[PATCH] Fix for follow-fork: followed child doesn't stop

Breazeal, Don donb@codesourcery.com
Fri Jun 13 01:07:00 GMT 2014


Hi Pedro,
Thanks for clarifying this for me.

On 6/6/2014 4:34 AM, Pedro Alves wrote:
> On 06/05/2014 10:44 PM, Breazeal, Don wrote:
>> +proc test_follow_fork { who detach cmd } {
--snip--
> Instead of explicitly prepending "$test_name: " to tests,
> please use with_test_prefix to make test messages unique.
> Below, runto_main, the untested line, this "set verbose", all
> these look like they'll print the same message in gdb.sum.
> 
Done.

--snip--
>> +    } elseif {$who == "parent" && $detach == "on"} {
>> +	set expected_re "Detaching after fork from .*set breakpoint here.*"
> 
> Something odd with indentation here.
> 
Fixed.

--snip--
>> +    # The first two tests should be sufficient to test the defaults.
>> +    # There is no need to test using the defaults in other permutations
>> +    # (e.g. "default" "on", "parent" "default", etc.).
>> +    set cases [list [list "default" "default" "next 2"]   \
>> +                    [list "default" "default" "continue"] \
>> +                    [list "parent"  "on"      "next 2"]   \
>> +                    [list "parent"  "on"      "continue"] \
>> +                    [list "child"   "on"      "next 2"]   \
>> +                    [list "child"   "on"      "continue"] \
>> +                    [list "parent"  "off"     "next 2"]   \
>> +                    [list "parent"  "off"     "continue"] \
>> +                    [list "child"   "off"     "next 2"]   \
>> +                    [list "child"   "off"     "continue"]]
>> +    foreach args $cases {
>> +        test_follow_fork [lindex $args 0] [lindex $args 1] [lindex $args 2]
>> +    }
> 
> This is still quite manual.  Please write:
> 
>     foreach cmd {"next 2" "continue"} {
>       test_follow_fork "default" "default" $cmd
>     }
> 
>     # Now test all explicit permutations.
>     foreach who {"parent" "child"} {
>       foreach detach {"on" "off"} {
> 	foreach cmd {"next 2" "continue"} {
> 	  test_follow_fork $who $detach $cmd
>         }
>       }
>     }
> 
Done.

On 6/6/2014 4:12 AM, Pedro Alves wrote:
> I think we should instead make the clone be disabled to begin with,
> so that it isn't marked as duplicate/inserted, and then re-enable it
> in follow_inferior_reset_breakpoints, _after_ it's been re_set
> to the child thread.

Here is the updated patch with (one hopes) fewer assumptions and errors.
 The problem statement has been corrected (detach-on-fork set to "off",
not "on", although in the end this was mostly irrelevant), the fix and
tests have been revised per comments, and the test slightly enhanced to
check that unfollowed inferiors can be resumed.

Symptom:

Using the test program gdb.base/foll-fork.c, with follow-fork-mode set
to "child" and detach-on-fork set to "off", stepping past the fork call
results in the child process running to completion, when it should just
finish the single step.  In addition, the single-step breakpoint is not
removed from the parent process, so if it is resumed it receives a SIGTRAP.

Cause:

No matter what the setting for detach-on-fork, when stepping past a
fork, the insertion state of the single-step breakpoint
(step_resume_breakpoint) is not handled correctly in the parent.  The SR
breakpoint is cloned for use with the child process. Before the clone is
associated with the child, it is treated as a duplicate of the original,
associated with the parent.  This results in the insertion state of the
original SR breakpoint and the clone being "swapped" by
breakpoint.c:update_global_location_list, so that the clone is marked as
inserted.

In the case where the parent is not detached, the two breakpoints remain
in the state (clone marked inserted, original not marked inserted).  As
a result the breakpoint is never inserted in the child, because although
the cloned SR breakpoint is eventually associated with the child, it is
marked as already inserted.  When the child is resumed, it runs to
completion.  Similarly, the breakpoint is never removed from the parent,
so that if it is resumed after the child exits, it gets a SIGTRAP.

In the case where the parent is detached (via target_detach), the clone
is still associated with the parent and marked as inserted, so the SR
breakpoint is removed from the parent and the clone's inserted flag is
cleared.  Although there are no visible symptoms, the maintenance of the
'inserted' flag is incorrect.

