tools: Comment and correctness fixes in dump_process_memory.
It is not necessary to kill the entire process group to dump it, and it is not
required to ptrace() it. Fix the first one, and document why ptrace() may still
be useful.
Reported by bgeffon@google.com.
Bug: 845459
Change-Id: I9f08e7208cfcfdec00bcaf80f83c585f1d4ea195
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1675649
Reviewed-by: Egor Pasko <pasko@chromium.org>
Commit-Queue: Benoit L <lizeb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#672040}
diff --git a/tools/dump_process_memory/dump_process.cc b/tools/dump_process_memory/dump_process.cc
index de6384be..ad94c87 100644
--- a/tools/dump_process_memory/dump_process.cc
+++ b/tools/dump_process_memory/dump_process.cc
@@ -45,9 +45,8 @@
ScopedPtracer(pid_t pid) : pid_(pid), is_attached_(false) {
// ptrace() delivers a SIGSTOP signal to one thread in the target process,
// unless it is already stopped. Since we want to stop the whole process,
- // send a signal to every thread in the process group.
- pid_t process_group_id = getpgid(pid);
- if (killpg(process_group_id, SIGSTOP)) {
+ // kill() it first.
+ if (kill(pid, SIGSTOP)) {
PLOG(ERROR) << "Cannot stop the process group of " << pid;
return;
}
@@ -206,6 +205,9 @@
// disk.
bool DumpMappings(pid_t pid) {
LOG(INFO) << "Attaching to " << pid;
+ // ptrace() is not required to read the process's memory, but the permissions
+ // to attach to the target process is.
+ // Attach anyway to make it clearer when this fails.
ScopedPtracer tracer(pid);
if (!tracer.IsAttached())
return false;
