Re: [PATCH, gdbserver] Scan for existing threads during attachment on Linux

[Luis Machado <lgustavo@codesourcery.com>]

[-- Attachment #1: Type: text/plain, Size: 3747 bytes --]

On 05/23/2011 07:37 AM, Pedro Alves wrote:
> On Friday 20 May 2011 19:42:02, Luis Machado wrote:
>
>
>> +/* Attach to PID.  If PID is the tgdi, attach to it and stop
>> +   all of its threads.  */
>> +
>>   int
>>   linux_attach (unsigned long pid)
>>   {
>> +  DIR *dir;
>> +  char pathname[128];
>> +  int is_tgid = 0;
>> +  FILE *fd;
>> +
>> +  /* Attach to PID.  We will check for other threads
>> +     soon.  */
>>     linux_attach_lwp_1 (pid, 1);
>> -  linux_add_process (pid, 1);
>>
>> -  if (!non_stop)
>> +  /* Find out what is the tgid of this lwp.  */
>> +  sprintf (pathname, "/proc/%ld/stat", pid);
>
> Can you factor this into a standalone function, something
> like
>
> static int
> tgid_of_pid (int pid)
> {
> ...
> }
>
> please?
>

Done.

>> +
>> +  fd = fopen (pathname, "r");
>> +
>> +  if (fd != NULL)
>>       {
>> -      struct thread_info *thread;
>> +      int proc_id, ppid, pgrp;
>> +      char comm[NAME_MAX + 1], state;
>>
>> -     /* Don't ignore the initial SIGSTOP if we just attached to this
>> -       process.  It will be collected by wait shortly.  */
>> -      thread = find_thread_ptid (ptid_build (pid, pid, 0));
>> -      thread->last_resume_kind = resume_stop;
>> +      fscanf (fd, "%d %s %c %d %d",&proc_id, comm,&state,&ppid,&pgrp);
>> +      fclose (fd);
>> +
>> +      if (pgrp == pid)
>> +       is_tgid = 1;
>> +    }
>> +  else
>> +    {
>> +      fprintf (stderr, "Could not open /proc/%ld/stat.\n", pid);
>> +      fflush (stderr);
>>       }
>>
>> +  sprintf (pathname, "/proc/%ld/task", pid);
>> +
>> +  dir = opendir (pathname);
>> +
>> +  if (!dir)
>> +    {
>> +      fprintf (stderr, "Could not open /proc/%ld/task.\n", pid);
>> +      fflush (stderr);
>> +    }
>
> Why even try to open the dir if is_tgid is false?
>

Doesn't make sense really. Fixed.

>> +  else if (is_tgid)
>> +    {
>> +      /* At this point we attached to the tgid.  Scan the task for
>> +        existing threads.  */
>> +      unsigned long lwp;
>> +      int new_threads_found;
>> +      int iterations = 0;
>> +      struct dirent *dp;
>> +
>> +      while (iterations<  2)
>> +       {
>> +         new_threads_found = 0;
>> +         /* Add all the other threads.  While we go through the
>> +            threads, new threads may be spawned.  Cycle through
>> +            the list of threads until we have done two iterations without
>> +            finding new threads.  */
>> +         while ((dp = readdir (dir)) != NULL)
>> +           {
>> +             /* Fetch one lwp.  */
>> +             lwp = strtoul (dp->d_name, NULL, 10);
>> +
>> +             /* Is this a new thread?  */
>> +             if (lwp&&  find_thread_ptid (ptid_build (pid, lwp, 0)) == NULL)
>> +               {
>> +                 linux_attach_lwp_1 (lwp, 0);
>> +                 new_threads_found++;
>> +
>> +                 if (debug_threads)
>> +                   fprintf (stderr, "Found and attached to new lwp %ld\n", lwp);
>> +               }
>> +           }
>> +
>> +         if (!new_threads_found)
>> +           iterations++;
>> +         else
>> +           iterations = 0;
>> +
>> +         rewinddir (dir);
>> +       }
>> +      closedir (dir);
>> +    }
>> +
>> +  linux_add_process (pid, 1);
>> +
>> +  /* Mark the threads as stopped.  */
>> +  if (!non_stop)
>> +    for_each_inferior (&all_threads, set_resume_kind_stop);
>> +
>>     return 0;
>>   }
>>
>
> Otherwise okay.
>

I've also addressed the two comments from the other reply. We're 
stopping only the threads from the PID we're attaching to.

The stale comments were updated to reflect the current situation. We can 
still have a lwp that isn't the tgid, but we handle the case of pid == 
tgid correctly now. The new comments reflect that.

Luis

[-- Attachment #2: stop_threads.diff --]
[-- Type: text/x-patch, Size: 4489 bytes --]

2011-05-23  Luis Machado  <lgustavo@codesourcery.com>

	* linux-low.c (tgid_of_pid): New function.
	(linux_attach): Scan for existing threads when attaching to a
	process that is the tgid.