Fix:

Set the cloned breakpoint as disabled from the moment it is created.
This is done by modifying clone_momentary_breakpoint to take an
additional argument, LOC_ENABLED, which is used as the value of the
bp_location->enabled member.  The clone must be disabled at that point,
rather than later, because clone_momentary_breakpoint calls
update_global_location_list, which will treat the clone as a duplicate
of the original breakpoint if it is enabled.

All the calls to clone_momentary_breakpoint had to be modified to pass
'1' or '0'.  I looked at implementing an enum for the enabled member,
but concluded that readability would suffer because there are so many
places it is used as a boolean, e.g. "if (bl->enabled)".

In follow_inferior_reset_breakpoints, the clone is set to enabled once
it has been associated with the child process.  With this, the
bp_location 'inserted' member is maintained correctly throughout the
follow-fork procedure and the behavior is as expected.

The same treatment is given to the exception_resume_breakpoint when
following a fork.

Testing:

Ran 'make check' on Linux x64.

Along with the fix above, the coverage of the follow-fork test
gdb.base/foll-fork.exp was expanded to:

1) cover all the combinations of values for follow-fork-mode and
detach-on-fork

2) make sure that both user breakpoints and single-step breakpoints are
propagated correctly to the child

3) check that the inferior list has the expected contents after
following the fork.

4) check that unfollowed, undetached inferiors can be resumed.


Thanks
--Don


gdb/

2014-06-12  Don Breazeal  <donb@codesourcery.com>

	* breakpoint.c (set_longjmp_breakpoint): Call
	momentary_breakpoint_from_master with additional argument.
	(set_longjmp_breakpoint_for_call_dummy): Call
	momentary_breakpoint_from_master with additional argument.
	(set_std_terminate_breakpoint): Call
	momentary_breakpoint_from_master with additional argument.
	(momentary_breakpoint_from_master): Add argument to function
	definition and use it to initialize structure member flag.
	(clone_momentary_breakpoint): Call
	momentary_breakpoint_from_master with additional argument.
	* infrun.c (follow_inferior_reset_breakpoints): Clear structure
	member flags set in momentary_breakpoint_from_master.

gdb/testsuite/

2014-06-12  Don Breazeal  <donb@codesourcery.com>

	* gdb.base/foll-fork.exp (default_fork_parent_follow):
	Deleted procedure.
	(explicit_fork_parent_follow): Deleted procedure.
	(explicit_fork_child_follow): Deleted procedure.
	(test_follow_fork): New procedure.
	(do_fork_tests): Replace calls to deleted procedures with
	calls to test_follow_fork and reset GDB for subsequent
	procedure calls.

---
 gdb/breakpoint.c                     |   20 ++--
 gdb/infrun.c                         |   15 ++-
 gdb/testsuite/gdb.base/foll-fork.exp |  193
+++++++++++++++++++++++-----------
 3 files changed, 156 insertions(+), 72 deletions(-)

diff --git a/gdb/breakpoint.c b/gdb/breakpoint.c
index 23c8895..11fe5a7 100644
--- a/gdb/breakpoint.c
+++ b/gdb/breakpoint.c
@@ -154,7 +154,8 @@ struct breakpoint *set_raw_breakpoint (struct
gdbarch *gdbarch,
 static struct breakpoint *
   momentary_breakpoint_from_master (struct breakpoint *orig,
 				    enum bptype type,
-				    const struct breakpoint_ops *ops);
+				    const struct breakpoint_ops *ops,
+				    int loc_enabled);

 static void breakpoint_adjustment_warning (CORE_ADDR, CORE_ADDR, int, int);

@@ -7390,7 +7391,7 @@ set_longjmp_breakpoint (struct thread_info *tp,
struct frame_id frame)
 	/* longjmp_breakpoint_ops ensures INITIATING_FRAME is cleared again
 	   after their removal.  */
 	clone = momentary_breakpoint_from_master (b, type,
-						  &longjmp_breakpoint_ops);
+						  &longjmp_breakpoint_ops, 1);
 	clone->thread = thread;
       }

@@ -7440,7 +7441,8 @@ set_longjmp_breakpoint_for_call_dummy (void)
 	struct breakpoint *new_b;

 	new_b = momentary_breakpoint_from_master (b, bp_longjmp_call_dummy,
-						  &momentary_breakpoint_ops);
+						  &momentary_breakpoint_ops,
+						  1);
 	new_b->thread = pid_to_thread_id (inferior_ptid);

