dm log userspace: group clear and mark requests
authorJonathan Brassow <jbrassow@redhat.com>
Thu, 13 Jan 2011 19:59:51 +0000 (19:59 +0000)
committerAlasdair G Kergon <agk@redhat.com>
Thu, 13 Jan 2011 19:59:51 +0000 (19:59 +0000)
Allow the device-mapper log's 'mark' and 'clear' requests to be
grouped and processed in a batch.  This can significantly reduce the
amount of traffic going between the kernel and userspace (where the
processing daemon resides).

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
drivers/md/dm-log-userspace-base.c

index 767adf300fa517c9f28d7676fabde333c6f81459..31e1687e7bf6dc8b5f23cdcbf9a903cefd3ef1e4 100644 (file)
@@ -18,6 +18,14 @@ struct flush_entry {
        struct list_head list;
 };
 
+/*
+ * This limit on the number of mark and clear request is, to a degree,
+ * arbitrary.  However, there is some basis for the choice in the limits
+ * imposed on the size of data payload by dm-log-userspace-transfer.c:
+ * dm_consult_userspace().
+ */
+#define MAX_FLUSH_GROUP_COUNT 32
+
 struct log_c {
        struct dm_target *ti;
        uint32_t region_size;
@@ -348,6 +356,71 @@ static int userspace_in_sync(struct dm_dirty_log *log, region_t region,
        return (r) ? 0 : (int)in_sync;
 }
 
+static int flush_one_by_one(struct log_c *lc, struct list_head *flush_list)
+{
+       int r = 0;
+       struct flush_entry *fe;
+
+       list_for_each_entry(fe, flush_list, list) {
+               r = userspace_do_request(lc, lc->uuid, fe->type,
+                                        (char *)&fe->region,
+                                        sizeof(fe->region),
+                                        NULL, NULL);
+               if (r)
+                       break;
+       }
+
+       return r;
+}
+
+static int flush_by_group(struct log_c *lc, struct list_head *flush_list)
+{
+       int r = 0;
+       int count;
+       uint32_t type = 0;
+       struct flush_entry *fe, *tmp_fe;
+       LIST_HEAD(tmp_list);
+       uint64_t group[MAX_FLUSH_GROUP_COUNT];
+
+       /*
+        * Group process the requests
+        */
+       while (!list_empty(flush_list)) {
+               count = 0;
+
+               list_for_each_entry_safe(fe, tmp_fe, flush_list, list) {
+                       group[count] = fe->region;
+                       count++;
+
+                       list_del(&fe->list);
+                       list_add(&fe->list, &tmp_list);
+
+                       type = fe->type;
+                       if (count >= MAX_FLUSH_GROUP_COUNT)
+                               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;
+               }
+       }
+
+       /*
+        * Must collect flush_entrys that were successfully processed
+        * as a group so that they will be free'd by the caller.
+        */
+       list_splice_init(&tmp_list, flush_list);
+
+       return r;
+}
+
 /*
  * userspace_flush
  *
@@ -382,30 +455,13 @@ static int userspace_flush(struct dm_dirty_log *log)
        if (list_empty(&mark_list) && list_empty(&clear_list))
                return 0;
 
-       /*
-        * FIXME: Count up requests, group request types,
-        * allocate memory to stick all requests in and
-        * send to server in one go.  Failing the allocation,
-        * do it one by one.
-        */
-
-       list_for_each_entry(fe, &mark_list, list) {
-               r = userspace_do_request(lc, lc->uuid, fe->type,
-                                        (char *)&fe->region,
-                                        sizeof(fe->region),
-                                        NULL, NULL);
-               if (r)
-                       goto fail;
-       }
+       r = flush_by_group(lc, &mark_list);
+       if (r)
+               goto fail;
 
-       list_for_each_entry(fe, &clear_list, list) {
-               r = userspace_do_request(lc, lc->uuid, fe->type,
-                                        (char *)&fe->region,
-                                        sizeof(fe->region),
-                                        NULL, NULL);
-               if (r)
-                       goto fail;
-       }
+       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);