]> git.ozlabs.org Git - ccan/blobdiff - ccan/tdb2/tdb1_lock.c
tdb2: make tests work in parallel.
[ccan] / ccan / tdb2 / tdb1_lock.c
index bed70ba2ec9031fd5721c115029e97a03c0afba0..5cc0ad656729c51877aebf5f1fe7489bb11b972a 100644 (file)
 
 #include "tdb1_private.h"
 
-_PUBLIC_ void tdb1_setalarm_sigptr(struct tdb1_context *tdb, volatile sig_atomic_t *ptr)
-{
-       tdb->interrupt_sig_ptr = ptr;
-}
-
-static int fcntl_lock(struct tdb1_context *tdb,
-                     int rw, off_t off, off_t len, bool waitflag)
-{
-       struct flock fl;
-
-       fl.l_type = rw;
-       fl.l_whence = SEEK_SET;
-       fl.l_start = off;
-       fl.l_len = len;
-       fl.l_pid = 0;
-
-       if (waitflag)
-               return fcntl(tdb->fd, F_SETLKW, &fl);
-       else
-               return fcntl(tdb->fd, F_SETLK, &fl);
-}
-
-static int fcntl_unlock(struct tdb1_context *tdb, int rw, off_t off, off_t len)
-{
-       struct flock fl;
-#if 0 /* Check they matched up locks and unlocks correctly. */
-       char line[80];
-       FILE *locks;
-       bool found = false;
-
-       locks = fopen("/proc/locks", "r");
-
-       while (fgets(line, 80, locks)) {
-               char *p;
-               int type, start, l;
-
-               /* eg. 1: FLOCK  ADVISORY  WRITE 2440 08:01:2180826 0 EOF */
-               p = strchr(line, ':') + 1;
-               if (strncmp(p, " POSIX  ADVISORY  ", strlen(" POSIX  ADVISORY  ")))
-                       continue;
-               p += strlen(" FLOCK  ADVISORY  ");
-               if (strncmp(p, "READ  ", strlen("READ  ")) == 0)
-                       type = F_RDLCK;
-               else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
-                       type = F_WRLCK;
-               else
-                       abort();
-               p += 6;
-               if (atoi(p) != getpid())
-                       continue;
-               p = strchr(strchr(p, ' ') + 1, ' ') + 1;
-               start = atoi(p);
-               p = strchr(p, ' ') + 1;
-               if (strncmp(p, "EOF", 3) == 0)
-                       l = 0;
-               else
-                       l = atoi(p) - start + 1;
-
-               if (off == start) {
-                       if (len != l) {
-                               fprintf(stderr, "Len %u should be %u: %s",
-                                       (int)len, l, line);
-                               abort();
-                       }
-                       if (type != rw) {
-                               fprintf(stderr, "Type %s wrong: %s",
-                                       rw == F_RDLCK ? "READ" : "WRITE", line);
-                               abort();
-                       }
-                       found = true;
-                       break;
-               }
-       }
-
-       if (!found) {
-               fprintf(stderr, "Unlock on %u@%u not found!\n",
-                       (int)off, (int)len);
-               abort();
-       }
-
-       fclose(locks);
-#endif
-
-       fl.l_type = F_UNLCK;
-       fl.l_whence = SEEK_SET;
-       fl.l_start = off;
-       fl.l_len = len;
-       fl.l_pid = 0;
-
-       return fcntl(tdb->fd, F_SETLKW, &fl);
-}
-
 /* list -1 is the alloc list, otherwise a hash chain. */
 static tdb1_off_t lock_offset(int list)
 {
@@ -133,184 +41,84 @@ static tdb1_off_t lock_offset(int list)
 
    note that a len of zero means lock to end of file
 */
-int tdb1_brlock(struct tdb1_context *tdb,
+int tdb1_brlock(struct tdb_context *tdb,
               int rw_type, tdb1_off_t offset, size_t len,
-              enum tdb1_lock_flags flags)
+              enum tdb_lock_flags flags)
 {
-       int ret;
-
-       if (tdb->flags & TDB1_NOLOCK) {
-               return 0;
-       }
-
-       if (flags & TDB1_LOCK_MARK_ONLY) {
+       enum TDB_ERROR ecode = tdb_brlock(tdb, rw_type, offset, len, flags
+                                         | TDB_LOCK_NOCHECK);
+       if (ecode == TDB_SUCCESS)
                return 0;
-       }
-
-       if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
-               tdb->ecode = TDB1_ERR_RDONLY;
-               return -1;
-       }
-
-       do {
-               ret = fcntl_lock(tdb, rw_type, offset, len,
-                                flags & TDB1_LOCK_WAIT);
-               /* Check for a sigalarm break. */
-               if (ret == -1 && errno == EINTR &&
-                               tdb->interrupt_sig_ptr &&
-                               *tdb->interrupt_sig_ptr) {
-                       break;
-               }
-       } while (ret == -1 && errno == EINTR);
-
-       if (ret == -1) {
-               tdb->ecode = TDB1_ERR_LOCK;
-               /* Generic lock error. errno set by fcntl.
-                * EAGAIN is an expected return from non-blocking
-                * locks. */
-               if (!(flags & TDB1_LOCK_PROBE) && errno != EAGAIN) {
-                       TDB1_LOG((tdb, TDB1_DEBUG_TRACE,"tdb1_brlock failed (fd=%d) at offset %d rw_type=%d flags=%d len=%d\n",
-                                tdb->fd, offset, rw_type, flags, (int)len));
-               }
-               return -1;
-       }
-       return 0;
+       tdb->last_error = ecode;
+       return -1;
 }
 
