implement in-kernel gendisk events handling

Currently, media presence polling for removeable block devices is done
from userland.  There are several issues with this.

* Polling is done by periodically opening the device.  For SCSI
  devices, the command sequence generated by such action involves a
  few different commands including TEST_UNIT_READY.  This behavior,
  while perfectly legal, is different from Windows which only issues
  single command, GET_EVENT_STATUS_NOTIFICATION.  Unfortunately, some
  ATAPI devices lock up after being periodically queried such command
  sequences.

* There is no reliable and unintrusive way for a userland program to
  tell whether the target device is safe for media presence polling.
  For example, polling for media presence during an on-going burning
  session can make it fail.  The polling program can avoid this by
  opening the device with O_EXCL but then it risks making a valid
  exclusive user of the device fail w/ -EBUSY.

* Userland polling is unnecessarily heavy and in-kernel implementation
  is lighter and better coordinated (workqueue, timer slack).

This patch implements framework for in-kernel disk event handling,
which includes media presence polling.

* bdops->check_events() is added, which supercedes ->media_changed().
  It should check whether there's any pending event and return if so.
  Currently, two events are defined - DISK_EVENT_MEDIA_CHANGE and
  DISK_EVENT_EJECT_REQUEST.  ->check_events() is guaranteed not to be
  called parallelly.

* gendisk->events and ->async_events are added.  These should be
  initialized by block driver before passing the device to add_disk().
  The former contains the mask of all supported events and the latter
  the mask of all events which the device can report without polling.
  /sys/block/*/events[_async] export these to userland.

* Kernel parameter block.events_dfl_poll_msecs controls the system
  polling interval (default is 0 which means disable) and
  /sys/block/*/events_poll_msecs control polling intervals for
  individual devices (default is -1 meaning use system setting).  Note
  that if a device can report all supported events asynchronously and
  its polling interval isn't explicitly set, the device won't be
  polled regardless of the system polling interval.

* If a device is opened exclusively with write access, event checking
  is automatically disabled until all write exclusive accesses are
  released.

* There are event 'clearing' events.  For example, both of currently
  defined events are cleared after the device has been successfully
  opened.  This information is passed to ->check_events() callback
  using @clearing argument as a hint.

* Event checking is always performed from system_nrt_wq and timer
  slack is set to 25% for polling.

* Nothing changes for drivers which implement ->media_changed() but
  not ->check_events().  Going forward, all drivers will be converted
  to ->check_events() and ->media_change() will be dropped.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Kay Sievers <kay.sievers@vrfy.org>
Cc: Jan Kara <jack@suse.cz>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>

block/genhd.c

@@ -18,6 +18,7 @@
 #include <linux/buffer_head.h>
 #include <linux/mutex.h>
 #include <linux/idr.h>
+#include <linux/log2.h>
 
 #include "blk.h"
 
@@ -35,6 +36,10 @@ static DEFINE_IDR(ext_devt_idr);
 
 static struct device_type disk_type;
 
+static void disk_add_events(struct gendisk *disk);
+static void disk_del_events(struct gendisk *disk);
+static void disk_release_events(struct gendisk *disk);
+
 /**
  * disk_get_part - get partition
  * @disk: disk to look partition from
@@ -609,6 +614,8 @@ void add_disk(struct gendisk *disk)
 	retval = sysfs_create_link(&disk_to_dev(disk)->kobj, &bdi->dev->kobj,
 				   "bdi");
 	WARN_ON(retval);
+
+	disk_add_events(disk);
 }
 EXPORT_SYMBOL(add_disk);
 
@@ -617,6 +624,8 @@ void del_gendisk(struct gendisk *disk)
 	struct disk_part_iter piter;
 	struct hd_struct *part;
 
+	disk_del_events(disk);
+
 	/* invalidate stuff */
 	disk_part_iter_init(&piter, disk,
 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
@@ -1089,6 +1098,7 @@ static void disk_release(struct device *dev)
 {
 	struct gendisk *disk = dev_to_disk(dev);
 
