Bug 253873

Description of problem:
This is the same problem found as in BZ#245343 (RHEL4)

quote:
Assuming the following case:

 - There is a multithread process.
 - One of non-main thread read() from tty.
 - That process belongs to background process group

Then read() returns by accident with invalid errno (ERESTARTSYS).

The cause of this problem is tty driver. On sending SIGTTIN,
tty driver tries to send the signal to main-thread's
task_struct and return -ERESTARTSYS for restarting read().
However, the driver doesn't set TIF_SIGPENDING of current.
Hence current doesn't handle signal and returns to
userspace with setting errno ERESTARTSYS.

Same problem might occur on "SIGTTOU and write()" case.

How reproducible:
Always

Step to Reproduce:

1. Run attached program on background.

$ ./mt-wrong-errno &

2. Move that process to foreground

$ fg

Actual Results:
read() returns with setting errno ERESTARTSYS by accident.
In addition, such errno should not be appeared on user
space.

Expected Result:
Wait user input on foreground.

Summary of actions taken to resolve issue:
None

Location of diagnostic data:
None

Hardware configuration:
None

Software:
None

Additional Info:
Community kernel has already fixed this problem since 2.6.22-rc4.
Patch is at:

http://lkml.org/lkml/2007/5/29/296

Original discussion is here:

http://www.mail-archive.com/linux-kernel@vger.kernel.org/msg162569.html

Business Impact:
We can't call read() from tty safety.

Other:
None