[PATCH v6 3/6] Extended-remote Linux follow fork
Pedro Alves
palves@redhat.com
Tue Mar 24 12:17:00 GMT 2015
Hi Don,
On 03/17/2015 08:56 PM, Don Breazeal wrote:
> * using a new predicate in gdbserver to control handling of the fork event
> (and eventually all extended events) in linux_wait_1. The predicate,
> extended_event_reported, checks a target_waitstatus.kind for an
> extended ptrace event.
>
> * implementing a new RSP 'T' Stop Reply Packet stop reason: "fork", in
> gdbserver/remote-utils.c and remote.c.
>
> * implementing new target and RSP support for target_follow_fork with
> target extended-remote. (The RSP components were actually defined in
> patch 4, but they see their first use here).
This description seems to be a bit out of date. We're in patch 3 still. :-)
>
> - remote target routine remote_follow_fork, which just sends the 'D;pid'
> detach packet to detach the new fork child cleanly. We can't just
> call target_detach because the data structures for the forked child
> have not been allocated on the host side.
>
> Tested on x64 Ubuntu Lucid, native, remote, extended-remote.
>
> gdb/doc/
> 2015-03-17 Don Breazeal <donb@codesourcery.com>
>
> * gdb.texinfo (Forks): List targets that have fork debugging
> support.
> (Stop Reply Packets): List new stop reason "xfork".
I know I was the one who suggested avoiding "fork" because it starts
with an hex char, but thinking further, if gdb must report
support for fork events for the server to report these, it's ok
to use "fork". I'm no thinking we can rename it back, thus
fending off questions about this in the future. Could you
do that? Sorry about the back and forth...
>
> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index b089b1b..b62b37f 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -3142,6 +3142,9 @@ create additional processes using the @code{fork} or @code{vfork} functions.
> Currently, the only platforms with this feature are HP-UX (11.x and later
> only?) and @sc{gnu}/Linux (kernel version 2.5.60 and later).
>
> +The fork debugging commands are supported in both native mode and when
> +connected to @code{gdbserver} using @kbd{target extended-remote}.
> +
> By default, when a program forks, @value{GDBN} will continue to debug
> the parent process and the child process will run unimpeded.
>
> @@ -35197,6 +35200,12 @@ The packet indicates that the loaded libraries have changed.
> @value{GDBN} should use @samp{qXfer:libraries:read} to fetch a new
> list of loaded libraries. The @var{r} part is ignored.
>
> +@cindex fork events, remote reply
> +@item xfork
> +The packet indicates that @code{fork} was called, and @var{r}
> +is the ptid of the new child process. This packet is only
> +applicable to targets that support fork events.
> +
Please mention here the need for reporting support in qSupported.
E.g., the new swbreak stop reason says:
~~~
This packet should not be sent by default; older @value{GDBN} versions
did not support it. @value{GDBN} requests it, by supplying an
appropriate @samp{qSupported} feature (@pxref{qSupported}). The
remote stub must also supply the appropriate @samp{qSupported} feature
indicating support.
~~~
> if (debug_threads)
> debug_printf ("HEW: Got clone event "
> "from LWP %ld, new child is LWP %ld\n",
> @@ -434,7 +487,12 @@ handle_extended_wait (struct lwp_info *event_child, int wstat)
> new_lwp->status_pending_p = 1;
> new_lwp->status_pending = status;
> }
> +
> + /* Don't report the event. */
> + return 1;
> }
> +
> + internal_error (__FILE__, __LINE__, _("unknown ptrace event %d"), event);
> }
>
> /* Return the PC as read from the regcache of LWP, without any
> @@ -1828,6 +1886,19 @@ check_stopped_by_watchpoint (struct lwp_info *child)
> return child->stop_reason == TARGET_STOPPED_BY_WATCHPOINT;
> }
>
> +/* Wrapper for linux_enable_event_reporting that disables any
> + supported events that we have determined should not be
> + reported (e.g. GDB did not request them). */
> +
> +static void
> +linux_low_enable_events (pid_t pid, int attached)
> +{
> + if (!report_fork_events)
> + linux_ptrace_clear_flags (PTRACE_O_TRACEFORK);
> +
> + linux_enable_event_reporting (pid, attached);
> +}
> +
> /* Do low-level handling of the event, and check if we should go on
> and pass it to caller code. Return the affected lwp if we are, or
> NULL otherwise. */
> @@ -1912,11 +1983,11 @@ linux_low_filter_event (int lwpid, int wstat)
> }
> }
>
> - if (WIFSTOPPED (wstat) && child->must_set_ptrace_flags)
> + if (WIFSTOPPED (wstat) && child->must_set_ptrace_flags && gdb_connected ())
> {
I don't really understand this. If the flag is set, why would it matter
whether gdb is connected?