+	disk_release_events(disk);
 	kfree(disk->random);
 	disk_replace_part_tbl(disk, NULL);
 	free_part_stats(&disk->part0);
@@ -1350,3 +1360,422 @@ int invalidate_partition(struct gendisk *disk, int partno)
 }
 
 EXPORT_SYMBOL(invalidate_partition);
+
+/*
+ * Disk events - monitor disk events like media change and eject request.
+ */
+struct disk_events {
+	struct list_head	node;		/* all disk_event's */
+	struct gendisk		*disk;		/* the associated disk */
+	spinlock_t		lock;
+
+	int			block;		/* event blocking depth */
+	unsigned int		pending;	/* events already sent out */
+	unsigned int		clearing;	/* events being cleared */
+
+	long			poll_msecs;	/* interval, -1 for default */
+	struct delayed_work	dwork;
+};
+
+static const char *disk_events_strs[] = {
+	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "media_change",
+	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "eject_request",
+};
+
+static char *disk_uevents[] = {
+	[ilog2(DISK_EVENT_MEDIA_CHANGE)]	= "DISK_MEDIA_CHANGE=1",
+	[ilog2(DISK_EVENT_EJECT_REQUEST)]	= "DISK_EJECT_REQUEST=1",
+};
+
+/* list of all disk_events */
+static DEFINE_MUTEX(disk_events_mutex);
+static LIST_HEAD(disk_events);
+
+/* disable in-kernel polling by default */
+static unsigned long disk_events_dfl_poll_msecs	= 0;
+
+static unsigned long disk_events_poll_jiffies(struct gendisk *disk)
+{
+	struct disk_events *ev = disk->ev;
+	long intv_msecs = 0;
+
+	/*
+	 * If device-specific poll interval is set, always use it.  If
+	 * the default is being used, poll iff there are events which
+	 * can't be monitored asynchronously.
+	 */
+	if (ev->poll_msecs >= 0)
+		intv_msecs = ev->poll_msecs;
+	else if (disk->events & ~disk->async_events)
+		intv_msecs = disk_events_dfl_poll_msecs;
+
+	return msecs_to_jiffies(intv_msecs);
+}
+
+static void __disk_block_events(struct gendisk *disk, bool sync)
+{
+	struct disk_events *ev = disk->ev;
+	unsigned long flags;
+	bool cancel;
+
+	spin_lock_irqsave(&ev->lock, flags);
+	cancel = !ev->block++;
+	spin_unlock_irqrestore(&ev->lock, flags);
+
+	if (cancel) {
+		if (sync)
+			cancel_delayed_work_sync(&disk->ev->dwork);
+		else
+			cancel_delayed_work(&disk->ev->dwork);
+	}
+}
+
+static void __disk_unblock_events(struct gendisk *disk, bool check_now)
+{
+	struct disk_events *ev = disk->ev;
+	unsigned long intv;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ev->lock, flags);
+
+	if (WARN_ON_ONCE(ev->block <= 0))
+		goto out_unlock;
+
+	if (--ev->block)
+		goto out_unlock;
+
+	/*
+	 * Not exactly a latency critical operation, set poll timer
+	 * slack to 25% and kick event check.
+	 */
+	intv = disk_events_poll_jiffies(disk);
+	set_timer_slack(&ev->dwork.timer, intv / 4);
+	if (check_now)
+		queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
+	else if (intv)
+		queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
+out_unlock:
+	spin_unlock_irqrestore(&ev->lock, flags);
+}
+
+/**
+ * disk_block_events - block and flush disk event checking
+ * @disk: disk to block events for
+ *
+ * On return from this function, it is guaranteed that event checking
+ * isn't in progress and won't happen until unblocked by
+ * disk_unblock_events().  Events blocking is counted and the actual
+ * unblocking happens after the matching number of unblocks are done.
+ *
+ * Note that this intentionally does not block event checking from
+ * disk_clear_events().
+ *
+ * CONTEXT:
+ * Might sleep.
+ */
+void disk_block_events(struct gendisk *disk)
+{
+	if (disk->ev)
+		__disk_block_events(disk, true);
+}
+
+/**
+ * disk_unblock_events - unblock disk event checking
+ * @disk: disk to unblock events for
+ *
+ * Undo disk_block_events().  When the block count reaches zero, it
+ * starts events polling if configured.
+ *
+ * CONTEXT:
+ * Don't care.  Safe to call from irq context.
+ */
+void disk_unblock_events(struct gendisk *disk)