-int tdb1_brunlock(struct tdb1_context *tdb,
+int tdb1_brunlock(struct tdb_context *tdb,
                 int rw_type, tdb1_off_t offset, size_t len)
 {
-       int ret;
-
-       if (tdb->flags & TDB1_NOLOCK) {
+       enum TDB_ERROR ecode = tdb_brunlock(tdb, rw_type, offset, len);
+       if (ecode == TDB_SUCCESS)
                return 0;
-       }
-
-       do {
-               ret = fcntl_unlock(tdb, rw_type, offset, len);
-       } while (ret == -1 && errno == EINTR);
-
-       if (ret == -1) {
-               TDB1_LOG((tdb, TDB1_DEBUG_TRACE,"tdb1_brunlock failed (fd=%d) at offset %d rw_type=%d len=%d\n",
-                        tdb->fd, offset, rw_type, (int)len));
-       }
-       return ret;
+       tdb->last_error = ecode;
+       return -1;
 }
 
-/*
-  upgrade a read lock to a write lock. This needs to be handled in a
-  special way as some OSes (such as solaris) have too conservative
-  deadlock detection and claim a deadlock when progress can be
-  made. For those OSes we may loop for a while.
-*/
-int tdb1_allrecord_upgrade(struct tdb1_context *tdb)
+int tdb1_allrecord_upgrade(struct tdb_context *tdb)
 {
-       int count = 1000;
-
-       if (tdb->allrecord_lock.count != 1) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR,
-                        "tdb1_allrecord_upgrade failed: count %u too high\n",
-                        tdb->allrecord_lock.count));
-               return -1;
-       }
-
-       if (tdb->allrecord_lock.off != 1) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR,
-                        "tdb1_allrecord_upgrade failed: already upgraded?\n"));
-               return -1;
-       }
-
-       while (count--) {
-               struct timeval tv;
-               if (tdb1_brlock(tdb, F_WRLCK, TDB1_FREELIST_TOP, 0,
-                              TDB1_LOCK_WAIT|TDB1_LOCK_PROBE) == 0) {
-                       tdb->allrecord_lock.ltype = F_WRLCK;
-                       tdb->allrecord_lock.off = 0;
-                       return 0;
-               }
-               if (errno != EDEADLK) {
-                       break;
-               }
-               /* sleep for as short a time as we can - more portable than usleep() */
-               tv.tv_sec = 0;
-               tv.tv_usec = 1;
-               select(0, NULL, NULL, NULL, &tv);
-       }
-       TDB1_LOG((tdb, TDB1_DEBUG_TRACE,"tdb1_allrecord_upgrade failed\n"));
+       enum TDB_ERROR ecode = tdb_allrecord_upgrade(tdb, TDB1_FREELIST_TOP);
+       if (ecode == TDB_SUCCESS)
+               return 0;
+       tdb->last_error = ecode;
        return -1;
 }
 
-static struct tdb1_lock_type *tdb1_find_nestlock(struct tdb1_context *tdb,
-                                                tdb1_off_t offset)
+static struct tdb_lock *tdb1_find_nestlock(struct tdb_context *tdb,
+                                          tdb1_off_t offset)
 {
        unsigned int i;
 
-       for (i=0; i<tdb->num_lockrecs; i++) {
-               if (tdb->lockrecs[i].off == offset) {
-                       return &tdb->lockrecs[i];
+       for (i=0; i<tdb->file->num_lockrecs; i++) {
+               if (tdb->file->lockrecs[i].off == offset) {
+                       return &tdb->file->lockrecs[i];
                }
        }
        return NULL;
 }
 
 /* lock an offset in the database. */
-int tdb1_nest_lock(struct tdb1_context *tdb, uint32_t offset, int ltype,
-                 enum tdb1_lock_flags flags)
+int tdb1_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
+                 enum tdb_lock_flags flags)
 {
-       struct tdb1_lock_type *new_lck;
-
-       if (offset >= lock_offset(tdb->header.hash_size)) {
-               tdb->ecode = TDB1_ERR_LOCK;
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR,"tdb1_lock: invalid offset %u for ltype=%d\n",
-                        offset, ltype));
-               return -1;
-       }
-       if (tdb->flags & TDB1_NOLOCK)
-               return 0;
-
-       new_lck = tdb1_find_nestlock(tdb, offset);
-       if (new_lck) {
-               /*
-                * Just increment the in-memory struct, posix locks
-                * don't stack.
-                */
-               new_lck->count++;
-               return 0;
-       }
+       enum TDB_ERROR ecode;
 
