This stand-alone program demonstrates the problem previously
reported.
If you execute this locally, the program will wait for any
input from the terminal (STDIN_FILENO) and then it will
terminate. This is the expected behavior.
If you execute this while logged in using telnet, using linux-2.4.18,
this program will also execute as expected.
However, if you execute using linux-2.4.19, when logged in using
telnet, the program will exit as soon as the child writes the
first '.' to the terminal because a SIGIO signal is being incorrectly
generated for both output and input.
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <signal.h>
#include <termios.h>
#include <fcntl.h>
#include <sys/wait.h>
#include <errno.h>
#include <sys/resource.h>
#define ERRORS(s) { \
fprintf(stderr, "Error from line %d, file %s, call %s, (%s)\n", \
__LINE__,__FILE__,(s), strerror(errno)); \
}
#define FAIL -1
int enab = 0;
int alive = 0;
static void set_sig(int sig, sig_t funct, int flags)
{
struct sigaction sa;
if(sigaction(sig, NULL, &sa) == FAIL)
ERRORS("sigaction");
sa.sa_flags = flags;
sa.sa_handler = funct;
if(sigaction(sig, &sa, NULL) == FAIL)
ERRORS("sigaction");
return;
}
static void iotrap(int unused)
{
enab = 0;
}
static void reaper(int unused)
{
alive = 0;
while(wait3(&unused, WNOHANG, NULL) > 0)
;
}
int main(int args, char *argv[]);
int main(int args, char *argv[])
{
int flags;
size_t i;
pid_t pid;
struct termios term, save;
set_sig(SIGCHLD, reaper, SA_INTERRUPT|SA_RESTART);
set_sig(SIGIO, iotrap, SA_INTERRUPT|SA_RESTART);
alive = 1;
switch((pid = fork()))
{
case 0: /* Child */
for(i=0; i < 0x10; i++)
{
(void)sleep(1);
fprintf(stderr, ".");
(void)sleep(1);
}
exit(EXIT_SUCCESS);
default:
break;
}
/*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
/*
* Save terminal characteristics and then set the terminal for raw
* input generating a signal upon any received character.
*/
if(tcgetattr(STDIN_FILENO, &term) == FAIL)
ERRORS("tcgetattr");
save = term;
term.c_lflag = ISIG;
term.c_iflag = 0;
if(tcsetattr(STDIN_FILENO, TCSANOW, &term) == FAIL)
ERRORS("tcsetattr");
if((flags = fcntl(STDIN_FILENO, F_GETFL)) == FAIL)
ERRORS("fcntl");
flags |= (FNDELAY|FASYNC);
if(fcntl(STDIN_FILENO, F_SETFL, flags) == FAIL)
ERRORS("fcntl");
if(fcntl(STDIN_FILENO, F_SETOWN, getpid()) == FAIL)
ERRORS("fcntl");
/*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
fprintf(stderr, "Waiting for input.......");
enab = 1;
while(enab)
{
pause();
fprintf(stderr, "Got out of pause\n");
}
if(alive) kill(pid, SIGINT);
fprintf(stderr, "Exit okay, cleaning up...\n");
/*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
/*
* Restore the terminal characteristics before we exit. Note, the
* terminal is shared. We can't just exit!
*/
flags &= ~(FNDELAY|FASYNC);
if(fcntl(STDIN_FILENO, F_SETFL, flags) == FAIL)
ERRORS("fcntl");
if(tcsetattr(STDIN_FILENO, TCSAFLUSH, &save) == FAIL)
ERRORS("tcsetattr");
set_sig(SIGIO, SIG_DFL, SA_INTERRUPT);
fprintf(stderr, "Done!\n");
return 0;
}
Cheers,
Dick Johnson
Penguin : Linux version 2.4.19 on an i686 machine (797.90 BogoMips).
The US military has given us many words, FUBAR, SNAFU, now ENRON.
Yes, top management were graduates of West Point and Annapolis.
This is a fix included since 2.4.19-pre5-aa1.
You're assuming that every SIGIO coming your way
1) is coming via fd 0
2) is a POLL_IN interrupt
You should either be polling the file descriptor (0) from the signal
handler, or using SA_SIGINFO (see sigaction(2)) to determine
which interrupt it is.
Regards,
Sapan
On Fri, Aug 16, 2002 at 03:27:33PM -0400, Richard B. Johnson wrote:
>
> This stand-alone program demonstrates the problem previously
> reported.
>
> If you execute this locally, the program will wait for any
> input from the terminal (STDIN_FILENO) and then it will
> terminate. This is the expected behavior.
>
> If you execute this while logged in using telnet, using linux-2.4.18,
> this program will also execute as expected.
>
> However, if you execute using linux-2.4.19, when logged in using
> telnet, the program will exit as soon as the child writes the
> first '.' to the terminal because a SIGIO signal is being incorrectly
> generated for both output and input.
