Merge branch 'for-3.5/drivers' of git://git.kernel.dk/linux-block

Pull block driver updates from Jens Axboe:
 "Here are the driver related changes for 3.5.  It contains:

   - The floppy changes from Jiri.  Jiri is now also marked as the
     maintainer of floppy.c, I shall be publically branding his forehead
     with red hot iron at the next opportune moment.

   - A batch of drbd updates and fixes from the linbit crew, as well as
     fixes from others.

   - Two small fixes for xen-blkfront courtesy of Jan."

* 'for-3.5/drivers' of git://git.kernel.dk/linux-block: (70 commits)
  floppy: take over maintainership
  floppy: remove floppy-specific O_EXCL handling
  floppy: convert to delayed work and single-thread wq
  xen-blkfront: module exit handling adjustments
  xen-blkfront: properly name all devices
  drbd: grammar fix in log message
  drbd: check MODULE for THIS_MODULE
  drbd: Restore the request restart logic
  drbd: introduce a bio_set to allocate housekeeping bios from
  drbd: remove unused define
  drbd: bm_page_async_io: properly initialize page->private
  drbd: use the newly introduced page pool for bitmap IO
  drbd: add page pool to be used for meta data IO
  drbd: allow bitmap to change during writeout from resync_finished
  drbd: fix race between drbdadm invalidate/verify and finishing resync
  drbd: fix resend/resubmit of frozen IO
  drbd: Ensure that data_size is not 0 before using data_size-1 as index
  drbd: Delay/reject other state changes while establishing a connection
  drbd: move put_ldev from __req_mod() to the endio callback
  drbd: fix WRITE_ACKED_BY_PEER_AND_SIS to not set RQ_NET_DONE
  ...

10 files changed, 741 insertions, 283 deletions

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index cf0e63dd97da..e54e31b02b88 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -65,39 +65,80 @@ struct drbd_atodb_wait {
 
 int w_al_write_transaction(struct drbd_conf *, struct drbd_work *, int);
 
+void *drbd_md_get_buffer(struct drbd_conf *mdev)
+{
+	int r;
+
+	wait_event(mdev->misc_wait,
+		   (r = atomic_cmpxchg(&mdev->md_io_in_use, 0, 1)) == 0 ||
+		   mdev->state.disk <= D_FAILED);
+
+	return r ? NULL : page_address(mdev->md_io_page);
+}
+
+void drbd_md_put_buffer(struct drbd_conf *mdev)
+{
+	if (atomic_dec_and_test(&mdev->md_io_in_use))
+		wake_up(&mdev->misc_wait);
+}
+
+static bool md_io_allowed(struct drbd_conf *mdev)
+{
+	enum drbd_disk_state ds = mdev->state.disk;
+	return ds >= D_NEGOTIATING || ds == D_ATTACHING;
+}
+
+void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
+				     unsigned int *done)
+{
+	long dt = bdev->dc.disk_timeout * HZ / 10;
+	if (dt == 0)
+		dt = MAX_SCHEDULE_TIMEOUT;
+
+	dt = wait_event_timeout(mdev->misc_wait, *done || !md_io_allowed(mdev), dt);
+	if (dt == 0)
+		dev_err(DEV, "meta-data IO operation timed out\n");
+}
+
 static int _drbd_md_sync_page_io(struct drbd_conf *mdev,
 				 struct drbd_backing_dev *bdev,
 				 struct page *page, sector_t sector,
 				 int rw, int size)
 {
 	struct bio *bio;
-	struct drbd_md_io md_io;
 	int ok;
 
-	md_io.mdev = mdev;
-	init_completion(&md_io.event);
-	md_io.error = 0;
+	mdev->md_io.done = 0;
+	mdev->md_io.error = -ENODEV;
 
 	if ((rw & WRITE) && !test_bit(MD_NO_FUA, &mdev->flags))
 		rw |= REQ_FUA | REQ_FLUSH;
 	rw |= REQ_SYNC;
 
-	bio = bio_alloc(GFP_NOIO, 1);
+	bio = bio_alloc_drbd(GFP_NOIO);
 	bio->bi_bdev = bdev->md_bdev;
 	bio->bi_sector = sector;
 	ok = (bio_add_page(bio, page, size, 0) == size);
 	if (!ok)
 		goto out;
-	bio->bi_private = &md_io;
+	bio->bi_private = &mdev->md_io;
 	bio->bi_end_io = drbd_md_io_complete;
 	bio->bi_rw = rw;
 
+	if (!get_ldev_if_state(mdev, D_ATTACHING)) {  /* Corresponding put_ldev in drbd_md_io_complete() */
+		dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in _drbd_md_sync_page_io()\n");
+		ok = 0;
+		goto out;
+	}
+
+	bio_get(bio); /* one bio_put() is in the completion handler */
+	atomic_inc(&mdev->md_io_in_use); /* drbd_md_put_buffer() is in the completion handler */
 	if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD))
 		bio_endio(bio, -EIO);
 	else
 		submit_bio(rw, bio);
-	wait_for_completion(&md_io.event);
-	ok = bio_flagged(bio, BIO_UPTODATE) && md_io.error == 0;
+	wait_until_done_or_disk_failure(mdev, bdev, &mdev->md_io.done);
+	ok = bio_flagged(bio, BIO_UPTODATE) && mdev->md_io.error == 0;
 
  out:
 	bio_put(bio);
@@ -111,7 +152,7 @@ int drbd_md_sync_page_io(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 	int offset = 0;
 	struct page *iop = mdev->md_io_page;
 
