nvme: introduce generic per-namespace chardev
authorMinwoo Im <minwoo.im.dev@gmail.com>
Wed, 21 Apr 2021 07:45:04 +0000 (16:45 +0900)
committerChristoph Hellwig <hch@lst.de>
Thu, 22 Apr 2021 05:25:17 +0000 (07:25 +0200)
Userspace has not been allowed to I/O to device that's failed to
be initialized.  This patch introduces generic per-namespace character
device to allow userspace to I/O regardless the block device is there or
not.

The chardev naming convention will similar to the existing blkdev naming,
using a ng prefix instead of nvme, i.e.

- /dev/ngXnY

It also supports multipath which means it will not expose chardev for the
hidden namespace blkdevs (e.g., nvmeXcYnZ).  If /dev/ngXnY is created for
a ns_head, then I/O request will be routed to a specific controller
selected by the iopolicy of the subsystem.

Signed-off-by: Minwoo Im <minwoo.im.dev@gmail.com>
Signed-off-by: Javier González <javier.gonz@samsung.com>
Reviewed-by: Keith Busch <kbusch@kernel.org>
Tested-by: Kanchan Joshi <joshi.k@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
drivers/nvme/host/core.c
drivers/nvme/host/ioctl.c
drivers/nvme/host/multipath.c
drivers/nvme/host/nvme.h

index b905f91f14eba6a9ea3c7b7f7a104cec66c99259..2f45e8fcdd7cbd8c882bee3938ed65fc2839046c 100644 (file)
@@ -89,6 +89,10 @@ static dev_t nvme_ctrl_base_chr_devt;
 static struct class *nvme_class;
 static struct class *nvme_subsys_class;
 
+static DEFINE_IDA(nvme_ns_chr_minor_ida);
+static dev_t nvme_ns_chr_devt;
+static struct class *nvme_ns_chr_class;
+
 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
                                           unsigned nsid);
@@ -3429,6 +3433,66 @@ static int __nvme_check_ids(struct nvme_subsystem *subsys,
        return 0;
 }
 
+void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
+{
+       cdev_device_del(cdev, cdev_device);
+       ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(cdev_device->devt));
+}
+
+int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
+               const struct file_operations *fops, struct module *owner)
+{
+       int minor, ret;
+
+       minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
+       if (minor < 0)
+               return minor;
+       cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
+       cdev_device->class = nvme_ns_chr_class;
+       device_initialize(cdev_device);
+       cdev_init(cdev, fops);
+       cdev->owner = owner;
+       ret = cdev_device_add(cdev, cdev_device);
+       if (ret)
+               ida_simple_remove(&nvme_ns_chr_minor_ida, minor);
+       return ret;
+}
+
+static int nvme_ns_chr_open(struct inode *inode, struct file *file)
+{
+       return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
+}
+
+static int nvme_ns_chr_release(struct inode *inode, struct file *file)
+{
+       nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
+       return 0;
+}
+
+static const struct file_operations nvme_ns_chr_fops = {
+       .owner          = THIS_MODULE,
+       .open           = nvme_ns_chr_open,
+       .release        = nvme_ns_chr_release,
+       .unlocked_ioctl = nvme_ns_chr_ioctl,
+       .compat_ioctl   = compat_ptr_ioctl,
+};
+
+static int nvme_add_ns_cdev(struct nvme_ns *ns)
+{
+       int ret;
+
+       ns->cdev_device.parent = ns->ctrl->device;
+       ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
+                          ns->ctrl->instance, ns->head->instance);
+       if (ret)
+               return ret;
+       ret = nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
+                           ns->ctrl->ops->module);
+       if (ret)
+               kfree_const(ns->cdev_device.kobj.name);
+       return ret;
+}
+
 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
                unsigned nsid, struct nvme_ns_ids *ids)
 {
@@ -3630,6 +3694,8 @@ static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
        nvme_get_ctrl(ctrl);
 
        device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
+       if (!nvme_ns_head_multipath(ns->head))
+               nvme_add_ns_cdev(ns);
 
        nvme_mpath_add_disk(ns, id);
        nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
@@ -3674,6 +3740,8 @@ static void nvme_ns_remove(struct nvme_ns *ns)
        synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
 
        if (ns->disk->flags & GENHD_FL_UP) {
+               if (!nvme_ns_head_multipath(ns->head))
+                       nvme_cdev_del(&ns->cdev, &ns->cdev_device);
                del_gendisk(ns->disk);
                blk_cleanup_queue(ns->queue);
                if (blk_get_integrity(ns->disk))
@@ -4464,8 +4532,24 @@ static int __init nvme_core_init(void)
                result = PTR_ERR(nvme_subsys_class);
                goto destroy_class;
        }
+
+       result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
+                                    "nvme-generic");
+       if (result < 0)
+               goto destroy_subsys_class;
+
+       nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
+       if (IS_ERR(nvme_ns_chr_class)) {
+               result = PTR_ERR(nvme_ns_chr_class);
+               goto unregister_generic_ns;
+       }
+
        return 0;
 
+unregister_generic_ns:
+       unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
+destroy_subsys_class:
+       class_destroy(nvme_subsys_class);
 destroy_class:
        class_destroy(nvme_class);
 unregister_chrdev:
@@ -4482,12 +4566,15 @@ out:
 
 static void __exit nvme_core_exit(void)
 {
+       class_destroy(nvme_ns_chr_class);
        class_destroy(nvme_subsys_class);
        class_destroy(nvme_class);
+       unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
        unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
        destroy_workqueue(nvme_delete_wq);
        destroy_workqueue(nvme_reset_wq);
        destroy_workqueue(nvme_wq);
+       ida_destroy(&nvme_ns_chr_minor_ida);
        ida_destroy(&nvme_instance_ida);
 }
 
index 8e05d65c9e9340b92799551df3d973ba448f2b8f..502f8e4a2a1f00dd9fecbb8dc8bc0263b44affcf 100644 (file)
@@ -346,15 +346,27 @@ static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned int cmd,
        }
 }
 
+static int __nvme_ioctl(struct nvme_ns *ns, unsigned int cmd, void __user *arg)
+{
+       if (is_ctrl_ioctl(cmd))
+               return nvme_ctrl_ioctl(ns->ctrl, cmd, arg);
+       return nvme_ns_ioctl(ns, cmd, arg);
+}
+
 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
                unsigned int cmd, unsigned long arg)
 {
        struct nvme_ns *ns = bdev->bd_disk->private_data;
-       void __user *argp = (void __user *)arg;
 
-       if (is_ctrl_ioctl(cmd))
-               return nvme_ctrl_ioctl(ns->ctrl, cmd, argp);
-       return nvme_ns_ioctl(ns, cmd, argp);
+       return __nvme_ioctl(ns, cmd, (void __user *)arg);
+}
+
+long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       struct nvme_ns *ns =
+               container_of(file_inode(file)->i_cdev, struct nvme_ns, cdev);
+
+       return __nvme_ioctl(ns, cmd, (void __user *)arg);
 }
 
 #ifdef CONFIG_NVME_MULTIPATH
@@ -388,10 +400,24 @@ int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
                unsigned int cmd, unsigned long arg)
 {
        struct nvme_ns_head *head = bdev->bd_disk->private_data;
+       void __user *argp = (void __user *)arg;
+
+       if (is_ctrl_ioctl(cmd))
+               return nvme_ns_head_ctrl_ioctl(head, cmd, argp);
+       return nvme_ns_head_ns_ioctl(head, cmd, argp);
+}
+
+long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
+               unsigned long arg)
+{
+       struct cdev *cdev = file_inode(file)->i_cdev;
+       struct nvme_ns_head *head =
+               container_of(cdev, struct nvme_ns_head, cdev);
+       void __user *argp = (void __user *)arg;
 
        if (is_ctrl_ioctl(cmd))
-               return nvme_ns_head_ctrl_ioctl(head, cmd, (void __user *)arg);
-       return nvme_ns_head_ns_ioctl(head, cmd, (void __user *)arg);
+               return nvme_ns_head_ctrl_ioctl(head, cmd, argp);
+       return nvme_ns_head_ns_ioctl(head, cmd, argp);
 }
 #endif /* CONFIG_NVME_MULTIPATH */
 
index 68918ea1d3d0989a3fbee76110758e02f463eca3..0d0de3433f3776a6ae8d7f153877b8746ade67cb 100644 (file)
@@ -357,6 +357,48 @@ const struct block_device_operations nvme_ns_head_ops = {
        .pr_ops         = &nvme_pr_ops,
 };
 
+static inline struct nvme_ns_head *cdev_to_ns_head(struct cdev *cdev)
+{
+       return container_of(cdev, struct nvme_ns_head, cdev);
+}
+
+static int nvme_ns_head_chr_open(struct inode *inode, struct file *file)
+{
+       if (!nvme_tryget_ns_head(cdev_to_ns_head(inode->i_cdev)))
+               return -ENXIO;
+       return 0;
+}
+
+static int nvme_ns_head_chr_release(struct inode *inode, struct file *file)
+{
+       nvme_put_ns_head(cdev_to_ns_head(inode->i_cdev));
+       return 0;
+}
+
+static const struct file_operations nvme_ns_head_chr_fops = {
+       .owner          = THIS_MODULE,
+       .open           = nvme_ns_head_chr_open,
+       .release        = nvme_ns_head_chr_release,
+       .unlocked_ioctl = nvme_ns_head_chr_ioctl,
+       .compat_ioctl   = compat_ptr_ioctl,
+};
+
+static int nvme_add_ns_head_cdev(struct nvme_ns_head *head)
+{
+       int ret;
+
+       head->cdev_device.parent = &head->subsys->dev;
+       ret = dev_set_name(&head->cdev_device, "ng%dn%d",
+                          head->subsys->instance, head->instance);
+       if (ret)
+               return ret;
+       ret = nvme_cdev_add(&head->cdev, &head->cdev_device,
+                           &nvme_ns_head_chr_fops, THIS_MODULE);
+       if (ret)
+               kfree_const(head->cdev_device.kobj.name);
+       return ret;
+}
+
 static void nvme_requeue_work(struct work_struct *work)
 {
        struct nvme_ns_head *head =
@@ -435,9 +477,11 @@ static void nvme_mpath_set_live(struct nvme_ns *ns)
        if (!head->disk)
                return;
 
-       if (!test_and_set_bit(NVME_NSHEAD_DISK_LIVE, &head->flags))
+       if (!test_and_set_bit(NVME_NSHEAD_DISK_LIVE, &head->flags)) {
                device_add_disk(&head->subsys->dev, head->disk,
                                nvme_ns_id_attr_groups);
+               nvme_add_ns_head_cdev(head);
+       }
 
        mutex_lock(&head->lock);
        if (nvme_path_is_optimized(ns)) {
@@ -714,8 +758,10 @@ void nvme_mpath_remove_disk(struct nvme_ns_head *head)
 {
        if (!head->disk)
                return;
-       if (head->disk->flags & GENHD_FL_UP)
+       if (head->disk->flags & GENHD_FL_UP) {
+               nvme_cdev_del(&head->cdev, &head->cdev_device);
                del_gendisk(head->disk);
+       }
        blk_set_queue_dying(head->disk->queue);
        /* make sure all pending bios are cleaned up */
        kblockd_schedule_work(&head->requeue_work);
@@ -785,4 +831,3 @@ void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
        kfree(ctrl->ana_log_buf);
        ctrl->ana_log_buf = NULL;
 }
-
index 49276186d5bd6eb2269a653ae78974423646392b..773dde5b231dab0e0580d71141a6535f6a1f5e23 100644 (file)
@@ -412,6 +412,10 @@ struct nvme_ns_head {
        bool                    shared;
        int                     instance;
        struct nvme_effects_log *effects;
+
+       struct cdev             cdev;
+       struct device           cdev_device;
+
        struct gendisk          *disk;
 #ifdef CONFIG_NVME_MULTIPATH
        struct bio_list         requeue_list;
@@ -464,6 +468,9 @@ struct nvme_ns {
 #define NVME_NS_ANA_PENDING    2
 #define NVME_NS_FORCE_RO       3
 
+       struct cdev             cdev;
+       struct device           cdev_device;
+
        struct nvme_fault_inject fault_inject;
 
 };
@@ -658,10 +665,16 @@ void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx);
 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
 void nvme_put_ns_head(struct nvme_ns_head *head);
 struct nvme_ctrl *nvme_find_get_live_ctrl(struct nvme_subsystem *subsys);
+int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
+               const struct file_operations *fops, struct module *owner);
+void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
                unsigned int cmd, unsigned long arg);
+long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
                unsigned int cmd, unsigned long arg);
+long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
+               unsigned long arg);
 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
                unsigned long arg);
 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);