>
>
> #include <stdio.h>
> #include <stdlib.h>
> #include <unistd.h>
> #include <string.h>
> #include <signal.h>
> #include <termios.h>
> #include <fcntl.h>
> #include <sys/wait.h>
> #include <errno.h>
> #include <sys/resource.h>
>
> #define ERRORS(s) { \
> fprintf(stderr, "Error from line %d, file %s, call %s, (%s)\n", \
> __LINE__,__FILE__,(s), strerror(errno)); \
> }
>
> #define FAIL -1
> int enab = 0;
> int alive = 0;
> static void set_sig(int sig, sig_t funct, int flags)
> {
> struct sigaction sa;
> if(sigaction(sig, NULL, &sa) == FAIL)
> ERRORS("sigaction");
> sa.sa_flags = flags;
> sa.sa_handler = funct;
> if(sigaction(sig, &sa, NULL) == FAIL)
> ERRORS("sigaction");
> return;
> }
> static void iotrap(int unused)
> {
> enab = 0;
> }
> static void reaper(int unused)
> {
> alive = 0;
> while(wait3(&unused, WNOHANG, NULL) > 0)
> ;
>
> }
> int main(int args, char *argv[]);
> int main(int args, char *argv[])
> {
> int flags;
> size_t i;
> pid_t pid;
> struct termios term, save;
> set_sig(SIGCHLD, reaper, SA_INTERRUPT|SA_RESTART);
> set_sig(SIGIO, iotrap, SA_INTERRUPT|SA_RESTART);
> alive = 1;
> switch((pid = fork()))
> {
> case 0: /* Child */
> for(i=0; i < 0x10; i++)
> {
> (void)sleep(1);
> fprintf(stderr, ".");
> (void)sleep(1);
> }
> exit(EXIT_SUCCESS);
> default:
> break;
> }
> /*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
> /*
> * Save terminal characteristics and then set the terminal for raw
> * input generating a signal upon any received character.
> */
> if(tcgetattr(STDIN_FILENO, &term) == FAIL)
> ERRORS("tcgetattr");
> save = term;
> term.c_lflag = ISIG;
> term.c_iflag = 0;
> if(tcsetattr(STDIN_FILENO, TCSANOW, &term) == FAIL)
> ERRORS("tcsetattr");
> if((flags = fcntl(STDIN_FILENO, F_GETFL)) == FAIL)
> ERRORS("fcntl");
> flags |= (FNDELAY|FASYNC);
> if(fcntl(STDIN_FILENO, F_SETFL, flags) == FAIL)
> ERRORS("fcntl");
> if(fcntl(STDIN_FILENO, F_SETOWN, getpid()) == FAIL)
> ERRORS("fcntl");
> /*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
> fprintf(stderr, "Waiting for input.......");
> enab = 1;
> while(enab)
> {
> pause();
> fprintf(stderr, "Got out of pause\n");
>
> }
> if(alive) kill(pid, SIGINT);
> fprintf(stderr, "Exit okay, cleaning up...\n");
> /*-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-*/
> /*
> * Restore the terminal characteristics before we exit. Note, the
> * terminal is shared. We can't just exit!
> */
> flags &= ~(FNDELAY|FASYNC);
> if(fcntl(STDIN_FILENO, F_SETFL, flags) == FAIL)
> ERRORS("fcntl");
> if(tcsetattr(STDIN_FILENO, TCSAFLUSH, &save) == FAIL)
> ERRORS("tcsetattr");
> set_sig(SIGIO, SIG_DFL, SA_INTERRUPT);
> fprintf(stderr, "Done!\n");
> return 0;
> }
>
>
>
> Cheers,
> Dick Johnson
> Penguin : Linux version 2.4.19 on an i686 machine (797.90 BogoMips).
> The US military has given us many words, FUBAR, SNAFU, now ENRON.
> Yes, top management were graduates of West Point and Annapolis.
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at http://www.tux.org/lkml/
On Wed, 21 Aug 2002 [email protected] wrote:
> This is a fix included since 2.4.19-pre5-aa1.
>
Well if this is a 'fix' there is a lot of legacy software that
just got broken. It was discovered when the new kernel was booted
on some very reliable file-servers.
> You're assuming that every SIGIO coming your way
> 1) is coming via fd 0
> 2) is a POLL_IN interrupt
>
> You should either be polling the file descriptor (0) from the signal
> handler, or using SA_SIGINFO (see sigaction(2)) to determine
> which interrupt it is.
>
I wrote the software to demonstrate the problem, not as an example
of how to poll the keyboard.
Cheers,
Dick Johnson
Penguin : Linux version 2.4.18 on an i686 machine (797.90 BogoMips).
The US military has given us many words, FUBAR, SNAFU, now ENRON.
Yes, top management were graduates of West Point and Annapolis.
[email protected] said:
> Well if this is a 'fix' there is a lot of legacy software that just
> got broken. It was discovered when the new kernel was booted on some
> very reliable file-servers.
Then your "very reliable" file-servers are broken and need fixing so they
no longer depend on this bug.
Jeff