Merge tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "Seems bigger than usual, a number of things were posted near/during
  the merg window:

   - Fix some compilation regressions related to the new DMABUF code

   - Close a race with ib_register_device() vs netdev events that causes
     GID table corruption

   - Compilation warnings with some compilers in bng_re

   - Correct error unwind in bng_re and the umem pinned dmabuf

   - Avoid NULL pointer crash in ionic during query_port()

   - Check the size for uAPI validation checks in EFA

   - Several system call stack leaks in drivers found with AI

   - Fix the new restricted_node_type so it works with wildcard listens
     too"

* tag 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/rdma/rdma:
  RDMA/uverbs: Import DMA-BUF module in uverbs_std_types_dmabuf file
  RDMA/umem: Fix double dma_buf_unpin in failure path
  RDMA/core: Check id_priv->restricted_node_type in cma_listen_on_dev()
  RDMA/ionic: Fix kernel stack leak in ionic_create_cq()
  RDMA/irdma: Fix kernel stack leak in irdma_create_user_ah()
  IB/mthca: Add missed mthca_unmap_user_db() for mthca_create_srq()
  RDMA/efa: Fix typo in efa_alloc_mr()
  RDMA/ionic: Fix potential NULL pointer dereference in ionic_query_port
  RDMA/bng_re: Unwind bng_re_dev_init properly
  RDMA/bng_re: Remove unnessary validity checks
  RDMA/core: Fix stale RoCE GIDs during netdev events at registration
  RDMA/uverbs: select CONFIG_DMA_SHARED_BUFFER

Author: torvalds

drivers/infiniband/Kconfig

@@ -6,6 +6,7 @@ menuconfig INFINIBAND
 	depends on INET
 	depends on m || IPV6 != m
 	depends on !ALPHA
+	select DMA_SHARED_BUFFER
 	select IRQ_POLL
 	select DIMLIB
 	help

drivers/infiniband/core/cache.c

@@ -926,6 +926,13 @@ static int gid_table_setup_one(struct ib_device *ib_dev)
 	if (err)
 		return err;
 
+	/*
+	 * Mark the device as ready for GID cache updates. This allows netdev
+	 * event handlers to update the GID cache even before the device is
+	 * fully registered.
+	 */
+	ib_device_enable_gid_updates(ib_dev);
+
 	rdma_roce_rescan_device(ib_dev);
 
 	return err;
@@ -1637,6 +1644,12 @@ void ib_cache_release_one(struct ib_device *device)
 
 void ib_cache_cleanup_one(struct ib_device *device)
 {
+	/*
+	 * Clear the GID updates mark first to prevent event handlers from
+	 * accessing the device while it's being torn down.
+	 */
+	ib_device_disable_gid_updates(device);
+
 	/* The cleanup function waits for all in-progress workqueue
 	 * elements and cleans up the GID cache. This function should be
 	 * called after the device was removed from the devices list and

drivers/infiniband/core/cma.c

@@ -2729,13 +2729,17 @@ static int cma_listen_on_dev(struct rdma_id_private *id_priv,
 	*to_destroy = NULL;
 	if (cma_family(id_priv) == AF_IB && !rdma_cap_ib_cm(cma_dev->device, 1))
 		return 0;
+	if (id_priv->restricted_node_type != RDMA_NODE_UNSPECIFIED &&
+	    id_priv->restricted_node_type != cma_dev->device->node_type)
+		return 0;
 
 	dev_id_priv =
 		__rdma_create_id(net, cma_listen_handler, id_priv,
 				 id_priv->id.ps, id_priv->id.qp_type, id_priv);
 	if (IS_ERR(dev_id_priv))
 		return PTR_ERR(dev_id_priv);
 
+	dev_id_priv->restricted_node_type = id_priv->restricted_node_type;
 	dev_id_priv->state = RDMA_CM_ADDR_BOUND;
 	memcpy(cma_src_addr(dev_id_priv), cma_src_addr(id_priv),
 	       rdma_addr_size(cma_src_addr(id_priv)));
@@ -4194,7 +4198,7 @@ int rdma_restrict_node_type(struct rdma_cm_id *id, u8 node_type)
 	}
 
 	mutex_lock(&lock);
-	if (id_priv->cma_dev)
+	if (READ_ONCE(id_priv->state) != RDMA_CM_IDLE)
 		ret = -EALREADY;
 	else
 		id_priv->restricted_node_type = node_type;

drivers/infiniband/core/core_priv.h

@@ -100,6 +100,9 @@ void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
 			      roce_netdev_callback cb,
 			      void *cookie);
 
