Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
From: Doug Evans <dje@google.com>
To: Tom Tromey <tromey@redhat.com>
Cc: Pedro Alves <palves@redhat.com>,
	Joel Brobecker <brobecker@adacore.com>,
		gdb-patches <gdb-patches@sourceware.org>
Subject: Re: [RFA] Ensure result of make_cleanup is never NULL.
Date: Thu, 19 Apr 2012 19:24:00 -0000	[thread overview]
Message-ID: <CADPb22SgZabeYBCRbGaj-ohFvfmmkBk-NkP9UNyS-VSRVPpi6Q@mail.gmail.com> (raw)
In-Reply-To: <87sjg1qf7o.fsf@fleche.redhat.com>

[-- Attachment #1: Type: text/plain, Size: 1572 bytes --]

On Wed, Apr 18, 2012 at 7:36 AM, Tom Tromey <tromey@redhat.com> wrote:
>>>>>> "Pedro" == Pedro Alves <palves@redhat.com> writes:
>
> Pedro> Well, aren't all places using the sentinel, using it throught the macro?
> Pedro> What does this make more robust?
>
> If you initialize the object it can end up in a read-only section, so
> writes to it will SEGV.
>
> In this particular case I don't see that this would help a lot, since
> the object is just a sentinel.  But, it wouldn't hurt and maybe it would
> catch some really unusual bug.

Hi.  fyi, here is what I committed.

I changed the name cleanup_sentinel -> sentinel_cleanup because I went
to print the value in gdb and spelled it the latter way and when that
failed I cursed because I was darn sure I had spelled it correctly.
That told me I liked the latter spelling better. :-)   Anyways ...

2012-04-19  Doug Evans  <dje@google.com>

        * cleanups.h (struct cleanup): Move to cleanups.c.
        (make_cleanup_dtor_ftype): New typedef.
        (make_cleanup_dtor): Use it.
        (ALL_CLEANUPS): Replace with ...
        (all_cleanups): ... this.  Declare.  All uses updated.
        * cleanups.c: #include "gdb_assert.h".
        (sentinel_cleanup): New static global.
        (SENTINEL_CLEANUP): Define.
        (cleanup_chain, final_cleanup_chain): Initialize to SENTINEL_CLEANUP.
        (make_my_cleanup2): Assert result is non-NULL.
        (all_cleanups): New function.
        (save_my_cleanups): Initialize new chain to SENTINEL_CLEANUP instead
        of NULL.

[-- Attachment #2: gdb-120419-cleanups-3b.patch.txt --]
[-- Type: text/plain, Size: 8427 bytes --]

2012-04-19  Doug Evans  <dje@google.com>

	* cleanups.h (struct cleanup): Move to cleanups.c.
	(make_cleanup_dtor_ftype): New typedef.
	(make_cleanup_dtor): Use it.
	(ALL_CLEANUPS): Replace with ...
	(all_cleanups): ... this.  Declare.  All uses updated.
	* cleanups.c: #include "gdb_assert.h".
	(sentinel_cleanup): New static global.
	(SENTINEL_CLEANUP): Define.
	(cleanup_chain, final_cleanup_chain): Initialize to SENTINEL_CLEANUP.
	(make_my_cleanup2): Assert result is non-NULL.
	(all_cleanups): New function.
	(save_my_cleanups): Initialize new chain to SENTINEL_CLEANUP instead
	of NULL.

Index: cleanups.c
===================================================================
RCS file: /cvs/src/src/gdb/cleanups.c,v
retrieving revision 1.2
diff -u -p -r1.2 cleanups.c
--- cleanups.c	17 Apr 2012 21:24:47 -0000	1.2
+++ cleanups.c	19 Apr 2012 17:52:44 -0000
@@ -18,15 +18,53 @@
    along with this program.  If not, see <http://www.gnu.org/licenses/>.  */
 
