cxl/mem: Do not rely on device_add() side effects for dev_set_name() failures
authorDan Williams <dan.j.williams@intel.com>
Thu, 1 Apr 2021 14:33:25 +0000 (07:33 -0700)
committerDan Williams <dan.j.williams@intel.com>
Tue, 6 Apr 2021 18:41:30 +0000 (11:41 -0700)
While device_add() will happen to catch dev_set_name() failures it is a
broken pattern to follow given that the core may try to fall back to a
different name.

Add explicit checking for dev_set_name() failures to be cleaned up by
put_device(). Skip cdev_device_add() and proceed directly to
put_device() if the name set fails.

This type of bug is easier to see if 'alloc' is split from 'add'
operations that require put_device() on failure. So cxl_memdev_alloc()
is split out as a result.

Fixes: b39cb1052a5c ("cxl/mem: Register CXL memX devices")
Reported-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Link: https://lore.kernel.org/r/161728760514.2474381.1163928273337158134.stgit@dwillia2-desk3.amr.corp.intel.com
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
drivers/cxl/mem.c

index 438f086..da93b63 100644 (file)
@@ -1180,7 +1180,7 @@ static void cxl_memdev_unregister(void *_cxlmd)
        put_device(dev);
 }
 
-static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
+static struct cxl_memdev *cxl_memdev_alloc(struct cxl_mem *cxlm)
 {
        struct pci_dev *pdev = cxlm->pdev;
        struct cxl_memdev *cxlmd;
@@ -1190,11 +1190,11 @@ static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
 
        cxlmd = kzalloc(sizeof(*cxlmd), GFP_KERNEL);
        if (!cxlmd)
-               return -ENOMEM;
+               return ERR_PTR(-ENOMEM);
 
        rc = ida_alloc_range(&cxl_memdev_ida, 0, CXL_MEM_MAX_DEVS, GFP_KERNEL);
        if (rc < 0)
-               goto err_id;
+               goto err;
        cxlmd->id = rc;
 
        dev = &cxlmd->dev;
@@ -1203,10 +1203,31 @@ static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
        dev->bus = &cxl_bus_type;
        dev->devt = MKDEV(cxl_mem_major, cxlmd->id);
        dev->type = &cxl_memdev_type;
-       dev_set_name(dev, "mem%d", cxlmd->id);
 
        cdev = &cxlmd->cdev;
        cdev_init(cdev, &cxl_memdev_fops);
+       return cxlmd;
+
+err:
+       kfree(cxlmd);
+       return ERR_PTR(rc);
+}
+
+static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
+{
+       struct cxl_memdev *cxlmd;
+       struct device *dev;
+       struct cdev *cdev;
+       int rc;
+
+       cxlmd = cxl_memdev_alloc(cxlm);
+       if (IS_ERR(cxlmd))
+               return PTR_ERR(cxlmd);
+
+       dev = &cxlmd->dev;
+       rc = dev_set_name(dev, "mem%d", cxlmd->id);
+       if (rc)
+               goto err;
 
        /*
         * Activate ioctl operations, no cxl_memdev_rwsem manipulation
@@ -1214,23 +1235,21 @@ static int cxl_mem_add_memdev(struct cxl_mem *cxlm)
         */
        cxlmd->cxlm = cxlm;
 
+       cdev = &cxlmd->cdev;
        rc = cdev_device_add(cdev, dev);
        if (rc)
-               goto err_add;
+               goto err;
 
        return devm_add_action_or_reset(dev->parent, cxl_memdev_unregister,
                                        cxlmd);
 
-err_add:
+err:
        /*
         * The cdev was briefly live, shutdown any ioctl operations that
         * saw that state.
         */
        cxl_memdev_shutdown(cxlmd);
-       ida_free(&cxl_memdev_ida, cxlmd->id);
-err_id:
-       kfree(cxlmd);
-
+       put_device(dev);
        return rc;
 }