Merge branch 'for-2.6.39/drivers' of git://git.kernel.dk/linux-2.6-block

* 'for-2.6.39/drivers' of git://git.kernel.dk/linux-2.6-block: (122 commits)
  cciss: fix lost command issue
  drbd: need include for bitops functions declarations
  Revert "cciss: Add missing allocation in scsi_cmd_stack_setup and  corresponding deallocation"
  cciss: fix missed command status value CMD_UNABORTABLE
  cciss: remove unnecessary casts
  cciss: Mask off error bits of c->busaddr in cmd_special_free when calling pci_free_consistent
  cciss: Inform controller we are using 32-bit tags.
  cciss: hoist tag masking out of loop
  cciss: Add missing allocation in scsi_cmd_stack_setup and  corresponding deallocation
  cciss: export resettable host attribute
  drbd: drop code present under #ifdef which is relevant to 2.6.28 and below
  drbd: Fixed handling of read errors on a 'VerifyS' node
  drbd: Fixed handling of read errors on a 'VerifyT' node
  drbd: Implemented real timeout checking for request processing time
  drbd: Remove unused function atodb_endio()
  drbd: improve log message if received sector offset exceeds local capacity
  drbd: kill dead code
  drbd: don't BUG_ON, if bio_add_page of a single page to an empty bio fails
  drbd: Removed left over, now wrong comments
  drbd: serialize admin requests for new verify run with pending bitmap io
  ...

12 files changed, 2132 insertions, 1376 deletions

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index aca302492ff..2a1642bc451 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -92,7 +92,7 @@ static int _drbd_md_sync_page_io(struct drbd_conf *mdev,
 	bio->bi_end_io = drbd_md_io_complete;
 	bio->bi_rw = rw;
 
-	if (FAULT_ACTIVE(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD))
+	if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD))
 		bio_endio(bio, -EIO);
 	else
 		submit_bio(rw, bio);
@@ -176,13 +176,17 @@ static struct lc_element *_al_get(struct drbd_conf *mdev, unsigned int enr)
 	struct lc_element *al_ext;
 	struct lc_element *tmp;
 	unsigned long     al_flags = 0;
+	int wake;
 
 	spin_lock_irq(&mdev->al_lock);
 	tmp = lc_find(mdev->resync, enr/AL_EXT_PER_BM_SECT);
 	if (unlikely(tmp != NULL)) {
 		struct bm_extent  *bm_ext = lc_entry(tmp, struct bm_extent, lce);
 		if (test_bit(BME_NO_WRITES, &bm_ext->flags)) {
+			wake = !test_and_set_bit(BME_PRIORITY, &bm_ext->flags);
 			spin_unlock_irq(&mdev->al_lock);
+			if (wake)
+				wake_up(&mdev->al_wait);
 			return NULL;
 		}
 	}
@@ -258,6 +262,33 @@ void drbd_al_complete_io(struct drbd_conf *mdev, sector_t sector)
 	spin_unlock_irqrestore(&mdev->al_lock, flags);
 }
 
+#if (PAGE_SHIFT + 3) < (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT)
+/* Currently BM_BLOCK_SHIFT, BM_EXT_SHIFT and AL_EXTENT_SHIFT
+ * are still coupled, or assume too much about their relation.
+ * Code below will not work if this is violated.
+ * Will be cleaned up with some followup patch.
+ */
+# error FIXME
+#endif
+
+static unsigned int al_extent_to_bm_page(unsigned int al_enr)
+{
+	return al_enr >>
+		/* bit to page */
+		((PAGE_SHIFT + 3) -
+		/* al extent number to bit */
+		 (AL_EXTENT_SHIFT - BM_BLOCK_SHIFT));
+}
+
+static unsigned int rs_extent_to_bm_page(unsigned int rs_enr)
+{
+	return rs_enr >>
+		/* bit to page */
+		((PAGE_SHIFT + 3) -
+		/* al extent number to bit */
+		 (BM_EXT_SHIFT - BM_BLOCK_SHIFT));
+}
+
 int
 w_al_write_transaction(struct drbd_conf *mdev, struct drbd_work *w, int unused)
 {
@@ -285,7 +316,7 @@ w_al_write_transaction(struct drbd_conf *mdev, struct drbd_work *w, int unused)
 	 * For now, we must not write the transaction,
 	 * if we cannot write out the bitmap of the evicted extent. */
 	if (mdev->state.conn < C_CONNECTED && evicted != LC_FREE)
-		drbd_bm_write_sect(mdev, evicted/AL_EXT_PER_BM_SECT);
+		drbd_bm_write_page(mdev, al_extent_to_bm_page(evicted));
 
 	/* The bitmap write may have failed, causing a state change. */
 	if (mdev->state.disk < D_INCONSISTENT) {
@@ -334,7 +365,7 @@ w_al_write_transaction(struct drbd_conf *mdev, struct drbd_work *w, int unused)
 		+ mdev->ldev->md.al_offset + mdev->al_tr_pos;
 
 	if (!drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE))
-		drbd_chk_io_error(mdev, 1, TRUE);
+		drbd_chk_io_error(mdev, 1, true);
 
 	if (++mdev->al_tr_pos >
 	    div_ceil(mdev->act_log->nr_elements, AL_EXTENTS_PT))
@@ -511,225 +542,6 @@ cancel:
 	return 1;
 }
 
-static void atodb_endio(struct bio *bio, int error)
-{
-	struct drbd_atodb_wait *wc = bio->bi_private;
-	struct drbd_conf *mdev = wc->mdev;
-	struct page *page;
-	int uptodate = bio_flagged(bio, BIO_UPTODATE);
-
-	/* strange behavior of some lower level drivers...
-	 * fail the request by clearing the uptodate flag,
-	 * but do not return any error?! */
-	if (!error && !uptodate)
-		error = -EIO;
-
-	drbd_chk_io_error(mdev, error, TRUE);
-	if (error && wc->error == 0)
-		wc->error = error;
-
-	if (atomic_dec_and_test(&wc->count))
-		complete(&wc->io_done);
-
-	page = bio->bi_io_vec[0].bv_page;
-	put_page(page);
-	bio_put(bio);
-	mdev->bm_writ_cnt++;
-	put_ldev(mdev);
-}
-
-/* sector to word */
-#define S2W(s)	((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
-
-/* activity log to on disk bitmap -- prepare bio unless that sector
- * is already covered by previously prepared bios */
-static int atodb_prepare_unless_covered(struct drbd_conf *mdev,
-					struct bio **bios,
-					unsigned int enr,
-					struct drbd_atodb_wait *wc) __must_hold(local)
-{
-	struct bio *bio;
-	struct page *page;
-	sector_t on_disk_sector;
-	unsigned int page_offset = PAGE_SIZE;
-	int offset;
-	int i = 0;
-	int err = -ENOMEM;
-
-	/* We always write aligned, full 4k blocks,
-	 * so we can ignore the logical_block_size (for now) */
-	enr &= ~7U;
-	on_disk_sector = enr + mdev->ldev->md.md_offset
-			     + mdev->ldev->md.bm_offset;
-
-	D_ASSERT(!(on_disk_sector & 7U));
-
-	/* Check if that enr is already covered by an already created bio.
-	 * Caution, bios[] is not NULL terminated,
-	 * but only initialized to all NULL.
-	 * For completely scattered activity log,
-	 * the last invocation iterates over all bios,
-	 * and finds the last NULL entry.
-	 */
-	while ((bio = bios[i])) {
-		if (bio->bi_sector == on_disk_sector)
-			return 0;
-		i++;
-	}
-	/* bios[i] == NULL, the next not yet used slot */
-
-	/* GFP_KERNEL, we are not in the write-out path */
-	bio = bio_alloc(GFP_KERNEL, 1);
-	if (bio == NULL)
-		return -ENOMEM;
-
-	if (i > 0) {
-		const struct bio_vec *prev_bv = bios[i-1]->bi_io_vec;
-		page_offset = prev_bv->bv_offset + prev_bv->bv_len;
-		page = prev_bv->bv_page;
-	}
-	if (page_offset == PAGE_SIZE) {
-		page = alloc_page(__GFP_HIGHMEM);
-		if (page == NULL)
-			goto out_bio_put;
-		page_offset = 0;
-	} else {
-		get_page(page);
-	}
-
-	offset = S2W(enr);
-	drbd_bm_get_lel(mdev, offset,
-			min_t(size_t, S2W(8), drbd_bm_words(mdev) - offset),
-			kmap(page) + page_offset);
-	kunmap(page);
-
-	bio->bi_private = wc;
-	bio->bi_end_io = atodb_endio;
-	bio->bi_bdev = mdev->ldev->md_bdev;
-	bio->bi_sector = on_disk_sector;
-
-	if (bio_add_page(bio, page, 4096, page_offset) != 4096)
-		goto out_put_page;
-
-	atomic_inc(&wc->count);
