[RFC 00/15] drbd: part 2 of adding multiple volume support to drbd

[RFC 00/15] drbd: part 2 of adding multiple volume support to drbd

[Philipp Reisner]

This the first request for review of drbd-8.4. The complete set has 
500 patches. This is the second installment containing 15 patches.
These 15 are actually 117 to 132 of the whole set.

The whole set is available here:
  git://git.drbd.org/linux-2.6-drbd.git for-jens

and is jens_for-3.2_drivers...for-jens
and this part is a96f0ec...ccb0def

The most noticeable change is the support for multiple replicated volumes in
a single DRBD connection.  Write-ordering is obeyed among all writes in all
volumes in a single connection.  This feature is really important for users
who use DRBD for mirroring over longer distances. (Protocol A).

[PATCH 01/15] drbd: Rename drbd_endio_{pri,sec} -> drbd_{,peer_}request_endio

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_int.h      |    2 +-
 drivers/block/drbd/drbd_receiver.c |   13 +++++++------
 drivers/block/drbd/drbd_req.c      |    4 ++--
 drivers/block/drbd/drbd_req.h      |    2 +-
 drivers/block/drbd/drbd_worker.c   |    8 ++++----
 drivers/block/drbd/drbd_wrappers.h |    4 ++--
 6 files changed, 17 insertions(+), 16 deletions(-)

diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index 29e32f7..de92e8b 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -661,7 +661,7 @@ struct drbd_request {
 	/* if local IO is not allowed, will be NULL.
 	 * if local IO _is_ allowed, holds the locally submitted bio clone,
 	 * or, after local IO completion, the ERR_PTR(error).
-	 * see drbd_endio_pri(). */
+	 * see drbd_request_endio(). */
 	struct bio *private_bio;
 
 	struct drbd_interval i;
diff --git a/drivers/block/drbd/drbd_receiver.c b/drivers/block/drbd/drbd_receiver.c
index 9b18cc9..deae7cb 100644
--- a/drivers/block/drbd/drbd_receiver.c
+++ b/drivers/block/drbd/drbd_receiver.c
@@ -1139,7 +1139,7 @@ next_bio:
 	bio->bi_bdev = mdev->ldev->backing_bdev;
 	bio->bi_rw = rw;
 	bio->bi_private = peer_req;
-	bio->bi_end_io = drbd_endio_sec;
+	bio->bi_end_io = drbd_peer_request_endio;
 
 	bio->bi_next = bios;
 	bios = bio;
@@ -1575,7 +1575,7 @@ static int receive_RSDataReply(struct drbd_conf *mdev, enum drbd_packet cmd,
 	if (get_ldev(mdev)) {
 		/* data is submitted to disk within recv_resync_read.
 		 * corresponding put_ldev done below on error,
-		 * or in drbd_endio_sec. */
+		 * or in drbd_peer_request_endio. */
 		ok = recv_resync_read(mdev, sector, data_size);
 	} else {
 		if (__ratelimit(&drbd_ratelimit_state))
@@ -1763,10 +1763,11 @@ static int receive_Data(struct drbd_conf *mdev, enum drbd_packet cmd,
 		return drbd_drain_block(mdev, data_size);
 	}
 
-	/* get_ldev(mdev) successful.
-	 * Corresponding put_ldev done either below (on various errors),
-	 * or in drbd_endio_sec, if we successfully submit the data at
-	 * the end of this function. */
+	/*
+	 * Corresponding put_ldev done either below (on various errors), or in
+	 * drbd_peer_request_endio, if we successfully submit the data at the
+	 * end of this function.
+	 */
 
 	sector = be64_to_cpu(p->sector);
 	peer_req = read_in_block(mdev, p->block_id, sector, data_size);
diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 45a543e..18eb3d1 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -116,7 +116,7 @@ static void _req_is_done(struct drbd_conf *mdev, struct drbd_request *req, const
 			drbd_set_in_sync(mdev, req->i.sector, req->i.size);
 
 		/* one might be tempted to move the drbd_al_complete_io
-		 * to the local io completion callback drbd_endio_pri.
+		 * to the local io completion callback drbd_request_endio.
 		 * but, if this was a mirror write, we may only
 		 * drbd_al_complete_io after this is RQ_NET_DONE,
 		 * otherwise the extent could be dropped from the al
@@ -252,7 +252,7 @@ void _req_may_be_done(struct drbd_request *req, struct bio_and_error *m)
 		 * what we need to do here is just: complete the master_bio.
 		 *
 		 * local completion error, if any, has been stored as ERR_PTR
-		 * in private_bio within drbd_endio_pri.
+		 * in private_bio within drbd_request_endio.
 		 */
 		int ok = (s & RQ_LOCAL_OK) || (s & RQ_NET_OK);
 		int error = PTR_ERR(req->private_bio);
diff --git a/drivers/block/drbd/drbd_req.h b/drivers/block/drbd/drbd_req.h
index e6232ce..e6f2361 100644
--- a/drivers/block/drbd/drbd_req.h
+++ b/drivers/block/drbd/drbd_req.h
@@ -230,7 +230,7 @@ static inline void drbd_req_make_private_bio(struct drbd_request *req, struct bi
 	req->private_bio = bio;
 
 	bio->bi_private  = req;
-	bio->bi_end_io   = drbd_endio_pri;
+	bio->bi_end_io   = drbd_request_endio;
 	bio->bi_next     = NULL;
 }
 
diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 2da2d23..01ab0bc 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -45,8 +45,8 @@ static int w_make_resync_request(struct drbd_work *w, int cancel);
 
 /* endio handlers:
  *   drbd_md_io_complete (defined here)
- *   drbd_endio_pri (defined here)
- *   drbd_endio_sec (defined here)
+ *   drbd_request_endio (defined here)
+ *   drbd_peer_request_endio (defined here)
  *   bm_async_io_complete (defined in drbd_bitmap.c)
  *
  * For all these callbacks, note the following:
@@ -151,7 +151,7 @@ static void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __rel
 /* writes on behalf of the partner, or resync writes,
  * "submitted" by the receiver.
  */
-void drbd_endio_sec(struct bio *bio, int error)
+void drbd_peer_request_endio(struct bio *bio, int error)
 {
 	struct drbd_peer_request *peer_req = bio->bi_private;
 	struct drbd_conf *mdev = peer_req->w.mdev;
@@ -187,7 +187,7 @@ void drbd_endio_sec(struct bio *bio, int error)
 
 /* read, readA or write requests on R_PRIMARY coming from drbd_make_request
  */
-void drbd_endio_pri(struct bio *bio, int error)
+void drbd_request_endio(struct bio *bio, int error)
 {
 	unsigned long flags;
 	struct drbd_request *req = bio->bi_private;
diff --git a/drivers/block/drbd/drbd_wrappers.h b/drivers/block/drbd/drbd_wrappers.h
index 151f1a3..decf9b2 100644
--- a/drivers/block/drbd/drbd_wrappers.h
+++ b/drivers/block/drbd/drbd_wrappers.h
@@ -20,8 +20,8 @@ static inline void drbd_set_my_capacity(struct drbd_conf *mdev,
 
 /* bi_end_io handlers */
 extern void drbd_md_io_complete(struct bio *bio, int error);
-extern void drbd_endio_sec(struct bio *bio, int error);
-extern void drbd_endio_pri(struct bio *bio, int error);
+extern void drbd_peer_request_endio(struct bio *bio, int error);
+extern void drbd_request_endio(struct bio *bio, int error);
 
 /*
  * used to submit our private bio
-- 
1.7.4.1

[PATCH 02/15] drbd: Iterate over all overlapping intervals in a tree

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

Add a macro and helper function for doing that.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_interval.c |   18 ++++++++++++++++++
 drivers/block/drbd/drbd_interval.h |   16 ++++++++++++----
 2 files changed, 30 insertions(+), 4 deletions(-)

diff --git a/drivers/block/drbd/drbd_interval.c b/drivers/block/drbd/drbd_interval.c
index 14dbe2d..0e53f10 100644
--- a/drivers/block/drbd/drbd_interval.c
+++ b/drivers/block/drbd/drbd_interval.c
@@ -157,3 +157,21 @@ drbd_find_overlap(struct rb_root *root, sector_t sector, unsigned int size)
 	}
 	return overlap;
 }
+
+struct drbd_interval *
+drbd_next_overlap(struct drbd_interval *i, sector_t sector, unsigned int size)
+{
+	sector_t end = sector + (size >> 9);
+	struct rb_node *node;
+
+	for (;;) {
+		node = rb_next(&i->rb);
+		if (!node)
+			return NULL;
+		i = rb_entry(node, struct drbd_interval, rb);
+		if (i->sector >= end)
+			return NULL;
+		if (sector < i->sector + (i->size >> 9))
+			return i;
+	}
+}
diff --git a/drivers/block/drbd/drbd_interval.h b/drivers/block/drbd/drbd_interval.h
index 4010ad9..f38fcb0 100644
--- a/drivers/block/drbd/drbd_interval.h
+++ b/drivers/block/drbd/drbd_interval.h
@@ -23,10 +23,18 @@ static inline bool drbd_interval_empty(struct drbd_interval *i)
 	return RB_EMPTY_NODE(&i->rb);
 }
 
-bool drbd_insert_interval(struct rb_root *, struct drbd_interval *);
-bool drbd_contains_interval(struct rb_root *, sector_t, struct drbd_interval *);
-void drbd_remove_interval(struct rb_root *, struct drbd_interval *);
-struct drbd_interval *drbd_find_overlap(struct rb_root *, sector_t,
+extern bool drbd_insert_interval(struct rb_root *, struct drbd_interval *);
+extern bool drbd_contains_interval(struct rb_root *, sector_t,
+				   struct drbd_interval *);
+extern void drbd_remove_interval(struct rb_root *, struct drbd_interval *);
+extern struct drbd_interval *drbd_find_overlap(struct rb_root *, sector_t,
 					unsigned int);
+extern struct drbd_interval *drbd_next_overlap(struct drbd_interval *, sector_t,
+					unsigned int);
+
+#define drbd_for_each_overlap(i, root, sector, size)		\
+	for (i = drbd_find_overlap(root, sector, size);		\
+	     i;							\
+	     i = drbd_next_overlap(i, sector, size))
 
 #endif  /* __DRBD_INTERVAL_H */
-- 
1.7.4.1

[PATCH 03/15] drbd: Remove obsolete comment

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_req.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 18eb3d1..157c737 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -667,10 +667,9 @@ static int drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int s
 		return 0;
 	if (mdev->state.disk <  D_INCONSISTENT)
 		return 0;
-	/* state.disk == D_INCONSISTENT   We will have a look at the BitMap */
-	nr_sectors = drbd_get_capacity(mdev->this_bdev);
 	esector = sector + (size >> 9) - 1;
 
+	nr_sectors = drbd_get_capacity(mdev->this_bdev);
 	D_ASSERT(sector  < nr_sectors);
 	D_ASSERT(esector < nr_sectors);
 
-- 
1.7.4.1

[PATCH 04/15] drbd: Use the IS_ALIGNED() macro in some more places

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_actlog.c   |    6 +++---
 drivers/block/drbd/drbd_receiver.c |    2 +-
 drivers/block/drbd/drbd_req.c      |    2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index 0748871..ad61863 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -767,7 +767,7 @@ 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_BIO_SIZE) {
+	if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "drbd_set_in_sync: sector=%llus size=%d nonsense!\n",
 				(unsigned long long)sector, size);
 		return;
@@ -832,7 +832,7 @@ int __drbd_set_out_of_sync(struct drbd_conf *mdev, sector_t sector, int size,
 	unsigned int enr, count = 0;
 	struct lc_element *e;
 
-	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
+	if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "sector: %llus, size: %d\n",
 			(unsigned long long)sector, size);
 		return 0;
@@ -1217,7 +1217,7 @@ 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_BIO_SIZE) {
+	if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "drbd_rs_failed_io: sector=%llus size=%d nonsense!\n",
 				(unsigned long long)sector, size);
 		return;
diff --git a/drivers/block/drbd/drbd_receiver.c b/drivers/block/drbd/drbd_receiver.c
index deae7cb..af3ee2f 100644
--- a/drivers/block/drbd/drbd_receiver.c
+++ b/drivers/block/drbd/drbd_receiver.c
@@ -2041,7 +2041,7 @@ static int receive_DataRequest(struct drbd_conf *mdev, enum drbd_packet cmd,
 	sector = be64_to_cpu(p->sector);
 	size   = be32_to_cpu(p->blksize);
 
