X-Git-Url: https://git.ozlabs.org/?p=ccan;a=blobdiff_plain;f=ccan%2Ftdb2%2Ftdb.c;h=f6ca01657518af29c1111700c530abda29ba7d43;hp=4b02755c28887667c1ca2adb157fb7520e8e4835;hb=a39bf3aca1b60365efaf8b1eeb2b2c58b09ffab6;hpb=e2cb71e5d46801a94a3f0f546b8dcc2b7a79b66c

diff --git a/ccan/tdb2/tdb.c b/ccan/tdb2/tdb.c
index 4b02755c..f6ca0165 100644
--- a/ccan/tdb2/tdb.c
+++ b/ccan/tdb2/tdb.c
@@ -1,8 +1,7 @@
 #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 };
@@ -10,13 +9,6 @@ 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;
@@ -30,6 +22,24 @@ static bool tdb_already_open(dev_t device, ino_t ino)
 	return false;
 }
 
+static bool read_all(int fd, void *buf, size_t len)
+{
+	while (len) {
+		ssize_t ret;
+		ret = read(fd, buf, len);
+		if (ret < 0)
+			return false;
+		if (ret == 0) {
+			/* ETOOSHORT? */
+			errno = EWOULDBLOCK;
+			return false;
+		}
+		buf = (char *)buf + ret;
+		len -= ret;
+	}
+	return true;
+}
+
 static uint64_t random_number(struct tdb_context *tdb)
 {
 	int fd;
@@ -38,9 +48,7 @@ static uint64_t random_number(struct tdb_context *tdb)
 
 	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");
+		if (read_all(fd, &ret, sizeof(ret))) {
 			close(fd);
 			return ret;
 		}
@@ -55,9 +63,6 @@ static uint64_t random_number(struct tdb_context *tdb)
 			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);
 				/* Copy at least some bytes. */
 				memcpy(&ret, reply+1, r - 1);
 				if (reply[0] == sizeof(uint64_t)
@@ -73,37 +78,26 @@ static uint64_t random_number(struct tdb_context *tdb)
 	/* 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_LOG_WARNING,
+		   "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_freetable ftable;
 };
 
 /* initialise a new database */
-static int tdb_new_database(struct tdb_context *tdb,
-			    struct tdb_attribute_seed *seed,
-			    struct tdb_header *hdr)
+static enum TDB_ERROR 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;
+	ssize_t rlen;
+	enum TDB_ERROR ecode;
 
 	/* Fill in the header */
 	newdb.hdr.version = TDB_VERSION;
@@ -116,30 +110,21 @@ static int tdb_new_database(struct tdb_context *tdb,
 					 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_table = offsetof(struct new_database, ftable);
+	memset(&newdb.ftable, 0, sizeof(newdb.ftable));
+	ecode = set_header(NULL, &newdb.ftable.hdr, TDB_FTABLE_MAGIC, 0,
+			   sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
+			   sizeof(newdb.ftable) - sizeof(newdb.ftable.hdr),
+			   0);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
+	}
 
 	/* Magic food */
 	memset(newdb.hdr.magic_food, 0, sizeof(newdb.hdr.magic_food));
@@ -148,32 +133,42 @@ static int tdb_new_database(struct tdb_context *tdb,
 	/* 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;
-			return -1;
+			return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+					  "tdb_new_database:"
+					  " failed to allocate");
 		}
 		memcpy(tdb->map_ptr, &newdb, tdb->map_size);
-		return 0;
+		return TDB_SUCCESS;
+	}
+	if (lseek(tdb->fd, 0, SEEK_SET) == -1) {
+		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				  "tdb_new_database:"
+				  " failed to seek: %s", strerror(errno));
 	}
-	if (lseek(tdb->fd, 0, SEEK_SET) == -1)
-		return -1;
 
-	if (ftruncate(tdb->fd, 0) == -1)
-		return -1;
+	if (ftruncate(tdb->fd, 0) == -1) {
+		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				  "tdb_new_database:"
+				  " failed to truncate: %s", strerror(errno));
+	}
 
