[PATCH weston,v2] refactor configuration API of fbdev-backend

Benoit Gschwind gschwind at gnu-log.net
Wed Apr 27 20:04:25 UTC 2016


Hello Pekka,

I was too confident when I submitted my patch, while I fixing the memory
corruption that you pointed out I found another issue:

composite-fbdev.c:504

	section = weston_config_get_section(backend->compositor->config,
					    "output", "name",
					    output->base.name);

Thus I will investigate to fix this issue, and submit again the patch :)

Thanks again for your reviews

Best regards

Le 27/04/2016 09:59, Pekka Paalanen a écrit :
> On Tue, 26 Apr 2016 21:10:55 +0200
> Benoit Gschwind <gschwind at gnu-log.net> wrote:
> 
>> Implement a "well" defined API to configure the fbdev backend.
>> Following and according to discution about libweston API
>> ---
>> Signed-off-by: Benoit Gschwind <gschwind at gnu-log.net>
> 
> This S-o-b is on the wrong side of the '---'. git-am will drop if it is
> below the '---'.
> 
>>
>> v1:
>>  - add src/compositor-fbdev.h in the Makefile.am
>>  - fix typo
>>
>>  Makefile.am            |  3 ++-
>>  src/compositor-fbdev.c | 45 +++++++++++++++++++++++----------------------
>>  src/compositor-fbdev.h | 49 +++++++++++++++++++++++++++++++++++++++++++++++++
>>  src/main.c             | 33 +++++++++++++++++++++++++++++++--
>>  4 files changed, 105 insertions(+), 25 deletions(-)
>>  create mode 100644 src/compositor-fbdev.h
> 
> Hi Benoit,
> 
> would have been nice to rebase this on upstream master branch, but I
> handled the conflicts for now to review it.
> 
>>
>> diff --git a/Makefile.am b/Makefile.am
>> index c042c68..d319786 100644
>> --- a/Makefile.am
>> +++ b/Makefile.am
>> @@ -72,7 +72,8 @@ weston_SOURCES =					\
>>  	src/log.c					\
>>  	src/compositor.c				\
>>  	src/compositor.h				\
>> -	src/compositor-headless.h		\
>> +	src/compositor-headless.h			\
>> +	src/compositor-fbdev.h				\
>>  	src/input.c					\
>>  	src/data-device.c				\
>>  	src/screenshooter.c				\
> 
> Makefile.am is still missing the other two places to add
> compositor-fbdev.h to.
> 
>> diff --git a/src/compositor-fbdev.c b/src/compositor-fbdev.c
>> index 19c5e3b..d0597a6 100644
>> --- a/src/compositor-fbdev.c
>> +++ b/src/compositor-fbdev.c
>> @@ -44,6 +44,7 @@
>>  
>>  #include "shared/helpers.h"
>>  #include "compositor.h"
>> +#include "compositor-fbdev.h"
>>  #include "launcher-util.h"
>>  #include "pixman-renderer.h"
>>  #include "libinput-seat.h"
>> @@ -93,12 +94,6 @@ struct fbdev_output {
>>  	uint8_t depth;
>>  };
>>  
>> -struct fbdev_parameters {
>> -	int tty;
>> -	char *device;
>> -	int use_gl;
>> -};
>> -
>>  struct gl_renderer_interface *gl_renderer;
>>  
>>  static const char default_seat[] = "seat0";
>> @@ -744,7 +739,7 @@ fbdev_restore(struct weston_compositor *compositor)
>>  static struct fbdev_backend *
>>  fbdev_backend_create(struct weston_compositor *compositor, int *argc, char *argv[],
>>                       struct weston_config *config,
>> -                     struct fbdev_parameters *param)
>> +                     struct weston_fbdev_backend_config *param)
>>  {
>>  	struct fbdev_backend *backend;
>>  	const char *seat_id = default_seat;
>> @@ -827,29 +822,35 @@ out_compositor:
>>  	return NULL;
>>  }
>>  
>> +static void
>> +config_init_to_defaults(struct weston_fbdev_backend_config *config)
>> +{
>> +	/* TODO: Ideally, available frame buffers should be enumerated using
>> +	 * udev, rather than passing a device node in as a parameter. */
>> +	config->tty = 0; /* default to current tty */
>> +	config->device = "/dev/fb0"; /* default frame buffer */
>> +	config->use_gl = 0;
>> +}
>> +
>>  WL_EXPORT int
>>  backend_init(struct weston_compositor *compositor, int *argc, char *argv[],
>> -	     struct weston_config *config,
>> +	     struct weston_config *wc,
>>  	     struct weston_backend_config *config_base)
>>  {
>>  	struct fbdev_backend *b;
>> -	/* TODO: Ideally, available frame buffers should be enumerated using
>> -	 * udev, rather than passing a device node in as a parameter. */
>> -	struct fbdev_parameters param = {
>> -		.tty = 0, /* default to current tty */
>> -		.device = "/dev/fb0", /* default frame buffer */
>> -		.use_gl = 0,
>> -	};
>> +	struct weston_fbdev_backend_config config = {{ 0, }};
>>  
>> -	const struct weston_option fbdev_options[] = {
>> -		{ WESTON_OPTION_INTEGER, "tty", 0, &param.tty },
>> -		{ WESTON_OPTION_STRING, "device", 0, &param.device },
>> -		{ WESTON_OPTION_BOOLEAN, "use-gl", 0, &param.use_gl },
>> -	};
>> +	if (config_base == NULL ||
>> +	    config_base->struct_version != WESTON_FBDEV_BACKEND_CONFIG_VERSION ||
>> +	    config_base->struct_size > sizeof(struct weston_fbdev_backend_config)) {
>> +		weston_log("fbdev backend config structure is invalid\n");
>> +		return -1;
>> +	}
>>  
>> -	parse_options(fbdev_options, ARRAY_LENGTH(fbdev_options), argc, argv);
>> +	config_init_to_defaults(&config);
>> +	memcpy(&config, config_base, config_base->struct_size);
>>  
>> -	b = fbdev_backend_create(compositor, argc, argv, config, &param);
>> +	b = fbdev_backend_create(compositor, argc, argv, wc, &config);
> 
> I am quoting the exact same comment I gave the last time:
> 
>> There is a small issue with config.device. It is ok for
>> config_init_to_defaults() to use a statically allocated const string
>> for it, because the pointer gets overwritten with a pointer from
>> main.c. However, fbdev_output_create() is storing the pointer instead
>> of calling strdup(). When main.c after init frees the pointer,
>> fbdev_output::device will contain a stale pointer.
>>
>> Could you send a preparation patch that makes fbdev_output_create() use
>> strdup() for the device string, please?
>>
>> I think it might have leaked the device string allocated by
>> parse_options() before your patch.
> 
> What we talked about in IRC was the other comment, not this one.
> 
> Please, look again at the call chain after applying your patch:
> 
> main.c:
> - load_fbdev_backend():
> 	The config.device string is malloc'd, and passed in to
> 	load_backend_new().
> 
> compositor-fbdev.c:
> - backend_init(): Just copies the pointer, does not reallocate the string.
> - fbdev_backend_create(): Calls fbdev_output_create().
> - fbdev_output_create(): stores the string pointer in struct
>   fbdev_output::device *without* copying.
> 
> main.c:
> - load_fbdev_backend(): calls free() on the pointer.
> 
> This results in struct fbdev_output::device containing a pointer to
> freed memory.
> 
>>  	if (b == NULL)
>>  		return -1;
>>  	return 0;
> 
> 
> Thanks,
> pq
> 


More information about the wayland-devel mailing list