-	if (size <= 0 || (size & 0x1ff) != 0 || size > DRBD_MAX_BIO_SIZE) {
+	if (size <= 0 || !IS_ALIGNED(size, 512) || size > DRBD_MAX_BIO_SIZE) {
 		dev_err(DEV, "%s:%d: sector: %llus, size: %u\n", __FILE__, __LINE__,
 				(unsigned long long)sector, size);
 		return false;
diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 157c737..48d313d 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -1023,7 +1023,7 @@ int drbd_make_request(struct request_queue *q, struct bio *bio)
 	 * what we "blindly" assume:
 	 */
 	D_ASSERT(bio->bi_size > 0);
-	D_ASSERT((bio->bi_size & 0x1ff) == 0);
+	D_ASSERT(IS_ALIGNED(bio->bi_size, 512));
 	D_ASSERT(bio->bi_idx == 0);
 
 	/* to make some things easier, force alignment of requests within the
-- 
1.7.4.1

[PATCH 05/15] drbd: simplify condition in drbd_may_do_local_read()

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

fold
	if (x >= (N+1))
		return 0;
	if (x < N)
		return 0;
into
	if (x != N)
		return 0;

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_req.c |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)

diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index 48d313d..7332198 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -663,9 +663,7 @@ static int drbd_may_do_local_read(struct drbd_conf *mdev, sector_t sector, int s
 
 	if (mdev->state.disk == D_UP_TO_DATE)
 		return 1;
-	if (mdev->state.disk >= D_OUTDATED)
-		return 0;
-	if (mdev->state.disk <  D_INCONSISTENT)
+	if (mdev->state.disk != D_INCONSISTENT)
 		return 0;
 	esector = sector + (size >> 9) - 1;
 
-- 
1.7.4.1

[PATCH 06/15] drbd: fix typo in comment

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_actlog.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index ad61863..1ce3de6 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -546,7 +546,7 @@ cancel:
 }
 
 /**
- * drbd_al_apply_to_bm() - Sets the bitmap to diry(1) where covered ba active AL extents
+ * drbd_al_apply_to_bm() - Sets the bitmap to dirty(1) where covered by active AL extents
  * @mdev:	DRBD device.
  */
 void drbd_al_apply_to_bm(struct drbd_conf *mdev)
-- 
1.7.4.1

[PATCH 07/15] drbd: in drbd_suspend_al, set AL_SUSPENDED before unlocking the activity log

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

As using an empty activity log is the whole point of the excercise,
make sure it is still empty when setting AL_SUSPENDED.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_nl.c |    8 +++-----
 1 files changed, 3 insertions(+), 5 deletions(-)

diff --git a/drivers/block/drbd/drbd_nl.c b/drivers/block/drbd/drbd_nl.c
index d6832f8..ae8f42e 100644
--- a/drivers/block/drbd/drbd_nl.c
+++ b/drivers/block/drbd/drbd_nl.c
@@ -901,19 +901,17 @@ static void drbd_suspend_al(struct drbd_conf *mdev)
 {
 	int s = 0;
 
-	if (lc_try_lock(mdev->act_log)) {
-		drbd_al_shrink(mdev);
-		lc_unlock(mdev->act_log);
-	} else {
+	if (!lc_try_lock(mdev->act_log)) {
 		dev_warn(DEV, "Failed to lock al in drbd_suspend_al()\n");
 		return;
 	}
 
+	drbd_al_shrink(mdev);
 	spin_lock_irq(&mdev->tconn->req_lock);
 	if (mdev->state.conn < C_CONNECTED)
 		s = !test_and_set_bit(AL_SUSPENDED, &mdev->flags);
-
 	spin_unlock_irq(&mdev->tconn->req_lock);
+	lc_unlock(mdev->act_log);
 
 	if (s)
 		dev_info(DEV, "Suspended AL updates\n");
-- 
1.7.4.1

[PATCH 08/15] drbd: use clear_bit_unlock() where appropriate

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Some open-coded clear_bit(); smp_mb__after_clear_bit();
should in fact have been smp_mb__before_clear_bit(); clear_bit();

Instead, use clear_bit_unlock() to annotate the intention,
and have it do the right thing.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_bitmap.c |    3 +--
 drivers/block/drbd/drbd_main.c   |    3 +--
 include/linux/lru_cache.h        |    3 +--
 lib/lru_cache.c                  |   10 ++++------
 4 files changed, 7 insertions(+), 12 deletions(-)

diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index e8d652f..4be7370 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -219,8 +219,7 @@ 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();
+	clear_bit_unlock(BM_PAGE_IO_LOCK, addr);
 	wake_up(&mdev->bitmap->bm_io_wait);
 }
 
diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c
index 910dff7..75b092b 100644
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -2819,8 +2819,7 @@ static int w_bitmap_io(struct drbd_work *w, int unused)
 		put_ldev(mdev);
 	}
 
-	clear_bit(BITMAP_IO, &mdev->flags);
-	smp_mb__after_clear_bit();
+	clear_bit_unlock(BITMAP_IO, &mdev->flags);
 	wake_up(&mdev->misc_wait);
 
 	if (work->done)
diff --git a/include/linux/lru_cache.h b/include/linux/lru_cache.h
index 7a71ffa..4cceafb 100644
--- a/include/linux/lru_cache.h
+++ b/include/linux/lru_cache.h
@@ -275,8 +275,7 @@ static inline int lc_try_lock(struct lru_cache *lc)
  */
 static inline void lc_unlock(struct lru_cache *lc)
 {
-	clear_bit(__LC_DIRTY, &lc->flags);
-	smp_mb__after_clear_bit();
+	clear_bit_unlock(__LC_DIRTY, &lc->flags);
 }
 
 static inline int lc_is_used(struct lru_cache *lc, unsigned int enr)
diff --git a/lib/lru_cache.c b/lib/lru_cache.c
index a07e726..9f353f7 100644
--- a/lib/lru_cache.c
+++ b/lib/lru_cache.c
@@ -44,8 +44,8 @@ MODULE_LICENSE("GPL");
 } while (0)
 
 #define RETURN(x...)     do { \
-	clear_bit(__LC_PARANOIA, &lc->flags); \
-	smp_mb__after_clear_bit(); return x ; } while (0)
+	clear_bit_unlock(__LC_PARANOIA, &lc->flags); \
+	return x ; } while (0)
 
 /* BUG() if e is not one of the elements tracked by lc */
 #define PARANOIA_LC_ELEMENT(lc, e) do {	\
@@ -438,8 +438,7 @@ void lc_changed(struct lru_cache *lc, struct lc_element *e)
 	hlist_add_head(&e->colision, lc_hash_slot(lc, lc->new_number));
 	lc->changing_element = NULL;
 	lc->new_number = LC_FREE;
-	clear_bit(__LC_DIRTY, &lc->flags);
-	smp_mb__after_clear_bit();
+	clear_bit_unlock(__LC_DIRTY, &lc->flags);
 	RETURN();
 }
 
@@ -463,8 +462,7 @@ unsigned int lc_put(struct lru_cache *lc, struct lc_element *e)
 		/* move it to the front of LRU. */
 		list_move(&e->list, &lc->lru);
 		lc->used--;
-		clear_bit(__LC_STARVING, &lc->flags);
-		smp_mb__after_clear_bit();
+		clear_bit_unlock(__LC_STARVING, &lc->flags);
 	}
 	RETURN(e->refcnt);
 }
-- 
1.7.4.1

[PATCH 09/15] lru_cache.h: fix comments referring to ts_ instead of lc_

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

For some time we contemplated calling the "struct lru_cache"
a "struct tracked_set", and some comments kept the ts_ prefix.

Fix those to match the member field names.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 lib/lru_cache.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/lib/lru_cache.c b/lib/lru_cache.c
index 9f353f7..4f638b8 100644
--- a/lib/lru_cache.c
+++ b/lib/lru_cache.c
@@ -378,7 +378,7 @@ struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr)
 	/* it was not present in the active set.
 	 * we are going to recycle an unused (or even "free") element.
 	 * user may need to commit a transaction to record that change.
-	 * we serialize on flags & TF_DIRTY */
+	 * we serialize on flags & LC_DIRTY */
 	if (test_and_set_bit(__LC_DIRTY, &lc->flags)) {
 		++lc->dirty;
 		RETURN(NULL);
-- 
1.7.4.1

[PATCH 10/15] lru_cache: consolidate lc_get and lc_try_get

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 lib/lru_cache.c |  120 ++++++++++++++++++++++++++++---------------------------
 1 files changed, 61 insertions(+), 59 deletions(-)

diff --git a/lib/lru_cache.c b/lib/lru_cache.c
index 4f638b8..1762168 100644
--- a/lib/lru_cache.c
+++ b/lib/lru_cache.c
@@ -308,45 +308,7 @@ static int lc_unused_element_available(struct lru_cache *lc)
 	return 0;
 }
 
-
-/**
- * lc_get - get element by label, maybe change the active set
- * @lc: the lru cache to operate on
- * @enr: the label to look up
- *
- * Finds an element in the cache, increases its usage count,
- * "touches" and returns it.
- *
- * In case the requested number is not present, it needs to be added to the
- * cache. Therefore it is possible that an other element becomes evicted from
- * the cache. In either case, the user is notified so he is able to e.g. keep
- * a persistent log of the cache changes, and therefore the objects in use.
- *
- * Return values:
- *  NULL
- *     The cache was marked %LC_STARVING,
- *     or the requested label was not in the active set
- *     and a changing transaction is still pending (@lc was marked %LC_DIRTY).
- *     Or no unused or free element could be recycled (@lc will be marked as
- *     %LC_STARVING, blocking further lc_get() operations).
- *
- *  pointer to the element with the REQUESTED element number.
- *     In this case, it can be used right away
- *
- *  pointer to an UNUSED element with some different element number,
- *          where that different number may also be %LC_FREE.
- *
- *          In this case, the cache is marked %LC_DIRTY (blocking further changes),
- *          and the returned element pointer is removed from the lru list and
- *          hash collision chains.  The user now should do whatever housekeeping
- *          is necessary.
- *          Then he must call lc_changed(lc,element_pointer), to finish
- *          the change.
- *
- * NOTE: The user needs to check the lc_number on EACH use, so he recognizes
- *       any cache set change.
- */
-struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr)
+static struct lc_element *__lc_get(struct lru_cache *lc, unsigned int enr, bool may_change)
 {
 	struct lc_element *e;
 
@@ -366,6 +328,8 @@ struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr)
 	}
 
 	++lc->misses;
+	if (!may_change)
+		RETURN(NULL);
 
 	/* In case there is nothing available and we can not kick out
 	 * the LRU element, we have to wait ...
@@ -397,29 +361,67 @@ struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr)
 	RETURN(e);
 }
 
-/* similar to lc_get,
- * but only gets a new reference on an existing element.
- * you either get the requested element, or NULL.
- * will be consolidated into one function.
+/**
+ * lc_get - get element by label, maybe change the active set
+ * @lc: the lru cache to operate on
+ * @enr: the label to look up
+ *
+ * Finds an element in the cache, increases its usage count,
+ * "touches" and returns it.
+ *
+ * In case the requested number is not present, it needs to be added to the
+ * cache. Therefore it is possible that an other element becomes evicted from
+ * the cache. In either case, the user is notified so he is able to e.g. keep
+ * a persistent log of the cache changes, and therefore the objects in use.
+ *
+ * Return values:
+ *  NULL
+ *     The cache was marked %LC_STARVING,
+ *     or the requested label was not in the active set
+ *     and a changing transaction is still pending (@lc was marked %LC_DIRTY).
+ *     Or no unused or free element could be recycled (@lc will be marked as
+ *     %LC_STARVING, blocking further lc_get() operations).
+ *
+ *  pointer to the element with the REQUESTED element number.
+ *     In this case, it can be used right away
+ *
+ *  pointer to an UNUSED element with some different element number,
+ *          where that different number may also be %LC_FREE.
+ *
+ *          In this case, the cache is marked %LC_DIRTY (blocking further changes),
+ *          and the returned element pointer is removed from the lru list and
+ *          hash collision chains.  The user now should do whatever housekeeping
+ *          is necessary.
+ *          Then he must call lc_changed(lc,element_pointer), to finish
+ *          the change.
+ *
+ * NOTE: The user needs to check the lc_number on EACH use, so he recognizes
+ *       any cache set change.
  */
-struct lc_element *lc_try_get(struct lru_cache *lc, unsigned int enr)
+struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr)
 {
-	struct lc_element *e;
-
-	PARANOIA_ENTRY();
-	if (lc->flags & LC_STARVING) {
-		++lc->starving;
-		RETURN(NULL);
-	}
+	return __lc_get(lc, enr, 1);
+}
 
-	e = lc_find(lc, enr);
-	if (e) {
-		++lc->hits;
-		if (e->refcnt++ == 0)
-			lc->used++;
-		list_move(&e->list, &lc->in_use); /* Not evictable... */
-	}
-	RETURN(e);
+/**
+ * lc_try_get - get element by label, if present; do not change the active set
+ * @lc: the lru cache to operate on
+ * @enr: the label to look up
+ *
+ * Finds an element in the cache, increases its usage count,
+ * "touches" and returns it.
+ *
+ * Return values:
+ *  NULL
+ *     The cache was marked %LC_STARVING,
+ *     or the requested label was not in the active set
+ *
+ *  pointer to the element with the REQUESTED element number.
+ *     In this case, it can be used right away
+ */
+struct lc_element *lc_try_get(struct lru_cache *lc, unsigned int enr)
+{
+	return __lc_get(lc, enr, 0);
 }
 
 /**
-- 
1.7.4.1

[PATCH 11/15] drbd: allow to select specific bitmap pages for writeout

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

We are about to allow several changes to the active set in one activity
log transaction. We have to write out the corresponding bitmap pages as
well, if changed.

Introduce drbd_bm_mark_for_writeout(), then re-use the existing bitmap
writeout path to submit all marked pages in one go.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_bitmap.c |   43 ++++++++++++++++++++++++++++++++-----
 drivers/block/drbd/drbd_int.h    |   13 ++++++++---
 2 files changed, 46 insertions(+), 10 deletions(-)

diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index 4be7370..2d00a8f 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -188,6 +188,9 @@ void drbd_bm_unlock(struct drbd_conf *mdev)
 /* 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
+/* pages marked with this "HINT" will be considered for writeout
+ * on activity log transactions */
+#define BM_PAGE_HINT_WRITEOUT	27
 
 /* store_page_idx uses non-atomic assignment. It is only used directly after
  * allocating the page.  All other bm_set_page_* and bm_clear_page_* need to
@@ -237,6 +240,21 @@ static void bm_set_page_need_writeout(struct page *page)
 	set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
 }
 
+/**
+ * drbd_bm_mark_for_writeout() - mark a page with a "hint" to be considered for writeout
+ * @mdev:	DRBD device.
+ * @page_nr:	the bitmap page to mark with the "hint" flag
+ *
+ * From within an activity log transaction, we mark a few pages with these
+ * hints, then call drbd_bm_write_hinted(), which will only write out changed
+ * pages which are flagged with this mark.
+ */
+void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr)
+{
+	struct page *page = mdev->bitmap->bm_pages[page_nr];
+	set_bit(BM_PAGE_HINT_WRITEOUT, &page_private(page));
+}
+
 static int bm_test_page_unchanged(struct page *page)
 {
 	volatile const unsigned long *addr = &page_private(page);
@@ -897,6 +915,7 @@ struct bm_aio_ctx {
 	struct completion done;
 	unsigned flags;
 #define BM_AIO_COPY_PAGES	1
+#define BM_AIO_WRITE_HINTED	2
 	int error;
 };
 
@@ -1007,13 +1026,13 @@ 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,
+		.flags = flags,
 	};
 	struct drbd_bitmap *b = mdev->bitmap;
 	int num_pages, i, count = 0;
@@ -1042,6 +1061,10 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_id
 		if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx)
 			break;
 		if (rw & WRITE) {
+			if ((flags & BM_AIO_WRITE_HINTED) &&
+			    !test_and_clear_bit(BM_PAGE_HINT_WRITEOUT,
+				    &page_private(b->bm_pages[i])))
+				continue;
 			if (bm_test_page_unchanged(b->bm_pages[i])) {
 				dynamic_dev_dbg(DEV, "skipped bm write for idx %u\n", i);
 				continue;
@@ -1099,7 +1122,7 @@ 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);
 }
 
 /**
@@ -1110,7 +1133,7 @@ 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);
 }
 
 /**
@@ -1120,12 +1143,20 @@ 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_hinted() - Write bitmap pages with "hint" marks, if they have changed.
+ * @mdev:	DRBD device.
+ */
+int drbd_bm_write_hinted(struct drbd_conf *mdev) __must_hold(local)
+{
+	return bm_rw(mdev, WRITE, BM_AIO_WRITE_HINTED | BM_AIO_COPY_PAGES, 0);
+}
 
 /**
- * drbd_bm_write_page: Writes a PAGE_SIZE aligned piece of bitmap
+ * drbd_bm_write_page() - Writes a PAGE_SIZE aligned piece of bitmap
  * @mdev:	DRBD device.
  * @idx:	bitmap page index
  *
diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index de92e8b..f125c0e 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1310,11 +1310,14 @@ struct bm_extent {
 
 #define SLEEP_TIME (HZ/10)
 
-#define BM_BLOCK_SHIFT  12			 /* 4k per bit */
+/* We do bitmap IO in units of 4k blocks.
+ * We also still have a hardcoded 4k per bit relation. */
+#define BM_BLOCK_SHIFT	12			 /* 4k per bit */
 #define BM_BLOCK_SIZE	 (1<<BM_BLOCK_SHIFT)
-/* (9+3) : 512 bytes @ 8 bits; representing 16M storage
- * per sector of on disk bitmap */
-#define BM_EXT_SHIFT	 (BM_BLOCK_SHIFT + MD_SECTOR_SHIFT + 3)  /* = 24 */
+/* mostly arbitrarily set the represented size of one bitmap extent,
+ * aka resync extent, to 16 MiB (which is also 512 Byte worth of bitmap
+ * at 4k per bit resolution) */
+#define BM_EXT_SHIFT	 24	/* 16 MiB per resync extent */
 #define BM_EXT_SIZE	 (1<<BM_EXT_SHIFT)
 
 #if (BM_EXT_SHIFT != 24) || (BM_BLOCK_SHIFT != 12)
@@ -1408,7 +1411,9 @@ extern int  drbd_bm_test_bit(struct drbd_conf *mdev, unsigned long bitnr);
 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 void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr);
 extern int  drbd_bm_write(struct drbd_conf *mdev) __must_hold(local);
+extern int  drbd_bm_write_hinted(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);
-- 
1.7.4.1

[PATCH 12/15] lru_cache: allow multiple changes per transaction

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Allow multiple changes to the active set of elements in lru_cache.
The only current user of lru_cache, drbd, is driving this generalisation.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_actlog.c |   50 ++------
 drivers/block/drbd/drbd_nl.c     |    4 +-
 include/linux/lru_cache.h        |   68 +++++++----
 lib/lru_cache.c                  |  243 ++++++++++++++++++++++++++------------
 4 files changed, 225 insertions(+), 140 deletions(-)

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index 1ce3de6..44097c8 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -175,7 +175,6 @@ 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);
@@ -190,19 +189,8 @@ static struct lc_element *_al_get(struct drbd_conf *mdev, unsigned int enr)
 			return NULL;
 		}
 	}
-	al_ext   = lc_get(mdev->act_log, enr);
-	al_flags = mdev->act_log->flags;
+	al_ext = lc_get(mdev->act_log, enr);
 	spin_unlock_irq(&mdev->al_lock);
-
-	/*
-	if (!al_ext) {
-		if (al_flags & LC_STARVING)
-			dev_warn(DEV, "Have to wait for LRU element (AL too small?)\n");
-		if (al_flags & LC_DIRTY)
-			dev_warn(DEV, "Ongoing AL update (AL device too slow?)\n");
-	}
-	*/
-
 	return al_ext;
 }
 
@@ -235,7 +223,7 @@ void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector)
 		mdev->al_writ_cnt++;
 
 		spin_lock_irq(&mdev->al_lock);
-		lc_changed(mdev->act_log, al_ext);
+		lc_committed(mdev->act_log);
 		spin_unlock_irq(&mdev->al_lock);
 		wake_up(&mdev->al_wait);
 	}
@@ -601,7 +589,7 @@ void drbd_al_shrink(struct drbd_conf *mdev)
 	struct lc_element *al_ext;
 	int i;
 
-	D_ASSERT(test_bit(__LC_DIRTY, &mdev->act_log->flags));
+	D_ASSERT(test_bit(__LC_LOCKED, &mdev->act_log->flags));
 
 	for (i = 0; i < mdev->act_log->nr_elements; i++) {
 		al_ext = lc_element_by_index(mdev->act_log, i);
@@ -708,7 +696,9 @@ static void drbd_try_clear_on_disk_bm(struct drbd_conf *mdev, sector_t sector,
 			}
 			ext->rs_left = rs_left;
 			ext->rs_failed = success ? 0 : count;
-			lc_changed(mdev->resync, &ext->lce);
+			/* we don't keep a persistent log of the resync lru,
+			 * we can commit any change right away. */
+			lc_committed(mdev->resync);
 		}
 		lc_put(mdev->resync, &ext->lce);
 		/* no race, we are within the al_lock! */
@@ -892,7 +882,7 @@ struct bm_extent *_bme_get(struct drbd_conf *mdev, unsigned int enr)
 		if (bm_ext->lce.lc_number != enr) {
 			bm_ext->rs_left = drbd_bm_e_weight(mdev, enr);
 			bm_ext->rs_failed = 0;
-			lc_changed(mdev->resync, &bm_ext->lce);
+			lc_committed(mdev->resync);
 			wakeup = 1;
 		}
 		if (bm_ext->lce.refcnt == 1)
@@ -908,7 +898,7 @@ struct bm_extent *_bme_get(struct drbd_conf *mdev, unsigned int enr)
 		if (rs_flags & LC_STARVING)
 			dev_warn(DEV, "Have to wait for element"
 			     " (resync LRU too small?)\n");
-		BUG_ON(rs_flags & LC_DIRTY);
+		BUG_ON(rs_flags & LC_LOCKED);
 	}
 
 	return bm_ext;
@@ -916,26 +906,12 @@ struct bm_extent *_bme_get(struct drbd_conf *mdev, unsigned int enr)
 
 static int _is_in_al(struct drbd_conf *mdev, unsigned int enr)
 {
-	struct lc_element *al_ext;
-	int rv = 0;
+	int rv;
 
 	spin_lock_irq(&mdev->al_lock);
-	if (unlikely(enr == mdev->act_log->new_number))
-		rv = 1;
-	else {
-		al_ext = lc_find(mdev->act_log, enr);
-		if (al_ext) {
-			if (al_ext->refcnt)
-				rv = 1;
-		}
-	}
+	rv = lc_is_used(mdev->act_log, enr);
 	spin_unlock_irq(&mdev->al_lock);
 
-	/*
-	if (unlikely(rv)) {
-		dev_info(DEV, "Delaying sync read until app's write is done\n");
-	}
-	*/
 	return rv;
 }
 
@@ -1065,13 +1041,13 @@ int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector)
 			if (rs_flags & LC_STARVING)
 				dev_warn(DEV, "Have to wait for element"
 				     " (resync LRU too small?)\n");
-			BUG_ON(rs_flags & LC_DIRTY);
+			BUG_ON(rs_flags & LC_LOCKED);
 			goto try_again;
 		}
 		if (bm_ext->lce.lc_number != enr) {
 			bm_ext->rs_left = drbd_bm_e_weight(mdev, enr);
 			bm_ext->rs_failed = 0;
-			lc_changed(mdev->resync, &bm_ext->lce);
+			lc_committed(mdev->resync);
 			wake_up(&mdev->al_wait);
 			D_ASSERT(test_bit(BME_LOCKED, &bm_ext->flags) == 0);
 		}
@@ -1082,8 +1058,6 @@ int drbd_try_rs_begin_io(struct drbd_conf *mdev, sector_t sector)
 	}
 check_al:
 	for (i = 0; i < AL_EXT_PER_BM_SECT; i++) {
-		if (unlikely(al_enr+i == mdev->act_log->new_number))
-			goto try_again;
 		if (lc_is_used(mdev->act_log, al_enr+i))
 			goto try_again;
 	}
diff --git a/drivers/block/drbd/drbd_nl.c b/drivers/block/drbd/drbd_nl.c
index ae8f42e..0a92f52 100644
--- a/drivers/block/drbd/drbd_nl.c
+++ b/drivers/block/drbd/drbd_nl.c
@@ -760,7 +760,7 @@ static int drbd_check_al_size(struct drbd_conf *mdev)
 
 	in_use = 0;
 	t = mdev->act_log;
-	n = lc_create("act_log", drbd_al_ext_cache,
+	n = lc_create("act_log", drbd_al_ext_cache, 1,
 		mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 
 	if (n == NULL) {
@@ -1016,7 +1016,7 @@ static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp
 	}
 
 	resync_lru = lc_create("resync", drbd_bm_ext_cache,
-			61, sizeof(struct bm_extent),
+			1, 61, sizeof(struct bm_extent),
 			offsetof(struct bm_extent, lce));
 	if (!resync_lru) {
 		retcode = ERR_NOMEM;
diff --git a/include/linux/lru_cache.h b/include/linux/lru_cache.h
index 4cceafb..cbafae4 100644
--- a/include/linux/lru_cache.h
+++ b/include/linux/lru_cache.h
@@ -166,9 +166,11 @@ struct lc_element {
 	/* if we want to track a larger set of objects,
 	 * it needs to become arch independend u64 */
 	unsigned lc_number;
-
 	/* special label when on free list */
 #define LC_FREE (~0U)
+
+	/* for pending changes */
+	unsigned lc_new_number;
 };
 
 struct lru_cache {
@@ -176,6 +178,7 @@ struct lru_cache {
 	struct list_head lru;
 	struct list_head free;
 	struct list_head in_use;
+	struct list_head to_be_changed;
 
 	/* the pre-created kmem cache to allocate the objects from */
 	struct kmem_cache *lc_cache;
@@ -186,7 +189,7 @@ struct lru_cache {
 	size_t element_off;
 
 	/* number of elements (indices) */
-	unsigned int  nr_elements;
+	unsigned int nr_elements;
 	/* Arbitrary limit on maximum tracked objects. Practical limit is much
 	 * lower due to allocation failures, probably. For typical use cases,
 	 * nr_elements should be a few thousand at most.
@@ -194,18 +197,19 @@ struct lru_cache {
 	 * 8 high bits of .lc_index to be overloaded with flags in the future. */
 #define LC_MAX_ACTIVE	(1<<24)
 
+	/* allow to accumulate a few (index:label) changes,
+	 * but no more than max_pending_changes */
+	unsigned int max_pending_changes;
+	/* number of elements currently on to_be_changed list */
+	unsigned int pending_changes;
+
 	/* statistics */
-	unsigned used; /* number of lelements currently on in_use list */
-	unsigned long hits, misses, starving, dirty, changed;
+	unsigned used; /* number of elements currently on in_use list */
+	unsigned long hits, misses, starving, locked, changed;
 
 	/* see below: flag-bits for lru_cache */
 	unsigned long flags;
 
-	/* when changing the label of an index element */
-	unsigned int  new_number;
-
-	/* for paranoia when changing the label of an index element */
-	struct lc_element *changing_element;
 
 	void  *lc_private;
 	const char *name;
@@ -221,10 +225,15 @@ enum {
 	/* debugging aid, to catch concurrent access early.
 	 * user needs to guarantee exclusive access by proper locking! */
 	__LC_PARANOIA,
-	/* if we need to change the set, but currently there is a changing
-	 * transaction pending, we are "dirty", and must deferr further
-	 * changing requests */
+
+	/* annotate that the set is "dirty", possibly accumulating further
+	 * changes, until a transaction is finally triggered */
 	__LC_DIRTY,
+
+	/* Locked, no further changes allowed.
+	 * Also used to serialize changing transactions. */
+	__LC_LOCKED,
+
 	/* if we need to change the set, but currently there is no free nor
 	 * unused element available, we are "starving", and must not give out
 	 * further references, to guarantee that eventually some refcnt will
@@ -236,9 +245,11 @@ enum {
 };
 #define LC_PARANOIA (1<<__LC_PARANOIA)
 #define LC_DIRTY    (1<<__LC_DIRTY)
+#define LC_LOCKED   (1<<__LC_LOCKED)
 #define LC_STARVING (1<<__LC_STARVING)
 
 extern struct lru_cache *lc_create(const char *name, struct kmem_cache *cache,
+		unsigned max_pending_changes,
 		unsigned e_count, size_t e_size, size_t e_off);
 extern void lc_reset(struct lru_cache *lc);
 extern void lc_destroy(struct lru_cache *lc);
@@ -249,7 +260,7 @@ extern struct lc_element *lc_try_get(struct lru_cache *lc, unsigned int enr);
 extern struct lc_element *lc_find(struct lru_cache *lc, unsigned int enr);
 extern struct lc_element *lc_get(struct lru_cache *lc, unsigned int enr);
 extern unsigned int lc_put(struct lru_cache *lc, struct lc_element *e);
-extern void lc_changed(struct lru_cache *lc, struct lc_element *e);
+extern void lc_committed(struct lru_cache *lc);
 
 struct seq_file;
 extern size_t lc_seq_printf_stats(struct seq_file *seq, struct lru_cache *lc);
@@ -258,31 +269,40 @@ extern void lc_seq_dump_details(struct seq_file *seq, struct lru_cache *lc, char
 				void (*detail) (struct seq_file *, struct lc_element *));
 
 /**
- * lc_try_lock - can be used to stop lc_get() from changing the tracked set
+ * lc_try_lock_for_transaction - can be used to stop lc_get() from changing the tracked set
  * @lc: the lru cache to operate on
  *
- * Note that the reference counts and order on the active and lru lists may
- * still change.  Returns true if we acquired the lock.
+ * Allows (expects) the set to be "dirty".  Note that the reference counts and
+ * order on the active and lru lists may still change.  Used to serialize
+ * changing transactions.  Returns true if we aquired the lock.
  */
-static inline int lc_try_lock(struct lru_cache *lc)
+static inline int lc_try_lock_for_transaction(struct lru_cache *lc)
 {
-	return !test_and_set_bit(__LC_DIRTY, &lc->flags);
+	return !test_and_set_bit(__LC_LOCKED, &lc->flags);
 }
 
 /**
+ * lc_try_lock - variant to stop lc_get() from changing the tracked set
+ * @lc: the lru cache to operate on
+ *
+ * Note that the reference counts and order on the active and lru lists may
+ * still change.  Only works on a "clean" set.  Returns true if we aquired the
+ * lock, which means there are no pending changes, and any further attempt to
+ * change the set will not succeed until the next lc_unlock().
+ */
+extern int lc_try_lock(struct lru_cache *lc);
+
+/**
  * lc_unlock - unlock @lc, allow lc_get() to change the set again
  * @lc: the lru cache to operate on
  */
 static inline void lc_unlock(struct lru_cache *lc)
 {
-	clear_bit_unlock(__LC_DIRTY, &lc->flags);
+	clear_bit(__LC_DIRTY, &lc->flags);
+	clear_bit_unlock(__LC_LOCKED, &lc->flags);
 }
 
-static inline int lc_is_used(struct lru_cache *lc, unsigned int enr)
-{
-	struct lc_element *e = lc_find(lc, enr);
-	return e && e->refcnt;
-}
+extern bool lc_is_used(struct lru_cache *lc, unsigned int enr);
 
 #define lc_entry(ptr, type, member) \
 	container_of(ptr, type, member)
diff --git a/lib/lru_cache.c b/lib/lru_cache.c
index 1762168..d71d894 100644
--- a/lib/lru_cache.c
+++ b/lib/lru_cache.c
@@ -55,9 +55,40 @@ MODULE_LICENSE("GPL");
 	BUG_ON(i >= lc_->nr_elements);	\
 	BUG_ON(lc_->lc_element[i] != e_); } while (0)
 
+
+/* We need to atomically
+ *  - try to grab the lock (set LC_LOCKED)
+ *  - only if there is no pending transaction
+ *    (neither LC_DIRTY nor LC_STARVING is set)
+ * Because of PARANOIA_ENTRY() above abusing lc->flags as well,
+ * it is not sufficient to just say
+ *	return 0 == cmpxchg(&lc->flags, 0, LC_LOCKED);
+ */
+int lc_try_lock(struct lru_cache *lc)
+{
+	unsigned long val;
+	do {
+		val = cmpxchg(&lc->flags, 0, LC_LOCKED);
+	} while (unlikely (val == LC_PARANOIA));
+	/* Spin until no-one is inside a PARANOIA_ENTRY()/RETURN() section. */
+	return 0 == val;
+#if 0
+	/* Alternative approach, spin in case someone enters or leaves a
+	 * PARANOIA_ENTRY()/RETURN() section. */
+	unsigned long old, new, val;
+	do {
+		old = lc->flags & LC_PARANOIA;
+		new = old | LC_LOCKED;
+		val = cmpxchg(&lc->flags, old, new);
+	} while (unlikely (val == (old ^ LC_PARANOIA)));
+	return old == val;
+#endif
+}
+
 /**
  * lc_create - prepares to track objects in an active set
  * @name: descriptive name only used in lc_seq_printf_stats and lc_seq_dump_details
+ * @max_pending_changes: maximum changes to accumulate until a transaction is required
  * @e_count: number of elements allowed to be active simultaneously
  * @e_size: size of the tracked objects
  * @e_off: offset to the &struct lc_element member in a tracked object
@@ -66,6 +97,7 @@ MODULE_LICENSE("GPL");
  * or NULL on (allocation) failure.
  */
 struct lru_cache *lc_create(const char *name, struct kmem_cache *cache,
+		unsigned max_pending_changes,
 		unsigned e_count, size_t e_size, size_t e_off)
 {
 	struct hlist_head *slot = NULL;
@@ -98,12 +130,13 @@ struct lru_cache *lc_create(const char *name, struct kmem_cache *cache,
 	INIT_LIST_HEAD(&lc->in_use);
 	INIT_LIST_HEAD(&lc->lru);
 	INIT_LIST_HEAD(&lc->free);
+	INIT_LIST_HEAD(&lc->to_be_changed);
 
 	lc->name = name;
 	lc->element_size = e_size;
 	lc->element_off = e_off;
 	lc->nr_elements = e_count;
-	lc->new_number = LC_FREE;
+	lc->max_pending_changes = max_pending_changes;
 	lc->lc_cache = cache;
 	lc->lc_element = element;
 	lc->lc_slot = slot;
@@ -117,6 +150,7 @@ struct lru_cache *lc_create(const char *name, struct kmem_cache *cache,
 		e = p + e_off;
 		e->lc_index = i;
 		e->lc_number = LC_FREE;
+		e->lc_new_number = LC_FREE;
 		list_add(&e->list, &lc->free);
 		element[i] = e;
 	}
@@ -175,15 +209,15 @@ void lc_reset(struct lru_cache *lc)
 	INIT_LIST_HEAD(&lc->in_use);
 	INIT_LIST_HEAD(&lc->lru);
 	INIT_LIST_HEAD(&lc->free);
+	INIT_LIST_HEAD(&lc->to_be_changed);
 	lc->used = 0;
 	lc->hits = 0;
 	lc->misses = 0;
 	lc->starving = 0;
-	lc->dirty = 0;
+	lc->locked = 0;
 	lc->changed = 0;
+	lc->pending_changes = 0;
 	lc->flags = 0;
-	lc->changing_element = NULL;
-	lc->new_number = LC_FREE;
 	memset(lc->lc_slot, 0, sizeof(struct hlist_head) * lc->nr_elements);
 
 	for (i = 0; i < lc->nr_elements; i++) {
@@ -194,6 +228,7 @@ void lc_reset(struct lru_cache *lc)
 		/* re-init it */
 		e->lc_index = i;
 		e->lc_number = LC_FREE;
+		e->lc_new_number = LC_FREE;
 		list_add(&e->list, &lc->free);
 	}
 }
@@ -208,14 +243,14 @@ size_t lc_seq_printf_stats(struct seq_file *seq, struct lru_cache *lc)
 	/* NOTE:
 	 * total calls to lc_get are
 	 * (starving + hits + misses)
-	 * misses include "dirty" count (update from an other thread in
+	 * misses include "locked" count (update from an other thread in
 	 * progress) and "changed", when this in fact lead to an successful
 	 * update of the cache.
 	 */
 	return seq_printf(seq, "\t%s: used:%u/%u "
-		"hits:%lu misses:%lu starving:%lu dirty:%lu changed:%lu\n",
+		"hits:%lu misses:%lu starving:%lu locked:%lu changed:%lu\n",
 		lc->name, lc->used, lc->nr_elements,
-		lc->hits, lc->misses, lc->starving, lc->dirty, lc->changed);
+		lc->hits, lc->misses, lc->starving, lc->locked, lc->changed);
 }
 
 static struct hlist_head *lc_hash_slot(struct lru_cache *lc, unsigned int enr)
@@ -224,16 +259,8 @@ static struct hlist_head *lc_hash_slot(struct lru_cache *lc, unsigned int enr)
 }
 
 
-/**
- * lc_find - find element by label, if present in the hash table
- * @lc: The lru_cache object
- * @enr: element number
- *
- * Returns the pointer to an element, if the element with the requested
- * "label" or element number is present in the hash table,
- * or NULL if not found. Does not change the refcnt.
- */
-struct lc_element *lc_find(struct lru_cache *lc, unsigned int enr)
+static struct lc_element *__lc_find(struct lru_cache *lc, unsigned int enr,
+		bool include_changing)
 {
 	struct hlist_node *n;
 	struct lc_element *e;
@@ -241,29 +268,48 @@ struct lc_element *lc_find(struct lru_cache *lc, unsigned int enr)
 	BUG_ON(!lc);
 	BUG_ON(!lc->nr_elements);
 	hlist_for_each_entry(e, n, lc_hash_slot(lc, enr), colision) {
-		if (e->lc_number == enr)
+		/* "about to be changed" elements, pending transaction commit,
+		 * are hashed by their "new number". "Normal" elements have
+		 * lc_number == lc_new_number. */
+		if (e->lc_new_number != enr)
+			continue;
+		if (e->lc_new_number == e->lc_number || include_changing)
 			return e;
+		break;
 	}
 	return NULL;
 }
 
-/* returned element will be "recycled" immediately */
-static struct lc_element *lc_evict(struct lru_cache *lc)
+/**
+ * lc_find - find element by label, if present in the hash table
+ * @lc: The lru_cache object
+ * @enr: element number
+ *
+ * Returns the pointer to an element, if the element with the requested
+ * "label" or element number is present in the hash table,
+ * or NULL if not found. Does not change the refcnt.
+ * Ignores elements that are "about to be used", i.e. not yet in the active
+ * set, but still pending transaction commit.
+ */
+struct lc_element *lc_find(struct lru_cache *lc, unsigned int enr)
 {
-	struct list_head  *n;
-	struct lc_element *e;
-
-	if (list_empty(&lc->lru))
-		return NULL;
-
-	n = lc->lru.prev;
-	e = list_entry(n, struct lc_element, list);
-
-	PARANOIA_LC_ELEMENT(lc, e);
+	return __lc_find(lc, enr, 0);
+}
 
-	list_del(&e->list);
-	hlist_del(&e->colision);
-	return e;
+/**
+ * lc_is_used - find element by label
+ * @lc: The lru_cache object
+ * @enr: element number
+ *
+ * Returns true, if the element with the requested "label" or element number is
+ * present in the hash table, and is used (refcnt > 0).
+ * Also finds elements that are not _currently_ used but only "about to be
+ * used", i.e. on the "to_be_changed" list, pending transaction commit.
+ */
+bool lc_is_used(struct lru_cache *lc, unsigned int enr)
+{
+	struct lc_element *e = __lc_find(lc, enr, 1);
+	return e && e->refcnt;
 }
 
 /**
@@ -280,22 +326,34 @@ void lc_del(struct lru_cache *lc, struct lc_element *e)
 	PARANOIA_LC_ELEMENT(lc, e);
 	BUG_ON(e->refcnt);
 
-	e->lc_number = LC_FREE;
+	e->lc_number = e->lc_new_number = LC_FREE;
 	hlist_del_init(&e->colision);
 	list_move(&e->list, &lc->free);
 	RETURN();
 }
 
-static struct lc_element *lc_get_unused_element(struct lru_cache *lc)
+static struct lc_element *lc_prepare_for_change(struct lru_cache *lc, unsigned new_number)
 {
 	struct list_head *n;
+	struct lc_element *e;
+
+	if (!list_empty(&lc->free))
+		n = lc->free.next;
+	else if (!list_empty(&lc->lru))
+		n = lc->lru.prev;
+	else
+		return NULL;
+
+	e = list_entry(n, struct lc_element, list);
+	PARANOIA_LC_ELEMENT(lc, e);
 
-	if (list_empty(&lc->free))
-		return lc_evict(lc);
+	e->lc_new_number = new_number;
+	if (!hlist_unhashed(&e->colision))
+		__hlist_del(&e->colision);
+	hlist_add_head(&e->colision, lc_hash_slot(lc, new_number));
+	list_move(&e->list, &lc->to_be_changed);
 
-	n = lc->free.next;
-	list_del(n);
-	return list_entry(n, struct lc_element, list);
+	return e;
 }
 
 static int lc_unused_element_available(struct lru_cache *lc)
@@ -318,8 +376,12 @@ static struct lc_element *__lc_get(struct lru_cache *lc, unsigned int enr, bool
 		RETURN(NULL);
 	}
 
-	e = lc_find(lc, enr);
-	if (e) {
+	e = __lc_find(lc, enr, 1);
+	/* if lc_new_number != lc_number,
+	 * this enr is currently being pulled in already,
+	 * and will be available once the pending transaction
+	 * has been committed. */
+	if (e && e->lc_new_number == e->lc_number) {
 		++lc->hits;
 		if (e->refcnt++ == 0)
 			lc->used++;
@@ -331,6 +393,24 @@ static struct lc_element *__lc_get(struct lru_cache *lc, unsigned int enr, bool
 	if (!may_change)
 		RETURN(NULL);
 
+	/* It has been found above, but on the "to_be_changed" list, not yet
+	 * committed.  Don't pull it in twice, wait for the transaction, then
+	 * try again */
+	if (e)
+		RETURN(NULL);
+
+	/* To avoid races with lc_try_lock(), first, mark us dirty
+	 * (using test_and_set_bit, as it implies memory barriers), ... */
+	test_and_set_bit(__LC_DIRTY, &lc->flags);
+
+	/* ... only then check if it is locked anyways. If lc_unlock clears
+	 * the dirty bit again, that's not a problem, we will come here again.
+	 */
+	if (test_bit(__LC_LOCKED, &lc->flags)) {
+		++lc->locked;
+		RETURN(NULL);
+	}
+
 	/* In case there is nothing available and we can not kick out
 	 * the LRU element, we have to wait ...
 	 */
@@ -339,24 +419,19 @@ static struct lc_element *__lc_get(struct lru_cache *lc, unsigned int enr, bool
 		RETURN(NULL);
 	}
 
-	/* it was not present in the active set.
-	 * we are going to recycle an unused (or even "free") element.
-	 * user may need to commit a transaction to record that change.
-	 * we serialize on flags & LC_DIRTY */
-	if (test_and_set_bit(__LC_DIRTY, &lc->flags)) {
-		++lc->dirty;
+	/* It was not present in the active set.  We are going to recycle an
+	 * unused (or even "free") element, but we won't accumulate more than
+	 * max_pending_changes changes.  */
+	if (lc->pending_changes >= lc->max_pending_changes)
 		RETURN(NULL);
-	}
 
-	e = lc_get_unused_element(lc);
+	e = lc_prepare_for_change(lc, enr);
 	BUG_ON(!e);
 
 	clear_bit(__LC_STARVING, &lc->flags);
 	BUG_ON(++e->refcnt != 1);
 	lc->used++;
-
-	lc->changing_element = e;
-	lc->new_number = enr;
+	lc->pending_changes++;
 
 	RETURN(e);
 }
@@ -388,12 +463,15 @@ static struct lc_element *__lc_get(struct lru_cache *lc, unsigned int enr, bool
  *  pointer to an UNUSED element with some different element number,
  *          where that different number may also be %LC_FREE.
  *
- *          In this case, the cache is marked %LC_DIRTY (blocking further changes),
- *          and the returned element pointer is removed from the lru list and
- *          hash collision chains.  The user now should do whatever housekeeping
- *          is necessary.
- *          Then he must call lc_changed(lc,element_pointer), to finish
- *          the change.
+ *          In this case, the cache is marked %LC_DIRTY,
+ *          so lc_try_lock() will no longer succeed.
+ *          The returned element pointer is moved to the "to_be_changed" list,
+ *          and registered with the new element number on the hash collision chains,
+ *          so it is possible to pick it up from lc_is_used().
+ *          Up to "max_pending_changes" (see lc_create()) can be accumulated.
+ *          The user now should do whatever housekeeping is necessary,
+ *          typically serialize on lc_try_lock_for_transaction(), then call
+ *          lc_committed(lc) and lc_unlock(), to finish the change.
  *
  * NOTE: The user needs to check the lc_number on EACH use, so he recognizes
  *       any cache set change.
@@ -425,22 +503,25 @@ struct lc_element *lc_try_get(struct lru_cache *lc, unsigned int enr)
 }
 
 /**
- * lc_changed - tell @lc that the change has been recorded
+ * lc_committed - tell @lc that pending changes have been recorded
  * @lc: the lru cache to operate on
- * @e: the element pending label change
+ *
+ * User is expected to serialize on explicit lc_try_lock_for_transaction()
+ * before the transaction is started, and later needs to lc_unlock() explicitly
+ * as well.
  */
-void lc_changed(struct lru_cache *lc, struct lc_element *e)
+void lc_committed(struct lru_cache *lc)
 {
+	struct lc_element *e, *tmp;
+
 	PARANOIA_ENTRY();
-	BUG_ON(e != lc->changing_element);
-	PARANOIA_LC_ELEMENT(lc, e);
-	++lc->changed;
-	e->lc_number = lc->new_number;
-	list_add(&e->list, &lc->in_use);
-	hlist_add_head(&e->colision, lc_hash_slot(lc, lc->new_number));
-	lc->changing_element = NULL;
-	lc->new_number = LC_FREE;
-	clear_bit_unlock(__LC_DIRTY, &lc->flags);
+	list_for_each_entry_safe(e, tmp, &lc->to_be_changed, list) {
+		/* count number of changes, not number of transactions */
+		++lc->changed;
+		e->lc_number = e->lc_new_number;
+		list_move(&e->list, &lc->in_use);
+	}
+	lc->pending_changes = 0;
 	RETURN();
 }
 
@@ -459,7 +540,7 @@ unsigned int lc_put(struct lru_cache *lc, struct lc_element *e)
 	PARANOIA_ENTRY();
 	PARANOIA_LC_ELEMENT(lc, e);
 	BUG_ON(e->refcnt == 0);
-	BUG_ON(e == lc->changing_element);
+	BUG_ON(e->lc_number != e->lc_new_number);
 	if (--e->refcnt == 0) {
 		/* move it to the front of LRU. */
 		list_move(&e->list, &lc->lru);
@@ -504,16 +585,24 @@ unsigned int lc_index_of(struct lru_cache *lc, struct lc_element *e)
 void lc_set(struct lru_cache *lc, unsigned int enr, int index)
 {
 	struct lc_element *e;
+	struct list_head *lh;
 
 	if (index < 0 || index >= lc->nr_elements)
 		return;
 
 	e = lc_element_by_index(lc, index);
-	e->lc_number = enr;
+	BUG_ON(e->lc_number != e->lc_new_number);
+	BUG_ON(e->refcnt != 0);
 
+	e->lc_number = e->lc_new_number = enr;
 	hlist_del_init(&e->colision);
-	hlist_add_head(&e->colision, lc_hash_slot(lc, enr));
-	list_move(&e->list, e->refcnt ? &lc->in_use : &lc->lru);
+	if (enr == LC_FREE)
+		lh = &lc->free;
+	else {
+		hlist_add_head(&e->colision, lc_hash_slot(lc, enr));
+		lh = &lc->lru;
+	}
+	list_move(&e->list, lh);
 }
 
 /**
@@ -553,8 +642,10 @@ EXPORT_SYMBOL(lc_try_get);
 EXPORT_SYMBOL(lc_find);
 EXPORT_SYMBOL(lc_get);
 EXPORT_SYMBOL(lc_put);
-EXPORT_SYMBOL(lc_changed);
+EXPORT_SYMBOL(lc_committed);
 EXPORT_SYMBOL(lc_element_by_index);
 EXPORT_SYMBOL(lc_index_of);
 EXPORT_SYMBOL(lc_seq_printf_stats);
 EXPORT_SYMBOL(lc_seq_dump_details);
+EXPORT_SYMBOL(lc_try_lock);
+EXPORT_SYMBOL(lc_is_used);
-- 
1.7.4.1

[PATCH 13/15] drbd: new on-disk activity log transaction format

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Use a new on-disk transaction format for the activity log, which allows
for multiple changes to the active set per transaction.

Using 4k transaction blocks, we can now get rid of the work-around code
to deal with devices not supporting 512 byte logical block size.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_actlog.c |  403 +++++++++++++++++++++++---------------
 drivers/block/drbd/drbd_int.h    |   44 +++--
 drivers/block/drbd/drbd_main.c   |    4 -
 drivers/block/drbd/drbd_nl.c     |   42 +---
 include/linux/drbd.h             |    4 +
 include/linux/drbd_limits.h      |    8 +-
 6 files changed, 294 insertions(+), 211 deletions(-)

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index 44097c8..55f5a7a 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -24,21 +24,67 @@
  */
 
 #include <linux/slab.h>
+#include <linux/crc32c.h>
 #include <linux/drbd.h>
+#include <linux/drbd_limits.h>
+#include <linux/dynamic_debug.h>
 #include "drbd_int.h"
 #include "drbd_wrappers.h"
 
-/* We maintain a trivial checksum in our on disk activity log.
- * With that we can ensure correct operation even when the storage
- * device might do a partial (last) sector write while losing power.
- */
-struct __packed al_transaction {
-	u32       magic;
-	u32       tr_number;
-	struct __packed {
-		u32 pos;
-		u32 extent; } updates[1 + AL_EXTENTS_PT];
-	u32       xor_sum;
+/* all fields on disc in big endian */
+struct __packed al_transaction_on_disk {
+	/* don't we all like magic */
+	__be32	magic;
+
+	/* to identify the most recent transaction block
+	 * in the on disk ring buffer */
+	__be32	tr_number;
+
+	/* checksum on the full 4k block, with this field set to 0. */
+	__be32	crc32c;
+
+	/* type of transaction, special transaction types like:
+	 * purge-all, set-all-idle, set-all-active, ... to-be-defined */
+	__be16	transaction_type;
+
+	/* we currently allow only a few thousand extents,
+	 * so 16bit will be enough for the slot number. */
+
+	/* how many updates in this transaction */
+	__be16	n_updates;
+
+	/* maximum slot number, "al-extents" in drbd.conf speak.
+	 * Having this in each transaction should make reconfiguration
+	 * of that parameter easier. */
+	__be16	context_size;
+
+	/* slot number the context starts with */
+	__be16	context_start_slot_nr;
+
+	/* Some reserved bytes.  Expected usage is a 64bit counter of
+	 * sectors-written since device creation, and other data generation tag
+	 * supporting usage */
+	__be32	__reserved[4];
+
+	/* --- 36 byte used --- */
+
+	/* Reserve space for up to AL_UPDATES_PER_TRANSACTION changes
+	 * in one transaction, then use the remaining byte in the 4k block for
+	 * context information.  "Flexible" number of updates per transaction
+	 * does not help, as we have to account for the case when all update
+	 * slots are used anyways, so it would only complicate code without
+	 * additional benefit.
+	 */
+	__be16	update_slot_nr[AL_UPDATES_PER_TRANSACTION];
+
+	/* but the extent number is 32bit, which at an extent size of 4 MiB
+	 * allows to cover device sizes of up to 2**54 Byte (16 PiB) */
+	__be32	update_extent_nr[AL_UPDATES_PER_TRANSACTION];
+
+	/* --- 420 bytes used (36 + 64*6) --- */
+
+	/* 4096 - 420 = 3676 = 919 * 4 */
+	__be32	context[AL_CONTEXT_PER_TRANSACTION];
 };
 
 struct update_odbm_work {
@@ -48,11 +94,8 @@ struct update_odbm_work {
 
 struct update_al_work {
 	struct drbd_work w;
-	struct lc_element *al_ext;
 	struct completion event;
-	unsigned int enr;
-	/* if old_enr != LC_FREE, write corresponding bitmap sector, too */
-	unsigned int old_enr;
+	int err;
 };
 
 struct drbd_atodb_wait {
@@ -107,67 +150,30 @@ static int _drbd_md_sync_page_io(struct drbd_conf *mdev,
 int drbd_md_sync_page_io(struct drbd_conf *mdev, struct drbd_backing_dev *bdev,
 			 sector_t sector, int rw)
 {
-	int logical_block_size, mask, ok;
-	int offset = 0;
+	int ok;
 	struct page *iop = mdev->md_io_page;
 
 	D_ASSERT(mutex_is_locked(&mdev->md_io_mutex));
 
 	BUG_ON(!bdev->md_bdev);
 
-	logical_block_size = bdev_logical_block_size(bdev->md_bdev);
-	if (logical_block_size == 0)
-		logical_block_size = MD_SECTOR_SIZE;
-
-	/* in case logical_block_size != 512 [ s390 only? ] */
-	if (logical_block_size != MD_SECTOR_SIZE) {
-		mask = (logical_block_size / MD_SECTOR_SIZE) - 1;
-		D_ASSERT(mask == 1 || mask == 3 || mask == 7);
-		D_ASSERT(logical_block_size == (mask+1) * MD_SECTOR_SIZE);
-		offset = sector & mask;
-		sector = sector & ~mask;
-		iop = mdev->md_io_tmpp;
-
-		if (rw & WRITE) {
-			/* these are GFP_KERNEL pages, pre-allocated
-			 * on device initialization */
-			void *p = page_address(mdev->md_io_page);
-			void *hp = page_address(mdev->md_io_tmpp);
-
-			ok = _drbd_md_sync_page_io(mdev, bdev, iop, sector,
-					READ, logical_block_size);
-
-			if (unlikely(!ok)) {
-				dev_err(DEV, "drbd_md_sync_page_io(,%llus,"
-				    "READ [logical_block_size!=512]) failed!\n",
-				    (unsigned long long)sector);
-				return 0;
-			}
-
-			memcpy(hp + offset*MD_SECTOR_SIZE, p, MD_SECTOR_SIZE);
-		}
-	}
+	dev_dbg(DEV, "meta_data io: %s [%d]:%s(,%llus,%s)\n",
+	     current->comm, current->pid, __func__,
+	     (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ");
 
 	if (sector < drbd_md_first_sector(bdev) ||
-	    sector > drbd_md_last_sector(bdev))
+	    sector + 7 > drbd_md_last_sector(bdev))
 		dev_alert(DEV, "%s [%d]:%s(,%llus,%s) out of range md access!\n",
 		     current->comm, current->pid, __func__,
 		     (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ");
 
-	ok = _drbd_md_sync_page_io(mdev, bdev, iop, sector, rw, logical_block_size);
+	ok = _drbd_md_sync_page_io(mdev, bdev, iop, sector, rw, MD_BLOCK_SIZE);
 	if (unlikely(!ok)) {
 		dev_err(DEV, "drbd_md_sync_page_io(,%llus,%s) failed!\n",
 		    (unsigned long long)sector, (rw & WRITE) ? "WRITE" : "READ");
 		return 0;
 	}
 
-	if (logical_block_size != MD_SECTOR_SIZE && !(rw & WRITE)) {
-		void *p = page_address(mdev->md_io_page);
-		void *hp = page_address(mdev->md_io_tmpp);
-
-		memcpy(p, hp + offset*MD_SECTOR_SIZE, MD_SECTOR_SIZE);
-	}
-
 	return ok;
 }
 
@@ -211,20 +217,34 @@ void drbd_al_begin_io(struct drbd_conf *mdev, sector_t sector)
 		 * current->bio_tail list now.
 		 * we have to delegate updates to the activity log
 		 * to the worker thread. */
-		init_completion(&al_work.event);
-		al_work.al_ext = al_ext;
-		al_work.enr = enr;
-		al_work.old_enr = al_ext->lc_number;
-		al_work.w.cb = w_al_write_transaction;
-		al_work.w.mdev = mdev;
-		drbd_queue_work_front(&mdev->tconn->data.work, &al_work.w);
-		wait_for_completion(&al_work.event);
 
-		mdev->al_writ_cnt++;
+		/* Serialize multiple transactions.
+		 * This uses test_and_set_bit, memory barrier is implicit.
+		 * Optimization potential:
+		 * first check for transaction number > old transaction number,
+		 * so not all waiters have to lock/unlock.  */
+		wait_event(mdev->al_wait, lc_try_lock_for_transaction(mdev->act_log));
 
-		spin_lock_irq(&mdev->al_lock);
-		lc_committed(mdev->act_log);
-		spin_unlock_irq(&mdev->al_lock);
+		/* Double check: it may have been committed by someone else,
+		 * while we have been waiting for the lock. */
+		if (al_ext->lc_number != enr) {
+			init_completion(&al_work.event);
+			al_work.w.cb = w_al_write_transaction;
+			al_work.w.mdev = mdev;
+			drbd_queue_work_front(&mdev->tconn->data.work, &al_work.w);
+			wait_for_completion(&al_work.event);
+
+			mdev->al_writ_cnt++;
+
+			spin_lock_irq(&mdev->al_lock);
+			/* FIXME
+			if (al_work.err)
+				we need an "lc_cancel" here;
+			*/
+			lc_committed(mdev->act_log);
+			spin_unlock_irq(&mdev->al_lock);
+		}
+		lc_unlock(mdev->act_log);
 		wake_up(&mdev->al_wait);
 	}
 }
@@ -283,95 +303,108 @@ w_al_write_transaction(struct drbd_work *w, int unused)
 {
 	struct update_al_work *aw = container_of(w, struct update_al_work, w);
 	struct drbd_conf *mdev = w->mdev;
-	struct lc_element *updated = aw->al_ext;
-	const unsigned int new_enr = aw->enr;
-	const unsigned int evicted = aw->old_enr;
-	struct al_transaction *buffer;
+	struct al_transaction_on_disk *buffer;
+	struct lc_element *e;
 	sector_t sector;
-	int i, n, mx;
-	unsigned int extent_nr;
-	u32 xor_sum = 0;
+	int i, mx;
+	unsigned extent_nr;
+	unsigned crc = 0;
 
 	if (!get_ldev(mdev)) {
-		dev_err(DEV,
-			"disk is %s, cannot start al transaction (-%d +%d)\n",
-			drbd_disk_str(mdev->state.disk), evicted, new_enr);
+		dev_err(DEV, "disk is %s, cannot start al transaction\n",
+			drbd_disk_str(mdev->state.disk));
+		aw->err = -EIO;
 		complete(&((struct update_al_work *)w)->event);
 		return 1;
 	}
-	/* do we have to do a bitmap write, first?
-	 * TODO reduce maximum latency:
-	 * submit both bios, then wait for both,
-	 * instead of doing two synchronous sector writes.
-	 * 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_page(mdev, al_extent_to_bm_page(evicted));
 
 	/* The bitmap write may have failed, causing a state change. */
 	if (mdev->state.disk < D_INCONSISTENT) {
 		dev_err(DEV,
-			"disk is %s, cannot write al transaction (-%d +%d)\n",
-			drbd_disk_str(mdev->state.disk), evicted, new_enr);
+			"disk is %s, cannot write al transaction\n",
+			drbd_disk_str(mdev->state.disk));
+		aw->err = -EIO;
 		complete(&((struct update_al_work *)w)->event);
 		put_ldev(mdev);
 		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 = page_address(mdev->md_io_page);
 
-	buffer->magic = __constant_cpu_to_be32(DRBD_MAGIC);
+	memset(buffer, 0, sizeof(*buffer));
+	buffer->magic = cpu_to_be32(DRBD_AL_MAGIC);
 	buffer->tr_number = cpu_to_be32(mdev->al_tr_number);
 
-	n = lc_index_of(mdev->act_log, updated);
-
-	buffer->updates[0].pos = cpu_to_be32(n);
-	buffer->updates[0].extent = cpu_to_be32(new_enr);
+	i = 0;
+	/* no need to grab any further lock, this list cannot change
+	 * while we hold the "LC_LOCKED" bit. */
+	list_for_each_entry(e, &mdev->act_log->to_be_changed, list) {
+		BUG_ON(i >= AL_UPDATES_PER_TRANSACTION);
+		buffer->update_slot_nr[i] = cpu_to_be16(e->lc_index);
+		buffer->update_extent_nr[i] = cpu_to_be32(e->lc_new_number);
+		if (e->lc_number != LC_FREE)
+			drbd_bm_mark_for_writeout(mdev,
+					al_extent_to_bm_page(e->lc_number));
+		i++;
+	}
+	buffer->n_updates = cpu_to_be16(i);
+	for ( ; i < AL_UPDATES_PER_TRANSACTION; i++) {
+		buffer->update_slot_nr[i] = cpu_to_be16(-1);
+		buffer->update_extent_nr[i] = cpu_to_be32(LC_FREE);
+	}
 