-	if (!tdb_pwrite_all(tdb->fd, &newdb, dbsize, 0)) {
-		tdb->ecode = TDB_ERR_IO;
-		return -1;
+	rlen = write(tdb->fd, &newdb, sizeof(newdb));
+	if (rlen != sizeof(newdb)) {
+		if (rlen >= 0)
+			errno = ENOSPC;
+		return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				  "tdb_new_database: %zi writing header: %s",
+				  rlen, strerror(errno));
 	}
-	return 0;
+	return TDB_SUCCESS;
 }
 
 struct tdb_context *tdb_open(const char *name, int tdb_flags,
@@ -182,38 +177,40 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 {
 	struct tdb_context *tdb;
 	struct stat st;
-	int save_errno;
+	int saved_errno = 0;
 	uint64_t hash_test;
 	unsigned v;
+	ssize_t rlen;
 	struct tdb_header hdr;
 	struct tdb_attribute_seed *seed = NULL;
+	tdb_bool_err berr;
+	enum TDB_ERROR ecode;
 
 	tdb = malloc(sizeof(*tdb));
 	if (!tdb) {
 		/* Can't log this */
 		errno = ENOMEM;
-		goto fail;
+		return NULL;
 	}
 	tdb->name = NULL;
 	tdb->map_ptr = NULL;
 	tdb->direct_access = 0;
 	tdb->fd = -1;
 	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->access = 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;
@@ -222,27 +219,32 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		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;
+			ecode = tdb_logerr(tdb, TDB_ERR_EINVAL,
+					   TDB_LOG_USE_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;
+		ecode = tdb_logerr(tdb, TDB_ERR_EINVAL, TDB_LOG_USE_ERROR,
+				   "tdb_open: can't open tdb %s write-only",
+				   name);
 		goto fail;
 	}
 
 	if ((open_flags & O_ACCMODE) == O_RDONLY) {
 		tdb->read_only = true;
-		/* read only databases don't do locking */
-		tdb->flags |= TDB_NOLOCK;
 		tdb->mmap_flags = PROT_READ;
 	} else {
 		tdb->read_only = false;
@@ -252,22 +254,23 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 	/* 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, seed, &hdr) != 0) {
-			tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
-				 "tdb_open: tdb_new_database failed!");
+		ecode = tdb_new_database(tdb, seed, &hdr);
+		if (ecode != TDB_SUCCESS) {
 			goto fail;
 		}
 		tdb_convert(tdb, &hdr.hash_seed, sizeof(hdr.hash_seed));
 		tdb->hash_seed = hdr.hash_seed;
-		tdb_zone_init(tdb);
+		tdb_ftable_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;
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				   "tdb_open: could not open file %s: %s",
+				   name, strerror(errno));
+		goto fail;
 	}
 
 	/* on exec, don't inherit the fd */
@@ -275,31 +278,39 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
         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_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, seed, &hdr) == -1) {
-			if (errno == 0) {
-				errno = EIO; /* ie bad format or something */
-			}
+	ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
+	if (ecode != TDB_SUCCESS) {
+		goto fail;
+	}
+
+	/* If they used O_TRUNC, read will return 0. */
+	rlen = read(tdb->fd, &hdr, sizeof(hdr));
+	if (rlen == 0 && (open_flags & O_CREAT)) {
+		ecode = tdb_new_database(tdb, seed, &hdr);
+		if (ecode != TDB_SUCCESS) {
 			goto fail;
 		}
-	} else if (hdr.version != TDB_VERSION) {
+	} else if (rlen < 0) {
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				   "tdb_open: error %s reading %s",
+				   strerror(errno), name);
+		goto fail;
+	} else if (rlen < sizeof(hdr)
+		   || strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+				   "tdb_open: %s is not a tdb file", name);
+		goto fail;
+	}
+
+	if (hdr.version != TDB_VERSION) {
 		if (hdr.version == bswap_64(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;
+			ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
+					   "tdb_open:"
+					   " %s is unknown version 0x%llx",
+					   name, (long long)hdr.version);
 			goto fail;
 		}
 	}
@@ -310,29 +321,35 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 	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;
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_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;
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_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;
+		ecode = tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_USE_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;
+		ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+				   "tdb_open: failed to allocate name");
 		goto fail;
 	}
 
