[Xen-changelog] Make xenstored use tdb, transactions can soft-fail (EAGAIN)

[Xen patchbot -unstable <patchbot-unstable@xxxxxxxxxxxxxxxxxxx>]

# HG changeset patch
# User Rusty Russell <rusty@xxxxxxxxxxxxxxx>
# Node ID 76af1a1df67cd8958fc40b1c574a8bd26d5026c2
# Parent  6aef7d1062bb6035a2a84603d9f054b24e14c9b2
Make xenstored use tdb, transactions can soft-fail (EAGAIN)
Transactions no longer take root dir, no longer lock & block: commit can fail 
spuriously with EAGAIN, not ETIMEDOUT.
Speeds up transactions by over 1000 times, should be NFS safe.
New program: xs_tdb_dump to dump raw TDB contents.
Don't do failure testing: we are no longer robust against all ENOMEM 8(
Introduce "struct node" which contains perms, children and data.
Make struct xs_permissions unpadded, so we can write to tdb w/o valgrind 
complaints.
Gently modify TDB to use talloc, not do alloc on tdb_delete.

Signed-off-by: Rusty Russell <rusty@xxxxxxxxxxxxxxx>

diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/Makefile
--- a/tools/xenstore/Makefile   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/Makefile   Fri Sep 23 13:25:01 2005
@@ -28,11 +28,11 @@
 CLIENTS += xenstore-write
 CLIENTS_OBJS := $(patsubst xenstore-%,xenstore_%.o,$(CLIENTS))
 
-all: libxenstore.so xenstored $(CLIENTS)
+all: libxenstore.so xenstored $(CLIENTS) xs_tdb_dump
 
 testcode: xs_test xenstored_test xs_random xs_dom0_test
 
-xenstored: xenstored_core.o xenstored_watch.o xenstored_domain.o 
xenstored_transaction.o xs_lib.o talloc.o utils.o
+xenstored: xenstored_core.o xenstored_watch.o xenstored_domain.o 
xenstored_transaction.o xs_lib.o talloc.o utils.o tdb.o
        $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -lxenctrl -o $@
 
 $(CLIENTS): libxenstore.so
@@ -42,7 +42,10 @@
 $(CLIENTS_OBJS): xenstore_%.o: xenstore_client.c
        $(COMPILE.c) -DCLIENT_$(*F) -o $@ $<
 
-xenstored_test: xenstored_core_test.o xenstored_watch_test.o 
xenstored_domain_test.o xenstored_transaction_test.o xs_lib.o talloc_test.o 
fake_libxc.o utils.o
+xenstored_test: xenstored_core_test.o xenstored_watch_test.o 
xenstored_domain_test.o xenstored_transaction_test.o xs_lib.o talloc_test.o 
fake_libxc.o utils.o tdb.o
+       $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
+
+xs_tdb_dump: xs_tdb_dump.o utils.o tdb.o talloc.o
        $(LINK.o) $^ $(LOADLIBES) $(LDLIBS) -o $@
 
 xs_test: xs_test.o xs_lib.o utils.o
@@ -103,7 +106,7 @@
 randomcheck: xs_random xenstored_test $(TESTDIR)
        $(TESTENV) ./xs_random --simple --fast /tmp/xs_random 200000 
$(RANDSEED) && echo
        $(TESTENV) ./xs_random --fast /tmp/xs_random 100000 $(RANDSEED) && echo
-       $(TESTENV) ./xs_random --fail /tmp/xs_random 10000 $(RANDSEED)
+#      $(TESTENV) ./xs_random --fail /tmp/xs_random 10000 $(RANDSEED)
 
 crashme:  xs_crashme xenstored_test $(TESTDIR)
        rm -rf $(TESTDIR)/store $(TESTDIR)/transactions /tmp/xs_crashme.vglog* 
/tmp/trace
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/testsuite/04rm.test
--- a/tools/xenstore/testsuite/04rm.test        Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/04rm.test        Fri Sep 23 13:25:01 2005
@@ -6,6 +6,8 @@
 # Create file and remove it
 write /test contents
 rm /test
+expect tool
+dir /
 
 # Create directory and remove it.
 mkdir /dir
@@ -15,3 +17,4 @@
 mkdir /dir
 write /dir/test contents
 rm /dir
+
diff -r 6aef7d1062bb -r 76af1a1df67c 
tools/xenstore/testsuite/08transaction.slowtest
--- a/tools/xenstore/testsuite/08transaction.slowtest   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/08transaction.slowtest   Fri Sep 23 13:25:01 2005
@@ -1,21 +1,43 @@
-# Test transaction timeouts.  Take a second each.
+# Test transaction clashes.
 
 mkdir /test
 write /test/entry1 contents
 
-# Transactions can take as long as the want...
-start /test
-sleep 1100
-rm /test/entry1
+# Start transaction, do read-only op, transaction succeeds
+1 start
+1 write /test/entry1 contents2
+expect contents
+read /test/entry1
+1 commit
+expect contents2
+read /test/entry1
+
+# Start transaction, abort other transaction, transaction succeeds.
+1 start
+1 write /test/entry1 contents3
+start
+write /test/entry1 contents
+abort
+1 commit
+expect contents3
+read /test/entry1
+
+# Start transaction, do write op, transaction fails
+1 start
+1 write /test/entry1 contents4
+write /test/entry1 contents
+expect 1: commit failed: Resource temporarily unavailable
+1 commit
+expect contents
+read /test/entry1
+
+# Start transaction, do other transaction, transaction fails
+1 start
+1 write /test/entry1 contents4
+start
+write /test/entry1 contents5
 commit
-dir /test
-
-# ... as long as noone is waiting.
-1 start /test
-notimeout
-2 mkdir /test/dir
-1 mkdir /test/dir
-expect 1:dir
-1 dir /test
-expect 1: commit failed: Connection timed out
+expect 1: commit failed: Resource temporarily unavailable
 1 commit
+expect contents5
+read /test/entry1
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/testsuite/08transaction.test
--- a/tools/xenstore/testsuite/08transaction.test       Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/08transaction.test       Fri Sep 23 13:25:01 2005
@@ -3,7 +3,7 @@
 mkdir /test
 
 # Simple transaction: create a file inside transaction.
-1 start /test
+1 start
 1 write /test/entry1 contents
 2 dir /test
 expect 1:entry1
@@ -15,7 +15,7 @@
 rm /test/entry1
 
 # Create a file and abort transaction.
-1 start /test
+1 start
 1 write /test/entry1 contents
 2 dir /test
 expect 1:entry1
@@ -25,7 +25,7 @@
 
 write /test/entry1 contents
 # Delete in transaction, commit
-1 start /test
+1 start
 1 rm /test/entry1
 expect 2:entry1
 2 dir /test
@@ -35,7 +35,7 @@
 
 # Delete in transaction, abort.
 write /test/entry1 contents
-1 start /test
+1 start
 1 rm /test/entry1
 expect 2:entry1
 2 dir /test
@@ -47,7 +47,7 @@
 # Events inside transactions don't trigger watches until (successful) commit.
 mkdir /test/dir
 1 watch /test token
-2 start /test
+2 start
 2 mkdir /test/dir/sub
 expect 1: waitwatch failed: Connection timed out
 1 waitwatch
@@ -55,7 +55,7 @@
 1 close
 
 1 watch /test token
-2 start /test
+2 start
 2 mkdir /test/dir/sub
 2 abort
 expect 1: waitwatch failed: Connection timed out
@@ -63,7 +63,7 @@
 1 close
 
 1 watch /test token
-2 start /test
+2 start
 2 mkdir /test/dir/sub
 2 commit
 expect 1:/test/dir/sub:token
@@ -73,7 +73,7 @@
 
 # Rm inside transaction works like rm outside: children get notified.
 1 watch /test/dir/sub token
-2 start /test
+2 start
 2 rm /test/dir
 2 commit
 expect 1:/test/dir/sub:token
@@ -83,7 +83,7 @@
 
 # Multiple events from single transaction don't trigger assert
 1 watch /test token
-2 start /test
+2 start
 2 write /test/1 contents
 2 write /test/2 contents
 2 commit
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/testsuite/12readonly.test
--- a/tools/xenstore/testsuite/12readonly.test  Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/12readonly.test  Fri Sep 23 13:25:01 2005
@@ -13,23 +13,23 @@
 getperm /test
 watch /test token
 unwatch /test token 
-start /
+start
 commit
-start /
+start
 abort
 
 # These don't work
-expect write failed: Read-only file system
+expect write failed: Permission denied
 write /test2 contents
-expect write failed: Read-only file system
+expect write failed: Permission denied
 write /test contents
-expect setperm failed: Read-only file system
+expect setperm failed: Permission denied
 setperm /test 100 NONE
-expect setperm failed: Read-only file system
+expect setperm failed: Permission denied
 setperm /test 100 NONE
-expect shutdown failed: Read-only file system
+expect shutdown failed: Permission denied
 shutdown
-expect introduce failed: Read-only file system
+expect introduce failed: Permission denied
 introduce 1 100 7 /home
 
 # Check that watches work like normal.
diff -r 6aef7d1062bb -r 76af1a1df67c 
tools/xenstore/testsuite/14complexperms.test
--- a/tools/xenstore/testsuite/14complexperms.test      Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/14complexperms.test      Fri Sep 23 13:25:01 2005
@@ -33,14 +33,6 @@
 expect *No such file or directory
 unwatch /dir/file token 
 expect *Permission denied
-start /dir/file
-expect *No such file or directory
-abort
-expect *Permission denied
-start /dir/file
-expect *No such file or directory
-commit
-expect *Permission denied
 introduce 2 100 7 /dir/file
 
 # Now it exists
@@ -73,12 +65,4 @@
 expect *No such file or directory
 unwatch /dir/file token 
 expect *Permission denied
-start /dir/file
-expect *No such file or directory
-abort
-expect *Permission denied
-start /dir/file
-expect *No such file or directory
-commit
-expect *Permission denied
 introduce 2 100 7 /dir/file
diff -r 6aef7d1062bb -r 76af1a1df67c 
tools/xenstore/testsuite/16block-watch-crash.test
--- a/tools/xenstore/testsuite/16block-watch-crash.test Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/testsuite/16block-watch-crash.test Fri Sep 23 13:25:01 2005
@@ -1,13 +1,14 @@
 # Test case where blocked connection gets sent watch.
 
-mkdir /test
-watch /test token
-1 start /test
-# This will block on above
-noackwrite /test/entry contents
-1 write /test/entry2 contents
-1 commit
-readack
-expect /test/entry2:token
-waitwatch
-ackwatch token
+# FIXME: We no longer block connections 
+# mkdir /test
+# watch /test token
+# 1 start
+# # This will block on above
+# noackwrite /test/entry contents
+# 1 write /test/entry2 contents
+# 1 commit
+# readack
+# expect /test/entry2:token
+# waitwatch
+# ackwatch token
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored.h
--- a/tools/xenstore/xenstored.h        Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored.h        Fri Sep 23 13:25:01 2005
@@ -75,7 +75,7 @@
        XSD_ERROR(ENOSYS),
        XSD_ERROR(EROFS),
        XSD_ERROR(EBUSY),
-       XSD_ERROR(ETIMEDOUT),
+       XSD_ERROR(EAGAIN),
        XSD_ERROR(EISCONN),
 };
 struct xsd_sockmsg
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_core.c
--- a/tools/xenstore/xenstored_core.c   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_core.c   Fri Sep 23 13:25:01 2005
@@ -50,10 +50,12 @@
 #include "xenstored_transaction.h"
 #include "xenstored_domain.h"
 #include "xenctrl.h"
+#include "tdb.h"
 
 static bool verbose;
 LIST_HEAD(connections);
 static int tracefd = -1;
+static TDB_CONTEXT *tdb_ctx;
 
 #ifdef TESTING
 static bool failtest = false;
@@ -124,6 +126,23 @@
               "xenstored corruption: connection id %i: err %s: %s",
               conn ? (int)conn->id : -1, strerror(saved_errno), str);
        _exit(2);
+}
+
+TDB_CONTEXT *tdb_context(struct connection *conn)
+{
+       /* conn = NULL used in manual_node at setup. */
+       if (!conn || !conn->transaction)
+               return tdb_ctx;
+       return tdb_transaction_context(conn->transaction);
+}
+
+bool replace_tdb(const char *newname, TDB_CONTEXT *newtdb)
+{
+       if (rename(newname, xs_daemon_tdb()) != 0)
+               return false;
+       tdb_close(tdb_ctx);
+       tdb_ctx = talloc_steal(talloc_autofree_context(), newtdb);
+       return true;
 }
 
 static char *sockmsg_string(enum xsd_sockmsg_type type)
@@ -202,37 +221,6 @@
        write(tracefd, string, strlen(string));
 }
 
-void trace_watch_timeout(const struct connection *conn, const char *node, 
const char *token)
-{
-       char string[64];
-       if (tracefd < 0)
-               return;
-       write(tracefd, "WATCH_TIMEOUT ", strlen("WATCH_TIMEOUT "));
-       sprintf(string, " %p ", conn);
-       write(tracefd, string, strlen(string));
-       write(tracefd, " (", 2);
-       write(tracefd, node, strlen(node));
-       write(tracefd, " ", 1);
-       write(tracefd, token, strlen(token));
-       write(tracefd, ")\n", 2);
-}
-
-static void trace_blocked(const struct connection *conn,
-                         const struct buffered_data *data)
-{
-       char string[64];
-
-       if (tracefd < 0)
-               return;
-
-       write(tracefd, "BLOCKED", strlen("BLOCKED"));
-       sprintf(string, " %p (", conn);
-       write(tracefd, string, strlen(string));
-       write(tracefd, sockmsg_string(data->hdr.msg.type),
-             strlen(sockmsg_string(data->hdr.msg.type)));
-       write(tracefd, ")\n", 2);
-}
-
 void trace(const char *fmt, ...)
 {
        va_list arglist;
@@ -253,7 +241,6 @@
        int ret;
        struct buffered_data *out = conn->out;
 
-       assert(conn->state != BLOCKED);
        if (out->inhdr) {
                if (verbose)
                        xprintf("Writing msg %s (%s) out to %p\n",
@@ -351,24 +338,6 @@
        return max;
 }
 
-/* Read everything from a talloc_open'ed fd. */
-void *read_all(int *fd, unsigned int *size)
-{
-       unsigned int max = 4;
-       int ret;
-       void *buffer = talloc_size(fd, max);
-
-       *size = 0;
-       while ((ret = read(*fd, buffer + *size, max - *size)) > 0) {
-               *size += ret;
-               if (*size == max)
-                       buffer = talloc_realloc_size(fd, buffer, max *= 2);
-       }
-       if (ret < 0)
-               return NULL;
-       return buffer;
-}
-
 static int destroy_fd(void *_fd)
 {
        int *fd = _fd;
@@ -409,42 +378,167 @@
        return child[len] == '/' || child[len] == '\0';
 }
 
-/* Answer never ends in /. */
-char *node_dir_outside_transaction(const char *node)
-{
-       if (streq(node, "/"))
-               return talloc_strdup(node, xs_daemon_store());
-       return talloc_asprintf(node, "%s%s", xs_daemon_store(), node);
-}
-
-static char *node_dir(struct transaction *trans, const char *node)
-{
-       if (!trans || !within_transaction(trans, node))
-               return node_dir_outside_transaction(node);
-       return node_dir_inside_transaction(trans, node);
-}
-
-static char *datafile(const char *dir)
-{
-       return talloc_asprintf(dir, "%s/.data", dir);
-}
-
-static char *node_datafile(struct transaction *trans, const char *node)
-{
-       return datafile(node_dir(trans, node));
-}
-
-static char *permfile(const char *dir)
-{
-       return talloc_asprintf(dir, "%s/.perms", dir);
-}
-
-static char *node_permfile(struct transaction *trans, const char *node)
-{
-       return permfile(node_dir(trans, node));
-}
-
-struct buffered_data *new_buffer(void *ctx)
+/* If it fails, returns NULL and sets errno. */
+static struct node *read_node(struct connection *conn, const char *name)
+{
+       TDB_DATA key, data;
+       u32 *p;
+       struct node *node;
+
+       key.dptr = (void *)name;
+       key.dsize = strlen(name);
+       data = tdb_fetch(tdb_context(conn), key);
+
+       if (data.dptr == NULL) {
+               if (tdb_error(tdb_context(conn)) == TDB_ERR_NOEXIST)
+                       errno = ENOENT;
+               else
+                       errno = EIO;
+               return NULL;
+       }
+
+       node = talloc(name, struct node);
+       node->name = talloc_strdup(node, name);
+       node->parent = NULL;
+       node->tdb = tdb_context(conn);
+       talloc_steal(node, data.dptr);
+
+       /* Datalen, childlen, number of permissions */
+       p = (u32 *)data.dptr;
+       node->num_perms = p[0];
+       node->datalen = p[1];
+       node->childlen = p[2];
+
+       /* Permissions are struct xs_permissions. */
+       node->perms = (void *)&p[3];
+       /* Data is binary blob (usually ascii, no nul). */
+       node->data = node->perms + node->num_perms;
+       /* Children is strings, nul separated. */
+       node->children = node->data + node->datalen;
+
+       return node;
+}
+
+static bool write_node(struct connection *conn, const struct node *node)
+{
+       TDB_DATA key, data;
+       void *p;
+
+       key.dptr = (void *)node->name;
+       key.dsize = strlen(node->name);
+
+       data.dsize = 3*sizeof(u32)
+               + node->num_perms*sizeof(node->perms[0])
+               + node->datalen + node->childlen;
+       data.dptr = talloc_size(node, data.dsize);
+       ((u32 *)data.dptr)[0] = node->num_perms;
+       ((u32 *)data.dptr)[1] = node->datalen;
+       ((u32 *)data.dptr)[2] = node->childlen;
+       p = data.dptr + 3 * sizeof(u32);
+
+       memcpy(p, node->perms, node->num_perms*sizeof(node->perms[0]));
+       p += node->num_perms*sizeof(node->perms[0]);
+       memcpy(p, node->data, node->datalen);
+       p += node->datalen;
+       memcpy(p, node->children, node->childlen);
+
+       /* TDB should set errno, but doesn't even set ecode AFAICT. */
+       if (tdb_store(tdb_context(conn), key, data, TDB_REPLACE) != 0) {
+               errno = ENOSPC;
+               return false;
+       }
+       return true;
+}
+
+static enum xs_perm_type perm_for_conn(struct connection *conn,
+                                      struct xs_permissions *perms,
+                                      unsigned int num)
+{
+       unsigned int i;
+       enum xs_perm_type mask = XS_PERM_READ|XS_PERM_WRITE|XS_PERM_OWNER;
+
+       if (!conn->can_write)
+               mask &= ~XS_PERM_WRITE;
+
+       /* Owners and tools get it all... */
+       if (!conn->id || perms[0].id == conn->id)
+               return (XS_PERM_READ|XS_PERM_WRITE|XS_PERM_OWNER) & mask;
+
+       for (i = 1; i < num; i++)
+               if (perms[i].id == conn->id)
+                       return perms[i].perms & mask;
+
+       return perms[0].perms & mask;
+}
+
+static char *get_parent(const char *node)
+{
+       char *slash = strrchr(node + 1, '/');
+       if (!slash)
+               return talloc_strdup(node, "/");
+       return talloc_asprintf(node, "%.*s", (int)(slash - node), node);
+}
+
+/* What do parents say? */
+static enum xs_perm_type ask_parents(struct connection *conn, const char *name)
+{
+       struct node *node;
+
+       do {
+               name = get_parent(name);
+               node = read_node(conn, name);
+               if (node)
+                       break;
+       } while (!streq(name, "/"));
+
+       /* No permission at root?  We're in trouble. */
+       if (!node)
+               corrupt(conn, "No permissions file at root");
+
+       return perm_for_conn(conn, node->perms, node->num_perms);
+}
+
+/* We have a weird permissions system.  You can allow someone into a
+ * specific node without allowing it in the parents.  If it's going to
+ * fail, however, we don't want the errno to indicate any information
+ * about the node. */
+static int errno_from_parents(struct connection *conn, const char *node,
+                             int errnum, enum xs_perm_type perm)
+{
+       /* We always tell them about memory failures. */
+       if (errnum == ENOMEM)
+               return errnum;
+
+       if (ask_parents(conn, node) & perm)
+               return errnum;
+       return EACCES;
+}
+
+/* If it fails, returns NULL and sets errno. */
+struct node *get_node(struct connection *conn,
+                     const char *name,
+                     enum xs_perm_type perm)
+{
+       struct node *node;
+
+       if (!name || !is_valid_nodename(name)) {
+               errno = EINVAL;
+               return NULL;
+       }
+       node = read_node(conn, name);
+       /* If we don't have permission, we don't have node. */
+       if (node) {
+               if ((perm_for_conn(conn, node->perms, node->num_perms) & perm)
+                   != perm)
+                       node = NULL;
+       }
+       /* Clean up errno if they weren't supposed to know. */
+       if (!node) 
+               errno = errno_from_parents(conn, name, errno, perm);
+       return node;
+}
+
+static struct buffered_data *new_buffer(void *ctx)
 {
        struct buffered_data *data;
 
@@ -457,7 +551,8 @@
 }
 
 /* Return length of string (including nul) at this offset. */
-unsigned int get_string(const struct buffered_data *data, unsigned int offset)
+static unsigned int get_string(const struct buffered_data *data,
+                              unsigned int offset)
 {
        const char *nul;
 
@@ -508,7 +603,6 @@
                conn->waiting_reply = bdata;
        } else
                conn->out = bdata;
-       assert(conn->state != BLOCKED);
        conn->state = BUSY;
 }
 
@@ -567,29 +661,6 @@
        return in->buffer;
 }
 
-/* If it fails, returns NULL and sets errno. */
-static struct xs_permissions *get_perms(const char *dir, unsigned int *num)
-{
-       unsigned int size;
-       char *strings;
-       struct xs_permissions *ret;
-       int *fd;
-
-       fd = talloc_open(permfile(dir), O_RDONLY, 0);
-       if (!fd)
-               return NULL;
-       strings = read_all(fd, &size);
-       if (!strings)
-               return NULL;
-
-       *num = xs_count_strings(strings, size);
-       ret = talloc_array(dir, struct xs_permissions, *num);
-       if (!xs_strings_to_perms(ret, *num, strings))
-               corrupt(NULL, "Permissions corrupt for %s", dir);
-
-       return ret;
-}
-
 static char *perms_to_strings(const void *ctx,
                              struct xs_permissions *perms, unsigned int num,
                              unsigned int *len)
@@ -608,173 +679,6 @@
                *len += strlen(buffer) + 1;
        }
        return strings;
-}
-
-/* Destroy this, and its children, and its children's children. */
-int destroy_path(void *path)
-{
-       DIR *dir;
-       struct dirent *dirent;
-
-       dir = opendir(path);
-       if (!dir) {
-               if (unlink(path) == 0 || errno == ENOENT)
-                       return 0;
-               corrupt(NULL, "Destroying path %s", path);
-       }
-
-       while ((dirent = readdir(dir)) != NULL) {
-               char fullpath[strlen(path) + 1 + strlen(dirent->d_name) + 1];
-               sprintf(fullpath, "%s/%s", (char *)path, dirent->d_name);
-               if (!streq(dirent->d_name,".") && !streq(dirent->d_name,".."))
-                       destroy_path(fullpath);
-       }
-       closedir(dir);
-       if (rmdir(path) != 0)
-               corrupt(NULL, "Destroying directory %s", path);
-       return 0;
-}
-
-/* Create a self-destructing temporary path */
-static char *temppath(const char *path)
-{
-       char *tmppath = talloc_asprintf(path, "%s.tmp", path);
-       talloc_set_destructor(tmppath, destroy_path);
-       return tmppath;
-}
-
-/* Create a self-destructing temporary file */
-static char *tempfile(const char *path, void *contents, unsigned int len)
-{
-       int *fd;
-       char *tmppath = temppath(path);
-
-       fd = talloc_open(tmppath, O_WRONLY|O_CREAT|O_EXCL, 0640);
-       if (!fd)
-               return NULL;
-       if (!xs_write_all(*fd, contents, len))
-               return NULL;
-
-       return tmppath;
-}
-
-static int destroy_opendir(void *_dir)
-{
-       DIR **dir = _dir;
-       closedir(*dir);
-       return 0;
-}
-
-/* Return a pointer to a DIR*, self-closing and attached to this pathname. */
-DIR **talloc_opendir(const char *pathname)
-{
-       DIR **dir;
-
-       dir = talloc(pathname, DIR *);
-       *dir = opendir(pathname);
-       if (!*dir) {
-               int saved_errno = errno;
-               talloc_free(dir);
-               errno = saved_errno;
-               return NULL;
-       }
-       talloc_set_destructor(dir, destroy_opendir);
-       return dir;
-}
-
-/* We assume rename() doesn't fail on moves in same dir. */
-static void commit_tempfile(const char *path)
-{
-       char realname[strlen(path) + 1];
-       unsigned int len = strrchr(path, '.') - path;
-
-       memcpy(realname, path, len);
-       realname[len] = '\0';
-       if (rename(path, realname) != 0)
-               corrupt(NULL, "Committing %s", realname);
-       talloc_set_destructor(path, NULL);
-}
-
-static bool set_perms(struct transaction *transaction,
-                     const char *node,
-                     struct xs_permissions *perms, unsigned int num)
-{
-       unsigned int len;
-       char *permpath, *strings;
-
-       strings = perms_to_strings(node, perms, num, &len);
-       if (!strings)
-               return false;
-
-       /* Create then move. */
-       permpath = tempfile(node_permfile(transaction, node), strings, len);
-       if (!permpath)
-               return false;
-
-       commit_tempfile(permpath);
-       return true;
-}
-
-static char *get_parent(const char *node)
-{
-       char *slash = strrchr(node + 1, '/');
-       if (!slash)
-               return talloc_strdup(node, "/");
-       return talloc_asprintf(node, "%.*s", (int)(slash - node), node);
-}
-
-static enum xs_perm_type perm_for_id(domid_t id,
-                                    struct xs_permissions *perms,
-                                    unsigned int num)
-{
-       unsigned int i;
-
-       /* Owners and tools get it all... */
-       if (!id || perms[0].id == id)
-               return XS_PERM_READ|XS_PERM_WRITE|XS_PERM_OWNER;
-
-       for (i = 1; i < num; i++)
-               if (perms[i].id == id)
-                       return perms[i].perms;
-
-       return perms[0].perms;
-}
-
-/* What do parents say? */
-static enum xs_perm_type ask_parents(struct connection *conn,
-                                    const char *node)
-{
-       struct xs_permissions *perms;
-       unsigned int num;
-
-       do {
-               node = get_parent(node);
-               perms = get_perms(node_dir(conn->transaction, node), &num);
-               if (perms)
-                       break;
-       } while (!streq(node, "/"));
-
-       /* No permission at root?  We're in trouble. */
-       if (!perms)
-               corrupt(conn, "No permissions file at root");
-
-       return perm_for_id(conn->id, perms, num);
-}
-
-/* We have a weird permissions system.  You can allow someone into a
- * specific node without allowing it in the parents.  If it's going to
- * fail, however, we don't want the errno to indicate any information
- * about the node. */
-static int errno_from_parents(struct connection *conn, const char *node,
-                             int errnum)
-{
-       /* We always tell them about memory failures. */
-       if (errnum == ENOMEM)
-               return errnum;
-
-       if (ask_parents(conn, node) & XS_PERM_READ)
-               return errnum;
-       return EACCES;
 }
 
 char *canonicalize(struct connection *conn, const char *node)
@@ -789,46 +693,6 @@
        return (char *)node;
 }
 
-bool check_node_perms(struct connection *conn, const char *node,
-                     enum xs_perm_type perm)
-{
-       struct xs_permissions *perms;
-       unsigned int num;
-
-       if (!node || !is_valid_nodename(node)) {
-               errno = EINVAL;
-               return false;
-       }
-
-       if (!conn->can_write && (perm & XS_PERM_WRITE)) {
-               errno = EROFS;
-               return false;
-       }
-
-       perms = get_perms(node_dir(conn->transaction, node), &num);
-
-       if (perms) {
-               if (perm_for_id(conn->id, perms, num) & perm)
-                       return true;
-               errno = EACCES;
-               return false;
-       }
-
-       /* If it's OK not to exist, we consult parents. */
-       if (errno == ENOENT && (perm & XS_PERM_ENOENT_OK)) {
-               if (ask_parents(conn, node) & perm)
-                       return true;
-               /* Parents say they should not know. */
-               errno = EACCES;
-               return false;
-       }
-
-       /* They might not have permission to even *see* this node, in
-        * which case we return EACCES even if it's ENOENT or EIO. */
-       errno = errno_from_parents(conn, node, errno);
-       return false;
-}
-
 bool check_event_node(const char *node)
 {
        if (!node || !strstarts(node, "@")) {
@@ -838,142 +702,144 @@
        return true;
 }
 
-static void send_directory(struct connection *conn, const char *node)
-{
-       char *path, *reply;
-       unsigned int reply_len = 0;
-       DIR **dir;
-       struct dirent *dirent;
-
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_READ)) {
+static void send_directory(struct connection *conn, const char *name)
+{
+       struct node *node;
+
+       name = canonicalize(conn, name);
+       node = get_node(conn, name, XS_PERM_READ);
+       if (!node) {
                send_error(conn, errno);
                return;
        }
 
-       path = node_dir(conn->transaction, node);
-       dir = talloc_opendir(path);
-       if (!dir) {
+       send_reply(conn, XS_DIRECTORY, node->children, node->childlen);
+}
+
+static void do_read(struct connection *conn, const char *name)
+{
+       struct node *node;
+
+       name = canonicalize(conn, name);
+       node = get_node(conn, name, XS_PERM_READ);
+       if (!node) {
                send_error(conn, errno);
                return;
        }
 
