Merge git://git.kernel.org/pub/scm/linux/kernel/git/agk/linux-2.6-dm
[sfrench/cifs-2.6.git] / drivers / md / dm.c
index 8d40f27cce894e4a3c489a7d84d321fb39b363d6..788ba96a6256aaed6de8625d306cc4d5e33a82f9 100644 (file)
@@ -99,19 +99,9 @@ union map_info *dm_get_mapinfo(struct bio *bio)
 /*
  * Work processed by per-device workqueue.
  */
-struct dm_wq_req {
-       enum {
-               DM_WQ_FLUSH_DEFERRED,
-       } type;
-       struct work_struct work;
-       struct mapped_device *md;
-       void *context;
-};
-
 struct mapped_device {
        struct rw_semaphore io_lock;
        struct mutex suspend_lock;
-       spinlock_t pushback_lock;
        rwlock_t map_lock;
        atomic_t holders;
        atomic_t open_count;
@@ -129,8 +119,9 @@ struct mapped_device {
         */
        atomic_t pending;
        wait_queue_head_t wait;
+       struct work_struct work;
        struct bio_list deferred;
-       struct bio_list pushback;
+       spinlock_t deferred_lock;
 
        /*
         * Processing queue (flush/barriers)
@@ -453,7 +444,9 @@ static int queue_io(struct mapped_device *md, struct bio *bio)
                return 1;
        }
 
+       spin_lock_irq(&md->deferred_lock);
        bio_list_add(&md->deferred, bio);
+       spin_unlock_irq(&md->deferred_lock);
 
        up_write(&md->io_lock);
        return 0;               /* deferred successfully */
@@ -537,16 +530,14 @@ static void dec_pending(struct dm_io *io, int error)
                if (io->error == DM_ENDIO_REQUEUE) {
                        /*
                         * Target requested pushing back the I/O.
-                        * This must be handled before the sleeper on
-                        * suspend queue merges the pushback list.
                         */
-                       spin_lock_irqsave(&md->pushback_lock, flags);
+                       spin_lock_irqsave(&md->deferred_lock, flags);
                        if (__noflush_suspending(md))
-                               bio_list_add(&md->pushback, io->bio);
+                               bio_list_add(&md->deferred, io->bio);
                        else
                                /* noflush suspend was interrupted. */
                                io->error = -EIO;
-                       spin_unlock_irqrestore(&md->pushback_lock, flags);
+                       spin_unlock_irqrestore(&md->deferred_lock, flags);
                }
 
                end_io_acct(io);
@@ -834,20 +825,22 @@ static int __clone_and_map(struct clone_info *ci)
 }
 
 /*
- * Split the bio into several clones.
+ * Split the bio into several clones and submit it to targets.
  */
-static int __split_bio(struct mapped_device *md, struct bio *bio)
+static void __split_and_process_bio(struct mapped_device *md, struct bio *bio)
 {
        struct clone_info ci;
        int error = 0;
 
        ci.map = dm_get_table(md);
-       if (unlikely(!ci.map))
-               return -EIO;
+       if (unlikely(!ci.map)) {
+               bio_io_error(bio);
+               return;
+       }
        if (unlikely(bio_barrier(bio) && !dm_table_barrier_ok(ci.map))) {
                dm_table_put(ci.map);
                bio_endio(bio, -EOPNOTSUPP);
-               return 0;
+               return;
        }
        ci.md = md;
        ci.bio = bio;
@@ -867,8 +860,6 @@ static int __split_bio(struct mapped_device *md, struct bio *bio)
        /* drop the extra reference count */
        dec_pending(ci.io, error);
        dm_table_put(ci.map);
-
-       return 0;
 }
 /*-----------------------------------------------------------------
  * CRUD END
@@ -959,8 +950,9 @@ static int dm_request(struct request_queue *q, struct bio *bio)
                down_read(&md->io_lock);
        }
 
-       r = __split_bio(md, bio);
+       __split_and_process_bio(md, bio);
        up_read(&md->io_lock);
+       return 0;
 
 out_req:
        if (r < 0)
@@ -1074,6 +1066,8 @@ out:
 
 static struct block_device_operations dm_blk_dops;
 
+static void dm_wq_work(struct work_struct *work);
+
 /*
  * Allocate and initialise a blank device with a given minor.
  */
@@ -1101,7 +1095,7 @@ static struct mapped_device *alloc_dev(int minor)
 
        init_rwsem(&md->io_lock);
        mutex_init(&md->suspend_lock);
-       spin_lock_init(&md->pushback_lock);
+       spin_lock_init(&md->deferred_lock);
        rwlock_init(&md->map_lock);
        atomic_set(&md->holders, 1);
        atomic_set(&md->open_count, 0);
@@ -1118,6 +1112,7 @@ static struct mapped_device *alloc_dev(int minor)
        md->queue->backing_dev_info.congested_fn = dm_any_congested;
        md->queue->backing_dev_info.congested_data = md;
        blk_queue_make_request(md->queue, dm_request);
+       blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN, NULL);
        blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
        md->queue->unplug_fn = dm_unplug_all;
        blk_queue_merge_bvec(md->queue, dm_merge_bvec);
@@ -1140,6 +1135,7 @@ static struct mapped_device *alloc_dev(int minor)
 
        atomic_set(&md->pending, 0);
        init_waitqueue_head(&md->wait);
+       INIT_WORK(&md->work, dm_wq_work);
        init_waitqueue_head(&md->eventq);
 
        md->disk->major = _major;
@@ -1379,18 +1375,24 @@ void dm_put(struct mapped_device *md)
 }
 EXPORT_SYMBOL_GPL(dm_put);
 
