startup.c:129 if daemon(0,0) fails, the error message is wrong.

[olahaye74]

daemon(0,0) does:

• fork
• setsid
• chdir / (1st arg = 0)
• redirects standard input, standard output and standard error to /dev/null (2nd arg = 0)

if daemon(0,0) fails, that does not mean that fork failed. I can fail at setsid, chdir or redirections to /dev/null

I got gpm failed errno=37. Doing strace show that fork succeed indeed. The problem was that /dev/null is not a char device and lock failed.
The correct error should be "daemonize failed" not "fork failed"!
(Because of the missleading message I lost time.)

[telmich]

No strong opinion from my side, please send a PR to change it

[olahaye74]

I have no time yet for that now.
Also, I gave a missleading message. I don't know errno in fact, the only real thing is that daemon fails after clone (fork). Looking at daemon source code (https://github.com/lattera/glibc/blob/master/misc/daemon.c), and looking at gpm strace, I see the fork success, the chdir success, the setsid and the open(/dev/null) success, thus it fails checking /dev/null is a char device (my system has a buggy /dev/null which is not a char device for some reason).
strace relevant part:
clone(child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD, child_tidptr=0x7f294c08ca10) = 21398
exit_group(0) = ?
+++ exited with 0 +++
strace: Process 21398 attached
setsid() = 21398
chdir("/") = 0
open("/dev/null", O_RDWR) = 3
fstat(3, {st_mode=S_IFREG|0644, st_size=0, ...}) = 0
close(3)
=> Then it reports fork failed.

[olahaye74]

Confirmed, If I do "rm -f /dev/null; mknode /dev/null c 1 3", gpm starts without problem.
It should report /dev/null: BAD DEVICE or something appropriate.

[olahaye74]

After speaking with glibc dev, it turns out that gpm doesn't handle errno.
daemon() will set errno to ENODEV if /dev/null is not a device.
Thus, the error message should be "daemon() failed: ENODEV" or something similar.