+void ib_device_enable_gid_updates(struct ib_device *device);
+void ib_device_disable_gid_updates(struct ib_device *device);
+
 typedef int (*nldev_callback)(struct ib_device *device,
 			      struct sk_buff *skb,
 			      struct netlink_callback *cb,

drivers/infiniband/core/device.c

@@ -93,6 +93,7 @@ static struct workqueue_struct *ib_unreg_wq;
 static DEFINE_XARRAY_FLAGS(devices, XA_FLAGS_ALLOC);
 static DECLARE_RWSEM(devices_rwsem);
 #define DEVICE_REGISTERED XA_MARK_1
+#define DEVICE_GID_UPDATES XA_MARK_2
 
 static u32 highest_client_id;
 #define CLIENT_REGISTERED XA_MARK_1
@@ -2412,11 +2413,42 @@ void ib_enum_all_roce_netdevs(roce_netdev_filter filter,
 	unsigned long index;
 
 	down_read(&devices_rwsem);
-	xa_for_each_marked (&devices, index, dev, DEVICE_REGISTERED)
+	xa_for_each_marked(&devices, index, dev, DEVICE_GID_UPDATES)
 		ib_enum_roce_netdev(dev, filter, filter_cookie, cb, cookie);
 	up_read(&devices_rwsem);
 }
 
+/**
+ * ib_device_enable_gid_updates - Mark device as ready for GID cache updates
+ * @device: Device to mark
+ *
+ * Called after GID table is allocated and initialized. After this mark is set,
+ * netdevice event handlers can update the device's GID cache. This allows
+ * events that arrive during device registration to be processed, avoiding
+ * stale GID entries when netdev properties change during the device
+ * registration process.
+ */
+void ib_device_enable_gid_updates(struct ib_device *device)
+{
+	down_write(&devices_rwsem);
+	xa_set_mark(&devices, device->index, DEVICE_GID_UPDATES);
+	up_write(&devices_rwsem);
+}
+
+/**
+ * ib_device_disable_gid_updates - Clear the GID updates mark
+ * @device: Device to unmark
+ *
+ * Called before GID table cleanup to prevent event handlers from accessing
+ * the device while it's being torn down.
+ */
+void ib_device_disable_gid_updates(struct ib_device *device)
+{
+	down_write(&devices_rwsem);
+	xa_clear_mark(&devices, device->index, DEVICE_GID_UPDATES);
+	up_write(&devices_rwsem);
+}
+
 /*
  * ib_enum_all_devs - enumerate all ib_devices
  * @cb: Callback to call for each found ib_device

drivers/infiniband/core/umem_dmabuf.c

@@ -218,13 +218,11 @@ ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device,
 
 	err = ib_umem_dmabuf_map_pages(umem_dmabuf);
 	if (err)
-		goto err_unpin;
+		goto err_release;
 	dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
 
 	return umem_dmabuf;
 
-err_unpin:
-	dma_buf_unpin(umem_dmabuf->attach);
 err_release:
 	dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv);
 	ib_umem_release(&umem_dmabuf->umem);

drivers/infiniband/core/uverbs_std_types_dmabuf.c

@@ -10,6 +10,8 @@
 #include "rdma_core.h"
 #include "uverbs.h"
 
+MODULE_IMPORT_NS("DMA_BUF");
+
 static int uverbs_dmabuf_attach(struct dma_buf *dmabuf,
 				struct dma_buf_attachment *attachment)
 {

drivers/infiniband/hw/bng_re/bng_dev.c

@@ -54,9 +54,6 @@ static void bng_re_destroy_chip_ctx(struct bng_re_dev *rdev)
 {
 	struct bng_re_chip_ctx *chip_ctx;
 
-	if (!rdev->chip_ctx)
-		return;
-
 	kfree(rdev->dev_attr);
 	rdev->dev_attr = NULL;
 
@@ -124,12 +121,6 @@ static int bng_re_net_ring_free(struct bng_re_dev *rdev,
 	struct bnge_fw_msg fw_msg = {};
 	int rc = -EINVAL;
 
-	if (!rdev)
-		return rc;
-
-	if (!aux_dev)
-		return rc;
-
 	bng_re_init_hwrm_hdr((void *)&req, HWRM_RING_FREE);
 	req.ring_type = type;
 	req.ring_id = cpu_to_le16(fw_ring_id);
@@ -150,10 +141,7 @@ static int bng_re_net_ring_alloc(struct bng_re_dev *rdev,
 	struct hwrm_ring_alloc_input req = {};
 	struct hwrm_ring_alloc_output resp;
 	struct bnge_fw_msg fw_msg = {};
-	int rc = -EINVAL;
-
-	if (!aux_dev)
-		return rc;
+	int rc;
 
 	bng_re_init_hwrm_hdr((void *)&req, HWRM_RING_ALLOC);
 	req.enables = 0;
@@ -184,10 +172,7 @@ static int bng_re_stats_ctx_free(struct bng_re_dev *rdev)
 	struct hwrm_stat_ctx_free_input req = {};
 	struct hwrm_stat_ctx_free_output resp = {};
 	struct bnge_fw_msg fw_msg = {};
-	int rc = -EINVAL;
-
-	if (!aux_dev)
-		return rc;
+	int rc;
 
 	bng_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_FREE);
 	req.stat_ctx_id = cpu_to_le32(rdev->stats_ctx.fw_id);
@@ -208,13 +193,10 @@ static int bng_re_stats_ctx_alloc(struct bng_re_dev *rdev)
 	struct hwrm_stat_ctx_alloc_output resp = {};
 	struct hwrm_stat_ctx_alloc_input req = {};
 	struct bnge_fw_msg fw_msg = {};
-	int rc = -EINVAL;
+	int rc;
 
 	stats->fw_id = BNGE_INVALID_STATS_CTX_ID;
 