-       reply = talloc_strdup(node, "");
-       while ((dirent = readdir(*dir)) != NULL) {
-               int len = strlen(dirent->d_name) + 1;
-
-               if (!valid_chars(dirent->d_name))
-                       continue;
-
-               reply = talloc_realloc(path, reply, char, reply_len + len);
-               strcpy(reply + reply_len, dirent->d_name);
-               reply_len += len;
-       }
-
-       send_reply(conn, XS_DIRECTORY, reply, reply_len);
-}
-
-static void do_read(struct connection *conn, const char *node)
-{
-       char *value;
-       unsigned int size;
-       int *fd;
-
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_READ)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       fd = talloc_open(node_datafile(conn->transaction, node), O_RDONLY, 0);
-       if (!fd) {
-               /* Data file doesn't exist?  We call that a directory */
-               if (errno == ENOENT)
-                       errno = EISDIR;
-               send_error(conn, errno);
-               return;
-       }
-
-       value = read_all(fd, &size);
-       if (!value)
-               send_error(conn, errno);
-       else
-               send_reply(conn, XS_READ, value, size);
-}
-
-/* Commit this directory, eg. comitting a/b.tmp/c causes a/b.tmp -> a.b */
-static bool commit_dir(char *dir)
-{
-       char *dot, *slash, *dest;
-
-       dot = strrchr(dir, '.');
-       slash = strchr(dot, '/');
-       if (slash)
-               *slash = '\0';
-
-       dest = talloc_asprintf(dir, "%.*s", (int)(dot - dir), dir);
-       return rename(dir, dest) == 0;
-}
-
-/* Create a temporary directory.  Put data in it (if data != NULL) */
-static char *tempdir(struct connection *conn,
-                    const char *node, void *data, unsigned int datalen)
-{
-       struct xs_permissions *perms;
-       char *permstr;
-       unsigned int num, len;
-       int *fd;
-       char *dir;
-
-       dir = temppath(node_dir(conn->transaction, node));
-       if (mkdir(dir, 0750) != 0) {
-               if (errno != ENOENT)
+       send_reply(conn, XS_READ, node->data, node->datalen);
+}
+
+static void delete_node_single(struct connection *conn, struct node *node)
+{
+       TDB_DATA key;
+
+       key.dptr = (void *)node->name;
+       key.dsize = strlen(node->name);
+
+       if (tdb_delete(tdb_context(conn), key) != 0)
+               corrupt(conn, "Could not delete '%s'", node->name);
+}
+
+/* Must not be / */
+static char *basename(const char *name)
+{
+       return strrchr(name, '/') + 1;
+}
+
+static struct node *construct_node(struct connection *conn, const char *name)
+{
+       const char *base;
+       unsigned int baselen;
+       struct node *parent, *node;
+       char *children, *parentname = get_parent(name);
+
+       /* If parent doesn't exist, create it. */
+       parent = read_node(conn, parentname);
+       if (!parent)
+               parent = construct_node(conn, parentname);
+       if (!parent)
+               return NULL;
+       
+       /* Add child to parent. */
+       base = basename(name);
+       baselen = strlen(base) + 1;
+       children = talloc_array(name, char, parent->childlen + baselen);
+       memcpy(children, parent->children, parent->childlen);
+       memcpy(children + parent->childlen, base, baselen);
+       parent->children = children;
+       parent->childlen += baselen;
+
+       /* Allocate node */
+       node = talloc(name, struct node);
+       node->tdb = tdb_context(conn);
+       node->name = talloc_strdup(node, name);
+
+       /* Inherit permissions, except domains own what they create */
+       node->num_perms = parent->num_perms;
+       node->perms = talloc_memdup(node, parent->perms,
+                                   node->num_perms * sizeof(node->perms[0]));
+       if (conn->id)
+               node->perms[0].id = conn->id;
+
+       /* No children, no data */
+       node->children = node->data = NULL;
+       node->childlen = node->datalen = 0;
+       node->parent = parent;
+       return node;
+}
+
+static int destroy_node(void *_node)
+{
+       struct node *node = _node;
+       TDB_DATA key;
+
+       if (streq(node->name, "/"))
+               corrupt(NULL, "Destroying root node!");
+
+       key.dptr = (void *)node->name;
+       key.dsize = strlen(node->name);
+
+       tdb_delete(node->tdb, key);
+       return 0;
+}
+
+/* Be careful: create heirarchy, put entry in existing parent *last*.
+ * This helps fsck if we die during this. */
+static struct node *create_node(struct connection *conn, 
+                               const char *name,
+                               void *data, unsigned int datalen)
+{
+       struct node *node, *i;
+
+       node = construct_node(conn, name);
+       if (!node)
+               return NULL;
+
+       node->data = data;
+       node->datalen = datalen;
+
+       /* We write out the nodes down, setting destructor in case
+        * something goes wrong. */
+       for (i = node; i; i = i->parent) {
+               if (!write_node(conn, i))
                        return NULL;
-
-               dir = tempdir(conn, get_parent(node), NULL, 0);
-               if (!dir)
-                       return NULL;
-
-               dir = talloc_asprintf(dir, "%s%s", dir, strrchr(node, '/'));
-               if (mkdir(dir, 0750) != 0)
-                       return NULL;
-               talloc_set_destructor(dir, destroy_path);
-       }
-
-       perms = get_perms(get_parent(dir), &num);
-       assert(perms);
-       /* Domains own what they create. */
-       if (conn->id)
-               perms->id = conn->id;
-
-       permstr = perms_to_strings(dir, perms, num, &len);
-       fd = talloc_open(permfile(dir), O_WRONLY|O_CREAT|O_EXCL, 0640);
-       if (!fd || !xs_write_all(*fd, permstr, len))
-               return NULL;
-
-       if (data) {
-               char *datapath = datafile(dir);
-
-               fd = talloc_open(datapath, O_WRONLY|O_CREAT|O_EXCL, 0640);
-               if (!fd || !xs_write_all(*fd, data, datalen))
-                       return NULL;
-       }
-       return dir;
-}
-
-static bool node_exists(struct connection *conn, const char *node)
-{
-       struct stat st;
-
-       return lstat(node_dir(conn->transaction, node), &st) == 0;
+               talloc_set_destructor(i, destroy_node);
+       }
+
+       /* OK, now remove destructors so they stay around */
+       for (i = node; i; i = i->parent)
+               talloc_set_destructor(i, NULL);
+       return node;
 }
 
 /* path, data... */
 static void do_write(struct connection *conn, struct buffered_data *in)
 {
        unsigned int offset, datalen;
+       struct node *node;
        char *vec[1] = { NULL }; /* gcc4 + -W + -Werror fucks code. */
-       char *node, *tmppath;
+       char *name;
 
        /* Extra "strings" can be created by binary data. */
        if (get_strings(in, vec, ARRAY_SIZE(vec)) < ARRAY_SIZE(vec)) {
@@ -981,99 +847,115 @@
                return;
        }
 
-       node = canonicalize(conn, vec[0]);
-       if (!within_transaction(conn->transaction, node)) {
-               send_error(conn, EROFS);
-               return;
-       }
-
-       if (transaction_block(conn, node))
-               return;
-
        offset = strlen(vec[0]) + 1;
        datalen = in->used - offset;
 
-       if (!check_node_perms(conn, node, XS_PERM_WRITE|XS_PERM_ENOENT_OK)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       if (!node_exists(conn, node)) {
-               char *dir;
-
-               /* Does not exist... */
+       name = canonicalize(conn, vec[0]);
+       node = get_node(conn, name, XS_PERM_WRITE);
+       if (!node) {
+               /* No permissions, invalid input? */
                if (errno != ENOENT) {
                        send_error(conn, errno);
                        return;
                }
-
-               dir = tempdir(conn, node, in->buffer + offset, datalen);
-               if (!dir || !commit_dir(dir)) {
+               node = create_node(conn, name, in->buffer + offset, datalen);
+               if (!node) {
                        send_error(conn, errno);
                        return;
                }
-               
        } else {
-               /* Exists... */
-               tmppath = tempfile(node_datafile(conn->transaction, node),
-                                  in->buffer + offset, datalen);
-               if (!tmppath) {
+               node->data = in->buffer + offset;
+               node->datalen = datalen;
+               if (!write_node(conn, node)){
                        send_error(conn, errno);
                        return;
                }
-
-               commit_tempfile(tmppath);
-       }
-
-       add_change_node(conn->transaction, node, false);
-       fire_watches(conn, node, false);
+       }
+
+       add_change_node(conn->transaction, name, false);
+       fire_watches(conn, name, false);
        send_ack(conn, XS_WRITE);
 }
 
-static void do_mkdir(struct connection *conn, const char *node)
-{
-       char *dir;
-
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_WRITE|XS_PERM_ENOENT_OK)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       if (!within_transaction(conn->transaction, node)) {
-               send_error(conn, EROFS);
-               return;
-       }
-
-       if (transaction_block(conn, node))
-               return;
+static void do_mkdir(struct connection *conn, const char *name)
+{
+       struct node *node;
+
+       name = canonicalize(conn, name);
+       node = get_node(conn, name, XS_PERM_WRITE);
 
        /* If it already exists, fine. */
-       if (node_exists(conn, node)) {
-               send_ack(conn, XS_MKDIR);
-               return;
-       }
-
-       dir = tempdir(conn, node, NULL, 0);
-       if (!dir || !commit_dir(dir)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       add_change_node(conn->transaction, node, false);
-       fire_watches(conn, node, false);
+       if (!node) {
+               /* No permissions? */
+               if (errno != ENOENT) {
+                       send_error(conn, errno);
+                       return;
+               }
+               node = create_node(conn, name, NULL, 0);
+               if (!node) {
+                       send_error(conn, errno);
+                       return;
+               }
+               add_change_node(conn->transaction, name, false);
+               fire_watches(conn, name, false);
+       }
        send_ack(conn, XS_MKDIR);
 }
 
-static void do_rm(struct connection *conn, const char *node)
-{
-       char *tmppath, *path;
-
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_WRITE)) {
+static void delete_node(struct connection *conn, struct node *node)
+{
+       unsigned int i;
+
+       /* Delete self, then delete children.  If something goes wrong,
+        * consistency check will clean up this way. */
+       delete_node_single(conn, node);
+
+       /* Delete children, too. */
+       for (i = 0; i < node->childlen; i += strlen(node->children+i) + 1) {
+               struct node *child;
+
+               child = read_node(conn, 
+                                 talloc_asprintf(node, "%s/%s", node->name,
+                                                 node->children + i));
+               if (!child)
+                       corrupt(conn, "No child '%s' found", child);
+               delete_node(conn, child);
+       }
+}
+
+/* Delete memory using memmove. */
+static void memdel(void *mem, unsigned off, unsigned len, unsigned total)
+{
+       memmove(mem + off, mem + off + len, total - off - len);
+}
+
+static bool delete_child(struct connection *conn,
+                        struct node *node, const char *childname)
+{
+       unsigned int i;
+
+       for (i = 0; i < node->childlen; i += strlen(node->children+i) + 1) {
+               if (streq(node->children+i, childname)) {
+                       memdel(node->children, i, strlen(childname) + 1,
+                              node->childlen);
+                       node->childlen -= strlen(childname) + 1;
+                       return write_node(conn, node);
+               }
+       }
+       corrupt(conn, "Can't find child '%s' in %s", childname, node->name);
+}
+
+static void do_rm(struct connection *conn, const char *name)
+{
+       struct node *node, *parent;
+
+       name = canonicalize(conn, name);
+       node = get_node(conn, name, XS_PERM_WRITE);
+       if (!node) {
                /* Didn't exist already?  Fine, if parent exists. */
                if (errno == ENOENT) {
-                       if (node_exists(conn, get_parent(node))) {
+                       node = read_node(conn, get_parent(name));
+                       if (node) {
                                send_ack(conn, XS_RM);
                                return;
                        }
@@ -1084,53 +966,43 @@
                return;
        }
 
-       if (!within_transaction(conn->transaction, node)) {
-               send_error(conn, EROFS);
-               return;
-       }
-
-       if (transaction_block(conn, node))
-               return;
-
-       if (streq(node, "/")) {
+       if (streq(name, "/")) {
                send_error(conn, EINVAL);
                return;
        }
 
-       /* We move the directory to temporary name, destructor cleans up. */
-       path = node_dir(conn->transaction, node);
-       tmppath = talloc_asprintf(node, "%s.tmp", path);
-       talloc_set_destructor(tmppath, destroy_path);
-
-       if (rename(path, tmppath) != 0) {
+       /* Delete from parent first, then if something explodes fsck cleans. */
+       parent = read_node(conn, get_parent(name));
+       if (!parent) {
+               send_error(conn, EINVAL);
+               return;
+       }
+
+       if (!delete_child(conn, parent, basename(name))) {
+               send_error(conn, EINVAL);
+               return;
+       }
+
+       delete_node(conn, node);
+       add_change_node(conn->transaction, name, true);
+       fire_watches(conn, name, true);
+       send_ack(conn, XS_RM);
+}
+
+static void do_get_perms(struct connection *conn, const char *name)
+{
+       struct node *node;
+       char *strings;
+       unsigned int len;
+
+       name = canonicalize(conn, name);
+       node = get_node(conn, name, XS_PERM_READ);
+       if (!node) {
                send_error(conn, errno);
                return;
        }
 
-       add_change_node(conn->transaction, node, true);
-       fire_watches(conn, node, true);
-       send_ack(conn, XS_RM);
-}
-
-static void do_get_perms(struct connection *conn, const char *node)
-{
-       struct xs_permissions *perms;
-       char *strings;
-       unsigned int len, num;
-
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_READ)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       perms = get_perms(node_dir(conn->transaction, node), &num);
-       if (!perms) {
-               send_error(conn, errno);
-               return;
-       }
-
-       strings = perms_to_strings(node, perms, num, &len);
+       strings = perms_to_strings(node, node->perms, node->num_perms, &len);
        if (!strings)
                send_error(conn, errno);
        else
@@ -1140,8 +1012,8 @@
 static void do_set_perms(struct connection *conn, struct buffered_data *in)
 {
        unsigned int num;
-       char *node, *permstr;
-       struct xs_permissions *perms;
+       char *name, *permstr;
+       struct node *node;
 
        num = xs_count_strings(in->buffer, in->used);
        if (num < 2) {
@@ -1150,37 +1022,30 @@
        }
 
        /* First arg is node name. */
-       node = canonicalize(conn, in->buffer);
+       name = canonicalize(conn, in->buffer);
        permstr = in->buffer + strlen(in->buffer) + 1;
        num--;
 
-       if (!within_transaction(conn->transaction, node)) {
-               send_error(conn, EROFS);
-               return;
-       }
-
-       if (transaction_block(conn, node))
-               return;
-
        /* We must own node to do this (tools can do this too). */
-       if (!check_node_perms(conn, node, XS_PERM_WRITE|XS_PERM_OWNER)) {
+       node = get_node(conn, name, XS_PERM_WRITE|XS_PERM_OWNER);
+       if (!node) {
                send_error(conn, errno);
                return;
        }
 
-       perms = talloc_array(node, struct xs_permissions, num);
-       if (!xs_strings_to_perms(perms, num, permstr)) {
+       node->perms = talloc_array(node, struct xs_permissions, num);
+       node->num_perms = num;
+       if (!xs_strings_to_perms(node->perms, num, permstr)) {
                send_error(conn, errno);
                return;
        }
-
-       if (!set_perms(conn->transaction, node, perms, num)) {
+       if (!write_node(conn, node)) {
                send_error(conn, errno);
                return;
        }
 
-       add_change_node(conn->transaction, node, false);
-       fire_watches(conn, node, false);
+       add_change_node(conn->transaction, name, false);
+       fire_watches(conn, name, false);
        send_ack(conn, XS_SET_PERMS);
 }
 
@@ -1221,12 +1086,8 @@
        case XS_SHUTDOWN:
                /* FIXME: Implement gentle shutdown too. */
                /* Only tools can do this. */
-               if (conn->id != 0) {
+               if (conn->id != 0 || !conn->can_write) {
                        send_error(conn, EACCES);
-                       break;
-               }
-               if (!conn->can_write) {
-                       send_error(conn, EROFS);
                        break;
                }
                send_ack(conn, XS_SHUTDOWN);
@@ -1263,7 +1124,7 @@
                break;
 
        case XS_TRANSACTION_START:
-               do_transaction_start(conn, onearg(in));
+               do_transaction_start(conn, in);
                break;
 
        case XS_TRANSACTION_END:
@@ -1309,6 +1170,8 @@
        /* For simplicity, we kill the connection on OOM. */
        talloc_set_fail_handler(out_of_mem, &talloc_fail);
        if (setjmp(talloc_fail)) {
+               /* Free in before conn, in case it needs something. */
+               talloc_free(in);
                talloc_free(conn);
                goto end;
        }
@@ -1330,16 +1193,8 @@
        conn->in = new_buffer(conn);
        process_message(conn, in);
 