--- .pc/stop_threads.diff/gdb/gdbserver/linux-low.c	2011-05-23 12:05:42.447095000 -0300
+++ gdb/gdbserver/linux-low.c	2011-05-23 12:40:47.735095000 -0300
@@ -495,6 +495,38 @@ get_stop_pc (struct lwp_info *lwp)
   return stop_pc;
 }
 
+/* Extract the tgid (thread group leader id) of PID.  Return
+   the tgid if successful and 0 otherwise.  */
+
+static unsigned long
+tgid_of_pid (unsigned long pid)
+{
+  char pathname[128];
+  FILE *fd;
+  unsigned long tgid = 0;
+
+  sprintf (pathname, "/proc/%ld/stat", pid);
+
+  fd = fopen (pathname, "r");
+
+  if (fd != NULL)
+    {
+      int proc_id, ppid, pgrp;
+      char comm[NAME_MAX + 1], state;
+
+      /* Extract the pgrp field.  */
+      fscanf (fd, "%d %s %c %d %d", &proc_id, comm, &state, &ppid, &pgrp);
+      fclose (fd);
+      tgid = pgrp;
+    }
+  else
+    {
+      fprintf (stderr, "Could not open /proc/%ld/stat.\n", pid);
+      fflush (stderr);
+    }
+    return tgid;
+}
+
 static void *
 add_lwp (ptid_t ptid)
 {
@@ -586,7 +618,9 @@ linux_attach_lwp_1 (unsigned long lwpid,
     }
 
   if (initial)
-    /* NOTE/FIXME: This lwp might have not been the tgid.  */
+    /* If lwp is the tgid, we handle adding existing threads later.
+       Otherwise we just add lwp without bothering about any other
+       threads.  */
     ptid = ptid_build (lwpid, lwpid, 0);
   else
     {
@@ -621,8 +655,10 @@ linux_attach_lwp_1 (unsigned long lwpid,
 	In this case we want the process thread to stop.
 	This is handled by having linux_attach set last_resume_kind ==
 	resume_stop after we return.
-	??? If the process already has several threads we leave the other
-	threads running.
+
+	If the pid we are attaching to is also the tgid, we attach to and
+	stop all the existing threads.  Otherwise, we attach to pid and
+	ignore any other threads in the same group as this pid.
 
      3) GDB is connecting to gdbserver and is requesting an enumeration of all
 	existing threads.
@@ -646,22 +682,93 @@ linux_attach_lwp (unsigned long lwpid)
   linux_attach_lwp_1 (lwpid, 0);
 }
 
+/* Attach to PID.  If PID is the tgdi, attach to it and stop
+   all of its threads.  */
+
 int
 linux_attach (unsigned long pid)
 {
+  struct thread_info *thread;
+
+  /* Attach to PID.  We will check for other threads
+     soon.  */
   linux_attach_lwp_1 (pid, 1);
   linux_add_process (pid, 1);
 
   if (!non_stop)
     {
-      struct thread_info *thread;
-
-     /* Don't ignore the initial SIGSTOP if we just attached to this
-	process.  It will be collected by wait shortly.  */
       thread = find_thread_ptid (ptid_build (pid, pid, 0));
       thread->last_resume_kind = resume_stop;
     }
 
+  if (tgid_of_pid (pid) == pid)
+    {
+      DIR *dir;
+      char pathname[128];
+
+      sprintf (pathname, "/proc/%ld/task", pid);
+
+      dir = opendir (pathname);
+
+      if (!dir)
+	{
+	  fprintf (stderr, "Could not open /proc/%ld/task.\n", pid);
+	  fflush (stderr);
+	}
+      else
+	{
+	  /* At this point we attached to the tgid.  Scan the task for
+	     existing threads.  */
+	  unsigned long lwp;
+	  int new_threads_found;
+	  int iterations = 0;
+	  struct dirent *dp;
+
+	  while (iterations < 2)
+	    {
+	      new_threads_found = 0;
+	      /* Add all the other threads.  While we go through the
+		 threads, new threads may be spawned.  Cycle through
+		 the list of threads until we have done two iterations without
+		 finding new threads.  */
+	      while ((dp = readdir (dir)) != NULL)
+		{
+		  /* Fetch one lwp.  */
+		  lwp = strtoul (dp->d_name, NULL, 10);
+
+		  /* Is this a new thread?  */
+		  if (lwp && find_thread_ptid (ptid_build (pid, lwp, 0)) == NULL)
+		    {
+		      linux_attach_lwp_1 (lwp, 0);
+		      new_threads_found++;
+
+		      /* Mark the threads as stopped.  */
+		      if (!non_stop)
+			{
+			  thread = find_thread_ptid (ptid_build (pid, lwp, 0));
+
+			  /* Check if we actually attached to this thread.  It
+			     may have failed.  */
+			  if (thread)
+			    thread->last_resume_kind = resume_stop;
+			}
+
+		      if (debug_threads)
+			fprintf (stderr, "Found and attached to new lwp %ld\n", lwp);
+		    }
+		}
+
+	      if (!new_threads_found)
+		iterations++;
+	      else
+		iterations = 0;
+
+	      rewinddir (dir);
+	    }
+	  closedir (dir);
+	}
+    }
+
   return 0;
 }