-       new_lck = (struct tdb1_lock_type *)realloc(
-               tdb->lockrecs,
-               sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
-       if (new_lck == NULL) {
-               errno = ENOMEM;
+       if (offset >= lock_offset(tdb->tdb1.header.hash_size)) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                       "tdb1_lock: invalid offset %u for"
+                                       " ltype=%d",
+                                       offset, ltype);
                return -1;
        }
-       tdb->lockrecs = new_lck;
 
-       /* Since fcntl locks don't nest, we do a lock for the first one,
-          and simply bump the count for future ones */
-       if (tdb1_brlock(tdb, ltype, offset, 1, flags)) {
+       ecode = tdb_nest_lock(tdb, offset, ltype, flags | TDB_LOCK_NOCHECK);
+       if (unlikely(ecode != TDB_SUCCESS)) {
+               tdb->last_error = ecode;
                return -1;
        }
-
-       tdb->lockrecs[tdb->num_lockrecs].off = offset;
-       tdb->lockrecs[tdb->num_lockrecs].count = 1;
-       tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
-       tdb->num_lockrecs++;
-
        return 0;
 }
 
-static int tdb1_lock_and_recover(struct tdb1_context *tdb)
+static int tdb1_lock_and_recover(struct tdb_context *tdb)
 {
        int ret;
 
        /* We need to match locking order in transaction commit. */
-       if (tdb1_brlock(tdb, F_WRLCK, TDB1_FREELIST_TOP, 0, TDB1_LOCK_WAIT)) {
+       if (tdb1_brlock(tdb, F_WRLCK, TDB1_FREELIST_TOP, 0,
+                       TDB_LOCK_WAIT|TDB_LOCK_NOCHECK)) {
                return -1;
        }
 
-       if (tdb1_brlock(tdb, F_WRLCK, TDB1_OPEN_LOCK, 1, TDB1_LOCK_WAIT)) {
+       if (tdb1_brlock(tdb, F_WRLCK, TDB1_OPEN_LOCK, 1,
+                       TDB_LOCK_WAIT|TDB_LOCK_NOCHECK)) {
                tdb1_brunlock(tdb, F_WRLCK, TDB1_FREELIST_TOP, 0);
                return -1;
        }
@@ -323,39 +131,56 @@ static int tdb1_lock_and_recover(struct tdb1_context *tdb)
        return ret;
 }
 
-static bool have_data_locks(const struct tdb1_context *tdb)
+static bool have_data_locks(const struct tdb_context *tdb)
 {
        unsigned int i;
 
-       for (i = 0; i < tdb->num_lockrecs; i++) {
-               if (tdb->lockrecs[i].off >= lock_offset(-1))
+       for (i = 0; i < tdb->file->num_lockrecs; i++) {
+               if (tdb->file->lockrecs[i].off >= lock_offset(-1))
                        return true;
        }
        return false;
 }
 
-static int tdb1_lock_list(struct tdb1_context *tdb, int list, int ltype,
-                        enum tdb1_lock_flags waitflag)
+static int tdb1_lock_list(struct tdb_context *tdb, int list, int ltype,
+                        enum tdb_lock_flags waitflag)
 {
        int ret;
        bool check = false;
 
        /* a allrecord lock allows us to avoid per chain locks */
-       if (tdb->allrecord_lock.count &&
-           (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
-               return 0;
+       if (tdb->file->allrecord_lock.count) {
+               if (!check_lock_pid(tdb, "tdb1_lock_list", true)) {
+                       tdb->last_error = TDB_ERR_LOCK;
+                       return -1;
+               }
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error = owner_conflict(tdb, "tdb1_lock_list");
+                       return -1;
+               }
+               if (ltype == tdb->file->allrecord_lock.ltype
+                   || ltype == F_RDLCK) {
+                       return 0;
+               }
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb1_lock_list:"
+                                            " already have read lock");
+               return -1;
        }
 
-       if (tdb->allrecord_lock.count) {
-               tdb->ecode = TDB1_ERR_LOCK;
-               ret = -1;
-       } else {
-               /* Only check when we grab first data lock. */
-               check = !have_data_locks(tdb);
-               ret = tdb1_nest_lock(tdb, lock_offset(list), ltype, waitflag);
+       /* Only check when we grab first data lock. */
+       check = !have_data_locks(tdb);
+       ret = tdb1_nest_lock(tdb, lock_offset(list), ltype, waitflag);
+
+       if (ret == 0 && check) {
+               tdb_bool_err berr = tdb1_needs_recovery(tdb);
 
-               if (ret == 0 && check && tdb1_needs_recovery(tdb)) {
-                       tdb1_nest_unlock(tdb, lock_offset(list), ltype, false);
+               if (berr < 0) {
+                       return -1;
+               }
+               if (berr == true) {
+                       tdb1_nest_unlock(tdb, lock_offset(list), ltype);
 
                        if (tdb1_lock_and_recover(tdb) == -1) {
                                return -1;
@@ -367,105 +192,65 @@ static int tdb1_lock_list(struct tdb1_context *tdb, int list, int ltype,
 }
 
 /* lock a list in the database. list -1 is the alloc list */
-int tdb1_lock(struct tdb1_context *tdb, int list, int ltype)
+int tdb1_lock(struct tdb_context *tdb, int list, int ltype)
 {
        int ret;
 
