Hi!
I wanted to look at the process table of a crashed kernel from a dump.
ps(1) then segfaulted.
It seems that
1) pr->ps_pgrp is kernel's (it's being kvm_read earlier in libkvm)
and so is ps_session
2) p_comm and s_login are part of their structs and not pointers,
therefore kvm_read() a. k. a. copy_str() doesn't work on them.
I couldn't fix argv to display properly, but this is already better than
a segfault. Tested only on i386.
Reproduce:
1) C-M-Esc, boot crash
2) ps -N /var/crash/bsd.? -M /var/crash/bsd.?.core
3) when trying the diff, don't forget to recompile libkvm AND ps
Anyone has anything to say or is willing to commit?
--
Martin Pelikan
Index: sys/sysctl.h
===================================================================
RCS file: /cvs/src/sys/sys/sysctl.h,v
retrieving revision 1.116
diff -u -p -r1.116 sysctl.h
--- sys/sysctl.h 8 Jul 2011 18:38:55 -0000 1.116
+++ sys/sysctl.h 18 Aug 2011 11:16:35 -0000
@@ -482,7 +482,7 @@ do {
\
(kp)->p_limit = PTRTOINT64((pr)->ps_limit); \
(kp)->p_vmspace = PTRTOINT64((p)->p_vmspace); \
(kp)->p_sigacts = PTRTOINT64((p)->p_sigacts); \
- (kp)->p_sess = PTRTOINT64((pr)->ps_session); \
+ (kp)->p_sess = PTRTOINT64(sess); \
(kp)->p_ru = PTRTOINT64((p)->p_ru); \
\
(kp)->p_exitsig = (p)->p_exitsig; \
@@ -528,11 +528,11 @@ do {
\
(kp)->p_xstat = (p)->p_xstat; \
(kp)->p_acflag = (p)->p_acflag; \
\
- /* XXX depends on p_emul being an array and not a pointer */ \
+ /* p_emul is a pointer and not an array */ \
copy_str((kp)->p_emul, (char *)(p)->p_emul + \
offsetof(struct emul, e_name), sizeof((kp)->p_emul)); \
- copy_str((kp)->p_comm, (p)->p_comm, sizeof((kp)->p_comm)); \
- copy_str((kp)->p_login, (sess)->s_login, \
+ (void) memcpy((kp)->p_comm, (p)->p_comm, sizeof((kp)->p_comm)); \
+ (void) memcpy((kp)->p_login, (sess)->s_login, \
MIN(sizeof((kp)->p_login) - 1, sizeof((sess)->s_login))); \
\
if ((sess)->s_ttyvp) \
