| Message ID | 20220327053943.3071-1-xiam0nd.tong@gmail.com (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | viresh kumar |
| Headers | show |
| Series | opp: fix a missing check on list iterator | expand |
On 27-03-22, 13:39, Xiaomeng Tong wrote: > The bug is here: > dev = new_dev->dev; > > The list iterator 'new_dev' will point to a bogus position containing > HEAD if the list is empty or no element is found. This case must > be checked before any use of the iterator, otherwise it will lead > to a invalid memory access. > > To fix this bug, add an check. Use a new variable 'iter' as the > list iterator, while use the old variable 'new_dev' as a dedicated > pointer to point to the found element. > > Cc: stable@vger.kernel.org > Fixes: deaa51465105a ("PM / OPP: Add debugfs support") > Signed-off-by: Xiaomeng Tong <xiam0nd.tong@gmail.com> > --- > drivers/opp/debugfs.c | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/drivers/opp/debugfs.c b/drivers/opp/debugfs.c > index 596c185b5dda..a4476985e4ce 100644 > --- a/drivers/opp/debugfs.c > +++ b/drivers/opp/debugfs.c > @@ -187,14 +187,19 @@ void opp_debug_register(struct opp_device *opp_dev, struct opp_table *opp_table) > static void opp_migrate_dentry(struct opp_device *opp_dev, > struct opp_table *opp_table) > { > - struct opp_device *new_dev; > + struct opp_device *new_dev = NULL, *iter; > const struct device *dev; > struct dentry *dentry; > > /* Look for next opp-dev */ > - list_for_each_entry(new_dev, &opp_table->dev_list, node) > - if (new_dev != opp_dev) > + list_for_each_entry(iter, &opp_table->dev_list, node) > + if (iter != opp_dev) { > + new_dev = iter; > break; > + } > + > + if (!new_dev) > + return; I think you missed this check in the parent function ? if (!list_is_singular(&opp_table->dev_list)) { i.e. this bug can never happen.
On Mon, 28 Mar 2022 08:47:39 +0530, Viresh Kumar wrote: > > diff --git a/drivers/opp/debugfs.c b/drivers/opp/debugfs.c > > index 596c185b5dda..a4476985e4ce 100644 > > --- a/drivers/opp/debugfs.c > > +++ b/drivers/opp/debugfs.c > > @@ -187,14 +187,19 @@ void opp_debug_register(struct opp_device *opp_dev, struct opp_table *opp_table) > > static void opp_migrate_dentry(struct opp_device *opp_dev, > > struct opp_table *opp_table) > > { > > - struct opp_device *new_dev; > > + struct opp_device *new_dev = NULL, *iter; > > const struct device *dev; > > struct dentry *dentry; > > > > /* Look for next opp-dev */ > > - list_for_each_entry(new_dev, &opp_table->dev_list, node) > > - if (new_dev != opp_dev) > > + list_for_each_entry(iter, &opp_table->dev_list, node) > > + if (iter != opp_dev) { > > + new_dev = iter; > > break; > > + } > > + > > + if (!new_dev) > > + return; > > I think you missed this check in the parent function ? > > if (!list_is_singular(&opp_table->dev_list)) { > > > i.e. this bug can never happen. > No. the conditon to call opp_migrate_dentry(opp_dev, opp_table); is: if (!list_is_singular(&opp_table->dev_list)), while list_is_singlular is: !list_empty(head) && (head->next == head->prev); so the condition is: list_empty(head) || (head->next != head->prev) if the list is empty, the bug can be triggered. -- Xiaomeng Tong
On 28-03-22, 15:43, Xiaomeng Tong wrote: > No. the conditon to call opp_migrate_dentry(opp_dev, opp_table); is: > if (!list_is_singular(&opp_table->dev_list)), > > while list_is_singlular is: !list_empty(head) && (head->next == head->prev); > > so the condition is: list_empty(head) || (head->next != head->prev) > > if the list is empty, the bug can be triggered. List can't be empty here by design. It will be a huge bug in that case, which should lead to crash somewhere.
