diff options
| -rw-r--r-- | include/linux/crush/crush.h | 58 | ||||
| -rw-r--r-- | include/linux/crush/mapper.h | 9 | ||||
| -rw-r--r-- | net/ceph/crush/mapper.c | 74 | ||||
| -rw-r--r-- | net/ceph/osdmap.c | 2 |
4 files changed, 119 insertions, 24 deletions
diff --git a/include/linux/crush/crush.h b/include/linux/crush/crush.h index fbecbd089d75..d8676e56fa23 100644 --- a/include/linux/crush/crush.h +++ b/include/linux/crush/crush.h @@ -137,6 +137,64 @@ struct crush_bucket { }; +/** @ingroup API + * + * Replacement weights for each item in a bucket. The size of the + * array must be exactly the size of the straw2 bucket, just as the + * item_weights array. + * + */ +struct crush_weight_set { + __u32 *weights; /*!< 16.16 fixed point weights + in the same order as items */ + __u32 size; /*!< size of the __weights__ array */ +}; + +/** @ingroup API + * + * Replacement weights and ids for a given straw2 bucket, for + * placement purposes. + * + * When crush_do_rule() chooses the Nth item from a straw2 bucket, the + * replacement weights found at __weight_set[N]__ are used instead of + * the weights from __item_weights__. If __N__ is greater than + * __weight_set_size__, the weights found at __weight_set_size-1__ are + * used instead. For instance if __weight_set__ is: + * + * [ [ 0x10000, 0x20000 ], // position 0 + * [ 0x20000, 0x40000 ] ] // position 1 + * + * choosing the 0th item will use position 0 weights [ 0x10000, 0x20000 ] + * choosing the 1th item will use position 1 weights [ 0x20000, 0x40000 ] + * choosing the 2th item will use position 1 weights [ 0x20000, 0x40000 ] + * etc. + * + */ +struct crush_choose_arg { + __s32 *ids; /*!< values to use instead of items */ + __u32 ids_size; /*!< size of the __ids__ array */ + struct crush_weight_set *weight_set; /*!< weight replacements for + a given position */ + __u32 weight_set_size; /*!< size of the __weight_set__ array */ +}; + +/** @ingroup API + * + * Replacement weights and ids for each bucket in the crushmap. The + * __size__ of the __args__ array must be exactly the same as the + * __map->max_buckets__. + * + * The __crush_choose_arg__ at index N will be used when choosing + * an item from the bucket __map->buckets[N]__ bucket, provided it + * is a straw2 bucket. + * + */ +struct crush_choose_arg_map { + struct crush_choose_arg *args; /*!< replacement for each bucket + in the crushmap */ + __u32 size; /*!< size of the __args__ array */ +}; + struct crush_bucket_uniform { struct crush_bucket h; __u32 item_weight; /* 16-bit fixed point; all items equally weighted */ diff --git a/include/linux/crush/mapper.h b/include/linux/crush/mapper.h index c95e19e1ff11..141edabb947e 100644 --- a/include/linux/crush/mapper.h +++ b/include/linux/crush/mapper.h @@ -11,11 +11,10 @@ #include "crush.h" extern int crush_find_rule(const struct crush_map *map, int ruleset, int type, int size); -extern int crush_do_rule(const struct crush_map *map, - int ruleno, - int x, int *result, int result_max, - const __u32 *weights, int weight_max, - void *cwin); +int crush_do_rule(const struct crush_map *map, + int ruleno, int x, int *result, int result_max, + const __u32 *weight, int weight_max, + void *cwin, const struct crush_choose_arg *choose_args); /* * Returns the exact amount of workspace that will need to be used diff --git a/net/ceph/crush/mapper.c b/net/ceph/crush/mapper.c index b5cd8c21bfdf..0b2646a9cc50 100644 --- a/net/ceph/crush/mapper.c +++ b/net/ceph/crush/mapper.c @@ -302,19 +302,42 @@ static __u64 crush_ln(unsigned int xin) * */ +static __u32 *get_choose_arg_weights(const struct crush_bucket_straw2 *bucket, + const struct crush_choose_arg *arg, + int position) +{ + if (!arg || !arg->weight_set || arg->weight_set_size == 0) + return bucket->item_weights; + + if (position >= arg->weight_set_size) + position = arg->weight_set_size - 1; + return arg->weight_set[position].weights; +} + +static __s32 *get_choose_arg_ids(const struct crush_bucket_straw2 *bucket, + const struct crush_choose_arg *arg) +{ + if (!arg || !arg->ids) + return bucket->h.items; + + return arg->ids; +} + static int bucket_straw2_choose(const struct crush_bucket_straw2 *bucket, - int x, int r) + int x, int r, + const struct crush_choose_arg *arg, + int position) { unsigned int i, high = 0; unsigned int u; - unsigned int