ASoC: Intel: Skylake: prevent memory leak in snd_skl_parse_uuids
Subsystems (3)
PATCH v2 1/1 ASoC: Intel: Skylake: prevent memory leak in snd_skl_parse_uuids
In snd_skl_parse_uuids if allocation for module->instance_id fails, the allocated memory for module shoulde be released. I changes the allocation for module to use devm_kzalloc to be resource_managed allocation and avoid the release in error path.
---
Changes in v2:
- Changed the allocation for module from kzalloc to devm_kzalloc
--- sound/soc/intel/skylake/skl-sst-utils.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
sound/soc/intel/skylake/skl-sst-utils.c +1 -1
@@ -284,7 +284,7 @@ int snd_skl_parse_uuids(struct sst_dsp *ctx, const struct firmware *fw,*/ for (i =0 ; i < num_entry; i++, mod_entry++) { - module =kzalloc (sizeof (*module),GFP_KERNEL ); + module =devm_kzalloc (ctx->dev,sizeof (*module),GFP_KERNEL );if (!module) { ret = -ENOMEM ; goto free_uuid_list;
Signature · 1 line
2.17.1
Quoted 4 lines · Navid Emamdoost
In
snd_skl_parse_uuidsif allocation formodule->instance_idfails, the allocated memory for module shoulde be released. I changes the allocation for module to usedevm_kzallocto beresource_managedallocation and avoid the release in error path.
if you use devm_, don't you need to fix the error path as well then, I see a kfree(uuid) in skl_freeup_uuid_list().
I am not very familiar with this code but the error seems to be that the list_add_tail() is called after the module->instance_id is allocated, so there is a risk that the module allocated earlier is not freed (since it's not yet added to the list). Freeing the module as done in patch 1 works, using devm_ without fixing the error path does not seem correct to me.
Quoted 23 lines
Signed-off-by Navid Emamdoost <[email protected]>---
Changes in v2:
- Changed the allocation for module from kzalloc todevm_kzalloc--- sound/soc/intel/skylake/skl-sst-utils.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)sound/soc/intel/skylake/skl-sst-utils.c +1
-1@@ -284,7 +284,7 @@ int snd_skl_parse_uuids(struct sst_dsp *ctx, const struct firmware *fw,*/ for (i =0 ; i < num_entry; i++, mod_entry++) { - module =kzalloc (sizeof (*module),GFP_KERNEL ); + module =devm_kzalloc (ctx->dev,sizeof (*module),GFP_KERNEL );if (!module) { ret = -ENOMEM ; goto free_uuid_list;
Quoted 6 lines · Pierre-Louis Bossart
Quoted 4 lines · Navid Emamdoost
In
snd_skl_parse_uuidsif allocation formodule->instance_idfails, the allocated memory for module shoulde be released. I changes the allocation for module to usedevm_kzallocto beresource_managedallocation and avoid the release in error path.
Quoted 9 lines
if you use
devm_, don't you need to fix the error path as well then, I see a kfree(uuid) inskl_freeup_uuid_list().I am not very familiar with this code but the error seems to be that the
list_add_tail()is called after themodule->instance_idis allocated, so there is a risk that the module allocated earlier is not freed (since it's not yet added to the list). Freeing the module as done in patch 1 works, usingdevm_without fixing the error path does not seem correct to me.
Thanks for the feedback, then it's your call if you can accept patch 1 as fix.
Navid.
Quoted 23 lines
Quoted 22 lines
Signed-off-by Navid Emamdoost <[email protected]>---
Changes in v2:
- Changed the allocation for module from kzalloc todevm_kzalloc--- sound/soc/intel/skylake/skl-sst-utils.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)sound/soc/intel/skylake/skl-sst-utils.c +1
-1@@ -284,7 +284,7 @@ int snd_skl_parse_uuids(struct sst_dsp *ctx, const struct firmware *fw,*/ for (i =0 ; i < num_entry; i++, mod_entry++) { - module =kzalloc (sizeof (*module),GFP_KERNEL ); + module =devm_kzalloc (ctx->dev,sizeof (*module),GFP_KERNEL );if (!module) { ret = -ENOMEM ; goto free_uuid_list;
Quoted 10 lines · Navid Emamdoost
Quoted 9 lines · Pierre-Louis Bossart
Quoted 4 lines · Navid Emamdoost
In
snd_skl_parse_uuidsif allocation formodule->instance_idfails, the allocated memory for module shoulde be released. I changes the allocation for module to usedevm_kzallocto beresource_managedallocation and avoid the release in error path.if you use
devm_, don't you need to fix the error path as well then, I see a kfree(uuid) inskl_freeup_uuid_list().
