*/
#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)
r = tdb_access_read(tdb, off, sizeof(*r), true);
if (TDB_PTR_IS_ERR(r)) {
- tdb->ecode = TDB_PTR_ERR(r);
/* FIXME */
return 0;
}
key = tdb_access_read(tdb, off + sizeof(*r), klen, false);
if (TDB_PTR_IS_ERR(key)) {
- tdb->ecode = TDB_PTR_ERR(key);
return 0;
}
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 (TDB_PTR_IS_ERR(rkey)) {
- tdb->ecode = TDB_PTR_ERR(rkey);
- return ret;
+ 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;
}
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;
ecode = tdb_read_convert(tdb, off, rec, sizeof(*rec));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return false;
+ 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;
}
h->group_start = off;
ecode = tdb_read_convert(tdb, off, h->group, sizeof(h->group));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return TDB_OFF_ERR;
+ return ecode;
}
for (i = 0; i < (1 << TDB_HASH_GROUP_BITS); i++) {
ecode = tdb_read_convert(tdb, recoff, rec,
sizeof(*rec));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return TDB_OFF_ERR;
+ 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) {
next = tdb_read_off(tdb, off
+ offsetof(struct tdb_chain, next));
if (TDB_OFF_IS_ERR(next)) {
- tdb->ecode = next;
- return TDB_OFF_ERR;
+ return next;
}
if (next)
next += sizeof(struct tdb_used_record);
}
/* 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. */
ecode = tdb_lock_hashes(tdb, h->hlock_start, h->hlock_range, ltype,
TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return TDB_OFF_ERR;
+ return ecode;
}
hashtable = offsetof(struct tdb_header, hashtable);
ecode = tdb_read_convert(tdb, h->group_start, &h->group,
sizeof(h->group));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
goto fail;
}
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;
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
}
/* 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)
+enum TDB_ERROR replace_in_hash(struct tdb_context *tdb,
+ struct hash_info *h,
+ tdb_off_t new_off)
{
- enum TDB_ERROR ecode;
-
- ecode = tdb_write_off(tdb, hbucket_off(h->group_start, h->found_bucket),
- encode_offset(new_off, h));
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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)
{
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)) {
- tdb->ecode = entry;
- return -1;
+ return entry;
}
if (entry == 1 << TDB_HASH_GROUP_BITS) {
next = tdb_read_off(tdb, subhash
+ offsetof(struct tdb_chain, next));
if (TDB_OFF_IS_ERR(next)) {
- tdb->ecode = next;
- return -1;
+ 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 (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) {
- tdb->ecode = ecode;
- return -1;
+ return ecode;
}
ecode = tdb_write_off(tdb, subhash
+ offsetof(struct tdb_chain,
next),
next);
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
+ return ecode;
}
}
return add_to_chain(tdb, next, new_off);
}
- ecode = tdb_write_off(tdb, subhash + entry * sizeof(tdb_off_t),
- new_off);
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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)
+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;
unsigned int gnum;
- enum TDB_ERROR ecode;
h.hash_used = hash_used;
group = tdb_access_write(tdb, h.group_start,
sizeof(*group) << TDB_HASH_GROUP_BITS, true);
if (TDB_PTR_IS_ERR(group)) {
- tdb->ecode = TDB_PTR_ERR(group);
- return -1;
+ return TDB_PTR_ERR(group);
}
force_into_group(group, h.home_bucket, encode_offset(off, &h));
- ecode = tdb_access_commit(tdb, group);
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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;
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;
+ }
ecode = zero_out(tdb, subhash + sizeof(struct tdb_used_record),
subsize);
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
+ return ecode;
}
/* Remove any which are destined for bucket or are in wrong place. */
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];
