[PATCH] python: Fix memleak in do_start_initialization

[PATCH] python: Fix memleak in do_start_initialization

[Simon Marchi]

While playing with valgrind, I noticed that the progname variable in
do_start_initialization is not being freed (concat returns a malloc'ed
string).  Use a unique_xmalloc_ptr to manage it.

gdb/ChangeLog:

	* python/python.c (do_start_initialization): Change progname
	type to gdb::unique_xmalloc_ptr.
---
 gdb/python/python.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/gdb/python/python.c b/gdb/python/python.c
index 5f152611e8..f17ac36569 100644
--- a/gdb/python/python.c
+++ b/gdb/python/python.c
@@ -1658,7 +1658,7 @@ finalize_python (void *ignore)
 static bool
 do_start_initialization ()
 {
-  char *progname;
+  gdb::unique_xmalloc_ptr<char> progname;
 #ifdef IS_PY3K
   int i;
   size_t progsize, count;
@@ -1672,19 +1672,19 @@ do_start_initialization ()
      /foo/bin/python
      /foo/lib/pythonX.Y/...
      This must be done before calling Py_Initialize.  */
-  progname = concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
-		     SLASH_STRING, "python", (char *) NULL);
+  progname.reset (concat (ldirname (python_libdir).c_str (), SLASH_STRING,
+			  "bin", SLASH_STRING, "python", (char *) NULL));
 #ifdef IS_PY3K
   std::string oldloc = setlocale (LC_ALL, NULL);
   setlocale (LC_ALL, "");
-  progsize = strlen (progname);
+  progsize = strlen (progname.get ());
   progname_copy = (wchar_t *) PyMem_Malloc ((progsize + 1) * sizeof (wchar_t));
   if (!progname_copy)
     {
       fprintf (stderr, "out of memory\n");
       return false;
     }
-  count = mbstowcs (progname_copy, progname, progsize + 1);
+  count = mbstowcs (progname_copy, progname.get (), progsize + 1);
   if (count == (size_t) -1)
     {
       fprintf (stderr, "Could not convert python path to string\n");
-- 
2.15.0

Re: [PATCH] python: Fix memleak in do_start_initialization

[Tom Tromey]

>>>>> "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:

Simon> While playing with valgrind, I noticed that the progname variable in
Simon> do_start_initialization is not being freed (concat returns a malloc'ed
Simon> string).  Use a unique_xmalloc_ptr to manage it.

I don't think this is quite correct.
On the Python 2 branch, the code does this:

  Py_SetProgramName (progname);

The docs for Py_SetProgramName say:

     The argument should point to a
     zero-terminated character string in static storage whose contents
     will not change for the duration of the program's execution.

I think a comment referring to the Python docs would be better.
And, freeing progname on the Python 3 branch is ok.
Maybe the Python 2 branch could pass "progname.release ()".

Tom

Re: [PATCH] python: Fix memleak in do_start_initialization

[Simon Marchi]

On 2017-11-26 02:00 PM, Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:
> 
> Simon> While playing with valgrind, I noticed that the progname variable in
> Simon> do_start_initialization is not being freed (concat returns a malloc'ed
> Simon> string).  Use a unique_xmalloc_ptr to manage it.
> 
> I don't think this is quite correct.
> On the Python 2 branch, the code does this:
> 
>   Py_SetProgramName (progname);
> 
> The docs for Py_SetProgramName say:
> 
>      The argument should point to a
>      zero-terminated character string in static storage whose contents
>      will not change for the duration of the program's execution.

Ahh, you are completely right, I always forget about Python 2.  And I relied
too much on my IDE to find usages of progname, it didn't find the usage in the
disabled #if branch.  At least it would have failed at compile time with
Python 2, not runtime.

> I think a comment referring to the Python docs would be better.
> And, freeing progname on the Python 3 branch is ok.
> Maybe the Python 2 branch could pass "progname.release ()".

There's already a comment in that regard:

  /* Note that Py_SetProgramName expects the string it is passed to
     remain alive for the duration of the program's execution, so
     it is not freed after this call.  */

When I read it, I thought it referred only to progname_copy, but it
also applies to progname for Python 2.  I integrated your suggestion,
wdyt?


