tdb2: allow multiple chain locks.
[ccan] / ccan / tdb2 / hash.c
index e1b1dfe34bdbfa071e8e4b8a43595c56af247b9d..1359cfecd66dc6280e3314e78fd2e7e5ea91c8bd 100644 (file)
@@ -1,7 +1,7 @@
- /* 
+ /*
    Trivial Database 2: hash handling
    Copyright (C) Rusty Russell 2010
-   
+
    This library is free software; you can redistribute it and/or
    modify it under the terms of the GNU Lesser General Public
    License as published by the Free Software Foundation; either
 */
 #include "private.h"
 #include <assert.h>
-#include <ccan/hash/hash.h>
-
-static uint64_t jenkins_hash(const void *key, size_t length, uint64_t seed,
-                            void *arg)
-{
-       uint64_t ret;
-       /* hash64_stable assumes lower bits are more important; they are a
-        * slightly better hash.  We use the upper bits first, so swap them. */
-       ret = hash64_stable((const unsigned char *)key, length, seed);
-       return (ret >> 32) | (ret << 32);
-}
-
-void tdb_hash_init(struct tdb_context *tdb)
-{
-       tdb->khash = jenkins_hash;
-       tdb->hash_priv = NULL;
-}
 
 uint64_t tdb_hash(struct tdb_context *tdb, const void *ptr, size_t len)
 {
-       return tdb->khash(ptr, len, tdb->hash_seed, tdb->hash_priv);
+       return tdb->hash_fn(ptr, len, tdb->hash_seed, tdb->hash_data);
 }
 
 uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off)
@@ -47,16 +30,18 @@ uint64_t hash_record(struct tdb_context *tdb, tdb_off_t off)
        uint64_t klen, hash;
 
        r = tdb_access_read(tdb, off, sizeof(*r), true);
-       if (!r)
+       if (TDB_PTR_IS_ERR(r)) {
                /* FIXME */
                return 0;
+       }
 
        klen = rec_key_length(r);
        tdb_access_release(tdb, r);
 
        key = tdb_access_read(tdb, off + sizeof(*r), klen, false);
-       if (!key)
+       if (TDB_PTR_IS_ERR(key)) {
                return 0;
+       }
 
        hash = tdb_hash(tdb, key, klen);
        tdb_access_release(tdb, key);
@@ -78,43 +63,45 @@ static uint32_t use_bits(struct hash_info *h, unsigned num)
        return bits_from(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)
+static tdb_bool_err key_matches(struct tdb_context *tdb,
+                               const struct tdb_used_record *rec,
+                               tdb_off_t off,
+                               const struct tdb_data *key)
 {
-       bool ret = false;
+       tdb_bool_err ret = false;
        const char *rkey;
 
        if (rec_key_length(rec) != key->dsize) {
-               add_stat(tdb, compare_wrong_keylen, 1);
+               tdb->stats.compare_wrong_keylen++;
                return ret;
        }
 
        rkey = tdb_access_read(tdb, off + sizeof(*rec), key->dsize, false);
-       if (!rkey)
-               return ret;
+       if (TDB_PTR_IS_ERR(rkey)) {
+               return TDB_PTR_ERR(rkey);
+       }
        if (memcmp(rkey, key->dptr, key->dsize) == 0)
                ret = true;
        else
-               add_stat(tdb, compare_wrong_keycmp, 1);
+               tdb->stats.compare_wrong_keycmp++;
        tdb_access_release(tdb, rkey);
        return ret;
 }
 
 /* Does entry match? */
-static bool match(struct tdb_context *tdb,
-                 struct hash_info *h,
-                 const struct tdb_data *key,
-                 tdb_off_t val,
-                 struct tdb_used_record *rec)
+static tdb_bool_err match(struct tdb_context *tdb,
+                         struct hash_info *h,
+                         const struct tdb_data *key,
+                         tdb_off_t val,
+                         struct tdb_used_record *rec)
 {
        tdb_off_t off;
+       enum TDB_ERROR ecode;
 
-       add_stat(tdb, compares, 1);
+       tdb->stats.compares++;
        /* Desired bucket must match. */
        if (h->home_bucket != (val & TDB_OFF_HASH_GROUP_MASK)) {
-               add_stat(tdb, compare_wrong_bucket, 1);
+               tdb->stats.compare_wrong_bucket++;
                return false;
        }
 
