modified workqueue calls on Hayakawa's comments
In the cluster evironment, cluster write has poor performance.
Because userspace_flush has to contact userspace program(cmirrord)
in clear/mark/flush request. But both mark and flush requests
require cmirrord to circle message to all the cluster nodes in each
flush call. This behave is realy slow. So the idea is merging
mark and flush request together to reduce the kernel-userspace-kernel
time. Add a new flag DM_INTEGRATED_FLUSH to tell userspace application
that kernel has a flush with mark_region data. Besides, when only sending
clear request, the flush request could be delayed.
Added a workqueue to run delayed flush request.
Signed-off-by: dongmao zhang <dmzhang@xxxxxxxx>
---
drivers/md/dm-log-userspace-base.c | 117 ++++++++++++++++++++++++++++-----
include/uapi/linux/dm-log-userspace.h | 10 ++-
2 files changed, 105 insertions(+), 22 deletions(-)
diff --git a/drivers/md/dm-log-userspace-base.c b/drivers/md/dm-log-userspace-base.c
index 9429159..08b0e69 100644
--- a/drivers/md/dm-log-userspace-base.c
+++ b/drivers/md/dm-log-userspace-base.c
@@ -11,9 +11,10 @@
#include <linux/dm-log-userspace.h>
#include <linux/module.h>
+#include <linux/workqueue.h>
#include "dm-log-userspace-transfer.h"
-#define DM_LOG_USERSPACE_VSN "1.1.0"
+#define DM_LOG_USERSPACE_VSN "1.2.0"
struct flush_entry {
int type;
@@ -58,6 +59,12 @@ struct log_c {
spinlock_t flush_lock;
struct list_head mark_list;
struct list_head clear_list;
+
+ /*work queue for flush clear region*/
+ struct workqueue_struct *dmlog_wq;
+ struct delayed_work flush_log_work;
+ atomic_t sched_flush;
+ uint32_t integrated_flush;
};
static mempool_t *flush_entry_pool;
@@ -141,6 +148,17 @@ static int build_constructor_string(struct dm_target *ti,
return str_size;
}
+static void do_flush(struct delayed_work *work)
+{
+ int r;
+ struct log_c *lc = container_of(work, struct log_c, flush_log_work);
+ r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
+ NULL, 0, NULL, NULL);
+ atomic_set(&lc->sched_flush, 0);
+ if (r)
+ dm_table_event(lc->ti->table);
+}
+
/*
* userspace_ctr
*
@@ -199,6 +217,10 @@ static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
return str_size;
}
+ lc->integrated_flush = 0;
+ if (strstr(ctr_str, "integrated_flush"))
+ lc->integrated_flush = 1;
+
devices_rdata = kzalloc(devices_rdata_size, GFP_KERNEL);
if (!devices_rdata) {
DMERR("Failed to allocate memory for device information");
@@ -246,6 +268,20 @@ static int userspace_ctr(struct dm_dirty_log *log, struct dm_target *ti,
DMERR("Failed to register %s with device-mapper",
devices_rdata);
}
+
+
+ if (lc->integrated_flush) {
+ lc->dmlog_wq = alloc_workqueue("dmlogd", WQ_MEM_RECLAIM, 0);
+ if (!lc->dmlog_wq) {
+ DMERR("couldn't start dmlogd");
+ r = -ENOMEM;
+ goto out;
+ }
+
+ INIT_DELAYED_WORK(&lc->flush_log_work, do_flush);
+ atomic_set(&lc->sched_flush, 0);
+ }
+
out:
kfree(devices_rdata);
if (r) {
@@ -264,6 +300,14 @@ static void userspace_dtr(struct dm_dirty_log *log)
{
struct log_c *lc = log->context;
+ if (lc->integrated_flush) {
+ /* flush workqueue */
+ if (atomic_read(&lc->sched_flush))
+ flush_delayed_work(&lc->flush_log_work);
+
+ destroy_workqueue(lc->dmlog_wq);
+ }
+
(void) dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_DTR,
NULL, 0,
NULL, NULL);
@@ -294,6 +338,10 @@ static int userspace_postsuspend(struct dm_dirty_log *log)
int r;
struct log_c *lc = log->context;
+ /* run planed flush earlier */
+ if (lc->integrated_flush && atomic_read(&lc->sched_flush))
+ flush_delayed_work(&lc->flush_log_work);
+
r = dm_consult_userspace(lc->uuid, lc->luid, DM_ULOG_POSTSUSPEND,
NULL, 0,
NULL, NULL);
@@ -405,7 +453,8 @@ static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
return r;
}