-       if (conn->state == BLOCKED) {
-               /* Blocked by transaction: queue for re-xmit. */
-               talloc_free(conn->in);
-               conn->in = in;
-               in = NULL;
-               trace_blocked(conn, conn->in);
-       }
-
+       talloc_free(in);
 end:
-       talloc_free(in);
        talloc_set_fail_handler(NULL, NULL);
        if (talloc_total_blocks(NULL)
            != talloc_total_blocks(talloc_autofree_context()) + 1) {
@@ -1350,7 +1205,7 @@
 
 /* Errors in reading or allocating here mean we get out of sync, so we
  * drop the whole client connection. */
-void handle_input(struct connection *conn)
+static void handle_input(struct connection *conn)
 {
        int bytes;
        struct buffered_data *in;
@@ -1402,39 +1257,10 @@
        talloc_free(conn);
 }
 
-void handle_output(struct connection *conn)
+static void handle_output(struct connection *conn)
 {
        if (!write_message(conn))
                talloc_free(conn);
-}
-
-/* If a transaction has ended, see if we can unblock any connections. */
-static void unblock_connections(void)
-{
-       struct connection *i, *tmp;
-
-       list_for_each_entry_safe(i, tmp, &connections, list) {
-               switch (i->state) {
-               case BLOCKED:
-                       if (!transaction_covering_node(i->blocked_by)) {
-                               talloc_free(i->blocked_by);
-                               i->blocked_by = NULL;
-                               i->state = OK;
-                               consider_message(i);
-                       }
-                       break;
-               case BUSY:
-               case OK:
-                       break;
-               }
-       }
-
-       /* To balance bias, move first entry to end. */
-       if (!list_empty(&connections)) {
-               i = list_top(&connections, struct connection, list);
-               list_del(&i->list);
-               list_add_tail(&i->list, &connections);
-       }
 }
 
 struct connection *new_connection(connwritefn_t *write, connreadfn_t *read)
@@ -1451,7 +1277,6 @@
                return NULL;
 
        new->state = OK;
-       new->blocked_by = NULL;
        new->out = new->waiting_reply = NULL;
        new->waiting_for_ack = NULL;
        new->fd = -1;
@@ -1504,25 +1329,9 @@
                close(fd);
 }
 
-/* Calc timespan from now to absolute time. */
-static void time_relative_to_now(struct timeval *tv)
-{
-       struct timeval now;
-
-       gettimeofday(&now, NULL);
-       if (timercmp(&now, tv, >))
-               timerclear(tv);
-       else {
-               tv->tv_sec -= now.tv_sec;
-               if (now.tv_usec > tv->tv_usec) {
-                       tv->tv_sec--;
-                       tv->tv_usec += 1000000;
-               }
-               tv->tv_usec -= now.tv_usec;
-       }
-}
-
 #ifdef TESTING
+/* Valgrind can check our writes better if we don't use mmap */
+#define TDB_FLAGS TDB_NOMMAP
 /* Useful for running under debugger. */
 void dump_connection(void)
 {
@@ -1532,13 +1341,10 @@
                printf("Connection %p:\n", i);
                printf("    state = %s\n",
                       i->state == OK ? "OK"
-                      : i->state == BLOCKED ? "BLOCKED"
                       : i->state == BUSY ? "BUSY"
                       : "INVALID");
                if (i->id)
                        printf("    id = %i\n", i->id);
-               if (i->blocked_by)
-                       printf("    blocked on = %s\n", i->blocked_by);
                if (!i->in->inhdr || i->in->used)
                        printf("    got %i bytes of %s\n",
                               i->in->used, i->in->inhdr ? "header" : "data");
@@ -1559,44 +1365,53 @@
                dump_watches(i);
        }
 }
+#else
+#define TDB_FLAGS 0
 #endif
 
+/* We create initial nodes manually. */
+static void manual_node(const char *name, const char *child)
+{
+       struct node *node;
+       struct xs_permissions perms = { .id = 0, .perms = XS_PERM_READ };
+
+       node = talloc(NULL, struct node);
+       node->name = name;
+       node->perms = &perms;
+       node->num_perms = 1;
+       node->data = NULL;
+       node->datalen = 0;
+       node->children = (char *)child;
+       if (child)
+               node->childlen = strlen(child) + 1;
+       else
+               node->childlen = 0;
+
+       if (!write_node(NULL, node))
+               barf_perror("Could not create initial node %s", name);
+       talloc_free(node);
+}
+
+#
+
 static void setup_structure(void)
 {
-       struct xs_permissions perms = { .id = 0, .perms = XS_PERM_READ };
-       char *root, *dir, *permfile;
-
-       /* Create root directory, with permissions. */
-       if (mkdir(xs_daemon_store(), 0750) != 0) {
-               if (errno != EEXIST)
-                       barf_perror("Could not create root %s",
-                                   xs_daemon_store());
-               return;
-       }
-       root = talloc_strdup(talloc_autofree_context(), "/");
-       if (!set_perms(NULL, root, &perms, 1))
-               barf_perror("Could not create permissions in root");
-
-       /* Create tool directory, with xenstored subdir. */
-       dir = talloc_asprintf(root, "%s/%s", xs_daemon_store(), "tool");
-       if (mkdir(dir, 0750) != 0)
-               barf_perror("Making dir %s", dir);
-       
-       permfile = talloc_strdup(root, "/tool");
-       if (!set_perms(NULL, permfile, &perms, 1))
-               barf_perror("Could not create permissions on %s", permfile);
-
-       dir = talloc_asprintf(root, "%s/%s", dir, "xenstored");
-       if (mkdir(dir, 0750) != 0)
-               barf_perror("Making dir %s", dir);
-       
-       permfile = talloc_strdup(root, "/tool/xenstored");
-       if (!set_perms(NULL, permfile, &perms, 1))
-               barf_perror("Could not create permissions on %s", permfile);
-       talloc_free(root);
-       if (mkdir(xs_daemon_transactions(), 0750) != 0)
-               barf_perror("Could not create transaction dir %s",
-                           xs_daemon_transactions());
+       char *tdbname;
+       tdbname = talloc_strdup(talloc_autofree_context(), xs_daemon_tdb());
+       tdb_ctx = tdb_open(tdbname, 0, TDB_FLAGS, O_RDWR, 0);
+
+       if (!tdb_ctx) {
+               tdb_ctx = tdb_open(tdbname, 7919, TDB_FLAGS, O_RDWR|O_CREAT,
+                                  0640);
+               if (!tdb_ctx)
+                       barf_perror("Could not create tdb file %s", tdbname);
+
+               manual_node("/", "tool");
+               manual_node("/tool", "xenstored");
+               manual_node("/tool/xenstored", NULL);
+       }
+
+       /* FIXME: Fsck */
 }
 
 static void write_pidfile(const char *pidfile)
@@ -1759,17 +1574,8 @@
        /* FIXME: Rewrite so noone can starve. */
        for (;;) {
                struct connection *i;
-               struct timeval *tvp = NULL, tv;
-
-               timerclear(&tv);
-               shortest_transaction_timeout(&tv);
-               shortest_watch_ack_timeout(&tv);
-               if (timerisset(&tv)) {
-                       time_relative_to_now(&tv);
-                       tvp = &tv;
-               }
-
-               if (select(max+1, &inset, &outset, NULL, tvp) < 0) {
+
+               if (select(max+1, &inset, &outset, NULL, NULL) < 0) {
                        if (errno == EINTR)
                                continue;
                        barf_perror("Select failed");
@@ -1818,14 +1624,6 @@
                        }
                }
 
-               if (tvp) {
-                       check_transaction_timeout();
-                       check_watch_ack_timeout();
-               }
-
-               /* If transactions ended, we might be able to do more work. */
-               unblock_connections();
-
                max = initialize_set(&inset, &outset, *sock, *ro_sock,
                                     event_fd);
        }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_core.h
--- a/tools/xenstore/xenstored_core.h   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_core.h   Fri Sep 23 13:25:01 2005
@@ -28,6 +28,7 @@
 #include "xs_lib.h"
 #include "xenstored.h"
 #include "list.h"
+#include "tdb.h"
 
 struct buffered_data
 {
@@ -49,8 +50,6 @@
 
 enum state
 {
-       /* Blocked by transaction. */
-       BLOCKED,
        /* Doing action, not listening */
        BUSY,
        /* Completed */
@@ -69,9 +68,6 @@
 
        /* Blocked on transaction?  Busy? */
        enum state state;
-
-       /* Node we are waiting for (if state == BLOCKED) */
-       char *blocked_by;
 
        /* Is this a read-only connection? */
        bool can_write;
@@ -103,9 +99,27 @@
 };
 extern struct list_head connections;
 
-/* Return length of string (including nul) at this offset. */
-unsigned int get_string(const struct buffered_data *data,
-                       unsigned int offset);
+struct node {
+       const char *name;
+
+       /* Database I came from */
+       TDB_CONTEXT *tdb;
+
+       /* Parent (optional) */
+       struct node *parent;
+
+       /* Permissions. */
+       unsigned int num_perms;
+       struct xs_permissions *perms;
+
+       /* Contents. */
+       unsigned int datalen;
+       void *data;
+
+       /* Children, each nul-terminated. */
+       unsigned int childlen;
+       char *children;
+};
 
 /* Break input into vectors, return the number, fill in up to num of them. */
 unsigned int get_strings(struct buffered_data *data,
@@ -113,9 +127,6 @@
 
 /* Is child node a child or equal to parent node? */
 bool is_child(const char *child, const char *parent);
-
-/* Create a new buffer with lifetime of context. */
-struct buffered_data *new_buffer(void *ctx);
 
 void send_reply(struct connection *conn, enum xsd_sockmsg_type type,
                const void *data, unsigned int len);
@@ -129,15 +140,22 @@
 /* Canonicalize this path if possible. */
 char *canonicalize(struct connection *conn, const char *node);
 
-/* Check permissions on this node. */
-bool check_node_perms(struct connection *conn, const char *node,
-                     enum xs_perm_type perm);
-
 /* Check if node is an event node. */
 bool check_event_node(const char *node);
 
-/* Path to this node outside transaction. */
-char *node_dir_outside_transaction(const char *node);
+/* Get this node, checking we have permissions. */
+struct node *get_node(struct connection *conn,
+                     const char *name,
+                     enum xs_perm_type perm);
+
+/* Get TDB context for this connection */
+TDB_CONTEXT *tdb_context(struct connection *conn);
+
+/* Destructor for tdbs: required for transaction code */
+int destroy_tdb(void *_tdb);
+
+/* Replace the tdb: required for transaction code */
+bool replace_tdb(const char *newname, TDB_CONTEXT *newtdb);
 
 /* Fail due to excessive corruption, capitalist pigdogs! */
 void __attribute__((noreturn)) corrupt(struct connection *conn,
@@ -145,23 +163,9 @@
 
 struct connection *new_connection(connwritefn_t *write, connreadfn_t *read);
 
-void handle_input(struct connection *conn);
-void handle_output(struct connection *conn);
-
 /* Is this a valid node name? */
 bool is_valid_nodename(const char *node);
 
-/* Return a pointer to an open dir, self-closig and attached to pathname. */
-DIR **talloc_opendir(const char *pathname);
-
-/* Return a pointer to an fd, self-closing and attached to this pathname. */
-int *talloc_open(const char *pathname, int flags, int mode);
-
-/* Convenient talloc-style destructor for paths. */
-int destroy_path(void *path);
-
-/* Read entire contents of a talloced fd. */
-void *read_all(int *fd, unsigned int *size);
 
 /* Tracing infrastructure. */
 void trace_create(const void *data, const char *type);
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_domain.c
--- a/tools/xenstore/xenstored_domain.c Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_domain.c Fri Sep 23 13:25:01 2005
@@ -309,13 +309,8 @@
                return;
        }
 
-       if (conn->id != 0) {
+       if (conn->id != 0 || !conn->can_write) {
                send_error(conn, EACCES);
-               return;
-       }
-
-       if (!conn->can_write) {
-               send_error(conn, EROFS);
                return;
        }
 
@@ -386,7 +381,7 @@
 
        talloc_free(domain->conn);
 
-       fire_watches(NULL, "@releaseDomain", false);
+       fire_watches(conn, "@releaseDomain", false);
 
        send_ack(conn, XS_RELEASE);
 }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_transaction.c
--- a/tools/xenstore/xenstored_transaction.c    Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_transaction.c    Fri Sep 23 13:25:01 2005
@@ -26,6 +26,7 @@
 #include <stdarg.h>
 #include <stdlib.h>
 #include <fcntl.h>
+#include <unistd.h>
 #include "talloc.h"
 #include "list.h"
 #include "xenstored_transaction.h"
@@ -51,74 +52,26 @@
        /* Global list of transactions. */
        struct list_head list;
 
+       /* Generation when transaction started. */
+       unsigned int generation;
+
        /* My owner (conn->transaction == me). */
        struct connection *conn;
 
-       /* Subtree this transaction covers */
-       char *node;
-
-       /* Base for this transaction. */
-       char *divert;
+       /* TDB to work on, and filename */
+       TDB_CONTEXT *tdb;
+       char *tdb_name;
 
        /* List of changed nodes. */
        struct list_head changes;
-
-       /* Someone's waiting: time limit. */
-       struct timeval timeout;
-
-       /* We've timed out. */
-       bool destined_to_fail;
 };
 static LIST_HEAD(transactions);
+static unsigned int generation;
 
-bool within_transaction(struct transaction *trans, const char *node)
+/* Return tdb context to use for this connection. */
+TDB_CONTEXT *tdb_transaction_context(struct transaction *trans)
 {
-       if (!trans)
-               return true;
-       return is_child(node, trans->node);
-}
-
-/* You are on notice: this transaction is blocking someone. */
-static void start_transaction_timeout(struct transaction *trans)
-{
-       if (timerisset(&trans->timeout))
-               return;
-
-       /* One second timeout. */
-       gettimeofday(&trans->timeout, NULL);
-       trans->timeout.tv_sec += 1;
-}
-
-struct transaction *transaction_covering_node(const char *node)
-{
-       struct transaction *i;
-
-       list_for_each_entry(i, &transactions, list) {
-               if (i->destined_to_fail)
-                       continue;
-               if (is_child(i->node, node) || is_child(node, i->node))
-                       return i;
-       }
-       return NULL;
-}
-
-bool transaction_block(struct connection *conn, const char *node)
-{
-       struct transaction *trans;
-
-       /* Transactions don't overlap, so we can't be blocked by
-        * others if we're in one. */
-       if (conn->transaction)
-               return false;
-
-       trans = transaction_covering_node(node);
-       if (trans) {
-               start_transaction_timeout(trans);
-               conn->state = BLOCKED;
-               conn->blocked_by = talloc_strdup(conn, node);
-               return true;
-       }
-       return false;
+       return trans->tdb;
 }
 
 /* Callers get a change node (which can fail) and only commit after they've
@@ -127,8 +80,11 @@
 {
        struct changed_node *i;
 
-       if (!trans)
+       if (!trans) {
+               /* They're changing the global database. */
+               generation++;
                return;
+       }
 
        list_for_each_entry(i, &trans->changes, list)
                if (streq(i->node, node))
@@ -140,167 +96,47 @@
        list_add_tail(&i->list, &trans->changes);
 }
 