-       ret = tdb1_lock_list(tdb, list, ltype, TDB1_LOCK_WAIT);
-       if (ret) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR, "tdb1_lock failed on list %d "
-                        "ltype=%d (%s)\n",  list, ltype, strerror(errno)));
+       ret = tdb1_lock_list(tdb, list, ltype, TDB_LOCK_WAIT);
+       /* Don't log for EAGAIN and EINTR: they could have overridden lock fns */
+       if (ret && errno != EAGAIN && errno != EINTR) {
+               tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
+                          "tdb1_lock failed on list %d "
+                          "ltype=%d (%s)",  list, ltype, strerror(errno));
        }
        return ret;
 }
 
-/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
-int tdb1_lock_nonblock(struct tdb1_context *tdb, int list, int ltype)
-{
-       return tdb1_lock_list(tdb, list, ltype, TDB1_LOCK_NOWAIT);
-}
-
-
-int tdb1_nest_unlock(struct tdb1_context *tdb, uint32_t offset, int ltype,
-                   bool mark_lock)
+int tdb1_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype)
 {
-       int ret = -1;
-       struct tdb1_lock_type *lck;
-
-       if (tdb->flags & TDB1_NOLOCK)
-               return 0;
+       enum TDB_ERROR ecode;
 
        /* Sanity checks */
-       if (offset >= lock_offset(tdb->header.hash_size)) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR, "tdb1_unlock: offset %u invalid (%d)\n", offset, tdb->header.hash_size));
-               return ret;
-       }
-
-       lck = tdb1_find_nestlock(tdb, offset);
-       if ((lck == NULL) || (lck->count == 0)) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR, "tdb1_unlock: count is 0\n"));
+       if (offset >= lock_offset(tdb->tdb1.header.hash_size)) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
+                                       "tdb1_unlock: offset %u invalid (%d)",
+                                       offset, tdb->tdb1.header.hash_size);
                return -1;
        }
 
-       if (lck->count > 1) {
-               lck->count--;
-               return 0;
-       }
-
-       /*
-        * This lock has count==1 left, so we need to unlock it in the
-        * kernel. We don't bother with decrementing the in-memory array
-        * element, we're about to overwrite it with the last array element
-        * anyway.
-        */
-
-       if (mark_lock) {
-               ret = 0;
-       } else {
-               ret = tdb1_brunlock(tdb, ltype, offset, 1);
-       }
-
-       /*
-        * Shrink the array by overwriting the element just unlocked with the
-        * last array element.
-        */
-       *lck = tdb->lockrecs[--tdb->num_lockrecs];
-
-       /*
-        * We don't bother with realloc when the array shrinks, but if we have
-        * a completely idle tdb we should get rid of the locked array.
-        */
-
-       if (tdb->num_lockrecs == 0) {
-               SAFE_FREE(tdb->lockrecs);
+       ecode = tdb_nest_unlock(tdb, offset, ltype);
+       if (unlikely(ecode != TDB_SUCCESS)) {
+               tdb->last_error = ecode;
+               return -1;
        }
-
-       if (ret)
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR, "tdb1_unlock: An error occurred unlocking!\n"));
-       return ret;
+       return 0;
 }
 
-int tdb1_unlock(struct tdb1_context *tdb, int list, int ltype)
+int tdb1_unlock(struct tdb_context *tdb, int list, int ltype)
 {
        /* a global lock allows us to avoid per chain locks */
-       if (tdb->allrecord_lock.count &&
-           (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
+       if (tdb->file->allrecord_lock.count &&
+           (ltype == tdb->file->allrecord_lock.ltype || ltype == F_RDLCK)) {
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error = owner_conflict(tdb, "tdb1_unlock");
+                       return -1;
+               }
                return 0;
        }
 
-       if (tdb->allrecord_lock.count) {
-               tdb->ecode = TDB1_ERR_LOCK;
+       if (tdb->file->allrecord_lock.count) {
+               tdb->last_error = TDB_ERR_LOCK;
                return -1;
        }
 
-       return tdb1_nest_unlock(tdb, lock_offset(list), ltype, false);
+       return tdb1_nest_unlock(tdb, lock_offset(list), ltype);
 }
 
 /*
   get the transaction lock
  */
-int tdb1_transaction_lock(struct tdb1_context *tdb, int ltype,
-                        enum tdb1_lock_flags lockflags)
+int tdb1_transaction_lock(struct tdb_context *tdb, int ltype,
+                        enum tdb_lock_flags lockflags)
 {
        return tdb1_nest_lock(tdb, TDB1_TRANSACTION_LOCK, ltype, lockflags);
 }
@@ -473,92 +258,52 @@ int tdb1_transaction_lock(struct tdb1_context *tdb, int ltype,
 /*
   release the transaction lock
  */
-int tdb1_transaction_unlock(struct tdb1_context *tdb, int ltype)
+int tdb1_transaction_unlock(struct tdb_context *tdb, int ltype)
 {
-       return tdb1_nest_unlock(tdb, TDB1_TRANSACTION_LOCK, ltype, false);
+       return tdb1_nest_unlock(tdb, TDB1_TRANSACTION_LOCK, ltype);
 }
 