Quoted 8 lines
Quoted 6 lines
I am not very familiar with this code but the error seems to be that the
list_add_tail()is called after themodule->instance_idis allocated, so there is a risk that the module allocated earlier is not freed (since it's not yet added to the list). Freeing the module as done in patch 1 works, usingdevm_without fixing the error path does not seem correct to me.Thanks for the feedback, then it's your call if you can accept patch 1 as fix.
Cezary, it's really your call.
Quoted 24 lines
Navid.
Quoted 23 lines
Quoted 22 lines
Signed-off-by Navid Emamdoost <[email protected]>---
Changes in v2:
- Changed the allocation for module from kzalloc todevm_kzalloc--- sound/soc/intel/skylake/skl-sst-utils.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)sound/soc/intel/skylake/skl-sst-utils.c +1
-1@@ -284,7 +284,7 @@ int snd_skl_parse_uuids(struct sst_dsp *ctx, const struct firmware *fw,*/ for (i =0 ; i < num_entry; i++, mod_entry++) { - module =kzalloc (sizeof (*module),GFP_KERNEL ); + module =devm_kzalloc (ctx->dev,sizeof (*module),GFP_KERNEL );if (!module) { ret = -ENOMEM ; goto free_uuid_list;
Cezary Rojewski
maintainerQuoted 11 lines · Pierre-Louis Bossart
Quoted 10 lines · Navid Emamdoost
Quoted 9 lines · Pierre-Louis Bossart
Quoted 4 lines · Navid Emamdoost
In
snd_skl_parse_uuidsif allocation formodule->instance_idfails, the allocated memory for module shoulde be released. I changes the allocation for module to usedevm_kzallocto beresource_managedallocation and avoid the release in error path.if you use
devm_, don't you need to fix the error path as well then, I see a kfree(uuid) inskl_freeup_uuid_list().
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
I am not very familiar with this code but the error seems to be that the
list_add_tail()is called after themodule->instance_idis allocated, so there is a risk that the module allocated earlier is not freed (since it's not yet added to the list). Freeing the module as done in patch 1 works, usingdevm_without fixing the error path does not seem correct to me.
Good catch, Pierre.
Quoted 5 lines
Quoted 2 lines
Thanks for the feedback, then it's your call if you can accept patch 1 as fix.
Cezary, it's really your call.
Actually, not the best person to ask about "objective decisions" here as my vision is clouded by changes done internally. This code no longer exists in our internal repo. It's better for host to send MODULE_INFO request rather than understanding firmware binary structure and parse it directly.
I'm fine with solution #1 as I guess asking to wait for refactor is not an option. Code deployment is delayed due to range of administrative decisions, some of which should be uncovered on alsa-devel soon enough.
Czarek
Quoted 3 lines · Cezary Rojewski
I'm fine with solution #1 as I guess asking to wait for refactor is not an option. Code deployment is delayed due to range of administrative decisions, some of which should be uncovered on alsa-devel soon enough.
The problem with solution #1 is freeing orphaned pointer. It will work, but it's simple is not okay from object life time prospective.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 4 lines
The problem with solution #1 is freeing orphaned pointer. It will work, but it's simple is not okay from object life time prospective.
?? I don't get your point at all Andy.
Two allocations happens in a loop and if the second fails, you free the first and then jump to free everything allocated in the previous iterations. what am I missing?
Quoted 7 lines · Pierre-Louis Bossart
Quoted 2 lines
The problem with solution #1 is freeing orphaned pointer. It will work, but it's simple is not okay from object life time prospective.
?? I don't get your point at all Andy.
Two allocations happens in a loop and if the second fails, you free the first and then jump to free everything allocated in the previous iterations. what am I missing?
Two things:
- one allocation is done with kzalloc(), while the other one withdevm_kcalloc()
- due to above the ordering of resources is reversed
Signature · 2 lines
With Best Regards, Andy Shevchenko
Quoted 10 lines · Andy Shevchenko
Quoted 7 lines · Pierre-Louis Bossart
Quoted 2 lines
The problem with solution #1 is freeing orphaned pointer. It will work, but it's simple is not okay from object life time prospective.
?? I don't get your point at all Andy.
Two allocations happens in a loop and if the second fails, you free the first and then jump to free everything allocated in the previous iterations. what am I missing?
Quoted 4 lines
Two things:
- one allocation is done withkzalloc(), while the other one withdevm_kcalloc()
- due to above the ordering of resources is reversed
Ah yes, I see your point now, sorry for being thick.
Indeed it'd make sense to use devm_ for both allocations, but then the kfree needs to be removed in the error handling.