[Drbd-dev] [PATCH 0/9] DRBD updates for 3.18

[Drbd-dev] [PATCH 0/9] DRBD updates for 3.18

[Philipp Reisner]

Hi Jens,

please consider to put these patches into your (future) for-3.18/drivers
branch. I have posted it for review on August 28. It consists mainly of
minor cleanups, and an important performance improvement. With that
DRBD was seen to write with 160k IOPS, on SSDs.

You may take the patches from this posting, or pull them from
http://git.drbd.org/linux-2.6-drbd.git for-3.18-rc1
the branch there is based on Linux 3.17-rc4.

Andreas Gruenbacher (5):
  drbd: Use better variable names
  drbd: Use consistent names for all the bi_end_io callbacks
  drbd: Avoid inconsistent locking warning
  drbd: Get rid of the __no_warn and __cond_lock macros
  drbd: Get rid of the WORK_PENDING macro

Lars Ellenberg (2):
  drbd: Improve asender performance
  drbd: reduce lock contention in drbd_worker

Philipp Marek (1):
  drbd: Remove superfluous newline from "resync_extents" debugfs entry.

Philipp Reisner (1):
  drbd: Add missing newline in resync progress display in /proc/drbd

 drivers/block/drbd/drbd_actlog.c   |  4 +--
 drivers/block/drbd/drbd_bitmap.c   |  6 ++---
 drivers/block/drbd/drbd_debugfs.c  |  2 +-
 drivers/block/drbd/drbd_int.h      | 19 +++++++-------
 drivers/block/drbd/drbd_main.c     | 28 ++++++++++----------
 drivers/block/drbd/drbd_proc.c     |  4 ++-
 drivers/block/drbd/drbd_receiver.c | 52 +++++++++++++++++++++-----------------
 drivers/block/drbd/drbd_req.c      |  2 +-
 drivers/block/drbd/drbd_state.c    | 18 ++++++-------
 drivers/block/drbd/drbd_worker.c   | 51 +++++++++++++++----------------------
 10 files changed, 93 insertions(+), 93 deletions(-)

-- 
1.9.1

[Drbd-dev] [PATCH 1/9] drbd: Use better variable names

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

Rename local variable 'ds' to 'disk_state' or 'data_size'.
'dgs' to 'digest_size'

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_int.h      |  6 ++---
 drivers/block/drbd/drbd_main.c     | 28 +++++++++++------------
 drivers/block/drbd/drbd_receiver.c | 46 +++++++++++++++++++-------------------
 drivers/block/drbd/drbd_state.c    | 18 +++++++--------
 4 files changed, 49 insertions(+), 49 deletions(-)

diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index 1a00001..9e1288e 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -2109,7 +2109,7 @@ static inline bool is_sync_state(enum drbd_conns connection_state)
 
 static inline void put_ldev(struct drbd_device *device)
 {
-	enum drbd_disk_state ds = device->state.disk;
+	enum drbd_disk_state disk_state = device->state.disk;
 	/* We must check the state *before* the atomic_dec becomes visible,
 	 * or we have a theoretical race where someone hitting zero,
 	 * while state still D_FAILED, will then see D_DISKLESS in the
@@ -2122,10 +2122,10 @@ static inline void put_ldev(struct drbd_device *device)
 	__release(local);
 	D_ASSERT(device, i >= 0);
 	if (i == 0) {
-		if (ds == D_DISKLESS)
+		if (disk_state == D_DISKLESS)
 			/* even internal references gone, safe to destroy */
 			drbd_device_post_work(device, DESTROY_DISK);
-		if (ds == D_FAILED)
+		if (disk_state == D_FAILED)
 			/* all application IO references gone. */
 			if (!test_and_set_bit(GOING_DISKLESS, &device->flags))
 				drbd_device_post_work(device, GO_DISKLESS);
diff --git a/drivers/block/drbd/drbd_main.c b/drivers/block/drbd/drbd_main.c
index 9b465bb..973c185 100644
--- a/drivers/block/drbd/drbd_main.c
+++ b/drivers/block/drbd/drbd_main.c
@@ -1622,13 +1622,13 @@ int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *
 	struct drbd_socket *sock;
 	struct p_data *p;
 	unsigned int dp_flags = 0;
-	int dgs;
+	int digest_size;
 	int err;
 
 	sock = &peer_device->connection->data;
 	p = drbd_prepare_command(peer_device, sock);
-	dgs = peer_device->connection->integrity_tfm ?
-	      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
+	digest_size = peer_device->connection->integrity_tfm ?
+		      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
 
 	if (!p)
 		return -EIO;
@@ -1659,9 +1659,9 @@ int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *
 
