Fix change 6093 which broke rbtdb when it grew too large

I misunderstood the purpose of the `heap_index` rdataset header
member; I thought it identified which heap to use, and could therefore
be smaller, the same size as `locknum` indexes. But in fact it is a
position within a heap, so it needs to be able to count up to the
total number of rdatasets in the rbtdb.

So this changes `heap_index` from `uint16_t` back to `unsigned int`.

To avoid re-embiggening the rdatasetheader, shrink the `count` member
from `uint32` to `uint16`. The `count` is used to rotate RRsets in
`dns_rdataset_towiresorted()`, so 16 bits is more than large enough.
This change also means we no longer need to avoid colliding with
`DNS_RDATASET_COUNT_UNDEFINED` i.e. UINT32_MAX.

Closes #3862

diff --git a/lib/dns/rbtdb.c b/lib/dns/rbtdb.c
index 45c282581e7a2b6764119c3043f80aa9873e4d39..b41f05d73d7c4cbe115a36217fe1902a1cded59d 100644 (file)
--- a/lib/dns/rbtdb.c
+++ b/lib/dns/rbtdb.c
@@ -240,23 +240,23 @@ typedef struct rdatasetheader {
        atomic_uint_least16_t attributes;
        dns_trust_t trust;
 
-       uint16_t heap_index;
+       unsigned int heap_index;
        /*%<
-        * Used for TTL-based cache cleaning. Matches type of dns_rbt_t->locknum
+        * Used for TTL-based cache cleaning.
         */
 
-       unsigned int resign_lsb : 1;
        isc_stdtime_t resign;
+       unsigned int resign_lsb : 1;
 
-       atomic_uint_fast32_t last_refresh_fail_ts;
-
-       atomic_uint_fast32_t count;
+       atomic_uint_fast16_t count;
        /*%<
         * Monotonically increased every time this rdataset is bound so that
         * it is used as the base of the starting point in DNS responses
         * when the "cyclic" rrset-order is required.
         */
 
+       atomic_uint_fast32_t last_refresh_fail_ts;
+
        struct noqname *noqname;
        struct noqname *closest;
        /*%<
@@ -781,14 +781,8 @@ setownercase(rdatasetheader_t *header, const dns_name_t *name);
  * 'init_count' is used to initialize 'newheader->count' which inturn
  * is used to determine where in the cycle rrset-order cyclic starts.
  * We don't lock this as we don't care about simultaneous updates.
- *
- * Note:
- *      Both init_count and header->count can be UINT32_MAX.
- *      The count on the returned rdataset however can't be as
- *      that indicates that the database does not implement cyclic
- *      processing.
  */
-static atomic_uint_fast32_t init_count = 0;
+static atomic_uint_fast16_t init_count = 0;
 
 /*
  * Locking
@@ -3112,9 +3106,6 @@ bind_rdataset(dns_rbtdb_t *rbtdb, dns_rbtnode_t *node, rdatasetheader_t *header,
        raw = (unsigned char *)header + sizeof(*header);
        rdataset->private3 = raw;
        rdataset->count = atomic_fetch_add_relaxed(&header->count, 1);
-       if (rdataset->count == UINT32_MAX) {
-               rdataset->count = 0;
-       }
 
        /*
         * Reset iterator state.