On Mon, 28 Mar 2022 14:20:57 +0530, Viresh Kumar wrote: > On 28-03-22, 15:43, Xiaomeng Tong wrote: > > No. the conditon to call opp_migrate_dentry(opp_dev, opp_table); is: > > if (!list_is_singular(&opp_table->dev_list)), > > > > while list_is_singlular is: !list_empty(head) && (head->next == head->prev); > > > > so the condition is: list_empty(head) || (head->next != head->prev) > > > > if the list is empty, the bug can be triggered. > > List can't be empty here by design. It will be a huge bug in that > case, which should lead to crash somewhere. > There is anther condition to trigger this bug: the list is not empty and no element found (if (iter != opp_dev)). -- Xiaomeng Tong
On 28-03-22, 17:13, Xiaomeng Tong wrote: > On Mon, 28 Mar 2022 14:20:57 +0530, Viresh Kumar wrote: > > On 28-03-22, 15:43, Xiaomeng Tong wrote: > > > No. the conditon to call opp_migrate_dentry(opp_dev, opp_table); is: > > > if (!list_is_singular(&opp_table->dev_list)), > > > > > > while list_is_singlular is: !list_empty(head) && (head->next == head->prev); > > > > > > so the condition is: list_empty(head) || (head->next != head->prev) > > > > > > if the list is empty, the bug can be triggered. > > > > List can't be empty here by design. It will be a huge bug in that > > case, which should lead to crash somewhere. > > > > There is anther condition to trigger this bug: the list is not empty and > no element found (if (iter != opp_dev)). I suggest reading the code again, considering opp_debug_unregister() as well. What's happening here is this: - Several devices share the OPP table. - One of them (devX) is going away and opp_debug_unregister() is called for this device. - If devX is the last device for this OPP table, then we don't migrate and just release all resources. - Otherwise, we migrate it to the next element in the list. i.e. any device which != devX. Please tell based on this where do you see a possibility of a bug. Surely there can be one, but I fail to see it at the moment and need more detail of the same. Thanks Xiaomeng.
On Mon, 28 Mar 2022 15:09:33 +0530, Viresh Kumar wrote: > On 28-03-22, 17:13, Xiaomeng Tong wrote: > > On Mon, 28 Mar 2022 14:20:57 +0530, Viresh Kumar wrote: > > > On 28-03-22, 15:43, Xiaomeng Tong wrote: > > > > No. the conditon to call opp_migrate_dentry(opp_dev, opp_table); is: > > > > if (!list_is_singular(&opp_table->dev_list)), > > > > > > > > while list_is_singlular is: !list_empty(head) && (head->next == head->prev); > > > > > > > > so the condition is: list_empty(head) || (head->next != head->prev) > > > > > > > > if the list is empty, the bug can be triggered. > > > > > > List can't be empty here by design. It will be a huge bug in that > > > case, which should lead to crash somewhere. > > > > > > > There is anther condition to trigger this bug: the list is not empty and > > no element found (if (iter != opp_dev)). > > I suggest reading the code again, considering opp_debug_unregister() > as well. > > What's happening here is this: > > - Several devices share the OPP table. > - One of them (devX) is going away and opp_debug_unregister() is called for this device. > - If devX is the last device for this OPP table, then we don't migrate > and just release all resources. > - Otherwise, we migrate it to the next element in the list. i.e. any > device which != devX. > > Please tell based on this where do you see a possibility of a bug. > Surely there can be one, but I fail to see it at the moment and need > more detail of the same. > Perhaps you are right. Anyway, It is a good choise to use list iterator only inside the loop as linus suggested [1], to avoid potential risk. I have also repost another patch with changed commit message. Please check it, thank you. [1]:https://lore.kernel.org/lkml/20220301075839.4156-1-xiam0nd.tong@gmail.com/ -- Xiaomeng Tong
diff --git a/drivers/opp/debugfs.c b/drivers/opp/debugfs.c index 596c185b5dda..a4476985e4ce 100644 --- a/drivers/opp/debugfs.c +++ b/drivers/opp/debugfs.c @@ -187,14 +187,19 @@ void opp_debug_register(struct opp_device *opp_dev, struct opp_table *opp_table) static void opp_migrate_dentry(struct opp_device *opp_dev, struct opp_table *opp_table) { - struct opp_device *new_dev; + struct opp_device *new_dev = NULL, *iter; const struct device *dev; struct dentry *dentry; /* Look for next opp-dev */ - list_for_each_entry(new_dev, &opp_table->dev_list, node) - if (new_dev != opp_dev) + list_for_each_entry(iter, &opp_table->dev_list, node) + if (iter != opp_dev) { + new_dev = iter; break; + } + + if (!new_dev) + return; /* new_dev is guaranteed to be valid here */ dev = new_dev->dev;
The bug is here: dev = new_dev->dev; The list iterator 'new_dev' will point to a bogus position containing HEAD if the list is empty or no element is found. This case must be checked before any use of the iterator, otherwise it will lead to a invalid memory access. To fix this bug, add an check. Use a new variable 'iter' as the list iterator, while use the old variable 'new_dev' as a dedicated pointer to point to the found element. Cc: stable@vger.kernel.org Fixes: deaa51465105a ("PM / OPP: Add debugfs support") Signed-off-by: Xiaomeng Tong <xiam0nd.tong@gmail.com> --- drivers/opp/debugfs.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)