ceph: use rbtree for pg pools; decode new osdmap format
authorSage Weil <sage@newdream.net>
Tue, 16 Feb 2010 23:55:03 +0000 (15:55 -0800)
committerSage Weil <sage@newdream.net>
Wed, 17 Feb 2010 18:02:49 +0000 (10:02 -0800)
Since we can now create and destroy pg pools, the pool ids will be sparse,
and an array no longer makes sense for looking up by pool id.  Use an
rbtree instead.

The OSDMap encoding also no longer has a max pool count (previously used to
allocate the array).  There is a new pool_max, that is the largest pool id
we've ever used, although we don't actually need it in the client.

Signed-off-by: Sage Weil <sage@newdream.net>
fs/ceph/debugfs.c
fs/ceph/osdmap.c
fs/ceph/osdmap.h
fs/ceph/rados.h

index 1a47b5c25b5fb3707d39696e1dfbe82ec60f9d26..e159f14151103a6ce97d5f12c4184207cdb08f4f 100644 (file)
@@ -78,6 +78,7 @@ static int osdmap_show(struct seq_file *s, void *p)
 {
        int i;
        struct ceph_client *client = s->private;
+       struct rb_node *n;
 
        if (client->osdc.osdmap == NULL)
                return 0;
@@ -87,11 +88,11 @@ static int osdmap_show(struct seq_file *s, void *p)
                   " NEARFULL" : "",
                   (client->osdc.osdmap->flags & CEPH_OSDMAP_FULL) ?
                   " FULL" : "");
-       for (i = 0; i < client->osdc.osdmap->num_pools; i++) {
+       for (n = rb_first(&client->osdc.osdmap->pg_pools); n; n = rb_next(n)) {
                struct ceph_pg_pool_info *pool =
-                       &client->osdc.osdmap->pg_pool[i];
+                       rb_entry(n, struct ceph_pg_pool_info, node);
                seq_printf(s, "pg_pool %d pg_num %d / %d, lpg_num %d / %d\n",
-                          i, pool->v.pg_num, pool->pg_num_mask,
+                          pool->id, pool->v.pg_num, pool->pg_num_mask,
                           pool->v.lpg_num, pool->lpg_num_mask);
        }
        for (i = 0; i < client->osdc.osdmap->max_osd; i++) {
index 443fdcdb19c49bdb26bc98b6f662578319ae9059..34b5696c84fdf7c30250f03695d041d95ed44d6b 100644 (file)
@@ -328,9 +328,15 @@ void ceph_osdmap_destroy(struct ceph_osdmap *map)
                rb_erase(&pg->node, &map->pg_temp);
                kfree(pg);
        }
+       while (!RB_EMPTY_ROOT(&map->pg_pools)) {
+               struct ceph_pg_pool_info *pi =
+                       rb_entry(rb_first(&map->pg_pools),
+                                struct ceph_pg_pool_info, node);
+               rb_erase(&pi->node, &map->pg_pools);
+               kfree(pi);
+       }
        kfree(map->osd_state);
        kfree(map->osd_weight);
-       kfree(map->pg_pool);
        kfree(map->osd_addr);
        kfree(map);
 }
@@ -432,6 +438,48 @@ static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
        return NULL;
 }
 
+/*
+ * rbtree of pg pool info
+ */
+static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
+{
+       struct rb_node **p = &root->rb_node;
+       struct rb_node *parent = NULL;
+       struct ceph_pg_pool_info *pi = NULL;
+
+       while (*p) {
+               parent = *p;
+               pi = rb_entry(parent, struct ceph_pg_pool_info, node);
+               if (new->id < pi->id)
+                       p = &(*p)->rb_left;
+               else if (new->id > pi->id)
+                       p = &(*p)->rb_right;
+               else
+                       return -EEXIST;
+       }
+
+       rb_link_node(&new->node, parent, p);
+       rb_insert_color(&new->node, root);
+       return 0;
+}
+
+static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
+{
+       struct ceph_pg_pool_info *pi;
+       struct rb_node *n = root->rb_node;
+
+       while (n) {
+               pi = rb_entry(n, struct ceph_pg_pool_info, node);
+               if (id < pi->id)
+                       n = n->rb_left;
+               else if (id > pi->id)
+                       n = n->rb_right;
+               else
+                       return pi;
+       }
+       return NULL;
+}
+
 /*
  * decode a full map.
  */
@@ -443,6 +491,7 @@ struct ceph_osdmap *osdmap_decode(void **p, void *end)
        u8 ev;
        int err = -EINVAL;
        void *start = *p;
+       struct ceph_pg_pool_info *pi;
 
        dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
 
@@ -464,32 +513,27 @@ struct ceph_osdmap *osdmap_decode(void **p, void *end)
        ceph_decode_copy(p, &map->created, sizeof(map->created));
        ceph_decode_copy(p, &map->modified, sizeof(map->modified));
 
-       map->num_pools = ceph_decode_32(p);
-       map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
-                              GFP_NOFS);
-       if (!map->pg_pool) {
-               err = -ENOMEM;
-               goto bad;
-       }
        ceph_decode_32_safe(p, end, max, bad);
        while (max--) {
-               ceph_decode_need(p, end, 4+1+sizeof(map->pg_pool->v), bad);
-               i = ceph_decode_32(p);
-               if (i >= map->num_pools)
+               ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
+               pi = kmalloc(sizeof(*pi), GFP_NOFS);
+               if (!pi)
                        goto bad;
+               pi->id = ceph_decode_32(p);
                ev = ceph_decode_8(p); /* encoding version */
                if (ev > CEPH_PG_POOL_VERSION) {
                        pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
                                   ev, CEPH_PG_POOL_VERSION);
                        goto bad;
                }
-               ceph_decode_copy(p, &map->pg_pool[i].v,
-                                sizeof(map->pg_pool->v));
-               calc_pg_masks(&map->pg_pool[i]);
-               p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
-               p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
+               ceph_decode_copy(p, &pi->v, sizeof(pi->v));
+               __insert_pg_pool(&map->pg_pools, pi);
+               calc_pg_masks(pi);
+               p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
+               p += le32_to_cpu(pi->v.num_removed_snap_intervals)
                        * sizeof(u64) * 2;
        }
+       ceph_decode_32_safe(p, end, map->pool_max, bad);
 
        ceph_decode_32_safe(p, end, map->flags, bad);
 
@@ -581,7 +625,7 @@ struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
        u32 epoch = 0;
        struct ceph_timespec modified;
        u32 len, pool;
-       __s32 new_flags, max;
+       __s32 new_pool_max, new_flags, max;
        void *start = *p;
        int err = -EINVAL;
        u16 version;
@@ -600,6 +644,7 @@ struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
        epoch = ceph_decode_32(p);
        BUG_ON(epoch != map->epoch+1);
        ceph_decode_copy(p, &modified, sizeof(modified));
+       new_pool_max = ceph_decode_32(p);
        new_flags = ceph_decode_32(p);
 
        /* full map? */
@@ -623,6 +668,8 @@ struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
        /* new flags? */
        if (new_flags >= 0)
                map->flags = new_flags;
+       if (new_pool_max >= 0)
+               map->pool_max = new_pool_max;
 
        ceph_decode_need(p, end, 5*sizeof(u32), bad);
 
@@ -647,37 +694,42 @@ struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
        ceph_decode_32_safe(p, end, len, bad);
        while (len--) {
                __u8 ev;
+               struct ceph_pg_pool_info *pi;
 
