Skip to contentSkip to conversation

rpmb: Fix request serialisation and teardown races

PATCH Applied Stanley Jhu 3 participants 7 messages Nov 12 – Nov 18

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

v6 cover rpmb: Fix request serialisation and teardown races

Jens, Martin,

Patch 1/3 serializes requests and provider unregistration in drivers/misc/rpmb-core.c. Patches 2/3 and 3/3 fix UFS RPMB device lifetime and drop the unregistered bus in drivers/ufs/core/ufs-rpmb.c so regions register again after commit 36f35b8df697 ("driver core: reject devices with unregistered buses").

Patches 2/3 and 3/3 depend on 1/3 at runtime to avoid a use-after-free on host unbind. Jens, if 1/3 looks good to you, could you give an Acked-by so Martin can take all three patches in order through the SCSI tree?

Changes since v5:
- Rebase onto mkp/scsi.git for-next.
- 1/3: clear descr.route_frames under rdev->lock instead of dead flag.
- 2/3: pin sdev_gendev via get_device(), drop Bean's tag on 1/3, 2/3.
- v5: https://lore.kernel.org/r/[email protected]

Stanley Jhu (3):
rpmb: core: Guard frame requests and teardown with mutex
scsi: ufs: rpmb: Decouple device lifecycle from devres to avoid UAF
scsi: ufs: rpmb: Drop the unregistered ufs_rpmb bus

 drivers/misc/rpmb-core.c    | 20 ++++++++-
 drivers/ufs/core/ufs-rpmb.c | 84 +++++++++++++++++++------------------
 include/linux/rpmb.h        |  3 ++
 3 files changed, 65 insertions(+), 42 deletions(-)

base-commit: 0fc91e802801819a0c51d30b6eecac63ab4f85d2

Signature · 1 line
2.56.0.567.gdb80ae8cd5-goog
#1 lore raw

PATCH v6 1/3 rpmb: core: Guard frame requests and teardown with mutex

rpmb_route_frames() invokes the provider callback without serialization, which causes two races:

1. Concurrent callers on the same rpmb_dev interleave multi-frame authenticated sequences on the region's single result register and receive each other's responses without error.
2. rpmb_dev_unregister() calls device_del() and allows the parent provider device to be freed while an in-flight rpmb_route_frames() call still dereferences rdev->dev.parent, causing a use-after-free.

Serialize rpmb_route_frames() and rpmb_dev_unregister() with a mutex in struct rpmb_dev and clear descr.route_frames on unregistration so new requests fail with -ENODEV. Also update the stale kerneldoc note advising providers to unregister from their parent release callback, which cannot run while the child rpmb_dev still holds a parent reference.

Fixes 1e9046e3a154 ("rpmb: add Replay Protected Memory Block (RPMB) subsystem")

---
Jens, if this looks good to you, an Acked-by would let Martin take the whole series through the SCSI tree. Patches 2/3 and 3/3 depend on this one.

 drivers/misc/rpmb-core.c | 20 ++++++++++++++++++--
 include/linux/rpmb.h     |  3 +++
 2 files changed, 21 insertions(+), 2 deletions(-)
drivers/misc/rpmb-core.c +18 -2
index ecf14acf230a..0fda893dd74d 100644
@@ -45,6 +45,8 @@ EXPORT_SYMBOL_GPL(rpmb_dev_put);
  * @rsp:	rpmb response frames
  * @rsp_len:	length of rpmb response frames in bytes
  *
+ * Context: Might sleep.
+ *
  * Returns: < 0 on failure
  */
 int rpmb_route_frames(struct rpmb_dev *rdev, u8 *req,
@@ -53,6 +55,10 @@ int rpmb_route_frames(struct rpmb_dev *rdev, u8 *req,
 	if (!req || !req_len || !rsp || !rsp_len)
 		return -EINVAL;
 
+	guard(mutex)(&rdev->lock);
+	if (!rdev->descr.route_frames)
+		return -ENODEV;
+
 	return rdev->descr.route_frames(rdev->dev.parent, req, req_len,
 					rsp, rsp_len);
 }
@@ -62,6 +68,7 @@ static void rpmb_dev_release(struct device *dev)
 {
 	struct rpmb_dev *rdev = to_rpmb_dev(dev);
 
+	mutex_destroy(&rdev->lock);
 	ida_free(&rpmb_ida, rdev->id);
 	kfree(rdev->descr.dev_id);
 	kfree(rdev);
@@ -123,8 +130,9 @@ EXPORT_SYMBOL_GPL(rpmb_interface_unregister);
  * rpmb_dev_unregister() - unregister RPMB partition from the RPMB subsystem
  * @rdev: the rpmb device to unregister
  *
- * This function should be called from the release function of the
- * underlying device used when the RPMB device was registered.
+ * This function should be called from the remove or unbind callback of the
+ * underlying device used when the RPMB device was registered, never from
+ * a device release callback.
  *
  * Returns: < 0 on failure
  */
@@ -133,6 +141,12 @@ int rpmb_dev_unregister(struct rpmb_dev *rdev)
 	if (!rdev)
 		return -EINVAL;
 
+	scoped_guard(mutex, &rdev->lock) {
+		if (!rdev->descr.route_frames)
+			return 0;
+		rdev->descr.route_frames = NULL;
+	}
+
 	device_del(&rdev->dev);
 
 	rpmb_dev_put(rdev);
@@ -164,6 +178,7 @@ struct rpmb_dev *rpmb_dev_register(struct device *dev,
 	rdev = kzalloc_obj(*rdev);
 	if (!rdev)
 		return ERR_PTR(-ENOMEM);
+	mutex_init(&rdev->lock);
 	rdev->descr = *descr;
 	rdev->descr.dev_id = kmemdup(descr->dev_id, descr->dev_id_len,
 				     GFP_KERNEL);
@@ -194,6 +209,7 @@ struct rpmb_dev *rpmb_dev_register(struct device *dev,
 err_free_dev_id:
 	kfree(rdev->descr.dev_id);
 err_free_rdev:
+	mutex_destroy(&rdev->lock);
 	kfree(rdev);
 	return ERR_PTR(ret);
 }
include/linux/rpmb.h +3 -0
index ed3f8e431eff..11a5d2646e38 100644
@@ -7,6 +7,7 @@
 #define __RPMB_H__
 
 #include <linux/device.h>
+#include <linux/mutex.h>
 #include <linux/types.h>
 
 /**
@@ -48,12 +49,14 @@ struct rpmb_descr {
  * struct rpmb_dev - device which can support RPMB partition
  *
  * @dev              : device
+ * @lock             : serialises requests and protects them against teardown
  * @id               : device_id
  * @list_node        : linked list node
  * @descr            : RPMB description
  */
 struct rpmb_dev {
 	struct device dev;
+	struct mutex lock;
 	int id;
 	struct list_head list_node;
 	struct rpmb_descr descr;
Signature · 1 line
2.56.0.567.gdb80ae8cd5-goog
Tested-by on v6 1/3
#2 lore raw

Re: [PATCH v6 1/3] rpmb: core: Guard frame requests and teardown with mutex

Quoted 18 lines · Stanley Jhu

rpmb_route_frames() invokes the provider callback without serialization, which causes two races:

1. Concurrent callers on the same rpmb_dev interleave multi-frame authenticated sequences on the region's single result register and receive each other's responses without error.
2. rpmb_dev_unregister() calls device_del() and allows the parent provider device to be freed while an in-flight rpmb_route_frames() call still dereferences rdev->dev.parent, causing a use-after-free.

Serialize rpmb_route_frames() and rpmb_dev_unregister() with a mutex in struct rpmb_dev and clear descr.route_frames on unregistration so new requests fail with -ENODEV. Also update the stale kerneldoc note advising providers to unregister from their parent release callback, which cannot run while the child rpmb_dev still holds a parent reference.

Quoted 7 lines
Fixes 1e9046e3a154 ("rpmb: add Replay Protected Memory Block (RPMB) subsystem")

---
Jens, if this looks good to you, an Acked-by would let Martin take the whole series through the SCSI tree. Patches 2/3 and 3/3 depend on this one.

Thanks a lot Stanley!

Works on my Rockchip RK3576 with two UFS 2.2 parts, FORESEE C2G071 and BIWIN BWU2A0526B128G. Two kthreads issuing GET_WRITE_COUNTER, 2000 requests each, without the series (bus registration restored by other means):

FORESEE: MISMATCH=3814 ERRORS=0 (plus wrong response types)
BIWIN: MISMATCH=0 ERRORS=0

The BIWIN part keeps the responses apart even though a trace on the RPMB WLUN shows the SECURITY PROTOCOL OUT/IN pairs interleaving, so the corruption is device dependent. With the series applied both parts report MISMATCH=0 ERRORS=0, and the trace shows each OUT/IN pair dispatched together.

Best regards,
Alexey

#3 lore raw

PATCH v6 2/3 scsi: ufs: rpmb: Decouple device lifecycle from devres to avoid UAF

In reply to Stanley Jhu

Region device lifetime in ufs-rpmb fails on host removal in three ways:

1. ufs_rpmb_probe() allocates struct ufs_rpmb_dev with devm_kzalloc() on the host device even though it embeds a reference-counted struct device, so devres frees the memory on host unbind regardless of live references.
2. Probe takes no reference on hba->ufs_rpmb_wlun, so scsi_remove_host() frees the scsi_device while an RPMB request is still in flight.
3. Child rpmb_dev holds a parent reference on ufs_rpmb->dev until rpmb_dev_unregister() runs, which sits inside ufs_rpmb_device_release() and therefore never runs on unbind.

Allocate struct ufs_rpmb_dev with kzalloc_obj() and free it in ufs_rpmb_device_release(), hold a get_device() reference on the WLUN across the region lifetime, call rpmb_dev_unregister() explicitly during removal and probe unwind, and reject requests when the WLUN is offline.

Fixes b06b8c421485 ("scsi: ufs: core: Add OP-TEE based RPMB driver for UFS devices")
---
 drivers/ufs/core/ufs-rpmb.c | 79 +++++++++++++++++++++----------------
 1 file changed, 44 insertions(+), 35 deletions(-)
drivers/ufs/core/ufs-rpmb.c +44 -35
index 65f2d71cf073..11d2d8e789d0 100644
@@ -14,6 +14,7 @@
 #include <linux/module.h>
 #include <linux/device.h>
 #include <linux/kernel.h>
+#include <linux/slab.h>
 #include <linux/types.h>
 #include <linux/rpmb.h>
 #include <linux/string.h>
@@ -36,13 +37,14 @@ struct ufs_rpmb_dev {
 	u8 region_id;
 	struct device dev;
 	struct rpmb_dev *rdev;
-	struct ufs_hba *hba;
+	struct scsi_device *sdev;
 	struct list_head node;
 };
 
-static int ufs_sec_submit(struct ufs_hba *hba, u16 spsp, void *buffer, size_t len, bool send)
+static int ufs_sec_submit(struct ufs_rpmb_dev *ufs_rpmb, u16 spsp,
+			  void *buffer, size_t len, bool send)
 {
-	struct scsi_device *sdev = hba->ufs_rpmb_wlun;
+	struct scsi_device *sdev = ufs_rpmb->sdev;
 	struct scsi_failure failure_defs[] = {
 		{
 			.sense_key = UNIT_ATTENTION,
@@ -62,6 +64,9 @@ static int ufs_sec_submit(struct ufs_hba *hba, u16 spsp, void *buffer, size_t le
 	};
 	u8 cdb[12] = { };
 
+	if (!sdev || !scsi_device_online(sdev))
+		return -ENODEV;
+
 	cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
 	cdb[1] = UFS_RPMB_SEC_PROTOCOL;
 	put_unaligned_be16(spsp, &cdb[2]);
@@ -74,13 +79,12 @@ static int ufs_sec_submit(struct ufs_hba *hba, u16 spsp, void *buffer, size_t le
 
 /* UFS RPMB route frames implementation */
 static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_len, u8 *resp,
-					unsigned int resp_len)
+				 unsigned int resp_len)
 {
 	struct ufs_rpmb_dev *ufs_rpmb = dev_get_drvdata(dev);
 	struct rpmb_frame *frm_out = (struct rpmb_frame *)req;
 	bool need_result_read = true;
 	u16 req_type, protocol_id;
-	struct ufs_hba *hba;
 	int ret;
 
 	if (!ufs_rpmb) {
@@ -88,8 +92,6 @@ static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_l
 		return -ENODEV;
 	}
 
-	hba = ufs_rpmb->hba;
-
 	/* req_resp is at the end of an RPMB frame. */
 	if (req_len < sizeof(*frm_out))
 		return -EINVAL;
@@ -122,7 +124,7 @@ static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_l
 
 	protocol_id = ufs_rpmb->region_id << 8 | UFS_RPMB_SEC_PROTOCOL_ID;
 
-	ret = ufs_sec_submit(hba, protocol_id, req, req_len, true);
+	ret = ufs_sec_submit(ufs_rpmb, protocol_id, req, req_len, true);
 	if (ret) {
 		dev_err(dev, "Command failed with ret=%d\n", ret);
 		return ret;
@@ -133,7 +135,8 @@ static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_l
 
 		memset(frm_resp, 0, sizeof(*frm_resp));
 		put_unaligned_be16(RPMB_RESULT_READ, &frm_resp->req_resp);
-		ret = ufs_sec_submit(hba, protocol_id, resp, resp_len, true);
+		ret = ufs_sec_submit(ufs_rpmb, protocol_id, resp, resp_len,
+				     true);
 		if (ret) {
 			dev_err(dev, "Result read request failed with ret=%d\n", ret);
 			return ret;
@@ -141,7 +144,8 @@ static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_l
 	}
 
 	if (!ret) {
-		ret = ufs_sec_submit(hba, protocol_id, resp, resp_len, false);
+		ret = ufs_sec_submit(ufs_rpmb, protocol_id, resp, resp_len,
+				     false);
 		if (ret)
 			dev_err(dev, "Response read failed with ret=%d\n", ret);
 	}
@@ -151,23 +155,31 @@ static int ufs_rpmb_route_frames(struct device *dev, u8 *req, unsigned int req_l
 
 static void ufs_rpmb_device_release(struct device *dev)
 {
-	struct ufs_rpmb_dev *ufs_rpmb = dev_get_drvdata(dev);
+	struct ufs_rpmb_dev *ufs_rpmb =
+		container_of(dev, struct ufs_rpmb_dev, dev);
 
-	rpmb_dev_unregister(ufs_rpmb->rdev);
+	put_device(&ufs_rpmb->sdev->sdev_gendev);
+	kfree(ufs_rpmb);
 }
 
 /* UFS RPMB device registration */
 int ufs_rpmb_probe(struct ufs_hba *hba)
 {
+	struct rpmb_descr descr = {
+		.type = RPMB_TYPE_UFS,
+		.route_frames = ufs_rpmb_route_frames,
+		.reliable_wr_count = hba->dev_info.rpmb_io_size,
+	};
+	struct scsi_device *sdev = hba->ufs_rpmb_wlun;
 	struct ufs_rpmb_dev *ufs_rpmb, *it, *tmp;
 	u8 dev_id[UFS_RPMB_ID_LEN];
 	struct rpmb_dev *rdev;
-	char *cid = NULL;
+	char *cid;
 	int region;
 	u32 cap;
 	int ret;
 
-	if (!hba->ufs_rpmb_wlun || hba->dev_info.b_advanced_rpmb_en) {
+	if (!sdev || hba->dev_info.b_advanced_rpmb_en) {
 		dev_info(hba->dev, "Skip OP-TEE RPMB registration\n");
 		return -ENODEV;
 	}
@@ -178,25 +190,22 @@ int ufs_rpmb_probe(struct ufs_hba *hba)
 		return -EINVAL;
 	}
 
-	struct rpmb_descr descr = {
-		.type = RPMB_TYPE_UFS,
-		.route_frames = ufs_rpmb_route_frames,
-		.reliable_wr_count = hba->dev_info.rpmb_io_size,
-	};
-
 	for (region = 0; region < ARRAY_SIZE(hba->dev_info.rpmb_region_size); region++) {
 		cap = hba->dev_info.rpmb_region_size[region];
 		if (!cap)
 			continue;
 
-		ufs_rpmb = devm_kzalloc(hba->dev, sizeof(*ufs_rpmb), GFP_KERNEL);
+		ufs_rpmb = kzalloc_obj(*ufs_rpmb);
 		if (!ufs_rpmb) {
 			ret = -ENOMEM;
 			goto err_out;
 		}
 
-		ufs_rpmb->hba = hba;
-		ufs_rpmb->dev.parent = &hba->ufs_rpmb_wlun->sdev_gendev;
+		get_device(&sdev->sdev_gendev);
+
+		ufs_rpmb->sdev = sdev;
+		ufs_rpmb->region_id = region;
+		ufs_rpmb->dev.parent = &sdev->sdev_gendev;
 		ufs_rpmb->dev.bus = &ufs_rpmb_bus_type;
 		ufs_rpmb->dev.release = ufs_rpmb_device_release;
 		dev_set_name(&ufs_rpmb->dev, "ufs_rpmb%d", region);
@@ -207,16 +216,14 @@ int ufs_rpmb_probe(struct ufs_hba *hba)
 		ret = device_register(&ufs_rpmb->dev);
 		if (ret) {
 			dev_err(hba->dev, "Failed to register UFS RPMB device %d\n", region);
-			put_device(&ufs_rpmb->dev);
-			goto err_out;
+			goto err_put;
 		}
 
 		/* Create unique ID by appending region number to device_id */
 		cid = kasprintf(GFP_KERNEL, "%s-R%d", hba->dev_info.device_id, region);
 		if (!cid) {
-			device_unregister(&ufs_rpmb->dev);
 			ret = -ENOMEM;
-			goto err_out;
+			goto err_unreg;
 		}
 
 		blake2b(NULL, 0, cid, strlen(cid), dev_id, UFS_RPMB_ID_LEN);
@@ -227,29 +234,30 @@ int ufs_rpmb_probe(struct ufs_hba *hba)
 
 		/* Register RPMB device */
 		rdev = rpmb_dev_register(&ufs_rpmb->dev, &descr);
+		kfree(cid);
 		if (IS_ERR(rdev)) {
 			dev_err(hba->dev, "Failed to register UFS RPMB device.\n");
-			device_unregister(&ufs_rpmb->dev);
 			ret = PTR_ERR(rdev);
-			goto err_out;
+			goto err_unreg;
 		}
 
-		kfree(cid);
-		cid = NULL;
-
 		ufs_rpmb->rdev = rdev;
-		ufs_rpmb->region_id = region;
-
 		list_add_tail(&ufs_rpmb->node, &hba->rpmbs);
 
 		dev_info(hba->dev, "UFS RPMB region %d registered (capacity=%u)\n", region, cap);
 	}
 
 	return 0;
+
+err_unreg:
+	device_unregister(&ufs_rpmb->dev);
+	goto err_out;
+err_put:
+	put_device(&ufs_rpmb->dev);
 err_out:
-	kfree(cid);
 	list_for_each_entry_safe(it, tmp, &hba->rpmbs, node) {
 		list_del(&it->node);
+		rpmb_dev_unregister(it->rdev);
 		device_unregister(&it->dev);
 	}
 
@@ -269,6 +277,7 @@ void ufs_rpmb_remove(struct ufs_hba *hba)
 		dev_info(hba->dev, "Removing UFS RPMB region %d\n", ufs_rpmb->region_id);
 		/* Remove from list first */
 		list_del(&ufs_rpmb->node);
+		rpmb_dev_unregister(ufs_rpmb->rdev);
 		/* Unregister device */
 		device_unregister(&ufs_rpmb->dev);
 	}
Signature · 1 line
2.56.0.567.gdb80ae8cd5-goog
#4 lore raw

PATCH v6 3/3 scsi: ufs: rpmb: Drop the unregistered ufs_rpmb bus

In reply to Stanley Jhu

ufs_rpmb_probe() assigns ufs_rpmb_bus_type to dev.bus without ever calling bus_register(). Since commit 36f35b8df697 ("driver core: reject devices with unregistered buses"), device_register() rejects unregistered buses with -EINVAL and UFS RPMB region registration fails.

ufs_rpmb_bus_type defines no callbacks and has no drivers, and consumers look up regions through /sys/class/rpmb. Drop ufs_rpmb_bus_type and leave dev.bus unset.

Fixes b06b8c421485 ("scsi: ufs: core: Add OP-TEE based RPMB driver for UFS devices")
---
 drivers/ufs/core/ufs-rpmb.c | 5 -----
 1 file changed, 5 deletions(-)
drivers/ufs/core/ufs-rpmb.c +0 -5
index 11d2d8e789d0..6803f3e62d78 100644
@@ -28,10 +28,6 @@
 #define UFS_RPMB_SEC_PROTOCOL		0xEC	/* JEDEC UFS application */
 #define UFS_RPMB_SEC_PROTOCOL_ID	0x01	/* JEDEC UFS RPMB protocol ID, CDB byte3 */
 
-static const struct bus_type ufs_rpmb_bus_type = {
-	.name = "ufs_rpmb",
-};
-
 /* UFS RPMB device structure */
 struct ufs_rpmb_dev {
 	u8 region_id;
@@ -206,7 +202,6 @@ int ufs_rpmb_probe(struct ufs_hba *hba)
 		ufs_rpmb->sdev = sdev;
 		ufs_rpmb->region_id = region;
 		ufs_rpmb->dev.parent = &sdev->sdev_gendev;
-		ufs_rpmb->dev.bus = &ufs_rpmb_bus_type;
 		ufs_rpmb->dev.release = ufs_rpmb_device_release;
 		dev_set_name(&ufs_rpmb->dev, "ufs_rpmb%d", region);
 
Signature · 1 line
2.56.0.567.gdb80ae8cd5-goog
Tested-by on v6 3/3
#5 lore raw

Re: [PATCH v6 3/3] scsi: ufs: rpmb: Drop the unregistered ufs_rpmb bus

Quoted 10 lines · Stanley Jhu

ufs_rpmb_probe() assigns ufs_rpmb_bus_type to dev.bus without ever calling bus_register(). Since commit 36f35b8df697 ("driver core: reject devices with unregistered buses"), device_register() rejects unregistered buses with -EINVAL and UFS RPMB region registration fails.

ufs_rpmb_bus_type defines no callbacks and has no drivers, and consumers look up regions through /sys/class/rpmb. Drop ufs_rpmb_bus_type and leave dev.bus unset.

Quoted 5 lines
Fixes b06b8c421485 ("scsi: ufs: core: Add OP-TEE based RPMB driver for UFS devices")

---

Thanks a lot, Stanley! Works like a charm on my Rockchip RK3576 with two UFS 2.2 parts, FORESEE C2G071 and BIWIN BWU2A0526B128G.

Best regards,
Alexey

maintainer Applied
#6 lore raw
In reply to Stanley Jhu
Quoted 9 lines · Stanley Jhu

Jens, Martin,

Patch 1/3 serializes requests and provider unregistration in drivers/misc/rpmb-core.c. Patches 2/3 and 3/3 fix UFS RPMB device lifetime and drop the unregistered bus in drivers/ufs/core/ufs-rpmb.c so regions register again after commit 36f35b8df697 ("driver core: reject devices with unregistered buses").

[...]

Applied to 7.4/scsi-fixes, thanks!

[1/3] rpmb: core: Guard frame requests and teardown with mutex

      https://git.kernel.org/mkp/scsi/c/e0696bf65006
[2/3] scsi: ufs: rpmb: Decouple device lifecycle from devres to avoid UAF
      https://git.kernel.org/mkp/scsi/c/7bc9b854cd68
[3/3] scsi: ufs: rpmb: Drop the unregistered ufs_rpmb bus
      https://git.kernel.org/mkp/scsi/c/b1de2bd5a4d6
Signature · 1 line
Martin K. Petersen