From 468fe44d21dfed10f547d790d233547ab0da2cf5 Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@polymtl.ca>
Date: Sat, 25 Nov 2017 22:51:02 -0500
Subject: [PATCH] python: Fix memleak in do_start_initialization

While playing with valgrind, I noticed that with Python 3, the progname
variable in do_start_initialization is not being freed (concat returns a
malloc'ed string).  This patch uses unique_xmalloc_ptr to manage it.
With Python 2, we pass progname it directly to Py_SetProgramName, so it
should not be freed.  We therefore release it before passing it.

gdb/ChangeLog:

	* python/python.c (do_start_initialization): Change progname
	type to gdb::unique_xmalloc_ptr.  Release the pointer when using
	Python 2.
---
 gdb/python/python.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/gdb/python/python.c b/gdb/python/python.c
index 5f152611e8..6b05f97b39 100644
--- a/gdb/python/python.c
+++ b/gdb/python/python.c
@@ -1658,7 +1658,7 @@ finalize_python (void *ignore)
 static bool
 do_start_initialization ()
 {
-  char *progname;
+  gdb::unique_xmalloc_ptr<char> progname;
 #ifdef IS_PY3K
   int i;
   size_t progsize, count;
@@ -1672,19 +1672,19 @@ do_start_initialization ()
      /foo/bin/python
      /foo/lib/pythonX.Y/...
      This must be done before calling Py_Initialize.  */
-  progname = concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
-		     SLASH_STRING, "python", (char *) NULL);
+  progname.reset (concat (ldirname (python_libdir).c_str (), SLASH_STRING,
+			  "bin", SLASH_STRING, "python", (char *) NULL));
 #ifdef IS_PY3K
   std::string oldloc = setlocale (LC_ALL, NULL);
   setlocale (LC_ALL, "");
-  progsize = strlen (progname);
+  progsize = strlen (progname.get ());
   progname_copy = (wchar_t *) PyMem_Malloc ((progsize + 1) * sizeof (wchar_t));
   if (!progname_copy)
     {
       fprintf (stderr, "out of memory\n");
       return false;
     }
-  count = mbstowcs (progname_copy, progname, progsize + 1);
+  count = mbstowcs (progname_copy, progname.get (), progsize + 1);
   if (count == (size_t) -1)
     {
       fprintf (stderr, "Could not convert python path to string\n");
@@ -1697,7 +1697,7 @@ do_start_initialization ()
      it is not freed after this call.  */
   Py_SetProgramName (progname_copy);
 #else
-  Py_SetProgramName (progname);
+  Py_SetProgramName (progname.release ());
 #endif
 #endif

-- 
2.15.0

Re: [PATCH] python: Fix memleak in do_start_initialization

[Tom Tromey]

>>>>> "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:

Simon> There's already a comment in that regard:

Hah, I totally missed this.

Simon> When I read it, I thought it referred only to progname_copy, but it
Simon> also applies to progname for Python 2.  I integrated your suggestion,
Simon> wdyt?

Simon> -  char *progname;
Simon> +  gdb::unique_xmalloc_ptr<char> progname;

I think the patch looks good, but would be even clearer if this
declaration were moved into the #if, so that..

Simon> -  progname = concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
Simon> -		     SLASH_STRING, "python", (char *) NULL);
Simon> +  progname.reset (concat (ldirname (python_libdir).c_str (), SLASH_STRING,
Simon> +			  "bin", SLASH_STRING, "python", (char *) NULL));

...this reset could be replaced with plain initialization.

thanks,
Tom

Re: [PATCH] python: Fix memleak in do_start_initialization

[Simon Marchi]