-	D_ASSERT(mutex_is_locked(&mdev->md_io_mutex));
+	D_ASSERT(atomic_read(&mdev->md_io_in_use) == 1);
 
 	BUG_ON(!bdev->md_bdev);
 
@@ -328,8 +369,13 @@ w_al_write_transaction(struct drbd_conf *mdev, struct drbd_work *w, int unused)
 		return 1;
 	}
 
-	mutex_lock(&mdev->md_io_mutex); /* protects md_io_buffer, al_tr_cycle, ... */
-	buffer = (struct al_transaction *)page_address(mdev->md_io_page);
+	buffer = drbd_md_get_buffer(mdev); /* protects md_io_buffer, al_tr_cycle, ... */
+	if (!buffer) {
+		dev_err(DEV, "disk failed while waiting for md_io buffer\n");
+		complete(&((struct update_al_work *)w)->event);
+		put_ldev(mdev);
+		return 1;
+	}
 
 	buffer->magic = __constant_cpu_to_be32(DRBD_MAGIC);
 	buffer->tr_number = cpu_to_be32(mdev->al_tr_number);
@@ -374,7 +420,7 @@ w_al_write_transaction(struct drbd_conf *mdev, struct drbd_work *w, int unused)
 	D_ASSERT(mdev->al_tr_pos < MD_AL_MAX_SIZE);
 	mdev->al_tr_number++;
 
-	mutex_unlock(&mdev->md_io_mutex);
+	drbd_md_put_buffer(mdev);
 
 	complete(&((struct update_al_work *)w)->event);
 	put_ldev(mdev);
@@ -443,8 +489,9 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 	/* lock out all other meta data io for now,
 	 * and make sure the page is mapped.
 	 */
-	mutex_lock(&mdev->md_io_mutex);
-	buffer = page_address(mdev->md_io_page);
+	buffer = drbd_md_get_buffer(mdev);
+	if (!buffer)
+		return 0;
 
 	/* Find the valid transaction in the log */
 	for (i = 0; i <= mx; i++) {
@@ -452,7 +499,7 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 		if (rv == 0)
 			continue;
 		if (rv == -1) {
-			mutex_unlock(&mdev->md_io_mutex);
+			drbd_md_put_buffer(mdev);
 			return 0;
 		}
 		cnr = be32_to_cpu(buffer->tr_number);
@@ -478,7 +525,7 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 
 	if (!found_valid) {
 		dev_warn(DEV, "No usable activity log found.\n");
-		mutex_unlock(&mdev->md_io_mutex);
+		drbd_md_put_buffer(mdev);
 		return 1;
 	}
 
@@ -493,7 +540,7 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 		rv = drbd_al_read_tr(mdev, bdev, buffer, i);
 		ERR_IF(rv == 0) goto cancel;
 		if (rv == -1) {
-			mutex_unlock(&mdev->md_io_mutex);
+			drbd_md_put_buffer(mdev);
 			return 0;
 		}
 
@@ -534,7 +581,7 @@ cancel:
 		mdev->al_tr_pos = 0;
 
 	/* ok, we are done with it */
-	mutex_unlock(&mdev->md_io_mutex);
+	drbd_md_put_buffer(mdev);
 
 	dev_info(DEV, "Found %d transactions (%d active extents) in activity log.\n",
 	     transactions, active_extents);
@@ -671,16 +718,20 @@ static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector,
 			else
 				ext->rs_failed += count;
 			if (ext->rs_left < ext->rs_failed) {
-				dev_err(DEV, "BAD! sector=%llus enr=%u rs_left=%d "
-				    "rs_failed=%d count=%d\n",
+				dev_warn(DEV, "BAD! sector=%llus enr=%u rs_left=%d "
+				    "rs_failed=%d count=%d cstate=%s\n",
 				     (unsigned long long)sector,
 				     ext->lce.lc_number, ext->rs_left,
-				     ext->rs_failed, count);
-				dump_stack();
-
-				lc_put(mdev->resync, &ext->lce);
-				drbd_force_state(mdev, NS(conn, C_DISCONNECTING));
-				return;
+				     ext->rs_failed, count,
+				     drbd_conn_str(mdev->state.conn));
+
+				/* We don't expect to be able to clear more bits
+				 * than have been set when we originally counted
+				 * the set bits to cache that value in ext->rs_left.
+				 * Whatever the reason (disconnect during resync,
+				 * delayed local completion of an application write),
+				 * try to fix it up by recounting here. */
+				ext->rs_left = drbd_bm_e_weight(mdev, enr);
 			}
 		} else {
 			/* Normally this element should be in the cache,
@@ -1192,6 +1243,7 @@ int drbd_rs_del_all(struct drbd_conf *mdev)
 		put_ldev(mdev);
 	}
 	spin_unlock_irq(&mdev->al_lock);
+	wake_up(&mdev->al_wait);
 