w; __s64 ln, draw, high_draw = 0; + __u32 *weights = get_choose_arg_weights(bucket, arg, position); + __s32 *ids = get_choose_arg_ids(bucket, arg); for (i = 0; i < bucket->h.size; i++) { - w = bucket->item_weights[i]; - if (w) { - u = crush_hash32_3(bucket->h.hash, x, - bucket->h.items[i], r); + dprintk("weight 0x%x item %d\n", weights[i], ids[i]); + if (weights[i]) { + u = crush_hash32_3(bucket->h.hash, x, ids[i], r); u &= 0xffff; /* @@ -335,7 +358,7 @@ static int bucket_straw2_choose(const struct crush_bucket_straw2 *bucket, * weight means a larger (less negative) value * for draw. */ - draw = div64_s64(ln, w); + draw = div64_s64(ln, weights[i]); } else { draw = S64_MIN; } @@ -352,7 +375,9 @@ static int bucket_straw2_choose(const struct crush_bucket_straw2 *bucket, static int crush_bucket_choose(const struct crush_bucket *in, struct crush_work_bucket *work, - int x, int r) + int x, int r, + const struct crush_choose_arg *arg, + int position) { dprintk(" crush_bucket_choose %d x=%d r=%d\n", in->id, x, r); BUG_ON(in->size == 0); @@ -374,7 +399,7 @@ static int crush_bucket_choose(const struct crush_bucket *in, case CRUSH_BUCKET_STRAW2: return bucket_straw2_choose( (const struct crush_bucket_straw2 *)in, - x, r); + x, r, arg, position); default: dprintk("unknown bucket %d alg %d\n", in->id, in->alg); return in->items[0]; @@ -436,7 +461,8 @@ static int crush_choose_firstn(const struct crush_map *map, unsigned int vary_r, unsigned int stable, int *out2, - int parent_r) + int parent_r, + const struct crush_choose_arg *choose_args) { int rep; unsigned int ftotal, flocal; @@ -486,7 +512,10 @@ static int crush_choose_firstn(const struct crush_map *map, else item = crush_bucket_choose( in, work->work[-1-in->id], - x, r); + x, r, + (choose_args ? + &choose_args[-1-in->id] : 0), + outpos); if (item >= map->max_devices) { dprintk(" bad item %d\n", item); skip_rep = 1; @@ -543,7 +572,8 @@ static int crush_choose_firstn(const struct crush_map *map, vary_r, stable, NULL, - sub_r) <= outpos) + sub_r, + choose_args) <= outpos) /* didn't get leaf */ reject = 1; } else { @@ -620,7 +650,8 @@ static void crush_choose_indep(const struct crush_map *map, unsigned int recurse_tries, int recurse_to_leaf, int *out2, - int parent_r) + int parent_r, + const struct crush_choose_arg *choose_args) { const struct crush_bucket *in = bucket; int endpos = outpos + left; @@ -692,7 +723,10 @@ static void crush_choose_indep(const struct crush_map *map, item = crush_bucket_choose( in, work->work[-1-in->id], - x, r); + x, r, + (choose_args ? + &choose_args[-1-in->id] : 0), + outpos); if (item >= map->max_devices) { dprintk(" bad item %d\n", item); out[rep] = CRUSH_ITEM_NONE; @@ -746,7 +780,8 @@ static void crush_choose_indep(const struct crush_map *map, x, 1, numrep, 0, out2, rep, recurse_tries, 0, - 0, NULL, r); + 0, NULL, r, + choose_args); if (out2[rep] == CRUSH_ITEM_NONE) { /* placed nothing; no leaf */ break; @@ -854,11 +889,12 @@ void crush_init_workspace(const struct crush_map *map, void *v) * @weight: weight vector (for map leaves) * @weight_max: size of weight vector * @cwin: pointer to at least crush_work_size() bytes of memory + * @choose_args: weights and ids for each known bucket */ int crush_do_rule(const struct crush_map *map, int ruleno, int x, int *result, int result_max, const __u32 *weight, int weight_max, - void *cwin) + void *cwin, const struct crush_choose_arg *choose_args) { int result_len; struct crush_work *cw = cwin; @@ -1013,7 +1049,8 @@ int crush_do_rule(const struct crush_map *map, vary_r, stable, c+osize, - 0); + 0, + choose_args); } else { out_size = ((numrep < (result_max-osize)) ? numrep : (result_max-osize)); @@ -1030,7 +1067,8 @@ int crush_do_rule(const struct crush_map *map, choose_leaf_tries : 1, recurse_to_leaf, c+osize, - 0); + 0, + choose_args); osize += out_size; } } diff --git a/net/ceph/osdmap.c b/net/ceph/osdmap.c index 93baa69407c5..9da0ee61aca5 100644 --- a/net/ceph/osdmap.c +++ b/net/ceph/osdmap.c @@ -2111,7 +2111,7 @@ static int do_crush(struct ceph_osdmap *map, int ruleno, int x, mutex_lock(&map->crush_workspace_mutex); r = crush_do_rule(map->crush, ruleno, x, result, result_max, - weight, weight_max, map->crush_workspace); + weight, weight_max, map->crush_workspace, NULL); mutex_unlock(&map->crush_workspace_mutex); return r; |