Message-ID: <20180712181846.GD13192@cisco.cisco.com>
Date:   Thu, 12 Jul 2018 12:18:46 -0600
From:   Tycho Andersen <tycho@...ho.ws>
To:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>
Cc:     Jiri Slaby <jslaby@...e.com>, linux-serial@...r.kernel.org,
        linux-kernel@...r.kernel.org, "Serge E . Hallyn" <serge@...lyn.com>
Subject: Re: [PATCH v4] uart: fix race between uart_put_char() and
 uart_shutdown()

On Thu, Jul 12, 2018 at 05:40:15PM +0200, Greg Kroah-Hartman wrote:
> On Thu, Jul 12, 2018 at 09:08:22AM -0600, Tycho Andersen wrote:
> > On Thu, Jul 12, 2018 at 05:04:38PM +0200, Greg Kroah-Hartman wrote:
> > > On Wed, Jul 11, 2018 at 10:07:44AM -0600, Tycho Andersen wrote:
> > > > +	if (uport)
> > > > +		spin_lock_irqsave(&uport->lock, flags);
> > > 
> > > That's the same thing as just calling uart_port_lock(), why aren't you
> > > doing that?
> > 
> > Because the compiler can't seem to "see" through the macros/ref calls,
> > and I get the warning I mentioned here if I use them:
> > 
> > https://lkml.org/lkml/2018/7/6/840
> 
> What horrible version of gcc are you using that give you that?  Don't
> open-code things just because of a broken compiler.

I've tried with both 7.3.0 and 5.4.0. I think the reason we see this
here but not elsewhere in the file is because there's an actual
function call (free_page()) in the critical section.

If we move that out, something like the below patch, it all works for
me.

Tycho


>From 532e82ceec67142b71c60ce74ce08c5339195a94 Mon Sep 17 00:00:00 2001
From: Tycho Andersen <tycho@...ho.ws>
Date: Mon, 4 Jun 2018 09:55:09 -0600
Subject: [PATCH] uart: fix race between uart_put_char() and uart_shutdown()
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

We have reports of the following crash:

    PID: 7 TASK: ffff88085c6d61c0 CPU: 1 COMMAND: "kworker/u25:0"
    #0 [ffff88085c6db710] machine_kexec at ffffffff81046239
    #1 [ffff88085c6db760] crash_kexec at ffffffff810fc248
    #2 [ffff88085c6db830] oops_end at ffffffff81008ae7
    #3 [ffff88085c6db860] no_context at ffffffff81050b8f
    #4 [ffff88085c6db8b0] __bad_area_nosemaphore at ffffffff81050d75
    #5 [ffff88085c6db900] bad_area_nosemaphore at ffffffff81050e83
    #6 [ffff88085c6db910] __do_page_fault at ffffffff8105132e
    #7 [ffff88085c6db9b0] do_page_fault at ffffffff8105152c
    #8 [ffff88085c6db9c0] page_fault at ffffffff81a3f122
    [exception RIP: uart_put_char+149]
    RIP: ffffffff814b67b5 RSP: ffff88085c6dba78 RFLAGS: 00010006
    RAX: 0000000000000292 RBX: ffffffff827c5120 RCX: 0000000000000081
    RDX: 0000000000000000 RSI: 000000000000005f RDI: ffffffff827c5120
    RBP: ffff88085c6dba98 R8: 000000000000012c R9: ffffffff822ea320
    R10: ffff88085fe4db04 R11: 0000000000000001 R12: ffff881059f9c000
    R13: 0000000000000001 R14: 000000000000005f R15: 0000000000000fba
    ORIG_RAX: ffffffffffffffff CS: 0010 SS: 0018
    #9 [ffff88085c6dbaa0] tty_put_char at ffffffff81497544
    #10 [ffff88085c6dbac0] do_output_char at ffffffff8149c91c
    #11 [ffff88085c6dbae0] __process_echoes at ffffffff8149cb8b
    #12 [ffff88085c6dbb30] commit_echoes at ffffffff8149cdc2
    #13 [ffff88085c6dbb60] n_tty_receive_buf_fast at ffffffff8149e49b
    #14 [ffff88085c6dbbc0] __receive_buf at ffffffff8149ef5a
    #15 [ffff88085c6dbc20] n_tty_receive_buf_common at ffffffff8149f016
    #16 [ffff88085c6dbca0] n_tty_receive_buf2 at ffffffff8149f194
    #17 [ffff88085c6dbcb0] flush_to_ldisc at ffffffff814a238a
    #18 [ffff88085c6dbd50] process_one_work at ffffffff81090be2
    #19 [ffff88085c6dbe20] worker_thread at ffffffff81091b4d
    #20 [ffff88085c6dbeb0] kthread at ffffffff81096384
    #21 [ffff88085c6dbf50] ret_from_fork at ffffffff81a3d69f​