@@ -122,16 +109,18 @@ static bool match(struct tdb_context *tdb,
        if (bits_from(val, TDB_OFF_HASH_EXTRA_BIT, TDB_OFF_UPPER_STEAL_EXTRA)
            != bits_from(h->h, 64 - h->hash_used - TDB_OFF_UPPER_STEAL_EXTRA,
                    TDB_OFF_UPPER_STEAL_EXTRA)) {
-               add_stat(tdb, compare_wrong_offsetbits, 1);
+               tdb->stats.compare_wrong_offsetbits++;
                return false;
        }
 
        off = val & TDB_OFF_MASK;
-       if (tdb_read_convert(tdb, off, rec, sizeof(*rec)) == -1)
-               return false;
+       ecode = tdb_read_convert(tdb, off, rec, sizeof(*rec));
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
        if ((h->h & ((1 << 11)-1)) != rec_hash(rec)) {
-               add_stat(tdb, compare_wrong_rechash, 1);
+               tdb->stats.compare_wrong_rechash++;
                return false;
        }
 
@@ -164,6 +153,7 @@ static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
                                    struct traverse_info *tinfo)
 {
        tdb_off_t off, next;
+       enum TDB_ERROR ecode;
 
        /* In case nothing is free, we set these to zero. */
        h->home_bucket = h->found_bucket = 0;
@@ -172,8 +162,10 @@ static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
                unsigned int i;
 
                h->group_start = off;
-               if (tdb_read_convert(tdb, off, h->group, sizeof(h->group)))
-                       return TDB_OFF_ERR;
+               ecode = tdb_read_convert(tdb, off, h->group, sizeof(h->group));
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
+               }
 
                for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
                        tdb_off_t recoff;
@@ -186,10 +178,17 @@ static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
                        /* 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;
+                       ecode = tdb_read_convert(tdb, recoff, rec,
+                                                sizeof(*rec));
+                       if (ecode != TDB_SUCCESS) {
+                               return ecode;
+                       }
 
-                       if (key_matches(tdb, rec, recoff, &key)) {
+                       ecode = key_matches(tdb, rec, recoff, &key);
+                       if (ecode < 0) {
+                               return ecode;
+                       }
+                       if (ecode == 1) {
                                h->home_bucket = h->found_bucket = i;
 
                                if (tinfo) {
@@ -207,8 +206,9 @@ static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
                }
                next = tdb_read_off(tdb, off
                                    + offsetof(struct tdb_chain, next));
-               if (next == TDB_OFF_ERR)
-                       return TDB_OFF_ERR;
+               if (TDB_OFF_IS_ERR(next)) {
+                       return next;
+               }
                if (next)
                        next += sizeof(struct tdb_used_record);
        }
@@ -216,7 +216,7 @@ static tdb_off_t COLD find_in_chain(struct tdb_context *tdb,
 }
 
 /* This is the core routine which searches the hashtable for an entry.
- * On error, no locks are held and TDB_OFF_ERR is returned.
+ * On error, no locks are held and -ve is returned.
  * Otherwise, hinfo is filled in (and the optional tinfo).
  * If not found, the return value is 0.
  * If found, the return value is the offset, and *rec is the record. */
@@ -229,6 +229,7 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
 {
        uint32_t i, group;
        tdb_off_t hashtable;
+       enum TDB_ERROR ecode;
 
        h->h = tdb_hash(tdb, key.dptr, key.dsize);
        h->hash_used = 0;
@@ -236,16 +237,18 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
        h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
 
        h->hlock_start = hlock_range(group, &h->hlock_range);
-       if (tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
-                           TDB_LOCK_WAIT))
-               return TDB_OFF_ERR;
+       ecode = tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
+                               TDB_LOCK_WAIT);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
        hashtable = offsetof(struct tdb_header, hashtable);
        if (tinfo) {
                tinfo->toplevel_group = group;
                tinfo->num_levels = 1;
                tinfo->levels[0].entry = 0;
-               tinfo->levels[0].hashtable = hashtable 
+               tinfo->levels[0].hashtable = hashtable
                        + (group << TDB_HASH_GROUP_BITS) * sizeof(tdb_off_t);
                tinfo->levels[0].total_buckets = 1 << TDB_HASH_GROUP_BITS;
        }
@@ -255,9 +258,11 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
                h->group_start = hashtable
                        + group * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
 
-               if (tdb_read_convert(tdb, h->group_start, &h->group,
-                                    sizeof(h->group)) == -1)
+               ecode = tdb_read_convert(tdb, h->group_start, &h->group,
+                                        sizeof(h->group));
+               if (ecode != TDB_SUCCESS) {
                        goto fail;
+               }
 