-	/* we already know that we may do this...
-	 * get_ldev_if_state(mdev,D_ATTACHING);
-	 * just get the extra reference, so that the local_cnt reflects
-	 * the number of pending IO requests DRBD at its backing device.
-	 */
-	atomic_inc(&mdev->local_cnt);
-
-	bios[i] = bio;
-
-	return 0;
-
-out_put_page:
-	err = -EINVAL;
-	put_page(page);
-out_bio_put:
-	bio_put(bio);
-	return err;
-}
-
-/**
- * drbd_al_to_on_disk_bm() -  * Writes bitmap parts covered by active AL extents
- * @mdev:	DRBD device.
- *
- * Called when we detach (unconfigure) local storage,
- * or when we go from R_PRIMARY to R_SECONDARY role.
- */
-void drbd_al_to_on_disk_bm(struct drbd_conf *mdev)
-{
-	int i, nr_elements;
-	unsigned int enr;
-	struct bio **bios;
-	struct drbd_atodb_wait wc;
-
-	ERR_IF (!get_ldev_if_state(mdev, D_ATTACHING))
-		return; /* sorry, I don't have any act_log etc... */
-
-	wait_event(mdev->al_wait, lc_try_lock(mdev->act_log));
-
-	nr_elements = mdev->act_log->nr_elements;
-
-	/* GFP_KERNEL, we are not in anyone's write-out path */
-	bios = kzalloc(sizeof(struct bio *) * nr_elements, GFP_KERNEL);
-	if (!bios)
-		goto submit_one_by_one;
-
-	atomic_set(&wc.count, 0);
-	init_completion(&wc.io_done);
-	wc.mdev = mdev;
-	wc.error = 0;
-
-	for (i = 0; i < nr_elements; i++) {
-		enr = lc_element_by_index(mdev->act_log, i)->lc_number;
-		if (enr == LC_FREE)
-			continue;
-		/* next statement also does atomic_inc wc.count and local_cnt */
-		if (atodb_prepare_unless_covered(mdev, bios,
-						enr/AL_EXT_PER_BM_SECT,
-						&wc))
-			goto free_bios_submit_one_by_one;
-	}
-
-	/* unnecessary optimization? */
-	lc_unlock(mdev->act_log);
-	wake_up(&mdev->al_wait);
-
-	/* all prepared, submit them */
-	for (i = 0; i < nr_elements; i++) {
-		if (bios[i] == NULL)
-			break;
-		if (FAULT_ACTIVE(mdev, DRBD_FAULT_MD_WR)) {
-			bios[i]->bi_rw = WRITE;
-			bio_endio(bios[i], -EIO);
-		} else {
-			submit_bio(WRITE, bios[i]);
-		}
-	}
-
-	/* always (try to) flush bitmap to stable storage */
-	drbd_md_flush(mdev);
-
-	/* In case we did not submit a single IO do not wait for
-	 * them to complete. ( Because we would wait forever here. )
-	 *
-	 * In case we had IOs and they are already complete, there
-	 * is not point in waiting anyways.
-	 * Therefore this if () ... */
-	if (atomic_read(&wc.count))
-		wait_for_completion(&wc.io_done);
-
-	put_ldev(mdev);
-
-	kfree(bios);
-	return;
-
- free_bios_submit_one_by_one:
-	/* free everything by calling the endio callback directly. */
-	for (i = 0; i < nr_elements && bios[i]; i++)
-		bio_endio(bios[i], 0);
-
-	kfree(bios);
-
- submit_one_by_one:
-	dev_warn(DEV, "Using the slow drbd_al_to_on_disk_bm()\n");
-
-	for (i = 0; i < mdev->act_log->nr_elements; i++) {
-		enr = lc_element_by_index(mdev->act_log, i)->lc_number;
-		if (enr == LC_FREE)
-			continue;
-		/* Really slow: if we have al-extents 16..19 active,
-		 * sector 4 will be written four times! Synchronous! */
-		drbd_bm_write_sect(mdev, enr/AL_EXT_PER_BM_SECT);
-	}
-
-	lc_unlock(mdev->act_log);
-	wake_up(&mdev->al_wait);
-	put_ldev(mdev);
-}
-
 /**
  * drbd_al_apply_to_bm() - Sets the bitmap to diry(1) where covered ba active AL extents
  * @mdev:	DRBD device.
@@ -809,7 +621,7 @@ static int w_update_odbm(struct drbd_conf *mdev, struct drbd_work *w, int unused
 		return 1;
 	}
 
-	drbd_bm_write_sect(mdev, udw->enr);
+	drbd_bm_write_page(mdev, rs_extent_to_bm_page(udw->enr));
 	put_ldev(mdev);
 
 	kfree(udw);
@@ -889,7 +701,6 @@ static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector,
 				dev_warn(DEV, "Kicking resync_lru element enr=%u "
 				     "out with rs_failed=%d\n",
 				     ext->lce.lc_number, ext->rs_failed);
-				set_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags);
 			}
 			ext->rs_left = rs_left;
 			ext->rs_failed = success ? 0 : count;
@@ -908,7 +719,6 @@ static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector,
 				drbd_queue_work_front(&mdev->data.work, &udw->w);
 			} else {
 				dev_warn(DEV, "Could not kmalloc an udw\n");
-				set_bit(WRITE_BM_AFTER_RESYNC, &mdev->flags);
 			}
 		}
 	} else {
@@ -919,6 +729,22 @@ static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector,
 	}
 }
 
+void drbd_advance_rs_marks(struct drbd_conf *mdev, unsigned long still_to_go)
+{
+	unsigned long now = jiffies;
+	unsigned long last = mdev->rs_mark_time[mdev->rs_last_mark];
+	int next = (mdev->rs_last_mark + 1) % DRBD_SYNC_MARKS;
+	if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) {
+		if (mdev->rs_mark_left[mdev->rs_last_mark] != still_to_go &&
+		    mdev->state.conn != C_PAUSED_SYNC_T &&
+		    mdev->state.conn != C_PAUSED_SYNC_S) {
+			mdev->rs_mark_time[next] = now;
+			mdev->rs_mark_left[next] = still_to_go;
+			mdev->rs_last_mark = next;
+		}
+	}
+}
+
 /* clear the bit corresponding to the piece of storage in question:
  * size byte of data starting from sector.  Only clear a bits of the affected
  * one ore more _aligned_ BM_BLOCK_SIZE blocks.
@@ -936,7 +762,7 @@ void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, int size,
 	int wake_up = 0;
 	unsigned long flags;
 
-	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
+	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "drbd_set_in_sync: sector=%llus size=%d nonsense!\n",
 				(unsigned long long)sector, size);
 		return;
@@ -969,21 +795,9 @@ void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, int size,
 	 */
 	count = drbd_bm_clear_bits(mdev, sbnr, ebnr);
 	if (count && get_ldev(mdev)) {
-		unsigned long now = jiffies;
-		unsigned long last = mdev->rs_mark_time[mdev->rs_last_mark];
-		int next = (mdev->rs_last_mark + 1) % DRBD_SYNC_MARKS;
-		if (time_after_eq(now, last + DRBD_SYNC_MARK_STEP)) {
-			unsigned long tw = drbd_bm_total_weight(mdev);
-			if (mdev->rs_mark_left[mdev->rs_last_mark] != tw &&
-			    mdev->state.conn != C_PAUSED_SYNC_T &&
-			    mdev->state.conn != C_PAUSED_SYNC_S) {
-				mdev->rs_mark_time[next] = now;
-				mdev->rs_mark_left[next] = tw;
-				mdev->rs_last_mark = next;
-			}
-		}
+		drbd_advance_rs_marks(mdev, drbd_bm_total_weight(mdev));
 		spin_lock_irqsave(&mdev->al_lock, flags);
-		drbd_try_clear_on_disk_bm(mdev, sector, count, TRUE);
+		drbd_try_clear_on_disk_bm(mdev, sector, count, true);
 		spin_unlock_irqrestore(&mdev->al_lock, flags);
 
 		/* just wake_up unconditional now, various lc_chaged(),
@@ -998,27 +812,27 @@ void __drbd_set_in_sync(struct drbd_conf *mdev, sector_t sector, int size,
 /*
  * this is intended to set one request worth of data out of sync.
  * affects at least 1 bit,
- * and at most 1+DRBD_MAX_SEGMENT_SIZE/BM_BLOCK_SIZE bits.
+ * and at most 1+DRBD_MAX_BIO_SIZE/BM_BLOCK_SIZE bits.
  *
  * called by tl_clear and drbd_send_dblock (==drbd_make_request).
  * so this can be _any_ process.
  */
-void __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size,
+int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size,
 			    const char *file, const unsigned int line)
 {
 	unsigned long sbnr, ebnr, lbnr, flags;
 	sector_t esector, nr_sectors;
-	unsigned int enr, count;
+	unsigned int enr, count = 0;
 	struct lc_element *e;
 
-	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
+	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "sector: %llus, size: %d\n",
 			(unsigned long long)sector, size);
-		return;
+		return 0;
 	}
 
 	if (!get_ldev(mdev))
-		return; /* no disk, no metadata, no bitmap to set bits in */
+		return 0; /* no disk, no metadata, no bitmap to set bits in */
 
 	nr_sectors = drbd_get_capacity(mdev->this_bdev);
 	esector = sector + (size >> 9) - 1;
@@ -1048,6 +862,8 @@ void __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size,
 
 out:
 	put_ldev(mdev);
+
+	return count;
 }
 
 static
@@ -1128,7 +944,10 @@ int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector)
 	unsigned int enr = BM_SECT_TO_EXT(sector);
 	struct bm_extent *bm_ext;
 	int i, sig;
+	int sa = 200; /* Step aside 200 times, then grab the extent and let app-IO wait.
+			 200 times -> 20 seconds. */
 
+retry:
 	sig = wait_event_interruptible(mdev->al_wait,
 			(bm_ext = _bme_get(mdev, enr)));
 	if (sig)
@@ -1139,16 +958,25 @@ int drbd_rs_begin_io(struct drbd_conf *mdev, sector_t sector)
 
 	for (i = 0; i < AL_EXT_PER_BM_SECT; i++) {
 		sig = wait_event_interruptible(mdev->al_wait,
-				!_is_in_al(mdev, enr * AL_EXT_PER_BM_SECT + i));
-		if (sig) {
+					       !_is_in_al(mdev, enr * AL_EXT_PER_BM_SECT + i) ||
+					       test_bit(BME_PRIORITY, &bm_ext->flags));
+
+		if (sig || (test_bit(BME_PRIORITY, &bm_ext->flags) && sa)) {
 			spin_lock_irq(&mdev->al_lock);
 			if (lc_put(mdev->resync, &bm_ext->lce) == 0) {
-				clear_bit(BME_NO_WRITES, &bm_ext->flags);
+				bm_ext->flags = 0; /* clears BME_NO_WRITES and eventually BME_PRIORITY */
 				mdev->resync_locked--;
 				wake_up(&mdev->al_wait);
 			}
 			spin_unlock_irq(&mdev->al_lock);
-			return -EINTR;
+			if (sig)
+				return -EINTR;
+			if (schedule_timeout_interruptible(HZ/10))
+				return -EINTR;
+			if (sa && --sa == 0)
+				dev_warn(DEV,"drbd_rs_begin_io() stepped aside for 20sec."
+					 "Resync stalled?\n");
+			goto retry;
 		}
 	}
 	set_bit(BME_LOCKED, &bm_ext->flags);
@@ -1291,8 +1119,7 @@ void drbd_rs_complete_io(struct drbd_conf *mdev, sector_t sector)
 	}
 
 	if (lc_put(mdev->resync, &bm_ext->lce) == 0) {
-		clear_bit(BME_LOCKED, &bm_ext->flags);
-		clear_bit(BME_NO_WRITES, &bm_ext->flags);
+		bm_ext->flags = 0; /* clear BME_LOCKED, BME_NO_WRITES and BME_PRIORITY */
 		mdev->resync_locked--;
 		wake_up(&mdev->al_wait);
 	}
@@ -1383,7 +1210,7 @@ void drbd_rs_failed_io(struct drbd_conf *mdev, sector_t sector, int size)
 	sector_t esector, nr_sectors;
 	int wake_up = 0;
 
-	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_SEGMENT_SIZE) {
+	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "drbd_rs_failed_io: sector=%llus size=%d nonsense!\n",
 				(unsigned long long)sector, size);
 		return;
@@ -1420,7 +1247,7 @@ void drbd_rs_failed_io(struct drbd_conf *mdev, sector_t sector, int size)
 		mdev->rs_failed += count;
 
 		if (get_ldev(mdev)) {
-			drbd_try_clear_on_disk_bm(mdev, sector, count, FALSE);
+			drbd_try_clear_on_disk_bm(mdev, sector, count, false);
 			put_ldev(mdev);
 		}
 
diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index 0645ca829a9..f0ae63d2df6 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -28,18 +28,58 @@
 #include <linux/drbd.h>
 #include <linux/slab.h>
 #include <asm/kmap_types.h>
+
+#include <asm-generic/bitops/le.h>
+
 #include "drbd_int.h"
 
+
 /* OPAQUE outside this file!
  * interface defined in drbd_int.h
 
  * convention:
  * function name drbd_bm_... => used elsewhere, "public".
  * function name      bm_... => internal to implementation, "private".
+ */
+
+
+/*
+ * LIMITATIONS:
+ * We want to support >= peta byte of backend storage, while for now still using
+ * a granularity of one bit per 4KiB of storage.
+ * 1 << 50		bytes backend storage (1 PiB)
+ * 1 << (50 - 12)	bits needed
+ *	38 --> we need u64 to index and count bits
+ * 1 << (38 - 3)	bitmap bytes needed
+ *	35 --> we still need u64 to index and count bytes
+ *			(that's 32 GiB of bitmap for 1 PiB storage)
+ * 1 << (35 - 2)	32bit longs needed
+ *	33 --> we'd even need u64 to index and count 32bit long words.
+ * 1 << (35 - 3)	64bit longs needed
+ *	32 --> we could get away with a 32bit unsigned int to index and count
+ *	64bit long words, but I rather stay with unsigned long for now.
+ *	We probably should neither count nor point to bytes or long words
+ *	directly, but either by bitnumber, or by page index and offset.
+ * 1 << (35 - 12)
+ *	22 --> we need that much 4KiB pages of bitmap.
+ *	1 << (22 + 3) --> on a 64bit arch,
+ *	we need 32 MiB to store the array of page pointers.
+ *
+ * Because I'm lazy, and because the resulting patch was too large, too ugly
+ * and still incomplete, on 32bit we still "only" support 16 TiB (minus some),
+ * (1 << 32) bits * 4k storage.
+ *
 
- * Note that since find_first_bit returns int, at the current granularity of
- * the bitmap (4KB per byte), this implementation "only" supports up to
- * 1<<(32+12) == 16 TB...
+ * bitmap storage and IO:
+ *	Bitmap is stored little endian on disk, and is kept little endian in
+ *	core memory. Currently we still hold the full bitmap in core as long
+ *	as we are "attached" to a local disk, which at 32 GiB for 1PiB storage
+ *	seems excessive.
+ *
+ *	We plan to reduce the amount of in-core bitmap pages by pageing them in
+ *	and out against their on-disk location as necessary, but need to make
+ *	sure we don't cause too much meta data IO, and must not deadlock in
+ *	tight memory situations. This needs some more work.
  */
 
 /*
@@ -55,13 +95,9 @@
 struct drbd_bitmap {
 	struct page **bm_pages;
 	spinlock_t bm_lock;
-	/* WARNING unsigned long bm_*:
-	 * 32bit number of bit offset is just enough for 512 MB bitmap.
-	 * it will blow up if we make the bitmap bigger...
-	 * not that it makes much sense to have a bitmap that large,
-	 * rather change the granularity to 16k or 64k or something.
-	 * (that implies other problems, however...)
-	 */
+
+	/* see LIMITATIONS: above */
+
 	unsigned long bm_set;       /* nr of set bits; THINK maybe atomic_t? */
 	unsigned long bm_bits;
 	size_t   bm_words;
@@ -69,29 +105,18 @@ struct drbd_bitmap {
 	sector_t bm_dev_capacity;
 	struct mutex bm_change; /* serializes resize operations */
 
-	atomic_t bm_async_io;
-	wait_queue_head_t bm_io_wait;
+	wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */
 
-	unsigned long  bm_flags;
+	enum bm_flag bm_flags;
 
 	/* debugging aid, in case we are still racy somewhere */
 	char          *bm_why;
 	struct task_struct *bm_task;
 };
 
-/* definition of bits in bm_flags */
-#define BM_LOCKED       0
-#define BM_MD_IO_ERROR  1
-#define BM_P_VMALLOCED  2
-
 static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
 			       unsigned long e, int val, const enum km_type km);
 
-static int bm_is_locked(struct drbd_bitmap *b)
-{
-	return test_bit(BM_LOCKED, &b->bm_flags);
-}
-
 #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__)
 static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func)
 {
@@ -108,7 +133,7 @@ static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func)
 	    b->bm_task == mdev->worker.task   ? "worker"   : "?");
 }
 
-void drbd_bm_lock(struct drbd_conf *mdev, char *why)
+void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags)
 {
 	struct drbd_bitmap *b = mdev->bitmap;
 	int trylock_failed;
@@ -131,8 +156,9 @@ void drbd_bm_lock(struct drbd_conf *mdev, char *why)
 		    b->bm_task == mdev->worker.task   ? "worker"   : "?");
 		mutex_lock(&b->bm_change);
 	}
-	if (__test_and_set_bit(BM_LOCKED, &b->bm_flags))
+	if (BM_LOCKED_MASK & b->bm_flags)
 		dev_err(DEV, "FIXME bitmap already locked in bm_lock\n");
+	b->bm_flags |= flags & BM_LOCKED_MASK;
 
 	b->bm_why  = why;
 	b->bm_task = current;
@@ -146,31 +172,137 @@ void drbd_bm_unlock(struct drbd_conf *mdev)
 		return;
 	}
 
-	if (!__test_and_clear_bit(BM_LOCKED, &mdev->bitmap->bm_flags))
+	if (!(BM_LOCKED_MASK & mdev->bitmap->bm_flags))
 		dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n");
 
+	b->bm_flags &= ~BM_LOCKED_MASK;
 	b->bm_why  = NULL;
 	b->bm_task = NULL;
 	mutex_unlock(&b->bm_change);
 }
 
