* [9fans] Pipes staying after sending note
@ 2023-07-08 15:46 Philip Silva via 9fans
2023-07-09 0:03 ` [9fans] " Anthony Martin
0 siblings, 1 reply; 2+ messages in thread
From: Philip Silva via 9fans @ 2023-07-08 15:46 UTC (permalink / raw)
To: 9fans
Hello,
I'm trying to understand how pipes work when terminating a forked process. It seems when sending kill to the forked process, connected pipes don't always break. At least a subsequent write might work. Is it possible to make it reliably fail anyway or is it necessary to close the file descriptors from the calling process?
#include <u.h>
#include <libc.h>
#include <stdio.h>
void sendnote(int cpid) {
int fd;
char fn[20], buf[20];
sprintf(fn, "/proc/%d/note", cpid);
if ((fd = open(fn, OWRITE)) <= 0) {
printf("open %s\n", fd);
}
if (write(fd, "kill", 5) == 0) {
printf("write error\n");
}
close(fd);
return;
}
int main() {
int cpid, infd[2], outfd[2], n;
char outbuf[20], inbuf[20];
char *args[] = {"/bin/cat", NULL};
pipe(infd);
pipe(outfd);
if ((cpid = fork()) != 0) {
close(infd[1]);
close(outfd[1]);
n = write(infd[0], "test", 4);
printf("check process: wrote %d bytes\n", n);
sendnote(cpid);
// close(infd[0]); // <--- uncomment to make write fail
// close(outfd[0]);
n = write(infd[0], "test2", 5);
printf("write: wrote %d bytes\n", n);
n = read(outfd[0], outbuf, 20);
printf("outbuf=%s (n=%d)\n", outbuf, n);
wait();
return 0;
}
dup(infd[1], 0);
close(infd[0]);
close(infd[1]);
dup(outfd[1], 1);
close(outfd[0]);
close(outfd[1]);
exec("/bin/cat", args);
return 0;
}
Will output:
check process: wrote 4 bytes
write: wrote 5 bytes
outbuf=test (n=4)
6.out 82435: main
Closing the FDs after sending the note:
check process: wrote 4 bytes
write: wrote -1 bytes
outbuf= (n=-1)
6.out 82423: main
On the other hand instead of closing just adding sleep(1) after sendnote() produces:
check process: wrote 4 bytes
6.out 82595: sys: write on closed pipe pc=0x202603
(That's more or less also reproducing a part of Go's os/exec TestContextCancel)
Greetings, Philip
------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T3392001a02ee7ec8-Ma261d9fb098775daa526c6b3
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
^ permalink raw reply [flat|nested] 2+ messages in thread
* [9fans] Re: Pipes staying after sending note
2023-07-08 15:46 [9fans] Pipes staying after sending note Philip Silva via 9fans
@ 2023-07-09 0:03 ` Anthony Martin
0 siblings, 0 replies; 2+ messages in thread
From: Anthony Martin @ 2023-07-09 0:03 UTC (permalink / raw)
To: 9fans
Philip Silva via 9fans <9fans@9fans.net> once said:
> if ((cpid = fork()) != 0) {
> close(infd[1]);
> close(outfd[1]);
>
> n = write(infd[0], "test", 4);
> printf("check process: wrote %d bytes\n", n);
>
> sendnote(cpid);
> // close(infd[0]); // <--- uncomment to make write fail
> // close(outfd[0]);
>
> n = write(infd[0], "test2", 5);
> printf("write: wrote %d bytes\n", n);
>
> n = read(outfd[0], outbuf, 20);
> printf("outbuf=%s (n=%d)\n", outbuf, n);
> wait();
> return 0;
Why aren't you waiting for the process to die
after sending it the note? It must be running
to see that is has a note pending. At the time
of your second write, the scheduler may not
have selected the child process to run.
The sequence could look like:
child: read from empty pipe, go to sleep
parent: write to pipe
parent: write kill to /proc/$cpid/note
parent: write to pipe
child: wakeup, see note, die
Cheers,
Anthony
------------------------------------------
9fans: 9fans
Permalink: https://9fans.topicbox.com/groups/9fans/T3392001a02ee7ec8-Mfe03d0d140063db1ea509f2a
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2023-07-09 0:04 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2023-07-08 15:46 [9fans] Pipes staying after sending note Philip Silva via 9fans
2023-07-09 0:03 ` [9fans] " Anthony Martin
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).