-/* Returns 0 if all done, -1 if error, 1 if ok. */
-static int tdb1_allrecord_check(struct tdb1_context *tdb, int ltype,
-                              enum tdb1_lock_flags flags, bool upgradable)
-{
-       /* There are no locks on read-only dbs */
-       if (tdb->read_only || tdb->traverse_read) {
-               tdb->ecode = TDB1_ERR_LOCK;
+/* lock/unlock entire database.  It can only be upgradable if you have some
+ * other way of guaranteeing exclusivity (ie. transaction write lock).
+ * We do the locking gradually to avoid being starved by smaller locks. */
+int tdb1_allrecord_lock(struct tdb_context *tdb, int ltype,
+                      enum tdb_lock_flags flags, bool upgradable)
+{
+       enum TDB_ERROR ecode;
+       tdb_bool_err berr;
+
+       /* tdb_lock_gradual() doesn't know about tdb->tdb1.traverse_read. */
+       if (tdb->tdb1.traverse_read && !(tdb->flags & TDB_NOLOCK)) {
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb1_allrecord_lock during"
+                                            " tdb1_read_traverse");
                return -1;
        }
 
-       if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
-               tdb->allrecord_lock.count++;
+       if (tdb->file->allrecord_lock.count
+           && tdb->file->allrecord_lock.ltype == ltype) {
+               tdb->file->allrecord_lock.count++;
                return 0;
        }
 
-       if (tdb->allrecord_lock.count) {
-               /* a global lock of a different type exists */
-               tdb->ecode = TDB1_ERR_LOCK;
-               return -1;
-       }
-
        if (tdb1_have_extra_locks(tdb)) {
                /* can't combine global and chain locks */
-               tdb->ecode = TDB1_ERR_LOCK;
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+                                            TDB_LOG_USE_ERROR,
+                                            "tdb1_allrecord_lock holding"
+                                            " other locks");
                return -1;
        }
 
        if (upgradable && ltype != F_RDLCK) {
                /* tdb error: you can't upgrade a write lock! */
-               tdb->ecode = TDB1_ERR_LOCK;
+               tdb->last_error = tdb_logerr(tdb, TDB_ERR_LOCK,
+                                            TDB_LOG_ERROR,
+                                            "tdb1_allrecord_lock cannot"
+                                            " have upgradable write lock");
                return -1;
        }
-       return 1;
-}
-
-/* We only need to lock individual bytes, but Linux merges consecutive locks
- * so we lock in contiguous ranges. */
-static int tdb1_chainlock_gradual(struct tdb1_context *tdb,
-                                int ltype, enum tdb1_lock_flags flags,
-                                size_t off, size_t len)
-{
-       int ret;
-       enum tdb1_lock_flags nb_flags = (flags & ~TDB1_LOCK_WAIT);
-
-       if (len <= 4) {
-               /* Single record.  Just do blocking lock. */
-               return tdb1_brlock(tdb, ltype, off, len, flags);
-       }
-
-       /* First we try non-blocking. */
-       ret = tdb1_brlock(tdb, ltype, off, len, nb_flags);
-       if (ret == 0) {
-               return 0;
-       }
-
-       /* Try locking first half, then second. */
-       ret = tdb1_chainlock_gradual(tdb, ltype, flags, off, len / 2);
-       if (ret == -1)
-               return -1;
-
-       ret = tdb1_chainlock_gradual(tdb, ltype, flags,
-                                   off + len / 2, len - len / 2);
-       if (ret == -1) {
-               tdb1_brunlock(tdb, ltype, off, len / 2);
-               return -1;
-       }
-       return 0;
-}
-
-/* lock/unlock entire database.  It can only be upgradable if you have some
- * other way of guaranteeing exclusivity (ie. transaction write lock).
- * We do the locking gradually to avoid being starved by smaller locks. */
-int tdb1_allrecord_lock(struct tdb1_context *tdb, int ltype,
-                      enum tdb1_lock_flags flags, bool upgradable)
-{
-       switch (tdb1_allrecord_check(tdb, ltype, flags, upgradable)) {
-       case -1:
-               return -1;
-       case 0:
-               return 0;
-       }
 
        /* We cover two kinds of locks:
         * 1) Normal chain locks.  Taken for almost all operations.
@@ -567,34 +312,36 @@ int tdb1_allrecord_lock(struct tdb1_context *tdb, int ltype,
         *
         * It is (1) which cause the starvation problem, so we're only
         * gradual for that. */
-       if (tdb1_chainlock_gradual(tdb, ltype, flags, TDB1_FREELIST_TOP,
-                                 tdb->header.hash_size * 4) == -1) {
+       ecode = tdb_lock_gradual(tdb, ltype, flags | TDB_LOCK_NOCHECK,
+                                TDB1_FREELIST_TOP, tdb->tdb1.header.hash_size * 4);
+       if (ecode != TDB_SUCCESS) {
+               tdb->last_error = ecode;
                return -1;
        }
 