+{
+	if (disk->ev)
+		__disk_unblock_events(disk, true);
+}
+
+/**
+ * disk_check_events - schedule immediate event checking
+ * @disk: disk to check events for
+ *
+ * Schedule immediate event checking on @disk if not blocked.
+ *
+ * CONTEXT:
+ * Don't care.  Safe to call from irq context.
+ */
+void disk_check_events(struct gendisk *disk)
+{
+	if (disk->ev) {
+		__disk_block_events(disk, false);
+		__disk_unblock_events(disk, true);
+	}
+}
+EXPORT_SYMBOL_GPL(disk_check_events);
+
+/**
+ * disk_clear_events - synchronously check, clear and return pending events
+ * @disk: disk to fetch and clear events from
+ * @mask: mask of events to be fetched and clearted
+ *
+ * Disk events are synchronously checked and pending events in @mask
+ * are cleared and returned.  This ignores the block count.
+ *
+ * CONTEXT:
+ * Might sleep.
+ */
+unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask)
+{
+	const struct block_device_operations *bdops = disk->fops;
+	struct disk_events *ev = disk->ev;
+	unsigned int pending;
+
+	if (!ev) {
+		/* for drivers still using the old ->media_changed method */
+		if ((mask & DISK_EVENT_MEDIA_CHANGE) &&
+		    bdops->media_changed && bdops->media_changed(disk))
+			return DISK_EVENT_MEDIA_CHANGE;
+		return 0;
+	}
+
+	/* tell the workfn about the events being cleared */
+	spin_lock_irq(&ev->lock);
+	ev->clearing |= mask;
+	spin_unlock_irq(&ev->lock);
+
+	/* uncondtionally schedule event check and wait for it to finish */
+	__disk_block_events(disk, true);
+	queue_delayed_work(system_nrt_wq, &ev->dwork, 0);
+	flush_delayed_work(&ev->dwork);
+	__disk_unblock_events(disk, false);
+
+	/* then, fetch and clear pending events */
+	spin_lock_irq(&ev->lock);
+	WARN_ON_ONCE(ev->clearing & mask);	/* cleared by workfn */
+	pending = ev->pending & mask;
+	ev->pending &= ~mask;
+	spin_unlock_irq(&ev->lock);
+
+	return pending;
+}
+
+static void disk_events_workfn(struct work_struct *work)
+{
+	struct delayed_work *dwork = to_delayed_work(work);
+	struct disk_events *ev = container_of(dwork, struct disk_events, dwork);
+	struct gendisk *disk = ev->disk;
+	char *envp[ARRAY_SIZE(disk_uevents) + 1] = { };
+	unsigned int clearing = ev->clearing;
+	unsigned int events;
+	unsigned long intv;
+	int nr_events = 0, i;
+
+	/* check events */
+	events = disk->fops->check_events(disk, clearing);
+
+	/* accumulate pending events and schedule next poll if necessary */
+	spin_lock_irq(&ev->lock);
+
+	events &= ~ev->pending;
+	ev->pending |= events;
+	ev->clearing &= ~clearing;
+
+	intv = disk_events_poll_jiffies(disk);
+	if (!ev->block && intv)
+		queue_delayed_work(system_nrt_wq, &ev->dwork, intv);
+
+	spin_unlock_irq(&ev->lock);
+
+	/* tell userland about new events */
+	for (i = 0; i < ARRAY_SIZE(disk_uevents); i++)
+		if (events & (1 << i))
+			envp[nr_events++] = disk_uevents[i];
+
+	if (nr_events)
+		kobject_uevent_env(&disk_to_dev(disk)->kobj, KOBJ_CHANGE, envp);
+}
+
+/*
+ * A disk events enabled device has the following sysfs nodes under
+ * its /sys/block/X/ directory.
+ *
+ * events		: list of all supported events
+ * events_async		: list of events which can be detected w/o polling
+ * events_poll_msecs	: polling interval, 0: disable, -1: system default
+ */
+static ssize_t __disk_events_show(unsigned int events, char *buf)
+{
+	const char *delim = "";
+	ssize_t pos = 0;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(disk_events_strs); i++)
+		if (events & (1 << i)) {
+			pos += sprintf(buf + pos, "%s%s",
+				       delim, disk_events_strs[i]);
+			delim = " ";
+		}
+	if (pos)
+		pos += sprintf(buf + pos, "\n");
+	return pos;
+}
+
+static ssize_t disk_events_show(struct device *dev,
+				struct device_attribute *attr, char *buf)
+{
+	struct gendisk *disk = dev_to_disk(dev);
+
+	return __disk_events_show(disk->events, buf);
+}
+
