File ltrace.trace_exec.patch of Package ltrace

Overview Repositories Revisions Requests Users Attributes Meta

File ltrace.trace_exec.patch of Package ltrace (Revision cf43f5e31b99b90de200abb3328a0741)

Currently displaying revision cf43f5e31b99b90de200abb3328a0741 , Show latest

Bug 201282 - ltrace for execl call does not work

Hello Ian,
While trying out different programs with trunk-68 level, observed
that ltrace is not able to trace properly a new process created by
"exec" call. Output is showing "unexpected breakpoints" and library
calls of the new process is not getting displayed in ppc64.

I have prepared a patch for this using PTRACE_O_TRACEEXEC option. Tested
this in ppc64 architecture with no make check errors.

Man page of execve says "There shall be no return from  a successful
exec, because the calling process image is overlaid by the new process
image".  In S390 arch, the return value of execve is not predictable. In
ppc64, the program is not even hitting Sys_ret of execve. So instead of
having the processing of exec in process_sysret(), moved that block to
wait_for_something.c file. I was intending to use the following
explained in man page, to capture the event of execl getting executed.

PTRACE_O_TRACEEXEC (since Linux 2.5.46)
Stop the child at the next exec()  call  with
SIGTRAP  | PTRACE_EVENT_EXEC << 8.

But unfortunately ltrace is not getting this signal. SIGTRAP is not
getting modified as mentioned here during an execl call.  Hence for now,
for confirming whether the SIGTRAP received by ltrace is the one after
execution of  an execl call, the last entry in the system call stack in
used.

---
 proc.c                    |    5 +++--
 process_event.c           |   19 -------------------
 sysdeps/linux-gnu/trace.c |    5 ++++-
 wait_for_something.c      |   23 +++++++++++++++++++++++
 4 files changed, 30 insertions(+), 22 deletions(-)

Index: ltrace-0.4/proc.c
===================================================================
--- ltrace-0.4.orig/proc.c
+++ ltrace-0.4/proc.c
@@ -22,10 +22,11 @@ struct process *open_program(char *filen
 	}
 	proc->filename = filename;
 	proc->breakpoints_enabled = -1;
+	proc->pid = 0;
+	breakpoints_init(proc);
 	if (pid) {
 		proc->pid = pid;
 	}
-	breakpoints_init(proc);
 
 	proc->next = list_of_processes;
 	list_of_processes = proc;
@@ -56,5 +57,5 @@ void open_pid(pid_t pid, int verbose)
 
 	proc = open_program(filename, pid);
 	continue_process (pid);
-	proc->breakpoints_enabled = 1;
+	proc->breakpoints_enabled = -1;
 }
Index: ltrace-0.4/process_event.c
===================================================================
--- ltrace-0.4.orig/process_event.c
+++ ltrace-0.4/process_event.c
@@ -241,25 +241,6 @@ static void process_sysret(struct event 
 		output_right(LT_TOF_SYSCALLR, event->proc,
 			     sysname(event->proc, event->e_un.sysnum));
 	}
-	if (exec_p(event->proc, event->e_un.sysnum)) {
-		arg_type_info info;
-		info.arg_num = -1; /* Return value */
-		info.type = ARGTYPE_LONG;
-		if (gimme_arg(LT_TOF_SYSCALLR, event->proc, &info) == 0) {
-			pid_t saved_pid;
-			event->proc->mask_32bit = 0;
-			event->proc->personality = 0;
-			/* FIXME: Leak, should have arch_dep_free.
-			   But we are leaking here much more than that.  */
-			event->proc->arch_ptr = NULL;
-			event->proc->filename = pid2name(event->proc->pid);
-			saved_pid = event->proc->pid;
-			event->proc->pid = 0;
-			breakpoints_init(event->proc);
-			event->proc->pid = saved_pid;
-		} else
-			enable_all_breakpoints(event->proc);
-	}
 	continue_process(event->proc->pid);
 }
 
Index: ltrace-0.4/sysdeps/linux-gnu/trace.c
===================================================================
--- ltrace-0.4.orig/sysdeps/linux-gnu/trace.c
+++ ltrace-0.4/sysdeps/linux-gnu/trace.c
@@ -101,9 +101,12 @@ void trace_set_options(struct process *p
 #ifndef PTRACE_O_TRACESYSGOOD
 #define PTRACE_O_TRACESYSGOOD 0x00000001
 #endif
+#ifndef PTRACE_O_TRACEEXEC
+#define PTRACE_O_TRACEEXEC 0x00000010
+#endif
 	if (proc->tracesysgood & 0x80)
 		return;
-	if (ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_TRACESYSGOOD) < 0 &&
+	if (ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_TRACESYSGOOD | PTRACE_O_TRACEEXEC) < 0 &&
 	    ptrace(PTRACE_OLDSETOPTIONS, pid, 0, PTRACE_O_TRACESYSGOOD) < 0) {
 		perror("PTRACE_SETOPTIONS");
 		return;
Index: ltrace-0.4/wait_for_something.c
===================================================================
--- ltrace-0.4.orig/wait_for_something.c
+++ ltrace-0.4/wait_for_something.c
@@ -92,6 +92,29 @@ struct event *wait_for_something(void)
 		event.e_un.signum = WSTOPSIG(status);
 		return &event;
 	}
+	if (WSTOPSIG(status) == SIGTRAP){
+	/* Check whether this SIGTRAP is received just after execve is called for this process */
+		struct callstack_element *elem;
+
+		elem = &event.proc->callstack[event.proc->callstack_depth - 1];
+		if( elem && elem->is_syscall &&  exec_p(event.proc, elem->c_un.syscall)){
+			pid_t saved_pid;
+
+			event.thing = LT_EV_NONE;
+			event.e_un.signum = WSTOPSIG(status);
+			event.proc->mask_32bit = 0;
+			event.proc->personality = 0;
+			event.proc->arch_ptr = NULL;
+			event.proc->filename = pid2name(event.proc->pid);
+			saved_pid = event.proc->pid;
+			event.proc->pid = 0;
+			breakpoints_init(event.proc);
+			event.proc->pid = saved_pid;
+			continue_after_signal(event.proc->pid, event.e_un.signum);
+			return &event;
+		}
+	}
+
 	event.thing = LT_EV_BREAKPOINT;
 	if (!event.proc->instruction_pointer) {
 		event.proc->instruction_pointer =

Places

File ltrace.trace_exec.patch of Package ltrace (Revision cf43f5e31b99b90de200abb3328a0741)

Places