@@ -343,19 +360,50 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 	/* 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. */
+	berr = tdb_needs_recovery(tdb);
+	if (unlikely(berr != false)) {
+		if (berr < 0) {
+			ecode = berr;
+			goto fail;
+		}
+		ecode = tdb_lock_and_recover(tdb);
+		if (ecode != TDB_SUCCESS) {
+			goto fail;
+		}
+	}
+
+	ecode = tdb_ftable_init(tdb);
+	if (ecode != TDB_SUCCESS) {
 		goto fail;
+	}
 
 	tdb->next = tdbs;
 	tdbs = tdb;
 	return tdb;
 
  fail:
-	save_errno = errno;
-
-	if (!tdb)
-		return NULL;
+	/* Map ecode to some logical errno. */
+	if (!saved_errno) {
+		switch (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;
+		default:
+			saved_errno = EINVAL;
+			break;
+		}
+	}
 
 #ifdef TDB_TRACE
 	close(tdb->tracefd);
@@ -366,91 +414,102 @@ struct tdb_context *tdb_open(const char *name, int tdb_flags,
 		} else
 			tdb_munmap(tdb);
 	}
+	free(tdb->lockrecs);
 	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_LOG_ERROR,
+				   "tdb_open: failed to close tdb->fd"
+				   " on error: %s", strerror(errno));
 	free(tdb);
-	errno = save_errno;
+	errno = saved_errno;
 	return NULL;
 }
 
-/* FIXME: modify, don't rewrite! */
-static int update_rec_hdr(struct tdb_context *tdb,
-			  tdb_off_t off,
-			  tdb_len_t keylen,
-			  tdb_len_t datalen,
-			  struct tdb_used_record *rec,
-			  uint64_t h)
+static enum TDB_ERROR update_rec_hdr(struct tdb_context *tdb,
+				     tdb_off_t off,
+				     tdb_len_t keylen,
+				     tdb_len_t datalen,
+				     struct tdb_used_record *rec,
+				     uint64_t h)
 {
 	uint64_t dataroom = rec_data_length(rec) + rec_extra_padding(rec);
+	enum TDB_ERROR ecode;
 
-	if (set_header(tdb, rec, keylen, datalen, keylen + dataroom, h,
-		       rec_zone_bits(rec)))
-		return -1;
-
-	return tdb_write_convert(tdb, off, rec, sizeof(*rec));
+	ecode = set_header(tdb, rec, TDB_USED_MAGIC, keylen, datalen,
+			   keylen + dataroom, h);
+	if (ecode == TDB_SUCCESS) {
+		ecode = tdb_write_convert(tdb, off, rec, sizeof(*rec));
+	}
+	return ecode;
 }
 
-/* Returns -1 on error, 0 on OK */
-static int replace_data(struct tdb_context *tdb,
-			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)
+static enum TDB_ERROR replace_data(struct tdb_context *tdb,
+				   struct hash_info *h,
+				   struct tdb_data key, struct tdb_data dbuf,
+				   tdb_off_t old_off, tdb_len_t old_room,
+				   bool growing)
 {
 	tdb_off_t new_off;
+	enum TDB_ERROR ecode;
 
 	/* Allocate a new record. */
-	new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, growing);
-	if (unlikely(new_off == TDB_OFF_ERR))
-		return -1;
+	new_off = alloc(tdb, key.dsize, dbuf.dsize, h->h, TDB_USED_MAGIC,
+			growing);
+	if (TDB_OFF_IS_ERR(new_off)) {
+		return new_off;
+	}
 