-static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
+static int flush_by_group(struct log_c *lc, struct list_head *flush_list,
+ int flush_with_payload)
{
int r = 0;
int count;
@@ -431,15 +480,25 @@ static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
break;
}
- r = userspace_do_request(lc, lc->uuid, type,
- (char *)(group),
- count * sizeof(uint64_t),
- NULL, NULL);
- if (r) {
- /* Group send failed. Attempt one-by-one. */
- list_splice_init(&tmp_list, flush_list);
- r = flush_one_by_one(lc, flush_list);
- break;
+ if (flush_with_payload) {
+ r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
+ (char *)(group),
+ count * sizeof(uint64_t),
+ NULL, NULL);
+ /* integrated flush failed */
+ if (r)
+ return r;
+ } else {
+ r = userspace_do_request(lc, lc->uuid, type,
+ (char *)(group),
+ count * sizeof(uint64_t),
+ NULL, NULL);
+ if (r) {
+ /* Group send failed. Attempt one-by-one. */
+ list_splice_init(&tmp_list, flush_list);
+ r = flush_one_by_one(lc, flush_list);
+ break;
+ }
}
}
@@ -474,6 +533,8 @@ static int userspace_flush(struct dm_dirty_log *log)
int r = 0;
unsigned long flags;
struct log_c *lc = log->context;
+ int is_mark_list_empty;
+ int is_clear_list_empty;
LIST_HEAD(mark_list);
LIST_HEAD(clear_list);
struct flush_entry *fe, *tmp_fe;
@@ -483,19 +544,39 @@ static int userspace_flush(struct dm_dirty_log *log)
list_splice_init(&lc->clear_list, &clear_list);
spin_unlock_irqrestore(&lc->flush_lock, flags);
- if (list_empty(&mark_list) && list_empty(&clear_list))
+ is_mark_list_empty = list_empty(&mark_list);
+ is_clear_list_empty = list_empty(&clear_list);
+
+ if (is_mark_list_empty && is_clear_list_empty)
return 0;
- r = flush_by_group(lc, &mark_list);
- if (r)
- goto fail;
+ r = flush_by_group(lc, &clear_list, 0);
- r = flush_by_group(lc, &clear_list);
if (r)
goto fail;
- r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
- NULL, 0, NULL, NULL);
+ if (lc->integrated_flush) {
+ /* send flush request with mark_list as payload*/
+ r = flush_by_group(lc, &mark_list, 1);
+ if (r)
+ goto fail;
+ /* when only has clear region,we plan a flush in the furture */
+ if (!is_clear_list_empty && is_mark_list_empty &&
+ !atomic_read(&lc->sched_flush)) {
+ queue_delayed_work(lc->dmlog_wq, &lc->flush_log_work, 3 * HZ);
+ atomic_set(&lc->sched_flush, 1);
+ /* cancel pending flush because we have already flushed in mark_region*/
+ } else {
+ cancel_delayed_work(&lc->flush_log_work);
+ atomic_set(&lc->sched_flush, 0);
+ }
+ } else {
+ r = flush_by_group(lc, &mark_list, 0);
+ if (r)
+ goto fail;
+ r = userspace_do_request(lc, lc->uuid, DM_ULOG_FLUSH,
+ NULL, 0, NULL, NULL);
+ }
fail:
/*
diff --git a/include/uapi/linux/dm-log-userspace.h b/include/uapi/linux/dm-log-userspace.h
index 0678c2a..d1352ae 100644
--- a/include/uapi/linux/dm-log-userspace.h
+++ b/include/uapi/linux/dm-log-userspace.h
@@ -201,12 +201,12 @@
* int (*flush)(struct dm_dirty_log *log);
*
* Payload-to-userspace:
- * None.
+ * if DM_INTEGRATED_FLUSH is set, payload is as same as DM_ULOG_MARK_REGION
+ * uint64_t [] - region(s) to mark
+ * else None
* Payload-to-kernel:
* None.
*
- * No incoming or outgoing payload. Simply flush log state to disk.
- *
* When the request has been processed, user-space must return the
* dm_ulog_request to the kernel - setting the 'error' field and clearing
* 'data_size' appropriately.
@@ -385,8 +385,10 @@
* version 2: DM_ULOG_CTR allowed to return a string containing a
* device name that is to be registered with DM via
* 'dm_get_device'.
+ * version 3: DM_ULOG_FLUSH is capable to carry payload for marking
+ * regions.
*/
-#define DM_ULOG_REQUEST_VERSION 2
+#define DM_ULOG_REQUEST_VERSION 3
struct dm_ulog_request {
/*
--
1.7.3.4
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