> struct process_info *proc = find_process_pid (pid_of (thread));
>
> - linux_enable_event_reporting (lwpid, proc->attached);
> + linux_low_enable_events (lwpid, proc->attached);
> child->must_set_ptrace_flags = 0;
> }
>
> @@ -1926,8 +1997,12 @@ linux_low_filter_event (int lwpid, int wstat)
> && linux_is_extended_waitstatus (wstat))
> {
> child->stop_pc = get_pc (child);
> - handle_extended_wait (child, wstat);
> - return NULL;
> + if (handle_extended_wait (child, wstat))
> + {
> + /* The event has been handled, so just return without
> + reporting it. */
> + return NULL;
> + }
> }
>
> /* Check first whether this was a SW/HW breakpoint before checking
> @@ -2502,6 +2577,18 @@ ignore_event (struct target_waitstatus *ourstatus)
> return null_ptid;
> }
>
> +/* Return non-zero if WAITSTATUS reflects an extended linux
> + event. Otherwise, return zero. */
> +
> +static int
> +extended_event_reported (const struct target_waitstatus *waitstatus)
> +{
> + if (waitstatus == NULL)
> + return 0;
> +
> + return (waitstatus->kind == TARGET_WAITKIND_FORKED);
> +}
> +
> /* Wait for process, returns status. */
>
> static ptid_t
> @@ -2868,7 +2955,8 @@ linux_wait_1 (ptid_t ptid,
> && !bp_explains_trap && !trace_event)
> || (gdb_breakpoint_here (event_child->stop_pc)
> && gdb_condition_true_at_breakpoint (event_child->stop_pc)
> - && gdb_no_commands_at_breakpoint (event_child->stop_pc)));
> + && gdb_no_commands_at_breakpoint (event_child->stop_pc))
> + || extended_event_reported (&event_child->waitstatus));
>
> run_breakpoint_commands (event_child->stop_pc);
>
> @@ -2890,6 +2978,13 @@ linux_wait_1 (ptid_t ptid,
> paddress (event_child->stop_pc),
> paddress (event_child->step_range_start),
> paddress (event_child->step_range_end));
> + if (extended_event_reported (&event_child->waitstatus))
> + {
> + char *str = target_waitstatus_to_string (ourstatus);
> + debug_printf ("LWP %ld: extended event with waitstatus %s\n",
> + lwpid_of (get_lwp_thread (event_child)), str);
> + xfree (str);
> + }
> }
>
> /* We're not reporting this breakpoint to GDB, so apply the
> @@ -2999,7 +3094,17 @@ linux_wait_1 (ptid_t ptid,
> unstop_all_lwps (1, event_child);
> }
>
> - ourstatus->kind = TARGET_WAITKIND_STOPPED;
> + if (extended_event_reported (&event_child->waitstatus))
> + {
> + /* If the reported event is a fork, vfork or exec, let GDB know. */
> + ourstatus->kind = event_child->waitstatus.kind;
> + ourstatus->value = event_child->waitstatus.value;
> +
> + /* Reset the event lwp's waitstatus since we handled it already. */
> + event_child->waitstatus.kind = TARGET_WAITKIND_SPURIOUS;
I'd think this should be TARGET_WAITKIND_IGNORE? That's the usual
status for "clear".
> + }
> + else
> + ourstatus->kind = TARGET_WAITKIND_STOPPED;
>
> /* Now that we've selected our final event LWP, un-adjust its PC if
> it was a software breakpoint, and the client doesn't know we can
> @@ -3032,7 +3137,7 @@ linux_wait_1 (ptid_t ptid,
> but, it stopped for other reasons. */
> ourstatus->value.sig = gdb_signal_from_host (WSTOPSIG (w));
> }
> - else
> + else if (ourstatus->kind == TARGET_WAITKIND_STOPPED)
> {
> ourstatus->value.sig = gdb_signal_from_host (WSTOPSIG (w));
> }
> @@ -4830,8 +4935,8 @@ linux_write_memory (CORE_ADDR memaddr, const unsigned char *myaddr, int len)
> val = val & 0xffff;
> else if (len == 3)
> val = val & 0xffffff;
> - debug_printf ("Writing %0*x to 0x%08lx\n", 2 * ((len < 4) ? len : 4),
> - val, (long)memaddr);
> + debug_printf ("Writing %0*x to 0x%08lx in process %d\n",
> + 2 * ((len < 4) ? len : 4), val, (long)memaddr, pid);
> }
>
> /* Fill start and end extra bytes of buffer with existing memory data. */
> @@ -5271,6 +5376,39 @@ linux_supports_vfork_events (void)
> return linux_supports_tracefork ();
> }
>
> +/* Callback for 'find_inferior'. Set the (possibly changed) ptrace
> + options for the specified lwp. */
> +
> +static int
> +reset_lwp_ptrace_options_callback (struct inferior_list_entry *entry,
> + void *args)
> +{
> + struct thread_info *thread = (struct thread_info *) entry;
> + struct lwp_info *lwp = get_thread_lwp (thread);
> + struct process_info *proc = find_process_pid (pid_of (thread));
> +
> + linux_low_enable_events (lwpid_of (thread), proc->attached);
> + lwp->must_set_ptrace_flags = 0;
> +
> + return 0;
> +}
> +
> +/* Target hook for 'handle_new_gdb_connection'. Causes a reset of the
> + ptrace flags for all inferiors. This is in case the new GDB connection
> + doesn't support the same set of events that the previous one did. */
> +
> +static void
> +linux_handle_new_gdb_connection (void)
> +{
> + pid_t pid;
> +
> + /* Reset the ptrace options to enable on the inferior(s). */
> + linux_reset_ptrace_options ();
> +
> + /* Request that all the lwps reset their ptrace options. */
> + find_inferior (&all_threads, reset_lwp_ptrace_options_callback , &pid);
Spurious space before ', &pid'. But, you can't do this if threads
are already running, such as when you reconnect after you left the
target running for disconnected tracing. Instead, you need to
force threads to momentarily pause and set their must_set_ptrace_flags
flag, much like when we need to change running thread's debug registers
for watchpoints. See linux-x86-low.c:update_debug_registers_callback.
>
> diff --git a/gdb/nat/linux-ptrace.c b/gdb/nat/linux-ptrace.c
> index aba3da8..3494bcf 100644
> --- a/gdb/nat/linux-ptrace.c
> +++ b/gdb/nat/linux-ptrace.c
> @@ -25,13 +25,16 @@
>
> #include <stdint.h>
>
> -/* Stores the currently supported ptrace options. A value of
> - -1 means we did not check for features yet. A value of 0 means
> - there are no supported features. */
> +/* Stores the ptrace options currently enabled by this gdbserver.
Talking about "this gdbserver" here doesn't make sense. This file
is used by native gdb too.
