[PATCH v2 0/2] fastrpc: fix probe-time races between rpmsg and userspace

[PATCH v2 0/2] fastrpc: fix probe-time races between rpmsg and userspace

[Vinayak Katoch]

This series fixes two races in fastrpc_rpmsg_probe() that can be
triggered when userspace opens a fastrpc device node concurrently with
the rpmsg probe sequence.

fastrpc_device_register() calls misc_register() before the channel
context is fully initialised. A concurrent open() in that window hits
kref_get() on a zero refcount, and because fastrpc_cb_devices_create()
has not yet run sesscount is 0, the error path calls kref_put() on the
saturated refcount, triggering kfree() of the channel context. A
subsequent IRQ-path fastrpc_rpmsg_callback() then dereferences the
freed spinlock, causing a kernel panic. A NULL dereference in the
fastrpc_device_open() error path when cctx->rpdev is concurrently
cleared by fastrpc_rpmsg_remove() is also addressed.

Signed-off-by: Vinayak Katoch <vinayak.katoch@oss.qualcomm.com>
---
Changes in v2:
- Add Cc: stable to both fixes
- Free cctx via fastrpc_channel_ctx_put() in the probe error path
- Clear rpdev and wake pending users there, as remove() does
- Move fastrpc_notify_users() above probe
- Link to v1: https://lore.kernel.org/r/20261001-fastrpc-probe-fixes-v1-0-1376d93b8b3e@oss.qualcomm.com

---
Vinayak Katoch (2):
      misc: fastrpc: fix NULL rpdev dereference when session alloc fails
      misc: fastrpc: fix init ordering race in rpmsg probe

 drivers/misc/fastrpc.c | 95 +++++++++++++++++++++++++++++---------------------
 1 file changed, 56 insertions(+), 39 deletions(-)
---
base-commit: 8f7df005a7564bd7ffa75da9c3085d5b996f1e16
change-id: 20260923-fastrpc-probe-fixes-30c1980b50e6
prerequisite-change-id: 20260609-dup-sessions-ea2acaac1994:v6
prerequisite-patch-id: c8b4a3873f9cf0d69beff0f608933ae83faa4c47
prerequisite-patch-id: 05d9299adf4184e0e56d97dc063ab5bf227cc5a5
prerequisite-patch-id: da0a3d55397a522ea4b77769e387df9faea8e024

Best regards,
-- 
Vinayak Katoch <vinayak.katoch@oss.qualcomm.com>

[PATCH v2 1/2] misc: fastrpc: fix NULL rpdev dereference when session alloc fails

[Vinayak Katoch]

When a fastrpc client process dies abnormally without closing its file
descriptor, fastrpc_device_release() is never called and the compute-cb
session slot is never freed. The abnormal client death on the DSP side
triggers a glink channel teardown which propagates up through the rpmsg
bus and calls fastrpc_rpmsg_remove().

fastrpc_rpmsg_remove() nulls cctx->rpdev before calling
misc_deregister(). If a concurrent open() enters fastrpc_device_open()
in this window and fastrpc_session_alloc() returns NULL (sessions
exhausted by the leaked slot), the error path dereferences
cctx->rpdev->dev causing a NULL pointer dereference:

  Unable to handle kernel NULL pointer dereference at virtual address 0000000000000000
  pc : fastrpc_device_open+0x1b8/0x258 [fastrpc]
  Call trace:
   fastrpc_device_open+0x1b8/0x258 [fastrpc] (P)
   misc_open+0xd8/0x1a0

Fix by moving misc_deregister() before rpdev = NULL. misc_open() and
misc_deregister() both take misc_mtx, so misc_deregister() will either
block until fastrpc_device_open() completes or prevent new opens from
entering it entirely.

Fixes: 7c11df42d0c7 ("misc: fastrpc: Fix device_open when no session is available")
Cc: stable@vger.kernel.org
Reviewed-by: Ekansh Gupta <ekansh.gupta@oss.qualcomm.com>
Signed-off-by: Vinayak Katoch <vinayak.katoch@oss.qualcomm.com>
---
 drivers/misc/fastrpc.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
index 5dd1990d3e2a..f280bff2edf0 100644
--- a/drivers/misc/fastrpc.c
+++ b/drivers/misc/fastrpc.c
@@ -2697,6 +2697,12 @@ static void fastrpc_rpmsg_remove(struct rpmsg_device *rpdev)
 	struct fastrpc_user *user;
 	unsigned long flags;
 
+	if (cctx->fdevice)
+		misc_deregister(&cctx->fdevice->miscdev);
+
+	if (cctx->secure_fdevice)
+		misc_deregister(&cctx->secure_fdevice->miscdev);
+
 	/* No invocations past this point */
 	spin_lock_irqsave(&cctx->lock, flags);
 	cctx->rpdev = NULL;
@@ -2704,12 +2710,6 @@ static void fastrpc_rpmsg_remove(struct rpmsg_device *rpdev)
 		fastrpc_notify_users(user);
 	spin_unlock_irqrestore(&cctx->lock, flags);
 