On 2017-11-26 04:54 PM, Tom Tromey wrote:
>>>>>> "Simon" == Simon Marchi <simon.marchi@polymtl.ca> writes:
> 
> Simon> There's already a comment in that regard:
> 
> Hah, I totally missed this.
> 
> Simon> When I read it, I thought it referred only to progname_copy, but it
> Simon> also applies to progname for Python 2.  I integrated your suggestion,
> Simon> wdyt?
> 
> Simon> -  char *progname;
> Simon> +  gdb::unique_xmalloc_ptr<char> progname;
> 
> I think the patch looks good, but would be even clearer if this
> declaration were moved into the #if, so that..
> 
> Simon> -  progname = concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
> Simon> -		     SLASH_STRING, "python", (char *) NULL);
> Simon> +  progname.reset (concat (ldirname (python_libdir).c_str (), SLASH_STRING,
> Simon> +			  "bin", SLASH_STRING, "python", (char *) NULL));
> 
> ...this reset could be replaced with plain initialization.
> 
> thanks,
> Tom
> 

Good idea, here's what I pushed.  Thanks for the comments!


From e8e7d10c39955e7ff99ff42f6f16d6befe2fa12e Mon Sep 17 00:00:00 2001
From: Simon Marchi <simon.marchi@polymtl.ca>
Date: Sun, 26 Nov 2017 19:32:47 -0500
Subject: [PATCH] python: Fix memleak in do_start_initialization

While playing with valgrind, I noticed that with Python 3, the progname
variable in do_start_initialization is not being freed (concat returns a
malloc'ed string).  This patch uses unique_xmalloc_ptr to manage it.
With Python 2, we pass progname it directly to Py_SetProgramName, so it
should not be freed.  We therefore release it before passing it.

gdb/ChangeLog:

	* python/python.c (do_start_initialization): Change progname
	type to gdb::unique_xmalloc_ptr.  Release the pointer when using
	Python 2.
---
 gdb/ChangeLog       |  6 ++++++
 gdb/python/python.c | 12 ++++++------
 2 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index a901409228..b3032ed7d8 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,9 @@
+2017-11-26  Simon Marchi  <simon.marchi@polymtl.ca>
+
+	* python/python.c (do_start_initialization): Change progname
+	type to gdb::unique_xmalloc_ptr.  Release the pointer when using
+	Python 2.
+
 2017-11-26  Tom Tromey  <tom@tromey.com>

 	* common/format.h: Add include guards.
diff --git a/gdb/python/python.c b/gdb/python/python.c
index 5f152611e8..9ed9b6b1ca 100644
--- a/gdb/python/python.c
+++ b/gdb/python/python.c
@@ -1658,7 +1658,6 @@ finalize_python (void *ignore)
 static bool
 do_start_initialization ()
 {
-  char *progname;
 #ifdef IS_PY3K
   int i;
   size_t progsize, count;
@@ -1672,19 +1671,20 @@ do_start_initialization ()
      /foo/bin/python
      /foo/lib/pythonX.Y/...
      This must be done before calling Py_Initialize.  */
-  progname = concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
-		     SLASH_STRING, "python", (char *) NULL);
+  gdb::unique_xmalloc_ptr<char> progname
+    (concat (ldirname (python_libdir).c_str (), SLASH_STRING, "bin",
+	      SLASH_STRING, "python", (char *) NULL));
 #ifdef IS_PY3K
   std::string oldloc = setlocale (LC_ALL, NULL);
   setlocale (LC_ALL, "");
-  progsize = strlen (progname);
+  progsize = strlen (progname.get ());
   progname_copy = (wchar_t *) PyMem_Malloc ((progsize + 1) * sizeof (wchar_t));
   if (!progname_copy)
     {
       fprintf (stderr, "out of memory\n");
       return false;
     }
-  count = mbstowcs (progname_copy, progname, progsize + 1);
+  count = mbstowcs (progname_copy, progname.get (), progsize + 1);
   if (count == (size_t) -1)
     {
       fprintf (stderr, "Could not convert python path to string\n");
@@ -1697,7 +1697,7 @@ do_start_initialization ()
      it is not freed after this call.  */
   Py_SetProgramName (progname_copy);
 #else
-  Py_SetProgramName (progname);
+  Py_SetProgramName (progname.release ());
 #endif
 #endif

-- 
2.15.0