+static ssize_t disk_events_async_show(struct device *dev,
+				      struct device_attribute *attr, char *buf)
+{
+	struct gendisk *disk = dev_to_disk(dev);
+
+	return __disk_events_show(disk->async_events, buf);
+}
+
+static ssize_t disk_events_poll_msecs_show(struct device *dev,
+					   struct device_attribute *attr,
+					   char *buf)
+{
+	struct gendisk *disk = dev_to_disk(dev);
+
+	return sprintf(buf, "%ld\n", disk->ev->poll_msecs);
+}
+
+static ssize_t disk_events_poll_msecs_store(struct device *dev,
+					    struct device_attribute *attr,
+					    const char *buf, size_t count)
+{
+	struct gendisk *disk = dev_to_disk(dev);
+	long intv;
+
+	if (!count || !sscanf(buf, "%ld", &intv))
+		return -EINVAL;
+
+	if (intv < 0 && intv != -1)
+		return -EINVAL;
+
+	__disk_block_events(disk, true);
+	disk->ev->poll_msecs = intv;
+	__disk_unblock_events(disk, true);
+
+	return count;
+}
+
+static const DEVICE_ATTR(events, S_IRUGO, disk_events_show, NULL);
+static const DEVICE_ATTR(events_async, S_IRUGO, disk_events_async_show, NULL);
+static const DEVICE_ATTR(events_poll_msecs, S_IRUGO|S_IWUSR,
+			 disk_events_poll_msecs_show,
+			 disk_events_poll_msecs_store);
+
+static const struct attribute *disk_events_attrs[] = {
+	&dev_attr_events.attr,
+	&dev_attr_events_async.attr,
+	&dev_attr_events_poll_msecs.attr,
+	NULL,
+};
+
+/*
+ * The default polling interval can be specified by the kernel
+ * parameter block.events_dfl_poll_msecs which defaults to 0
+ * (disable).  This can also be modified runtime by writing to
+ * /sys/module/block/events_dfl_poll_msecs.
+ */
+static int disk_events_set_dfl_poll_msecs(const char *val,
+					  const struct kernel_param *kp)
+{
+	struct disk_events *ev;
+	int ret;
+
+	ret = param_set_ulong(val, kp);
+	if (ret < 0)
+		return ret;
+
+	mutex_lock(&disk_events_mutex);
+
+	list_for_each_entry(ev, &disk_events, node)
+		disk_check_events(ev->disk);
+
+	mutex_unlock(&disk_events_mutex);
+
+	return 0;
+}
+
+static const struct kernel_param_ops disk_events_dfl_poll_msecs_param_ops = {
+	.set	= disk_events_set_dfl_poll_msecs,
+	.get	= param_get_ulong,
+};
+
+#undef MODULE_PARAM_PREFIX
+#define MODULE_PARAM_PREFIX	"block."
+
+module_param_cb(events_dfl_poll_msecs, &disk_events_dfl_poll_msecs_param_ops,
+		&disk_events_dfl_poll_msecs, 0644);
+
+/*
+ * disk_{add|del|release}_events - initialize and destroy disk_events.
+ */
+static void disk_add_events(struct gendisk *disk)
+{
+	struct disk_events *ev;
+
+	if (!disk->fops->check_events || !(disk->events | disk->async_events))
+		return;
+
+	ev = kzalloc(sizeof(*ev), GFP_KERNEL);
+	if (!ev) {
+		pr_warn("%s: failed to initialize events\n", disk->disk_name);
+		return;
+	}
+
+	if (sysfs_create_files(&disk_to_dev(disk)->kobj,
+			       disk_events_attrs) < 0) {
+		pr_warn("%s: failed to create sysfs files for events\n",
+			disk->disk_name);
+		kfree(ev);
+		return;
+	}
+
+	disk->ev = ev;
+
+	INIT_LIST_HEAD(&ev->node);
+	ev->disk = disk;
+	spin_lock_init(&ev->lock);
+	ev->block = 1;
+	ev->poll_msecs = -1;
+	INIT_DELAYED_WORK(&ev->dwork, disk_events_workfn);
+
+	mutex_lock(&disk_events_mutex);
+	list_add_tail(&ev->node, &disk_events);
+	mutex_unlock(&disk_events_mutex);
+
+	/*
+	 * Block count is initialized to 1 and the following initial
+	 * unblock kicks it into action.
+	 */
+	__disk_unblock_events(disk, true);
+}
+
+static void disk_del_events(struct gendisk *disk)
+{
+	if (!disk->ev)
+		return;
+
+	__disk_block_events(disk, true);
+
+	mutex_lock(&disk_events_mutex);
+	list_del_init(&disk->ev->node);
+	mutex_unlock(&disk_events_mutex);
+
+	sysfs_remove_files(&disk_to_dev(disk)->kobj, disk_events_attrs);
+}
+
+static void disk_release_events(struct gendisk *disk)
+{
+	/* the block count should be 1 from disk_del_events() */
+	WARN_ON_ONCE(disk->ev && disk->ev->block != 1);
+	kfree(disk->ev);
+}