 	return 0;
 }
diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index 3030201c69d8..b5c5ff53cb57 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -205,7 +205,7 @@ void drbd_bm_unlock(struct drbd_conf *mdev)
 static void bm_store_page_idx(struct page *page, unsigned long idx)
 {
 	BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK));
-	page_private(page) |= idx;
+	set_page_private(page, idx);
 }
 
 static unsigned long bm_page_to_idx(struct page *page)
@@ -886,12 +886,21 @@ void drbd_bm_clear_all(struct drbd_conf *mdev)
 struct bm_aio_ctx {
 	struct drbd_conf *mdev;
 	atomic_t in_flight;
-	struct completion done;
+	unsigned int done;
 	unsigned flags;
 #define BM_AIO_COPY_PAGES	1
 	int error;
+	struct kref kref;
 };
 
+static void bm_aio_ctx_destroy(struct kref *kref)
+{
+	struct bm_aio_ctx *ctx = container_of(kref, struct bm_aio_ctx, kref);
+
+	put_ldev(ctx->mdev);
+	kfree(ctx);
+}
+
 /* bv_page may be a copy, or may be the original */
 static void bm_async_io_complete(struct bio *bio, int error)
 {
@@ -930,20 +939,21 @@ static void bm_async_io_complete(struct bio *bio, int error)
 
 	bm_page_unlock_io(mdev, idx);
 
-	/* FIXME give back to page pool */
 	if (ctx->flags & BM_AIO_COPY_PAGES)
-		put_page(bio->bi_io_vec[0].bv_page);
+		mempool_free(bio->bi_io_vec[0].bv_page, drbd_md_io_page_pool);
 
 	bio_put(bio);
 
-	if (atomic_dec_and_test(&ctx->in_flight))
-		complete(&ctx->done);
+	if (atomic_dec_and_test(&ctx->in_flight)) {
+		ctx->done = 1;
+		wake_up(&mdev->misc_wait);
+		kref_put(&ctx->kref, &bm_aio_ctx_destroy);
+	}
 }
 
 static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must_hold(local)
 {
-	/* we are process context. we always get a bio */
-	struct bio *bio = bio_alloc(GFP_KERNEL, 1);
+	struct bio *bio = bio_alloc_drbd(GFP_NOIO);
 	struct drbd_conf *mdev = ctx->mdev;
 	struct drbd_bitmap *b = mdev->bitmap;
 	struct page *page;
@@ -966,10 +976,8 @@ static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must
 	bm_set_page_unchanged(b->bm_pages[page_nr]);
 
 	if (ctx->flags & BM_AIO_COPY_PAGES) {
-		/* FIXME alloc_page is good enough for now, but actually needs
-		 * to use pre-allocated page pool */
 		void *src, *dest;
-		page = alloc_page(__GFP_HIGHMEM|__GFP_WAIT);
+		page = mempool_alloc(drbd_md_io_page_pool, __GFP_HIGHMEM|__GFP_WAIT);
 		dest = kmap_atomic(page);
 		src = kmap_atomic(b->bm_pages[page_nr]);
 		memcpy(dest, src, PAGE_SIZE);
@@ -981,6 +989,8 @@ static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must
 
 	bio->bi_bdev = mdev->ldev->md_bdev;
 	bio->bi_sector = on_disk_sector;
+	/* bio_add_page of a single page to an empty bio will always succeed,
+	 * according to api.  Do we want to assert that? */
 	bio_add_page(bio, page, len, 0);
 	bio->bi_private = ctx;
 	bio->bi_end_io = bm_async_io_complete;
@@ -999,14 +1009,9 @@ static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must
 /*
  * bm_rw: read/write the whole bitmap from/to its on disk location.
  */
-static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_idx) __must_hold(local)
+static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_writeout_upper_idx) __must_hold(local)
 {
-	struct bm_aio_ctx ctx = {
-		.mdev = mdev,
-		.in_flight = ATOMIC_INIT(1),
-		.done = COMPLETION_INITIALIZER_ONSTACK(ctx.done),
-		.flags = lazy_writeout_upper_idx ? BM_AIO_COPY_PAGES : 0,
-	};
+	struct bm_aio_ctx *ctx;
 	struct drbd_bitmap *b = mdev->bitmap;
 	int num_pages, i, count = 0;
 	unsigned long now;
@@ -1021,7 +1026,27 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
 	 * For lazy writeout, we don't care for ongoing changes to the bitmap,
 	 * as we submit copies of pages anyways.
 	 */
-	if (!ctx.flags)
+
+	ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO);
+	if (!ctx)
+		return -ENOMEM;
+
+	*ctx = (struct bm_aio_ctx) {
+		.mdev = mdev,
+		.in_flight = ATOMIC_INIT(1),
+		.done = 0,
+		.flags = flags,
+		.error = 0,
+		.kref = { ATOMIC_INIT(2) },
+	};
+
+	if (!get_ldev_if_state(mdev, D_ATTACHING)) {  /* put is in bm_aio_ctx_destroy() */
+		dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in bm_rw()\n");
+		kfree(ctx);
+		return -ENODEV;
+	}
+
+	if (!ctx->flags)
 		WARN_ON(!(BM_LOCKED_MASK & b->bm_flags));
 