-char *node_dir_inside_transaction(struct transaction *trans, const char *node)
-{
-       return talloc_asprintf(node, "%s/%s", trans->divert,
-                              node + strlen(trans->node));
-}
-
-void shortest_transaction_timeout(struct timeval *tv)
-{
-       struct transaction *i;
-
-       list_for_each_entry(i, &transactions, list) {
-               if (!timerisset(&i->timeout))
-                       continue;
-
-               if (!timerisset(tv) || timercmp(&i->timeout, tv, <))
-                       *tv = i->timeout;
-       }
-}      
-
-void check_transaction_timeout(void)
-{
-       struct transaction *i;
-       struct timeval now;
-
-       gettimeofday(&now, NULL);
-
-       list_for_each_entry(i, &transactions, list) {
-               if (!timerisset(&i->timeout))
-                       continue;
-
-               if (timercmp(&i->timeout, &now, <))
-                       i->destined_to_fail = true;
-       }
-}
-
 static int destroy_transaction(void *_transaction)
 {
        struct transaction *trans = _transaction;
 
        list_del(&trans->list);
        trace_destroy(trans, "transaction");
-       return destroy_path(trans->divert);
+       if (trans->tdb)
+               tdb_close(trans->tdb);
+       unlink(trans->tdb_name);
+       return 0;
 }
 
-static bool copy_file(const char *src, const char *dst)
+void do_transaction_start(struct connection *conn, struct buffered_data *in)
 {
-       int *infd, *outfd;
-       void *data;
-       unsigned int size;
-
-       infd = talloc_open(src, O_RDONLY, 0);
-       if (!infd)
-               return false;
-       outfd = talloc_open(dst, O_WRONLY|O_CREAT|O_EXCL, 0640);
-       if (!outfd)
-               return false;
-       data = read_all(infd, &size);
-       if (!data)
-               return false;
-       return xs_write_all(*outfd, data, size);
-}
-
-static bool copy_dir(const char *src, const char *dst)
-{
-       DIR **dir;
-       struct dirent *dirent;
-
-       if (mkdir(dst, 0750) != 0)
-               return false;
-
-       dir = talloc_opendir(src);
-       if (!dir)
-               return false;
-
-       while ((dirent = readdir(*dir)) != NULL) {
-               struct stat st;
-               char *newsrc, *newdst;
-
-               if (streq(dirent->d_name, ".") || streq(dirent->d_name, ".."))
-                       continue;
-
-               newsrc = talloc_asprintf(src, "%s/%s", src, dirent->d_name);
-               newdst = talloc_asprintf(src, "%s/%s", dst, dirent->d_name);
-               if (stat(newsrc, &st) != 0)
-                       return false;
-               
-               if (S_ISDIR(st.st_mode)) {
-                       if (!copy_dir(newsrc, newdst))
-                               return false;
-               } else {
-                       if (!copy_file(newsrc, newdst))
-                               return false;
-               }
-               /* Free now so we don't run out of file descriptors */
-               talloc_free(newsrc);
-               talloc_free(newdst);
-       }
-       return true;
-}
-
-void do_transaction_start(struct connection *conn, const char *node)
-{
-       struct transaction *transaction;
-       char *dir;
+       struct transaction *trans;
 
        if (conn->transaction) {
                send_error(conn, EBUSY);
                return;
        }
 
-       node = canonicalize(conn, node);
-       if (!check_node_perms(conn, node, XS_PERM_READ)) {
+       /* Attach transaction to input for autofree until it's complete */
+       trans = talloc(in, struct transaction);
+       INIT_LIST_HEAD(&trans->changes);
+       trans->conn = conn;
+       trans->generation = generation;
+       trans->tdb_name = talloc_asprintf(trans, "%s.%p",
+                                         xs_daemon_tdb(), trans);
+       trans->tdb = tdb_copy(tdb_context(conn), trans->tdb_name);
+       if (!trans->tdb) {
                send_error(conn, errno);
                return;
        }
+       /* Make it close if we go away. */
+       talloc_steal(trans, trans->tdb);
 
-       if (transaction_block(conn, node))
-               return;
-
-       dir = node_dir_outside_transaction(node);
-
-       /* Attach transaction to node for autofree until it's complete */
-       transaction = talloc(node, struct transaction);
-       transaction->node = talloc_strdup(transaction, node);
-       transaction->divert = talloc_asprintf(transaction, "%s/%p", 
-                                             xs_daemon_transactions(),
-                                             transaction);
-       INIT_LIST_HEAD(&transaction->changes);
-       transaction->conn = conn;
-       timerclear(&transaction->timeout);
-       transaction->destined_to_fail = false;
-       list_add_tail(&transaction->list, &transactions);
-       talloc_set_destructor(transaction, destroy_transaction);
-       trace_create(transaction, "transaction");
-
-       if (!copy_dir(dir, transaction->divert)) {
-               send_error(conn, errno);
-               return;
-       }
-
-       talloc_steal(conn, transaction);
-       conn->transaction = transaction;
-       send_ack(transaction->conn, XS_TRANSACTION_START);
-}
-
-static bool commit_transaction(struct transaction *trans)
-{
-       char *tmp, *dir;
-
-       /* Move: orig -> .old, repl -> orig.  Cleanup deletes .old. */
-       dir = node_dir_outside_transaction(trans->node);
-       tmp = talloc_asprintf(trans, "%s.old", dir);
-
-       if (rename(dir, tmp) != 0)
-               return false;
-       if (rename(trans->divert, dir) != 0)
-               corrupt(trans->conn, "Failed rename %s to %s",
-                       trans->divert, dir);
-
-       trans->divert = tmp;
-       return true;
+       /* Now we own it. */
+       conn->transaction = talloc_steal(conn, trans);
+       list_add_tail(&trans->list, &transactions);
+       talloc_set_destructor(trans, destroy_transaction);
+       send_ack(conn, XS_TRANSACTION_START);
 }
 
 void do_transaction_end(struct connection *conn, const char *arg)
@@ -318,25 +154,29 @@
                return;
        }
 
-       /* Set to NULL so fire_watches sends events. */
+       /* Set to NULL so fire_watches sends events, tdb_context works. */
        trans = conn->transaction;
        conn->transaction = NULL;
        /* Attach transaction to arg for auto-cleanup */
        talloc_steal(arg, trans);
 
        if (streq(arg, "T")) {
-               if (trans->destined_to_fail) {
-                       send_error(conn, ETIMEDOUT);
+               /* FIXME: Merge, rather failing on any change. */
+               if (trans->generation != generation) {
+                       send_error(conn, EAGAIN);
                        return;
                }
-               if (!commit_transaction(trans)) {
+               if (!replace_tdb(trans->tdb_name, trans->tdb)) {
                        send_error(conn, errno);
                        return;
                }
+               /* Don't close this: we won! */
+               trans->tdb = NULL;
 
                /* Fire off the watches for everything that changed. */
                list_for_each_entry(i, &trans->changes, list)
                        fire_watches(conn, i->node, i->recurse);
+               generation++;
        }
        send_ack(conn, XS_TRANSACTION_END);
 }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_transaction.h
--- a/tools/xenstore/xenstored_transaction.h    Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_transaction.h    Fri Sep 23 13:25:01 2005
@@ -22,29 +22,14 @@
 
 struct transaction;
 
-void do_transaction_start(struct connection *conn, const char *node);
+void do_transaction_start(struct connection *conn, struct buffered_data *node);
 void do_transaction_end(struct connection *conn, const char *arg);
 
-/* Is node covered by this transaction? */
-bool within_transaction(struct transaction *trans, const char *node);
-
-/* If a write op on this node blocked by another connections' transaction,
- * mark conn, setup transaction timeout and return true.
- */
-bool transaction_block(struct connection *conn, const char *node);
-
-/* Return transaction which covers this node. */
-struct transaction *transaction_covering_node(const char *node);
-
-/* Return directory of node within transaction t. */
-char *node_dir_inside_transaction(struct transaction *t, const char *node);
+bool transaction_block(struct connection *conn);
 
 /* This node was changed: can fail and longjmp. */
 void add_change_node(struct transaction *trans, const char *node, bool 
recurse);
 
-/* Get shortest timeout: leave tv unset if none. */
-void shortest_transaction_timeout(struct timeval *tv);
-
-/* Have any transactions timed out yet? */
-void check_transaction_timeout(void);
+/* Return tdb context to use for this connection. */
+TDB_CONTEXT *tdb_transaction_context(struct transaction *trans);
 #endif /* _XENSTORED_TRANSACTION_H */
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_watch.c
--- a/tools/xenstore/xenstored_watch.c  Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_watch.c  Fri Sep 23 13:25:01 2005
@@ -96,36 +96,38 @@
 }
 
 static void add_event(struct connection *conn,
-                     struct watch *watch, const char *node)
-{
-       struct watch_event *event;
-
-       /* Check read permission: no permission, no watch event.
-        * If it doesn't exist, we need permission to read parent.
-        */
-       if (!check_node_perms(conn, node, XS_PERM_READ|XS_PERM_ENOENT_OK) &&
-           !check_event_node(node)) {
-               return;
+                     struct watch *watch,
+                     const char *name)
+{
+       struct watch_event *event;
+
+       if (!check_event_node(name)) {
+               /* Can this conn load node, or see that it doesn't exist? */
+               struct node *node;
+
+               node = get_node(conn, name, XS_PERM_READ);
+               if (!node && errno != ENOENT)
+                       return;
        }
 
        if (watch->relative_path) {
-               node += strlen(watch->relative_path);
-               if (*node == '/') /* Could be "" */
-                       node++;
+               name += strlen(watch->relative_path);
+               if (*name == '/') /* Could be "" */
+                       name++;
        }
 
        event = talloc(watch, struct watch_event);
-       event->len = strlen(node) + 1 + strlen(watch->token) + 1;
+       event->len = strlen(name) + 1 + strlen(watch->token) + 1;
        event->data = talloc_array(event, char, event->len);
-       strcpy(event->data, node);
-       strcpy(event->data + strlen(node) + 1, watch->token);
+       strcpy(event->data, name);
+       strcpy(event->data + strlen(name) + 1, watch->token);
        talloc_set_destructor(event, destroy_watch_event);
        list_add_tail(&event->list, &watch->events);
        trace_create(event, "watch_event");
 }
 
 /* FIXME: we fail to fire on out of memory.  Should drop connections. */
-void fire_watches(struct connection *conn, const char *node, bool recurse)
+void fire_watches(struct connection *conn, const char *name, bool recurse)
 {
        struct connection *i;
        struct watch *watch;
@@ -137,9 +139,9 @@
        /* Create an event for each watch. */
        list_for_each_entry(i, &connections, list) {
                list_for_each_entry(watch, &i->watches, list) {
-                       if (is_child(node, watch->node))
-                               add_event(i, watch, node);
-                       else if (recurse && is_child(watch->node, node))
+                       if (is_child(name, watch->node))
+                               add_event(i, watch, name);
+                       else if (recurse && is_child(watch->node, name))
                                add_event(i, watch, watch->node);
                        else
                                continue;
@@ -154,49 +156,6 @@
 {
        trace_destroy(_watch, "watch");
        return 0;
-}
-
-void shortest_watch_ack_timeout(struct timeval *tv)
-{
-       (void)tv;
-#if 0 /* FIXME */
-       struct watch *watch;
-
-       list_for_each_entry(watch, &watches, list) {
-               struct watch_event *i;
-               list_for_each_entry(i, &watch->events, list) {
-                       if (!timerisset(&i->timeout))
-                               continue;
-                       if (!timerisset(tv) || timercmp(&i->timeout, tv, <))
-                               *tv = i->timeout;
-               }
-       }
-#endif
-}      
-
-void check_watch_ack_timeout(void)
-{
-#if 0
-       struct watch *watch;
-       struct timeval now;
-
-       gettimeofday(&now, NULL);
-       list_for_each_entry(watch, &watches, list) {
-               struct watch_event *i, *tmp;
-               list_for_each_entry_safe(i, tmp, &watch->events, list) {
-                       if (!timerisset(&i->timeout))
-                               continue;
-                       if (timercmp(&i->timeout, &now, <)) {
-                               xprintf("Warning: timeout on watch event %s"
-                                       " token %s\n",
-                                       i->node, watch->token);
-                               trace_watch_timeout(watch->conn, i->node,
-                                                   watch->token);
-                               timerclear(&i->timeout);
-                       }
-               }
-       }
-#endif
 }
 
 void do_watch(struct connection *conn, struct buffered_data *in)
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xenstored_watch.h
--- a/tools/xenstore/xenstored_watch.h  Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xenstored_watch.h  Fri Sep 23 13:25:01 2005
@@ -32,15 +32,9 @@
 /* Look through our watches: if any of them have an event, queue it. */
 void queue_next_event(struct connection *conn);
 
-/* Fire all watches: recurse means all the children are effected (ie. rm).
+/* Fire all watches: recurse means all the children are affected (ie. rm).
  */
-void fire_watches(struct connection *conn, const char *node, bool recurse);
-
-/* Find shortest timeout: if any, reduce tv (may already be set). */
-void shortest_watch_ack_timeout(struct timeval *tv);
-
-/* Check for watches which may have timed out. */
-void check_watch_ack_timeout(void);
+void fire_watches(struct connection *conn, const char *name, bool recurse);
 
 void dump_watches(struct connection *conn);
 
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs.c
--- a/tools/xenstore/xs.c       Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs.c       Fri Sep 23 13:25:01 2005
@@ -497,13 +497,12 @@
 
 /* Start a transaction: changes by others will not be seen during this
  * transaction, and changes will not be visible to others until end.
- * Transaction only applies to the given subtree.
  * You can only have one transaction at any time.
  * Returns false on failure.
  */
-bool xs_transaction_start(struct xs_handle *h, const char *subtree)
-{
-       return xs_bool(xs_single(h, XS_TRANSACTION_START, subtree, NULL));
+bool xs_transaction_start(struct xs_handle *h)
+{
+       return xs_bool(xs_single(h, XS_TRANSACTION_START, "", NULL));
 }
 
 /* End a transaction.
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs.h
--- a/tools/xenstore/xs.h       Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs.h       Fri Sep 23 13:25:01 2005
@@ -109,11 +109,10 @@
 
 /* Start a transaction: changes by others will not be seen during this
  * transaction, and changes will not be visible to others until end.
- * Transaction only applies to the given subtree.
  * You can only have one transaction at any time.
  * Returns false on failure.
  */
-bool xs_transaction_start(struct xs_handle *h, const char *subtree);
+bool xs_transaction_start(struct xs_handle *h);
 
 /* End a transaction.
  * If abandon is true, transaction is discarded instead of committed.
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_lib.c
--- a/tools/xenstore/xs_lib.c   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_lib.c   Fri Sep 23 13:25:01 2005
@@ -50,6 +50,13 @@
        return buf;
 }
 
+const char *xs_daemon_tdb(void)
+{
+       static char buf[PATH_MAX];
+       sprintf(buf, "%s/tdb", xs_daemon_rootdir());
+       return buf;
+}
+
 const char *xs_daemon_socket(void)
 {
        return xs_daemon_path();
@@ -62,24 +69,6 @@
        if (s == NULL)
                return NULL;
        if (snprintf(buf, PATH_MAX, "%s_ro", s) >= PATH_MAX)
-               return NULL;
-       return buf;
-}
-
-const char *xs_daemon_store(void)
-{
-       static char buf[PATH_MAX];
-       if (snprintf(buf, PATH_MAX, "%s/store",
-                    xs_daemon_rootdir()) >= PATH_MAX)
-               return NULL;
-       return buf;
-}
-
-const char *xs_daemon_transactions(void)
-{
-       static char buf[PATH_MAX];
-       if (snprintf(buf, PATH_MAX, "%s/transactions",
-                    xs_daemon_rootdir()) >= PATH_MAX)
                return NULL;
        return buf;
 }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_lib.h
--- a/tools/xenstore/xs_lib.h   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_lib.h   Fri Sep 23 13:25:01 2005
@@ -36,7 +36,7 @@
 
 struct xs_permissions
 {
-       domid_t id;
+       unsigned int id;
        enum xs_perm_type perms;
 };
 
