static int bucket_tree_choose(struct crush_bucket_tree *bucket,
int x, int r)
{
- int n, l;
+ int n;
__u32 w;
__u64 t;
n = bucket->num_nodes >> 1;
while (!terminal(n)) {
+ int l;
/* pick point in [0, w) */
w = bucket->node_weights[n];
t = (__u64)crush_hash32_4(bucket->h.hash, x, n, r,
* true if device is marked "out" (failed, fully offloaded)
* of the cluster
*/
-static int is_out(const struct crush_map *map, const __u32 *weight, int item, int x)
+static int is_out(const struct crush_map *map,
+ const __u32 *weight, int weight_max,
+ int item, int x)
{
+ if (item >= weight_max)
+ return 1;
if (weight[item] >= 0x10000)
return 0;
if (weight[item] == 0)
}
/**
- * crush_choose - choose numrep distinct items of given type
+ * crush_choose_firstn - choose numrep distinct items of given type
* @map: the crush_map
* @bucket: the bucket we are choose an item from
* @x: crush input value
* @type: the type of item to choose
* @out: pointer to output vector
* @outpos: our position in that vector
- * @firstn: true if choosing "first n" items, false if choosing "indep"
* @recurse_to_leaf: true if we want one device under each item of given type
* @descend_once: true if we should only try one descent before giving up
* @out2: second output vector for leaf items (if @recurse_to_leaf)
*/
-static int crush_choose(const struct crush_map *map,
- struct crush_bucket *bucket,
- const __u32 *weight,
- int x, int numrep, int type,
- int *out, int outpos,
- int firstn, int recurse_to_leaf,
- int descend_once, int *out2)
+static int crush_choose_firstn(const struct crush_map *map,
+ struct crush_bucket *bucket,
+ const __u32 *weight, int weight_max,
+ int x, int numrep, int type,
+ int *out, int outpos,
+ int recurse_to_leaf,
+ int descend_once, int *out2)
{
int rep;
unsigned int ftotal, flocal;
collide = 0;
retry_bucket = 0;
r = rep;
- if (in->alg == CRUSH_BUCKET_UNIFORM) {
- /* be careful */
- if (firstn || (__u32)numrep >= in->size)
- /* r' = r + f_total */
- r += ftotal;
- else if (in->size % numrep == 0)
- /* r'=r+(n+1)*f_local */
- r += (numrep+1) *
- (flocal+ftotal);
- else
- /* r' = r + n*f_local */
- r += numrep * (flocal+ftotal);
- } else {
- if (firstn)
- /* r' = r + f_total */
- r += ftotal;
- else
- /* r' = r + n*f_local */
- r += numrep * (flocal+ftotal);
- }
+ /* r' = r + f_total */
+ r += ftotal;
/* bucket choose */
if (in->size == 0) {
reject = 0;
if (!collide && recurse_to_leaf) {
if (item < 0) {
- if (crush_choose(map,
+ if (crush_choose_firstn(map,
map->buckets[-1-item],
- weight,
+ weight, weight_max,
x, outpos+1, 0,
out2, outpos,
- firstn, 0,
+ 0,
map->chooseleaf_descend_once,
NULL) <= outpos)
/* didn't get leaf */
/* out? */
if (itemtype == 0)
reject = is_out(map, weight,
+ weight_max,
item, x);
else
reject = 0;
}
+/**
+ * crush_choose_indep: alternative breadth-first positionally stable mapping
+ *
+ */
+static void crush_choose_indep(const struct crush_map *map,
+ struct crush_bucket *bucket,
+ const __u32 *weight, int weight_max,
+ int x, int left, int numrep, int type,
+ int *out, int outpos,
+ int recurse_to_leaf,
+ int *out2)
+{
+ struct crush_bucket *in = bucket;
+ int endpos = outpos + left;
+ int rep;
+ unsigned int ftotal;
+ int r;
+ int i;
+ int item = 0;
+ int itemtype;
+ int collide;
+
+ dprintk("CHOOSE%s INDEP bucket %d x %d outpos %d numrep %d\n", recurse_to_leaf ? "_LEAF" : "",
+ bucket->id, x, outpos, numrep);
+
+ /* initially my result is undefined */
+ for (rep = outpos; rep < endpos; rep++) {
+ out[rep] = CRUSH_ITEM_UNDEF;
+ if (out2)
+ out2[rep] = CRUSH_ITEM_UNDEF;
+ }
+
+ for (ftotal = 0; left > 0 && ftotal < map->choose_total_tries; ftotal++) {
+ for (rep = outpos; rep < endpos; rep++) {
+ if (out[rep] != CRUSH_ITEM_UNDEF)
+ continue;
+
+ in = bucket; /* initial bucket */
+
+ /* choose through intervening buckets */
+ for (;;) {
+ /* note: we base the choice on the position
+ * even in the nested call. that means that
+ * if the first layer chooses the same bucket
+ * in a different position, we will tend to
+ * choose a different item in that bucket.
+ * this will involve more devices in data
+ * movement and tend to distribute the load.