                /* Pointer to another hash table?  Go down... */
                if (is_subhash(h->group[h->home_bucket])) {
@@ -288,14 +293,20 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
                     i < (1 << TDB_HASH_GROUP_BITS);
                     i++, h->found_bucket = ((h->found_bucket+1)
                                             % (1 << TDB_HASH_GROUP_BITS))) {
+                       tdb_bool_err berr;
                        if (is_subhash(h->group[h->found_bucket]))
                                continue;
 
                        if (!h->group[h->found_bucket])
                                break;
 
-                       if (match(tdb, h, &key, h->group[h->found_bucket],
-                                 rec)) {
+                       berr = match(tdb, h, &key, h->group[h->found_bucket],
+                                    rec);
+                       if (berr < 0) {
+                               ecode = berr;
+                               goto fail;
+                       }
+                       if (berr) {
                                if (tinfo) {
                                        tinfo->levels[tinfo->num_levels-1].entry
                                                += h->found_bucket;
@@ -311,7 +322,7 @@ tdb_off_t find_and_lock(struct tdb_context *tdb,
 
 fail:
        tdb_unlock_hashes(tdb, h->hlock_start, h->hlock_range, ltype);
-       return TDB_OFF_ERR;
+       return ecode;
 }
 
 /* I wrote a simple test, expanding a hash to 2GB, for the following
@@ -384,42 +395,55 @@ static tdb_off_t encode_offset(tdb_off_t new_off, struct hash_info *h)
                   << TDB_OFF_HASH_EXTRA_BIT);
 }
 
-/* Simply overwrite the hash entry we found before. */ 
-int replace_in_hash(struct tdb_context *tdb,
-                   struct hash_info *h,
-                   tdb_off_t new_off)
+/* Simply overwrite the hash entry we found before. */
+enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
+                              struct hash_info *h,
+                              tdb_off_t new_off)
 {
        return tdb_write_off(tdb, hbucket_off(h->group_start, h->found_bucket),
                             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)
+static enum TDB_ERROR 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);
+       tdb_off_t entry;
+       enum TDB_ERROR ecode;
+
+       entry = tdb_find_zero_off(tdb, subhash, 1<<TDB_HASH_GROUP_BITS);
+       if (TDB_OFF_IS_ERR(entry)) {
+               return entry;
+       }
 
        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 (TDB_OFF_IS_ERR(next)) {
+                       return next;
+               }
 
                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;
+                       if (TDB_OFF_IS_ERR(next))
+                               return next;
+                       ecode = zero_out(tdb,
+                                        next+sizeof(struct tdb_used_record),
+                                        sizeof(struct tdb_chain));
+                       if (ecode != TDB_SUCCESS) {
+                               return ecode;
+                       }
+                       ecode = tdb_write_off(tdb, subhash
+                                             + offsetof(struct tdb_chain,
+                                                        next),
+                                             next);
+                       if (ecode != TDB_SUCCESS) {
+                               return ecode;
+                       }
                }
                return add_to_chain(tdb, next, new_off);
        }
@@ -429,8 +453,8 @@ static int COLD add_to_chain(struct tdb_context *tdb,
 }
 
 /* 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)
+static enum TDB_ERROR add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash,
+                                    unsigned hash_used, tdb_off_t val)
 {
        tdb_off_t off = (val & TDB_OFF_MASK), *group;
        struct hash_info h;
@@ -449,38 +473,44 @@ static int add_to_subhash(struct tdb_context *tdb, tdb_off_t subhash,
 
        group = tdb_access_write(tdb, h.group_start,
                                 sizeof(*group) << TDB_HASH_GROUP_BITS, true);
-       if (!group)
-               return -1;
+       if (TDB_PTR_IS_ERR(group)) {
+               return TDB_PTR_ERR(group);
+       }
        force_into_group(group, h.home_bucket, encode_offset(off, &h));
        return tdb_access_commit(tdb, group);
 }
 
-static int expand_group(struct tdb_context *tdb, struct hash_info *h)
+static enum TDB_ERROR expand_group(struct tdb_context *tdb, struct hash_info *h)
 {
        unsigned bucket, num_vals, i, magic;
        size_t subsize;
        tdb_off_t subhash;
        tdb_off_t vals[1 << TDB_HASH_GROUP_BITS];
+       enum TDB_ERROR ecode;
 