> + A value of -1 means we did not check for features yet. A value
> + of 0 means there are no enabled features. */
> static int current_ptrace_options = -1;
>
> -/* Additional flags to test. */
> +/* Stores the fork and exec ptrace options supported by the operating
> + system. */
> +static int supported_additional_flags = 0;
>
> +/* Additional flags to test. */
> static int additional_flags;
>
> /* Find all possible reasons we could fail to attach PID and append
> @@ -344,6 +347,7 @@ linux_check_ptrace_features (void)
>
> /* Initialize the options. */
> current_ptrace_options = 0;
> + supported_additional_flags = 0;
>
> /* Fork a child so we can do some testing. The child will call
> linux_child_function and will get traced. The child will
> @@ -394,7 +398,7 @@ linux_test_for_tracesysgood (int child_pid)
> (PTRACE_TYPE_ARG4) PTRACE_O_TRACESYSGOOD);
>
> if (ret == 0)
> - current_ptrace_options |= PTRACE_O_TRACESYSGOOD;
> + supported_additional_flags |= PTRACE_O_TRACESYSGOOD;
> }
>
> /* Determine if PTRACE_O_TRACEFORK can be used to follow fork
> @@ -421,7 +425,7 @@ linux_test_for_tracefork (int child_pid)
> (PTRACE_TYPE_ARG4) (PTRACE_O_TRACEFORK
> | PTRACE_O_TRACEVFORKDONE));
> if (ret == 0)
> - current_ptrace_options |= PTRACE_O_TRACEVFORKDONE;
> + supported_additional_flags |= PTRACE_O_TRACEVFORKDONE;
> }
>
> /* Setting PTRACE_O_TRACEFORK did not cause an error, however we
> @@ -459,9 +463,16 @@ linux_test_for_tracefork (int child_pid)
> /* We got the PID from the grandchild, which means fork
> tracing is supported. */
> current_ptrace_options |= PTRACE_O_TRACECLONE;
> - current_ptrace_options |= (additional_flags & (PTRACE_O_TRACEFORK
> - | PTRACE_O_TRACEVFORK
> - | PTRACE_O_TRACEEXEC));
> +
> + /* Save the "extended" options in case we need to reset
> + the options later for a connect from a different GDB. */
> + supported_additional_flags |= (PTRACE_O_TRACEFORK
> + | PTRACE_O_TRACEVFORK
> + | PTRACE_O_TRACEEXEC);
> +
> + /* Enable only those options requested by GDB. */
> + current_ptrace_options |= (supported_additional_flags
> + & additional_flags);
>
> /* Do some cleanup and kill the grandchild. */
> my_waitpid (second_pid, &second_status, 0);
> @@ -518,6 +529,25 @@ linux_enable_event_reporting (pid_t pid, int attached)
> (PTRACE_TYPE_ARG4) (uintptr_t) ptrace_options);
> }
>
> +/* Reset the ptrace options using potentially different
> + additional_flags. */
> +
> +void
> +linux_reset_ptrace_options (void)
> +{
> + /* Check if we have initialized the ptrace features for this
> + target. If not, do it now. */
> + if (current_ptrace_options == -1)
> + linux_check_ptrace_features ();
> +
> + /* Clear only the "extended" options. */
> + linux_ptrace_clear_flags (all_additional_flags);
> +
> + /* Add the new requested flags back in. */
> + current_ptrace_options |= (supported_additional_flags
> + & additional_flags);
> +}
> +
> /* Disable reporting of all currently supported ptrace events. */
>
> void
> @@ -603,9 +633,20 @@ linux_ptrace_init_warnings (void)
> void
> linux_ptrace_set_additional_flags (int flags)
> {
> + int invalid_flags = (flags & ~all_additional_flags);
> +
> + gdb_assert (invalid_flags == 0);
> additional_flags = flags;
> }
>
> +/* Clear FLAGS in current_ptrace_options. */
> +
> +void
> +linux_ptrace_clear_flags (int flags)
> +{
> + current_ptrace_options &= ~flags;
> +}
> +
This still looks a bit messy and mostly unnecessary to me. :-/
Why not something based on this? In linux-ptrace.c:
/* Enable reporting of all currently supported ptrace events.
OPTIONS is a bit mask of extended features we want enabled,
if supported by the kernel. PTRACE_O_TRACECLONE is always
enabled, if supported. */
void
linux_enable_event_reporting (pid_t pid, int options)
{
/* Check if we have initialized the ptrace features for this
target. If not, do it now. */
if (current_ptrace_options == -1)
linux_check_ptrace_features ();
/* We always want clone events. */
options |= PTRACE_O_TRACECLONE;
/* Filter out unsupported options. */
options &= current_ptrace_options;
/* Set the options. */
ptrace (PTRACE_SETOPTIONS, pid, (PTRACE_TYPE_ARG3) 0,
(PTRACE_TYPE_ARG4) (uintptr_t) options);
}
(current_ptrace_options would be renamed to
supported_ptrace_options or some such afterwards.)
and then in linux-low.c:
if (WIFSTOPPED (wstat) && child->must_set_ptrace_flags)
{
struct process_info *proc = find_process_pid (pid_of (thread));
int options = 0;
if (!proc->attached)
options |= PTRACE_O_EXITKILL;
if (report_fork_events)
options |= PTRACE_O_TRACEFORK;
if (report_vfork_events)
options |= (PTRACE_O_TRACEVFORKDONE | PTRACE_O_TRACEVFORK);
if (report_exec_events)
options |= PTRACE_O_TRACEEXEC;
linux_enable_event_reporting (pid, options);
child->must_set_ptrace_flags = 0;
}
That'd cope with a different gdb reconnecting and requesting
different options too.