 	/* our digest is still only over the payload.
 	 * TRIM does not carry any payload. */
-	if (dgs)
+	if (digest_size)
 		drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, p + 1);
-	err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + dgs, NULL, req->i.size);
+	err = __send_command(peer_device->connection, device->vnr, sock, P_DATA, sizeof(*p) + digest_size, NULL, req->i.size);
 	if (!err) {
 		/* For protocol A, we have to memcpy the payload into
 		 * socket buffers, as we may complete right away
@@ -1674,23 +1674,23 @@ int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *
 		 * out ok after sending on this side, but does not fit on the
 		 * receiving side, we sure have detected corruption elsewhere.
 		 */
-		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || dgs)
+		if (!(req->rq_state & (RQ_EXP_RECEIVE_ACK | RQ_EXP_WRITE_ACK)) || digest_size)
 			err = _drbd_send_bio(peer_device, req->master_bio);
 		else
 			err = _drbd_send_zc_bio(peer_device, req->master_bio);
 
 		/* double check digest, sometimes buffers have been modified in flight. */
-		if (dgs > 0 && dgs <= 64) {
+		if (digest_size > 0 && digest_size <= 64) {
 			/* 64 byte, 512 bit, is the largest digest size
 			 * currently supported in kernel crypto. */
 			unsigned char digest[64];
 			drbd_csum_bio(peer_device->connection->integrity_tfm, req->master_bio, digest);
-			if (memcmp(p + 1, digest, dgs)) {
+			if (memcmp(p + 1, digest, digest_size)) {
 				drbd_warn(device,
 					"Digest mismatch, buffer modified by upper layers during write: %llus +%u\n",
 					(unsigned long long)req->i.sector, req->i.size);
 			}
-		} /* else if (dgs > 64) {
+		} /* else if (digest_size > 64) {
 		     ... Be noisy about digest too large ...
 		} */
 	}
@@ -1711,13 +1711,13 @@ int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
 	struct drbd_socket *sock;
 	struct p_data *p;
 	int err;
-	int dgs;
+	int digest_size;
 
 	sock = &peer_device->connection->data;
 	p = drbd_prepare_command(peer_device, sock);
 
-	dgs = peer_device->connection->integrity_tfm ?
-	      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
+	digest_size = peer_device->connection->integrity_tfm ?
+		      crypto_hash_digestsize(peer_device->connection->integrity_tfm) : 0;
 
 	if (!p)
 		return -EIO;
@@ -1725,9 +1725,9 @@ int drbd_send_block(struct drbd_peer_device *peer_device, enum drbd_packet cmd,
 	p->block_id = peer_req->block_id;
 	p->seq_num = 0;  /* unused */
 	p->dp_flags = 0;
-	if (dgs)
+	if (digest_size)
 		drbd_csum_ee(peer_device->connection->integrity_tfm, peer_req, p + 1);
-	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + dgs, NULL, peer_req->i.size);
+	err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*p) + digest_size, NULL, peer_req->i.size);
 	if (!err)
 		err = _drbd_send_zc_ee(peer_device, peer_req);
 	mutex_unlock(&sock->mutex);  /* locked by drbd_prepare_command() */
diff --git a/drivers/block/drbd/drbd_receiver.c b/drivers/block/drbd/drbd_receiver.c
index 9342b8d..3ae769e 100644
--- a/drivers/block/drbd/drbd_receiver.c
+++ b/drivers/block/drbd/drbd_receiver.c
@@ -1371,9 +1371,9 @@ int drbd_submit_peer_request(struct drbd_device *device,
 	struct bio *bio;
 	struct page *page = peer_req->pages;
 	sector_t sector = peer_req->i.sector;
-	unsigned ds = peer_req->i.size;
+	unsigned data_size = peer_req->i.size;
 	unsigned n_bios = 0;
-	unsigned nr_pages = (ds + PAGE_SIZE -1) >> PAGE_SHIFT;
+	unsigned nr_pages = (data_size + PAGE_SIZE -1) >> PAGE_SHIFT;
 	int err = -ENOMEM;
 
