tdb2: update documentation
[ccan] / ccan / tdb2 / hash.c
index f8d6b5812ebfb5307e35f3203dbe0ded9abdbde5..18f6b1720eb662b2a0f140a6d23f7c844269b223 100644 (file)
@@ -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,
@@ -85,38 +109,33 @@ static bool match(struct tdb_context *tdb,
                  tdb_off_t val,
                  struct tdb_used_record *rec)
 {
-       bool ret;
-       const unsigned char *rkey;
        tdb_off_t off;
 
-       /* FIXME: Handle hash value truncated. */
-       if (bits(val, TDB_OFF_HASH_TRUNCATED_BIT, 1))
-               abort();
-
+       add_stat(tdb, compares, 1);
        /* Desired bucket must match. */
-       if (h->home_bucket != (val & TDB_OFF_HASH_GROUP_MASK))
+       if (h->home_bucket != (val & TDB_OFF_HASH_GROUP_MASK)) {
+               add_stat(tdb, compare_wrong_bucket, 1);
                return false;
+       }
 
        /* Top bits of offset == next bits of hash. */
        if (bits(val, TDB_OFF_HASH_EXTRA_BIT, TDB_OFF_UPPER_STEAL_EXTRA)
            != bits(h->h, 64 - h->hash_used - TDB_OFF_UPPER_STEAL_EXTRA,
-                   TDB_OFF_UPPER_STEAL_EXTRA))
+                   TDB_OFF_UPPER_STEAL_EXTRA)) {
+               add_stat(tdb, compare_wrong_offsetbits, 1);
                return false;
+       }
 
        off = val & TDB_OFF_MASK;
        if (tdb_read_convert(tdb, off, rec, sizeof(*rec)) == -1)
                return false;
 
-       /* FIXME: check extra bits in header? */
-       if (rec_key_length(rec) != key->dsize)
+       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 false;
-       ret = (memcmp(rkey, key->dptr, key->dsize) == 0);
-       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)
@@ -125,10 +144,9 @@ static tdb_off_t hbucket_off(tdb_off_t group_start, unsigned bucket)
                + (bucket % (1 << TDB_HASH_GROUP_BITS)) * sizeof(tdb_off_t);
 }
 
-/* Truncated hashes can't be all 1: that's how we spot a sub-hash */
 bool is_subhash(tdb_off_t val)
 {
-       return val >> (64-TDB_OFF_UPPER_STEAL) == (1<<TDB_OFF_UPPER_STEAL) - 1;
+       return (val >> TDB_OFF_UPPER_STEAL_SUBHASH_BIT) & 1;
 }
 
 /* FIXME: Guess the depth, don't over-lock! */
@@ -138,6 +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).
@@ -173,7 +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);
@@ -230,8 +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);
@@ -241,8 +317,8 @@ fail:
 /* I wrote a simple test, expanding a hash to 2GB, for the following
  * cases:
  * 1) Expanding all the buckets at once,
- * 2) Expanding the most-populated bucket,
- * 3) Expanding the bucket we wanted to place the new entry ito.
+ * 2) Expanding the bucket we wanted to place the new entry into.
+ * 3) Expanding the most-populated bucket,
  *
  * I measured the worst/average/best density during this process.
  * 1) 3%/16%/30%
@@ -317,6 +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<<TDB_HASH_GROUP_BITS);
+
+       if (entry == 1 << TDB_HASH_GROUP_BITS) {
+               tdb_off_t next;
+
+               next = tdb_read_off(tdb, subhash
+                                   + offsetof(struct tdb_chain, next));
+               if (next == TDB_OFF_ERR)
+                       return -1;
+
+               if (!next) {
+                       next = alloc(tdb, 0, sizeof(struct tdb_chain), 0,
+                                    TDB_CHAIN_MAGIC, false);
+                       if (next == TDB_OFF_ERR)
+                               return -1;
+                       if (zero_out(tdb, next+sizeof(struct tdb_used_record),
+                                    sizeof(struct tdb_chain)))
+                               return -1;
+                       if (tdb_write_off(tdb, subhash
+                                         + offsetof(struct tdb_chain, next),
+                                         next) != 0)
+                               return -1;
+               }
+               return add_to_chain(tdb, next, new_off);
+       }
+
+       return tdb_write_off(tdb, subhash + entry * sizeof(tdb_off_t),
+                            new_off);
+}
+
 /* Add into a newly created subhash. */
 static int add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash,
                          unsigned hash_used, tdb_off_t val)
@@ -327,14 +438,12 @@ static int add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash,
 
        h.hash_used = hash_used;
 
-       /* FIXME chain if hash_used == 64 */
        if (hash_used + TDB_SUBLEVEL_HASH_BITS > 64)
-               abort();
+               return add_to_chain(tdb, subhash, off);
 
-       /* FIXME: Do truncated hash bits if we can! */
        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);
 
@@ -348,21 +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. */
@@ -380,7 +497,10 @@ static int expand_group(struct tdb_context *tdb, struct hash_info *h)
        /* assert(num_vals); */
 
        /* Overwrite expanded bucket with subhash pointer. */
-       h->group[bucket] = subhash | ~((1ULL << (64 - TDB_OFF_UPPER_STEAL))-1);
+       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++) {
@@ -436,10 +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);
@@ -448,6 +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;
@@ -526,7 +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;
        }
 
@@ -534,6 +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--;
@@ -566,11 +703,12 @@ int next_in_hash(struct tdb_context *tdb, int ltype,
                                                  ltype);
                                return -1;
                        }
-                       if (rec_magic(&rec) != TDB_MAGIC) {
-                               tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
-                                        "next_in_hash:"
-                                        " corrupt record at %llu\n",
-                                        (long long)off);
+                       if (rec_magic(&rec) != TDB_USED_MAGIC) {
+                               tdb_logerr(tdb, TDB_ERR_CORRUPT,
+                                          TDB_DEBUG_FATAL,
+                                          "next_in_hash:"
+                                          " corrupt record at %llu",
+                                          (long long)off);
                                return -1;
                        }