 	/* We didn't like the existing one: remove it. */
 	if (old_off) {
-		add_free_record(tdb, old_zone, old_off,
-				sizeof(struct tdb_used_record)
-				+ key.dsize + old_room);
-		if (replace_in_hash(tdb, h, new_off) == -1)
-			return -1;
+		add_stat(tdb, frees, 1);
+		ecode = add_free_record(tdb, old_off,
+					sizeof(struct tdb_used_record)
+					+ key.dsize + old_room);
+		if (ecode == TDB_SUCCESS)
+			ecode = replace_in_hash(tdb, h, new_off);
 	} else {
-		if (add_to_hash(tdb, h, new_off) == -1)
-			return -1;
+		ecode = add_to_hash(tdb, h, new_off);
+	}
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
 	}
 
 	new_off += sizeof(struct tdb_used_record);
-	if (tdb->methods->write(tdb, new_off, key.dptr, key.dsize) == -1)
-		return -1;
+	ecode = tdb->methods->twrite(tdb, new_off, key.dptr, key.dsize);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
+	}
 
 	new_off += key.dsize;
-	if (tdb->methods->write(tdb, new_off, dbuf.dptr, dbuf.dsize) == -1)
-		return -1;
+	ecode = tdb->methods->twrite(tdb, new_off, dbuf.dptr, dbuf.dsize);
+	if (ecode != TDB_SUCCESS) {
+		return ecode;
+	}
 
 	/* FIXME: tdb_increment_seqnum(tdb); */
-	return 0;
+	return TDB_SUCCESS;
 }
 
-int tdb_store(struct tdb_context *tdb,
-	      struct tdb_data key, struct tdb_data dbuf, int flag)
+enum TDB_ERROR tdb_store(struct tdb_context *tdb,
+			 struct tdb_data key, struct tdb_data dbuf, int flag)
 {
 	struct hash_info h;
 	tdb_off_t off;
 	tdb_len_t old_room = 0;
 	struct tdb_used_record rec;
-	int ret;
+	enum TDB_ERROR ecode;
 
 	off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
-	if (unlikely(off == TDB_OFF_ERR))
-		return -1;
+	if (TDB_OFF_IS_ERR(off)) {
+		return off;
+	}
 
 	/* Now we have lock on this hash bucket. */
 	if (flag == TDB_INSERT) {
 		if (off) {
-			tdb->ecode = TDB_ERR_EXISTS;
-			goto fail;
+			ecode = TDB_ERR_EXISTS;
+			goto out;
 		}
 	} else {
 		if (off) {
@@ -458,43 +517,44 @@ int tdb_store(struct tdb_context *tdb,
 				+ rec_extra_padding(&rec);
 			if (old_room >= dbuf.dsize) {
 				/* Can modify in-place.  Easy! */
-				if (update_rec_hdr(tdb, off,
-						   key.dsize, dbuf.dsize,
-						   &rec, h.h))
-					goto fail;
-				if (tdb->methods->write(tdb, off + sizeof(rec)
-							+ key.dsize,
-							dbuf.dptr, dbuf.dsize))
-					goto fail;
+				ecode = update_rec_hdr(tdb, off,
+						       key.dsize, dbuf.dsize,
+						       &rec, h.h);
+				if (ecode != TDB_SUCCESS) {
+					goto out;
+				}
+				ecode = tdb->methods->twrite(tdb,
+							     off + sizeof(rec)
+							     + key.dsize,
+							     dbuf.dptr,
+							     dbuf.dsize);
+				if (ecode != TDB_SUCCESS) {
+					goto out;
+				}
 				tdb_unlock_hashes(tdb, h.hlock_start,
 						  h.hlock_range, F_WRLCK);
-				return 0;
+				return TDB_SUCCESS;
 			}
-			/* FIXME: See if right record is free? */
 		} else {
 			if (flag == TDB_MODIFY) {
 				/* if the record doesn't exist and we
 				   are in TDB_MODIFY mode then we should fail
 				   the store */
-				tdb->ecode = TDB_ERR_NOEXIST;
-				goto fail;
+				ecode = TDB_ERR_NOEXIST;
+				goto out;
 			}
 		}
 	}
 
 	/* 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);
-	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	return ret;
-
-fail:
+	ecode = replace_data(tdb, &h, key, dbuf, off, old_room, off);
+out:
 	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	return -1;
+	return ecode;
 }
 
