X-Git-Url: http://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Fhash.c;h=18f6b1720eb662b2a0f140a6d23f7c844269b223;hp=7ade36d35d24521c8e62645b33da35d9d373fda4;hb=a42bba8ec446284256a7c9146ba3525404de474c;hpb=115648946c7a78e87fb45eac6c1759a347d4c7f8 diff --git a/ccan/tdb2/hash.c b/ccan/tdb2/hash.c index 7ade36d3..18f6b172 100644 --- a/ccan/tdb2/hash.c +++ b/ccan/tdb2/hash.c @@ -78,6 +78,30 @@ static uint32_t use_bits(struct hash_info *h, unsigned num) return bits(h->h, 64 - h->hash_used, num); } +static bool key_matches(struct tdb_context *tdb, + const struct tdb_used_record *rec, + tdb_off_t off, + const struct tdb_data *key) +{ + bool ret = false; + const char *rkey; + + if (rec_key_length(rec) != key->dsize) { + add_stat(tdb, compare_wrong_keylen, 1); + return ret; + } + + rkey = tdb_access_read(tdb, off + sizeof(*rec), key->dsize, false); + if (!rkey) + return ret; + if (memcmp(rkey, key->dptr, key->dsize) == 0) + ret = true; + else + add_stat(tdb, compare_wrong_keycmp, 1); + tdb_access_release(tdb, rkey); + return ret; +} + /* Does entry match? */ static bool match(struct tdb_context *tdb, struct hash_info *h, @@ -85,15 +109,13 @@ static bool match(struct tdb_context *tdb, tdb_off_t val, struct tdb_used_record *rec) { - bool ret = false; - const unsigned char *rkey; tdb_off_t off; add_stat(tdb, compares, 1); /* Desired bucket must match. */ if (h->home_bucket != (val & TDB_OFF_HASH_GROUP_MASK)) { add_stat(tdb, compare_wrong_bucket, 1); - return ret; + return false; } /* Top bits of offset == next bits of hash. */ @@ -101,32 +123,19 @@ static bool match(struct tdb_context *tdb, != bits(h->h, 64 - h->hash_used - TDB_OFF_UPPER_STEAL_EXTRA, TDB_OFF_UPPER_STEAL_EXTRA)) { add_stat(tdb, compare_wrong_offsetbits, 1); - return ret; + return false; } off = val & TDB_OFF_MASK; if (tdb_read_convert(tdb, off, rec, sizeof(*rec)) == -1) - return ret; - - if (rec_key_length(rec) != key->dsize) { - add_stat(tdb, compare_wrong_keylen, 1); - return ret; - } + return false; if ((h->h & ((1 << 11)-1)) != rec_hash(rec)) { add_stat(tdb, compare_wrong_rechash, 1); return false; } - rkey = tdb_access_read(tdb, off + sizeof(*rec), key->dsize, false); - if (!rkey) - return ret; - if (memcmp(rkey, key->dptr, key->dsize) == 0) - ret = true; - else - add_stat(tdb, compare_wrong_keycmp, 1); - tdb_access_release(tdb, rkey); - return ret; + return key_matches(tdb, rec, off, key); } static tdb_off_t hbucket_off(tdb_off_t group_start, unsigned bucket) @@ -147,6 +156,65 @@ static tdb_off_t hlock_range(tdb_off_t group, tdb_off_t *size) return group << (64 - (TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS)); } +static tdb_off_t COLD find_in_chain(struct tdb_context *tdb, + struct tdb_data key, + tdb_off_t chain, + struct hash_info *h, + struct tdb_used_record *rec, + struct traverse_info *tinfo) +{ + tdb_off_t off, next; + + /* In case nothing is free, we set these to zero. */ + h->home_bucket = h->found_bucket = 0; + + for (off = chain; off; off = next) { + unsigned int i; + + h->group_start = off; + if (tdb_read_convert(tdb, off, h->group, sizeof(h->group))) + return TDB_OFF_ERR; + + for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) { + tdb_off_t recoff; + if (!h->group[i]) { + /* Remember this empty bucket. */ + h->home_bucket = h->found_bucket = i; + continue; + } + + /* We can insert extra bits via add_to_hash + * empty bucket logic. */ + recoff = h->group[i] & TDB_OFF_MASK; + if (tdb_read_convert(tdb, recoff, rec, sizeof(*rec))) + return TDB_OFF_ERR; + + if (key_matches(tdb, rec, recoff, &key)) { + h->home_bucket = h->found_bucket = i; + + if (tinfo) { + tinfo->levels[tinfo->num_levels] + .hashtable = off; + tinfo->levels[tinfo->num_levels] + .total_buckets + = 1 << TDB_HASH_GROUP_BITS; + tinfo->levels[tinfo->num_levels].entry + = i; + tinfo->num_levels++; + } + return recoff; + } + } + next = tdb_read_off(tdb, off + + offsetof(struct tdb_chain, next)); + if (next == TDB_OFF_ERR) + return TDB_OFF_ERR; + if (next) + next += sizeof(struct tdb_used_record); + } + return 0; +} + /* This is the core routine which searches the hashtable for an entry. * On error, no locks are held and TDB_OFF_ERR is returned. * Otherwise, hinfo is filled in (and the optional tinfo). @@ -182,7 +250,7 @@ tdb_off_t find_and_lock(struct tdb_context *tdb, tinfo->levels[0].total_buckets = 1 << TDB_HASH_GROUP_BITS; } - while (likely(h->hash_used < 64)) { + while (h->hash_used <= 64) { /* Read in the