 	/* Link NEW_B into the chain of RETVAL breakpoints.  */
@@ -7533,7 +7535,7 @@ set_std_terminate_breakpoint (void)
 	&& b->type == bp_std_terminate_master)
       {
 	momentary_breakpoint_from_master (b, bp_std_terminate,
-					  &momentary_breakpoint_ops);
+					  &momentary_breakpoint_ops, 1);
       }
 }

@@ -9051,13 +9053,14 @@ set_momentary_breakpoint (struct gdbarch
*gdbarch, struct symtab_and_line sal,
 }

 /* Make a momentary breakpoint based on the master breakpoint ORIG.
-   The new breakpoint will have type TYPE, and use OPS as it
-   breakpoint_ops.  */
+   The new breakpoint will have type TYPE, use OPS as its
+   breakpoint_ops, and will set enabled to LOC_ENABLED.  */

 static struct breakpoint *
 momentary_breakpoint_from_master (struct breakpoint *orig,
 				  enum bptype type,
-				  const struct breakpoint_ops *ops)
+				  const struct breakpoint_ops *ops,
+				  int loc_enabled)
 {
   struct breakpoint *copy;

@@ -9073,6 +9076,7 @@ momentary_breakpoint_from_master (struct
breakpoint *orig,
   copy->loc->probe = orig->loc->probe;
   copy->loc->line_number = orig->loc->line_number;
   copy->loc->symtab = orig->loc->symtab;
+  copy->loc->enabled = loc_enabled;
   copy->frame_id = orig->frame_id;
   copy->thread = orig->thread;
   copy->pspace = orig->pspace;
@@ -9095,7 +9099,7 @@ clone_momentary_breakpoint (struct breakpoint *orig)
   if (orig == NULL)
     return NULL;

-  return momentary_breakpoint_from_master (orig, orig->type, orig->ops);
+  return momentary_breakpoint_from_master (orig, orig->type, orig->ops, 0);
 }

 struct breakpoint *
diff --git a/gdb/infrun.c b/gdb/infrun.c
index 47604c7..4e808d9 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -569,7 +569,9 @@ follow_inferior_reset_breakpoints (void)

   /* Was there a step_resume breakpoint?  (There was if the user
      did a "next" at the fork() call.)  If so, explicitly reset its
-     thread number.
+     thread number.  Cloned step_resume breakpoints are disabled on
+     creation, so enable it here now that it is associated with the
+     correct thread.

      step_resumes are a form of bp that are made to be per-thread.
      Since we created the step_resume bp when the parent process
@@ -579,10 +581,17 @@ follow_inferior_reset_breakpoints (void)
      it is for, or it'll be ignored when it triggers.  */

   if (tp->control.step_resume_breakpoint)
-    breakpoint_re_set_thread (tp->control.step_resume_breakpoint);
+    {
+      breakpoint_re_set_thread (tp->control.step_resume_breakpoint);
+      tp->control.step_resume_breakpoint->loc->enabled = 1;
+    }

+  /* Treat exception_resume breakpoints like step_resume breakpoints.  */
   if (tp->control.exception_resume_breakpoint)
-    breakpoint_re_set_thread (tp->control.exception_resume_breakpoint);
+    {
+      breakpoint_re_set_thread (tp->control.exception_resume_breakpoint);
+      tp->control.exception_resume_breakpoint->loc->enabled = 1;
+    }

   /* Reinsert all breakpoints in the child.  The user may have set
      breakpoints after catching the fork, in which case those
diff --git a/gdb/testsuite/gdb.base/foll-fork.exp
b/gdb/testsuite/gdb.base/foll-fork.exp
index e1201d7..ad8b750 100644
--- a/gdb/testsuite/gdb.base/foll-fork.exp
+++ b/gdb/testsuite/gdb.base/foll-fork.exp
@@ -53,60 +53,130 @@ proc check_fork_catchpoints {} {
   }
 }