-/* word offset to long pointer */
-static unsigned long *__bm_map_paddr(struct drbd_bitmap *b, unsigned long offset, const enum km_type km)
+/* we store some "meta" info about our pages in page->private */
+/* at a granularity of 4k storage per bitmap bit:
+ * one peta byte storage: 1<<50 byte, 1<<38 * 4k storage blocks
+ *  1<<38 bits,
+ *  1<<23 4k bitmap pages.
+ * Use 24 bits as page index, covers 2 peta byte storage
+ * at a granularity of 4k per bit.
+ * Used to report the failed page idx on io error from the endio handlers.
+ */
+#define BM_PAGE_IDX_MASK	((1UL<<24)-1)
+/* this page is currently read in, or written back */
+#define BM_PAGE_IO_LOCK		31
+/* if there has been an IO error for this page */
+#define BM_PAGE_IO_ERROR	30
+/* this is to be able to intelligently skip disk IO,
+ * set if bits have been set since last IO. */
+#define BM_PAGE_NEED_WRITEOUT	29
+/* to mark for lazy writeout once syncer cleared all clearable bits,
+ * we if bits have been cleared since last IO. */
+#define BM_PAGE_LAZY_WRITEOUT	28
+
+/* store_page_idx uses non-atomic assingment. It is only used directly after
+ * allocating the page.  All other bm_set_page_* and bm_clear_page_* need to
+ * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap
+ * changes) may happen from various contexts, and wait_on_bit/wake_up_bit
+ * requires it all to be atomic as well. */
+static void bm_store_page_idx(struct page *page, unsigned long idx)
 {
-	struct page *page;
-	unsigned long page_nr;
+	BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK));
+	page_private(page) |= idx;
+}
+
+static unsigned long bm_page_to_idx(struct page *page)
+{
+	return page_private(page) & BM_PAGE_IDX_MASK;
+}
+
+/* As is very unlikely that the same page is under IO from more than one
+ * context, we can get away with a bit per page and one wait queue per bitmap.
+ */
+static void bm_page_lock_io(struct drbd_conf *mdev, int page_nr)
+{
+	struct drbd_bitmap *b = mdev->bitmap;
+	void *addr = &page_private(b->bm_pages[page_nr]);
+	wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr));
+}
+
+static void bm_page_unlock_io(struct drbd_conf *mdev, int page_nr)
+{
+	struct drbd_bitmap *b = mdev->bitmap;
+	void *addr = &page_private(b->bm_pages[page_nr]);
+	clear_bit(BM_PAGE_IO_LOCK, addr);
+	smp_mb__after_clear_bit();
+	wake_up(&mdev->bitmap->bm_io_wait);
+}
+
+/* set _before_ submit_io, so it may be reset due to being changed
+ * while this page is in flight... will get submitted later again */
+static void bm_set_page_unchanged(struct page *page)
+{
+	/* use cmpxchg? */
+	clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
+	clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
+}
 
+static void bm_set_page_need_writeout(struct page *page)
+{
+	set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
+}
+
+static int bm_test_page_unchanged(struct page *page)
+{
+	volatile const unsigned long *addr = &page_private(page);
+	return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0;
+}
+
+static void bm_set_page_io_err(struct page *page)
+{
+	set_bit(BM_PAGE_IO_ERROR, &page_private(page));
+}
+
+static void bm_clear_page_io_err(struct page *page)
+{
+	clear_bit(BM_PAGE_IO_ERROR, &page_private(page));
+}
+
+static void bm_set_page_lazy_writeout(struct page *page)
+{
+	set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
+}
+
+static int bm_test_page_lazy_writeout(struct page *page)
+{
+	return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
+}
+
+/* on a 32bit box, this would allow for exactly (2<<38) bits. */
+static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr)
+{
 	/* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */
-	page_nr = offset >> (PAGE_SHIFT - LN2_BPL + 3);
+	unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3);
 	BUG_ON(page_nr >= b->bm_number_of_pages);
-	page = b->bm_pages[page_nr];
+	return page_nr;
+}
 
+static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr)
+{
+	/* page_nr = (bitnr/8) >> PAGE_SHIFT; */
+	unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3);
+	BUG_ON(page_nr >= b->bm_number_of_pages);
+	return page_nr;
+}
+
+static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx, const enum km_type km)
+{
+	struct page *page = b->bm_pages[idx];
 	return (unsigned long *) kmap_atomic(page, km);
 }
 
-static unsigned long * bm_map_paddr(struct drbd_bitmap *b, unsigned long offset)
+static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx)
 {
-	return __bm_map_paddr(b, offset, KM_IRQ1);
+	return __bm_map_pidx(b, idx, KM_IRQ1);
 }
 
 static void __bm_unmap(unsigned long *p_addr, const enum km_type km)
@@ -202,6 +334,7 @@ static void bm_unmap(unsigned long *p_addr)
  * to be able to report device specific.
  */
 