+ */
+ r = rep;
+
+ /* be careful */
+ if (in->alg == CRUSH_BUCKET_UNIFORM &&
+ in->size % numrep == 0)
+ /* r'=r+(n+1)*f_total */
+ r += (numrep+1) * ftotal;
+ else
+ /* r' = r + n*f_total */
+ r += numrep * ftotal;
+
+ /* bucket choose */
+ if (in->size == 0) {
+ dprintk(" empty bucket\n");
+ break;
+ }
+
+ item = crush_bucket_choose(in, x, r);
+ if (item >= map->max_devices) {
+ dprintk(" bad item %d\n", item);
+ out[rep] = CRUSH_ITEM_NONE;
+ if (out2)
+ out2[rep] = CRUSH_ITEM_NONE;
+ left--;
+ break;
+ }
+
+ /* desired type? */
+ if (item < 0)
+ itemtype = map->buckets[-1-item]->type;
+ else
+ itemtype = 0;
+ dprintk(" item %d type %d\n", item, itemtype);
+
+ /* keep going? */
+ if (itemtype != type) {
+ if (item >= 0 ||
+ (-1-item) >= map->max_buckets) {
+ dprintk(" bad item type %d\n", type);
+ out[rep] = CRUSH_ITEM_NONE;
+ if (out2)
+ out2[rep] =
+ CRUSH_ITEM_NONE;
+ left--;
+ break;
+ }
+ in = map->buckets[-1-item];
+ continue;
+ }
+
+ /* collision? */
+ collide = 0;
+ for (i = outpos; i < endpos; i++) {
+ if (out[i] == item) {
+ collide = 1;
+ break;
+ }
+ }
+ if (collide)
+ break;
+
+ if (recurse_to_leaf) {
+ if (item < 0) {
+ crush_choose_indep(map,
+ map->buckets[-1-item],
+ weight, weight_max,
+ x, 1, numrep, 0,
+ out2, rep,
+ 0, NULL);
+ if (out2[rep] == CRUSH_ITEM_NONE) {
+ /* placed nothing; no leaf */
+ break;
+ }
+ } else {
+ /* we already have a leaf! */
+ out2[rep] = item;
+ }
+ }
+
+ /* out? */
+ if (itemtype == 0 &&
+ is_out(map, weight, weight_max, item, x))
+ break;
+
+ /* yay! */
+ out[rep] = item;
+ left--;
+ break;
+ }
+ }
+ }
+ for (rep = outpos; rep < endpos; rep++) {
+ if (out[rep] == CRUSH_ITEM_UNDEF) {
+ out[rep] = CRUSH_ITEM_NONE;
+ }
+ if (out2 && out2[rep] == CRUSH_ITEM_UNDEF) {
+ out2[rep] = CRUSH_ITEM_NONE;
+ }
+ }
+}
+
/**
* crush_do_rule - calculate a mapping with the given input and rule
* @map: the crush_map
* @x: hash input
* @result: pointer to result vector
* @result_max: maximum result size
+ * @weight: weight vector (for map leaves)
+ * @weight_max: size of weight vector
+ * @scratch: scratch vector for private use; must be >= 3 * result_max
*/
int crush_do_rule(const struct crush_map *map,
int ruleno, int x, int *result, int result_max,
- const __u32 *weight)
+ const __u32 *weight, int weight_max,
+ int *scratch)
{
int result_len;
- int a[CRUSH_MAX_SET];
- int b[CRUSH_MAX_SET];
- int c[CRUSH_MAX_SET];
+ int *a = scratch;
+ int *b = scratch + result_max;
+ int *c = scratch + result_max*2;
int recurse_to_leaf;
int *w;
int wsize = 0;
__u32 step;
int i, j;
int numrep;
- int firstn;
const int descend_once = 0;
if ((__u32)ruleno >= map->max_rules) {
o = b;
for (step = 0; step < rule->len; step++) {
+ int firstn = 0;
struct crush_rule_step *curstep = &rule->steps[step];
- firstn = 0;
switch (curstep->op) {
case CRUSH_RULE_TAKE:
w[0] = curstep->arg1;
continue;
}
j = 0;
- osize += crush_choose(map,
- map->buckets[-1-w[i]],
- weight,
- x, numrep,
- curstep->arg2,
- o+osize, j,
- firstn,
- recurse_to_leaf,
- descend_once, c+osize);
+ if (firstn) {
+ osize += crush_choose_firstn(
+ map,
+ map->buckets[-1-w[i]],
+ weight, weight_max,
+ x, numrep,
+ curstep->arg2,
+ o+osize, j,
+ recurse_to_leaf,
+ descend_once, c+osize);
+ } else {
+ crush_choose_indep(
+ map,
+ map->buckets[-1-w[i]],
+ weight, weight_max,
+ x, numrep, numrep,
+ curstep->arg2,
+ o+osize, j,
+ recurse_to_leaf,
+ c+osize);
+ osize += numrep;
+ }
}
if (recurse_to_leaf)
/* copy final _leaf_ values to output set */
memcpy(o, c, osize*sizeof(*o));
- /* swap t and w arrays */
+ /* swap o and w arrays */
tmp = o;
o = w;
w = tmp;