        /* Attach new empty subhash under fullest bucket. */
        bucket = fullest_bucket(tdb, h->group, h->home_bucket);
 
        if (h->hash_used == 64) {
-               add_stat(tdb, alloc_chain, 1);
+               tdb->stats.alloc_chain++;
                subsize = sizeof(struct tdb_chain);
                magic = TDB_CHAIN_MAGIC;
        } else {
-               add_stat(tdb, alloc_subhash, 1);
+               tdb->stats.alloc_subhash++;
                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;
+       if (TDB_OFF_IS_ERR(subhash)) {
+               return subhash;
+       }
 
-       if (zero_out(tdb, subhash + sizeof(struct tdb_used_record), subsize))
-               return -1;
+       ecode = zero_out(tdb, subhash + sizeof(struct tdb_used_record),
+                        subsize);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
        /* Remove any which are destined for bucket or are in wrong place. */
        num_vals = 0;
@@ -507,17 +537,19 @@ static int expand_group(struct tdb_context *tdb, struct hash_info *h)
                unsigned this_bucket = vals[i] & TDB_OFF_HASH_GROUP_MASK;
 
                if (this_bucket == bucket) {
-                       if (add_to_subhash(tdb, subhash, h->hash_used, vals[i]))
-                               return -1;
+                       ecode = add_to_subhash(tdb, subhash, h->hash_used,
+                                              vals[i]);
+                       if (ecode != TDB_SUCCESS)
+                               return ecode;
                } else {
                        /* There should be room to put this back. */
                        force_into_group(h->group, this_bucket, vals[i]);
                }
        }
-       return 0;
+       return TDB_SUCCESS;
 }
 
-int delete_from_hash(struct tdb_context *tdb, struct hash_info *h)
+enum TDB_ERROR delete_from_hash(struct tdb_context *tdb, struct hash_info *h)
 {
        unsigned int i, num_movers = 0;
        tdb_off_t movers[1 << TDB_HASH_GROUP_BITS];
@@ -554,8 +586,11 @@ int delete_from_hash(struct tdb_context *tdb, struct hash_info *h)
                                 h->group, sizeof(h->group));
 }
 
-int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off)
+enum TDB_ERROR add_to_hash(struct tdb_context *tdb, struct hash_info *h,
+                          tdb_off_t new_off)
 {
+       enum TDB_ERROR ecode;
+
        /* 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);
@@ -568,8 +603,10 @@ int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off)
                return add_to_chain(tdb, h->group_start, new_off);
 
        /* We're full.  Expand. */
-       if (expand_group(tdb, h) == -1)
-               return -1;
+       ecode = expand_group(tdb, h);
+       if (ecode != TDB_SUCCESS) {
+               return ecode;
+       }
 
        if (is_subhash(h->group[h->home_bucket])) {
                /* We were expanded! */
@@ -577,9 +614,11 @@ int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off)
                unsigned int gnum;
 
                /* Write back the modified group. */
-               if (tdb_write_convert(tdb, h->group_start, h->group,
-                                     sizeof(h->group)))
-                       return -1;
+               ecode = tdb_write_convert(tdb, h->group_start, h->group,
+                                         sizeof(h->group));
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
+               }
 
                /* Move hashinfo down a level. */
                hashtable = (h->group[h->home_bucket] & TDB_OFF_MASK)
@@ -588,28 +627,30 @@ int add_to_hash(struct tdb_context *tdb, struct hash_info *h, tdb_off_t new_off)
                h->home_bucket = use_bits(h, TDB_HASH_GROUP_BITS);
                h->group_start = hashtable
                        + gnum * (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
-               if (tdb_read_convert(tdb, h->group_start, &h->group,
-                                    sizeof(h->group)) == -1)
-                       return -1;
+               ecode = tdb_read_convert(tdb, h->group_start, &h->group,
+                                        sizeof(h->group));
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
+               }
        }
 
        /* Expanding the group must have made room if it didn't choose this
         * bucket. */
-       if (put_into_group(h->group, h->home_bucket, encode_offset(new_off, h)))
+       if (put_into_group(h->group, h->home_bucket, encode_offset(new_off,h))){
                return tdb_write_convert(tdb, h->group_start,
                                         h->group, sizeof(h->group));
+       }
 
        /* This can happen if all hashes in group (and us) dropped into same
         * group in subhash. */
        return add_to_hash(tdb, h, new_off);
 }
 