+
 static void bm_free_pages(struct page **pages, unsigned long number)
 {
 	unsigned long i;
@@ -269,6 +402,9 @@ static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want)
 				bm_vk_free(new_pages, vmalloced);
 				return NULL;
 			}
+			/* we want to know which page it is
+			 * from the endio handlers */
+			bm_store_page_idx(page, i);
 			new_pages[i] = page;
 		}
 	} else {
@@ -280,9 +416,9 @@ static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want)
 	}
 
 	if (vmalloced)
-		set_bit(BM_P_VMALLOCED, &b->bm_flags);
+		b->bm_flags |= BM_P_VMALLOCED;
 	else
-		clear_bit(BM_P_VMALLOCED, &b->bm_flags);
+		b->bm_flags &= ~BM_P_VMALLOCED;
 
 	return new_pages;
 }
@@ -319,7 +455,7 @@ void drbd_bm_cleanup(struct drbd_conf *mdev)
 {
 	ERR_IF (!mdev->bitmap) return;
 	bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages);
-	bm_vk_free(mdev->bitmap->bm_pages, test_bit(BM_P_VMALLOCED, &mdev->bitmap->bm_flags));
+	bm_vk_free(mdev->bitmap->bm_pages, (BM_P_VMALLOCED & mdev->bitmap->bm_flags));
 	kfree(mdev->bitmap);
 	mdev->bitmap = NULL;
 }
@@ -329,22 +465,39 @@ void drbd_bm_cleanup(struct drbd_conf *mdev)
  * this masks out the remaining bits.
  * Returns the number of bits cleared.
  */
+#define BITS_PER_PAGE		(1UL << (PAGE_SHIFT + 3))
+#define BITS_PER_PAGE_MASK	(BITS_PER_PAGE - 1)
+#define BITS_PER_LONG_MASK	(BITS_PER_LONG - 1)
 static int bm_clear_surplus(struct drbd_bitmap *b)
 {
-	const unsigned long mask = (1UL << (b->bm_bits & (BITS_PER_LONG-1))) - 1;
-	size_t w = b->bm_bits >> LN2_BPL;
-	int cleared = 0;
+	unsigned long mask;
 	unsigned long *p_addr, *bm;
+	int tmp;
+	int cleared = 0;
 
-	p_addr = bm_map_paddr(b, w);
-	bm = p_addr + MLPP(w);
-	if (w < b->bm_words) {
+	/* number of bits modulo bits per page */
+	tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
+	/* mask the used bits of the word containing the last bit */
+	mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
+	/* bitmap is always stored little endian,
+	 * on disk and in core memory alike */
+	mask = cpu_to_lel(mask);
+
+	p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
+	bm = p_addr + (tmp/BITS_PER_LONG);
+	if (mask) {
+		/* If mask != 0, we are not exactly aligned, so bm now points
+		 * to the long containing the last bit.
+		 * If mask == 0, bm already points to the word immediately
+		 * after the last (long word aligned) bit. */
 		cleared = hweight_long(*bm & ~mask);
 		*bm &= mask;
-		w++; bm++;
+		bm++;
 	}
 
-	if (w < b->bm_words) {
+	if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
+		/* on a 32bit arch, we may need to zero out
+		 * a padding long to align with a 64bit remote */
 		cleared += hweight_long(*bm);
 		*bm = 0;
 	}
@@ -354,66 +507,75 @@ static int bm_clear_surplus(struct drbd_bitmap *b)
 
 static void bm_set_surplus(struct drbd_bitmap *b)
 {
-	const unsigned long mask = (1UL << (b->bm_bits & (BITS_PER_LONG-1))) - 1;
-	size_t w = b->bm_bits >> LN2_BPL;
+	unsigned long mask;
 	unsigned long *p_addr, *bm;
-
-	p_addr = bm_map_paddr(b, w);
-	bm = p_addr + MLPP(w);
-	if (w < b->bm_words) {
+	int tmp;
+
+	/* number of bits modulo bits per page */
+	tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
+	/* mask the used bits of the word containing the last bit */
+	mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
+	/* bitmap is always stored little endian,
+	 * on disk and in core memory alike */
+	mask = cpu_to_lel(mask);
+
+	p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
+	bm = p_addr + (tmp/BITS_PER_LONG);
+	if (mask) {
+		/* If mask != 0, we are not exactly aligned, so bm now points
+		 * to the long containing the last bit.
+		 * If mask == 0, bm already points to the word immediately
+		 * after the last (long word aligned) bit. */
 		*bm |= ~mask;
-		bm++; w++;
+		bm++;
 	}
 
-	if (w < b->bm_words) {
-		*bm = ~(0UL);
+	if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
+		/* on a 32bit arch, we may need to zero out
+		 * a padding long to align with a 64bit remote */
+		*bm = ~0UL;
 	}
 	bm_unmap(p_addr);
 }
 
-static unsigned long __bm_count_bits(struct drbd_bitmap *b, const int swap_endian)
+/* you better not modify the bitmap while this is running,
+ * or its results will be stale */
+static unsigned long bm_count_bits(struct drbd_bitmap *b)
 {
-	unsigned long *p_addr, *bm, offset = 0;
+	unsigned long *p_addr;
 	unsigned long bi