 	if (peer_req->flags & EE_IS_TRIM_USE_ZEROOUT) {
@@ -1388,7 +1388,7 @@ int drbd_submit_peer_request(struct drbd_device *device,
 		list_add_tail(&peer_req->w.list, &device->active_ee);
 		spin_unlock_irq(&device->resource->req_lock);
 		if (blkdev_issue_zeroout(device->ldev->backing_bdev,
-			sector, ds >> 9, GFP_NOIO))
+			sector, data_size >> 9, GFP_NOIO))
 			peer_req->flags |= EE_WAS_ERROR;
 		drbd_endio_write_sec_final(peer_req);
 		return 0;
@@ -1426,12 +1426,12 @@ next_bio:
 	++n_bios;
 
 	if (rw & REQ_DISCARD) {
-		bio->bi_iter.bi_size = ds;
+		bio->bi_iter.bi_size = data_size;
 		goto submit;
 	}
 
 	page_chain_for_each(page) {
-		unsigned len = min_t(unsigned, ds, PAGE_SIZE);
+		unsigned len = min_t(unsigned, data_size, PAGE_SIZE);
 		if (!bio_add_page(bio, page, len, 0)) {
 			/* A single page must always be possible!
 			 * But in case it fails anyways,
@@ -1446,11 +1446,11 @@ next_bio:
 			}
 			goto next_bio;
 		}
-		ds -= len;
+		data_size -= len;
 		sector += len >> 9;
 		--nr_pages;
 	}
-	D_ASSERT(device, ds == 0);
+	D_ASSERT(device, data_size == 0);
 submit:
 	D_ASSERT(device, page == NULL);
 
@@ -1591,24 +1591,24 @@ read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector,
 	const sector_t capacity = drbd_get_capacity(device->this_bdev);
 	struct drbd_peer_request *peer_req;
 	struct page *page;
-	int dgs, ds, err;
-	unsigned int data_size = pi->size;
+	int digest_size, err;
+	unsigned int data_size = pi->size, ds;
 	void *dig_in = peer_device->connection->int_dig_in;
 	void *dig_vv = peer_device->connection->int_dig_vv;
 	unsigned long *data;
 	struct p_trim *trim = (pi->cmd == P_TRIM) ? pi->data : NULL;
 
-	dgs = 0;
+	digest_size = 0;
 	if (!trim && peer_device->connection->peer_integrity_tfm) {
-		dgs = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
+		digest_size = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
 		/*
 		 * FIXME: Receive the incoming digest into the receive buffer
 		 *	  here, together with its struct p_data?
 		 */
-		err = drbd_recv_all_warn(peer_device->connection, dig_in, dgs);
+		err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size);
 		if (err)
 			return NULL;
-		data_size -= dgs;
+		data_size -= digest_size;
 	}
 
 	if (trim) {
@@ -1661,16 +1661,16 @@ read_in_block(struct drbd_peer_device *peer_device, u64 id, sector_t sector,
 		ds -= len;
 	}
 
-	if (dgs) {
+	if (digest_size) {
 		drbd_csum_ee(peer_device->connection->peer_integrity_tfm, peer_req, dig_vv);
-		if (memcmp(dig_in, dig_vv, dgs)) {
+		if (memcmp(dig_in, dig_vv, digest_size)) {
 			drbd_err(device, "Digest integrity check FAILED: %llus +%u\n",
 				(unsigned long long)sector, data_size);
 			drbd_free_peer_req(device, peer_req);
 			return NULL;
 		}
 	}
-	device->recv_cnt += data_size>>9;
+	device->recv_cnt += data_size >> 9;
 	return peer_req;
 }
 
@@ -1708,17 +1708,17 @@ static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_req
 	struct bio_vec bvec;
 	struct bvec_iter iter;
 	struct bio *bio;
-	int dgs, err, expect;
+	int digest_size, err, expect;
 	void *dig_in = peer_device->connection->int_dig_in;
 	void *dig_vv = peer_device->connection->int_dig_vv;
 
-	dgs = 0;
+	digest_size = 0;
 	if (peer_device->connection->peer_integrity_tfm) {
-		dgs = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
-		err = drbd_recv_all_warn(peer_device->connection, dig_in, dgs);
+		digest_size = crypto_hash_digestsize(peer_device->connection->peer_integrity_tfm);
+		err = drbd_recv_all_warn(peer_device->connection, dig_in, digest_size);
 		if (err)
 			return err;
-		data_size -= dgs;
+		data_size -= digest_size;
 	}
 
 	/* optimistically update recv_cnt.  if receiving fails below,
@@ -1738,9 +1738,9 @@ static int recv_dless_read(struct drbd_peer_device *peer_device, struct drbd_req
 		data_size -= expect;
 	}
 
-	if (dgs) {
+	if (digest_size) {
 		drbd_csum_bio(peer_device->connection->peer_integrity_tfm, bio, dig_vv);
-		if (memcmp(dig_in, dig_vv, dgs)) {
+		if (memcmp(dig_in, dig_vv, digest_size)) {
 			drbd_err(peer_device, "Digest integrity check FAILED. Broken NICs?\n");
 			return -EINVAL;
 		}
diff --git a/drivers/block/drbd/drbd_state.c b/drivers/block/drbd/drbd_state.c
index c35c0f0..84b11f8 100644
--- a/drivers/block/drbd/drbd_state.c
+++ b/drivers/block/drbd/drbd_state.c
@@ -136,50 +136,50 @@ enum drbd_role conn_highest_peer(struct drbd_connection *connection)
 
 enum drbd_disk_state conn_highest_disk(struct drbd_connection *connection)
 {
-	enum drbd_disk_state ds = D_DISKLESS;
+	enum drbd_disk_state disk_state = D_DISKLESS;
 	struct drbd_peer_device *peer_device;
 	int vnr;
 
 	rcu_read_lock();
 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 		struct drbd_device *device = peer_device->device;
-		ds = max_t(enum drbd_disk_state, ds, device->state.disk);
+		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.disk);
 	}
 	rcu_read_unlock();
 
-	return ds;
+	return disk_state;
 }
 
 enum drbd_disk_state conn_lowest_disk(struct drbd_connection *connection)
 {
-	enum drbd_disk_state ds = D_MASK;
+	enum drbd_disk_state disk_state = D_MASK;
 	struct drbd_peer_device *peer_device;
 	int vnr;
 
 	rcu_read_lock();
 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 		struct drbd_device *device = peer_device->device;
-		ds = min_t(enum drbd_disk_state, ds, device->state.disk);
+		disk_state = min_t(enum drbd_disk_state, disk_state, device->state.disk);
 	}
 	rcu_read_unlock();
 
-	return ds;
+	return disk_state;
 }
 
 enum drbd_disk_state conn_highest_pdsk(struct drbd_connection *connection)
 {
-	enum drbd_disk_state ds = D_DISKLESS;
+	enum drbd_disk_state disk_state = D_DISKLESS;
 	struct drbd_peer_device *peer_device;
 	int vnr;
 
 	rcu_read_lock();
 	idr_for_each_entry(&connection->peer_devices, peer_device, vnr) {
 		struct drbd_device *device = peer_device->device;
-		ds = max_t(enum drbd_disk_state, ds, device->state.pdsk);
+		disk_state = max_t(enum drbd_disk_state, disk_state, device->state.pdsk);
 	}
 	rcu_read_unlock();
 
-	return ds;
+	return disk_state;
 }
 
 enum drbd_conns conn_lowest_conn(struct drbd_connection *connection)
-- 
1.9.1

[Drbd-dev] [PATCH 2/9] drbd: Use consistent names for all the bi_end_io callbacks

[Philipp Reisner]

From: Andreas Gruenbacher <andreas.gruenbacher@gmail.com>

Now they follow the _endio naming sheme.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_actlog.c | 4 ++--
 drivers/block/drbd/drbd_bitmap.c | 6 +++---
 drivers/block/drbd/drbd_int.h    | 2 +-
 drivers/block/drbd/drbd_worker.c | 6 +++---
 4 files changed, 9 insertions(+), 9 deletions(-)

diff --git a/drivers/block/drbd/drbd_actlog.c b/drivers/block/drbd/drbd_actlog.c
index d26a3fa..a2dfa16 100644
--- a/drivers/block/drbd/drbd_actlog.c
+++ b/drivers/block/drbd/drbd_actlog.c
@@ -158,14 +158,14 @@ static int _drbd_md_sync_page_io(struct drbd_device *device,
 	if (bio_add_page(bio, device->md_io.page, size, 0) != size)
 		goto out;
 	bio->bi_private = device;
-	bio->bi_end_io = drbd_md_io_complete;
+	bio->bi_end_io = drbd_md_endio;
 	bio->bi_rw = rw;
 
 	if (!(rw & WRITE) && device->state.disk == D_DISKLESS && device->ldev == NULL)
 		/* special case, drbd_md_read() during drbd_adm_attach(): no get_ldev */
 		;
 	else if (!get_ldev_if_state(device, D_ATTACHING)) {
-		/* Corresponding put_ldev in drbd_md_io_complete() */
+		/* Corresponding put_ldev in drbd_md_endio() */
 		drbd_err(device, "ASSERT FAILED: get_ldev_if_state() == 1 in _drbd_md_sync_page_io()\n");
 		err = -ENODEV;
 		goto out;
diff --git a/drivers/block/drbd/drbd_bitmap.c b/drivers/block/drbd/drbd_bitmap.c
index 426c97a..434c77d 100644
--- a/drivers/block/drbd/drbd_bitmap.c
+++ b/drivers/block/drbd/drbd_bitmap.c
@@ -941,7 +941,7 @@ static void drbd_bm_aio_ctx_destroy(struct kref *kref)
 }
 
 /* bv_page may be a copy, or may be the original */
-static void bm_async_io_complete(struct bio *bio, int error)
+static void drbd_bm_endio(struct bio *bio, int error)
 {
 	struct drbd_bm_aio_ctx *ctx = bio->bi_private;
 	struct drbd_device *device = ctx->device;
@@ -1027,7 +1027,7 @@ static void bm_page_io_async(struct drbd_bm_aio_ctx *ctx, int page_nr) __must_ho
 	 * according to api.  Do we want to assert that? */
 	bio_add_page(bio, page, len, 0);
 	bio->bi_private = ctx;
-	bio->bi_end_io = bm_async_io_complete;
+	bio->bi_end_io = drbd_bm_endio;
 
 	if (drbd_insert_fault(device, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) {
 		bio->bi_rw |= rw;
@@ -1125,7 +1125,7 @@ static int bm_rw(struct drbd_device *device, const unsigned int flags, unsigned
 	}
 
 	/*
-	 * We initialize ctx->in_flight to one to make sure bm_async_io_complete
+	 * We initialize ctx->in_flight to one to make sure drbd_bm_endio
 	 * will not set ctx->done early, and decrement / test it here.  If there
 	 * are still some bios in flight, we need to wait for them here.
 	 * If all IO is done already (or nothing had been submitted), there is
diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index 9e1288e..f424dc0 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -1483,7 +1483,7 @@ extern int drbd_khelper(struct drbd_device *device, char *cmd);
 
 /* drbd_worker.c */
 /* bi_end_io handlers */
-extern void drbd_md_io_complete(struct bio *bio, int error);
+extern void drbd_md_endio(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);
 extern int drbd_worker(struct drbd_thread *thi);
diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 50776b3..6e01e62 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -43,10 +43,10 @@ static int make_ov_request(struct drbd_device *, int);
 static int make_resync_request(struct drbd_device *, int);
 
 /* endio handlers:
- *   drbd_md_io_complete (defined here)
+ *   drbd_md_endio (defined here)
  *   drbd_request_endio (defined here)
  *   drbd_peer_request_endio (defined here)
- *   bm_async_io_complete (defined in drbd_bitmap.c)
+ *   drbd_bm_endio (defined in drbd_bitmap.c)
  *
  * For all these callbacks, note the following:
  * The callbacks will be called in irq context by the IDE drivers,
@@ -65,7 +65,7 @@ rwlock_t global_state_lock;
 /* used for synchronous meta data and bitmap IO
  * submitted by drbd_md_sync_page_io()
  */
-void drbd_md_io_complete(struct bio *bio, int error)
+void drbd_md_endio(struct bio *bio, int error)
 {
 	struct drbd_device *device;
 
-- 
1.9.1

[Drbd-dev] [PATCH 3/9] drbd: Remove superfluous newline from "resync_extents" debugfs entry.

[Philipp Reisner]

From: Philipp Marek <philipp.marek@linbit.com>

See "drbd/resources/*/volumes/*/resync_extents".

