| Message ID | 1428801145-5203-1-git-send-email-o-takashi@sakamocchi.jp (mailing list archive) |
|---|---|
| State | Accepted |
| Commit | e1c78df1da112f2644058af2425dd5ca3eb1a96a |
| Headers | show |
At Sun, 12 Apr 2015 10:12:25 +0900, Takashi Sakamoto wrote: > > An element instance can have several elements with the same feature. > Some userspace applications can add such an element instance by add > operation with the number of elements. Then, the element instance > gets a memory object to keep states of these elements. > > But the element instance has just one memory object for the elements. > This causes the same result to each read/write operations to the > different elements. > > This commit fixes this bug by allocating enough memory objects to the > element instance for each of elements. > > Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> > --- > sound/core/control.c | 16 +++++++++++----- > 1 file changed, 11 insertions(+), 5 deletions(-) > > diff --git a/sound/core/control.c b/sound/core/control.c > index ccb1ca2..5b2c352 100644 > --- a/sound/core/control.c > +++ b/sound/core/control.c > @@ -1029,7 +1029,7 @@ static int snd_ctl_elem_unlock(struct snd_ctl_file *file, > struct user_element { > struct snd_ctl_elem_info info; > struct snd_card *card; > - void *elem_data; /* element data */ > + char *elem_data; /* element data */ > unsigned long elem_data_size; /* size of element data in bytes */ > void *tlv_data; /* TLV data */ > unsigned long tlv_data_size; /* TLV data size */ > @@ -1078,9 +1078,12 @@ static int snd_ctl_elem_user_get(struct snd_kcontrol *kcontrol, > struct snd_ctl_elem_value *ucontrol) > { > struct user_element *ue = kcontrol->private_data; > + unsigned int size = ue->elem_data_size; > + char *src = ue->elem_data + > + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; You still need to check the validity of snd_ctl_get_ioff() value. The user-space can give any value here, and accessing without the check will expose the kernel memory to user-space at get or even corrupt at put. thanks, Takashi > > mutex_lock(&ue->card->user_ctl_lock); > - memcpy(&ucontrol->value, ue->elem_data, ue->elem_data_size); > + memcpy(&ucontrol->value, src, size); > mutex_unlock(&ue->card->user_ctl_lock); > return 0; > } > @@ -1090,11 +1093,14 @@ static int snd_ctl_elem_user_put(struct snd_kcontrol *kcontrol, > { > int change; > struct user_element *ue = kcontrol->private_data; > + unsigned int size = ue->elem_data_size; > + char *dst = ue->elem_data + > + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; > > mutex_lock(&ue->card->user_ctl_lock); > - change = memcmp(&ucontrol->value, ue->elem_data, ue->elem_data_size) != 0; > + change = memcmp(&ucontrol->value, dst, size) != 0; > if (change) > - memcpy(ue->elem_data, &ucontrol->value, ue->elem_data_size); > + memcpy(dst, &ucontrol->value, size); > mutex_unlock(&ue->card->user_ctl_lock); > return change; > } > @@ -1278,7 +1284,7 @@ static int snd_ctl_elem_add(struct snd_ctl_file *file, > if (err < 0) > return err; > memcpy(&kctl->id, &info->id, sizeof(kctl->id)); > - kctl->private_data = kzalloc(sizeof(struct user_element) + private_size, > + kctl->private_data = kzalloc(sizeof(struct user_element) + private_size * count, > GFP_KERNEL); > if (kctl->private_data == NULL) { > kfree(kctl); > -- > 2.1.0 >
Iwai-san, On Apr 12 2015 14:51, Takashi Iwai wrote: > At Sun, 12 Apr 2015 10:12:25 +0900, > Takashi Sakamoto wrote: >> >> An element instance can have several elements with the same feature. >> Some userspace applications can add such an element instance by add >> operation with the number of elements. Then, the element instance >> gets a memory object to keep states of these elements. >> >> But the element instance has just one memory object for the elements. >> This causes the same result to each read/write operations to the >> different elements. >> >> This commit fixes this bug by allocating enough memory objects to the >> element instance for each of elements. >> >> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> >> --- >> sound/core/control.c | 16 +++++++++++----- >> 1 file changed, 11 insertions(+), 5 deletions(-) >> >> diff --git a/sound/core/control.c b/sound/core/control.c >> index ccb1ca2..5b2c352 100644 >> --- a/sound/core/control.c >> +++ b/sound/core/control.c >> @@ -1029,7 +1029,7 @@ static int snd_ctl_elem_unlock(struct snd_ctl_file *file, >> struct user_element { >> struct snd_ctl_elem_info