        /* Grab individual record locks. */
-       if (tdb1_brlock(tdb, ltype, lock_offset(tdb->header.hash_size), 0,
+       if (tdb1_brlock(tdb, ltype, lock_offset(tdb->tdb1.header.hash_size), 0,
                       flags) == -1) {
                tdb1_brunlock(tdb, ltype, TDB1_FREELIST_TOP,
-                            tdb->header.hash_size * 4);
+                            tdb->tdb1.header.hash_size * 4);
                return -1;
        }
 
-       tdb->allrecord_lock.count = 1;
+       tdb->file->allrecord_lock.owner = tdb;
+       tdb->file->allrecord_lock.count = 1;
+       tdb->file->locker = getpid();
        /* If it's upgradable, it's actually exclusive so we can treat
         * it as a write lock. */
-       tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
-       tdb->allrecord_lock.off = upgradable;
-
-       if (tdb1_needs_recovery(tdb)) {
-               bool mark = flags & TDB1_LOCK_MARK_ONLY;
-               tdb1_allrecord_unlock(tdb, ltype, mark);
-               if (mark) {
-                       tdb->ecode = TDB1_ERR_LOCK;
-                       TDB1_LOG((tdb, TDB1_DEBUG_ERROR,
-                                "tdb1_lockall_mark cannot do recovery\n"));
-                       return -1;
-               }
+       tdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
+       tdb->file->allrecord_lock.off = upgradable;
+
+       berr = tdb1_needs_recovery(tdb);
+       if (berr < 0) {
+               return -1;
+       }
+
+       if (berr == true) {
+               tdb1_allrecord_unlock(tdb, ltype);
                if (tdb1_lock_and_recover(tdb) == -1) {
                        return -1;
                }
@@ -607,148 +354,94 @@ int tdb1_allrecord_lock(struct tdb1_context *tdb, int ltype,
 
 
 /* unlock entire db */
-int tdb1_allrecord_unlock(struct tdb1_context *tdb, int ltype, bool mark_lock)
+int tdb1_allrecord_unlock(struct tdb_context *tdb, int ltype)
 {
-       /* There are no locks on read-only dbs */
-       if (tdb->read_only || tdb->traverse_read) {
-               tdb->ecode = TDB1_ERR_LOCK;
+       /* Don't try this during r/o traversal! */
+       if (tdb->tdb1.traverse_read) {
+               tdb->last_error = TDB_ERR_LOCK;
                return -1;
        }
 
-       if (tdb->allrecord_lock.count == 0) {
-               tdb->ecode = TDB1_ERR_LOCK;
+       if (tdb->file->allrecord_lock.count == 0) {
+               tdb->last_error = TDB_ERR_LOCK;
                return -1;
        }
 
        /* Upgradable locks are marked as write locks. */
-       if (tdb->allrecord_lock.ltype != ltype
-           && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
-               tdb->ecode = TDB1_ERR_LOCK;
+       if (tdb->file->allrecord_lock.ltype != ltype
+           && (!tdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
+               tdb->last_error = TDB_ERR_LOCK;
                return -1;
        }
 
-       if (tdb->allrecord_lock.count > 1) {
-               tdb->allrecord_lock.count--;
+       if (tdb->file->allrecord_lock.count > 1) {
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error
+                               = owner_conflict(tdb, "tdb1_allrecord_unlock");
+                       return -1;
+               }
+               tdb->file->allrecord_lock.count--;
                return 0;
        }
 
-       if (!mark_lock && tdb1_brunlock(tdb, ltype, TDB1_FREELIST_TOP, 0)) {
-               TDB1_LOG((tdb, TDB1_DEBUG_ERROR, "tdb1_unlockall failed (%s)\n", strerror(errno)));
+       tdb->file->allrecord_lock.count = 0;
+       tdb->file->allrecord_lock.ltype = 0;
+
+       if (tdb1_brunlock(tdb, ltype, TDB1_FREELIST_TOP, 0)) {
+               tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
+                          "tdb1_unlockall failed (%s)", strerror(errno));
                return -1;
        }
 
-       tdb->allrecord_lock.count = 0;
-       tdb->allrecord_lock.ltype = 0;
-
        return 0;
 }
 
