usb: gadget: serial: fix Tx stall after buffer overflow
Subsystems (2)
PATCH 1/1 usb: gadget: serial: fix Tx stall after buffer overflow
Symptom: application opens /dev/ttyGS0 and starts sending (writing) to it while either USB cable is not connected, or nobody listens on the other side of the cable. If driver circular buffer overflows before connection is established, no data will be written to the USB layer until/unless /dev/ttyGS0 is closed and re-opened again by the application (the latter besides having no means of being notified about the event of establishing of the connection.)
Fix: on open and/or connect, kick Tx to flush circular buffer data to USB layer.
NOTE: current version of the driver leaks data from one connection to another through its internal circular buffer. It might be a good idea to clear the buffer on open/close/connect/disconnect, in which case the problem this patch solves would have been fixed in a different manner. However, not only that's a more dramatic change, but to do it right TTY-layer buffers are to be considered as well.
---
drivers/usb/gadget/function/u_serial.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/usb/gadget/function/u_serial.c +2 -0
@@ -563,6 +563,8 @@ static int gs_start_io(struct gs_port *port)/* unblock any pending writes into our circular buffer */ if (started) { +pr_debug ("gs_start_tx: ttyGS%d\n" , port->port_num); +gs_start_tx (port);tty_wakeup (port->port.tty); }else {gs_free_requests (ep, head, &port->read_allocated);
Signature · 1 line
2.10.0.1.g57b01a3
On Fri, Jan 17, 2020 at 08:29:33AM +0300, Sergey Organov wrote:
[...]
Quoted 6 lines
NOTE: current version of the driver leaks data from one connection to another through its internal circular buffer. It might be a good idea to clear the buffer on open/close/connect/disconnect, in which case the problem this patch solves would have been fixed in a different manner. However, not only that's a more dramatic change, but to do it right TTY-layer buffers are to be considered as well.
This is normal for serial devices, as they don't have any means to signal connection and will usually transmit anyway when not connected. In case of a console on the USB gadget-emulated serial port, it might actually be convenient that the data is kept until connection.
drivers/usb/gadget/function/u_serial.c static int gs_start_io(struct gs_port *port) · line 568
Quoted 9 lines
drivers/usb/gadget/function/u_serial.c +2
-0@@ -563,6 +563,8 @@ static int gs_start_io(struct gs_port *port)/* unblock any pending writes into our circular buffer */ if (started) { +pr_debug ("gs_start_tx: ttyGS%d\n" , port->port_num); +gs_start_tx (port);tty_wakeup (port->port.tty);
The tty_wakeup() will be called from gs_start_tx(), so should be removed from here.
The pr_debug() in other callers of gs_start_tx() say:
"caller: start ttyGS%d".
Best Regards,
Michał Mirosław
Quoted 9 lines · Michał Mirosław
On Fri, Jan 17, 2020 at 08:29:33AM +0300, Sergey Organov wrote:
[...]Quoted 6 lines
NOTE: current version of the driver leaks data from one connection to another through its internal circular buffer. It might be a good idea to clear the buffer on open/close/connect/disconnect, in which case the problem this patch solves would have been fixed in a different manner. However, not only that's a more dramatic change, but to do it right TTY-layer buffers are to be considered as well.
Quoted 4 lines
This is normal for serial devices, as they don't have any means to signal connection and will usually transmit anyway when not connected. In case of a console on the USB gadget-emulated serial port, it might actually be convenient that the data is kept until connection.
Yeah, just wanted to make sure I did select the right way of fixing the issue.
Quoted 5 lines
Quoted 5 lines
drivers/usb/gadget/function/u_serial.c +0
-0@@ -563,6 +563,8 @@ static int gs_start_io(struct gs_port *port)
Quoted 8 lines
drivers/usb/gadget/function/u_serial.c static int gs_start_io(struct gs_port *port) · line 568
Quoted 5 lines
drivers/usb/gadget/function/u_serial.c +2
-0/* unblock any pending writes into our circular buffer */ if (started) { +pr_debug ("gs_start_tx: ttyGS%d\n" , port->port_num); +gs_start_tx (port);tty_wakeup (port->port.tty);The
tty_wakeup()will be called fromgs_start_tx(), so should be removed from here.
Not exactly. tty_wakeup() will be called from gs_start_tx() only when there has been something actually transferred from the buffer. I didn't want to change behavior when the buffer is empty, so I kept the explicit tty_wakeup() call in place, intentionally. Please let me know if you still think it should be removed.
Quoted 2 lines
The
pr_debug()in other callers ofgs_start_tx()say:
"caller: start ttyGS%d".
???