-proc default_fork_parent_follow {} {
+# Test follow-fork to ensure that the correct process is followed, that
+# the followed process stops where it is expected to stop, that processes
+# are detached (or not) as expected, and that the inferior list has the
+# expected contents after following the fork.  WHO is the argument to
+# the 'set follow-fork-mode' command, DETACH is the argument to the
+# 'set detach-on-fork' command, and CMD is the GDB command used to
+# execute the program past the fork.  If the value of WHO or DETACH is
+# 'default', the corresponding GDB command is skipped for that test.
+# The value of CMD must be either 'next 2' or 'continue'.
+proc test_follow_fork { who detach cmd } {
     global gdb_prompt
+    global srcfile
+    global testfile

-    gdb_test "show follow-fork" \
-	"Debugger response to a program call of fork or vfork is \"parent\".*" \
-	"default show parent follow, no catchpoints"
+    with_test_prefix "follow $who, detach $detach, command \"$cmd\"" {

-    gdb_test "next 2" \
-	"Detaching after fork from.*" \
-	"default parent follow, no catchpoints"
+	# Start a new debugger session each time so defaults are legitimate.
+	clean_restart $testfile

-    # The child has been detached; allow time for any output it might
-    # generate to arrive, so that output doesn't get confused with
-    # any expected debugger output from a subsequent testpoint.
-    #
-    exec sleep 1
-}
+	if ![runto_main] {
+	    untested "could not run to main"
+	    return -1
+	}

-proc explicit_fork_parent_follow {} {
-    global gdb_prompt
+	# The "Detaching..." and "Attaching..." messages may be hidden by
+	# default.
+	gdb_test_no_output "set verbose"

-    gdb_test_no_output "set follow-fork parent"
+	# Set follow-fork-mode if we aren't using the default.
+	if {$who == "default"} {
+	    set who "parent"
+	} else {
+	    gdb_test_no_output "set follow-fork $who"
+	}

-    gdb_test "show follow-fork" \
-	"Debugger response to a program call of fork or vfork is \"parent\"." \
-	"explicit show parent follow, no catchpoints"
+	gdb_test "show follow-fork" \
+	"Debugger response to a program call of fork or vfork is \"$who\"." \
+	"show follow-fork"

-    gdb_test "next 2" "Detaching after fork from.*" \
-	"explicit parent follow, no catchpoints"
+	# Set detach-on-fork mode if we aren't using the default.
+	if {$detach == "default"} {
+	    set detach "on"
+	} else {
+	    gdb_test_no_output "set detach-on-fork $detach"
+	}

-    # The child has been detached; allow time for any output it might
-    # generate to arrive, so that output doesn't get confused with
-    # any expected debugger output from a subsequent testpoint.
-    #
-    exec sleep 1
-}
+	gdb_test "show detach-on-fork" \
+		 "Whether gdb will detach.* fork is $detach." \
+		 "show detach-on-fork"
+
+	# Set a breakpoint after the fork if we aren't single-stepping
+	# past the fork.
+	if {$cmd == "continue"} {
+	    set bp_after_fork [gdb_get_line_number "set breakpoint here"]
+	    gdb_test "break ${srcfile}:$bp_after_fork" \
+		     "Breakpoint.*, line $bp_after_fork.*" \
+		     "set breakpoint after fork"
+	}

-proc explicit_fork_child_follow {} {
-    global gdb_prompt
+	# Set up the output we expect to see after we run.
+	set expected_re ""
+	if {$who == "child"} {
+	    set expected_re "Attaching after.* fork to.*set breakpoint here.*"
+	} elseif {$who == "parent" && $detach == "on"} {
+	    set expected_re "Detaching after fork from .*set breakpoint here.*"
+	} else {
+	    set expected_re ".*set breakpoint here.*"
+	}

-    gdb_test_no_output "set follow-fork child"
+	# Test running past and following the fork, using the parameters
+	# set above.
+	gdb_test $cmd $expected_re "$cmd past fork"

-    gdb_test "show follow-fork" \
-	"Debugger response to a program call of fork or vfork is \"child\"." \
-	"explicit show child follow, no catchpoints"
+	# Check that we have the inferiors arranged correctly after
+	# following the fork.
+	set resume_unfollowed 0
+	if {$who == "parent" && $detach == "on"} {

-    gdb_test "next 2" "Attaching after.* fork to.*" \
-	"explicit child follow, no catchpoints"
+	    # Follow parent / detach child: the only inferior is the parent.
+	    gdb_test "info inferiors" "\\* 1 .* process.*" \
+		     "info inferiors"