 	num_pages = b->bm_number_of_pages;
@@ -1046,29 +1071,38 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
 				continue;
 			}
 		}
-		atomic_inc(&ctx.in_flight);
-		bm_page_io_async(&ctx, i, rw);
+		atomic_inc(&ctx->in_flight);
+		bm_page_io_async(ctx, i, rw);
 		++count;
 		cond_resched();
 	}
 
 	/*
-	 * We initialize ctx.in_flight to one to make sure bm_async_io_complete
-	 * will not complete() early, and decrement / test it here.  If there
+	 * We initialize ctx->in_flight to one to make sure bm_async_io_complete
+	 * will not set ctx->done early, and decrement / test it here.  If there
 	 * are still some bios in flight, we need to wait for them here.
+	 * If all IO is done already (or nothing had been submitted), there is
+	 * no need to wait.  Still, we need to put the kref associated with the
+	 * "in_flight reached zero, all done" event.
 	 */
-	if (!atomic_dec_and_test(&ctx.in_flight))
-		wait_for_completion(&ctx.done);
+	if (!atomic_dec_and_test(&ctx->in_flight))
+		wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done);
+	else
+		kref_put(&ctx->kref, &bm_aio_ctx_destroy);
+
 	dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
 			rw == WRITE ? "WRITE" : "READ",
 			count, jiffies - now);
 
-	if (ctx.error) {
+	if (ctx->error) {
 		dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n");
 		drbd_chk_io_error(mdev, 1, true);
-		err = -EIO; /* ctx.error ? */
+		err = -EIO; /* ctx->error ? */
 	}
 
+	if (atomic_read(&ctx->in_flight))
+		err = -EIO; /* Disk failed during IO... */
+
 	now = jiffies;
 	if (rw == WRITE) {
 		drbd_md_flush(mdev);
@@ -1082,6 +1116,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
 	dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
 	     ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
 
+	kref_put(&ctx->kref, &bm_aio_ctx_destroy);
 	return err;
 }
 
@@ -1091,7 +1126,7 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
  */
 int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local)
 {
-	return bm_rw(mdev, READ, 0);
+	return bm_rw(mdev, READ, 0, 0);
 }
 
 /**
@@ -1102,7 +1137,7 @@ int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local)
  */
 int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local)
 {
-	return bm_rw(mdev, WRITE, 0);
+	return bm_rw(mdev, WRITE, 0, 0);
 }
 
 /**
@@ -1112,7 +1147,23 @@ int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local)
  */
 int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local)
 {
-	return bm_rw(mdev, WRITE, upper_idx);
+	return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, upper_idx);
+}
+
+/**
+ * drbd_bm_write_copy_pages() - Write the whole bitmap to its on disk location.
+ * @mdev:	DRBD device.
+ *
+ * Will only write pages that have changed since last IO.
+ * In contrast to drbd_bm_write(), this will copy the bitmap pages
+ * to temporary writeout pages. It is intended to trigger a full write-out
+ * while still allowing the bitmap to change, for example if a resync or online
+ * verify is aborted due to a failed peer disk, while local IO continues, or
+ * pending resync acks are still being processed.
+ */
+int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local)
+{
+	return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, 0);
 }
 
 
@@ -1130,28 +1181,45 @@ int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(l
  */
 int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local)
 {
-	struct bm_aio_ctx ctx = {
+	struct bm_aio_ctx *ctx;
+	int err;
+
+	if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) {
+		dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx);
+		return 0;
+	}
+
+	ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO);
+	if (!ctx)
+		return -ENOMEM;
+
+	*ctx = (struct bm_aio_ctx) {
 		.mdev = mdev,
 		.in_flight = ATOMIC_INIT(1),
-		.done = COMPLETION_INITIALIZER_ONSTACK(ctx.done),
+		.done = 0,
 		.flags = BM_AIO_COPY_PAGES,
+		.error = 0,
+		.kref = { ATOMIC_INIT(2) },
 	};
 
-	if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) {
-		dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx);
-		return 0;
+	if (!get_ldev_if_state(mdev, D_ATTACHING)) {  /* put is in bm_aio_ctx_destroy() */
+		dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in drbd_bm_write_page()\n");
+		kfree(ctx);
+		return -ENODEV;
 	}
 
-	bm_page_io_async(&ctx, idx, WRITE_SYNC);
-	wait_for_completion(&ctx.done);
+	bm_page_io_async(ctx, idx, WRITE_SYNC);
+	wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done);
 