-/* lock entire database with write lock */
-_PUBLIC_ int tdb1_lockall(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_lock(tdb, F_WRLCK, TDB1_LOCK_WAIT, false);
-}
-
-/* lock entire database with write lock - mark only */
-_PUBLIC_ int tdb1_lockall_mark(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_lock(tdb, F_WRLCK, TDB1_LOCK_MARK_ONLY, false);
-}
-
-/* unlock entire database with write lock - unmark only */
-_PUBLIC_ int tdb1_lockall_unmark(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_unlock(tdb, F_WRLCK, true);
-}
-
-/* lock entire database with write lock - nonblocking varient */
-_PUBLIC_ int tdb1_lockall_nonblock(struct tdb1_context *tdb)
-{
-       int ret = tdb1_allrecord_lock(tdb, F_WRLCK, TDB1_LOCK_NOWAIT, false);
-       return ret;
-}
-
-/* unlock entire database with write lock */
-_PUBLIC_ int tdb1_unlockall(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_unlock(tdb, F_WRLCK, false);
-}
-
-/* lock entire database with read lock */
-_PUBLIC_ int tdb1_lockall_read(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_lock(tdb, F_RDLCK, TDB1_LOCK_WAIT, false);
-}
-
-/* lock entire database with read lock - nonblock varient */
-_PUBLIC_ int tdb1_lockall_read_nonblock(struct tdb1_context *tdb)
-{
-       int ret = tdb1_allrecord_lock(tdb, F_RDLCK, TDB1_LOCK_NOWAIT, false);
-       return ret;
-}
-
-/* unlock entire database with read lock */
-_PUBLIC_ int tdb1_unlockall_read(struct tdb1_context *tdb)
-{
-       return tdb1_allrecord_unlock(tdb, F_RDLCK, false);
-}
-
 /* lock/unlock one hash chain. This is meant to be used to reduce
    contention - it cannot guarantee how many records will be locked */
-_PUBLIC_ int tdb1_chainlock(struct tdb1_context *tdb, TDB1_DATA key)
+int tdb1_chainlock(struct tdb_context *tdb, TDB_DATA key)
 {
-       int ret = tdb1_lock(tdb, TDB1_BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+       int ret = tdb1_lock(tdb,
+                           TDB1_BUCKET(tdb_hash(tdb, key.dptr, key.dsize)),
+                           F_WRLCK);
        return ret;
 }
 
-/* lock/unlock one hash chain, non-blocking. This is meant to be used
-   to reduce contention - it cannot guarantee how many records will be
-   locked */
-_PUBLIC_ int tdb1_chainlock_nonblock(struct tdb1_context *tdb, TDB1_DATA key)
+int tdb1_chainunlock(struct tdb_context *tdb, TDB_DATA key)
 {
-       int ret = tdb1_lock_nonblock(tdb, TDB1_BUCKET(tdb->hash_fn(&key)), F_WRLCK);
-       return ret;
-}
-
-/* mark a chain as locked without actually locking it. Warning! use with great caution! */
-_PUBLIC_ int tdb1_chainlock_mark(struct tdb1_context *tdb, TDB1_DATA key)
-{
-       int ret = tdb1_nest_lock(tdb, lock_offset(TDB1_BUCKET(tdb->hash_fn(&key))),
-                               F_WRLCK, TDB1_LOCK_MARK_ONLY);
-       return ret;
-}
-
-/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
-_PUBLIC_ int tdb1_chainlock_unmark(struct tdb1_context *tdb, TDB1_DATA key)
-{
-       return tdb1_nest_unlock(tdb, lock_offset(TDB1_BUCKET(tdb->hash_fn(&key))),
-                              F_WRLCK, true);
-}
-
-_PUBLIC_ int tdb1_chainunlock(struct tdb1_context *tdb, TDB1_DATA key)
-{
-       return tdb1_unlock(tdb, TDB1_BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+       return tdb1_unlock(tdb, TDB1_BUCKET(tdb_hash(tdb, key.dptr, key.dsize)),
+                          F_WRLCK);
 }
 
-_PUBLIC_ int tdb1_chainlock_read(struct tdb1_context *tdb, TDB1_DATA key)
+int tdb1_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
 {
        int ret;
-       ret = tdb1_lock(tdb, TDB1_BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+       ret = tdb1_lock(tdb, TDB1_BUCKET(tdb_hash(tdb, key.dptr, key.dsize)),
+                       F_RDLCK);
        return ret;
 }
 
-_PUBLIC_ int tdb1_chainunlock_read(struct tdb1_context *tdb, TDB1_DATA key)
+int tdb1_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
 {
-       return tdb1_unlock(tdb, TDB1_BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+       return tdb1_unlock(tdb, TDB1_BUCKET(tdb_hash(tdb, key.dptr, key.dsize)),
+                          F_RDLCK);
 }
 
 /* record lock stops delete underneath */
-int tdb1_lock_record(struct tdb1_context *tdb, tdb1_off_t off)
+int tdb1_lock_record(struct tdb_context *tdb, tdb1_off_t off)
 {
-       if (tdb->allrecord_lock.count) {
+       if (tdb->file->allrecord_lock.count) {
+               if (!check_lock_pid(tdb, "tdb1_lock_record", true)) {
+                       tdb->last_error = TDB_ERR_LOCK;
+                       return -1;
+               }
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error = owner_conflict(tdb,
+                                                        "tdb1_lock_record");
+                       return -1;
+               }
                return 0;
        }
-       return off ? tdb1_brlock(tdb, F_RDLCK, off, 1, TDB1_LOCK_WAIT) : 0;
+       return off ? tdb1_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
 }
 