info; >> struct snd_card *card; >> - void *elem_data; /* element data */ >> + char *elem_data; /* element data */ >> unsigned long elem_data_size; /* size of element data in bytes */ >> void *tlv_data; /* TLV data */ >> unsigned long tlv_data_size; /* TLV data size */ >> @@ -1078,9 +1078,12 @@ static int snd_ctl_elem_user_get(struct snd_kcontrol *kcontrol, >> struct snd_ctl_elem_value *ucontrol) >> { >> struct user_element *ue = kcontrol->private_data; >> + unsigned int size = ue->elem_data_size; >> + char *src = ue->elem_data + >> + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; > > You still need to check the validity of snd_ctl_get_ioff() value. > The user-space can give any value here, and accessing without the > check will expose the kernel memory to user-space at get or even > corrupt at put. I don't agree this indication. The 'ucontrol' has ID information re-calculated by the caller (snd_ctl_elem_read() / snd_ctl_elem_write()), thus it's not directly passed from userspace. If the information includes something wrong, the caller returns error and don't execute callee. The snd_ctl_elem_add() has already allocated enough memory, so there're no wrong reference as long as the re-calculated ID includes larger numid than (kctl->numid + kctl->count), or larger index than kctl->count. Are there any cases that the re-calculated ID includes such wrong information after passing these functions below? * snd_ctl_find_id() -> find kctl with given ID info * snd_ctl_get_ioff() -> calculate offset with kctl and given ID * snd_ctl_build_ioff() -> re-calculate ID with kctl and the offset If it still includes wrong information, it's the caller's fault and should be fixed. Thanks Takashi Sakamoto > thanks, > > Takashi > >> >> mutex_lock(&ue->card->user_ctl_lock); >> - memcpy(&ucontrol->value, ue->elem_data, ue->elem_data_size); >> + memcpy(&ucontrol->value, src, size); >> mutex_unlock(&ue->card->user_ctl_lock); >> return 0; >> } >> @@ -1090,11 +1093,14 @@ static int snd_ctl_elem_user_put(struct snd_kcontrol *kcontrol, >> { >> int change; >> struct user_element *ue = kcontrol->private_data; >> + unsigned int size = ue->elem_data_size; >> + char *dst = ue->elem_data + >> + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; >> >> mutex_lock(&ue->card->user_ctl_lock); >> - change = memcmp(&ucontrol->value, ue->elem_data, ue->elem_data_size) != 0; >> + change = memcmp(&ucontrol->value, dst, size) != 0; >> if (change) >> - memcpy(ue->elem_data, &ucontrol->value, ue->elem_data_size); >> + memcpy(dst, &ucontrol->value, size); >> mutex_unlock(&ue->card->user_ctl_lock); >> return change; >> } >> @@ -1278,7 +1284,7 @@ static int snd_ctl_elem_add(struct snd_ctl_file *file, >> if (err < 0) >> return err; >> memcpy(&kctl->id, &info->id, sizeof(kctl->id)); >> - kctl->private_data = kzalloc(sizeof(struct user_element) + private_size, >> + kctl->private_data = kzalloc(sizeof(struct user_element) + private_size * count, >> GFP_KERNEL); >> if (kctl->private_data == NULL) { >> kfree(kctl); >> -- >> 2.1.0
At Sun, 12 Apr 2015 18:51:16 +0900, Takashi Sakamoto wrote: > > Iwai-san, > > On Apr 12 2015 14:51, Takashi Iwai wrote: > > At Sun, 12 Apr 2015 10:12:25 +0900, > > Takashi Sakamoto wrote: > >> > >> An element instance can have several elements with the same feature. > >> Some userspace applications can add such an element instance by add > >> operation with the number of elements. Then, the element instance > >> gets a memory object to keep states of these elements. > >> > >> But the element instance has just one memory object for the elements. > >> This causes the same result to each read/write operations to the > >> different elements. > >> > >> This commit fixes this bug by allocating enough memory objects to the > >> element instance for each of elements. > >> > >> Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> > >> --- > >> sound/core/control.c | 16 +++++++++++----- > >> 1 file changed, 11 insertions(+), 5 deletions(-) > >> > >> diff --git a/sound/core/control.c b/sound/core/control.c > >> index ccb1ca2..5b2c352 100644 > >> --- a/sound/core/control.c > >> +++ b/sound/core/control.c > >> @@ -1029,7 +1029,7 @@ static int snd_ctl_elem_unlock(struct snd_ctl_file *file, > >> struct user_element { > >> struct snd_ctl_elem_info info; > >> struct snd_card *card; > >> - void *elem_data; /* element data */ > >> + char *elem_data; /* element data */ > >> unsigned long elem_data_size; /* size of element