Some fixes for nosys()/SIGSYS

[kib]

Summary

This changeset ensures that we call nosys() for each situation where ENOSYS is returned, which fixes situations with missed SIGSYS from e.g. too high syscall number.

In nosys(), allow ABI to indicate that SIGSYS should be not sent at all. Also allow to disable SIGSYS for FreeBSD native processes.

Right now only amd64 is handled.

[dchagin]

Nice idea

[dchagin]

sys/kern/kern_sig.c

163

SYSCTL_INT(_kern, OID_AUTO, signosys, CTLFLAG_RWTUN, &kern_signosys, 0,

[kib]

Fix sysctl var name.
Add tests.

[kib]

Add FF copyright for the tests.

[jhb]

sys/kern/kern_sig.c

4226–4230

It might be more readable if we had a SV_SIGSYS that opted in to getting the signals that the FreeBSD ABIs set in their flags and then this would be SV_PROC_FLAG(p, SV_SIGSYS) != 0

tests/sys/kern/sigsys.c

20

static volatile?

62

Do you want another version of this test that adjusts the sysctl and then verifies there is no sigsys? Or perhaps you should fetch the sysctl and alter the expected test results based on its value?

[kib]

tests/sys/kern/sigsys.c

62

I remember there is some atf knob to allow/disallow changing global system state during the test. It would be needed there.

In fact Dmitry has another version of the tests, that I hope he will incorporate into this draft after I commit my parts. D41979

[kib]

SV_SIGSYS
Use sig_atomic_t for signal handler counter
Plug !x86 places

[dchagin]

btw, you didn't change the arm64 and i386 Linuxulator

tests/sys/kern/sigsys.c

62

I remember there is some atf knob to allow/disallow changing global system state during the test. It would be needed there.

Well, I don't know such knob,
my test stupidly returned the knob state to the default state because the kyua is ugly in that case, however that's enought in my POV

In fact Dmitry has another version of the tests, that I hope he will incorporate into this draft after I commit my parts. D41979

ok

[kib]

Handle i386 and arm64 linuxolators

[dchagin]

tests/sys/kern/sigsys.c

17

<errno.h>, <stdbool.h> and <stdlib.h> seem redundant

[kib]

tests/sys/kern/sigsys.c

17

Only stdlib.h, if tests written properly.

[kib]

Improve tests, use ATF in more correct way.

[dchagin]

sys/i386/i386/trap.c

1116

[markj]

tests/sys/kern/sigsys.c

62

See tests/sys/kern/kern_descrip_test.c for an example. You need to set "allow_sysctl_side_effects" and presumably also "require.user".

Alternately, at the beginning of the test, check whether the sysctl is set, and skip the test if not. Since kern_signosys = 1 by default, this is probably sufficient.

[dchagin]

tests/sys/kern/sigsys.c

62

See tests/sys/kern/kern_descrip_test.c for an example. You need to set "allow_sysctl_side_effects" and presumably also "require.user".

Alternately, at the beginning of the test, check whether the sysctl is set, and skip the test if not. Since kern_signosys = 1 by default, this is probably sufficient.

Thank you, D41979 updated