> -/* This takes a program previously attached to and detaches it. After
> - this is done, GDB can be used to debug some other program. We
> - better not have left any breakpoints in the target program or it'll
> - die when it hits one. */
> +/* This detaches a program to which we previously attached, using
> + inferior_ptid to identify the process. After this is done, GDB
> + can be used to debug some other program. We better not have left
> + any breakpoints in the target program or it'll die when it hits
> + one. If IS_FORK_CHILD is true, then inferior_ptid is the child
> + of an unfollowed fork, and we need to avoid deleting breakpoints
> + still needed by the parent. */
>
> static void
> -remote_detach_1 (const char *args, int from_tty, int extended)
> +remote_detach_1 (struct target_ops *ops, const char *args,
> + int from_tty, int is_fork_child)
> {
> int pid = ptid_get_pid (inferior_ptid);
> struct remote_state *rs = get_remote_state ();
> + struct thread_info *tp = find_thread_ptid (inferior_ptid);
> + int is_fork_parent;
>
> if (args)
> error (_("Argument given to \"detach\" when remotely debugging."));
> @@ -4468,22 +4482,75 @@ remote_detach_1 (const char *args, int from_tty, int extended)
> else
> error (_("Can't detach process."));
>
> - if (from_tty && !extended)
> + if (from_tty && !rs->extended)
> puts_filtered (_("Ending remote debugging.\n"));
>
> - target_mourn_inferior ();
> + /* Check to see if we are detaching a fork parent. Note that if we
> + are detaching a fork child, tp == NULL. */
> + if (tp != NULL)
> + is_fork_parent = tp->pending_follow.kind == TARGET_WAITKIND_FORKED;
> +
> + /* If doing detach-on-fork, we don't mourn, because that will delete
> + breakpoints that should be available for the followed inferior. */
> + if (!is_fork_child && !is_fork_parent)
> + target_mourn_inferior ();
> + else
> + {
> + inferior_ptid = null_ptid;
> + detach_inferior (pid);
> + }
> }
This looks quite confusing to me. Please instead simply factor out
the bits that actually do the remote detach to a separate
function, like:
static void
remote_detach_pid (int pid)
{
struct remote_state *rs = get_remote_state ();
if (remote_multi_process_p (rs))
xsnprintf (rs->buf, get_remote_packet_size (), "D;%x", pid);
else
strcpy (rs->buf, "D");
putpkt (rs->buf);
getpkt (&rs->buf, &rs->buf_size, 0);
if (rs->buf[0] == 'O' && rs->buf[1] == 'K')
;
else if (rs->buf[0] == '\0')
error (_("Remote doesn't know how to detach"));
else
error (_("Can't detach process."));
}
And use that from both remote_detach_1 and the detach-child-fork
case.
> +static int
> +remote_follow_fork (struct target_ops *ops, int follow_child,
> + int detach_fork)
> +{
> + struct remote_state *rs = get_remote_state ();
> +
> + if (remote_fork_event_p (rs))
> + {
> + if (detach_fork && !follow_child)
Aren't we missing the "detach_fork && follow_child" case?
> + {
> + ptid_t parent_ptid;
> + ptid_t child_ptid;
> +
> + gdb_assert (inferior_thread ()->pending_follow.kind
> + == TARGET_WAITKIND_FORKED);
> +
> + /* remote_detach_1 detaches inferior_ptid, which is currently
> + the ptid of the parent. Switch inferior_ptid to the ptid
> + of the child for the duration of the call. */
> + parent_ptid = inferior_ptid;
> + child_ptid = inferior_thread ()->pending_follow.value.related_pid;
> + inferior_ptid = child_ptid;
> + gdb_assert (!ptid_equal (parent_ptid, child_ptid));
> +
> + remote_detach_1 (ops, NULL, 0, 1);
This then ends up being just remote_detach_pid + detach_inferior, AFAICS.
