[PATCH] drbd: Support zeroout device instead of initial full sync
From: Nick Wang
Date: Thu Aug 06 2015 - 06:05:07 EST
Patch set for zeroing out device on both side
instead of initial full sync. Useful for high
latency network environment.
Implement --zeroout-devices and --discard-devices
for new-current-uuid
Signed-off-by: Nick Wang <nwang@xxxxxxxx>
CC: Philipp Reisner <philipp.reisner@xxxxxxxxxx>
CC: Lars Ellenberg <lars.ellenberg@xxxxxxxxxx>
CC: drbd-dev@xxxxxxxxxxxxxxxx
CC: linux-kernel@xxxxxxxxxxxxxxx
---
drbd/drbd_int.h | 15 +++++++
drbd/drbd_main.c | 60 +++++++++++++++++++++++++++-
drbd/drbd_nl.c | 41 +++++++++++++++++--
drbd/drbd_protocol.h | 2 +
drbd/drbd_receiver.c | 86 +++++++++++++++++++++++++++++++++++++++-
drbd/drbd_worker.c | 105 +++++++++++++++++++++++++++++++++++++++++++++++++
drbd/linux/drbd_genl.h | 2 +
7 files changed, 305 insertions(+), 6 deletions(-)
diff --git a/drbd/drbd_int.h b/drbd/drbd_int.h
index d1e2bc0..555b24c 100644
--- a/drbd/drbd_int.h
+++ b/drbd/drbd_int.h
@@ -621,6 +621,13 @@ enum {
RS_START, /* tell worker to start resync/OV */
RS_PROGRESS, /* tell worker that resync made significant progress */
RS_DONE, /* tell worker that resync is done */
+ P_ZERO_START, /* tell worker to zero out device */
+ S_ZERO_START, /* tell worker to zero out device as requested*/
+
+ /* used for zero out/discard device */
+ DISCARD_DISK, /* flag to discard device */
+ ZERO_DONE, /* succeed on zero out a device */
+ ZERO_FAIL, /* fail to zero out a device */
};
struct drbd_bitmap; /* opaque for drbd_device */
@@ -1204,6 +1211,11 @@ extern int __drbd_send_protocol(struct drbd_connection *connection, enum drbd_pa
extern int drbd_send_protocol(struct drbd_connection *connection);
extern int drbd_send_uuids(struct drbd_peer_device *);
extern int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *);
+extern int drbd_send_zeroout_start(struct drbd_peer_device *);
+extern int drbd_send_discard_start(struct drbd_peer_device *);
+extern int drbd_send_zeroout_finish(struct drbd_peer_device *);
+extern int drbd_send_zeroout_ok(struct drbd_peer_device *);
+extern int drbd_send_zeroout_fail(struct drbd_peer_device *);
extern void drbd_gen_and_send_sync_uuid(struct drbd_peer_device *);
extern int drbd_send_sizes(struct drbd_peer_device *peer_device, int trigger_reply, enum dds_flags flags);
extern int drbd_send_state(struct drbd_peer_device *, union drbd_state);
@@ -1661,6 +1673,9 @@ extern void drbd_send_acks_wf(struct work_struct *ws);
extern bool drbd_rs_c_min_rate_throttle(struct drbd_device *device);
extern bool drbd_rs_should_slow_down(struct drbd_device *device, sector_t sector,
bool throttle_if_app_is_waiting);
+extern int zeroout_local_device(struct drbd_device *device, bool discard);
+extern void require_zeroout_local_device(struct drbd_device *device);
+extern void receive_zeroout_local_device(struct drbd_device *device);
extern int drbd_submit_peer_request(struct drbd_device *,
struct drbd_peer_request *, const unsigned,
const int);
diff --git a/drbd/drbd_main.c b/drbd/drbd_main.c
index 31bf43f..badc719 100644
--- a/drbd/drbd_main.c
+++ b/drbd/drbd_main.c
@@ -908,6 +908,62 @@ int drbd_send_uuids_skip_initial_sync(struct drbd_peer_device *peer_device)
return _drbd_send_uuids(peer_device, 8);
}
+
+/**
+ * drbd_send_zeroout_start() - Notify peer node to zeroout device
+ * @peer_device: DRBD peer device.
+ */
+int drbd_send_zeroout_start(struct drbd_peer_device *peer_device)
+{
+ return _drbd_send_uuids(peer_device, 16);
+}
+
+/**
+ * drbd_send_discard_start() - Notify peer node to discard device
+ * @peer_device: DRBD peer device.