after slogging through some dissasembly:

ffffffff814b6720 <uart_put_char>:
ffffffff814b6720:	55                   	push   %rbp
ffffffff814b6721:	48 89 e5             	mov    %rsp,%rbp
ffffffff814b6724:	48 83 ec 20          	sub    $0x20,%rsp
ffffffff814b6728:	48 89 1c 24          	mov    %rbx,(%rsp)
ffffffff814b672c:	4c 89 64 24 08       	mov    %r12,0x8(%rsp)
ffffffff814b6731:	4c 89 6c 24 10       	mov    %r13,0x10(%rsp)
ffffffff814b6736:	4c 89 74 24 18       	mov    %r14,0x18(%rsp)
ffffffff814b673b:	e8 b0 8e 58 00       	callq  ffffffff81a3f5f0 <mcount>
ffffffff814b6740:	4c 8b a7 88 02 00 00 	mov    0x288(%rdi),%r12
ffffffff814b6747:	45 31 ed             	xor    %r13d,%r13d
ffffffff814b674a:	41 89 f6             	mov    %esi,%r14d
ffffffff814b674d:	49 83 bc 24 70 01 00 	cmpq   $0x0,0x170(%r12)
ffffffff814b6754:	00 00
ffffffff814b6756:	49 8b 9c 24 80 01 00 	mov    0x180(%r12),%rbx
ffffffff814b675d:	00
ffffffff814b675e:	74 2f                	je     ffffffff814b678f <uart_put_char+0x6f>
ffffffff814b6760:	48 89 df             	mov    %rbx,%rdi
ffffffff814b6763:	e8 a8 67 58 00       	callq  ffffffff81a3cf10 <_raw_spin_lock_irqsave>
ffffffff814b6768:	41 8b 8c 24 78 01 00 	mov    0x178(%r12),%ecx
ffffffff814b676f:	00
ffffffff814b6770:	89 ca                	mov    %ecx,%edx
ffffffff814b6772:	f7 d2                	not    %edx
ffffffff814b6774:	41 03 94 24 7c 01 00 	add    0x17c(%r12),%edx
ffffffff814b677b:	00
ffffffff814b677c:	81 e2 ff 0f 00 00    	and    $0xfff,%edx
ffffffff814b6782:	75 23                	jne    ffffffff814b67a7 <uart_put_char+0x87>
ffffffff814b6784:	48 89 c6             	mov    %rax,%rsi
ffffffff814b6787:	48 89 df             	mov    %rbx,%rdi
ffffffff814b678a:	e8 e1 64 58 00       	callq  ffffffff81a3cc70 <_raw_spin_unlock_irqrestore>
ffffffff814b678f:	44 89 e8             	mov    %r13d,%eax
ffffffff814b6792:	48 8b 1c 24          	mov    (%rsp),%rbx
ffffffff814b6796:	4c 8b 64 24 08       	mov    0x8(%rsp),%r12
ffffffff814b679b:	4c 8b 6c 24 10       	mov    0x10(%rsp),%r13
ffffffff814b67a0:	4c 8b 74 24 18       	mov    0x18(%rsp),%r14
ffffffff814b67a5:	c9                   	leaveq
ffffffff814b67a6:	c3                   	retq
ffffffff814b67a7:	49 8b 94 24 70 01 00 	mov    0x170(%r12),%rdx
ffffffff814b67ae:	00
ffffffff814b67af:	48 63 c9             	movslq %ecx,%rcx
ffffffff814b67b2:	41 b5 01             	mov    $0x1,%r13b
ffffffff814b67b5:	44 88 34 0a          	mov    %r14b,(%rdx,%rcx,1)
ffffffff814b67b9:	41 8b 94 24 78 01 00 	mov    0x178(%r12),%edx
ffffffff814b67c0:	00
ffffffff814b67c1:	83 c2 01             	add    $0x1,%edx
ffffffff814b67c4:	81 e2 ff 0f 00 00    	and    $0xfff,%edx
ffffffff814b67ca:	41 89 94 24 78 01 00 	mov    %edx,0x178(%r12)
ffffffff814b67d1:	00
ffffffff814b67d2:	eb b0                	jmp    ffffffff814b6784 <uart_put_char+0x64>
ffffffff814b67d4:	66 66 66 2e 0f 1f 84 	data32 data32 nopw %cs:0x0(%rax,%rax,1)
ffffffff814b67db:	00 00 00 00 00

for our build, this is crashing at:

    circ->buf[circ->head] = c;