-/* Traverse support: returns offset of record, or 0 or TDB_OFF_ERR. */
+/* Traverse support: returns offset of record, or 0 or -ve error. */
 static tdb_off_t iterate_hash(struct tdb_context *tdb,
                              struct traverse_info *tinfo)
 {
-       tdb_off_t off, val;
-       unsigned int i;
+       tdb_off_t off, val, i;
        struct traverse_level *tlevel;
 
        tlevel = &tinfo->levels[tinfo->num_levels-1];
@@ -620,9 +661,14 @@ again:
             i != tlevel->total_buckets;
             i = tdb_find_nonzero_off(tdb, tlevel->hashtable,
                                      i+1, tlevel->total_buckets)) {
+               if (TDB_OFF_IS_ERR(i)) {
+                       return i;
+               }
+
                val = tdb_read_off(tdb, tlevel->hashtable+sizeof(tdb_off_t)*i);
-               if (unlikely(val == TDB_OFF_ERR))
-                       return TDB_OFF_ERR;
+               if (TDB_OFF_IS_ERR(val)) {
+                       return val;
+               }
 
                off = val & TDB_OFF_MASK;
 
@@ -662,8 +708,9 @@ again:
                tlevel->hashtable = tdb_read_off(tdb, tlevel->hashtable
                                                 + offsetof(struct tdb_chain,
                                                            next));
-               if (tlevel->hashtable == TDB_OFF_ERR)
-                       return TDB_OFF_ERR;
+               if (TDB_OFF_IS_ERR(tlevel->hashtable)) {
+                       return tlevel->hashtable;
+               }
                if (tlevel->hashtable) {
                        tlevel->hashtable += sizeof(struct tdb_used_record);
                        tlevel->entry = 0;
@@ -677,38 +724,45 @@ again:
        goto again;
 }
 
-/* Return 1 if we find something, 0 if not, -1 on error. */
-int next_in_hash(struct tdb_context *tdb, int ltype,
-                struct traverse_info *tinfo,
-                TDB_DATA *kbuf, size_t *dlen)
+/* Return success if we find something, TDB_ERR_NOEXIST if none. */
+enum TDB_ERROR next_in_hash(struct tdb_context *tdb,
+                           struct traverse_info *tinfo,
+                           TDB_DATA *kbuf, size_t *dlen)
 {
        const unsigned group_bits = TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS;
        tdb_off_t hl_start, hl_range, off;
+       enum TDB_ERROR ecode;
 
        while (tinfo->toplevel_group < (1 << group_bits)) {
                hl_start = (tdb_off_t)tinfo->toplevel_group
                        << (64 - group_bits);
                hl_range = 1ULL << group_bits;
-               if (tdb_lock_hashes(tdb, hl_start, hl_range, ltype,
-                                   TDB_LOCK_WAIT) != 0)
-                       return -1;
+               ecode = tdb_lock_hashes(tdb, hl_start, hl_range, F_RDLCK,
+                                       TDB_LOCK_WAIT);
+               if (ecode != TDB_SUCCESS) {
+                       return ecode;
+               }
 
                off = iterate_hash(tdb, tinfo);
                if (off) {
                        struct tdb_used_record rec;
 
-                       if (tdb_read_convert(tdb, off, &rec, sizeof(rec))) {
-                               tdb_unlock_hashes(tdb,
-                                                 hl_start, hl_range, ltype);
-                               return -1;
+                       if (TDB_OFF_IS_ERR(off)) {
+                               ecode = off;
+                               goto fail;
+                       }
+
+                       ecode = tdb_read_convert(tdb, off, &rec, sizeof(rec));
+                       if (ecode != TDB_SUCCESS) {
+                               goto fail;
                        }
                        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;
+                               ecode = tdb_logerr(tdb, TDB_ERR_CORRUPT,
+                                                  TDB_LOG_ERROR,
+                                                  "next_in_hash:"
+                                                  " corrupt record at %llu",
+                                                  (long long)off);
+                               goto fail;
                        }
 