$ git co gregkh/tty-next && grep -r 'caller: start tty' . HEAD is now at 7788f54... serial_core: Remove unused member in uart_port $
-- Sergey
Quoted 16 lines · Sergey Organov
Quoted 5 lines · Michał Mirosław
Quoted 4 lines · Sergey Organov
drivers/usb/gadget/function/u_serial.c +0
-0@@ -563,6 +563,8 @@ static int gs_start_io(struct gs_port *port)Quoted 8 lines
drivers/usb/gadget/function/u_serial.c static int gs_start_io(struct gs_port *port) · line 568
Quoted 5 lines
drivers/usb/gadget/function/u_serial.c +2
-0/* unblock any pending writes into our circular buffer */ if (started) { +pr_debug ("gs_start_tx: ttyGS%d\n" , port->port_num); +gs_start_tx (port);tty_wakeup (port->port.tty);The
tty_wakeup()will be called fromgs_start_tx(), so should be removed from here.
Quoted 5 lines
Not exactly.
tty_wakeup()will be called fromgs_start_tx()only when there has been something actually transferred from the buffer. I didn't want to change behavior when the buffer is empty, so I kept the explicittty_wakeup()call in place, intentionally. Please let me know if you still think it should be removed.
Indeed it is as you describe. You might add an argument that initializes do_tty_wake, but I'm not sure saving one tty_wakeup() on open is worth the trouble.
Quoted 8 lines
Quoted 2 lines
The
pr_debug()in other callers ofgs_start_tx()say:
"caller: start ttyGS%d".???
$ git co gregkh/tty-next && grep -r 'caller: start tty' . HEAD is now at 7788f54... serial_core: Remove unused member in uart_port $
Replace 'caller' with a function calling gs_start_io().
Best Regards,
Michał Mirosław
Quoted 23 lines · Michał Mirosław
Quoted 16 lines · Sergey Organov
Quoted 5 lines · Michał Mirosław
Quoted 4 lines · Sergey Organov
drivers/usb/gadget/function/u_serial.c +0
-0@@ -563,6 +563,8 @@ static int gs_start_io(struct gs_port *port)Quoted 8 lines
drivers/usb/gadget/function/u_serial.c static int gs_start_io(struct gs_port *port) · line 568
Quoted 5 lines
drivers/usb/gadget/function/u_serial.c +2
-0/* unblock any pending writes into our circular buffer */ if (started) { +pr_debug ("gs_start_tx: ttyGS%d\n" , port->port_num); +gs_start_tx (port);tty_wakeup (port->port.tty);The
tty_wakeup()will be called fromgs_start_tx(), so should be removed from here.Quoted 5 lines
Not exactly.
tty_wakeup()will be called fromgs_start_tx()only when there has been something actually transferred from the buffer. I didn't want to change behavior when the buffer is empty, so I kept the explicittty_wakeup()call in place, intentionally. Please let me know if you still think it should be removed.
Quoted 3 lines
Indeed it is as you describe. You might add an argument that initializes
do_tty_wake, but I'm not sure saving onetty_wakeup()on open is worth the trouble.
OK, so let's leave it as is, at least for now.
Quoted 11 lines
Quoted 9 lines
Quoted 3 lines
The
pr_debug()in other callers ofgs_start_tx()say:
"caller: start ttyGS%d".???
$ git co gregkh/tty-next && grep -r 'caller: start tty' . HEAD is now at 7788f54... serial_core: Remove unused member in uart_port $Replace 'caller' with a function calling
gs_start_io().
Thanks, now I see... Do you prefer:
pr_debug("gs_start_io: start Tx on ttyGS%d\n", port->port_num);then?
Alternatively, I'm OK with removing this new debug print.
What do you think?
Thanks,
-- Sergey Organov
Quoted 15 lines · Sergey Organov
Quoted 5 lines · Michał Mirosław
Quoted 4 lines · Sergey Organov
Quoted 2 lines · Michał Mirosław
The
pr_debug()in other callers ofgs_start_tx()say:
"caller: start ttyGS%d".Quoted 7 lines
Quoted 5 lines
???
$ git co gregkh/tty-next && grep -r 'caller: start tty' . HEAD is now at 7788f54... serial_core: Remove unused member in uart_port $Replace 'caller' with a function calling
gs_start_io().
Quoted 7 lines
Thanks, now I see... Do you prefer:
pr_debug("gs_start_io: start Tx on ttyGS%d\n", port->port_num);then?
Alternatively, I'm OK with removing this new debug print.
Let's remove it. I was convinced that this is a caller of gs_start_io() and not the function itself. In this case callers already do the print.
BTW, the callers silently ignore (error) returns from this function. It might be useful to add pr_err() catching the errors.