-	if (ctx.error)
+	if (ctx->error)
 		drbd_chk_io_error(mdev, 1, true);
 		/* that should force detach, so the in memory bitmap will be
 		 * gone in a moment as well. */
 
 	mdev->bm_writ_cnt++;
-	return ctx.error;
+	err = atomic_read(&ctx->in_flight) ? -EIO : ctx->error;
+	kref_put(&ctx->kref, &bm_aio_ctx_destroy);
+	return err;
 }
 
 /* NOTE
diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index 8d680562ba73..02f013a073a7 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -712,7 +712,6 @@ struct drbd_request {
 	struct list_head tl_requests; /* ring list in the transfer log */
 	struct bio *master_bio;       /* master bio pointer */
 	unsigned long rq_state; /* see comments above _req_mod() */
-	int seq_num;
 	unsigned long start_time;
 };
 
@@ -851,6 +850,7 @@ enum {
 	NEW_CUR_UUID,		/* Create new current UUID when thawing IO */
 	AL_SUSPENDED,		/* Activity logging is currently suspended. */
 	AHEAD_TO_SYNC_SOURCE,   /* Ahead -> SyncSource queued */
+	STATE_SENT,		/* Do not change state/UUIDs while this is set */
 };
 
 struct drbd_bitmap; /* opaque for drbd_conf */
@@ -862,31 +862,30 @@ enum bm_flag {
 	BM_P_VMALLOCED = 0x10000, /* internal use only, will be masked out */
 
 	/* currently locked for bulk operation */
-	BM_LOCKED_MASK = 0x7,
+	BM_LOCKED_MASK = 0xf,
 
 	/* in detail, that is: */
 	BM_DONT_CLEAR = 0x1,
 	BM_DONT_SET   = 0x2,
 	BM_DONT_TEST  = 0x4,
 
+	/* so we can mark it locked for bulk operation,
+	 * and still allow all non-bulk operations */
+	BM_IS_LOCKED  = 0x8,
+
 	/* (test bit, count bit) allowed (common case) */
-	BM_LOCKED_TEST_ALLOWED = 0x3,
+	BM_LOCKED_TEST_ALLOWED = BM_DONT_CLEAR | BM_DONT_SET | BM_IS_LOCKED,
 
 	/* testing bits, as well as setting new bits allowed, but clearing bits
 	 * would be unexpected.  Used during bitmap receive.  Setting new bits
 	 * requires sending of "out-of-sync" information, though. */
-	BM_LOCKED_SET_ALLOWED = 0x1,
+	BM_LOCKED_SET_ALLOWED = BM_DONT_CLEAR | BM_IS_LOCKED,
 
-	/* clear is not expected while bitmap is locked for bulk operation */
+	/* for drbd_bm_write_copy_pages, everything is allowed,
+	 * only concurrent bulk operations are locked out. */
+	BM_LOCKED_CHANGE_ALLOWED = BM_IS_LOCKED,
 };
 
-
-/* TODO sort members for performance
- * MAYBE group them further */
-
-/* THINK maybe we actually want to use the default "event/%s" worker threads
- * or similar in linux 2.6, which uses per cpu data and threads.
- */
 struct drbd_work_queue {
 	struct list_head q;
 	struct semaphore s; /* producers up it, worker down()s it */
@@ -938,8 +937,7 @@ struct drbd_backing_dev {
 };
 
 struct drbd_md_io {
-	struct drbd_conf *mdev;
-	struct completion event;
+	unsigned int done;
 	int error;
 };
 
@@ -1022,6 +1020,7 @@ struct drbd_conf {
 	struct drbd_tl_epoch *newest_tle;
 	struct drbd_tl_epoch *oldest_tle;
 	struct list_head out_of_sequence_requests;
+	struct list_head barrier_acked_requests;
 	struct hlist_head *tl_hash;
 	unsigned int tl_hash_s;
 
@@ -1056,6 +1055,8 @@ struct drbd_conf {
 	struct crypto_hash *csums_tfm;
 	struct crypto_hash *verify_tfm;
 