-static int dm_wait_for_completion(struct mapped_device *md)
+static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
 {
        int r = 0;
+       DECLARE_WAITQUEUE(wait, current);
+
+       dm_unplug_all(md->queue);
+
+       add_wait_queue(&md->wait, &wait);
 
        while (1) {
-               set_current_state(TASK_INTERRUPTIBLE);
+               set_current_state(interruptible);
 
                smp_mb();
                if (!atomic_read(&md->pending))
                        break;
 
-               if (signal_pending(current)) {
+               if (interruptible == TASK_INTERRUPTIBLE &&
+                   signal_pending(current)) {
                        r = -EINTR;
                        break;
                }
@@ -1399,67 +1401,40 @@ static int dm_wait_for_completion(struct mapped_device *md)
        }
        set_current_state(TASK_RUNNING);
 
+       remove_wait_queue(&md->wait, &wait);
+
        return r;
 }
 
 /*
  * Process the deferred bios
  */
-static void __flush_deferred_io(struct mapped_device *md)
+static void dm_wq_work(struct work_struct *work)
 {
+       struct mapped_device *md = container_of(work, struct mapped_device,
+                                               work);
        struct bio *c;
 
-       while ((c = bio_list_pop(&md->deferred))) {
-               if (__split_bio(md, c))
-                       bio_io_error(c);
-       }
-
-       clear_bit(DMF_BLOCK_IO, &md->flags);
-}
+       down_write(&md->io_lock);
 
-static void __merge_pushback_list(struct mapped_device *md)
-{
-       unsigned long flags;
+next_bio:
+       spin_lock_irq(&md->deferred_lock);
+       c = bio_list_pop(&md->deferred);
+       spin_unlock_irq(&md->deferred_lock);
 
-       spin_lock_irqsave(&md->pushback_lock, flags);
-       clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
-       bio_list_merge_head(&md->deferred, &md->pushback);
-       bio_list_init(&md->pushback);
-       spin_unlock_irqrestore(&md->pushback_lock, flags);
-}
+       if (c) {
+               __split_and_process_bio(md, c);
+               goto next_bio;
+       }
 
-static void dm_wq_work(struct work_struct *work)
-{
-       struct dm_wq_req *req = container_of(work, struct dm_wq_req, work);
-       struct mapped_device *md = req->md;
+       clear_bit(DMF_BLOCK_IO, &md->flags);
 
-       down_write(&md->io_lock);
-       switch (req->type) {
-       case DM_WQ_FLUSH_DEFERRED:
-               __flush_deferred_io(md);
-               break;
-       default:
-               DMERR("dm_wq_work: unrecognised work type %d", req->type);
-               BUG();
-       }
        up_write(&md->io_lock);
 }
 
-static void dm_wq_queue(struct mapped_device *md, int type, void *context,
-                       struct dm_wq_req *req)
-{
-       req->type = type;
-       req->md = md;
-       req->context = context;
-       INIT_WORK(&req->work, dm_wq_work);
-       queue_work(md->wq, &req->work);
-}
-
-static void dm_queue_flush(struct mapped_device *md, int type, void *context)
+static void dm_queue_flush(struct mapped_device *md)
 {
-       struct dm_wq_req req;
-
-       dm_wq_queue(md, type, context, &req);
+       queue_work(md->wq, &md->work);
        flush_workqueue(md->wq);
 }
 
@@ -1534,7 +1509,6 @@ static void unlock_fs(struct mapped_device *md)
 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
 {
        struct dm_table *map = NULL;
-       DECLARE_WAITQUEUE(wait, current);
        int r = 0;
        int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
        int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
@@ -1584,28 +1558,22 @@ int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
        down_write(&md->io_lock);
        set_bit(DMF_BLOCK_IO, &md->flags);
 
-       add_wait_queue(&md->wait, &wait);
        up_write(&md->io_lock);
 
-       /* unplug */
-       if (map)
-               dm_table_unplug_all(map);
-
        /*
         * Wait for the already-mapped ios to complete.
         */
-       r = dm_wait_for_completion(md);
+       r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
 
        down_write(&md->io_lock);
-       remove_wait_queue(&md->wait, &wait);
 
        if (noflush)
-               __merge_pushback_list(md);
+               clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
        up_write(&md->io_lock);
 
        /* were we interrupted ? */
        if (r < 0) {
-               dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
+               dm_queue_flush(md);
 
                unlock_fs(md);
                goto out; /* pushback list is already flushed, so skip flush */
@@ -1645,7 +1613,7 @@ int dm_resume(struct mapped_device *md)
        if (r)
                goto out;
 
-       dm_queue_flush(md, DM_WQ_FLUSH_DEFERRED, NULL);
+       dm_queue_flush(md);
 
        unlock_fs(md);