Best Regards,
Michał Mirosław
PATCH v2 1/1 usb: gadget: serial: fix Tx stall after buffer overflow
In reply to Sergey OrganovSymptom: application opens /dev/ttyGS0 and starts sending (writing) to it while either USB cable is not connected, or nobody listens on the other side of the cable. If driver circular buffer overflows before connection is established, no data will be written to the USB layer until/unless /dev/ttyGS0 is closed and re-opened again by the application (the latter besides having no means of being notified about the event of establishing of the connection.)
Fix: on open and/or connect, kick Tx to flush circular buffer data to USB layer.
---
Changes in v2:
- Add comment to document why tty_wakeup() is kept in place
- Don't add debug print
- Remove NOTE from description
drivers/usb/gadget/function/u_serial.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/usb/gadget/function/u_serial.c +3 -1
@@ -561,8 +561,10 @@ static int gs_start_io(struct gs_port *port) port->n_read =0 ; started =gs_start_rx (port); -/* unblock any pending writes into our circular buffer */ if (started) { +gs_start_tx (port); +/* Unblock any pending writes into our circular buffer, in case +* we didn't in gs_start_tx() */ tty_wakeup (port->port.tty); }else {gs_free_requests (ep, head, &port->read_allocated);
Signature · 1 line
2.10.0.1.g57b01a3
Quoted 20 lines · Michał Mirosław
Quoted 19 lines · Sergey Organov
Quoted 5 lines · Michał Mirosław
Quoted 4 lines · Sergey Organov
Quoted 2 lines · Michał Mirosław
The
pr_debug()in other callers ofgs_start_tx()say:
"caller: start ttyGS%d".Quoted 7 lines
Quoted 5 lines
???
$ git co gregkh/tty-next && grep -r 'caller: start tty' . HEAD is now at 7788f54... serial_core: Remove unused member in uart_port $Replace 'caller' with a function calling
gs_start_io().Thanks, now I see... Do you prefer:
pr_debug("gs_start_io: start Tx on ttyGS%d\n", port->port_num);
Quoted 7 lines
Quoted 3 lines
then?
Alternatively, I'm OK with removing this new debug print.
Let's remove it. I was convinced that this is a caller of
gs_start_io()and not the function itself. In this case callers already do the print.
OK.
Quoted 2 lines
BTW, the callers silently ignore (error) returns from this function. It might be useful to add
pr_err()catching the errors.
... or actually handle the error returns? Anyway, that's beyond the scope of the patch and my expertise.
I've sent re-roll of the patch according to our discussion.
Thanks,
-- Sergey
Quoted 11 lines · Sergey Organov
Symptom: application opens
/dev/ttyGS0and starts sending (writing) to it while either USB cable is not connected, or nobody listens on the other side of the cable. If driver circular buffer overflows before connection is established, no data will be written to the USB layer until/unless/dev/ttyGS0is closed and re-opened again by the application (the latter besides having no means of being notified about the event of establishing of the connection.)Fix: on open and/or connect, kick Tx to flush circular buffer data to USB layer.
Quoted 8 lines
Signed-off-by Sergey Organov <[email protected]>---
Changes in v2:
- Add comment to document why
tty_wakeup()is kept in place
- Don't add debug print
- Remove NOTE from description
Best Regards,
Michał Mirosław
PATCH v3 1/1 usb: gadget: serial: fix Tx stall after buffer overflow
In reply to Sergey OrganovSymptom: application opens /dev/ttyGS0 and starts sending (writing) to it while either USB cable is not connected, or nobody listens on the other side of the cable. If driver circular buffer overflows before connection is established, no data will be written to the USB layer until/unless /dev/ttyGS0 is closed and re-opened again by the application (the latter besides having no means of being notified about the event of establishing of the connection.)
Fix: on open and/or connect, kick Tx to flush circular buffer data to USB layer.
---
Changes in v3:
- Add Reviewed-by: Michał Mirosław <[email protected]>
Changes in v2:
- Add comment to document why tty_wakeup() is kept in place
- Don't add debug print
- Remove NOTE from description
drivers/usb/gadget/function/u_serial.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/usb/gadget/function/u_serial.c +3 -1
@@ -561,8 +561,10 @@ static int gs_start_io(struct gs_port *port) port->n_read =0 ; started =gs_start_rx (port); -/* unblock any pending writes into our circular buffer */ if (started) { +gs_start_tx (port); +/* Unblock any pending writes into our circular buffer, in case +* we didn't in gs_start_tx() */ tty_wakeup (port->port.tty); }else {gs_free_requests (ep, head, &port->read_allocated);
Signature · 1 line
2.10.0.1.g57b01a3