Skip to contentSkip to conversation

Is this a bug that n_tty_set_room not serialized? If so please correct it.

DISCUSSION [email protected] 2 participants 2 messages Jan 23, 2013

lore
Share & tools
lore
#0 lore raw

Hi experts,

Seems there is an bug related to function n_tty_set_roomin source file drivers/tty/n_tty.c. This function is used to set receive_room based on the read_cntby both the consumer and the producer of the tty's read buffer.

But this function is not serialized, so I am afraid there is
a risk like: The producer make the buffer full and then call this function to set the receive_room to 0; while it is preempted by the consumer that empty the buffer(read_cnt is 0)and then call this very function to set the receive_room to a non-zero value. But when the producer continue, receive_room was set to 0 again. Thus both read_cnt and receive_room is 0.For consumer theread_cnt is 0 that no data can be read; while for the producer, thebuffer is full since the receive_room is 0. Both of them blockhere...

Two methods to avoid it:
A) Addor find a lock to serialize this function;
B) Once the producer find the buffer is full, double check/set the buffer with n_tty_set_room. For example in function "flush_to_ldisc".

Please correct me if I misunderstand something. Or please fix it.

Thanks
Xiao

#1 lore raw
Quoted 9 lines · On Wed, 23 Jan 2013 12:40:38 +0800 [email protected] wrote:

Hi experts,

Seems there is an bug related to function n_tty_set_roomin source file drivers/tty/n_tty.c. This function is used to set receive_room based on the read_cntby both the consumer and the producer of the tty's read buffer.

But this function is not serialized, so I am afraid there is a risk like: The producer make the buffer full and then call this function to

It's not entirely robust - it should be ok for current usage. The proper fix IMHO is to make the ldisc receive function return the number of bytes actually consumed. That cures the problem rather than patching around it.

It's an interesting little project for someone, and as I think only n_tty uses that level of flow control probably not too horrible now.

Alan