-int tdb_append(struct tdb_context *tdb,
-	       struct tdb_data key, struct tdb_data dbuf)
+enum TDB_ERROR tdb_append(struct tdb_context *tdb,
+			  struct tdb_data key, struct tdb_data dbuf)
 {
 	struct hash_info h;
 	tdb_off_t off;
@@ -502,11 +562,12 @@ int tdb_append(struct tdb_context *tdb,
 	tdb_len_t old_room = 0, old_dlen;
 	unsigned char *newdata;
 	struct tdb_data new_dbuf;
-	int ret;
+	enum TDB_ERROR ecode;
 
 	off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
-	if (unlikely(off == TDB_OFF_ERR))
-		return -1;
+	if (TDB_OFF_IS_ERR(off)) {
+		return off;
+	}
 
 	if (off) {
 		old_dlen = rec_data_length(&rec);
@@ -514,35 +575,33 @@ int tdb_append(struct tdb_context *tdb,
 
 		/* Fast path: can append in place. */
 		if (rec_extra_padding(&rec) >= dbuf.dsize) {
-			if (update_rec_hdr(tdb, off, key.dsize,
-					   old_dlen + dbuf.dsize, &rec, h.h))
-				goto fail;
+			ecode = update_rec_hdr(tdb, off, key.dsize,
+					       old_dlen + dbuf.dsize, &rec,
+					       h.h);
+			if (ecode != TDB_SUCCESS) {
+				goto out;
+			}
 
 			off += sizeof(rec) + key.dsize + old_dlen;
-			if (tdb->methods->write(tdb, off, dbuf.dptr,
-						dbuf.dsize) == -1)
-				goto fail;
-
-			/* FIXME: tdb_increment_seqnum(tdb); */
-			tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range,
-					  F_WRLCK);
-			return 0;
+			ecode = tdb->methods->twrite(tdb, off, dbuf.dptr,
+						     dbuf.dsize);
+			goto out;
 		}
-		/* FIXME: Check right record free? */
 
 		/* 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);
-			goto fail;
+			ecode = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+					   "tdb_append:"
+					   " failed to allocate %zu bytes",
+					   (size_t)(key.dsize + old_dlen
+						    + dbuf.dsize));
+			goto out;
 		}
-		if (tdb->methods->read(tdb, off + sizeof(rec) + key.dsize,
-				       newdata, old_dlen) != 0) {
-			free(newdata);
-			goto fail;
+		ecode = tdb->methods->tread(tdb, off + sizeof(rec) + key.dsize,
+					    newdata, old_dlen);
+		if (ecode != TDB_SUCCESS) {
+			goto out_free_newdata;
 		}
 		memcpy(newdata + old_dlen, dbuf.dptr, dbuf.dsize);
 		new_dbuf.dptr = newdata;
@@ -553,75 +612,77 @@ int tdb_append(struct tdb_context *tdb,
 	}
 
 	/* 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);
-	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	free(newdata);
-
-	return ret;
+	ecode = replace_data(tdb, &h, key, new_dbuf, off, old_room, true);
 
-fail:
+out_free_newdata:
+	free(newdata);
+out:
 	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	return -1;
+	return ecode;
 }
 
-struct tdb_data tdb_fetch(struct tdb_context *tdb, struct tdb_data key)
+enum TDB_ERROR tdb_fetch(struct tdb_context *tdb, struct tdb_data key,
+			 struct tdb_data *data)
 {
 	tdb_off_t off;
 	struct tdb_used_record rec;
 	struct hash_info h;
-	struct tdb_data ret;
+	enum TDB_ERROR ecode;
 
 	off = find_and_lock(tdb, key, F_RDLCK, &h, &rec, NULL);
-	if (unlikely(off == TDB_OFF_ERR))
-		return tdb_null;
+	if (TDB_OFF_IS_ERR(off)) {
+		return off;
+	}
 
 	if (!off) {
-		tdb->ecode = TDB_ERR_NOEXIST;
-		ret = tdb_null;
+		ecode = TDB_ERR_NOEXIST;
 	} else {
-		ret.dsize = rec_data_length(&rec);
-		ret.dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
-					  ret.dsize);
+		data->dsize = rec_data_length(&rec);
+		data->dptr = tdb_alloc_read(tdb, off + sizeof(rec) + key.dsize,
+					    data->dsize);
+		if (TDB_PTR_IS_ERR(data->dptr)) {
+			ecode = TDB_PTR_ERR(data->dptr);
+		} else
+			ecode = TDB_SUCCESS;
 	}
 
