aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--fs/ceph/osd_client.c19
-rw-r--r--fs/ceph/osd_client.h2
-rw-r--r--fs/ceph/osdmap.c29
-rw-r--r--fs/ceph/osdmap.h2
-rw-r--r--fs/ceph/rados.h1
5 files changed, 44 insertions, 9 deletions
diff --git a/fs/ceph/osd_client.c b/fs/ceph/osd_client.c
index 8128082a028e..3514f71ff85f 100644
--- a/fs/ceph/osd_client.c
+++ b/fs/ceph/osd_client.c
@@ -565,7 +565,8 @@ static int __map_osds(struct ceph_osd_client *osdc,
565{ 565{
566 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base; 566 struct ceph_osd_request_head *reqhead = req->r_request->front.iov_base;
567 struct ceph_pg pgid; 567 struct ceph_pg pgid;
568 int o = -1; 568 int acting[CEPH_PG_MAX_SIZE];
569 int o = -1, num = 0;
569 int err; 570 int err;
570 571
571 dout("map_osds %p tid %lld\n", req, req->r_tid); 572 dout("map_osds %p tid %lld\n", req, req->r_tid);
@@ -576,10 +577,16 @@ static int __map_osds(struct ceph_osd_client *osdc,
576 pgid = reqhead->layout.ol_pgid; 577 pgid = reqhead->layout.ol_pgid;
577 req->r_pgid = pgid; 578 req->r_pgid = pgid;
578 579
579 o = ceph_calc_pg_primary(osdc->osdmap, pgid); 580 err = ceph_calc_pg_acting(osdc->osdmap, pgid, acting);
581 if (err > 0) {
582 o = acting[0];
583 num = err;
584 }
580 585
581 if ((req->r_osd && req->r_osd->o_osd == o && 586 if ((req->r_osd && req->r_osd->o_osd == o &&
582 req->r_sent >= req->r_osd->o_incarnation) || 587 req->r_sent >= req->r_osd->o_incarnation &&
588 req->r_num_pg_osds == num &&
589 memcmp(req->r_pg_osds, acting, sizeof(acting[0])*num) == 0) ||
583 (req->r_osd == NULL && o == -1)) 590 (req->r_osd == NULL && o == -1))
584 return 0; /* no change */ 591 return 0; /* no change */
585 592
@@ -587,6 +594,10 @@ static int __map_osds(struct ceph_osd_client *osdc,
587 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o, 594 req->r_tid, le32_to_cpu(pgid.pool), le16_to_cpu(pgid.ps), o,
588 req->r_osd ? req->r_osd->o_osd : -1); 595 req->r_osd ? req->r_osd->o_osd : -1);
589 596
597 /* record full pg acting set */
598 memcpy(req->r_pg_osds, acting, sizeof(acting[0]) * num);
599 req->r_num_pg_osds = num;
600
590 if (req->r_osd) { 601 if (req->r_osd) {
591 __cancel_request(req); 602 __cancel_request(req);
592 list_del_init(&req->r_osd_item); 603 list_del_init(&req->r_osd_item);
@@ -612,7 +623,7 @@ static int __map_osds(struct ceph_osd_client *osdc,
612 __remove_osd_from_lru(req->r_osd); 623 __remove_osd_from_lru(req->r_osd);
613 list_add(&req->r_osd_item, &req->r_osd->o_requests); 624 list_add(&req->r_osd_item, &req->r_osd->o_requests);
614 } 625 }
615 err = 1; /* osd changed */ 626 err = 1; /* osd or pg changed */
616 627
617out: 628out:
618 return err; 629 return err;
diff --git a/fs/ceph/osd_client.h b/fs/ceph/osd_client.h
index c5191d62f243..ce776989ef6a 100644
--- a/fs/ceph/osd_client.h
+++ b/fs/ceph/osd_client.h
@@ -48,6 +48,8 @@ struct ceph_osd_request {
48 struct list_head r_osd_item; 48 struct list_head r_osd_item;
49 struct ceph_osd *r_osd; 49 struct ceph_osd *r_osd;
50 struct ceph_pg r_pgid; 50 struct ceph_pg r_pgid;
51 int r_pg_osds[CEPH_PG_MAX_SIZE];
52 int r_num_pg_osds;
51 53
52 struct ceph_connection *r_con_filling_msg; 54 struct ceph_connection *r_con_filling_msg;
53 55
diff --git a/fs/ceph/osdmap.c b/fs/ceph/osdmap.c
index 2e2c15eed82a..cfdd8f4388b7 100644
--- a/fs/ceph/osdmap.c
+++ b/fs/ceph/osdmap.c
@@ -1041,12 +1041,33 @@ static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1041} 1041}
1042 1042
1043/* 1043/*
1044 * Return acting set for given pgid.