-	xor_sum ^= new_enr;
+	buffer->context_size = cpu_to_be16(mdev->act_log->nr_elements);
+	buffer->context_start_slot_nr = cpu_to_be16(mdev->al_tr_cycle);
 
-	mx = min_t(int, AL_EXTENTS_PT,
+	mx = min_t(int, AL_CONTEXT_PER_TRANSACTION,
 		   mdev->act_log->nr_elements - mdev->al_tr_cycle);
 	for (i = 0; i < mx; i++) {
 		unsigned idx = mdev->al_tr_cycle + i;
 		extent_nr = lc_element_by_index(mdev->act_log, idx)->lc_number;
-		buffer->updates[i+1].pos = cpu_to_be32(idx);
-		buffer->updates[i+1].extent = cpu_to_be32(extent_nr);
-		xor_sum ^= extent_nr;
-	}
-	for (; i < AL_EXTENTS_PT; i++) {
-		buffer->updates[i+1].pos = __constant_cpu_to_be32(-1);
-		buffer->updates[i+1].extent = __constant_cpu_to_be32(LC_FREE);
-		xor_sum ^= LC_FREE;
+		buffer->context[i] = cpu_to_be32(extent_nr);
 	}
-	mdev->al_tr_cycle += AL_EXTENTS_PT;
+	for (; i < AL_CONTEXT_PER_TRANSACTION; i++)
+		buffer->context[i] = cpu_to_be32(LC_FREE);
+
+	mdev->al_tr_cycle += AL_CONTEXT_PER_TRANSACTION;
 	if (mdev->al_tr_cycle >= mdev->act_log->nr_elements)
 		mdev->al_tr_cycle = 0;
 
-	buffer->xor_sum = cpu_to_be32(xor_sum);
-
 	sector =  mdev->ldev->md.md_offset
-		+ 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);
+		+ mdev->ldev->md.al_offset
+		+ mdev->al_tr_pos * (MD_BLOCK_SIZE>>9);
 
-	if (++mdev->al_tr_pos >
-	    div_ceil(mdev->act_log->nr_elements, AL_EXTENTS_PT))
-		mdev->al_tr_pos = 0;
+	crc = crc32c(0, buffer, 4096);
+	buffer->crc32c = cpu_to_be32(crc);
 
-	D_ASSERT(mdev->al_tr_pos < MD_AL_MAX_SIZE);
-	mdev->al_tr_number++;
+	if (drbd_bm_write_hinted(mdev))
+		aw->err = -EIO;
+		/* drbd_chk_io_error done already */
+	else if (!drbd_md_sync_page_io(mdev, mdev->ldev, sector, WRITE)) {
+		aw->err = -EIO;
+		drbd_chk_io_error(mdev, 1, true);
+	} else {
+		/* advance ringbuffer position and transaction counter */
+		mdev->al_tr_pos = (mdev->al_tr_pos + 1) % (MD_AL_SECTORS*512/MD_BLOCK_SIZE);
+		mdev->al_tr_number++;
+	}
 
 	mutex_unlock(&mdev->md_io_mutex);
-
 	complete(&((struct update_al_work *)w)->event);
 	put_ldev(mdev);
 
 	return 1;
 }
 
+/* FIXME
+ * reading of the activity log,
+ * and potentially dirtying of the affected bitmap regions,
+ * should be done from userland only.
+ * DRBD would simply always attach with an empty activity log,
+ * and refuse to attach to something that looks like a crashed primary.
+ */
+
 /**
  * drbd_al_read_tr() - Read a single transaction from the on disk activity log
  * @mdev:	DRBD device.
@@ -383,27 +416,39 @@ w_al_write_transaction(struct drbd_work *w, int unused)
  */
 static int drbd_al_read_tr(struct drbd_conf *mdev,
 			   struct drbd_backing_dev *bdev,
-			   struct al_transaction *b,
 			   int index)
 {
+	struct al_transaction_on_disk *b = page_address(mdev->md_io_page);
 	sector_t sector;
-	int rv, i;
-	u32 xor_sum = 0;
+	u32 crc;
 
-	sector = bdev->md.md_offset + bdev->md.al_offset + index;
+	sector =  bdev->md.md_offset
+		+ bdev->md.al_offset
+		+ index * (MD_BLOCK_SIZE>>9);
 
 	/* Dont process error normally,
 	 * as this is done before disk is attached! */
 	if (!drbd_md_sync_page_io(mdev, bdev, sector, READ))
 		return -1;
 
-	rv = (b->magic == cpu_to_be32(DRBD_MAGIC));
+	if (!expect(b->magic == cpu_to_be32(DRBD_AL_MAGIC)))
+		return 0;
 
-	for (i = 0; i < AL_EXTENTS_PT + 1; i++)
-		xor_sum ^= be32_to_cpu(b->updates[i].extent);
-	rv &= (xor_sum == be32_to_cpu(b->xor_sum));
+	if (!expect(be16_to_cpu(b->n_updates) <= AL_UPDATES_PER_TRANSACTION))
+		return 0;
 
-	return rv;
+	if (!expect(be16_to_cpu(b->context_size) <= DRBD_AL_EXTENTS_MAX))
+		return 0;
+
+	if (!expect(be16_to_cpu(b->context_start_slot_nr) < DRBD_AL_EXTENTS_MAX))
+		return 0;
+
+	crc = be32_to_cpu(b->crc32c);
+	b->crc32c = 0;
+	if (!expect(crc == crc32c(0, b, 4096)))
+		return 0;
+
+	return 1;
 }
 
 /**
@@ -415,7 +460,7 @@ static int drbd_al_read_tr(struct drbd_conf *mdev,
  */
 int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 {
-	struct al_transaction *buffer;
+	struct al_transaction_on_disk *b;
 	int i;
 	int rv;
 	int mx;
@@ -428,25 +473,36 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 	u32 to_tnr = 0;
 	u32 cnr;
 
-	mx = div_ceil(mdev->act_log->nr_elements, AL_EXTENTS_PT);
+	/* Note that this is expected to be called with a newly created,
+	 * clean and all unused activity log of the "expected size".
+	 */
 
 	/* 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);
+	b = page_address(mdev->md_io_page);
+
+	/* Always use the full ringbuffer space for now.
+	 * possible optimization: read in all of it,
+	 * then scan the in-memory pages. */
+
+	mx = (MD_AL_SECTORS*512/MD_BLOCK_SIZE);
 
 	/* Find the valid transaction in the log */
-	for (i = 0; i <= mx; i++) {
-		rv = drbd_al_read_tr(mdev, bdev, buffer, i);
+	for (i = 0; i < mx; i++) {
+		rv = drbd_al_read_tr(mdev, bdev, i);
+		/* invalid data in that block */
 		if (rv == 0)
 			continue;
+
+		/* IO error */
 		if (rv == -1) {
 			mutex_unlock(&mdev->md_io_mutex);
 			return 0;
 		}
-		cnr = be32_to_cpu(buffer->tr_number);
 
+		cnr = be32_to_cpu(b->tr_number);
 		if (++found_valid == 1) {
 			from = i;
 			to = i;
@@ -454,8 +510,11 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 			to_tnr = cnr;
 			continue;
 		}
+
+		D_ASSERT(cnr != to_tnr);
+		D_ASSERT(cnr != from_tnr);
 		if ((int)cnr - (int)from_tnr < 0) {
-			D_ASSERT(from_tnr - cnr + i - from == mx+1);
+			D_ASSERT(from_tnr - cnr + i - from == mx);
 			from = i;
 			from_tnr = cnr;
 		}
@@ -476,11 +535,10 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 	 * dev_info(DEV, "Reading from %d to %d.\n",from,to); */
 	i = from;
 	while (1) {
-		int j, pos;
-		unsigned int extent_nr;
-		unsigned int trn;
+		struct lc_element *e;
+		unsigned j, n, slot, extent_nr;
 
-		rv = drbd_al_read_tr(mdev, bdev, buffer, i);
+		rv = drbd_al_read_tr(mdev, bdev, i);
 		if (!expect(rv != 0))
 			goto cancel;
 		if (rv == -1) {
@@ -488,23 +546,55 @@ int drbd_al_read_log(struct drbd_conf *mdev, struct drbd_backing_dev *bdev)
 			return 0;
 		}
 
-		trn = be32_to_cpu(buffer->tr_number);
+		/* deal with different transaction types.
+		 * not yet implemented */
+		if (!expect(b->transaction_type == 0))
+			goto cancel;
 
-		spin_lock_irq(&mdev->al_lock);
+		/* on the fly re-create/resize activity log?
+		 * will be a special transaction type flag. */
+		if (!expect(be16_to_cpu(b->context_size) == mdev->act_log->nr_elements))
+			goto cancel;
+		if (!expect(be16_to_cpu(b->context_start_slot_nr) < mdev->act_log->nr_elements))
+			goto cancel;
 
-		/* This loop runs backwards because in the cyclic
-		   elements there might be an old version of the
-		   updated element (in slot 0). So the element in slot 0
-		   can overwrite old versions. */
-		for (j = AL_EXTENTS_PT; j >= 0; j--) {
-			pos = be32_to_cpu(buffer->updates[j].pos);
-			extent_nr = be32_to_cpu(buffer->updates[j].extent);
+		/* We are the only user of the activity log right now,
+		 * don't actually need to take that lock. */
+		spin_lock_irq(&mdev->al_lock);
 
-			if (extent_nr == LC_FREE)
-				continue;
+		/* first, apply the context, ... */
+		for (j = 0, slot = be16_to_cpu(b->context_start_slot_nr);
+		     j < AL_CONTEXT_PER_TRANSACTION &&
+		     slot < mdev->act_log->nr_elements; j++, slot++) {
+			extent_nr = be32_to_cpu(b->context[j]);
+			e = lc_element_by_index(mdev->act_log, slot);
+			if (e->lc_number != extent_nr) {
+				if (extent_nr != LC_FREE)
+					active_extents++;
+				else
+					active_extents--;
+			}
+			lc_set(mdev->act_log, extent_nr, slot);
+		}
 
-			lc_set(mdev->act_log, extent_nr, pos);
-			active_extents++;
+		/* ... then apply the updates,
+		 * which override the context information.
+		 * drbd_al_read_tr already did the rangecheck
+		 * on n <= AL_UPDATES_PER_TRANSACTION */
+		n = be16_to_cpu(b->n_updates);
+		for (j = 0; j < n; j++) {
+			slot = be16_to_cpu(b->update_slot_nr[j]);
+			extent_nr = be32_to_cpu(b->update_extent_nr[j]);
+			if (!expect(slot < mdev->act_log->nr_elements))
+				break;
+			e = lc_element_by_index(mdev->act_log, slot);
+			if (e->lc_number != extent_nr) {
+				if (extent_nr != LC_FREE)
+					active_extents++;
+				else
+					active_extents--;
+			}
+			lc_set(mdev->act_log, extent_nr, slot);
 		}
 		spin_unlock_irq(&mdev->al_lock);
 
@@ -514,15 +604,12 @@ cancel:
 		if (i == to)
 			break;
 		i++;
-		if (i > mx)
+		if (i >= mx)
 			i = 0;
 	}
 
 	mdev->al_tr_number = to_tnr+1;
-	mdev->al_tr_pos = to;
-	if (++mdev->al_tr_pos >
-	    div_ceil(mdev->act_log->nr_elements, AL_EXTENTS_PT))
-		mdev->al_tr_pos = 0;
+	mdev->al_tr_pos = (to + 1) % (MD_AL_SECTORS*512/MD_BLOCK_SIZE);
 
 	/* ok, we are done with it */
 	mutex_unlock(&mdev->md_io_mutex);
diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index f125c0e..482bd7e 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1069,7 +1069,6 @@ struct drbd_conf {
 	atomic_t pp_in_use_by_net;	/* sendpage()d, still referenced by tcp */
 	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 */
 	spinlock_t al_lock;
 	wait_queue_head_t al_wait;
@@ -1259,22 +1258,39 @@ extern void drbd_ldev_destroy(struct drbd_conf *mdev);
    * either at the end of the backing device
    * or on a separate meta data device. */
 