- enum TDB_ERROR ecode;
h->group[h->found_bucket] = 0;
for (i = 1; i < (1 << TDB_HASH_GROUP_BITS); i++) {
}
/* Now we write back the hash group */
- ecode = tdb_write_convert(tdb, h->group_start,
- h->group, sizeof(h->group));
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ return tdb_write_convert(tdb, h->group_start,
+ 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;
if (!h->group[h->found_bucket]) {
h->group[h->found_bucket] = encode_offset(new_off, h);
/* Write back the modified group. */
- ecode = tdb_write_convert(tdb, h->group_start,
- h->group, sizeof(h->group));
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ return tdb_write_convert(tdb, h->group_start,
+ 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;
+ ecode = expand_group(tdb, h);
+ if (ecode != TDB_SUCCESS) {
+ return ecode;
+ }
if (is_subhash(h->group[h->home_bucket])) {
/* We were expanded! */
ecode = tdb_write_convert(tdb, h->group_start, h->group,
sizeof(h->group));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
+ return ecode;
}
/* Move hashinfo down a level. */
ecode = tdb_read_convert(tdb, h->group_start, &h->group,
sizeof(h->group));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
+ 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))){
- ecode = tdb_write_convert(tdb, h->group_start,
- h->group, sizeof(h->group));
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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
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)
{
i = tdb_find_nonzero_off(tdb, tlevel->hashtable,
i+1, tlevel->total_buckets)) {
if (TDB_OFF_IS_ERR(i)) {
- tdb->ecode = i;
- return TDB_OFF_ERR;
+ return i;
}
val = tdb_read_off(tdb, tlevel->hashtable+sizeof(tdb_off_t)*i);
if (TDB_OFF_IS_ERR(val)) {
- tdb->ecode = val;
- return TDB_OFF_ERR;
+ return val;
}
off = val & TDB_OFF_MASK;
+ offsetof(struct tdb_chain,
next));
if (TDB_OFF_IS_ERR(tlevel->hashtable)) {
- tdb->ecode = tlevel->hashtable;
- return TDB_OFF_ERR;
+ return tlevel->hashtable;
}
if (tlevel->hashtable) {
tlevel->hashtable += sizeof(struct tdb_used_record);
goto again;
}
-/* Return 1 if we find something, 0 if not, -1 on error. */
-int next_in_hash(struct tdb_context *tdb,
- 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;
ecode = tdb_lock_hashes(tdb, hl_start, hl_range, F_RDLCK,
TDB_LOCK_WAIT);
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
+ return ecode;
}
off = iterate_hash(tdb, tinfo);
if (off) {
struct tdb_used_record rec;
+ if (TDB_OFF_IS_ERR(off)) {
+ ecode = off;
+ goto fail;
+ }
+
ecode = tdb_read_convert(tdb, off, &rec, sizeof(rec));
if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- tdb_unlock_hashes(tdb,
- hl_start, hl_range, F_RDLCK);
- return -1;
+ goto fail;
}
if (rec_magic(&rec) != TDB_USED_MAGIC) {
- tdb_logerr(tdb, TDB_ERR_CORRUPT,
- TDB_LOG_ERROR,
- "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);
}
tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
if (TDB_PTR_IS_ERR(kbuf->dptr)) {
- tdb->ecode = TDB_PTR_ERR(kbuf->dptr);
- return -1;
+ return TDB_PTR_ERR(kbuf->dptr);
}
- return 1;
+ return TDB_SUCCESS;
}
tdb_unlock_hashes(tdb, hl_start, hl_range, F_RDLCK);
+= (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,
- 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;
/* 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)
{
enum TDB_ERROR ecode;
uint64_t h = tdb_hash(tdb, key->dptr, key->dsize);
ecode = tdb_lock_hashes(tdb, lockstart, locksize, ltype, waitflag);
tdb_trace_1rec(tdb, func, *key);
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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;
unsigned int group, gbits;
- enum TDB_ERROR ecode;
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", key);
- ecode = tdb_unlock_hashes(tdb, lockstart, locksize, F_WRLCK);
- if (ecode != TDB_SUCCESS) {
- tdb->ecode = ecode;
- return -1;
- }
- return 0;
+ 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);
}
projects / ccan / blobdiff