@@ -46,9 +46,8 @@
 /* Path for various daemon things: env vars can override. */
 const char *xs_daemon_socket(void);
 const char *xs_daemon_socket_ro(void);
-const char *xs_daemon_store(void);
-const char *xs_daemon_transactions(void);
 const char *xs_domain_dev(void);
+const char *xs_daemon_tdb(void);
 
 /* Simple write function: loops for you. */
 bool xs_write_all(int fd, const void *data, unsigned int len);
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_random.c
--- a/tools/xenstore/xs_random.c        Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_random.c        Fri Sep 23 13:25:01 2005
@@ -41,7 +41,7 @@
                          struct xs_permissions *perms,
                          unsigned int num);
 
-       bool (*transaction_start)(void *h, const char *subtree);
+       bool (*transaction_start)(void *h);
        bool (*transaction_end)(void *h, bool abort);
 
        /* Create and destroy a new handle. */
@@ -53,7 +53,6 @@
 {
        const char *base;
        char *transact_base;
-       char *transact;
 };
 
 static void convert_to_dir(const char *dirname)
@@ -95,31 +94,6 @@
        maybe_convert_to_directory(filename);
        return filename;
 }
-
-/* Is child a subnode of parent, or equal? */
-static bool is_child(const char *child, const char *parent)
-{
-       unsigned int len = strlen(parent);
-
-       /* / should really be "" for this algorithm to work, but that's a
-        * usability nightmare. */
-       if (streq(parent, "/"))
-               return true;
-
-       if (strncmp(child, parent, len) != 0)
-               return false;
-
-       return child[len] == '/' || child[len] == '\0';
-}
-
-static bool write_ok(struct file_ops_info *info, const char *path)
-{
-       if (info->transact && !is_child(path, info->transact)) {
-               errno = EROFS;
-               return false;
-       }
-       return true;
-}      
 
 static char **file_directory(struct file_ops_info *info,
                             const char *path, unsigned int *num)
@@ -184,8 +158,10 @@
 
        ret = grab_file(filename, &size);
        /* Directory exists, .DATA doesn't. */
-       if (!ret && errno == ENOENT && strends(filename, ".DATA"))
-               errno = EISDIR;
+       if (!ret && errno == ENOENT && strends(filename, ".DATA")) {
+               ret = strdup("");
+               size = 0;
+       }
        *len = size;
        return ret;
 }
@@ -270,9 +246,6 @@
                return false;
        }
 
-       if (!write_ok(info, path))
-               return false;
-
        /* Check non-perm file exists/ */
        if (lstat(filename, &st) != 0)
                return false;
@@ -338,9 +311,6 @@
        char *filename = filename_to_data(path_to_name(info, path));
        int fd;
 
-       if (!write_ok(info, path))
-               return false;
-
        make_dirs(parent_filename(filename));
        fd = open(filename, O_CREAT|O_TRUNC|O_WRONLY, 0600);
        if (fd < 0)
@@ -358,9 +328,6 @@
 {
        char *dirname = path_to_name(info, path);
 
-       if (!write_ok(info, path))
-               return false;
-
        make_dirs(parent_filename(dirname));
        if (mkdir(dirname, 0700) != 0)
                return (errno == EEXIST);
@@ -373,11 +340,6 @@
 {
        char *filename = path_to_name(info, path);
        struct stat st;
-
-       if (info->transact && streq(info->transact, path)) {
-               errno = EINVAL;
-               return false;
-       }
 
        if (lstat(filename, &st) != 0) {
                if (lstat(parent_filename(filename), &st) != 0)
@@ -385,9 +347,6 @@
                return true;
        }
 
-       if (!write_ok(info, path))
-               return false;
-
        if (streq(path, "/")) {
                errno = EINVAL;
                return false;
@@ -398,28 +357,20 @@
        return true;
 }
 
-static bool file_transaction_start(struct file_ops_info *info,
-                                  const char *subtree)
+static bool file_transaction_start(struct file_ops_info *info)
 {
        char *cmd;
-       char *filename = path_to_name(info, subtree);
-       struct stat st;
-
-       if (info->transact) {
+
+       if (info->transact_base) {
                errno = EBUSY;
                return false;
        }
 
-       if (lstat(filename, &st) != 0)
-               return false;
-
-       cmd = talloc_asprintf(NULL, "cp -r %s %s.transact",
-                             info->base, info->base);
+       info->transact_base = talloc_asprintf(NULL, "%s.transact", info->base);
+       cmd = talloc_asprintf(NULL, "cp -r %s %s",
+                             info->base, info->transact_base);
        do_command(cmd);
        talloc_free(cmd);
-
-       info->transact_base = talloc_asprintf(NULL, "%s.transact", info->base);
-       info->transact = talloc_strdup(NULL, subtree);
        return true;
 }
 
@@ -427,7 +378,7 @@
 {
        char *old, *cmd;
 
-       if (!info->transact) {
+       if (!info->transact_base) {
                errno = ENOENT;
                return false;
        }
@@ -448,9 +399,7 @@
 
 success:
        talloc_free(cmd);
-       talloc_free(info->transact);
        talloc_free(info->transact_base);
-       info->transact = NULL;
        info->transact_base = NULL;
        return true;
 }
@@ -461,7 +410,6 @@
 
        info->base = dir;
        info->transact_base = NULL;
-       info->transact = NULL;
        return info;
 }
 
@@ -898,11 +846,10 @@
        case 7: {
                if (verbose)
                        printf("START %s\n", name);
-               ret = bool_to_errstring(ops->transaction_start(h, name));
+               ret = bool_to_errstring(ops->transaction_start(h));
                if (streq(ret, "OK")) {
                        talloc_free(ret);
-                       ret = talloc_asprintf(NULL, "OK:START-TRANSACT:%s",
-                                             name);
+                       ret = talloc_asprintf(NULL, "OK:START-TRANSACT");
                }
 
                break;
@@ -978,6 +925,8 @@
                barf_perror("Creating directory %s/tool", dir);
        if (!file_set_perms(h, talloc_strdup(h, "/"), &perm, 1))
                barf_perror("Setting root perms in %s", dir);
+       if (!file_set_perms(h, talloc_strdup(h, "/tool"), &perm, 1))
+               barf_perror("Setting root perms in %s/tool", dir);
        file_close(h);
 }
 
@@ -1071,7 +1020,7 @@
                        goto out;
 
                if (!data->fast) {
-                       if (strstarts(ret, "OK:START-TRANSACT:")) {
+                       if (streq(ret, "OK:START-TRANSACT")) {
                                void *pre = data->ops->handle(data->dir);
 
                                snapshot = dump(data->ops, pre);
@@ -1303,7 +1252,7 @@
                             void *_data)
 {
        void *fileh, *xsh;
-       char *transact = NULL;
+       bool transact = false;
        struct ops *fail;
        struct diff_data *data = _data;
        unsigned int i, print;
@@ -1348,13 +1297,9 @@
                        goto out;
 
                if (strstarts(file, "OK:START-TRANSACT:"))
-                       transact = talloc_strdup(NULL,
-                                                file +
-                                                strlen("OK:START-TRANSACT:"));
-               else if (streq(file, "OK:STOP-TRANSACT")) {
-                       talloc_free(transact);
-                       transact = NULL;
-               }
+                       transact = true;
+               else if (streq(file, "OK:STOP-TRANSACT"))
+                       transact = false;
 
                talloc_free(file);
                talloc_free(xs);
@@ -1379,7 +1324,7 @@
 
                        fail = NULL;
                        if (!ops_equal(&xs_ops, xsh_pre, &file_ops, fileh_pre,
-                                      transact, &fail)) {
+                                      "/", &fail)) {
                                if (fail)
                                        barf("%s failed during transact\n",
                                             fail->name);
@@ -1455,9 +1400,6 @@
 
        fileh = file_handle(data->dir);
        xsh = xs_handle(data->dir);
-
-       sprintf(seed, "%i", data->seed);
-       free(xs_debug_command(xsh, "failtest", seed, strlen(seed)+1));
 
        print = number / 76;
        if (!print)
@@ -1491,8 +1433,12 @@
                if (trymap && !trymap[i])
                        continue;
 
-               if (verbose)
-                       printf("(%i) ", i);
+               /* Turn on failure. */
+               sprintf(seed, "%i", data->seed + i);
+               free(xs_debug_command(xsh, "failtest",seed,strlen(seed)+1));
+
+               if (verbose)
+                       printf("(%i) seed %s ", i, seed);
                ret = do_next_op(&xs_ops, xsh, i + data->seed, verbose);
                if (streq(ret, "FAILED:Connection reset by peer")
                    || streq(ret, "FAILED:Bad file descriptor")
@@ -1549,8 +1495,6 @@
                fail = NULL;
                if (!ops_equal(&xs_ops, tmpxsh, &file_ops, tmpfileh, "/",
                               &fail)) {
-                       xs_close(tmpxsh);
-                       file_close(tmpfileh);
                        if (fail) {
                                if (verbose)
                                        printf("%s failed\n", fail->name);
@@ -1561,10 +1505,16 @@
                                failed = 0;
                                if (verbose)
                                        printf("(Looks like it succeeded)\n");
+                               xs_close(tmpxsh);
+                               file_close(tmpfileh);
                                goto try_applying;
                        }
                        if (verbose)
-                               printf("Two backends not equal\n");
+                               printf("Trees differ:\nXS:%s\nFILE:%s\n",
+                                      dump(&xs_ops, tmpxsh),
+                                      dump(&file_ops, tmpfileh));
+                       xs_close(tmpxsh);
+                       file_close(tmpfileh);
                        goto out;
                }
 
@@ -1572,8 +1522,6 @@
                if (!xsh)
                        file_transaction_end(fileh, true);
 
-               /* Turn failures back on. */
-               free(xs_debug_command(tmpxsh, "failtest",  NULL, 0));
                xs_close(tmpxsh);
                file_close(tmpfileh);
        }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_stress.c
--- a/tools/xenstore/xs_stress.c        Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_stress.c        Fri Sep 23 13:25:01 2005
@@ -8,6 +8,7 @@
 #include <sys/stat.h>
 #include <fcntl.h>
 #include <unistd.h>
+#include <errno.h>
 
 #define NUM_HANDLES 2
 #define DIR_FANOUT 3
@@ -36,24 +37,18 @@
 
        srandom(childnum);
        for (i = 0; i < cycles; i++) {
-               unsigned int lockdepth, j, len;
-               char file[100] = "", lockdir[100];
+               unsigned int j, len;
+               char file[100] = "";
                char *contents, tmp[100];
                struct xs_handle *h = handles[random() % NUM_HANDLES];
 
-               lockdepth = random() % DIR_DEPTH;
-               for (j = 0; j < DIR_DEPTH; j++) {
-                       if (j == lockdepth)
-                               strcpy(lockdir, file);
+               for (j = 0; j < DIR_DEPTH; j++)
                        sprintf(file + strlen(file), "/%li",
                                random()%DIR_FANOUT);
-               }
-               if (streq(lockdir, ""))
-                       strcpy(lockdir, "/");
-
-               if (!xs_transaction_start(h, lockdir))
-                       barf_perror("%i: starting transaction %i on %s",
-                                   childnum, i, lockdir);
+
+               if (!xs_transaction_start(h))
+                       barf_perror("%i: starting transaction %i",
+                                   childnum, i);
 
                sprintf(file + strlen(file), "/count");
                contents = xs_read(h, file, &len);
@@ -68,18 +63,23 @@
                /* Abandon 1 in 10 */
                if (random() % 10 == 0) {
                        if (!xs_transaction_end(h, true))
-                               barf_perror("%i: can't abort transact %s",
-                                           childnum, lockdir);
+                               barf_perror("%i: can't abort transact",
+                                           childnum);
                        i--;
                } else {
-                       if (!xs_transaction_end(h, false))
-                               barf_perror("%i: can't commit transact %s",
-                                           childnum, lockdir);
-
-                       /* Offset when we print . so kids don't all
-                        * print at once. */
-                       if ((i + print/(childnum+1)) % print == 0)
-                               write(STDOUT_FILENO, &id, 1);
+                       if (!xs_transaction_end(h, false)) {
+                               if (errno == EAGAIN) {
+                                       write(STDOUT_FILENO, "!", 1);
+                                       i--;
+                               } else
+                                       barf_perror("%i: can't commit trans",
+                                                   childnum);
+                       } else {
+                               /* Offset when we print . so kids don't all
+                                * print at once. */
+                               if ((i + print/(childnum+1)) % print == 0)
+                                       write(STDOUT_FILENO, &id, 1);
+                       }
                }
        }
 }
@@ -201,7 +201,7 @@
        printf("\nCounting results...\n");
        i = tally_counts();
        if (i != (unsigned)atoi(argv[1]))
-               barf("Total counts %i not %s", i, atoi(argv[1]));
+               barf("Total counts %i not %s", i, argv[1]);
        printf("Success!\n");
        exit(0);
 }
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_test.c
--- a/tools/xenstore/xs_test.c  Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_test.c  Fri Sep 23 13:25:01 2005
@@ -562,9 +562,9 @@
                failed(handle);
 }
 
-static void do_start(unsigned int handle, const char *node)
-{
-       if (!xs_transaction_start(handles[handle], node))
+static void do_start(unsigned int handle)
+{
+       if (!xs_transaction_start(handles[handle]))
                failed(handle);
 }
 