+ */
+int drbd_send_discard_start(struct drbd_peer_device *peer_device)
+{
+ return _drbd_send_uuids(peer_device, 32);
+}
+
+/**
+ * drbd_send_zeroout_finish() - Notify both node finished zeroing out
+ * @peer_device: DRBD peer device.
+ */
+int drbd_send_zeroout_finish(struct drbd_peer_device *peer_device)
+{
+ return _drbd_send_uuids(peer_device, 64);
+}
+
+/**
+ * _drbd_send_zeroout_state() - Sends the drbd state to the peer
+ * @peer_device: DRBD peer device.
+ * @state: Device zero out status.
+ */
+static int _drbd_send_zeroout_state(struct drbd_peer_device *peer_device, unsigned int status)
+{
+ struct drbd_socket *sock;
+ struct p_state *p;
+
+ sock = &peer_device->connection->data;
+ p = drbd_prepare_command(peer_device, sock);
+ if (!p)
+ return -EIO;
+ p->state = cpu_to_be32(status);
+ return drbd_send_command(peer_device, sock, P_ZERO_OUT, sizeof(*p), NULL, 0);
+}
+
+int drbd_send_zeroout_ok(struct drbd_peer_device *peer_device)
+{
+ return _drbd_send_zeroout_state(peer_device, 0);
+}
+
+int drbd_send_zeroout_fail(struct drbd_peer_device *peer_device)
+{
+ return _drbd_send_zeroout_state(peer_device, 1);
+}
+
void drbd_print_uuids(struct drbd_device *device, const char *text)
{
if (get_ldev_if_state(device, D_NEGOTIATING)) {
@@ -3466,8 +3522,8 @@ void drbd_uuid_move_history(struct drbd_device *device) __must_hold(local)
{
int i;
- for (i = UI_HISTORY_START; i < UI_HISTORY_END; i++)
- device->ldev->md.uuid[i+1] = device->ldev->md.uuid[i];
+ for (i = UI_HISTORY_END; i > UI_HISTORY_START; i--)
+ device->ldev->md.uuid[i] = device->ldev->md.uuid[i-1];
}
void __drbd_uuid_set(struct drbd_device *device, int idx, u64 val) __must_hold(local)
diff --git a/drbd/drbd_nl.c b/drbd/drbd_nl.c
index bb7e1b0..dc2bfec 100644
--- a/drbd/drbd_nl.c
+++ b/drbd/drbd_nl.c
@@ -4015,8 +4015,11 @@ int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
{
struct drbd_config_context adm_ctx;
struct drbd_device *device;
+ struct drbd_peer_device *peer_device = NULL;
enum drbd_ret_code retcode;
int skip_initial_sync = 0;
+ int zeroout_devices = 0;
+ int discard_devices = 0;
int err;
struct new_c_uuid_parms args;
@@ -4045,12 +4048,28 @@ int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
goto out;
}
+ peer_device = first_peer_device(device);
+
/* this is "skip initial sync", assume to be clean */
if (device->state.conn == C_CONNECTED &&
- first_peer_device(device)->connection->agreed_pro_version >= 90 &&
+ peer_device->connection->agreed_pro_version >= 90 &&
device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED && args.clear_bm) {
drbd_info(device, "Preparing to skip initial sync\n");
skip_initial_sync = 1;
+ /* this is "zero out/discard" devices to make it all zero.
+ * ignore "zero out" if both "clear_bm" and "zeroout_devices/discard_devices" set. */
+ } else if (device->state.conn == C_CONNECTED &&
+ peer_device->connection->agreed_features & FF_DISCARD &&
+ device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
+ args.zeroout_devices) {
+ drbd_info(device, "Preparing to zero out devices, will take a long time\n");
+ zeroout_devices = 1;
+ } else if (device->state.conn == C_CONNECTED &&
+ peer_device->connection->agreed_features & FF_DISCARD &&
+ device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
+ args.discard_devices) {
+ drbd_info(device, "Preparing to discard devices, will take a long time\n");
+ discard_devices = 1;
} else if (device->state.conn != C_STANDALONE) {
retcode = ERR_CONNECTED;
goto out_dec;
@@ -4059,7 +4078,7 @@ int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
drbd_uuid_set(device, UI_BITMAP, 0); /* Rotate UI_BITMAP to History 1, etc... */
drbd_uuid_new_current(device); /* New current, previous to UI_BITMAP */
- if (args.clear_bm) {
+ if (args.clear_bm || args.zeroout_devices || args.discard_devices) {
err = drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
"clear_n_write from new_c_uuid", BM_LOCKED_MASK);
if (err) {
@@ -4067,7 +4086,7 @@ int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
retcode = ERR_IO_MD_DISK;
}
if (skip_initial_sync) {
- drbd_send_uuids_skip_initial_sync(first_peer_device(device));
+ drbd_send_uuids_skip_initial_sync(peer_device);
_drbd_uuid_set(device, UI_BITMAP, 0);
drbd_print_uuids(device, "cleared bitmap UUID");
spin_lock_irq(&device->resource->req_lock);
@@ -4075,6 +4094,22 @@ int drbd_adm_new_c_uuid(struct sk_buff *skb, struct genl_info *info)
CS_VERBOSE, NULL);
spin_unlock_irq(&device->resource->req_lock);
}
+ if (zeroout_devices || discard_devices) {
+ if (discard_devices) {
+ drbd_send_discard_start(peer_device);
+ set_bit(DISCARD_DISK, &device->flags);
+ } else {
+ drbd_send_zeroout_start(peer_device);
+ clear_bit(DISCARD_DISK, &device->flags);
+ }
+ _drbd_uuid_set(device, UI_BITMAP, 0);
+ drbd_print_uuids(device, "cleared bitmap UUID for zeroing device");
+
+ /* CLear bit flag of zero out */
+ clear_bit(ZERO_DONE, &device->flags);
+ clear_bit(ZERO_FAIL, &device->flags);
+ drbd_device_post_work(device, P_ZERO_START);
+ }
}
drbd_md_sync(device);
diff --git a/drbd/drbd_protocol.h b/drbd/drbd_protocol.h
index 405b181..d9db0ce 100644
--- a/drbd/drbd_protocol.h
+++ b/drbd/drbd_protocol.h
@@ -59,6 +59,7 @@ enum drbd_packet {
/* REQ_DISCARD. We used "discard" in different contexts before,
* which is why I chose TRIM here, to disambiguate. */
P_TRIM = 0x31,
+ P_ZERO_OUT = 0x32,
P_MAY_IGNORE = 0x100, /* Flag to test if (cmd > P_MAY_IGNORE) ... */
P_MAX_OPT_CMD = 0x101,
@@ -161,6 +162,7 @@ struct p_block_req {
*/
#define FF_TRIM 1
+#define FF_DISCARD 4
struct p_connection_features {
u32 protocol_min;
diff --git a/drbd/drbd_receiver.c b/drbd/drbd_receiver.c
index 06e5667..ec324ef 100644
--- a/drbd/drbd_receiver.c
+++ b/drbd/drbd_receiver.c
@@ -47,7 +47,7 @@
#include "drbd_vli.h"
#include <linux/scatterlist.h>
-#define PRO_FEATURES (FF_TRIM)
+#define PRO_FEATURES (FF_TRIM | FF_DISCARD)
struct flush_work {
struct drbd_work w;
@@ -4317,6 +4317,20 @@ static int receive_uuids(struct drbd_connection *connection, struct packet_info
peer_device->connection->agreed_pro_version >= 90 &&
device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
(p_uuid[UI_FLAGS] & 8);
+ int zeroout_devices =
+ device->state.conn == C_CONNECTED &&
+ peer_device->connection->agreed_pro_version >= 90 &&
+ device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
+ (p_uuid[UI_FLAGS] & 16);
+ int discard_devices =
+ device->state.conn == C_CONNECTED &&
+ peer_device->connection->agreed_pro_version >= 90 &&
+ device->ldev->md.uuid[UI_CURRENT] == UUID_JUST_CREATED &&
+ (p_uuid[UI_FLAGS] & 32);
+ int zeroout_finish =
+ device->state.conn == C_CONNECTED &&
+ peer_device->connection->agreed_pro_version >= 90 &&
+ (p_uuid[UI_FLAGS] & 64);
if (skip_initial_sync) {
drbd_info(device, "Accepted new current UUID, preparing to skip initial sync\n");
drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
@@ -4324,11 +4338,42 @@ static int receive_uuids(struct drbd_connection *connection, struct packet_info
BM_LOCKED_TEST_ALLOWED);
_drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]);
_drbd_uuid_set(device, UI_BITMAP, 0);
+ spin_lock_irq(&device->resource->req_lock);
_drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
CS_VERBOSE, NULL);
+ spin_unlock_irq(&device->resource->req_lock);
drbd_md_sync(device);
updated_uuids = 1;
}
+
+ if (zeroout_devices || discard_devices) {
+ if (discard_devices) {
+ drbd_info(device, "Accepted to discard devices, will take a long time\n");
+ set_bit(DISCARD_DISK, &device->flags);
+ } else {
+ drbd_info(device, "Accepted to zeroout devices, will take a long time\n");
+ clear_bit(DISCARD_DISK, &device->flags);
+ }
+