-#define MD_RESERVED_SECT (128LU << 11)  /* 128 MB, unit sectors */
 /* The following numbers are sectors */
-#define MD_AL_OFFSET 8	    /* 8 Sectors after start of meta area */
-#define MD_AL_MAX_SIZE 64   /* = 32 kb LOG  ~ 3776 extents ~ 14 GB Storage */
-/* Allows up to about 3.8TB */
-#define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_MAX_SIZE)
-
-/* Since the smalles IO unit is usually 512 byte */
-#define MD_SECTOR_SHIFT	 9
-#define MD_SECTOR_SIZE	 (1<<MD_SECTOR_SHIFT)
-
-/* activity log */
-#define AL_EXTENTS_PT ((MD_SECTOR_SIZE-12)/8-1) /* 61 ; Extents per 512B sector */
-#define AL_EXTENT_SHIFT 22		 /* One extent represents 4M Storage */
+/* Allows up to about 3.8TB, so if you want more,
+ * you need to use the "flexible" meta data format. */
+#define MD_RESERVED_SECT (128LU << 11)  /* 128 MB, unit sectors */
+#define MD_AL_OFFSET	8    /* 8 Sectors after start of meta area */
+#define MD_AL_SECTORS	64   /* = 32 kB on disk activity log ring buffer */
+#define MD_BM_OFFSET (MD_AL_OFFSET + MD_AL_SECTORS)
+
+/* we do all meta data IO in 4k blocks */
+#define MD_BLOCK_SHIFT	12
+#define MD_BLOCK_SIZE	(1<<MD_BLOCK_SHIFT)
+
+/* One activity log extent represents 4M of storage */
+#define AL_EXTENT_SHIFT 22
 #define AL_EXTENT_SIZE (1<<AL_EXTENT_SHIFT)
 
+/* We could make these currently hardcoded constants configurable
+ * variables at create-md time (or even re-configurable at runtime?).
+ * Which will require some more changes to the DRBD "super block"
+ * and attach code.
+ *
+ * updates per transaction:
+ *   This many changes to the active set can be logged with one transaction.
+ *   This number is arbitrary.
+ * context per transaction:
+ *   This many context extent numbers are logged with each transaction.
+ *   This number is resulting from the transaction block size (4k), the layout
+ *   of the transaction header, and the number of updates per transaction.
+ *   See drbd_actlog.c:struct al_transaction_on_disk
+ * */
+#define AL_UPDATES_PER_TRANSACTION	 64	// arbitrary
+#define AL_CONTEXT_PER_TRANSACTION	919	// (4096 - 36 - 6*64)/4
+
 #if BITS_PER_LONG == 32
 #define LN2_BPL 5
 #define cpu_to_lel(A) cpu_to_le32(A)
diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c
index 75b092b..47c4bae 100644
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -2842,10 +2842,6 @@ void drbd_ldev_destroy(struct drbd_conf *mdev)
 		drbd_free_bc(mdev->ldev);
 		mdev->ldev = NULL;);
 
-	if (mdev->md_io_tmpp) {
-		__free_page(mdev->md_io_tmpp);
-		mdev->md_io_tmpp = NULL;
-	}
 	clear_bit(GO_DISKLESS, &mdev->flags);
 }
 
diff --git a/drivers/block/drbd/drbd_nl.c b/drivers/block/drbd/drbd_nl.c
index 0a92f52..90d7317 100644
--- a/drivers/block/drbd/drbd_nl.c
+++ b/drivers/block/drbd/drbd_nl.c
@@ -527,7 +527,7 @@ static void drbd_md_set_sector_offsets(struct drbd_conf *mdev,
 	case DRBD_MD_INDEX_FLEX_INT:
 		bdev->md.md_offset = drbd_md_ss__(mdev, bdev);
 		/* al size is still fixed */
-		bdev->md.al_offset = -MD_AL_MAX_SIZE;
+		bdev->md.al_offset = -MD_AL_SECTORS;
 		/* we need (slightly less than) ~ this much bitmap sectors: */
 		md_size_sect = drbd_get_capacity(bdev->backing_bdev);
 		md_size_sect = ALIGN(md_size_sect, BM_SECT_PER_EXT);
@@ -751,8 +751,8 @@ static int drbd_check_al_size(struct drbd_conf *mdev)
 	unsigned int in_use;
 	int i;
 
-	if (!expect(mdev->sync_conf.al_extents >= 7))
-		mdev->sync_conf.al_extents = 127;
+	if (!expect(mdev->sync_conf.al_extents >= DRBD_AL_EXTENTS_MIN))
+		mdev->sync_conf.al_extents = DRBD_AL_EXTENTS_MIN;
 
 	if (mdev->act_log &&
 	    mdev->act_log->nr_elements == mdev->sync_conf.al_extents)
@@ -760,7 +760,7 @@ static int drbd_check_al_size(struct drbd_conf *mdev)
 
 	in_use = 0;
 	t = mdev->act_log;
-	n = lc_create("act_log", drbd_al_ext_cache, 1,
+	n = lc_create("act_log", drbd_al_ext_cache, AL_UPDATES_PER_TRANSACTION,
 		mdev->sync_conf.al_extents, sizeof(struct lc_element), 0);
 
 	if (n == NULL) {
@@ -932,7 +932,6 @@ static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp
 	union drbd_state ns, os;
 	enum drbd_state_rv rv;
 	int cp_discovered = 0;
-	int logical_block_size;
 
 	drbd_reconfig_start(mdev);
 
@@ -1087,25 +1086,6 @@ static int drbd_nl_disk_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *nlp
 
 	drbd_md_set_sector_offsets(mdev, nbc);
 
-	/* allocate a second IO page if logical_block_size != 512 */
-	logical_block_size = bdev_logical_block_size(nbc->md_bdev);
-	if (logical_block_size == 0)
-		logical_block_size = MD_SECTOR_SIZE;
-
-	if (logical_block_size != MD_SECTOR_SIZE) {
-		if (!mdev->md_io_tmpp) {
-			struct page *page = alloc_page(GFP_NOIO);
-			if (!page)
-				goto force_diskless_dec;
-
-			dev_warn(DEV, "Meta data's bdev logical_block_size = %d != %d\n",
-			     logical_block_size, MD_SECTOR_SIZE);
-			dev_warn(DEV, "Workaround engaged (has performance impact).\n");
-
-			mdev->md_io_tmpp = page;
-		}
-	}
-
 	if (!mdev->bitmap) {
 		if (drbd_bm_init(mdev)) {
 			retcode = ERR_NOMEM;
@@ -1804,14 +1784,12 @@ static int drbd_nl_syncer_conf(struct drbd_conf *mdev, struct drbd_nl_cfg_req *n
 
 	if (!expect(sc.rate >= 1))
 		sc.rate = 1;
-	if (!expect(sc.al_extents >= 7))
-		sc.al_extents = 127; /* arbitrary minimum */
-#define AL_MAX ((MD_AL_MAX_SIZE-1) * AL_EXTENTS_PT)
-	if (sc.al_extents > AL_MAX) {
-		dev_err(DEV, "sc.al_extents > %d\n", AL_MAX);
-		sc.al_extents = AL_MAX;
-	}
-#undef AL_MAX
+
+	/* clip to allowed range */
+	if (!expect(sc.al_extents >= DRBD_AL_EXTENTS_MIN))
+		sc.al_extents = DRBD_AL_EXTENTS_MIN;
+	if (!expect(sc.al_extents <= DRBD_AL_EXTENTS_MAX))
+		sc.al_extents = DRBD_AL_EXTENTS_MAX;
 
 	/* to avoid spurious errors when configuring minors before configuring
 	 * the minors they depend on: if necessary, first create the minor we
diff --git a/include/linux/drbd.h b/include/linux/drbd.h
index 35fc08a..70a688b 100644
--- a/include/linux/drbd.h
+++ b/include/linux/drbd.h
@@ -336,6 +336,10 @@ enum drbd_timeout_flag {
 #define DRBD_MAGIC 0x83740267
 #define DRBD_MAGIC_BIG 0x835a
 
+/* how I came up with this magic?
+ * base64 decode "actlog==" ;) */
+#define DRBD_AL_MAGIC 0x69cb65a2
+
 /* these are of type "int" */
 #define DRBD_MD_INDEX_INTERNAL -1
 #define DRBD_MD_INDEX_FLEX_EXT -2
diff --git a/include/linux/drbd_limits.h b/include/linux/drbd_limits.h
index 447c367..75f05af 100644
--- a/include/linux/drbd_limits.h
+++ b/include/linux/drbd_limits.h
@@ -102,10 +102,12 @@
 #define DRBD_RATE_DEF 250  /* kb/second */
 
   /* less than 7 would hit performance unnecessarily.
-   * 3833 is the largest prime that still does fit
-   * into 64 sectors of activity log */
+   * 919 slots context information per transaction,
+   * 32k activity log, 4k transaction size,
+   * one transaction in flight:
+   * 919 * 7 = 6433 */
 #define DRBD_AL_EXTENTS_MIN  7
-#define DRBD_AL_EXTENTS_MAX  3833
+#define DRBD_AL_EXTENTS_MAX  6433
 #define DRBD_AL_EXTENTS_DEF  127
 
 #define DRBD_AFTER_MIN  -1
-- 
1.7.4.1

[PATCH 14/15] drbd: silence some log messages on bitmap IO

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Summary log messages meant for global bitmap IO
should not be printed for bitmap IO caused by
activity log transactions.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_bitmap.c |   14 +++++++++-----
 1 files changed, 9 insertions(+), 5 deletions(-)

diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index 2d00a8f..2fc3dfa 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -1090,9 +1090,12 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_w
 	 */
 	if (!atomic_dec_and_test(&ctx.in_flight))
 		wait_for_completion(&ctx.done);
-	dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
-			rw == WRITE ? "WRITE" : "READ",
-			count, jiffies - now);
+
+	/* summary for global bitmap IO */
+	if (flags == 0)
+		dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
+				rw == WRITE ? "WRITE" : "READ",
+				count, jiffies - now);
 
 	if (ctx.error) {
 		dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n");
@@ -1110,8 +1113,9 @@ static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_w
 	}
 	now = b->bm_set;
 
-	dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
-	     ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
+	if (flags == 0)
+		dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
+		     ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
 
 	return err;
 }
-- 
1.7.4.1

[PATCH 15/15] drbd: fix a wrong likely(), updated comments

[Philipp Reisner]

From: Lars Ellenberg <lars.ellenberg@linbit.com>

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_worker.c |   15 ++++++++-------
 1 files changed, 8 insertions(+), 7 deletions(-)

diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 01ab0bc..8ee5c4f 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -295,6 +295,7 @@ void drbd_csum_bio(struct drbd_conf *mdev, struct crypto_hash *tfm, struct bio *
 	crypto_hash_final(&desc, digest);
 }
 
+/* MAYBE merge common code with w_e_end_ov_req */
 static int w_e_send_csum(struct drbd_work *w, int cancel)
 {
 	struct drbd_peer_request *peer_req = container_of(w, struct drbd_peer_request, w);
@@ -306,7 +307,7 @@ static int w_e_send_csum(struct drbd_work *w, int cancel)
 	if (unlikely(cancel))
 		goto out;
 
-	if (likely((peer_req->flags & EE_WAS_ERROR) != 0))
+	if (unlikely((peer_req->flags & EE_WAS_ERROR) != 0))
 		goto out;
 
 	digest_size = crypto_hash_digestsize(mdev->csums_tfm);
@@ -315,7 +316,7 @@ static int w_e_send_csum(struct drbd_work *w, int cancel)
 		sector_t sector = peer_req->i.sector;
 		unsigned int size = peer_req->i.size;
 		drbd_csum_ee(mdev, mdev->csums_tfm, peer_req, digest);
-		/* Free e and pages before send.
+		/* Free peer_req and pages before send.
 		 * In case we block on congestion, we could otherwise run into
 		 * some distributed deadlock, if the other side blocks on
 		 * congestion as well, because our receiver blocks in
@@ -1151,11 +1152,11 @@ int w_e_end_ov_reply(struct drbd_work *w, int cancel)
 		}
 	}
 
-		/* Free e and pages before send.
-		 * In case we block on congestion, we could otherwise run into
-		 * some distributed deadlock, if the other side blocks on
-		 * congestion as well, because our receiver blocks in
-		 * drbd_pp_alloc due to pp_in_use > max_buffers. */
+	/* Free peer_req and pages before send.
+	 * In case we block on congestion, we could otherwise run into
+	 * some distributed deadlock, if the other side blocks on
+	 * congestion as well, because our receiver blocks in
+	 * drbd_pp_alloc due to pp_in_use > max_buffers. */
 	drbd_free_ee(mdev, peer_req);
 	if (!eq)
 		drbd_ov_oos_found(mdev, sector, size);
-- 
1.7.4.1