-    # The child has been detached; allow time for any output it might
-    # generate to arrive, so that output doesn't get confused with
-    # any gdb_expected debugger output from a subsequent testpoint.
-    #
-    exec sleep 1
+	} elseif {$who == "parent" && $detach == "off"} {
+
+	    # Follow parent / keep child: two inferiors under debug, the
+	    # parent is the current inferior.
+	    gdb_test "info inferiors" " 2 .*process.*\\* 1 .*process.*" \
+		     "info inferiors"
+
+	    gdb_test "inferior 2" "Switching to inferior 2 .*"
+	    set resume_unfollowed 1
+
+	} elseif {$who == "child" && $detach == "on"} {
+
+	    # Follow child / detach parent: the child is under debug and is
+	    # the current inferior.  The parent is listed but is not under
+	    # debug.
+	    gdb_test "info inferiors" "\\* 2 .*process.* 1 .*<null>.*" \
+		     "info inferiors"
+
+	} elseif {$who == "child" && $detach == "off"} {
+
+	    # Follow child / keep parent: two inferiors under debug, the
+	    # child is the current inferior.
+	    gdb_test "info inferiors" "\\* 2 .*process.* 1 .*process.*" \
+		     "info inferiors"
+
+	    gdb_test "inferior 1" "Switching to inferior 1 .*"
+	    set resume_unfollowed 1
+	}
+
+	if {$resume_unfollowed == 1} {
+	    if {$cmd == "next 2"} {
+
+		gdb_continue_to_end "continue unfollowed inferior to end"
+
+	    } elseif {$cmd == "continue"} {
+
+		gdb_continue_to_breakpoint \
+		    "continue unfollowed inferior to bp" \
+		    ".* set breakpoint here.*"
+	    }
+	}
+    }
 }

 proc catch_fork_child_follow {} {
@@ -254,6 +324,7 @@ proc tcatch_fork_parent_follow {} {

 proc do_fork_tests {} {
     global gdb_prompt
+    global testfile

     # Verify that help is available for "set follow-fork-mode".
     #
@@ -286,31 +357,31 @@ By default, the debugger will follow the parent
process..*" \
     # fork-following is supported.
     if [runto_main] then { check_fork_catchpoints }

-    # Test the default behaviour, which is to follow the parent of a
-    # fork, and detach from the child.  Do this without catchpoints.
+    # Test the basic follow-fork functionality using all combinations of
+    # values for follow-fork-mode and detach-on-fork, using either a
+    # breakpoint or single-step to execute past the fork.
     #
-    if [runto_main] then { default_fork_parent_follow }
-
-    # Test the ability to explicitly follow the parent of a fork, and
-    # detach from the child.  Do this without catchpoints.
-    #
-    if [runto_main] then { explicit_fork_parent_follow }
+    # The first loop should be sufficient to test the defaults.  There
+    # is no need to test using the defaults in other permutations (e.g.
+    # "default" "on", "parent" "default", etc.).
+    foreach cmd {"next 2" "continue"} {
+        test_follow_fork "default" "default" $cmd
+    }

-    # Test the ability to follow the child of a fork, and detach from
-    # the parent.  Do this without catchpoints.
-    #
-    if [runto_main] then { explicit_fork_child_follow }
+    # Now test all explicit permutations.
+    foreach who {"parent" "child"} {
+	foreach detach {"on" "off"} {
+	    foreach cmd {"next 2" "continue"} {
+		test_follow_fork $who $detach $cmd
+	    }
+	}
+    }

-    # Test the ability to follow both child and parent of a fork.  Do
-    # this without catchpoints.
-    # ??rehrauer: NYI.  Will add testpoints here when implemented.
-    #
+    # Catchpoint tests.

-    # Test the ability to have the debugger ask the user at fork-time
-    # whether to follow the parent, child or both.  Do this without
-    # catchpoints.
-    # ??rehrauer: NYI.  Will add testpoints here when implemented.
-    #
+    # Restart to eliminate any effects of the follow-fork tests.
+    clean_restart $testfile
+    gdb_test_no_output "set verbose"

     # Test the ability to catch a fork, specify that the child be
     # followed, and continue.  Make the catchpoint permanent.
-- 
1.7.0.4




More information about the Gdb-patches mailing list