+ drbd_bitmap_io(device, &drbd_bmio_clear_n_write,
+ "clear_n_write from receive_uuids",
+ BM_LOCKED_TEST_ALLOWED);
+ _drbd_uuid_set(device, UI_CURRENT, p_uuid[UI_CURRENT]);
+ _drbd_uuid_set(device, UI_BITMAP, 0);
+ drbd_print_uuids(device, "cleared bitmap UUID for zeroing device");
+
+ drbd_device_post_work(device, S_ZERO_START);
+ updated_uuids = 1;
+ }
+
+ if (zeroout_finish) {
+ drbd_info(device, "Both side finished zero out devices.\n");
+ spin_lock_irq(&device->resource->req_lock);
+ _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE, pdsk, D_UP_TO_DATE),
+ CS_VERBOSE, NULL);
+ spin_unlock_irq(&device->resource->req_lock);
+ }
+
put_ldev(device);
} else if (device->state.disk < D_INCONSISTENT &&
device->state.role == R_PRIMARY) {
@@ -4383,6 +4428,41 @@ static union drbd_state convert_state(union drbd_state ps)
return ms;
}
+static int receive_zeroout_state(struct drbd_connection *connection, struct packet_info *pi)
+{
+ struct drbd_peer_device *peer_device;
+ struct drbd_device *device;
+ struct p_state *p = pi->data;
+ unsigned int isfail;
+
+ peer_device = conn_peer_device(connection, pi->vnr);
+ if (!peer_device)
+ return -EIO;
+ device = peer_device->device;
+
+ isfail = be32_to_cpu(p->state);
+
+ if (isfail) {
+ drbd_info(device, "Failed to zero out peer device\n");
+ set_bit(ZERO_FAIL, &device->flags);
+ } else {
+ drbd_info(device, "Finished zero out peer device\n");
+ if (test_and_clear_bit(ZERO_DONE, &device->flags)) {
+ drbd_info(device, "Both side finished zeroing.\n");
+ spin_lock_irq(&device->resource->req_lock);
+ _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE,
+ pdsk, D_UP_TO_DATE), CS_VERBOSE, NULL);
+ spin_unlock_irq(&device->resource->req_lock);
+ drbd_send_zeroout_finish(peer_device);
+ } else {
+ /* waiting for local device finished */
+ set_bit(ZERO_DONE, &device->flags);
+ }
+ }
+
+ return 0;
+}
+
static int receive_req_state(struct drbd_connection *connection, struct packet_info *pi)
{
struct drbd_peer_device *peer_device;
@@ -5008,6 +5088,7 @@ static struct data_cmd drbd_cmd_handler[] = {
[P_CONN_ST_CHG_REQ] = { 0, sizeof(struct p_req_state), receive_req_conn_state },
[P_PROTOCOL_UPDATE] = { 1, sizeof(struct p_protocol), receive_protocol },
[P_TRIM] = { 0, sizeof(struct p_trim), receive_Data },
+ [P_ZERO_OUT] = { 0, sizeof(struct p_state), receive_zeroout_state },
};
static void drbdd(struct drbd_connection *connection)
@@ -5287,6 +5368,9 @@ static int drbd_do_features(struct drbd_connection *connection)
drbd_info(connection, "Agreed to%ssupport TRIM on protocol level\n",
connection->agreed_features & FF_TRIM ? " " : " not ");
+ drbd_info(connection, "Agreed to%ssupport DISCARD DEVICE on protocol level\n",
+ connection->agreed_features & FF_DISCARD ? " " : " not ");
+
return 1;
incompat:
diff --git a/drbd/drbd_worker.c b/drbd/drbd_worker.c
index 2a15aeb..a8e231f 100644
--- a/drbd/drbd_worker.c
+++ b/drbd/drbd_worker.c
@@ -1653,6 +1653,105 @@ void drbd_rs_controller_reset(struct drbd_device *device)
rcu_read_unlock();
}
+/**
+ * zeroout_local_device()
+ * @device: DRBD device.
+ * @discard: whether to discard the block range.
+ *
+ * Description:
+ * Zero out drbd backing device when creating new uuid.
+ *
+**/
+int zeroout_local_device(struct drbd_device *device, bool discard)
+{
+ struct block_device *bdev;
+
+ bdev = device->ldev->backing_bdev;
+ if (device->ldev->known_size != drbd_get_capacity(bdev))
+ device->ldev->known_size = drbd_get_capacity(bdev);
+
+ if (discard){
+ /* zero out the backing device by discarding blocks */
+ return blkdev_issue_zeroout(bdev, 0,
+ device->ldev->known_size, GFP_NOIO, true);
+ } else {
+ /* zero out the backing device with WRITE call*/
+ return blkdev_issue_zeroout(bdev, 0,
+ device->ldev->known_size, GFP_NOIO, false);
+ }
+}
+
+/**
+ * require_zeroout_local_device()
+ * @device: DRBD device.