> +
> + /* Restore the parent ptid. */
> + inferior_ptid = parent_ptid;
> + }
> + }
> + return 0;
> }
>
> /* Same as remote_detach, but don't send the "D" packet; just disconnect. */
> @@ -5652,6 +5719,11 @@ Packet: '%s'\n"),
> p = unpack_varlen_hex (++p1, &c);
> event->core = c;
> }
> + else if (strncmp (p, "xfork", p1 - p) == 0)
> + {
> + event->ws.value.related_pid = read_ptid (++p1, &p);
> + event->ws.kind = TARGET_WAITKIND_FORKED;
> + }
> else
> {
> ULONGEST pnum;
> @@ -9514,8 +9586,11 @@ remote_pid_to_str (struct target_ops *ops, ptid_t ptid)
> if (ptid_equal (magic_null_ptid, ptid))
> xsnprintf (buf, sizeof buf, "Thread <main>");
> else if (rs->extended && remote_multi_process_p (rs))
> - xsnprintf (buf, sizeof buf, "Thread %d.%ld",
> - ptid_get_pid (ptid), ptid_get_lwp (ptid));
> + if (ptid_get_lwp (ptid) == 0)
> + return normal_pid_to_str (ptid);
> + else
> + xsnprintf (buf, sizeof buf, "Thread %d.%ld",
> + ptid_get_pid (ptid), ptid_get_lwp (ptid));
> else
> xsnprintf (buf, sizeof buf, "Thread %ld",
> ptid_get_lwp (ptid));
> @@ -11938,6 +12013,7 @@ Specify the serial device it is connected to (e.g. /dev/ttya).";
> extended_remote_ops.to_kill = extended_remote_kill;
> extended_remote_ops.to_supports_disable_randomization
> = extended_remote_supports_disable_randomization;
> + extended_remote_ops.to_follow_fork = remote_follow_fork;
> }
>
> static int
> diff --git a/gdb/testsuite/gdb.base/multi-forks.exp b/gdb/testsuite/gdb.base/multi-forks.exp
> index e95cb4b..2a71e63 100644
> --- a/gdb/testsuite/gdb.base/multi-forks.exp
> +++ b/gdb/testsuite/gdb.base/multi-forks.exp
> @@ -62,6 +62,23 @@ proc continue_to_exit_bp_loc {} {
> set seen_break 0
> set seen_prompt 0
> set seen_timeout 0
> +
> + # If we are running with a native gdbserver, the output ($decimal done)
> + # will come via the spawn_id of gdbserver, not the spawn_id of gdb (the
> + # default). So we grab the spawn_id of gdbserver, if it exists, and
> + # add it to the gdb_expect statement below using "-i", allowing us to
> + # apply the expect statement to the output of both spawn_ids.
> + #
> + # If we are running with a truly remote gdbserver (gdb,noinferiorio),
> + # the output of the program will be inaccessible. In this case we do
> + # not check for the ($decimal done) output, but just look for the gdb
> + # prompt.
> + global server_spawn_id
> + set current_spawn_id [board_info host fileid]
> + if {![info exists server_spawn_id]} {
> + set server_spawn_id ""
> + }
> +
> while { ($seen_done < 16 || ! $seen_prompt) && ! $seen_timeout } {
> # We don't know what order the interesting things will arrive in.
> # Using a pattern of the form 'x|y|z' instead of -re x ... -re y
> @@ -70,7 +87,8 @@ proc continue_to_exit_bp_loc {} {
> # first in the script that occurs anywhere in the input, so that
> # we don't skip anything.
> gdb_expect {
> - -re "($decimal done)|(Breakpoint)|($gdb_prompt)" {
> + -i "$current_spawn_id $server_spawn_id" \
Note this is currently working by chance. We need this one in:
https://sourceware.org/ml/gdb-patches/2015-02/msg00659.html
in order to safely be able to do '-i "$server_spawn_id"'.
> + -re "($decimal done)|(Breakpoint)|($gdb_prompt)" {
> if {[info exists expect_out(1,string)]} {
> incr seen_done
> } elseif {[info exists expect_out(2,string)]} {
Thanks,
Pedro Alves
More information about the Gdb-patches
mailing list