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

diff --git a/drivers/block/drbd/drbd_debugfs.c b/drivers/block/drbd/drbd_debugfs.c
index 5c20b18..900d4d3 100644
--- a/drivers/block/drbd/drbd_debugfs.c
+++ b/drivers/block/drbd/drbd_debugfs.c
@@ -695,7 +695,7 @@ static void resync_dump_detail(struct seq_file *m, struct lc_element *e)
 {
        struct bm_extent *bme = lc_entry(e, struct bm_extent, lce);
 
-       seq_printf(m, "%5d %s %s %s\n", bme->rs_left,
+       seq_printf(m, "%5d %s %s %s", bme->rs_left,
 		  test_bit(BME_NO_WRITES, &bme->flags) ? "NO_WRITES" : "---------",
 		  test_bit(BME_LOCKED, &bme->flags) ? "LOCKED" : "------",
 		  test_bit(BME_PRIORITY, &bme->flags) ? "PRIORITY" : "--------"
-- 
1.9.1

[Drbd-dev] [PATCH 4/9] drbd: Avoid inconsistent locking warning

[Philipp Reisner]

From: Andreas Gruenbacher <agruen@linbit.com>

request_timer_fn() takes resource->req_lock via the device and releases it via
the connection.  Avoid this as it is confusing static code checkers.

Reported-by: "Dan Carpenter" <dan.carpenter@oracle.com>
Signed-off-by: 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 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/block/drbd/drbd_req.c b/drivers/block/drbd/drbd_req.c
index c67717d..5a01c53 100644
--- a/drivers/block/drbd/drbd_req.c
+++ b/drivers/block/drbd/drbd_req.c
@@ -1645,6 +1645,6 @@ void request_timer_fn(unsigned long data)
 		? oldest_submit_jif + dt : now + et;
 	nt = time_before(ent, dt) ? ent : dt;
 out:
-	spin_unlock_irq(&connection->resource->req_lock);
+	spin_unlock_irq(&device->resource->req_lock);
 	mod_timer(&device->request_timer, nt);
 }
-- 
1.9.1

[Drbd-dev] [PATCH 5/9] drbd: Get rid of the __no_warn and __cond_lock macros

[Philipp Reisner]

From: Andreas Gruenbacher <andreas.gruenbacher@gmail.com>

These macros can easily be replaced with its definition.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_int.h    | 11 ++++++-----
 drivers/block/drbd/drbd_worker.c |  9 ++++++---
 2 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/drivers/block/drbd/drbd_int.h b/drivers/block/drbd/drbd_int.h
index f424dc0..9b22f8f 100644
--- a/drivers/block/drbd/drbd_int.h
+++ b/drivers/block/drbd/drbd_int.h
@@ -61,8 +61,6 @@
 # define __must_hold(x)
 #endif
 
-#define __no_warn(lock, stmt) do { __acquire(lock); stmt; __release(lock); } while (0)
-
 /* module parameter, defined in drbd_main.c */
 extern unsigned int minor_count;
 extern bool disable_sendpage;
@@ -2100,12 +2098,15 @@ static inline bool is_sync_state(enum drbd_conns connection_state)
 
 /**
  * get_ldev() - Increase the ref count on device->ldev. Returns 0 if there is no ldev
- * @M:		DRBD device.
+ * @_device:		DRBD device.
+ * @_min_state:		Minimum device state required for success.
  *
  * You have to call put_ldev() when finished working with device->ldev.
  */
-#define get_ldev(M) __cond_lock(local, _get_ldev_if_state(M,D_INCONSISTENT))
-#define get_ldev_if_state(M,MINS) __cond_lock(local, _get_ldev_if_state(M,MINS))
+#define get_ldev_if_state(_device, _min_state)				\
+	(_get_ldev_if_state((_device), (_min_state)) ?			\
+	 ({ __acquire(x); true; }) : false)
+#define get_ldev(_device) get_ldev_if_state(_device, D_INCONSISTENT)
 
 static inline void put_ldev(struct drbd_device *device)
 {
diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 6e01e62..b20cd21 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -1853,9 +1853,12 @@ static void drbd_ldev_destroy(struct drbd_device *device)
 	device->resync = NULL;
 	lc_destroy(device->act_log);
 	device->act_log = NULL;
-	__no_warn(local,
-		drbd_free_ldev(device->ldev);
-		device->ldev = NULL;);
+
+	__acquire(local);
+	drbd_free_ldev(device->ldev);
+	device->ldev = NULL;
+	__release(local);
+
 	clear_bit(GOING_DISKLESS, &device->flags);
 	wake_up(&device->misc_wait);
 }
-- 
1.9.1

[Drbd-dev] [PATCH 6/9] drbd: Get rid of the WORK_PENDING macro

[Philipp Reisner]

From: Andreas Gruenbacher <andreas.gruenbacher@gmail.com>

This macro doesn't add any value; just use test_bit() instead.

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 file changed, 7 insertions(+), 8 deletions(-)

diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index b20cd21..3b74f08 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -1931,19 +1931,18 @@ void __update_timing_details(
 	++(*cb_nr);
 }
 
-#define WORK_PENDING(work_bit, todo)	(todo & (1UL << work_bit))
 static void do_device_work(struct drbd_device *device, const unsigned long todo)
 {
-	if (WORK_PENDING(MD_SYNC, todo))
+	if (test_bit(MD_SYNC, &todo))
 		do_md_sync(device);
-	if (WORK_PENDING(RS_DONE, todo) ||
-	    WORK_PENDING(RS_PROGRESS, todo))
-		update_on_disk_bitmap(device, WORK_PENDING(RS_DONE, todo));
-	if (WORK_PENDING(GO_DISKLESS, todo))
+	if (test_bit(RS_DONE, &todo) ||
+	    test_bit(RS_PROGRESS, &todo))
+		update_on_disk_bitmap(device, test_bit(RS_DONE, &todo));
+	if (test_bit(GO_DISKLESS, &todo))
 		go_diskless(device);
-	if (WORK_PENDING(DESTROY_DISK, todo))
+	if (test_bit(DESTROY_DISK, &todo))
 		drbd_ldev_destroy(device);
-	if (WORK_PENDING(RS_START, todo))
+	if (test_bit(RS_START, &todo))
 		do_start_resync(device);
 }
 
-- 
1.9.1

[Drbd-dev] [PATCH 7/9] drbd: Improve asender performance

[Philipp Reisner]

From: Lars Ellenberg <lars@linbit.com>

Shorten receive path in the asender thread. Reduces CPU utilisation
of asender when receiving packets, and with that increases IOPs.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_receiver.c |  6 ++++++
 drivers/block/drbd/drbd_worker.c   | 11 +----------
 2 files changed, 7 insertions(+), 10 deletions(-)

diff --git a/drivers/block/drbd/drbd_receiver.c b/drivers/block/drbd/drbd_receiver.c
index 3ae769e..6960fb0 100644
--- a/drivers/block/drbd/drbd_receiver.c
+++ b/drivers/block/drbd/drbd_receiver.c
@@ -5561,6 +5561,7 @@ int drbd_asender(struct drbd_thread *thi)
 		 * rv <  expected: "woken" by signal during receive
 		 * rv == 0	 : "connection shut down by peer"
 		 */
+received_more:
 		if (likely(rv > 0)) {
 			received += rv;
 			buf	 += rv;
@@ -5636,6 +5637,11 @@ int drbd_asender(struct drbd_thread *thi)
 			expect	 = header_size;
 			cmd	 = NULL;
 		}
+		if (test_bit(SEND_PING, &connection->flags))
+			continue;
+		rv = drbd_recv_short(connection->meta.socket, buf, expect-received, MSG_DONTWAIT);
+		if (rv > 0)
+			goto received_more;
 	}
 
 	if (0) {
diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 3b74f08..3ed2d87 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -1994,22 +1994,13 @@ static bool dequeue_work_batch(struct drbd_work_queue *queue, struct list_head *
 	return !list_empty(work_list);
 }
 
-static bool dequeue_work_item(struct drbd_work_queue *queue, struct list_head *work_list)
-{
-	spin_lock_irq(&queue->q_lock);
-	if (!list_empty(&queue->q))
-		list_move(queue->q.next, work_list);
-	spin_unlock_irq(&queue->q_lock);
-	return !list_empty(work_list);
-}
-
 static void wait_for_work(struct drbd_connection *connection, struct list_head *work_list)
 {
 	DEFINE_WAIT(wait);
 	struct net_conf *nc;
 	int uncork, cork;
 