-	if (cctx->fdevice)
-		misc_deregister(&cctx->fdevice->miscdev);
-
-	if (cctx->secure_fdevice)
-		misc_deregister(&cctx->secure_fdevice->miscdev);
-
 	list_for_each_entry_safe(buf, b, &cctx->invoke_interrupted_mmaps, node)
 		list_del(&buf->node);
 

-- 
2.34.1

[PATCH v2 2/2] misc: fastrpc: fix init ordering race in rpmsg probe

[Vinayak Katoch]

fastrpc_device_register() calls misc_register() which makes the device
node visible to userspace. However, kref_init(), spin_lock_init(),
INIT_LIST_HEAD(), dev_set_drvdata() and fastrpc_cb_devices_create()
were all called after the switch/misc_register block, leaving a window
where userspace can open the device and call fastrpc_device_open() on
a partially initialised channel context.

On a concurrent open(), fastrpc_channel_ctx_get() calls kref_get() on
a kzalloc-zeroed, never kref_init()'d refcount of 0, triggering:

  refcount_t: addition on 0; use-after-free.
  Call trace:
   refcount_warn_saturate+0x120/0x148 (P)
   fastrpc_device_open+0x204/0x258 [fastrpc]
   misc_open+0xd8/0x1a0

Additionally, fastrpc_cb_devices_create() not having run means
sesscount is still 0, so fastrpc_session_alloc() returns NULL and
fastrpc_device_open() hits the error path which calls
fastrpc_channel_ctx_put() on the already-bogus refcount. This
triggers the free callback, kfree()-ing the channel context. A
subsequent IRQ-path fastrpc_rpmsg_callback() then dereferences the
freed spinlock:

  Unable to handle kernel paging request at virtual address ffffcd5841510ca0
  Internal error: Oops: 0000000096000047 [#1] SMP
  Call trace:
   queued_spin_lock_slowpath+0x414/0x5e0 (P)
   _raw_spin_lock_irqsave+0x74/0x90
   fastrpc_rpmsg_callback+0x4c/0xf0 [fastrpc]
   qcom_glink_native_rx+0x6c4/0xf10
   qcom_glink_smem_intr+0x1c/0x38 [qcom_glink_smem]
  Kernel panic - not syncing: Oops: Fatal exception in interrupt

Fix by moving all channel context initialisation (kref_init,
spin_lock_init, INIT_LIST_HEAD, idr_init, dev_set_drvdata) and
fastrpc_cb_devices_create() before fastrpc_device_register(), ensuring
the struct is fully ready before the device node is visible to
userspace.

Add err_depopulate label in the error path since
fastrpc_cb_devices_create() now runs before misc_register(); a
fastrpc_device_register() failure must call fastrpc_cb_devices_destroy()
to clean up the compute-cb children.

On CDSP/GDSP the secure node is registered before the non-secure one, so
an application may already have opened the first node and hold a channel
context reference when the second registration fails. Freeing the
context with kfree() there would leave that user with a dangling
pointer. Handle it like fastrpc_rpmsg_remove(): after the nodes are
deregistered, clear rpdev, wake pending invocations with -EPIPE and drop
only the probe reference with fastrpc_channel_ctx_put(), so the last
holder frees the context. Move fastrpc_notify_users() above probe so it
can be used there.

Fixes: 3abe3ab3cdab ("misc: fastrpc: add secure domain support")
Cc: stable@vger.kernel.org
Signed-off-by: Vinayak Katoch <vinayak.katoch@oss.qualcomm.com>
---
 drivers/misc/fastrpc.c | 83 ++++++++++++++++++++++++++++++--------------------
 1 file changed, 50 insertions(+), 33 deletions(-)

diff --git a/drivers/misc/fastrpc.c b/drivers/misc/fastrpc.c
index f280bff2edf0..fc09cd27af4b 100644
--- a/drivers/misc/fastrpc.c
+++ b/drivers/misc/fastrpc.c
@@ -2548,6 +2548,18 @@ static int fastrpc_init_reserved_mem(struct fastrpc_channel_ctx *cctx,
 				  cctx->vmperms, cctx->vmcount);
 }
 
+static void fastrpc_notify_users(struct fastrpc_user *user)
+{
+	struct fastrpc_invoke_ctx *ctx;
+
+	spin_lock(&user->lock);
+	list_for_each_entry(ctx, &user->pending, node) {
+		ctx->retval = -EPIPE;
+		complete(&ctx->work);
+	}
+	spin_unlock(&user->lock);
+}
+
 static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 {
 	struct device *rdev = &rpdev->dev;
@@ -2557,6 +2569,8 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 	bool secure_dsp;
 	unsigned int vmids[FASTRPC_MAX_VMIDS];
 	const struct fastrpc_soc_data *soc_data;
+	struct fastrpc_user *user;
+	unsigned long flags;
 
 	soc_data = device_get_match_data(rdev);
 
@@ -2605,6 +2619,23 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 	data->poll_mode_supported = soc_data->poll_mode_supported ||
 		of_machine_get_match(fastrpc_poll_supported_machines);
 
+	kref_init(&data->refcount);
+	atomic_set(&data->ctx_seq, 0);
+
+	rdev->dma_mask = &data->dma_mask;
+	dma_set_mask_and_coherent(rdev, DMA_BIT_MASK(32));
+	INIT_LIST_HEAD(&data->users);
+	INIT_LIST_HEAD(&data->invoke_interrupted_mmaps);
+	spin_lock_init(&data->lock);
+	idr_init(&data->ctx_idr);
+	data->domain_id = domain_id;
+	data->rpdev = rpdev;
+	dev_set_drvdata(&rpdev->dev, data);
+
+	err = fastrpc_cb_devices_create(rpdev);
+	if (err)
+		goto err_free_data;
+
 	switch (domain_id) {
 	case ADSP_DOMAIN_ID:
 	case MDSP_DOMAIN_ID:
@@ -2613,7 +2644,7 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 		data->unsigned_support = false;
 		err = fastrpc_device_register(rdev, data, secure_dsp, domain);
 		if (err)
-			goto err_free_data;
+			goto err_depopulate;
 		break;
 	case CDSP_DOMAIN_ID:
 	case GDSP_DOMAIN_ID:
@@ -2621,7 +2652,7 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 		/* Create both device nodes so that we can allow both Signed and Unsigned PD */
 		err = fastrpc_device_register(rdev, data, true, domain);
 		if (err)
-			goto err_free_data;
+			goto err_depopulate;
 
 		err = fastrpc_device_register(rdev, data, false, domain);
 		if (err)
@@ -2629,30 +2660,12 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 		break;
 	default:
 		err = -EINVAL;
-		goto err_free_data;
+		goto err_depopulate;
 	}
 
-	kref_init(&data->refcount);
-	atomic_set(&data->ctx_seq, 0);
-
-	rdev->dma_mask = &data->dma_mask;
-	dma_set_mask_and_coherent(rdev, DMA_BIT_MASK(32));
-	INIT_LIST_HEAD(&data->users);
-	INIT_LIST_HEAD(&data->invoke_interrupted_mmaps);
-	spin_lock_init(&data->lock);
-	idr_init(&data->ctx_idr);
-	data->domain_id = domain_id;
-	data->rpdev = rpdev;
-	dev_set_drvdata(&rpdev->dev, data);
-
-	err = fastrpc_cb_devices_create(rpdev);
-	if (err)
-		goto err_deregister_fdev;
-
 	if (data->domain_id == ADSP_DOMAIN_ID && data->sesscount > 0) {
 		struct fastrpc_session_ctx *last_sess;
 		struct fastrpc_session_ctx *dup_sess;
-		unsigned long flags;
 
 		spin_lock_irqsave(&data->lock, flags);
 		last_sess = &data->session[data->sesscount - 1];
@@ -2673,23 +2686,27 @@ static int fastrpc_rpmsg_probe(struct rpmsg_device *rpdev)
 	if (data->secure_fdevice)
 		misc_deregister(&data->secure_fdevice->miscdev);
 
+err_depopulate:
+	/*
+	 * A registered device node may already have been opened, so a user
+	 * can hold a cctx reference. Drop only our reference instead of
+	 * freeing the context directly.
+	 */
+	spin_lock_irqsave(&data->lock, flags);
+	data->rpdev = NULL;
+	list_for_each_entry(user, &data->users, user)
+		fastrpc_notify_users(user);
+	spin_unlock_irqrestore(&data->lock, flags);
+
+	fastrpc_cb_devices_destroy(rpdev);
+	fastrpc_channel_ctx_put(data);
+	return err;
+
 err_free_data:
 	kfree(data);
 	return err;
 }
 
-static void fastrpc_notify_users(struct fastrpc_user *user)
-{
-	struct fastrpc_invoke_ctx *ctx;
-
-	spin_lock(&user->lock);
-	list_for_each_entry(ctx, &user->pending, node) {
-		ctx->retval = -EPIPE;
-		complete(&ctx->work);
-	}
-	spin_unlock(&user->lock);
-}
-
 static void fastrpc_rpmsg_remove(struct rpmsg_device *rpdev)
 {
 	struct fastrpc_channel_ctx *cctx = dev_get_drvdata(&rpdev->dev);

-- 
2.34.1