                ceph_decode_32_safe(p, end, pool, bad);
-               if (pool >= map->num_pools) {
-                       void *pg_pool = kcalloc(pool + 1,
-                                               sizeof(*map->pg_pool),
-                                               GFP_NOFS);
-                       if (!pg_pool) {
-                               err = -ENOMEM;
-                               goto bad;
-                       }
-                       memcpy(pg_pool, map->pg_pool,
-                              map->num_pools * sizeof(*map->pg_pool));
-                       kfree(map->pg_pool);
-                       map->pg_pool = pg_pool;
-                       map->num_pools = pool+1;
-               }
-               ceph_decode_need(p, end, 1 + sizeof(map->pg_pool->v), bad);
+               ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
                ev = ceph_decode_8(p);  /* encoding version */
                if (ev > CEPH_PG_POOL_VERSION) {
                        pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
                                   ev, CEPH_PG_POOL_VERSION);
                        goto bad;
                }
-               ceph_decode_copy(p, &map->pg_pool[pool].v,
-                                sizeof(map->pg_pool->v));
-               calc_pg_masks(&map->pg_pool[pool]);
+               pi = __lookup_pg_pool(&map->pg_pools, pool);
+               if (!pi) {
+                       pi = kmalloc(sizeof(*pi), GFP_NOFS);
+                       if (!pi) {
+                               err = -ENOMEM;
+                               goto bad;
+                       }
+                       pi->id = pool;
+                       __insert_pg_pool(&map->pg_pools, pi);
+               }
+               ceph_decode_copy(p, &pi->v, sizeof(pi->v));
+               calc_pg_masks(pi);
        }
 
-       /* old_pool (ignore) */
+       /* old_pool */
        ceph_decode_32_safe(p, end, len, bad);
-       *p += len * sizeof(u32);
+       while (len--) {
+               struct ceph_pg_pool_info *pi;
+
+               ceph_decode_32_safe(p, end, pool, bad);
+               pi = __lookup_pg_pool(&map->pg_pools, pool);
+               if (pi) {
+                       rb_erase(&pi->node, &map->pg_pools);
+                       kfree(pi);
+               }
+       }
 
        /* new_up */
        err = -EINVAL;
@@ -861,10 +913,10 @@ int ceph_calc_object_layout(struct ceph_object_layout *ol,
        unsigned ps;
 
        BUG_ON(!osdmap);
-       if (poolid >= osdmap->num_pools)
-               return -EIO;
 
-       pool = &osdmap->pg_pool[poolid];
+       pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
+       if (!pool)
+               return -EIO;
        ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
        if (preferred >= 0) {
                ps += preferred;
@@ -919,9 +971,9 @@ static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
            preferred >= osdmap->crush->max_devices)
                preferred = -1;
 
-       if (poolid >= osdmap->num_pools)
+       pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
+       if (!pool)
                return NULL;
-       pool = &osdmap->pg_pool[poolid];
        ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
                                 pool->v.type, pool->v.size);
        if (ruleno < 0) {
index c4af8418aa002a2aef6db77248bb21cebe6cbc70..1fb55afb264261c87ee0c4c8a63588269493f1f8 100644 (file)
@@ -19,6 +19,8 @@
  * the change between two successive epochs, or as a fully encoded map.
  */
 struct ceph_pg_pool_info {
+       struct rb_node node;
+       int id;
        struct ceph_pg_pool v;
        int pg_num_mask, pgp_num_mask, lpg_num_mask, lpgp_num_mask;
 };
@@ -44,9 +46,8 @@ struct ceph_osdmap {
        struct ceph_entity_addr *osd_addr;
 
        struct rb_root pg_temp;
-
-       u32 num_pools;
-       struct ceph_pg_pool_info *pg_pool;
+       struct rb_root pg_pools;
+       u32 pool_max;
 
        /* the CRUSH map specifies the mapping of placement groups to
         * the list of osds that store+replicate them. */
index 1f4c78640541922a7f2b8f0df46fe07ad21eb59f..26ac8b89a67678e7cf703b4409de984b0ea137fe 100644 (file)
@@ -11,8 +11,8 @@
 /*
  * osdmap encoding versions
  */
-#define CEPH_OSDMAP_INC_VERSION 3
-#define CEPH_OSDMAP_VERSION     3
+#define CEPH_OSDMAP_INC_VERSION 4
+#define CEPH_OSDMAP_VERSION     4
 
 /*
  * fs id