+	unsigned long last_reattach_jif;
+	unsigned long last_reconnect_jif;
 	struct drbd_thread receiver;
 	struct drbd_thread worker;
 	struct drbd_thread asender;
@@ -1094,7 +1095,8 @@ struct drbd_conf {
 	wait_queue_head_t ee_wait;
 	struct page *md_io_page;	/* one page buffer for md_io */
 	struct page *md_io_tmpp;	/* for logical_block_size != 512 */
-	struct mutex md_io_mutex;	/* protects the md_io_buffer */
+	struct drbd_md_io md_io;
+	atomic_t md_io_in_use;		/* protects the md_io, md_io_page and md_io_tmpp */
 	spinlock_t al_lock;
 	wait_queue_head_t al_wait;
 	struct lru_cache *act_log;	/* activity log */
@@ -1228,8 +1230,8 @@ extern int drbd_send_uuids(struct drbd_conf *mdev);
 extern int drbd_send_uuids_skip_initial_sync(struct drbd_conf *mdev);
 extern int drbd_gen_and_send_sync_uuid(struct drbd_conf *mdev);
 extern int drbd_send_sizes(struct drbd_conf *mdev, int trigger_reply, enum dds_flags flags);
-extern int _drbd_send_state(struct drbd_conf *mdev);
-extern int drbd_send_state(struct drbd_conf *mdev);
+extern int drbd_send_state(struct drbd_conf *mdev, union drbd_state s);
+extern int drbd_send_current_state(struct drbd_conf *mdev);
 extern int _drbd_send_cmd(struct drbd_conf *mdev, struct socket *sock,
 			enum drbd_packets cmd, struct p_header80 *h,
 			size_t size, unsigned msg_flags);
@@ -1461,6 +1463,7 @@ extern int  drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr);
 extern int  drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local);
 extern int  drbd_bm_read(struct drbd_conf *mdev) __must_hold(local);
 extern int  drbd_bm_write(struct drbd_conf *mdev) __must_hold(local);
+extern int  drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local);
 extern unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev,
 		unsigned long al_enr);
 extern size_t	     drbd_bm_words(struct drbd_conf *mdev);
@@ -1493,11 +1496,38 @@ extern struct kmem_cache *drbd_al_ext_cache;	/* activity log extents */
 extern mempool_t *drbd_request_mempool;
 extern mempool_t *drbd_ee_mempool;
 
-extern struct page *drbd_pp_pool; /* drbd's page pool */
+/* drbd's page pool, used to buffer data received from the peer,
+ * or data requested by the peer.
+ *
+ * This does not have an emergency reserve.
+ *
+ * When allocating from this pool, it first takes pages from the pool.
+ * Only if the pool is depleted will try to allocate from the system.
+ *
+ * The assumption is that pages taken from this pool will be processed,
+ * and given back, "quickly", and then can be recycled, so we can avoid
+ * frequent calls to alloc_page(), and still will be able to make progress even
+ * under memory pressure.
+ */
+extern struct page *drbd_pp_pool;
 extern spinlock_t   drbd_pp_lock;
 extern int	    drbd_pp_vacant;
 extern wait_queue_head_t drbd_pp_wait;
 
+/* We also need a standard (emergency-reserve backed) page pool
+ * for meta data IO (activity log, bitmap).
+ * We can keep it global, as long as it is used as "N pages at a time".
+ * 128 should be plenty, currently we probably can get away with as few as 1.
+ */
+#define DRBD_MIN_POOL_PAGES	128
+extern mempool_t *drbd_md_io_page_pool;
+
+/* We also need to make sure we get a bio
+ * when we need it for housekeeping purposes */
+extern struct bio_set *drbd_md_io_bio_set;
+/* to allocate from that set */
+extern struct bio *bio_alloc_drbd(gfp_t gfp_mask);
+
 extern rwlock_t global_state_lock;
 
 extern struct drbd_conf *drbd_new_device(unsigned int minor);
@@ -1536,8 +1566,12 @@ extern void resume_next_sg(struct drbd_conf *mdev);
 extern void suspend_other_sg(struct drbd_conf *mdev);
 extern int drbd_resync_finished(struct drbd_conf *mdev);
 /* maybe rather drbd_main.c ? */
+extern void *drbd_md_get_buffer(struct drbd_conf *mdev);
+extern void drbd_md_put_buffer(struct drbd_conf *mdev);
 extern int drbd_md_sync_page_io(struct drbd_conf *mdev,
-		struct drbd_backing_dev *bdev, sector_t sector, int rw);
+				struct drbd_backing_dev *bdev, sector_t sector, int rw);
+extern void wait_until_done_or_disk_failure(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
+					    unsigned int *done);
 extern void drbd_ov_oos_found(struct drbd_conf*, sector_t, int);
 extern void drbd_rs_controller_reset(struct drbd_conf *mdev);
 
@@ -1754,19 +1788,6 @@ static inline struct page *page_chain_next(struct page *page)
 #define page_chain_for_each_safe(page, n) \
 	for (; page && ({ n = page_chain_next(page); 1; }); page = n)
 
-static inline int drbd_bio_has_active_page(struct bio *bio)
-{
-	struct bio_vec *bvec;
-	int i;
-
-	__bio_for_each_segment(bvec, bio, i, 0) {
-		if (page_count(bvec->bv_page) > 1)
-			return 1;
-	}
-
-	return 0;
-}
-
 static inline int drbd_ee_has_active_page(struct drbd_epoch_entry *e)
 {
 	struct page *page = e->pages;
@@ -1777,7 +1798,6 @@ static inline int drbd_ee_has_active_page(struct drbd_epoch_entry *e)
 	return 0;
 }
 
