Re: [PATCH] drm/i915: Fix potential context UAFs

From: Tvrtko Ursulin
Date: Fri Jan 06 2023 - 05:15:25 EST



On 05/01/2023 16:00, Tvrtko Ursulin wrote:

On 05/01/2023 15:52, Andi Shyti wrote:
Hi Rob,

On Tue, Jan 03, 2023 at 03:49:46PM -0800, Rob Clark wrote:
From: Rob Clark <robdclark@xxxxxxxxxxxx>

gem_context_register() makes the context visible to userspace, and which
point a separate thread can trigger the I915_GEM_CONTEXT_DESTROY ioctl.
So we need to ensure that nothing uses the ctx ptr after this.  And we
need to ensure that adding the ctx to the xarray is the *last* thing
that gem_context_register() does with the ctx pointer.

Signed-off-by: Rob Clark <robdclark@xxxxxxxxxxxx>

Reviewed-by: Andi Shyti <andi.shyti@xxxxxxxxxxxxxxx>

I also agree with Tvrtko that we should add Stable: and Fixes:.

Yeah I'll add them all when merging. Just waiting for full CI results. It will be like this:

Fixes: eb4dedae920a ("drm/i915/gem: Delay tracking the GEM context until it is registered")
Fixes: a4c1cdd34e2c ("drm/i915/gem: Delay context creation (v3)")
Fixes: 49bd54b390c2 ("drm/i915: Track all user contexts per client")
Cc: <stable@xxxxxxxxxxxxxxx> # v5.10+

Pushed to drm-intel-gt-next - thanks for the fix and reviews.

Regards,

Tvrtko


Regards,

Tvrtko

One little thing, "user after free" is clearer that UAF :)

Thanks,
Andi

---
  drivers/gpu/drm/i915/gem/i915_gem_context.c | 24 +++++++++++++++------
  1 file changed, 18 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_context.c b/drivers/gpu/drm/i915/gem/i915_gem_context.c
index 7f2831efc798..6250de9b9196 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_context.c
@@ -1688,6 +1688,10 @@ void i915_gem_init__contexts(struct drm_i915_private *i915)
      init_contexts(&i915->gem.contexts);
  }
+/*
+ * Note that this implicitly consumes the ctx reference, by placing
+ * the ctx in the context_xa.
+ */
  static void gem_context_register(struct i915_gem_context *ctx,
                   struct drm_i915_file_private *fpriv,
                   u32 id)
@@ -1703,10 +1707,6 @@ static void gem_context_register(struct i915_gem_context *ctx,
      snprintf(ctx->name, sizeof(ctx->name), "%s[%d]",
           current->comm, pid_nr(ctx->pid));
-    /* And finally expose ourselves to userspace via the idr */
-    old = xa_store(&fpriv->context_xa, id, ctx, GFP_KERNEL);
-    WARN_ON(old);
-
      spin_lock(&ctx->client->ctx_lock);
      list_add_tail_rcu(&ctx->client_link, &ctx->client->ctx_list);
      spin_unlock(&ctx->client->ctx_lock);
@@ -1714,6 +1714,10 @@ static void gem_context_register(struct i915_gem_context *ctx,
      spin_lock(&i915->gem.contexts.lock);
      list_add_tail(&ctx->link, &i915->gem.contexts.list);
      spin_unlock(&i915->gem.contexts.lock);
+
+    /* And finally expose ourselves to userspace via the idr */
+    old = xa_store(&fpriv->context_xa, id, ctx, GFP_KERNEL);
+    WARN_ON(old);
  }
  int i915_gem_context_open(struct drm_i915_private *i915,
@@ -2199,14 +2203,22 @@ finalize_create_context_locked(struct drm_i915_file_private *file_priv,
      if (IS_ERR(ctx))
          return ctx;
+    /*
+     * One for the xarray and one for the caller.  We need to grab
+     * the reference *prior* to making the ctx visble to userspace
+     * in gem_context_register(), as at any point after that
+     * userspace can try to race us with another thread destroying
+     * the context under our feet.
+     */
+    i915_gem_context_get(ctx);
+
      gem_context_register(ctx, file_priv, id);
      old = xa_erase(&file_priv->proto_context_xa, id);
      GEM_BUG_ON(old != pc);
      proto_context_close(file_priv->dev_priv, pc);
-    /* One for the xarray and one for the caller */
-    return i915_gem_context_get(ctx);
+    return ctx;
  }
  struct i915_gem_context *
--
2.38.1