 #include "defs.h"
+#include "gdb_assert.h"
+
+/* The cleanup list records things that have to be undone
+   if an error happens (descriptors to be closed, memory to be freed, etc.)
+   Each link in the chain records a function to call and an
+   argument to give it.
+
+   Use make_cleanup to add an element to the cleanup chain.
+   Use do_cleanups to do all cleanup actions back to a given
+   point in the chain.  Use discard_cleanups to remove cleanups
+   from the chain back to a given point, not doing them.
+
+   If the argument is pointer to allocated memory, then you need
+   to additionally set the 'free_arg' member to a function that will
+   free that memory.  This function will be called both when the cleanup
+   is executed and when it's discarded.  */
+
+struct cleanup
+{
+  struct cleanup *next;
+  void (*function) (void *);
+  void (*free_arg) (void *);
+  void *arg;
+};
+
+/* Used to mark the end of a cleanup chain.
+   The value is chosen so that it:
+   - is non-NULL so that make_cleanup never returns NULL,
+   - causes a segv if dereferenced
+     [though this won't catch errors that a value of, say,
+     ((struct cleanup *) -1) will]
+   - displays as something useful when printed in gdb.
+   This is const for a bit of extra robustness.
+   It is initialized to coax gcc into putting it into .rodata.
+   All fields are initialized to survive -Wextra.  */
+static const struct cleanup sentinel_cleanup = { 0, 0, 0, 0 };
+
+/* Handy macro to use when referring to sentinel_cleanup.  */
+#define SENTINEL_CLEANUP ((struct cleanup *) &sentinel_cleanup)
 
 /* Chain of cleanup actions established with make_cleanup,
    to be executed if an error happens.  */
+static struct cleanup *cleanup_chain = SENTINEL_CLEANUP;
 
-/* Cleaned up after a failed command.  */
-static struct cleanup *cleanup_chain;
-
-/* Cleaned up when gdb exits.  */
-static struct cleanup *final_cleanup_chain;
+/* Chain of cleanup actions established with make_final_cleanup,
+   to be executed when gdb exits.  */
+static struct cleanup *final_cleanup_chain = SENTINEL_CLEANUP;
 
 /* Main worker routine to create a cleanup.
    PMY_CHAIN is a pointer to either cleanup_chain or final_cleanup_chain.
@@ -51,6 +89,7 @@ make_my_cleanup2 (struct cleanup **pmy_c
   new->arg = arg;
   *pmy_chain = new;
 
+  gdb_assert (old_chain != NULL);
   return old_chain;
 }
 
@@ -120,6 +159,15 @@ do_my_cleanups (struct cleanup **pmy_cha
     }
 }
 
+/* Return a value that can be passed to do_cleanups, do_final_cleanups to
+   indicate perform all cleanups.  */
+
+struct cleanup *
+all_cleanups (void)
+{
+  return SENTINEL_CLEANUP;
+}
+
 /* Discard cleanups and do the actions they describe
    until we get back to the point OLD_CHAIN in the cleanup_chain.  */
 
@@ -185,7 +233,7 @@ save_my_cleanups (struct cleanup **pmy_c
 {
   struct cleanup *old_chain = *pmy_chain;
 
-  *pmy_chain = 0;
+  *pmy_chain = SENTINEL_CLEANUP;
   return old_chain;
 }
 
Index: cleanups.h
===================================================================
RCS file: /cvs/src/src/gdb/cleanups.h,v
retrieving revision 1.2
diff -u -p -r1.2 cleanups.h
--- cleanups.h	17 Apr 2012 21:24:47 -0000	1.2
+++ cleanups.h	19 Apr 2012 17:52:44 -0000
@@ -19,28 +19,8 @@
 #ifndef CLEANUPS_H
 #define CLEANUPS_H
 
-/* The cleanup list records things that have to be undone
-   if an error happens (descriptors to be closed, memory to be freed, etc.)
-   Each link in the chain records a function to call and an
-   argument to give it.
-
-   Use make_cleanup to add an element to the cleanup chain.
-   Use do_cleanups to do all cleanup actions back to a given
-   point in the chain.  Use discard_cleanups to remove cleanups
-   from the chain back to a given point, not doing them.
-
-   If the argument is pointer to allocated memory, then you need
-   to additionally set the 'free_arg' member to a function that will
-   free that memory.  This function will be called both when the cleanup
-   is executed and when it's discarded.  */
-
-struct cleanup
-  {
-    struct cleanup *next;
-    void (*function) (void *);
-    void (*free_arg) (void *);
-    void *arg;
-  };
+/* Outside of cleanups.c, this is an opaque type.  */
+struct cleanup;
 
 /* NOTE: cagney/2000-03-04: This typedef is strictly for the
    make_cleanup function declarations below.  Do not use this typedef
@@ -49,21 +29,25 @@ struct cleanup
    Calling a f(char*) function with f(void*) is non-portable.  */
 typedef void (make_cleanup_ftype) (void *);
 