 	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_RDLCK);
-	return ret;
+	return ecode;
 }
 
-int tdb_delete(struct tdb_context *tdb, struct tdb_data key)
+enum TDB_ERROR tdb_delete(struct tdb_context *tdb, struct tdb_data key)
 {
 	tdb_off_t off;
 	struct tdb_used_record rec;
 	struct hash_info h;
+	enum TDB_ERROR ecode;
 
 	off = find_and_lock(tdb, key, F_WRLCK, &h, &rec, NULL);
-	if (unlikely(off == TDB_OFF_ERR))
-		return -1;
+	if (TDB_OFF_IS_ERR(off)) {
+		return off;
+	}
 
 	if (!off) {
-		tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-		tdb->ecode = TDB_ERR_NOEXIST;
-		return -1;
+		ecode = TDB_ERR_NOEXIST;
+		goto unlock;
 	}
 
-	if (delete_from_hash(tdb, &h) == -1)
-		goto unlock_err;
+	ecode = delete_from_hash(tdb, &h);
+	if (ecode != TDB_SUCCESS) {
+		goto unlock;
+	}
 
 	/* Free the deleted entry. */
-	if (add_free_record(tdb, rec_zone_bits(&rec), off,
-			    sizeof(struct tdb_used_record)
-			    + rec_key_length(&rec)
-			    + rec_data_length(&rec)
-			    + rec_extra_padding(&rec)) != 0)
-		goto unlock_err;
-
-	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	return 0;
+	add_stat(tdb, frees, 1);
+	ecode = add_free_record(tdb, off,
+				sizeof(struct tdb_used_record)
+				+ rec_key_length(&rec)
+				+ rec_data_length(&rec)
+				+ rec_extra_padding(&rec));
 
-unlock_err:
+unlock:
 	tdb_unlock_hashes(tdb, h.hlock_start, h.hlock_range, F_WRLCK);
-	return -1;
+	return ecode;
 }
 
 int tdb_close(struct tdb_context *tdb)
@@ -629,12 +690,11 @@ int tdb_close(struct tdb_context *tdb)
 	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)
@@ -665,25 +725,54 @@ int tdb_close(struct tdb_context *tdb)
 	return ret;
 }
 
-enum TDB_ERROR tdb_error(struct tdb_context *tdb)
-{
-	return tdb->ecode;
-}
-
-const char *tdb_errorstr(struct tdb_context *tdb)
+const char *tdb_errorstr(enum TDB_ERROR ecode)
 {
 	/* Gcc warns if you miss a case in the switch, so use that. */
-	switch (tdb->ecode) {
+	switch (ecode) {
 	case TDB_SUCCESS: return "Success";
 	case TDB_ERR_CORRUPT: return "Corrupt database";
 	case TDB_ERR_IO: return "IO Error";
 	case TDB_ERR_LOCK: return "Locking error";
 	case TDB_ERR_OOM: return "Out of memory";
 	case TDB_ERR_EXISTS: return "Record exists";
-	case TDB_ERR_NESTING: return "Transaction already started";
 	case TDB_ERR_EINVAL: return "Invalid parameter";
 	case TDB_ERR_NOEXIST: return "Record does not exist";
 	case TDB_ERR_RDONLY: return "write not permitted";
 	}
 	return "Invalid error code";
 }
+
+enum TDB_ERROR COLD tdb_logerr(struct tdb_context *tdb,
+			       enum TDB_ERROR ecode,
+			       enum tdb_log_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;
+
+	if (!tdb->logfn)
+		return ecode;
+
+	/* 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, TDB_LOG_ERROR, tdb->log_private,
+			   "out of memory formatting message:");
+		tdb->logfn(tdb, level, tdb->log_private, fmt);
+		return ecode;
+	}
+	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;
+	return ecode;
+}