Chris Wilson wrote:
> On Tue, Dec 24, 2013 at 08:50:23PM +0900, Tetsuo Handa wrote:
> > >From 482be6384379072eb4c0d45d0ab8a25df4f59ed7 Mon Sep 17 00:00:00 2001
> > From: Tetsuo Handa <[email protected]>
> > Date: Tue, 24 Dec 2013 18:04:14 +0900
> > Subject: [PATCH] drm/i915: Fix refcount leak and possible NULL pointer dereference.
> >
> > Since get_pid_task() grabs a reference on the task_struct, we have to drop the
> > refcount after reading that task's comm name. Also, directly reading like
> > get_pid_task()->comm can trigger an oops when get_pid_task() returned NULL.
>
> The second issue is moot as file itself cannot exist if the task_struct
> is NULL, and the task_struct cannot be destroyed until we finish the
> function. The simpler fix would appear to be s/get_pid_task/pid_task/
If I understand correctly, priv->pid = get_pid(task_pid(current)); in
drm_open_helper() grabs a reference on "struct pid" before adding to
&dev->filelist, and put_pid(file_priv->pid); in drm_release() releases that
reference after removing from &dev->filelist.
So, you meant that mutex_lock_interruptible(&dev->struct_mutex); in
i915_gem_object_info() prevents drm_release() from calling put_pid() ?
Then, this file->pid in &dev->filelist keeps at least one reference.
OK. Updated patch follows.
----------
>From 5ea824bc84f65d2265addc81e1adacc8baf82d48 Mon Sep 17 00:00:00 2001
From: Tetsuo Handa <[email protected]>
Date: Fri, 3 Jan 2014 20:30:41 +0900
Subject: [PATCH v2] drm/i915: Fix refcount leak and possible NULL pointer dereference.
Since get_pid_task() grabs a reference on the task_struct, we have to drop the
refcount after reading that task's comm name. Use pid_task() with RCU instead.
Also, avoid directly reading like pid_task()->comm because
pid_task() will return NULL if the task have already exit()ed.
This patch fixes both problems.
Signed-off-by: Tetsuo Handa <[email protected]>
---
drivers/gpu/drm/i915/i915_debugfs.c | 12 +++++++++++-
1 files changed, 11 insertions(+), 1 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 6ed45a9..91c26b6 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -406,16 +406,26 @@ static int i915_gem_object_info(struct seq_file *m, void* data)
seq_putc(m, '\n');
list_for_each_entry_reverse(file, &dev->filelist, lhead) {
struct file_stats stats;
+ struct task_struct *task;
memset(&stats, 0, sizeof(stats));
idr_for_each(&file->object_idr, per_file_stats, &stats);
+ /*
+ * Although we have a valid reference on file->pid, that does
+ * not guarantee that the task_struct who called get_pid() is
+ * still alive (e.g. get_pid(current) => fork() => exit()).
+ * Therefore, we need to protect this ->comm access using RCU.
+ */
+ rcu_read_lock();
+ task = pid_task(file->pid, PIDTYPE_PID);
seq_printf(m, "%s: %u objects, %zu bytes (%zu active, %zu inactive, %zu unbound)\n",
- get_pid_task(file->pid, PIDTYPE_PID)->comm,
+ task ? task->comm : "<unknown>",
stats.count,
stats.total,
stats.active,
stats.inactive,
stats.unbound);
+ rcu_read_unlock();
}
mutex_unlock(&dev->struct_mutex);
--
1.7.1
On Fri, Jan 03, 2014 at 08:42:18PM +0900, Tetsuo Handa wrote:
> Chris Wilson wrote:
> > On Tue, Dec 24, 2013 at 08:50:23PM +0900, Tetsuo Handa wrote:
> > > >From 482be6384379072eb4c0d45d0ab8a25df4f59ed7 Mon Sep 17 00:00:00 2001
> > > From: Tetsuo Handa <[email protected]>
> > > Date: Tue, 24 Dec 2013 18:04:14 +0900
> > > Subject: [PATCH] drm/i915: Fix refcount leak and possible NULL pointer dereference.
> > >
> > > Since get_pid_task() grabs a reference on the task_struct, we have to drop the
> > > refcount after reading that task's comm name. Also, directly reading like
> > > get_pid_task()->comm can trigger an oops when get_pid_task() returned NULL.
> >
> > The second issue is moot as file itself cannot exist if the task_struct
> > is NULL, and the task_struct cannot be destroyed until we finish the
> > function. The simpler fix would appear to be s/get_pid_task/pid_task/
>
> If I understand correctly, priv->pid = get_pid(task_pid(current)); in
> drm_open_helper() grabs a reference on "struct pid" before adding to
> &dev->filelist, and put_pid(file_priv->pid); in drm_release() releases that
> reference after removing from &dev->filelist.
>
> So, you meant that mutex_lock_interruptible(&dev->struct_mutex); in
> i915_gem_object_info() prevents drm_release() from calling put_pid() ?
Right, my understanding is that since we take the struct_mutex during
release of the filp (and across this list walker) that is sufficient
serialisation to prevent the task struct from disappearing from
underneath us.
> Then, this file->pid in &dev->filelist keeps at least one reference.
> OK. Updated patch follows.
Looks good to me, and the comment is great.
-Chris
--
Chris Wilson, Intel Open Source Technology Centre
Chris Wilson wrote:
> > Then, this file->pid in &dev->filelist keeps at least one reference.
> > OK. Updated patch follows.
>
> Looks good to me, and the comment is great.
> -Chris
Please pick up this patch via your git tree.
Also, please change from "<unknown>" to whatever you like. For example,
dump_holder() in fs/gfs2/glock.c uses "(ended)" if comm is not available.
On Sun, Jan 05, 2014 at 10:08:42AM +0900, Tetsuo Handa wrote:
> Chris Wilson wrote:
> > > Then, this file->pid in &dev->filelist keeps at least one reference.
> > > OK. Updated patch follows.
> >
> > Looks good to me, and the comment is great.
> > -Chris
>
> Please pick up this patch via your git tree.
>
> Also, please change from "<unknown>" to whatever you like. For example,
> dump_holder() in fs/gfs2/glock.c uses "(ended)" if comm is not available.
Queued for -next, thanks for the patch.
-Daniel
--
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch