drbd: introduce P_ZEROES (REQ_OP_WRITE_ZEROES on the "wire")

And also re-enable partial-zero-out + discard aligned.

With the introduction of REQ_OP_WRITE_ZEROES,
we started to use that for both WRITE_ZEROES and DISCARDS,
hoping that WRITE_ZEROES would "do what we want",
UNMAP if possible, zero-out the rest.

The example scenario is some LVM "thin" backend.

While an un-allocated block on dm-thin reads as zeroes, on a dm-thin
with "skip_block_zeroing=true", after a partial block write allocated
that block, that same block may well map "undefined old garbage" from
the backends on LBAs that have not yet been written to.

If we cannot distinguish between zero-out and discard on the receiving
side, to avoid "undefined old garbage" to pop up randomly at later times
on supposedly zero-initialized blocks, we'd need to map all discards to
zero-out on the receiving side.  But that would potentially do a full
alloc on thinly provisioned backends, even when the expectation was to
unmap/trim/discard/de-allocate.

We need to distinguish on the protocol level, whether we need to guarantee
zeroes (and thus use zero-out, potentially doing the mentioned full-alloc),
or if we want to put the emphasis on discard, and only do a "best effort
zeroing" (by "discarding" blocks aligned to discard-granularity, and zeroing
only potential unaligned head and tail clippings to at least *try* to
avoid "false positives" in an online-verify later), hoping that someone
set skip_block_zeroing=false.

For some discussion regarding this on dm-devel, see also
https://www.mail-archive.com/dm-devel%40redhat.com/msg07965.html
https://www.redhat.com/archives/dm-devel/2018-January/msg00271.html

For backward compatibility, P_TRIM means zero-out, unless the
DRBD_FF_WZEROES feature flag is agreed upon during handshake.

To have upper layers even try to submit WRITE ZEROES requests,
we need to announce "efficient zeroout" independently.

We need to fixup max_write_zeroes_sectors after blk_queue_stack_limits():
if we can handle "zeroes" efficiently on the protocol,
we want to do that, even if our backend does not announce
max_write_zeroes_sectors itself.
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

drivers/block/drbd/drbd_debugfs.c

@@ -237,6 +237,8 @@ static void seq_print_peer_request_flags(struct seq_file *m, struct drbd_peer_re
 	seq_print_rq_state_bit(m, f & EE_CALL_AL_COMPLETE_IO, &sep, "in-AL");
 	seq_print_rq_state_bit(m, f & EE_SEND_WRITE_ACK, &sep, "C");
 	seq_print_rq_state_bit(m, f & EE_MAY_SET_IN_SYNC, &sep, "set-in-sync");
+	seq_print_rq_state_bit(m, f & EE_TRIM, &sep, "trim");
+	seq_print_rq_state_bit(m, f & EE_ZEROOUT, &sep, "zero-out");
 	seq_print_rq_state_bit(m, f & EE_WRITE_SAME, &sep, "write-same");
 	seq_putc(m, '\n');
 }

drivers/block/drbd/drbd_int.h

@@ -430,7 +430,11 @@ enum {
 	__EE_MAY_SET_IN_SYNC,
 
 	/* is this a TRIM aka REQ_OP_DISCARD? */
-	__EE_IS_TRIM,
+	__EE_TRIM,
+	/* explicit zero-out requested, or
+	 * our lower level cannot handle trim,
+	 * and we want to fall back to zeroout instead */
+	__EE_ZEROOUT,
 
 	/* In case a barrier failed,
 	 * we need to resubmit without the barrier flag. */
@@ -472,7 +476,8 @@ enum {
 };
 #define EE_CALL_AL_COMPLETE_IO (1<<__EE_CALL_AL_COMPLETE_IO)
 #define EE_MAY_SET_IN_SYNC     (1<<__EE_MAY_SET_IN_SYNC)
-#define EE_IS_TRIM             (1<<__EE_IS_TRIM)
+#define EE_TRIM                (1<<__EE_TRIM)
+#define EE_ZEROOUT             (1<<__EE_ZEROOUT)
 #define EE_RESUBMITTED         (1<<__EE_RESUBMITTED)
 #define EE_WAS_ERROR           (1<<__EE_WAS_ERROR)
 #define EE_HAS_DIGEST          (1<<__EE_HAS_DIGEST)
@@ -1556,6 +1561,8 @@ extern void start_resync_timer_fn(struct timer_list *t);
 extern void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req);
 
 /* drbd_receiver.c */
+extern int drbd_issue_discard_or_zero_out(struct drbd_device *device,
+		sector_t start, unsigned int nr_sectors, int flags);
 extern int drbd_receiver(struct drbd_thread *thi);
 extern int drbd_ack_receiver(struct drbd_thread *thi);
 extern void drbd_send_ping_wf(struct work_struct *ws);

drivers/block/drbd/drbd_main.c

@@ -1668,7 +1668,11 @@ static u32 bio_flags_to_wire(struct drbd_connection *connection,
 			(bio->bi_opf & REQ_PREFLUSH ? DP_FLUSH : 0) |
 			(bio_op(bio) == REQ_OP_WRITE_SAME ? DP_WSAME : 0) |
 			(bio_op(bio) == REQ_OP_DISCARD ? DP_DISCARD : 0) |
-			(bio_op(bio) == REQ_OP_WRITE_ZEROES ? DP_DISCARD : 0);
+			(bio_op(bio) == REQ_OP_WRITE_ZEROES ?
+			  ((connection->agreed_features & DRBD_FF_WZEROES) ?
+			   (DP_ZEROES |(!(bio->bi_opf & REQ_NOUNMAP) ? DP_DISCARD : 0))
+			   : DP_DISCARD)
+			: 0);
 	else
 		return bio->bi_opf & REQ_SYNC ? DP_RW_SYNC : 0;
 }
@@ -1712,10 +1716,11 @@ int drbd_send_dblock(struct drbd_peer_device *peer_device, struct drbd_request *
 	}
 	p->dp_flags = cpu_to_be32(dp_flags);
 
-	if (dp_flags & DP_DISCARD) {
+	if (dp_flags & (DP_DISCARD|DP_ZEROES)) {
+		enum drbd_packet cmd = (dp_flags & DP_ZEROES) ? P_ZEROES : P_TRIM;
 		struct p_trim *t = (struct p_trim*)p;
 		t->size = cpu_to_be32(req->i.size);
-		err = __send_command(peer_device->connection, device->vnr, sock, P_TRIM, sizeof(*t), NULL, 0);
+		err = __send_command(peer_device->connection, device->vnr, sock, cmd, sizeof(*t), NULL, 0);
 		goto out;
 	}
 	if (dp_flags & DP_WSAME) {

drivers/block/drbd/drbd_nl.c

@@ -1261,6 +1261,21 @@ static void fixup_discard_if_not_supported(struct request_queue *q)
 	}
 }
 
+static void fixup_write_zeroes(struct drbd_device *device, struct request_queue *q)
+{
+	/* Fixup max_write_zeroes_sectors after blk_queue_stack_limits():
+	 * if we can handle "zeroes" efficiently on the protocol,
+	 * we want to do that, even if our backend does not announce
+	 * max_write_zeroes_sectors itself. */
+	struct drbd_connection *connection = first_peer_device(device)->connection;
+	/* If the peer announces WZEROES support, use it.  Otherwise, rather
+	 * send explicit zeroes than rely on some discard-zeroes-data magic. */
+	if (connection->agreed_features & DRBD_FF_WZEROES)
+		q->limits.max_write_zeroes_sectors = DRBD_MAX_BBIO_SECTORS;
+	else
+		q->limits.max_write_zeroes_sectors = 0;
+}
+
 static void decide_on_write_same_support(struct drbd_device *device,
 			struct request_queue *q,
 			struct request_queue *b, struct o_qlim *o,
@@ -1371,6 +1386,7 @@ static void drbd_setup_queue_param(struct drbd_device *device, struct drbd_backi
 		}
 	}
 	fixup_discard_if_not_supported(q);
+	fixup_write_zeroes(device, q);
 }
 
 void drbd_reconsider_queue_parameters(struct drbd_device *device, struct drbd_backing_dev *bdev, struct o_qlim *o)

drivers/block/drbd/drbd_protocol.h

@@ -70,6 +70,11 @@ enum drbd_packet {
 	 * we may fall back to an opencoded loop instead. */
 	P_WSAME               = 0x34,
 
+	/* 0x35 already claimed in DRBD 9 */
+	P_ZEROES              = 0x36, /* data sock: zero-out, WRITE_ZEROES */
+
+	/* 0x40 .. 0x48 already claimed in DRBD 9 */
+
 	P_MAY_IGNORE	      = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */
 	P_MAX_OPT_CMD	      = 0x101,
 
@@ -130,6 +135,12 @@ struct p_header100 {
 #define DP_SEND_RECEIVE_ACK 128 /* This is a proto B write request */
 #define DP_SEND_WRITE_ACK   256 /* This is a proto C write request */
 #define DP_WSAME            512 /* equiv. REQ_WRITE_SAME */
+#define DP_ZEROES          1024 /* equiv. REQ_OP_WRITE_ZEROES */
+
+/* possible combinations:
+ * REQ_OP_WRITE_ZEROES:  DP_DISCARD | DP_ZEROES
+ * REQ_OP_WRITE_ZEROES + REQ_NOUNMAP: DP_ZEROES
+ */
 
 struct p_data {
 	u64	    sector;    /* 64 bits sector number */
@@ -197,6 +208,42 @@ struct p_block_req {
  */
 #define DRBD_FF_WSAME 4
 
+/* supports REQ_OP_WRITE_ZEROES on the "wire" protocol.
+ *
+ * We used to map that to "discard" on the sending side, and if we cannot
+ * guarantee that discard zeroes data, the receiving side would map discard
+ * back to zero-out.
+ *
+ * With the introduction of REQ_OP_WRITE_ZEROES,
+ * we started to use that for both WRITE_ZEROES and DISCARDS,
+ * hoping that WRITE_ZEROES would "do what we want",
+ * UNMAP if possible, zero-out the rest.
+ *
+ * The example scenario is some LVM "thin" backend.
+ *
+ * While an un-allocated block on dm-thin reads as zeroes, on a dm-thin
+ * with "skip_block_zeroing=true", after a partial block write allocated
+ * that block, that same block may well map "undefined old garbage" from
+ * the backends on LBAs that have not yet been written to.
+ *
+ * If we cannot distinguish between zero-out and discard on the receiving
+ * side, to avoid "undefined old garbage" to pop up randomly at later times
+ * on supposedly zero-initialized blocks, we'd need to map all discards to
+ * zero-out on the receiving side.  But that would potentially do a full
+ * alloc on thinly provisioned backends, even when the expectation was to
+ * unmap/trim/discard/de-allocate.
+ *
+ * We need to distinguish on the protocol level, whether we need to guarantee
+ * zeroes (and thus use zero-out, potentially doing the mentioned full-alloc),
+ * or if we want to put the emphasis on discard, and only do a "best effort
+ * zeroing" (by "discarding" blocks aligned to discard-granularity, and zeroing
+ * only potential unaligned head and tail clippings), to at least *try* to
+ * avoid "false positives" in an online-verify later, hoping that someone
+ * set skip_block_zeroing=false.
+ */
+#define DRBD_FF_WZEROES 8
+
+
 struct p_connection_features {
 	u32 protocol_min;
 	u32 feature_flags;

drivers/block/drbd/drbd_req.c

@@ -63,7 +63,7 @@ static struct drbd_request *drbd_req_new(struct drbd_device *device, struct bio
 	drbd_req_make_private_bio(req, bio_src);
 	req->rq_state = (bio_data_dir(bio_src) == WRITE ? RQ_WRITE : 0)
 		      | (bio_op(bio_src) == REQ_OP_WRITE_SAME ? RQ_WSAME : 0)
-		      | (bio_op(bio_src) == REQ_OP_WRITE_ZEROES ? RQ_UNMAP : 0)
+		      | (bio_op(bio_src) == REQ_OP_WRITE_ZEROES ? RQ_ZEROES : 0)
 		      | (bio_op(bio_src) == REQ_OP_DISCARD ? RQ_UNMAP : 0);
 	req->device = device;
 	req->master_bio = bio_src;
@@ -1155,12 +1155,11 @@ static int drbd_process_write_request(struct drbd_request *req)
 	return remote;
 }
 