hash group. */ h->group_start = hashtable + group * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS); @@ -239,8 +307,7 @@ tdb_off_t find_and_lock(struct tdb_context *tdb, return 0; } - /* FIXME: We hit the bottom. Chain! */ - abort(); + return find_in_chain(tdb, key, hashtable, h, rec, tinfo); fail: tdb_unlock_hashes(tdb, h->hlock_start, h->hlock_range, ltype); @@ -326,6 +393,41 @@ int replace_in_hash(struct tdb_context *tdb, encode_offset(new_off, h)); } +/* We slot in anywhere that's empty in the chain. */ +static int COLD add_to_chain(struct tdb_context *tdb, + tdb_off_t subhash, + tdb_off_t new_off) +{ + size_t entry = tdb_find_zero_off(tdb, subhash, 1< 64) - abort(); + return add_to_chain(tdb, subhash, off); h.h = hash_record(tdb, off); gnum = use_bits(&h, TDB_SUBLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS); - h.group_start = subhash + sizeof(struct tdb_used_record) + h.group_start = subhash + gnum * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS); h.home_bucket = use_bits(&h, TDB_HASH_GROUP_BITS); @@ -356,21 +457,29 @@ static int add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash, static int expand_group(struct tdb_context *tdb, struct hash_info *h) { - unsigned bucket, num_vals, i; + unsigned bucket, num_vals, i, magic; + size_t subsize; tdb_off_t subhash; tdb_off_t vals[1 << TDB_HASH_GROUP_BITS]; /* Attach new empty subhash under fullest bucket. */ bucket = fullest_bucket(tdb, h->group, h->home_bucket); - subhash = alloc(tdb, 0, sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS, - 0, false); + if (h->hash_used == 64) { + add_stat(tdb, alloc_chain, 1); + subsize = sizeof(struct tdb_chain); + magic = TDB_CHAIN_MAGIC; + } else { + add_stat(tdb, alloc_subhash, 1); + subsize = (sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS); + magic = TDB_HTABLE_MAGIC; + } + + subhash = alloc(tdb, 0, subsize, 0, magic, false); if (subhash == TDB_OFF_ERR) return -1; - add_stat(tdb, alloc_subhash, 1); - if (zero_out(tdb, subhash + sizeof(struct tdb_used_record), - sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS) == -1) + if (zero_out(tdb, subhash + sizeof(struct tdb_used_record), subsize)) return -1; /* Remove any which are destined for bucket or are in wrong place. */ @@ -390,6 +499,9 @@ static int expand_group(struct tdb_context *tdb, struct hash_info *h) /* Overwrite expanded bucket with subhash pointer. */ h->group[bucket] = subhash | (1ULL << TDB_OFF_UPPER_STEAL_SUBHASH_BIT); + /* Point to actual contents of record. */ + subhash += sizeof(struct tdb_used_record); + /* Put values back. */ for (i = 0; i < num_vals; i++) { unsigned this_bucket = vals[i] & TDB_OFF_HASH_GROUP_MASK; @@ -444,10 +556,6 @@ int delete_from_hash(struct tdb_context *tdb, struct hash_info *h) int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off) { - /* FIXME: chain! */ - if (h->hash_used >= 64) - abort(); - /* We hit an empty bucket during search? That's where it goes. */ if (!h->group[h->found_bucket]) { h->group[h->found_bucket] = encode_offset(new_off, h); @@ -456,6 +564,9 @@ int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off) h->group, sizeof(h->group)); } + if (h->hash_used > 64) + return add_to_chain(tdb, h->group_start, new_off); + /* We're full. Expand. */ if (expand_group(tdb, h) == -1) return -1; @@ -534,7 +645,11 @@ again: tlevel++; tlevel->hashtable = off + sizeof(struct tdb_used_record); tlevel->entry = 0; - tlevel->total_buckets = (1 << TDB_SUBLEVEL_HASH_BITS); + /* Next level is a chain? */ + if (unlikely(tinfo->num_levels == TDB_MAX_LEVELS + 1)) + tlevel->total_buckets = (1 << TDB_HASH_GROUP_BITS); + else + tlevel->total_buckets = (1 << TDB_SUBLEVEL_HASH_BITS); goto again; } @@ -542,6 +657,20 @@ again: if (tinfo->num_levels == 1) return 0; + /* Handle chained entries. */ + if (unlikely(tinfo->num_levels == TDB_MAX_LEVELS + 1)) { + tlevel->hashtable = tdb_read_off(tdb, tlevel->hashtable + + offsetof(struct tdb_chain, + next)); + if (tlevel->hashtable == TDB_OFF_ERR) + return TDB_OFF_ERR; + if (tlevel->hashtable) { + tlevel->hashtable += sizeof(struct tdb_used_record); + tlevel->entry = 0; + goto again; + } + } + /* Go back up and keep searching. */ tinfo->num_levels--; tlevel--; @@ -574,7 +703,7 @@ int next_in_hash(struct tdb_context *tdb, int ltype, ltype); return -1; } - if (rec_magic(&rec) != TDB_MAGIC) { + if (rec_magic(&rec) != TDB_USED_MAGIC) { tdb_logerr(tdb, TDB_ERR_CORRUPT, TDB_DEBUG_FATAL, "next_in_hash:"