-
 static inline void drbd_state_lock(struct drbd_conf *mdev)
 {
 	wait_event(mdev->misc_wait,
@@ -2230,7 +2250,7 @@ static inline void drbd_get_syncer_progress(struct drbd_conf *mdev,
 		 * Note: currently we don't support such large bitmaps on 32bit
 		 * arch anyways, but no harm done to be prepared for it here.
 		 */
-		unsigned int shift = mdev->rs_total >= (1ULL << 32) ? 16 : 10;
+		unsigned int shift = mdev->rs_total > UINT_MAX ? 16 : 10;
 		unsigned long left = *bits_left >> shift;
 		unsigned long total = 1UL + (mdev->rs_total >> shift);
 		unsigned long tmp = 1000UL - left * 1000UL/total;
@@ -2306,12 +2326,12 @@ static inline int drbd_state_is_stable(struct drbd_conf *mdev)
 	case D_OUTDATED:
 	case D_CONSISTENT:
 	case D_UP_TO_DATE:
+	case D_FAILED:
 		/* disk state is stable as well. */
 		break;
 
 	/* no new io accepted during tansitional states */
 	case D_ATTACHING:
-	case D_FAILED:
 	case D_NEGOTIATING:
 	case D_UNKNOWN:
 	case D_MASK:
diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c
index 211fc44f84be..920ede2829d6 100644
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -139,6 +139,8 @@ struct kmem_cache *drbd_bm_ext_cache;	/* bitmap extents */
 struct kmem_cache *drbd_al_ext_cache;	/* activity log extents */
 mempool_t *drbd_request_mempool;
 mempool_t *drbd_ee_mempool;
+mempool_t *drbd_md_io_page_pool;
+struct bio_set *drbd_md_io_bio_set;
 
 /* I do not use a standard mempool, because:
    1) I want to hand out the pre-allocated objects first.
@@ -159,7 +161,24 @@ static const struct block_device_operations drbd_ops = {
 	.release = drbd_release,
 };
 
-#define ARRY_SIZE(A) (sizeof(A)/sizeof(A[0]))
+static void bio_destructor_drbd(struct bio *bio)
+{
+	bio_free(bio, drbd_md_io_bio_set);
+}
+
+struct bio *bio_alloc_drbd(gfp_t gfp_mask)
+{
+	struct bio *bio;
+
+	if (!drbd_md_io_bio_set)
+		return bio_alloc(gfp_mask, 1);
+
+	bio = bio_alloc_bioset(gfp_mask, 1, drbd_md_io_bio_set);
+	if (!bio)
+		return NULL;
+	bio->bi_destructor = bio_destructor_drbd;
+	return bio;
+}
 
 #ifdef __CHECKER__
 /* When checking with sparse, and this is an inline function, sparse will
@@ -208,6 +227,7 @@ static int tl_init(struct drbd_conf *mdev)
 	mdev->oldest_tle = b;
 	mdev->newest_tle = b;
 	INIT_LIST_HEAD(&mdev->out_of_sequence_requests);
+	INIT_LIST_HEAD(&mdev->barrier_acked_requests);
 
 	mdev->tl_hash = NULL;
 	mdev->tl_hash_s = 0;
@@ -246,9 +266,7 @@ void _tl_add_barrier(struct drbd_conf *mdev, struct drbd_tl_epoch *new)
 	new->n_writes = 0;
 
 	newest_before = mdev->newest_tle;
-	/* never send a barrier number == 0, because that is special-cased
-	 * when using TCQ for our write ordering code */
-	new->br_number = (newest_before->br_number+1) ?: 1;
+	new->br_number = newest_before->br_number+1;
 	if (mdev->newest_tle != new) {
 		mdev->newest_tle->next = new;
 		mdev->newest_tle = new;
@@ -311,7 +329,7 @@ void tl_release(struct drbd_conf *mdev, unsigned int barrier_nr,
 	   These have been list_move'd to the out_of_sequence_requests list in
 	   _req_mod(, barrier_acked) above.
 	   */
-	list_del_init(&b->requests);
+	list_splice_init(&b->requests, &mdev->barrier_acked_requests);
 
 	nob = b->next;
 	if (test_and_clear_bit(CREATE_BARRIER, &mdev->flags)) {
@@ -411,6 +429,23 @@ static void _tl_restart(struct drbd_conf *mdev, enum drbd_req_event what)
 		b = tmp;
 		list_splice(&carry_reads, &b->requests);
 	}
+
+	/* Actions operating on the disk state, also want to work on
+	   requests that got barrier acked. */
+	switch (what) {
+	case fail_frozen_disk_io:
+	case restart_frozen_disk_io:
+		list_for_each_safe(le, tle, &mdev->barrier_acked_requests) {
+			req = list_entry(le, struct drbd_request, tl_requests);
+			_req_mod(req, what);
+		}
+
+	case connection_lost_while_pending:
+	case resend:
+		break;
+	default:
+		dev_err(DEV, "what = %d in _tl_restart()\n", what);
+	}
 }
 
 
@@ -458,6 +493,38 @@ void tl_restart(struct drbd_conf *mdev, enum drbd_req_event what)
 }
 