@@ -791,7 +791,7 @@
                xs_daemon_close(handles[handle]);
                handles[handle] = NULL;
        } else if (streq(command, "start"))
-               do_start(handle, arg(line, 1));
+               do_start(handle);
        else if (streq(command, "commit"))
                do_end(handle, false);
        else if (streq(command, "abort"))
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/tdb.c
--- /dev/null   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/tdb.c      Fri Sep 23 13:25:01 2005
@@ -0,0 +1,2151 @@
+ /* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell              1999-2004
+   Copyright (C) Paul `Rusty' Russell             2000
+   Copyright (C) Jeremy Allison                           2000-2003
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+
+#ifndef _SAMBA_BUILD_
+#if HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <unistd.h>
+#include <string.h>
+#include <fcntl.h>
+#include <errno.h>
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include "tdb.h"
+#include <stdarg.h>
+#include "talloc.h"
+#define HAVE_MMAP
+#else
+#include "includes.h"
+#include "lib/tdb/include/tdb.h"
+#include "system/time.h"
+#include "system/shmem.h"
+#include "system/filesys.h"
+#endif
+
+#define TDB_MAGIC_FOOD "TDB file\n"
+#define TDB_VERSION (0x26011967 + 6)
+#define TDB_MAGIC (0x26011999U)
+#define TDB_FREE_MAGIC (~TDB_MAGIC)
+#define TDB_DEAD_MAGIC (0xFEE1DEAD)
+#define TDB_ALIGNMENT 4
+#define MIN_REC_SIZE (2*sizeof(struct list_struct) + TDB_ALIGNMENT)
+#define DEFAULT_HASH_SIZE 131
+#define TDB_PAGE_SIZE 0x2000
+#define FREELIST_TOP (sizeof(struct tdb_header))
+#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
+#define TDB_BYTEREV(x) 
(((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
+#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
+#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
+#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off))
+#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1))
+
+
+/* NB assumes there is a local variable called "tdb" that is the
+ * current context, also takes doubly-parenthesized print-style
+ * argument. */
+#define TDB_LOG(x) tdb->log_fn x
+
+/* lock offsets */
+#define GLOBAL_LOCK 0
+#define ACTIVE_LOCK 4
+
+#ifndef MAP_FILE
+#define MAP_FILE 0
+#endif
+
+#ifndef MAP_FAILED
+#define MAP_FAILED ((void *)-1)
+#endif
+
+#ifndef discard_const_p
+# if defined(__intptr_t_defined) || defined(HAVE_INTPTR_T)
+#  define discard_const(ptr) ((void *)((intptr_t)(ptr)))
+# else
+#  define discard_const(ptr) ((void *)(ptr))
+# endif
+# define discard_const_p(type, ptr) ((type *)discard_const(ptr))
+#endif
+
+/* free memory if the pointer is valid and zero the pointer */
+#ifndef SAFE_FREE
+#define SAFE_FREE(x) do { if ((x) != NULL) {talloc_free(discard_const_p(void 
*, (x))); (x)=NULL;} } while(0)
+#endif
+
+#define BUCKET(hash) ((hash) % tdb->header.hash_size)
+TDB_DATA tdb_null;
+
+/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
+static TDB_CONTEXT *tdbs = NULL;
+
+static int tdb_munmap(TDB_CONTEXT *tdb)
+{
+       if (tdb->flags & TDB_INTERNAL)
+               return 0;
+
+#ifdef HAVE_MMAP
+       if (tdb->map_ptr) {
+               int ret = munmap(tdb->map_ptr, tdb->map_size);
+               if (ret != 0)
+                       return ret;
+       }
+#endif
+       tdb->map_ptr = NULL;
+       return 0;
+}
+
+static void tdb_mmap(TDB_CONTEXT *tdb)
+{
+       if (tdb->flags & TDB_INTERNAL)
+               return;
+
+#ifdef HAVE_MMAP
+       if (!(tdb->flags & TDB_NOMMAP)) {
+               tdb->map_ptr = mmap(NULL, tdb->map_size, 
+                                   PROT_READ|(tdb->read_only? 0:PROT_WRITE), 
+                                   MAP_SHARED|MAP_FILE, tdb->fd, 0);
+
+               /*
+                * NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
+                */
+
+               if (tdb->map_ptr == MAP_FAILED) {
+                       tdb->map_ptr = NULL;
+                       TDB_LOG((tdb, 2, "tdb_mmap failed for size %d (%s)\n", 
+                                tdb->map_size, strerror(errno)));
+               }
+       } else {
+               tdb->map_ptr = NULL;
+       }
+#else
+       tdb->map_ptr = NULL;
+#endif
+}
+
+/* Endian conversion: we only ever deal with 4 byte quantities */
+static void *convert(void *buf, u32 size)
+{
+       u32 i, *p = buf;
+       for (i = 0; i < size / 4; i++)
+               p[i] = TDB_BYTEREV(p[i]);
+       return buf;
+}
+#define DOCONV() (tdb->flags & TDB_CONVERT)
+#define CONVERT(x) (DOCONV() ? convert(&x, sizeof(x)) : &x)
+
+/* the body of the database is made of one list_struct for the free space
+   plus a separate data list for each hash value */
+struct list_struct {
+       tdb_off next; /* offset of the next record in the list */
+       tdb_len rec_len; /* total byte length of record */
+       tdb_len key_len; /* byte length of key */
+       tdb_len data_len; /* byte length of data */
+       u32 full_hash; /* the full 32 bit hash of the key */
+       u32 magic;   /* try to catch errors */
+       /* the following union is implied:
+               union {
+                       char record[rec_len];
+                       struct {
+                               char key[key_len];
+                               char data[data_len];
+                       }
+                       u32 totalsize; (tailer)
+               }
+       */
+};
+
+/* a byte range locking function - return 0 on success
+   this functions locks/unlocks 1 byte at the specified offset.
+
+   On error, errno is also set so that errors are passed back properly
+   through tdb_open(). */
+static int tdb_brlock(TDB_CONTEXT *tdb, tdb_off offset, 
+                     int rw_type, int lck_type, int probe)
+{
+       struct flock fl;
+       int ret;
+
+       if (tdb->flags & TDB_NOLOCK)
+               return 0;
+       if ((rw_type == F_WRLCK) && (tdb->read_only)) {
+               errno = EACCES;
+               return -1;
+       }
+
+       fl.l_type = rw_type;
+       fl.l_whence = SEEK_SET;
+       fl.l_start = offset;
+       fl.l_len = 1;
+       fl.l_pid = 0;
+
+       do {
+               ret = fcntl(tdb->fd,lck_type,&fl);
+       } while (ret == -1 && errno == EINTR);
+
+       if (ret == -1) {
+               if (!probe && lck_type != F_SETLK) {
+                       /* Ensure error code is set for log fun to examine. */
+                       tdb->ecode = TDB_ERR_LOCK;
+                       TDB_LOG((tdb, 5,"tdb_brlock failed (fd=%d) at offset %d 
rw_type=%d lck_type=%d\n", 
+                                tdb->fd, offset, rw_type, lck_type));
+               }
+               /* Generic lock error. errno set by fcntl.
+                * EAGAIN is an expected return from non-blocking
+                * locks. */
+               if (errno != EAGAIN) {
+               TDB_LOG((tdb, 5, "tdb_brlock failed (fd=%d) at offset %d 
rw_type=%d lck_type=%d: %s\n", 
+                                tdb->fd, offset, rw_type, lck_type, 
+                                strerror(errno)));
+               }
+               return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+       }
+       return 0;
+}
+
+/* lock a list in the database. list -1 is the alloc list */
+static int tdb_lock(TDB_CONTEXT *tdb, int list, int ltype)
+{
+       if (list < -1 || list >= (int)tdb->header.hash_size) {
+               TDB_LOG((tdb, 0,"tdb_lock: invalid list %d for ltype=%d\n", 
+                          list, ltype));
+               return -1;
+       }
+       if (tdb->flags & TDB_NOLOCK)
+               return 0;
+
+       /* Since fcntl locks don't nest, we do a lock for the first one,
+          and simply bump the count for future ones */
+       if (tdb->locked[list+1].count == 0) {
+               if (tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
+                       TDB_LOG((tdb, 0,"tdb_lock failed on list %d ltype=%d 
(%s)\n", 
+                                          list, ltype, strerror(errno)));
+                       return -1;
+               }
+               tdb->locked[list+1].ltype = ltype;
+       }
+       tdb->locked[list+1].count++;
+       return 0;
+}
+
+/* unlock the database: returns void because it's too late for errors. */
+       /* changed to return int it may be interesting to know there
+          has been an error  --simo */
+static int tdb_unlock(TDB_CONTEXT *tdb, int list,
+                     int ltype __attribute__((unused)))
+{
+       int ret = -1;
+
+       if (tdb->flags & TDB_NOLOCK)
+               return 0;
+
+       /* Sanity checks */
+       if (list < -1 || list >= (int)tdb->header.hash_size) {
+               TDB_LOG((tdb, 0, "tdb_unlock: list %d invalid (%d)\n", list, 
tdb->header.hash_size));
+               return ret;
+       }
+
+       if (tdb->locked[list+1].count==0) {
+               TDB_LOG((tdb, 0, "tdb_unlock: count is 0\n"));
+               return ret;
+       }
+
+       if (tdb->locked[list+1].count == 1) {
+               /* Down to last nested lock: unlock underneath */
+               ret = tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 
0);
+       } else {
+               ret = 0;
+       }
+       tdb->locked[list+1].count--;
+
+       if (ret)
+               TDB_LOG((tdb, 0,"tdb_unlock: An error occurred unlocking!\n")); 
+       return ret;
+}
+
+/* This is based on the hash algorithm from gdbm */
+static u32 default_tdb_hash(TDB_DATA *key)
+{
+       u32 value;      /* Used to compute the hash value.  */
+       u32   i;        /* Used to cycle through random values. */
+
+       /* Set the initial value from the key size. */
+       for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
+               value = (value + (key->dptr[i] << (i*5 % 24)));
+
+       return (1103515243 * value + 12345);  
+}
+
+/* check for an out of bounds access - if it is out of bounds then
+   see if the database has been expanded by someone else and expand
+   if necessary 
+   note that "len" is the minimum length needed for the db
+*/
+static int tdb_oob(TDB_CONTEXT *tdb, tdb_off len, int probe)
+{
+       struct stat st;
+       if (len <= tdb->map_size)
+               return 0;
+       if (tdb->flags & TDB_INTERNAL) {
+               if (!probe) {
+                       /* Ensure ecode is set for log fn. */
+                       tdb->ecode = TDB_ERR_IO;
+                       TDB_LOG((tdb, 0,"tdb_oob len %d beyond internal malloc 
size %d\n",
+                                (int)len, (int)tdb->map_size));
+               }
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+       }
+
+       if (fstat(tdb->fd, &st) == -1)
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+
+       if (st.st_size < (off_t)len) {
+               if (!probe) {
+                       /* Ensure ecode is set for log fn. */
+                       tdb->ecode = TDB_ERR_IO;
+                       TDB_LOG((tdb, 0,"tdb_oob len %d beyond eof at %d\n",
+                                (int)len, (int)st.st_size));
+               }
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+       }
+
+       /* Unmap, update size, remap */
+       if (tdb_munmap(tdb) == -1)
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+       tdb->map_size = st.st_size;
+       tdb_mmap(tdb);
+       return 0;
+}
+
+/* write a lump of data at a specified offset */
+static int tdb_write(TDB_CONTEXT *tdb, tdb_off off, void *buf, tdb_len len)
+{
+       if (tdb_oob(tdb, off + len, 0) != 0)
+               return -1;
+
+       if (tdb->map_ptr)
+               memcpy(off + (char *)tdb->map_ptr, buf, len);
+#ifdef HAVE_PWRITE
+       else if (pwrite(tdb->fd, buf, len, off) != (ssize_t)len) {
+#else
+       else if (lseek(tdb->fd, off, SEEK_SET) != (off_t)off
+                || write(tdb->fd, buf, len) != (off_t)len) {
+#endif
+               /* Ensure ecode is set for log fn. */
+               tdb->ecode = TDB_ERR_IO;
+               TDB_LOG((tdb, 0,"tdb_write failed at %d len=%d (%s)\n",
+                          off, len, strerror(errno)));
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+       }
+       return 0;
+}
+
+/* read a lump of data at a specified offset, maybe convert */
+static int tdb_read(TDB_CONTEXT *tdb,tdb_off off,void *buf,tdb_len len,int cv)
+{
+       if (tdb_oob(tdb, off + len, 0) != 0)
+               return -1;
+
+       if (tdb->map_ptr)
+               memcpy(buf, off + (char *)tdb->map_ptr, len);
+#ifdef HAVE_PREAD
+       else if (pread(tdb->fd, buf, len, off) != (off_t)len) {
+#else
+       else if (lseek(tdb->fd, off, SEEK_SET) != (off_t)off
+                || read(tdb->fd, buf, len) != (off_t)len) {
+#endif
+               /* Ensure ecode is set for log fn. */
+               tdb->ecode = TDB_ERR_IO;
+               TDB_LOG((tdb, 0,"tdb_read failed at %d len=%d (%s)\n",
+                          off, len, strerror(errno)));
+               return TDB_ERRCODE(TDB_ERR_IO, -1);
+       }
+       if (cv)
+               convert(buf, len);
+       return 0;
+}
+
+/* don't allocate memory: used in tdb_delete path. */
+static int tdb_key_eq(TDB_CONTEXT *tdb, tdb_off off, TDB_DATA key)
+{
+       char buf[64];
+       u32 len;
+
+       if (tdb_oob(tdb, off + key.dsize, 0) != 0)
+               return -1;
+
+       if (tdb->map_ptr)
+               return !memcmp(off + (char*)tdb->map_ptr, key.dptr, key.dsize);
+
+       while (key.dsize) {
+               len = key.dsize;
+               if (len > sizeof(buf))
+                       len = sizeof(buf);
+               if (tdb_read(tdb, off, buf, len, 0) != 0)
+                       return -1;
+               if (memcmp(buf, key.dptr, len) != 0)
+                       return 0;
+               key.dptr += len;
+               key.dsize -= len;
+               off += len;
+       }
+       return 1;
+}
+
+/* read a lump of data, allocating the space for it */
+static char *tdb_alloc_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_len len)
+{
+       char *buf;
+
+       if (!(buf = talloc_size(tdb, len))) {
+               /* Ensure ecode is set for log fn. */
+               tdb->ecode = TDB_ERR_OOM;
+               TDB_LOG((tdb, 0,"tdb_alloc_read malloc failed len=%d (%s)\n",
+                          len, strerror(errno)));
+               return TDB_ERRCODE(TDB_ERR_OOM, buf);
+       }
+       if (tdb_read(tdb, offset, buf, len, 0) == -1) {
+               SAFE_FREE(buf);
+               return NULL;
+       }
+       return buf;
+}
+
+/* read/write a tdb_off */
+static int ofs_read(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
+{
+       return tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
+}
+static int ofs_write(TDB_CONTEXT *tdb, tdb_off offset, tdb_off *d)
+{
+       tdb_off off = *d;
+       return tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
+}
+
+/* read/write a record */
+static int rec_read(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
+{
+       if (tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
+               return -1;
+       if (TDB_BAD_MAGIC(rec)) {
+               /* Ensure ecode is set for log fn. */
+               tdb->ecode = TDB_ERR_CORRUPT;
+               TDB_LOG((tdb, 0,"rec_read bad magic 0x%x at offset=%d\n", 
rec->magic, offset));
+               return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+       }
+       return tdb_oob(tdb, rec->next+sizeof(*rec), 0);
+}
+static int rec_write(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
+{
+       struct list_struct r = *rec;
+       return tdb_write(tdb, offset, CONVERT(r), sizeof(r));
+}
+
+/* read a freelist record and check for simple errors */
+static int rec_free_read(TDB_CONTEXT *tdb, tdb_off off, struct list_struct 
*rec)
+{
+       if (tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
+               return -1;
+
+       if (rec->magic == TDB_MAGIC) {
+               /* this happens when a app is showdown while deleting a record 
- we should
+                  not completely fail when this happens */
+               TDB_LOG((tdb, 0,"rec_free_read non-free magic 0x%x at offset=%d 
- fixing\n", 
+                        rec->magic, off));
+               rec->magic = TDB_FREE_MAGIC;
+               if (tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
+                       return -1;