Looking in uart_port_startup(), it seems that circ->buf (state->xmit.buf)
protected by the "per-port mutex", which based on uart_port_check() is
state->port.mutex. Indeed, the lock acquired in uart_put_char() is
uport->lock, i.e. not the same lock.

Anyway, since the lock is not acquired, if uart_shutdown() is called, the
last chunk of that function may release state->xmit.buf before its assigned
to null, and cause the race above.

To fix it, let's lock uport->lock when allocating/deallocating
state->xmit.buf in addition to the per-port mutex.

v2: switch to locking uport->lock on allocation/deallocation instead of
    locking the per-port mutex in uart_put_char. Note that since
    uport->lock is a spin lock, we have to switch the allocation to
    GFP_ATOMIC.
v3: move the allocation outside the lock, so we can switch back to
    GFP_KERNEL
v4: * switch to positive condition of state->xmit.buf in
      uart_port_startup()
    * declare `flags` on its own line
    * use the result of uart_port_lock() in uart_shutdown() to avoid
      uninitialized warning
    * don't use the uart_port_lock/unlock macros in uart_port_startup,
      instead test against uport directly; the compiler can't seem to "see"
      through the macros/ref/unref calls to not warn about uninitialized
      flags. We don't need to do a ref here since we hold the per-port
      mutex anyway.
v5: use uart_port_lock/unlock, but free the page outside the critical
    section if it is unused

Signed-off-by: Tycho Andersen <tycho@...ho.ws>
---
 drivers/tty/serial/serial_core.c | 20 +++++++++++++++-----
 1 file changed, 15 insertions(+), 5 deletions(-)

diff --git a/drivers/tty/serial/serial_core.c b/drivers/tty/serial/serial_core.c
index 9c14a453f73c..91b292bbda91 100644
--- a/drivers/tty/serial/serial_core.c
+++ b/drivers/tty/serial/serial_core.c
@@ -182,6 +182,8 @@ static int uart_port_startup(struct tty_struct *tty, struct uart_state *state,
 {
 	struct uart_port *uport = uart_port_check(state);
 	unsigned long page;
+	unsigned long flags;
+	bool page_used = false;
 	int retval = 0;
 
 	if (uport->type == PORT_UNKNOWN)
@@ -196,15 +198,20 @@ static int uart_port_startup(struct tty_struct *tty, struct uart_state *state,
 	 * Initialise and allocate the transmit and temporary
 	 * buffer.
 	 */
-	if (!state->xmit.buf) {
-		/* This is protected by the per port mutex */
-		page = get_zeroed_page(GFP_KERNEL);
-		if (!page)
-			return -ENOMEM;
+	page = get_zeroed_page(GFP_KERNEL);
+	if (!page)
+		return -ENOMEM;
 
+	uport = uart_port_lock(state, flags);
+	if (!state->xmit.buf) {
 		state->xmit.buf = (unsigned char *) page;
 		uart_circ_clear(&state->xmit);
+		page_used = true;
 	}
+	uart_port_unlock(uport, flags);
+
+	if (!page_used)
+		free_page(page);
 
 	retval = uport->ops->startup(uport);
 	if (retval == 0) {
@@ -263,6 +270,7 @@ static void uart_shutdown(struct tty_struct *tty, struct uart_state *state)
 {
 	struct uart_port *uport = uart_port_check(state);
 	struct tty_port *port = &state->port;
+	unsigned long flags;
 
 	/*
 	 * Set the TTY IO error marker
@@ -295,10 +303,12 @@ static void uart_shutdown(struct tty_struct *tty, struct uart_state *state)
 	/*
 	 * Free the transmit buffer page.
 	 */
+	uport = uart_port_lock(state, flags);
 	if (state->xmit.buf) {
 		free_page((unsigned long)state->xmit.buf);
 		state->xmit.buf = NULL;
 	}
+	uart_port_unlock(uport, flags);
 }
 
 /**
-- 
2.17.1

Powered by blists - more mailing lists