-static void drbd_process_discard_req(struct drbd_request *req)
+static void drbd_process_discard_or_zeroes_req(struct drbd_request *req, int flags)
 {
-	struct block_device *bdev = req->device->ldev->backing_bdev;
-
-	if (blkdev_issue_zeroout(bdev, req->i.sector, req->i.size >> 9,
-			GFP_NOIO, 0))
+	int err = drbd_issue_discard_or_zero_out(req->device,
+				req->i.sector, req->i.size >> 9, flags);
+	if (err)
 		req->private_bio->bi_status = BLK_STS_IOERR;
 	bio_endio(req->private_bio);
 }
@@ -1189,9 +1188,11 @@ drbd_submit_req_private_bio(struct drbd_request *req)
 	if (get_ldev(device)) {
 		if (drbd_insert_fault(device, type))
 			bio_io_error(bio);
-		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES ||
-			 bio_op(bio) == REQ_OP_DISCARD)
-			drbd_process_discard_req(req);
+		else if (bio_op(bio) == REQ_OP_WRITE_ZEROES)
+			drbd_process_discard_or_zeroes_req(req, EE_ZEROOUT |
+			    ((bio->bi_opf & REQ_NOUNMAP) ? 0 : EE_TRIM));
+		else if (bio_op(bio) == REQ_OP_DISCARD)
+			drbd_process_discard_or_zeroes_req(req, EE_TRIM);
 		else
 			generic_make_request(bio);
 		put_ldev(device);

drivers/block/drbd/drbd_req.h

@@ -208,6 +208,7 @@ enum drbd_req_state_bits {
 	__RQ_WRITE,
 	__RQ_WSAME,
 	__RQ_UNMAP,
+	__RQ_ZEROES,
 
 	/* Should call drbd_al_complete_io() for this request... */
 	__RQ_IN_ACT_LOG,
@@ -253,6 +254,7 @@ enum drbd_req_state_bits {
 #define RQ_WRITE           (1UL << __RQ_WRITE)
 #define RQ_WSAME           (1UL << __RQ_WSAME)
 #define RQ_UNMAP           (1UL << __RQ_UNMAP)
+#define RQ_ZEROES          (1UL << __RQ_ZEROES)
 #define RQ_IN_ACT_LOG      (1UL << __RQ_IN_ACT_LOG)
 #define RQ_UNPLUG          (1UL << __RQ_UNPLUG)
 #define RQ_POSTPONED	   (1UL << __RQ_POSTPONED)

drivers/block/drbd/drbd_worker.c

@@ -153,7 +153,7 @@ void drbd_endio_write_sec_final(struct drbd_peer_request *peer_req) __releases(l
 	do_wake = list_empty(block_id == ID_SYNCER ? &device->sync_ee : &device->active_ee);
 
 	/* FIXME do we want to detach for failed REQ_OP_DISCARD?
-	 * ((peer_req->flags & (EE_WAS_ERROR|EE_IS_TRIM)) == EE_WAS_ERROR) */
+	 * ((peer_req->flags & (EE_WAS_ERROR|EE_TRIM)) == EE_WAS_ERROR) */
 	if (peer_req->flags & EE_WAS_ERROR)
 		__drbd_chk_io_error(device, DRBD_WRITE_ERROR);
 

include/linux/drbd.h

@@ -51,7 +51,7 @@
 #endif
 
 extern const char *drbd_buildtag(void);
-#define REL_VERSION "8.4.10"
+#define REL_VERSION "8.4.11"
 #define API_VERSION 1
 #define PRO_VERSION_MIN 86
 #define PRO_VERSION_MAX 101