fs/block_dev.c

@@ -948,10 +948,11 @@ int check_disk_change(struct block_device *bdev)
 {
 	struct gendisk *disk = bdev->bd_disk;
 	const struct block_device_operations *bdops = disk->fops;
+	unsigned int events;
 
-	if (!bdops->media_changed)
-		return 0;
-	if (!bdops->media_changed(bdev->bd_disk))
+	events = disk_clear_events(disk, DISK_EVENT_MEDIA_CHANGE |
+				   DISK_EVENT_EJECT_REQUEST);
+	if (!(events & DISK_EVENT_MEDIA_CHANGE))
 		return 0;
 
 	flush_disk(bdev);
@@ -1158,9 +1159,10 @@ int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 
 	if (whole) {
 		/* finish claiming */
+		mutex_lock(&bdev->bd_mutex);
 		spin_lock(&bdev_lock);
 
-		if (res == 0) {
+		if (!res) {
 			BUG_ON(!bd_may_claim(bdev, whole, holder));
 			/*
 			 * Note that for a whole device bd_holders
@@ -1180,6 +1182,20 @@ int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 		wake_up_bit(&whole->bd_claiming, 0);
 
 		spin_unlock(&bdev_lock);
+
+		/*
+		 * Block event polling for write claims.  Any write
+		 * holder makes the write_holder state stick until all
+		 * are released.  This is good enough and tracking
+		 * individual writeable reference is too fragile given
+		 * the way @mode is used in blkdev_get/put().
+		 */
+		if (!res && (mode & FMODE_WRITE) && !bdev->bd_write_holder) {
+			bdev->bd_write_holder = true;
+			disk_block_events(bdev->bd_disk);
+		}
+
+		mutex_unlock(&bdev->bd_mutex);
 		bdput(whole);
 	}
 
@@ -1353,12 +1369,23 @@ int blkdev_put(struct block_device *bdev, fmode_t mode)
 
 		spin_unlock(&bdev_lock);
 
