#include "private.h"
#include <ccan/tdb2/tdb2.h>
-#include <ccan/build_assert/build_assert.h>
-#include <ccan/likely/likely.h>
#include <assert.h>
+#include <stdarg.h>
/* The null return. */
struct tdb_data tdb_null = { .dptr = NULL, .dsize = 0 };
/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
static struct tdb_context *tdbs = NULL;
-PRINTF_ATTRIBUTE(4, 5) static void
-null_log_fn(struct tdb_context *tdb,
- enum tdb_debug_level level, void *priv,
- const char *fmt, ...)
-{
-}
-
static bool tdb_already_open(dev_t device, ino_t ino)
{
struct tdb_context *i;
fd = open("/dev/urandom", O_RDONLY);
if (fd >= 0) {
if (tdb_read_all(fd, &ret, sizeof(ret))) {
- tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
- "tdb_open: random from /dev/urandom\n");
+ tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
+ "tdb_open: random from /dev/urandom");
close(fd);
return ret;
}
char reply[1 + sizeof(uint64_t)];
int r = read(fd, reply, sizeof(reply));
if (r > 1) {
- tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
- "tdb_open: %u random bytes from"
- " /dev/egd-pool\n", r-1);
+ tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
+ "tdb_open: %u random bytes from"
+ " /dev/egd-pool", r-1);
/* Copy at least some bytes. */
memcpy(&ret, reply+1, r - 1);
if (reply[0] == sizeof(uint64_t)
/* Fallback: pid and time. */
gettimeofday(&now, NULL);
ret = getpid() * 100132289ULL + now.tv_sec * 1000000ULL + now.tv_usec;
- tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
- "tdb_open: random from getpid and time\n");
+ tdb_logerr(tdb, TDB_SUCCESS, TDB_DEBUG_TRACE,
+ "tdb_open: random from getpid and time");
return ret;
}
struct new_database {
struct tdb_header hdr;
- /* Initial free zone. */
- struct free_zone_header zhdr;
- tdb_off_t free[BUCKETS_FOR_ZONE(INITIAL_ZONE_BITS) + 1];
- struct tdb_free_record frec;
- /* Rest up to 1 << INITIAL_ZONE_BITS is empty. */
- char space[(1 << INITIAL_ZONE_BITS)
- - sizeof(struct free_zone_header)
- - sizeof(tdb_off_t) * (BUCKETS_FOR_ZONE(INITIAL_ZONE_BITS)+1)
- - sizeof(struct tdb_free_record)];
- uint8_t tailer;
- /* Don't count final padding! */
+ struct tdb_freelist flist;
};
/* initialise a new database */
-static int tdb_new_database(struct tdb_context *tdb, struct tdb_header *hdr)
+static int tdb_new_database(struct tdb_context *tdb,
+ struct tdb_attribute_seed *seed,
+ struct tdb_header *hdr)
{
/* We make it up in memory, then write it out if not internal */
struct new_database newdb;
- unsigned int bucket, magic_len, dbsize;
-
- /* Don't want any extra padding! */
- dbsize = offsetof(struct new_database, tailer) + sizeof(newdb.tailer);
+ unsigned int magic_len;
/* Fill in the header */
newdb.hdr.version = TDB_VERSION;
- newdb.hdr.hash_seed = random_number(tdb);
+ if (seed)
+ newdb.hdr.hash_seed = seed->seed;
+ else
+ newdb.hdr.hash_seed = random_number(tdb);
newdb.hdr.hash_test = TDB_HASH_MAGIC;
newdb.hdr.hash_test = tdb->khash(&newdb.hdr.hash_test,
sizeof(newdb.hdr.hash_test),
newdb.hdr.hash_seed,
tdb->hash_priv);
+ newdb.hdr.recovery = 0;
memset(newdb.hdr.reserved, 0, sizeof(newdb.hdr.reserved));
/* Initial hashes are empty. */
memset(newdb.hdr.hashtable, 0, sizeof(newdb.hdr.hashtable));
- /* Free is mostly empty... */
- newdb.zhdr.zone_bits = INITIAL_ZONE_BITS;
- memset(newdb.free, 0, sizeof(newdb.free));
-
- /* Create the single free entry. */
- newdb.frec.magic_and_meta = TDB_FREE_MAGIC | INITIAL_ZONE_BITS;
- newdb.frec.data_len = (sizeof(newdb.frec)
- - sizeof(struct tdb_used_record)
- + sizeof(newdb.space));
-
- /* Add it to the correct bucket. */
- bucket = size_to_bucket(INITIAL_ZONE_BITS, newdb.frec.data_len);
- newdb.free[bucket] = offsetof(struct new_database, frec);
- newdb.frec.next = newdb.frec.prev = 0;
-
- /* Clear free space to keep valgrind happy, and avoid leaking stack. */
- memset(newdb.space, 0, sizeof(newdb.space));
-
- /* Tailer contains maximum number of free_zone bits. */
- newdb.tailer = INITIAL_ZONE_BITS;
+ /* Free is empty. */
+ newdb.hdr.free_list = offsetof(struct new_database, flist);
+ memset(&newdb.flist, 0, sizeof(newdb.flist));
+ set_used_header(NULL, &newdb.flist.hdr, 0,
+ sizeof(newdb.flist) - sizeof(newdb.flist.hdr),
+ sizeof(newdb.flist) - sizeof(newdb.flist.hdr), 1);
/* Magic food */
memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
/* This creates an endian-converted database, as if read from disk */
magic_len = sizeof(newdb.hdr.magic_food);
tdb_convert(tdb,
- (char *)&newdb.hdr + magic_len,
- offsetof(struct new_database, space) - magic_len);
+ (char *)&newdb.hdr + magic_len, sizeof(newdb) - magic_len);
*hdr = newdb.hdr;
if (tdb->flags & TDB_INTERNAL) {
- tdb->map_size = dbsize;
+ tdb->map_size = sizeof(newdb);
tdb->map_ptr = malloc(tdb->map_size);
if (!tdb->map_ptr) {
- tdb->ecode = TDB_ERR_OOM;
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
+ "tdb_new_database: failed to allocate");
return -1;
}
memcpy(tdb->map_ptr, &newdb, tdb->map_size);
if (ftruncate(tdb->fd, 0) == -1)
return -1;
- if (!tdb_pwrite_all(tdb->fd, &newdb, dbsize, 0)) {
- tdb->ecode = TDB_ERR_IO;
+ if (!tdb_pwrite_all(tdb->fd, &newdb, sizeof(newdb), 0)) {
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_FATAL,
+ "tdb_new_database: failed to write: %s",
+ strerror(errno));
return -1;
}
return 0;
{
struct tdb_context *tdb;
struct stat st;
- int save_errno;
+ int saved_errno = 0;
uint64_t hash_test;
unsigned v;
struct tdb_header hdr;
+ struct tdb_attribute_seed *seed = NULL;
tdb = malloc(sizeof(*tdb));
if (!tdb) {
/* Can't log this */
errno = ENOMEM;
- goto fail;
+ return NULL;
}
tdb->name = NULL;
tdb->map_ptr = NULL;
tdb->map_size = sizeof(struct tdb_header);
tdb->ecode = TDB_SUCCESS;
tdb->flags = tdb_flags;
- tdb->log = null_log_fn;
- tdb->log_priv = NULL;
+ tdb->logfn = NULL;
tdb->transaction = NULL;
+ tdb->stats = NULL;
tdb_hash_init(tdb);
- /* last_zone will be set below. */
tdb_io_init(tdb);
tdb_lock_init(tdb);
while (attr) {
switch (attr->base.attr) {
case TDB_ATTRIBUTE_LOG:
- tdb->log = attr->log.log_fn;
- tdb->log_priv = attr->log.log_private;
+ tdb->logfn = attr->log.log_fn;
+ tdb->log_private = attr->log.log_private;
break;
case TDB_ATTRIBUTE_HASH:
tdb->khash = attr->hash.hash_fn;
tdb->hash_priv = attr->hash.hash_private;
break;
+ case TDB_ATTRIBUTE_SEED:
+ seed = &attr->seed;
+ break;
+ case TDB_ATTRIBUTE_STATS:
+ tdb->stats = &attr->stats;
+ /* They have stats we don't know about? Tell them. */
+ if (tdb->stats->size > sizeof(attr->stats))
+ tdb->stats->size = sizeof(attr->stats);
+ break;
default:
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: unknown attribute type %u\n",
- attr->base.attr);
- errno = EINVAL;
+ tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
+ "tdb_open: unknown attribute type %u",
+ attr->base.attr);
goto fail;
}
attr = attr->base.next;
}
if ((open_flags & O_ACCMODE) == O_WRONLY) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: can't open tdb %s write-only\n", name);
- errno = EINVAL;
+ tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_DEBUG_ERROR,
+ "tdb_open: can't open tdb %s write-only", name);
goto fail;
}
/* internal databases don't need any of the rest. */
if (tdb->flags & TDB_INTERNAL) {
tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
- if (tdb_new_database(tdb, &hdr) != 0) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: tdb_new_database failed!");
+ if (tdb_new_database(tdb, seed, &hdr) != 0) {
goto fail;
}
tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
tdb->hash_seed = hdr.hash_seed;
- tdb_zone_init(tdb);
+ tdb_flist_init(tdb);
return tdb;
}
if ((tdb->fd = open(name, open_flags, mode)) == -1) {
- tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
- "tdb_open: could not open file %s: %s\n",
- name, strerror(errno));
- goto fail; /* errno set by open(2) */
+ /* errno set by open(2) */
+ saved_errno = errno;
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: could not open file %s: %s",
+ name, strerror(errno));
+ goto fail;
}
/* on exec, don't inherit the fd */
fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
/* ensure there is only one process initialising at once */
- if (tdb_lock_open(tdb) == -1) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: failed to get open lock on %s: %s\n",
- name, strerror(errno));
- goto fail; /* errno set by tdb_brlock */
+ if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
+ /* errno set by tdb_brlock */
+ saved_errno = errno;
+ goto fail;
}
if (!tdb_pread_all(tdb->fd, &hdr, sizeof(hdr), 0)
|| strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
- if (!(open_flags & O_CREAT) || tdb_new_database(tdb, &hdr) == -1) {
- if (errno == 0) {
- errno = EIO; /* ie bad format or something */
- }
+ if (!(open_flags & O_CREAT)) {
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: %s is not a tdb file", name);
+ goto fail;
+ }
+ if (tdb_new_database(tdb, seed, &hdr) == -1) {
goto fail;
}
} else if (hdr.version != TDB_VERSION) {
tdb->flags |= TDB_CONVERT;
else {
/* wrong version */
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: %s is unknown version 0x%llx\n",
- name, (long long)hdr.version);
- errno = EIO;
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: %s is unknown version 0x%llx",
+ name, (long long)hdr.version);
goto fail;
}
}
hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
if (hdr.hash_test != hash_test) {
/* wrong hash variant */
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: %s uses a different hash function\n",
- name);
- errno = EIO;
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: %s uses a different hash function",
+ name);
goto fail;
}
- if (fstat(tdb->fd, &st) == -1)
+ if (fstat(tdb->fd, &st) == -1) {
+ saved_errno = errno;
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: could not stat open %s: %s",
+ name, strerror(errno));
goto fail;
+ }
/* Is it already in the open list? If so, fail. */
if (tdb_already_open(st.st_dev, st.st_ino)) {
/* FIXME */
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: %s (%d,%d) is already open in this process\n",
- name, (int)st.st_dev, (int)st.st_ino);
- errno = EBUSY;
+ tdb_logerr(tdb, TDB_ERR_NESTING, TDB_DEBUG_ERROR,
+ "tdb_open: %s (%d,%d) is already open in this"
+ " process",
+ name, (int)st.st_dev, (int)st.st_ino);
goto fail;
}
tdb->name = strdup(name);
if (!tdb->name) {
- errno = ENOMEM;
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
+ "tdb_open: failed to allocate name");
goto fail;
}
/* This make sure we have current map_size and mmap. */
tdb->methods->oob(tdb, tdb->map_size + 1, true);
- /* Now we can pick a random free zone to start from. */
- if (tdb_zone_init(tdb) == -1)
+ /* Now it's fully formed, recover if necessary. */
+ if (tdb_needs_recovery(tdb) && tdb_lock_and_recover(tdb) == -1) {
+ goto fail;
+ }
+
+ if (tdb_flist_init(tdb) == -1)
goto fail;
tdb->next = tdbs;
return tdb;
fail:
- save_errno = errno;
-
- if (!tdb)
- return NULL;
+ /* Map ecode to some logical errno. */
+ if (!saved_errno) {
+ switch (tdb->ecode) {
+ case TDB_ERR_CORRUPT:
+ case TDB_ERR_IO:
+ saved_errno = EIO;
+ break;
+ case TDB_ERR_LOCK:
+ saved_errno = EWOULDBLOCK;
+ break;
+ case TDB_ERR_OOM:
+ saved_errno = ENOMEM;
+ break;
+ case TDB_ERR_EINVAL:
+ saved_errno = EINVAL;
+ break;
+ case TDB_ERR_NESTING:
+ saved_errno = EBUSY;
+ break;
+ default:
+ saved_errno = EINVAL;
+ break;
+ }
+ }
#ifdef TDB_TRACE
close(tdb->tracefd);
free((char *)tdb->name);
if (tdb->fd != -1)
if (close(tdb->fd) != 0)
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_open: failed to close tdb->fd"
- " on error!\n");
+ tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
+ "tdb_open: failed to close tdb->fd"
+ " on error!");
free(tdb);
- errno = save_errno;
+ errno = saved_errno;
return NULL;
}
{
uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
- if (set_header(tdb, rec, keylen, datalen, keylen + dataroom, h,
- rec_zone_bits(rec)))
+ if (set_used_header(tdb, rec, keylen, datalen, keylen + dataroom, h))
return -1;
return tdb_write_convert(tdb, off, rec, sizeof(*rec));
struct hash_info *h,
struct tdb_data key, struct tdb_data dbuf,
tdb_off_t old_off, tdb_len_t old_room,
- unsigned old_zone,
bool growing)
{
tdb_off_t new_off;
/* We didn't like the existing one: remove it. */
if (old_off) {
- add_free_record(tdb, old_zone, old_off,
+ add_stat(tdb, frees, 1);
+ add_free_record(tdb, old_off,
sizeof(struct tdb_used_record)
+ key.dsize + old_room);
if (replace_in_hash(tdb, h, new_off) == -1)
}
/* If we didn't use the old record, this implies we're growing. */
- ret = replace_data(tdb, &h, key, dbuf, off, old_room,
- rec_zone_bits(&rec), off != 0);
+ ret = replace_data(tdb, &h, key, dbuf, off, old_room, off != 0);
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
return ret;
/* Slow path. */
newdata = malloc(key.dsize + old_dlen + dbuf.dsize);
if (!newdata) {
- tdb->ecode = TDB_ERR_OOM;
- tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
- "tdb_append: cannot allocate %llu bytes!\n",
- (long long)key.dsize + old_dlen + dbuf.dsize);
+ tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_FATAL,
+ "tdb_append: failed to allocate %zu bytes",
+ (size_t)(key.dsize+old_dlen+dbuf.dsize));
goto fail;
}
if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
}
/* If they're using tdb_append(), it implies they're growing record. */
- ret = replace_data(tdb, &h, key, new_dbuf, off,
- old_room, rec_zone_bits(&rec), true);
+ ret = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
free(newdata);
goto unlock_err;
/* Free the deleted entry. */
- if (add_free_record(tdb, rec_zone_bits(&rec), off,
+ add_stat(tdb, frees, 1);
+ if (add_free_record(tdb, off,
sizeof(struct tdb_used_record)
+ rec_key_length(&rec)
+ rec_data_length(&rec)
struct tdb_context **i;
int ret = 0;
- /* FIXME:
+ tdb_trace(tdb, "tdb_close");
+
if (tdb->transaction) {
tdb_transaction_cancel(tdb);
}
- */
- tdb_trace(tdb, "tdb_close");
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
return ret;
}
-enum TDB_ERROR tdb_error(struct tdb_context *tdb)
+enum TDB_ERROR tdb_error(const struct tdb_context *tdb)
{
return tdb->ecode;
}
-const char *tdb_errorstr(struct tdb_context *tdb)
+const char *tdb_errorstr(const struct tdb_context *tdb)
{
/* Gcc warns if you miss a case in the switch, so use that. */
switch (tdb->ecode) {
}
return "Invalid error code";
}
+
+void COLD tdb_logerr(struct tdb_context *tdb,
+ enum TDB_ERROR ecode,
+ enum tdb_debug_level level,
+ const char *fmt, ...)
+{
+ char *message;
+ va_list ap;
+ size_t len;
+ /* tdb_open paths care about errno, so save it. */
+ int saved_errno = errno;
+
+ tdb->ecode = ecode;
+
+ if (!tdb->logfn)
+ return;
+
+ /* FIXME: Doesn't assume asprintf. */
+ va_start(ap, fmt);
+ len = vsnprintf(NULL, 0, fmt, ap);
+ va_end(ap);
+
+ message = malloc(len + 1);
+ if (!message) {
+ tdb->logfn(tdb, level, tdb->log_private,
+ "out of memory formatting message");
+ return;
+ }
+ va_start(ap, fmt);
+ len = vsprintf(message, fmt, ap);
+ va_end(ap);
+ tdb->logfn(tdb, level, tdb->log_private, message);
+ free(message);
+ errno = saved_errno;
+}