 /*
@@ -756,53 +449,73 @@ int tdb1_lock_record(struct tdb1_context *tdb, tdb1_off_t off)
   Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
   an error to fail to get the lock here.
 */
-int tdb1_write_lock_record(struct tdb1_context *tdb, tdb1_off_t off)
+int tdb1_write_lock_record(struct tdb_context *tdb, tdb1_off_t off)
 {
        struct tdb1_traverse_lock *i;
-       for (i = &tdb->travlocks; i; i = i->next)
+       for (i = &tdb->tdb1.travlocks; i; i = i->next)
                if (i->off == off)
                        return -1;
-       if (tdb->allrecord_lock.count) {
-               if (tdb->allrecord_lock.ltype == F_WRLCK) {
+       if (tdb->file->allrecord_lock.count) {
+               if (!check_lock_pid(tdb, "tdb1_write_lock_record", true)) {
+                       tdb->last_error = TDB_ERR_LOCK;
+                       return -1;
+               }
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error
+                               = owner_conflict(tdb, "tdb1_write_lock_record");
+                       return -1;
+               }
+               if (tdb->file->allrecord_lock.ltype == F_WRLCK) {
                        return 0;
                }
                return -1;
        }
-       return tdb1_brlock(tdb, F_WRLCK, off, 1, TDB1_LOCK_NOWAIT|TDB1_LOCK_PROBE);
+       return tdb1_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
 }
 
-int tdb1_write_unlock_record(struct tdb1_context *tdb, tdb1_off_t off)
+int tdb1_write_unlock_record(struct tdb_context *tdb, tdb1_off_t off)
 {
-       if (tdb->allrecord_lock.count) {
+       if (tdb->file->allrecord_lock.count) {
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error
+                               = owner_conflict(tdb,
+                                                "tdb1_write_unlock_record");
+                       return -1;
+               }
                return 0;
        }
        return tdb1_brunlock(tdb, F_WRLCK, off, 1);
 }
 
 /* fcntl locks don't stack: avoid unlocking someone else's */
-int tdb1_unlock_record(struct tdb1_context *tdb, tdb1_off_t off)
+int tdb1_unlock_record(struct tdb_context *tdb, tdb1_off_t off)
 {
        struct tdb1_traverse_lock *i;
        uint32_t count = 0;
 
-       if (tdb->allrecord_lock.count) {
+       if (tdb->file->allrecord_lock.count) {
+               if (tdb->file->allrecord_lock.owner != tdb) {
+                       tdb->last_error = owner_conflict(tdb,
+                                                        "tdb1_unlock_record");
+                       return -1;
+               }
                return 0;
        }
 
        if (off == 0)
                return 0;
-       for (i = &tdb->travlocks; i; i = i->next)
+       for (i = &tdb->tdb1.travlocks; i; i = i->next)
                if (i->off == off)
                        count++;
        return (count == 1 ? tdb1_brunlock(tdb, F_RDLCK, off, 1) : 0);
 }
 
-bool tdb1_have_extra_locks(struct tdb1_context *tdb)
+bool tdb1_have_extra_locks(struct tdb_context *tdb)
 {
-       unsigned int extra = tdb->num_lockrecs;
+       unsigned int extra = tdb->file->num_lockrecs;
 
        /* A transaction holds the lock for all records. */
-       if (!tdb->transaction && tdb->allrecord_lock.count) {
+       if (!tdb->tdb1.transaction && tdb->file->allrecord_lock.count) {
                return true;
        }
 
@@ -812,7 +525,7 @@ bool tdb1_have_extra_locks(struct tdb1_context *tdb)
        }
 
        /* In a transaction, we expect to hold the transaction lock */
-       if (tdb->transaction
+       if (tdb->tdb1.transaction
            && tdb1_find_nestlock(tdb, TDB1_TRANSACTION_LOCK)) {
                extra--;
        }
@@ -821,27 +534,27 @@ bool tdb1_have_extra_locks(struct tdb1_context *tdb)
 }
 
 /* The transaction code uses this to remove all locks. */
-void tdb1_release_transaction_locks(struct tdb1_context *tdb)
+void tdb1_release_transaction_locks(struct tdb_context *tdb)
 {
        unsigned int i, active = 0;
 
-       if (tdb->allrecord_lock.count != 0) {
-               tdb1_brunlock(tdb, tdb->allrecord_lock.ltype, TDB1_FREELIST_TOP, 0);
-               tdb->allrecord_lock.count = 0;
+       if (tdb->file->allrecord_lock.count != 0) {
+               tdb1_brunlock(tdb, tdb->file->allrecord_lock.ltype, TDB1_FREELIST_TOP, 0);
+               tdb->file->allrecord_lock.count = 0;
        }
 
-       for (i=0;i<tdb->num_lockrecs;i++) {
-               struct tdb1_lock_type *lck = &tdb->lockrecs[i];
+       for (i=0;i<tdb->file->num_lockrecs;i++) {
+               struct tdb_lock *lck = &tdb->file->lockrecs[i];
 
                /* Don't release the active lock!  Copy it to first entry. */
                if (lck->off == TDB1_ACTIVE_LOCK) {
-                       tdb->lockrecs[active++] = *lck;
+                       tdb->file->lockrecs[active++] = *lck;
                } else {
                        tdb1_brunlock(tdb, lck->ltype, lck->off, 1);
                }
        }
-       tdb->num_lockrecs = active;
-       if (tdb->num_lockrecs == 0) {
-               SAFE_FREE(tdb->lockrecs);
+       tdb->file->num_lockrecs = active;
+       if (tdb->file->num_lockrecs == 0) {
+               SAFE_FREE(tdb->file->lockrecs);
        }
 }