-		/* if this was the last claim, holder link should go too */
-		if (bdev_free)
+		/*
+		 * If this was the last claim, remove holder link and
+		 * unblock evpoll if it was a write holder.
+		 */
+		if (bdev_free) {
 			bd_unlink_disk_holder(bdev);
+			if (bdev->bd_write_holder) {
+				disk_unblock_events(bdev->bd_disk);
+				bdev->bd_write_holder = false;
+			} else
+				disk_check_events(bdev->bd_disk);
+		}
 
 		mutex_unlock(&bdev->bd_mutex);
-	}
+	} else
+		disk_check_events(bdev->bd_disk);
+
 	return __blkdev_put(bdev, mode, 0);
 }
 EXPORT_SYMBOL(blkdev_put);

include/linux/blkdev.h

@@ -1251,6 +1251,9 @@ struct block_device_operations {
 	int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
 	int (*direct_access) (struct block_device *, sector_t,
 						void **, unsigned long *);
+	unsigned int (*check_events) (struct gendisk *disk,
+				      unsigned int clearing);
+	/* ->media_changed() is DEPRECATED, use ->check_events() instead */
 	int (*media_changed) (struct gendisk *);
 	void (*unlock_native_capacity) (struct gendisk *);
 	int (*revalidate_disk) (struct gendisk *);

include/linux/fs.h

@@ -662,6 +662,7 @@ struct block_device {
 	void *			bd_claiming;
 	void *			bd_holder;
 	int			bd_holders;
+	bool			bd_write_holder;
 #ifdef CONFIG_SYSFS
 	struct gendisk *	bd_holder_disk;	/* for sysfs slave linkng */
 #endif

include/linux/genhd.h

@@ -127,6 +127,11 @@ struct hd_struct {
 #define GENHD_FL_EXT_DEVT			64 /* allow extended devt */
 #define GENHD_FL_NATIVE_CAPACITY		128
 
+enum {
+	DISK_EVENT_MEDIA_CHANGE			= 1 << 0, /* media changed */
+	DISK_EVENT_EJECT_REQUEST		= 1 << 1, /* eject requested */
+};
+
 #define BLK_SCSI_MAX_CMDS	(256)
 #define BLK_SCSI_CMD_PER_LONG	(BLK_SCSI_MAX_CMDS / (sizeof(long) * 8))
 
@@ -143,6 +148,8 @@ struct disk_part_tbl {
 	struct hd_struct __rcu *part[];
 };
 
+struct disk_events;
+
 struct gendisk {
 	/* major, first_minor and minors are input parameters only,
 	 * don't use directly.  Use disk_devt() and disk_max_parts().
@@ -154,6 +161,10 @@ struct gendisk {
 
 	char disk_name[DISK_NAME_LEN];	/* name of major driver */
 	char *(*devnode)(struct gendisk *gd, mode_t *mode);
+
+	unsigned int events;		/* supported events */
+	unsigned int async_events;	/* async events, subset of all */
+
 	/* Array of pointers to partitions indexed by partno.
 	 * Protected with matching bdev lock but stat and other
 	 * non-critical accesses use RCU.  Always access through
@@ -171,8 +182,8 @@ struct gendisk {
 	struct kobject *slave_dir;
 
 	struct timer_rand_state *random;
-
 	atomic_t sync_io;		/* RAID */
+	struct disk_events *ev;
 #ifdef  CONFIG_BLK_DEV_INTEGRITY
 	struct blk_integrity *integrity;
 #endif
@@ -405,6 +416,11 @@ static inline int get_disk_ro(struct gendisk *disk)
 	return disk->part0.policy;
 }
 
+extern void disk_block_events(struct gendisk *disk);
+extern void disk_unblock_events(struct gendisk *disk);
+extern void disk_check_events(struct gendisk *disk);
+extern unsigned int disk_clear_events(struct gendisk *disk, unsigned int mask);
+
 /* drivers/char/random.c */
 extern void add_disk_randomness(struct gendisk *disk);
 extern void rand_initialize_disk(struct gendisk *disk);