+ *
+ * Description:
+ * Start to zero out local device. Update
+ * status if peer node (secondary) finished
+ * zeroing.
+ *
+**/
+void require_zeroout_local_device(struct drbd_device *device)
+{
+ int zeroout_err = 0;
+
+ if (test_and_clear_bit(DISCARD_DISK, &device->flags)) {
+ zeroout_err = zeroout_local_device(device, true);
+ } else {
+ zeroout_err = zeroout_local_device(device, false);
+ }
+
+ if (zeroout_err) {
+ drbd_err(device, "Failed to zero out local device\n");
+ set_bit(ZERO_FAIL, &device->flags);
+ drbd_chk_io_error(device, 1, DRBD_WRITE_ERROR);
+ } else {
+ drbd_info(device, "Finished zero out local device.\n");
+
+ if (test_and_clear_bit(ZERO_DONE, &device->flags)) {
+ spin_lock_irq(&device->resource->req_lock);
+ _drbd_set_state(_NS2(device, disk, D_UP_TO_DATE,
+ pdsk, D_UP_TO_DATE), CS_VERBOSE, NULL);
+ spin_unlock_irq(&device->resource->req_lock);
+ drbd_send_zeroout_finish(first_peer_device(device));
+ } else if (test_and_clear_bit(ZERO_FAIL, &device->flags)) {
+ drbd_info(device, "Peer device has already failed on zero out\n");
+ } else {
+ /* waiting for peer device finished */
+ set_bit(ZERO_DONE, &device->flags);
+ }
+ }
+}
+
+/**
+ * receive_zeroout_local_device()
+ * @device: DRBD device.
+ *
+ * Description:
+ * Start to zero out local device.
+ * Notify peer node the zeroing result.
+ *
+**/
+void receive_zeroout_local_device(struct drbd_device *device)
+{
+ int zeroout_err = 0;
+ struct drbd_peer_device *const peer_device = first_peer_device(device);
+
+ if (test_and_clear_bit(DISCARD_DISK, &device->flags)) {
+ zeroout_err = zeroout_local_device(device, true);
+ } else {
+ zeroout_err = zeroout_local_device(device, false);
+ }
+ if (zeroout_err) {
+ drbd_err(device, "Failed to zero out local device\n");
+ drbd_send_zeroout_fail(peer_device);
+ drbd_chk_io_error(device, 1, DRBD_WRITE_ERROR);
+ } else {
+ drbd_info(device, "Finished zero out local device.\n");
+ drbd_send_zeroout_ok(peer_device);
+ }
+}
+
void start_resync_timer_fn(unsigned long data)
{
struct drbd_device *device = (struct drbd_device *) data;
@@ -1986,6 +2085,10 @@ static void do_device_work(struct drbd_device *device, const unsigned long todo)
drbd_ldev_destroy(device);
if (test_bit(RS_START, &todo))
do_start_resync(device);
+ if (test_bit(P_ZERO_START, &todo))
+ require_zeroout_local_device(device);
+ if (test_bit(S_ZERO_START, &todo))
+ receive_zeroout_local_device(device);
}
#define DRBD_DEVICE_WORK_MASK \
@@ -1995,6 +2098,8 @@ static void do_device_work(struct drbd_device *device, const unsigned long todo)
|(1UL << RS_START) \
|(1UL << RS_PROGRESS) \
|(1UL << RS_DONE) \
+ |(1UL << P_ZERO_START) \
+ |(1UL << S_ZERO_START) \
)
static unsigned long get_work_bits(unsigned long *flags)
diff --git a/drbd/linux/drbd_genl.h b/drbd/linux/drbd_genl.h
index 5db53f5..5a31a5c 100644
--- a/drbd/linux/drbd_genl.h
+++ b/drbd/linux/drbd_genl.h
@@ -240,6 +240,8 @@ GENL_struct(DRBD_NLA_START_OV_PARMS, 9, start_ov_parms,
GENL_struct(DRBD_NLA_NEW_C_UUID_PARMS, 10, new_c_uuid_parms,
__flg_field(1, DRBD_GENLA_F_MANDATORY, clear_bm)
+ __flg_field(2, 0, zeroout_devices)
+ __flg_field(3, 0, discard_devices)
)
GENL_struct(DRBD_NLA_TIMEOUT_PARMS, 11, timeout_parms,
--
2.1.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/