-	if (!aux_dev)
-		return rc;
-
 	bng_re_init_hwrm_hdr((void *)&req, HWRM_STAT_CTX_ALLOC);
 	req.update_period_ms = cpu_to_le32(1000);
 	req.stats_dma_addr = cpu_to_le64(stats->dma_map);
@@ -303,7 +285,7 @@ static int bng_re_dev_init(struct bng_re_dev *rdev)
 	if (rc) {
 		ibdev_err(&rdev->ibdev,
 				"Failed to register with netedev: %#x\n", rc);
-		return -EINVAL;
+		goto reg_netdev_fail;
 	}
 
 	set_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
@@ -312,19 +294,16 @@ static int bng_re_dev_init(struct bng_re_dev *rdev)
 		ibdev_err(&rdev->ibdev,
 			  "RoCE requires minimum 2 MSI-X vectors, but only %d reserved\n",
 			  rdev->aux_dev->auxr_info->msix_requested);
-		bnge_unregister_dev(rdev->aux_dev);
-		clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
-		return -EINVAL;
+		rc = -EINVAL;
+		goto msix_ctx_fail;
 	}
 	ibdev_dbg(&rdev->ibdev, "Got %d MSI-X vectors\n",
 		  rdev->aux_dev->auxr_info->msix_requested);
 
 	rc = bng_re_setup_chip_ctx(rdev);
 	if (rc) {
-		bnge_unregister_dev(rdev->aux_dev);
-		clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
 		ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
-		return -EINVAL;
+		goto msix_ctx_fail;
 	}
 
 	bng_re_query_hwrm_version(rdev);
@@ -333,16 +312,14 @@ static int bng_re_dev_init(struct bng_re_dev *rdev)
 	if (rc) {
 		ibdev_err(&rdev->ibdev,
 			  "Failed to allocate RCFW Channel: %#x\n", rc);
-		goto fail;
+		goto alloc_fw_chl_fail;
 	}
 
 	/* Allocate nq record memory */
 	rdev->nqr = kzalloc_obj(*rdev->nqr);
 	if (!rdev->nqr) {
-		bng_re_destroy_chip_ctx(rdev);
-		bnge_unregister_dev(rdev->aux_dev);
-		clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
-		return -ENOMEM;
+		rc = -ENOMEM;
+		goto nq_alloc_fail;
 	}
 
 	rdev->nqr->num_msix = rdev->aux_dev->auxr_info->msix_requested;
@@ -411,9 +388,15 @@ static int bng_re_dev_init(struct bng_re_dev *rdev)
 free_ring:
 	bng_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
 free_rcfw:
+	kfree(rdev->nqr);
+nq_alloc_fail:
 	bng_re_free_rcfw_channel(&rdev->rcfw);
-fail:
-	bng_re_dev_uninit(rdev);
+alloc_fw_chl_fail:
+	bng_re_destroy_chip_ctx(rdev);
+msix_ctx_fail:
+	bnge_unregister_dev(rdev->aux_dev);
+	clear_bit(BNG_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
+reg_netdev_fail:
 	return rc;
 }
 
@@ -486,8 +469,7 @@ static void bng_re_remove(struct auxiliary_device *adev)
 
 	rdev = dev_info->rdev;
 
-	if (rdev)
-		bng_re_remove_device(rdev, adev);
+	bng_re_remove_device(rdev, adev);
 	kfree(dev_info);
 }
 

drivers/infiniband/hw/efa/efa_verbs.c

@@ -1661,7 +1661,7 @@ static struct efa_mr *efa_alloc_mr(struct ib_pd *ibpd, int access_flags,
 	struct efa_mr *mr;
 
 	if (udata && udata->inlen &&
-	    !ib_is_udata_cleared(udata, 0, sizeof(udata->inlen))) {
+	    !ib_is_udata_cleared(udata, 0, udata->inlen)) {
 		ibdev_dbg(&dev->ibdev,
 			  "Incompatible ABI params, udata not cleared\n");
 		return ERR_PTR(-EINVAL);

drivers/infiniband/hw/ionic/ionic_controlpath.c

@@ -1218,7 +1218,7 @@ int ionic_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 		rdma_udata_to_drv_context(udata, struct ionic_ctx, ibctx);
 	struct ionic_vcq *vcq = to_ionic_vcq(ibcq);
 	struct ionic_tbl_buf buf = {};
-	struct ionic_cq_resp resp;
+	struct ionic_cq_resp resp = {};
 	struct ionic_cq_req req;
 	int udma_idx = 0, rc;