1045 */
1046int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
1047 int *acting)
1048{
1049 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1050 int i, o, num = CEPH_PG_MAX_SIZE;
1051
1052 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1053 if (!osds)
1054 return -1;
1055
1056 /* primary is first up osd */
1057 o = 0;
1058 for (i = 0; i < num; i++)
1059 if (ceph_osd_is_up(osdmap, osds[i]))
1060 acting[o++] = osds[i];
1061 return o;
1062}
1063
1064/*
1044 * Return primary osd for given pgid, or -1 if none. 1065 * Return primary osd for given pgid, or -1 if none.
1045 */ 1066 */
1046int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid) 1067int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1047{ 1068{
1048 int rawosds[10], *osds; 1069 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1049 int i, num = ARRAY_SIZE(rawosds); 1070 int i, num = CEPH_PG_MAX_SIZE;
1050 1071
1051 osds = calc_pg_raw(osdmap, pgid, rawosds, &num); 1072 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1052 if (!osds) 1073 if (!osds)
@@ -1054,9 +1075,7 @@ int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
1054 1075
1055 /* primary is first up osd */ 1076 /* primary is first up osd */
1056 for (i = 0; i < num; i++) 1077 for (i = 0; i < num; i++)
1057 if (ceph_osd_is_up(osdmap, osds[i])) { 1078 if (ceph_osd_is_up(osdmap, osds[i]))
1058 return osds[i]; 1079 return osds[i];
1059 break;
1060 }
1061 return -1; 1080 return -1;
1062} 1081}
diff --git a/fs/ceph/osdmap.h b/fs/ceph/osdmap.h
index 8bc9f1e4f562..970b547e510d 100644
--- a/fs/ceph/osdmap.h
+++ b/fs/ceph/osdmap.h
@@ -120,6 +120,8 @@ extern int ceph_calc_object_layout(struct ceph_object_layout *ol,
120 const char *oid, 120 const char *oid,
121 struct ceph_file_layout *fl, 121 struct ceph_file_layout *fl,
122 struct ceph_osdmap *osdmap); 122 struct ceph_osdmap *osdmap);
123extern int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
124 int *acting);
123extern int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, 125extern int ceph_calc_pg_primary(struct ceph_osdmap *osdmap,
124 struct ceph_pg pgid); 126 struct ceph_pg pgid);
125 127
diff --git a/fs/ceph/rados.h b/fs/ceph/rados.h
index a1fc1d017b58..fd56451a871f 100644
--- a/fs/ceph/rados.h
+++ b/fs/ceph/rados.h
@@ -58,6 +58,7 @@ struct ceph_timespec {
58#define CEPH_PG_LAYOUT_LINEAR 2 58#define CEPH_PG_LAYOUT_LINEAR 2
59#define CEPH_PG_LAYOUT_HYBRID 3 59#define CEPH_PG_LAYOUT_HYBRID 3
60 60
61#define CEPH_PG_MAX_SIZE 16 /* max # osds in a single pg */
61 62
62/* 63/*
63 * placement group. 64 * placement group.