data in bytes */ > >> void *tlv_data; /* TLV data */ > >> unsigned long tlv_data_size; /* TLV data size */ > >> @@ -1078,9 +1078,12 @@ static int snd_ctl_elem_user_get(struct snd_kcontrol *kcontrol, > >> struct snd_ctl_elem_value *ucontrol) > >> { > >> struct user_element *ue = kcontrol->private_data; > >> + unsigned int size = ue->elem_data_size; > >> + char *src = ue->elem_data + > >> + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; > > > > You still need to check the validity of snd_ctl_get_ioff() value. > > The user-space can give any value here, and accessing without the > > check will expose the kernel memory to user-space at get or even > > corrupt at put. > > I don't agree this indication. > > The 'ucontrol' has ID information re-calculated by the caller > (snd_ctl_elem_read() / snd_ctl_elem_write()), thus it's not directly > passed from userspace. If the information includes something wrong, the > caller returns error and don't execute callee. > > The snd_ctl_elem_add() has already allocated enough memory, so there're > no wrong reference as long as the re-calculated ID includes larger numid > than (kctl->numid + kctl->count), or larger index than kctl->count. > > Are there any cases that the re-calculated ID includes such wrong > information after passing these functions below? > * snd_ctl_find_id() -> find kctl with given ID info > * snd_ctl_get_ioff() -> calculate offset with kctl and given ID > * snd_ctl_build_ioff() -> re-calculate ID with kctl and the offset > > If it still includes wrong information, it's the caller's fault and > should be fixed. Hrm, OK, although I prefer having a safer check in the callee side as a pessimist (there are lots of direct calls of put callbacks in many driver codes). Let's cross fingers. Applied now. Thanks. Takashi
diff --git a/sound/core/control.c b/sound/core/control.c index ccb1ca2..5b2c352 100644 --- a/sound/core/control.c +++ b/sound/core/control.c @@ -1029,7 +1029,7 @@ static int snd_ctl_elem_unlock(struct snd_ctl_file *file, struct user_element { struct snd_ctl_elem_info info; struct snd_card *card; - void *elem_data; /* element data */ + char *elem_data; /* element data */ unsigned long elem_data_size; /* size of element data in bytes */ void *tlv_data; /* TLV data */ unsigned long tlv_data_size; /* TLV data size */ @@ -1078,9 +1078,12 @@ static int snd_ctl_elem_user_get(struct snd_kcontrol *kcontrol, struct snd_ctl_elem_value *ucontrol) { struct user_element *ue = kcontrol->private_data; + unsigned int size = ue->elem_data_size; + char *src = ue->elem_data + + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; mutex_lock(&ue->card->user_ctl_lock); - memcpy(&ucontrol->value, ue->elem_data, ue->elem_data_size); + memcpy(&ucontrol->value, src, size); mutex_unlock(&ue->card->user_ctl_lock); return 0; } @@ -1090,11 +1093,14 @@ static int snd_ctl_elem_user_put(struct snd_kcontrol *kcontrol, { int change; struct user_element *ue = kcontrol->private_data; + unsigned int size = ue->elem_data_size; + char *dst = ue->elem_data + + snd_ctl_get_ioff(kcontrol, &ucontrol->id) * size; mutex_lock(&ue->card->user_ctl_lock); - change = memcmp(&ucontrol->value, ue->elem_data, ue->elem_data_size) != 0; + change = memcmp(&ucontrol->value, dst, size) != 0; if (change) - memcpy(ue->elem_data, &ucontrol->value, ue->elem_data_size); + memcpy(dst, &ucontrol->value, size); mutex_unlock(&ue->card->user_ctl_lock); return change; } @@ -1278,7 +1284,7 @@ static int snd_ctl_elem_add(struct snd_ctl_file *file, if (err < 0) return err; memcpy(&kctl->id, &info->id, sizeof(kctl->id)); - kctl->private_data = kzalloc(sizeof(struct user_element) + private_size, + kctl->private_data = kzalloc(sizeof(struct user_element) + private_size * count, GFP_KERNEL); if (kctl->private_data == NULL) { kfree(kctl);
An element instance can have several elements with the same feature. Some userspace applications can add such an element instance by add operation with the number of elements. Then, the element instance gets a memory object to keep states of these elements. But the element instance has just one memory object for the elements. This causes the same result to each read/write operations to the different elements. This commit fixes this bug by allocating enough memory objects to the element instance for each of elements. Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp> --- sound/core/control.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)