-	dequeue_work_item(&connection->sender_work, work_list);
+	dequeue_work_batch(&connection->sender_work, work_list);
 	if (!list_empty(work_list))
 		return;
 
-- 
1.9.1

[Drbd-dev] [PATCH 8/9] drbd: reduce lock contention in drbd_worker

[Philipp Reisner]

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

The worker may now dequeue work items in batches.
This should reduce lock contention during busy periods.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
---
 drivers/block/drbd/drbd_worker.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)

diff --git a/drivers/block/drbd/drbd_worker.c b/drivers/block/drbd/drbd_worker.c
index 3ed2d87..d2d1f97 100644
--- a/drivers/block/drbd/drbd_worker.c
+++ b/drivers/block/drbd/drbd_worker.c
@@ -2026,8 +2026,6 @@ static void wait_for_work(struct drbd_connection *connection, struct list_head *
 		prepare_to_wait(&connection->sender_work.q_wait, &wait, TASK_INTERRUPTIBLE);
 		spin_lock_irq(&connection->resource->req_lock);
 		spin_lock(&connection->sender_work.q_lock);	/* FIXME get rid of this one? */
-		/* dequeue single item only,
-		 * we still use drbd_queue_work_front() in some places */
 		if (!list_empty(&connection->sender_work.q))
 			list_splice_tail_init(&connection->sender_work.q, work_list);
 		spin_unlock(&connection->sender_work.q_lock);	/* FIXME get rid of this one? */
@@ -2114,7 +2112,7 @@ int drbd_worker(struct drbd_thread *thi)
 		if (get_t_state(thi) != RUNNING)
 			break;
 
-		while (!list_empty(&work_list)) {
+		if (!list_empty(&work_list)) {
 			w = list_first_entry(&work_list, struct drbd_work, list);
 			list_del_init(&w->list);
 			update_worker_timing_details(connection, w->cb);
@@ -2130,13 +2128,13 @@ int drbd_worker(struct drbd_thread *thi)
 			update_worker_timing_details(connection, do_unqueued_work);
 			do_unqueued_work(connection);
 		}
-		while (!list_empty(&work_list)) {
+		if (!list_empty(&work_list)) {
 			w = list_first_entry(&work_list, struct drbd_work, list);
 			list_del_init(&w->list);
 			update_worker_timing_details(connection, w->cb);
 			w->cb(w, 1);
-		}
-		dequeue_work_batch(&connection->sender_work, &work_list);
+		} else
+			dequeue_work_batch(&connection->sender_work, &work_list);
 	} while (!list_empty(&work_list) || test_bit(DEVICE_WORK_PENDING, &connection->flags));
 
 	rcu_read_lock();
-- 
1.9.1

[Drbd-dev] [PATCH 9/9] drbd: Add missing newline in resync progress display in /proc/drbd

[Philipp Reisner]

Was broken in 2010 with commit 4b0715f096

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

diff --git a/drivers/block/drbd/drbd_proc.c b/drivers/block/drbd/drbd_proc.c
index 06e6147..3b10fa6 100644
--- a/drivers/block/drbd/drbd_proc.c
+++ b/drivers/block/drbd/drbd_proc.c
@@ -142,10 +142,12 @@ static void drbd_syncer_progress(struct drbd_device *device, struct seq_file *se
 			    (unsigned long) Bit2KB(rs_left >> 10),
 			    (unsigned long) Bit2KB(rs_total >> 10));
 	else
-		seq_printf(seq, "(%lu/%lu)K\n\t",
+		seq_printf(seq, "(%lu/%lu)K",
 			    (unsigned long) Bit2KB(rs_left),
 			    (unsigned long) Bit2KB(rs_total));
 
+	seq_printf(seq, "\n\t");
+
 	/* see drivers/md/md.c
 	 * We do not want to overflow, so the order of operands and
 	 * the * 100 / 100 trick are important. We do a +1 to be
-- 
1.9.1

Re: [Drbd-dev] [PATCH 0/9] DRBD updates for 3.18

[Jens Axboe]

On 2014-09-11 06:29, Philipp Reisner wrote:
> Hi Jens,
>
> please consider to put these patches into your (future) for-3.18/drivers
> branch. I have posted it for review on August 28. It consists mainly of
> minor cleanups, and an important performance improvement. With that
> DRBD was seen to write with 160k IOPS, on SSDs.

Applied, thanks Philipp.

-- 
Jens Axboe