+/* Function type for the dtor in make_cleanup_dtor.  */
+typedef void (make_cleanup_dtor_ftype) (void *);
+
 /* WARNING: The result of the "make cleanup" routines is not the intuitive
    choice of being a handle on the just-created cleanup.  Instead it is an
    opaque handle of the cleanup mechanism and represents all cleanups created
-   from that point onwards.  */
+   from that point onwards.
+   The result is guaranteed to be non-NULL though.  */
 
 extern struct cleanup *make_cleanup (make_cleanup_ftype *, void *);
 
 extern struct cleanup *make_cleanup_dtor (make_cleanup_ftype *, void *,
-					  void (*dtor) (void *));
+					  make_cleanup_dtor_ftype *);
 
 extern struct cleanup *make_final_cleanup (make_cleanup_ftype *, void *);
 
 /* A special value to pass to do_cleanups and do_final_cleanups
    to tell them to do all cleanups.  */
-#define	ALL_CLEANUPS	((struct cleanup *)0)
+extern struct cleanup *all_cleanups (void);
 
 extern void do_cleanups (struct cleanup *);
 extern void do_final_cleanups (struct cleanup *);
Index: exceptions.c
===================================================================
RCS file: /cvs/src/src/gdb/exceptions.c,v
retrieving revision 1.50
diff -u -p -r1.50 exceptions.c
--- exceptions.c	4 Jan 2012 08:17:01 -0000	1.50
+++ exceptions.c	19 Apr 2012 17:52:44 -0000
@@ -224,7 +224,7 @@ throw_exception (struct gdb_exception ex
   quit_flag = 0;
   immediate_quit = 0;
 
-  do_cleanups (ALL_CLEANUPS);
+  do_cleanups (all_cleanups ());
 
   /* Jump to the containing catch_errors() call, communicating REASON
      to that call via setjmp's return value.  Note that REASON can't
Index: main.c
===================================================================
RCS file: /cvs/src/src/gdb/main.c,v
retrieving revision 1.108
diff -u -p -r1.108 main.c
--- main.c	17 Apr 2012 15:56:21 -0000	1.108
+++ main.c	19 Apr 2012 17:52:44 -0000
@@ -231,7 +231,7 @@ captured_command_loop (void *data)
      are not that well behaved.  do_cleanups should either be replaced
      with a do_cleanups call (to cover the problem) or an assertion
      check to detect bad FUNCs code.  */
-  do_cleanups (ALL_CLEANUPS);
+  do_cleanups (all_cleanups ());
   /* If the command_loop returned, normally (rather than threw an
      error) we try to quit.  If the quit is aborted, catch_errors()
      which called this catch the signal and restart the command
Index: top.c
===================================================================
RCS file: /cvs/src/src/gdb/top.c,v
retrieving revision 1.214
diff -u -p -r1.214 top.c
--- top.c	1 Mar 2012 19:30:20 -0000	1.214
+++ top.c	19 Apr 2012 17:52:44 -0000
@@ -1297,8 +1297,9 @@ quit_target (void *arg)
   if (write_history_p && history_filename)
     write_history (history_filename);
 
-  do_final_cleanups (ALL_CLEANUPS);    /* Do any final cleanups before
-					  exiting.  */
+  /* Do any final cleanups before exiting.  */
+  do_final_cleanups (all_cleanups ());
+
   return 0;
 }
 

  reply	other threads:[~2012-04-19 19:18 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-04-16  5:01 Doug Evans
2012-04-16 10:58 ` Pedro Alves
2012-04-16 14:44 ` Joel Brobecker
2012-04-16 14:58   ` Jan Kratochvil
2012-04-16 15:03   ` Pedro Alves
2012-04-16 20:16     ` Tom Tromey
     [not found]     ` <CADPb22SfUk5s9JSSBvUTWVyhoiEqO4Gi+VNO-9MwH6rqW8qQ3g@mail.gmail.com>
2012-04-17 23:12       ` Doug Evans
2012-04-18  1:06         ` Doug Evans
2012-04-18  1:58           ` Joel Brobecker
2012-04-18  9:21             ` Pedro Alves
2012-04-18  9:29           ` Pedro Alves
2012-04-18 14:14             ` Doug Evans
2012-04-18 14:25               ` Pedro Alves
2012-04-18 14:32                 ` Doug Evans
2012-04-18 14:36                   ` Pedro Alves
2012-04-18 14:38                 ` Tom Tromey
2012-04-19 19:24                   ` Doug Evans [this message]
2012-04-18  9:26         ` Pedro Alves

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CADPb22SgZabeYBCRbGaj-ohFvfmmkBk-NkP9UNyS-VSRVPpi6Q@mail.gmail.com \
    --to=dje@google.com \
    --cc=brobecker@adacore.com \
    --cc=gdb-patches@sourceware.org \
    --cc=palves@redhat.com \
    --cc=tromey@redhat.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox