nullb: add interface to power on disk
authorShaohua Li <shli@fb.com>
Mon, 14 Aug 2017 22:04:54 +0000 (15:04 -0700)
committerJens Axboe <axboe@kernel.dk>
Wed, 23 Aug 2017 14:54:03 +0000 (08:54 -0600)
The device created in nullb configfs interface isn't power on by
default. After user configures the device, user can do 'echo 1 >
xxx/nullb/device_name/power' to power on the device, which will create a
disk. the xxx/nullb/device_name/index is the disk index, so if the index
is 2, the new created disk should be named as /dev/nullb2. Note, the
'index' is only valid after disk is power on.

'echo 0 > xxx/nullb/device_name/power' will remove the disk. Note, this
doesn't remove the device. To remove the device, user should do 'rmdir
xxx/nullb/device_name'. Removing the device will remove the disk too.

Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/block/null_blk.c

index c782492c00997e809760982054566fd0f9f729a8..cf14c46d34625c87af58f09648308ccbb8a5004d 100644 (file)
@@ -59,9 +59,11 @@ struct nullb_device {
        unsigned int blocksize; /* block size */
        unsigned int irqmode; /* IRQ completion handler */
        unsigned int hw_queue_depth; /* queue depth */
+       unsigned int index; /* index of the disk, only valid with a disk */
        bool use_lightnvm; /* register as a LightNVM device */
        bool blocking; /* blocking blk-mq device */
        bool use_per_node_hctx; /* use per-node allocation for hardware context */
+       bool power; /* power on/off the device */
 };
 
 struct nullb {
@@ -193,6 +195,8 @@ MODULE_PARM_DESC(use_per_node_hctx, "Use per-node allocation for hardware contex
 
 static struct nullb_device *null_alloc_dev(void);
 static void null_free_dev(struct nullb_device *dev);
+static void null_del_dev(struct nullb *nullb);
+static int null_add_dev(struct nullb_device *dev);
 
 static inline struct nullb_device *to_nullb_device(struct config_item *item)
 {
@@ -284,10 +288,50 @@ NULLB_DEVICE_ATTR(queue_mode, uint);
 NULLB_DEVICE_ATTR(blocksize, uint);
 NULLB_DEVICE_ATTR(irqmode, uint);
 NULLB_DEVICE_ATTR(hw_queue_depth, uint);
+NULLB_DEVICE_ATTR(index, uint);
 NULLB_DEVICE_ATTR(use_lightnvm, bool);
 NULLB_DEVICE_ATTR(blocking, bool);
 NULLB_DEVICE_ATTR(use_per_node_hctx, bool);
 
+static ssize_t nullb_device_power_show(struct config_item *item, char *page)
+{
+       return nullb_device_bool_attr_show(to_nullb_device(item)->power, page);
+}
+
+static ssize_t nullb_device_power_store(struct config_item *item,
+                                    const char *page, size_t count)
+{
+       struct nullb_device *dev = to_nullb_device(item);
+       bool newp = false;
+       ssize_t ret;
+
+       ret = nullb_device_bool_attr_store(&newp, page, count);
+       if (ret < 0)
+               return ret;
+
+       if (!dev->power && newp) {
+               if (test_and_set_bit(NULLB_DEV_FL_UP, &dev->flags))
+                       return count;
+               if (null_add_dev(dev)) {
+                       clear_bit(NULLB_DEV_FL_UP, &dev->flags);
+                       return -ENOMEM;
+               }
+
+               set_bit(NULLB_DEV_FL_CONFIGURED, &dev->flags);
+               dev->power = newp;
+       } else if (to_nullb_device(item)->power && !newp) {
+               mutex_lock(&lock);
+               dev->power = newp;
+               null_del_dev(dev->nullb);
+               mutex_unlock(&lock);
+               clear_bit(NULLB_DEV_FL_UP, &dev->flags);
+       }
+
+       return count;
+}
+
+CONFIGFS_ATTR(nullb_device_, power);
+
 static struct configfs_attribute *nullb_device_attrs[] = {
        &nullb_device_attr_size,
        &nullb_device_attr_completion_nsec,
@@ -297,9 +341,11 @@ static struct configfs_attribute *nullb_device_attrs[] = {
        &nullb_device_attr_blocksize,
        &nullb_device_attr_irqmode,
        &nullb_device_attr_hw_queue_depth,
+       &nullb_device_attr_index,
        &nullb_device_attr_use_lightnvm,
        &nullb_device_attr_blocking,
        &nullb_device_attr_use_per_node_hctx,
+       &nullb_device_attr_power,
        NULL,
 };
 
@@ -335,6 +381,15 @@ config_item *nullb_group_make_item(struct config_group *group, const char *name)
 static void
 nullb_group_drop_item(struct config_group *group, struct config_item *item)
 {
+       struct nullb_device *dev = to_nullb_device(item);
+
+       if (test_and_clear_bit(NULLB_DEV_FL_UP, &dev->flags)) {
+               mutex_lock(&lock);
+               dev->power = false;
+               null_del_dev(dev->nullb);
+               mutex_unlock(&lock);
+       }
+
        config_item_put(item);
 }
 
@@ -973,11 +1028,35 @@ static int null_init_tag_set(struct nullb *nullb, struct blk_mq_tag_set *set)
        return blk_mq_alloc_tag_set(set);
 }
 
+static void null_validate_conf(struct nullb_device *dev)
+{
+       dev->blocksize = round_down(dev->blocksize, 512);
+       dev->blocksize = clamp_t(unsigned int, dev->blocksize, 512, 4096);
+       if (dev->use_lightnvm && dev->blocksize != 4096)
+               dev->blocksize = 4096;
+
+       if (dev->use_lightnvm && dev->queue_mode != NULL_Q_MQ)
+               dev->queue_mode = NULL_Q_MQ;
+
+       if (dev->queue_mode == NULL_Q_MQ && dev->use_per_node_hctx) {
+               if (dev->submit_queues != nr_online_nodes)
+                       dev->submit_queues = nr_online_nodes;
+       } else if (dev->submit_queues > nr_cpu_ids)
+               dev->submit_queues = nr_cpu_ids;
+       else if (dev->submit_queues == 0)
+               dev->submit_queues = 1;
+
+       dev->queue_mode = min_t(unsigned int, dev->queue_mode, NULL_Q_MQ);
+       dev->irqmode = min_t(unsigned int, dev->irqmode, NULL_IRQ_TIMER);
+}
+
 static int null_add_dev(struct nullb_device *dev)
 {
        struct nullb *nullb;
        int rv;
 
+       null_validate_conf(dev);
+
        nullb = kzalloc_node(sizeof(*nullb), GFP_KERNEL, dev->home_node);
        if (!nullb) {
                rv = -ENOMEM;
@@ -1040,6 +1119,7 @@ static int null_add_dev(struct nullb_device *dev)
 
        mutex_lock(&lock);
        nullb->index = nullb_indexes++;
+       dev->index = nullb->index;
        mutex_unlock(&lock);
 
        blk_queue_logical_block_size(nullb->q, dev->blocksize);