                        kbuf->dsize = rec_key_length(&rec);
@@ -716,33 +770,40 @@ int next_in_hash(struct tdb_context *tdb, int ltype,
                        /* They want data as well? */
                        if (dlen) {
                                *dlen = rec_data_length(&rec);
-                               kbuf->dptr = tdb_alloc_read(tdb, 
+                               kbuf->dptr = tdb_alloc_read(tdb,
                                                            off + sizeof(rec),
                                                            kbuf->dsize
                                                            + *dlen);
                        } else {
-                               kbuf->dptr = tdb_alloc_read(tdb, 
+                               kbuf->dptr = tdb_alloc_read(tdb,
                                                            off + sizeof(rec),
                                                            kbuf->dsize);
                        }
-                       tdb_unlock_hashes(tdb, hl_start, hl_range, ltype);
-                       return kbuf->dptr ? 1 : -1;
+                       tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
+                       if (TDB_PTR_IS_ERR(kbuf->dptr)) {
+                               return TDB_PTR_ERR(kbuf->dptr);
+                       }
+                       return TDB_SUCCESS;
                }
 
-               tdb_unlock_hashes(tdb, hl_start, hl_range, ltype);
+               tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
 
                tinfo->toplevel_group++;
                tinfo->levels[0].hashtable
                        += (sizeof(tdb_off_t) << TDB_HASH_GROUP_BITS);
                tinfo->levels[0].entry = 0;
        }
-       return 0;
+       return TDB_ERR_NOEXIST;
+
+fail:
+       tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
+       return ecode;
+
 }
 
-/* Return 1 if we find something, 0 if not, -1 on error. */
-int first_in_hash(struct tdb_context *tdb, int ltype,
-                 struct traverse_info *tinfo,
-                 TDB_DATA *kbuf, size_t *dlen)
+enum TDB_ERROR first_in_hash(struct tdb_context *tdb,
+                            struct traverse_info *tinfo,
+                            TDB_DATA *kbuf, size_t *dlen)
 {
        tinfo->prev = 0;
        tinfo->toplevel_group = 0;
@@ -751,16 +812,16 @@ int first_in_hash(struct tdb_context *tdb, int ltype,
        tinfo->levels[0].entry = 0;
        tinfo->levels[0].total_buckets = (1 << TDB_HASH_GROUP_BITS);
 
-       return next_in_hash(tdb, ltype, tinfo, kbuf, dlen);
+       return next_in_hash(tdb, tinfo, kbuf, dlen);
 }
 
 /* Even if the entry isn't in this hash bucket, you'd have to lock this
  * bucket to find it. */
-static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
-                    int ltype, enum tdb_lock_flags waitflag,
-                    const char *func)
+static enum TDB_ERROR chainlock(struct tdb_context *tdb, const TDB_DATA *key,
+                               int ltype, enum tdb_lock_flags waitflag,
+                               const char *func)
 {
-       int ret;
+       enum TDB_ERROR ecode;
        uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
        tdb_off_t lockstart, locksize;
        unsigned int group, gbits;
@@ -770,19 +831,20 @@ static int chainlock(struct tdb_context *tdb, const TDB_DATA *key,
 
        lockstart = hlock_range(group, &locksize);
 
-       ret = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
+       ecode = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
        tdb_trace_1rec(tdb, func, *key);
-       return ret;
+       return ecode;
 }
 
 /* lock/unlock one hash chain. This is meant to be used to reduce
    contention - it cannot guarantee how many records will be locked */
-int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
+enum TDB_ERROR tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
 {
-       return chainlock(tdb, &key, F_WRLCK, TDB_LOCK_WAIT, "tdb_chainlock");
+       return tdb->last_error = chainlock(tdb, &key, F_WRLCK, TDB_LOCK_WAIT,
+                                          "tdb_chainlock");
 }
 
-int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
+void tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
 {
        uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
        tdb_off_t lockstart, locksize;
@@ -794,5 +856,26 @@ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
        lockstart = hlock_range(group, &locksize);
 
        tdb_trace_1rec(tdb, "tdb_chainunlock", key);
-       return tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
+       tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
+}
+
+enum TDB_ERROR tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+       return tdb->last_error = chainlock(tdb, &key, F_RDLCK, TDB_LOCK_WAIT,
+                                          "tdb_chainlock_read");
+}
+
+void tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
+{
+       uint64_t h = tdb_hash(tdb, key.dptr, key.dsize);
+       tdb_off_t lockstart, locksize;
+       unsigned int group, gbits;
+
+       gbits = TDB_TOPLEVEL_HASH_BITS - TDB_HASH_GROUP_BITS;
+       group = bits_from(h, 64 - gbits, gbits);
+
+       lockstart = hlock_range(group, &locksize);
+
+       tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
+       tdb_unlock_hashes(tdb, lockstart, locksize, F_RDLCK);
 }