -8,8 +8,20 @@ void scsi_show_rq(struct seq_file *m, struct
> > > > request
> > > > *rq)
> > > > �	struct scsi_cmnd *cmd = container_of(scsi_req(rq),
> > > > typeof(*cmd), req);
> > > > �	int msecs = jiffies_to_msecs(jiffies - cmd-
> > > > > 
> > > > > jiffies_at_alloc);
> > > > �	char buf[80];
> > > > +	unsigned char *cdb = cmd->cmnd;
> > > > +	unsigned short cdb_len = cmd->cmd_len;
> > > > 
> > > > �
> > > > -	__scsi_format_command(buf, sizeof(buf), cmd->cmnd, cmd-
> > > > > 
> > > > > cmd_len);
> > > > +	/*
> > > > +	�* This rq may have been freed, so don't be surprised if
> > > > +	�* read-after-free is reported.
> > > 
> > > I said fix this in sd.c:sd_uninit_command() by setting SCpnt->cmnd
> > > to NULL before calling mempool_free().
> > 
> > Hi James,
> > 
> > That can't avoid the use-after-free in both reality and theory.
> 
> I believe I already gave you the theory of how to avoid it.
> 
> > When 'cdb = cmd->cmnd' is executed in CPU1, 'cbd' is a valid pointer,
> > just after this instruction is done, this request is completed in
> > CPU2, and mempool_free() is run to complete this buffer.
> > 
> > Firstly,��both preemption and IRQ can happen inside
> > __scsi_format_command(),
> > when this function is scheduled to rerun, this buffer is freed since
> > mempool_free() has completed already, which is run in irq context.�
> 
> so disable irq's before taking the cmnd pointer copy ...

It isn't enough.

> 
> > Secondly, the use-after-free can't be avoided too if we disable irq
> > when checking and reading the variable. See following:
> > 
> > mempool_free(cmd->cmnd) is often run much quick because we complete
> > request on same CPU with the submission CPU(the CPU for allocation of
> > cmd->cmnd).
> 
> only in the add element case, which won't cause a use after free from

I don't understand why that is only in add-element case, could you explain
it a bit?

For slub/slab..., I believe there is much optimization in case that the
allocation/free is done on one local CPU, which should be quicker than
memory read from a remote NUMA node.

> any tool's point of view. �If you want greater assurance, take the

For example of KASAN[1], each byte of memory is covered by its shadow
memory. Once pool->free() in mempool_free(cmd->cmnd) is run before
__scsi_format_command() returns, the use-after-free is triggered.

And there are several access to this buffer in __scsi_format_command().
KASAN checks if each access is safe before the actual access, and the
check is added by GCC, see below words from [1]:

	Compile-time instrumentation used for checking memory accesses. Compiler inserts
	function calls (__asan_load*(addr), __asan_store*(addr)) before each memory
	access of size 1, 2, 4, 8 or 16. These functions check whether memory access is
	valid or not by checking corresponding shadow memory.


[1] Documentation/dev-tools/kasan.rst

> pool->lock (with encapsulation so we're not poking around in mempool
> internals).

The pool->lock isn't needed in fast path of 'pool->free()', and it is required
only when 'if (unlikely(pool->curr_nr < pool->min_nr))' is true, so this way
doesn't work too.

> 
> > So from CPU1's review, cmd->cmnd is in a remote NUMA node,
> > __scsi_format_command() is executed much slower than mempool_free().
> > So when mempool_free() returns, __scsi_format_command() may not
> > fetched the buffer in L1 cache yet, then use-after-free
> > is still triggered.
> > 
> > That is why I say this use-after-free is inevitable no matter
> > 'setting SCpnt->cmnd to NULL before calling mempool_free()' or not.
> 
> The bottom line is that there are several creative ways around this but
> the proposed code is currently broken and simply putting a comment in
> saying so doesn't make it acceptable.

As I explained above, I didn't see one really workable way. Or please
correct it if I am wrong.

Thanks,
Ming