 /**
+ * tl_abort_disk_io() - Abort disk I/O for all requests for a certain mdev in the TL
+ * @mdev:	DRBD device.
+ */
+void tl_abort_disk_io(struct drbd_conf *mdev)
+{
+	struct drbd_tl_epoch *b;
+	struct list_head *le, *tle;
+	struct drbd_request *req;
+
+	spin_lock_irq(&mdev->req_lock);
+	b = mdev->oldest_tle;
+	while (b) {
+		list_for_each_safe(le, tle, &b->requests) {
+			req = list_entry(le, struct drbd_request, tl_requests);
+			if (!(req->rq_state & RQ_LOCAL_PENDING))
+				continue;
+			_req_mod(req, abort_disk_io);
+		}
+		b = b->next;
+	}
+
+	list_for_each_safe(le, tle, &mdev->barrier_acked_requests) {
+		req = list_entry(le, struct drbd_request, tl_requests);
+		if (!(req->rq_state & RQ_LOCAL_PENDING))
+			continue;
+		_req_mod(req, abort_disk_io);
+	}
+
+	spin_unlock_irq(&mdev->req_lock);
+}
+
+/**
  * cl_wide_st_chg() - true if the state change is a cluster wide one
  * @mdev:	DRBD device.
  * @os:		old (current) state.
@@ -470,7 +537,7 @@ static int cl_wide_st_chg(struct drbd_conf *mdev,
 		 ((os.role != R_PRIMARY && ns.role == R_PRIMARY) ||
 		  (os.conn != C_STARTING_SYNC_T && ns.conn == C_STARTING_SYNC_T) ||
 		  (os.conn != C_STARTING_SYNC_S && ns.conn == C_STARTING_SYNC_S) ||
-		  (os.disk != D_DISKLESS && ns.disk == D_DISKLESS))) ||
+		  (os.disk != D_FAILED && ns.disk == D_FAILED))) ||
 		(os.conn >= C_CONNECTED && ns.conn == C_DISCONNECTING) ||
 		(os.conn == C_CONNECTED && ns.conn == C_VERIFY_S);
 }
@@ -509,8 +576,16 @@ static enum drbd_state_rv is_valid_state(struct drbd_conf *, union drbd_state);
 static enum drbd_state_rv is_valid_state_transition(struct drbd_conf *,
 						    union drbd_state,
 						    union drbd_state);
+enum sanitize_state_warnings {
+	NO_WARNING,
+	ABORTED_ONLINE_VERIFY,
+	ABORTED_RESYNC,
+	CONNECTION_LOST_NEGOTIATING,
+	IMPLICITLY_UPGRADED_DISK,
+	IMPLICITLY_UPGRADED_PDSK,
+};
 static union drbd_state sanitize_state(struct drbd_conf *mdev, union drbd_state os,
-				       union drbd_state ns, const char **warn_sync_abort);
+				       union drbd_state ns, enum sanitize_state_warnings *warn);
 int drbd_send_state_req(struct drbd_conf *,
 			union drbd_state, union drbd_state);
 
@@ -785,6 +860,13 @@ is_valid_state_transition(struct drbd_conf *mdev, union drbd_state ns,
 	if (ns.conn == os.conn && ns.conn == C_WF_REPORT_PARAMS)
 		rv = SS_IN_TRANSIENT_STATE;
 
+	/* While establishing a connection only allow cstate to change.
+	   Delay/refuse role changes, detach attach etc... */
+	if (test_bit(STATE_SENT, &mdev->flags) &&
+	    !(os.conn == C_WF_REPORT_PARAMS ||
+	      (ns.conn == C_WF_REPORT_PARAMS && os.conn == C_WF_CONNECTION)))
+		rv = SS_IN_TRANSIENT_STATE;
+
 	if ((ns.conn == C_VERIFY_S || ns.conn == C_VERIFY_T) && os.conn < C_CONNECTED)
 		rv = SS_NEED_CONNECTION;
 
@@ -803,6 +885,21 @@ is_valid_state_transition(struct drbd_conf *mdev, union drbd_state ns,
 	return rv;
 }
 
+static void print_sanitize_warnings(struct drbd_conf *mdev, enum sanitize_state_warnings warn)
+{
+	static const char *msg_table[] = {
+		[NO_WARNING] = "",
+		[ABORTED_ONLINE_VERIFY] = "Online-verify aborted.",
+		[ABORTED_RESYNC] = "Resync aborted.",
+		[CONNECTION_LOST_NEGOTIATING] = "Connection lost while negotiating, no data!",
+		[IMPLICITLY_UPGRADED_DISK] = "Implicitly upgraded disk",
+		[IMPLICITLY_UPGRADED_PDSK] = "Implicitly upgraded pdsk",
+	};
+
+	if (warn != NO_WARNING)
+		dev_warn(DEV, "%s\n", msg_table[warn]);
+}
+
 /**
  * sanitize_state() - Resolves implicitly necessary additional changes to a state transition
  * @mdev:	DRBD device.
@@ -814,11 +911,14 @@ is_valid_state_transition(struct drbd_conf *mdev, union drbd_state ns,
  * to D_UNKNOWN. This rule and many more along those lines are in this function.
  */
 static union drbd_state