+       }
+
+       if (rec->magic != TDB_FREE_MAGIC) {
+               /* Ensure ecode is set for log fn. */
+               tdb->ecode = TDB_ERR_CORRUPT;
+               TDB_LOG((tdb, 0,"rec_free_read bad magic 0x%x at offset=%d\n", 
+                          rec->magic, off));
+               return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+       }
+       if (tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
+               return -1;
+       return 0;
+}
+
+/* update a record tailer (must hold allocation lock) */
+static int update_tailer(TDB_CONTEXT *tdb, tdb_off offset,
+                        const struct list_struct *rec)
+{
+       tdb_off totalsize;
+
+       /* Offset of tailer from record header */
+       totalsize = sizeof(*rec) + rec->rec_len;
+       return ofs_write(tdb, offset + totalsize - sizeof(tdb_off),
+                        &totalsize);
+}
+
+static tdb_off tdb_dump_record(TDB_CONTEXT *tdb, tdb_off offset)
+{
+       struct list_struct rec;
+       tdb_off tailer_ofs, tailer;
+
+       if (tdb_read(tdb, offset, (char *)&rec, sizeof(rec), DOCONV()) == -1) {
+               printf("ERROR: failed to read record at %u\n", offset);
+               return 0;
+       }
+
+       printf(" rec: offset=0x%08x next=0x%08x rec_len=%d key_len=%d 
data_len=%d full_hash=0x%x magic=0x%x\n",
+              offset, rec.next, rec.rec_len, rec.key_len, rec.data_len, 
rec.full_hash, rec.magic);
+
+       tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off);
+       if (ofs_read(tdb, tailer_ofs, &tailer) == -1) {
+               printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
+               return rec.next;
+       }
+
+       if (tailer != rec.rec_len + sizeof(rec)) {
+               printf("ERROR: tailer does not match record! tailer=%u 
totalsize=%u\n",
+                               (unsigned int)tailer, (unsigned 
int)(rec.rec_len + sizeof(rec)));
+       }
+       return rec.next;
+}
+
+static int tdb_dump_chain(TDB_CONTEXT *tdb, int i)
+{
+       tdb_off rec_ptr, top;
+
+       top = TDB_HASH_TOP(i);
+
+       if (tdb_lock(tdb, i, F_WRLCK) != 0)
+               return -1;
+
+       if (ofs_read(tdb, top, &rec_ptr) == -1)
+               return tdb_unlock(tdb, i, F_WRLCK);
+
+       if (rec_ptr)
+               printf("hash=%d\n", i);
+
+       while (rec_ptr) {
+               rec_ptr = tdb_dump_record(tdb, rec_ptr);
+       }
+
+       return tdb_unlock(tdb, i, F_WRLCK);
+}
+
+void tdb_dump_all(TDB_CONTEXT *tdb)
+{
+       unsigned int i;
+       for (i=0;i<tdb->header.hash_size;i++) {
+               tdb_dump_chain(tdb, i);
+       }
+       printf("freelist:\n");
+       tdb_dump_chain(tdb, -1);
+}
+
+int tdb_printfreelist(TDB_CONTEXT *tdb)
+{
+       int ret;
+       long total_free = 0;
+       tdb_off offset, rec_ptr;
+       struct list_struct rec;
+
+       if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
+               return ret;
+
+       offset = FREELIST_TOP;
+
+       /* read in the freelist top */
+       if (ofs_read(tdb, offset, &rec_ptr) == -1) {
+               tdb_unlock(tdb, -1, F_WRLCK);
+               return 0;
+       }
+
+       printf("freelist top=[0x%08x]\n", rec_ptr );
+       while (rec_ptr) {
+               if (tdb_read(tdb, rec_ptr, (char *)&rec, sizeof(rec), DOCONV()) 
== -1) {
+                       tdb_unlock(tdb, -1, F_WRLCK);
+                       return -1;
+               }
+
+               if (rec.magic != TDB_FREE_MAGIC) {
+                       printf("bad magic 0x%08x in free list\n", rec.magic);
+                       tdb_unlock(tdb, -1, F_WRLCK);
+                       return -1;
+               }
+
+               printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)] (end 
= 0x%08x)\n", 
+                      rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + 
rec.rec_len);
+               total_free += rec.rec_len;
+
+               /* move to the next record */
+               rec_ptr = rec.next;
+       }
+       printf("total rec_len = [0x%08x (%d)]\n", (int)total_free, 
+               (int)total_free);
+
+       return tdb_unlock(tdb, -1, F_WRLCK);
+}
+
+/* Remove an element from the freelist.  Must have alloc lock. */
+static int remove_from_freelist(TDB_CONTEXT *tdb, tdb_off off, tdb_off next)
+{
+       tdb_off last_ptr, i;
+
+       /* read in the freelist top */
+       last_ptr = FREELIST_TOP;
+       while (ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
+               if (i == off) {
+                       /* We've found it! */
+                       return ofs_write(tdb, last_ptr, &next);
+               }
+               /* Follow chain (next offset is at start of record) */
+               last_ptr = i;
+       }
+       TDB_LOG((tdb, 0,"remove_from_freelist: not on list at off=%d\n", off));
+       return TDB_ERRCODE(TDB_ERR_CORRUPT, -1);
+}
+
+/* Add an element into the freelist. Merge adjacent records if
+   neccessary. */
+static int tdb_free(TDB_CONTEXT *tdb, tdb_off offset, struct list_struct *rec)
+{
+       tdb_off right, left;
+
+       /* Allocation and tailer lock */
+       if (tdb_lock(tdb, -1, F_WRLCK) != 0)
+               return -1;
+
+       /* set an initial tailer, so if we fail we don't leave a bogus record */
+       if (update_tailer(tdb, offset, rec) != 0) {
+               TDB_LOG((tdb, 0, "tdb_free: upfate_tailer failed!\n"));
+               goto fail;
+       }
+
+       /* Look right first (I'm an Australian, dammit) */
+       right = offset + sizeof(*rec) + rec->rec_len;
+       if (right + sizeof(*rec) <= tdb->map_size) {
+               struct list_struct r;
+
+               if (tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
+                       TDB_LOG((tdb, 0, "tdb_free: right read failed at %u\n", 
right));
+                       goto left;
+               }
+
+               /* If it's free, expand to include it. */
+               if (r.magic == TDB_FREE_MAGIC) {
+                       if (remove_from_freelist(tdb, right, r.next) == -1) {
+                               TDB_LOG((tdb, 0, "tdb_free: right free failed 
at %u\n", right));
+                               goto left;
+                       }
+                       rec->rec_len += sizeof(r) + r.rec_len;
+               }
+       }
+
+left:
+       /* Look left */
+       left = offset - sizeof(tdb_off);
+       if (left > TDB_DATA_START(tdb->header.hash_size)) {
+               struct list_struct l;
+               tdb_off leftsize;
+               
+               /* Read in tailer and jump back to header */
+               if (ofs_read(tdb, left, &leftsize) == -1) {
+                       TDB_LOG((tdb, 0, "tdb_free: left offset read failed at 
%u\n", left));
+                       goto update;
+               }
+               left = offset - leftsize;
+
+               /* Now read in record */
+               if (tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
+                       TDB_LOG((tdb, 0, "tdb_free: left read failed at %u 
(%u)\n", left, leftsize));
+                       goto update;
+               }
+
+               /* If it's free, expand to include it. */
+               if (l.magic == TDB_FREE_MAGIC) {
+                       if (remove_from_freelist(tdb, left, l.next) == -1) {
+                               TDB_LOG((tdb, 0, "tdb_free: left free failed at 
%u\n", left));
+                               goto update;
+                       } else {
+                               offset = left;
+                               rec->rec_len += leftsize;
+                       }
+               }
+       }
+
+update:
+       if (update_tailer(tdb, offset, rec) == -1) {
+               TDB_LOG((tdb, 0, "tdb_free: update_tailer failed at %u\n", 
offset));
+               goto fail;
+       }
+
+       /* Now, prepend to free list */
+       rec->magic = TDB_FREE_MAGIC;
+
+       if (ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
+           rec_write(tdb, offset, rec) == -1 ||
+           ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
+               TDB_LOG((tdb, 0, "tdb_free record write failed at offset=%d\n", 
offset));
+               goto fail;
+       }
+
+       /* And we're done. */
+       tdb_unlock(tdb, -1, F_WRLCK);
+       return 0;
+
+ fail:
+       tdb_unlock(tdb, -1, F_WRLCK);
+       return -1;
+}
+
+
+/* expand a file.  we prefer to use ftruncate, as that is what posix
+  says to use for mmap expansion */
+static int expand_file(TDB_CONTEXT *tdb, tdb_off size, tdb_off addition)
+{
+       char buf[1024];
+#if HAVE_FTRUNCATE_EXTEND
+       if (ftruncate(tdb->fd, size+addition) != 0) {
+               TDB_LOG((tdb, 0, "expand_file ftruncate to %d failed (%s)\n", 
+                          size+addition, strerror(errno)));
+               return -1;
+       }
+#else
+       char b = 0;
+
+#ifdef HAVE_PWRITE
+       if (pwrite(tdb->fd,  &b, 1, (size+addition) - 1) != 1) {
+#else
+       if (lseek(tdb->fd, (size+addition) - 1, SEEK_SET) != 
(off_t)(size+addition) - 1 || 
+           write(tdb->fd, &b, 1) != 1) {
+#endif
+               TDB_LOG((tdb, 0, "expand_file to %d failed (%s)\n", 
+                          size+addition, strerror(errno)));
+               return -1;
+       }
+#endif
+
+       /* now fill the file with something. This ensures that the file isn't 
sparse, which would be
+          very bad if we ran out of disk. This must be done with write, not 
via mmap */
+       memset(buf, 0x42, sizeof(buf));
+       while (addition) {
+               int n = addition>sizeof(buf)?sizeof(buf):addition;
+#ifdef HAVE_PWRITE
+               int ret = pwrite(tdb->fd, buf, n, size);
+#else
+               int ret;
+               if (lseek(tdb->fd, size, SEEK_SET) != (off_t)size)
+                       return -1;
+               ret = write(tdb->fd, buf, n);
+#endif
+               if (ret != n) {
+                       TDB_LOG((tdb, 0, "expand_file write of %d failed 
(%s)\n", 
+                                  n, strerror(errno)));
+                       return -1;
+               }
+               addition -= n;
+               size += n;
+       }
+       return 0;
+}
+
+
+/* expand the database at least size bytes by expanding the underlying
+   file and doing the mmap again if necessary */
+static int tdb_expand(TDB_CONTEXT *tdb, tdb_off size)
+{
+       struct list_struct rec;
+       tdb_off offset;
+
+       if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
+               TDB_LOG((tdb, 0, "lock failed in tdb_expand\n"));
+               return -1;
+       }
+
+       /* must know about any previous expansions by another process */
+       tdb_oob(tdb, tdb->map_size + 1, 1);
+
+       /* always make room for at least 10 more records, and round
+           the database up to a multiple of TDB_PAGE_SIZE */
+       size = TDB_ALIGN(tdb->map_size + size*10, TDB_PAGE_SIZE) - 
tdb->map_size;
+
+       if (!(tdb->flags & TDB_INTERNAL))
+               tdb_munmap(tdb);
+
+       /*
+        * We must ensure the file is unmapped before doing this
+        * to ensure consistency with systems like OpenBSD where
+        * writes and mmaps are not consistent.
+        */
+
+       /* expand the file itself */
+       if (!(tdb->flags & TDB_INTERNAL)) {
+               if (expand_file(tdb, tdb->map_size, size) != 0)
+                       goto fail;
+       }
+
+       tdb->map_size += size;
+
+       if (tdb->flags & TDB_INTERNAL) {
+               char *new_map_ptr = talloc_realloc_size(tdb, tdb->map_ptr,
+                                                       tdb->map_size);
+               if (!new_map_ptr) {
+                       tdb->map_size -= size;
+                       goto fail;
+               }
+               tdb->map_ptr = new_map_ptr;
+       } else {
+               /*
+                * We must ensure the file is remapped before adding the space
+                * to ensure consistency with systems like OpenBSD where
+                * writes and mmaps are not consistent.
+                */
+
+               /* We're ok if the mmap fails as we'll fallback to read/write */
+               tdb_mmap(tdb);
+       }
+
+       /* form a new freelist record */
+       memset(&rec,'\0',sizeof(rec));
+       rec.rec_len = size - sizeof(rec);
+
+       /* link it into the free list */
+       offset = tdb->map_size - size;
+       if (tdb_free(tdb, offset, &rec) == -1)
+               goto fail;
+
+       tdb_unlock(tdb, -1, F_WRLCK);
+       return 0;
+ fail:
+       tdb_unlock(tdb, -1, F_WRLCK);
+       return -1;
+}
+
+
+/* 
+   the core of tdb_allocate - called when we have decided which
+   free list entry to use
+ */
+static tdb_off tdb_allocate_ofs(TDB_CONTEXT *tdb, tdb_len length, tdb_off 
rec_ptr,
+                               struct list_struct *rec, tdb_off last_ptr)
+{
+       struct list_struct newrec;
+       tdb_off newrec_ptr;
+
+       memset(&newrec, '\0', sizeof(newrec));
+
+       /* found it - now possibly split it up  */
+       if (rec->rec_len > length + MIN_REC_SIZE) {
+               /* Length of left piece */
+               length = TDB_ALIGN(length, TDB_ALIGNMENT);
+               
+               /* Right piece to go on free list */
+               newrec.rec_len = rec->rec_len - (sizeof(*rec) + length);
+               newrec_ptr = rec_ptr + sizeof(*rec) + length;
+               
+               /* And left record is shortened */
+               rec->rec_len = length;
+       } else {
+               newrec_ptr = 0;
+       }
+       
+       /* Remove allocated record from the free list */
+       if (ofs_write(tdb, last_ptr, &rec->next) == -1) {
+               return 0;
+       }
+       
+       /* Update header: do this before we drop alloc
+          lock, otherwise tdb_free() might try to
+          merge with us, thinking we're free.
+          (Thanks Jeremy Allison). */
+       rec->magic = TDB_MAGIC;
+       if (rec_write(tdb, rec_ptr, rec) == -1) {
+               return 0;
+       }
+       
+       /* Did we create new block? */
+       if (newrec_ptr) {
+               /* Update allocated record tailer (we
+                  shortened it). */
+               if (update_tailer(tdb, rec_ptr, rec) == -1) {
+                       return 0;
+               }
+               
+               /* Free new record */
+               if (tdb_free(tdb, newrec_ptr, &newrec) == -1) {
+                       return 0;
+               }
+       }
+       
+       /* all done - return the new record offset */
+       return rec_ptr;
+}
+
+/* allocate some space from the free list. The offset returned points
+   to a unconnected list_struct within the database with room for at
+   least length bytes of total data
+
+   0 is returned if the space could not be allocated
+ */
+static tdb_off tdb_allocate(TDB_CONTEXT *tdb, tdb_len length,
+                           struct list_struct *rec)
+{
+       tdb_off rec_ptr, last_ptr, newrec_ptr;
+       struct {
+               tdb_off rec_ptr, last_ptr;
+               tdb_len rec_len;
+       } bestfit = { 0, 0, 0 };
+
+       if (tdb_lock(tdb, -1, F_WRLCK) == -1)
+               return 0;
+
+       /* Extra bytes required for tailer */
+       length += sizeof(tdb_off);
+
+ again:
+       last_ptr = FREELIST_TOP;
+
+       /* read in the freelist top */
+       if (ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
+               goto fail;
+
+       bestfit.rec_ptr = 0;
+
+       /* 
+          this is a best fit allocation strategy. Originally we used
+          a first fit strategy, but it suffered from massive fragmentation
+          issues when faced with a slowly increasing record size.
+        */
+       while (rec_ptr) {
+               if (rec_free_read(tdb, rec_ptr, rec) == -1) {
+                       goto fail;
+               }
+
+               if (rec->rec_len >= length) {
+                       if (bestfit.rec_ptr == 0 ||
+                           rec->rec_len < bestfit.rec_len) {
+                               bestfit.rec_len = rec->rec_len;
+                               bestfit.rec_ptr = rec_ptr;
+                               bestfit.last_ptr = last_ptr;
+                               /* consider a fit to be good enough if we 
aren't wasting more than half the space */
+                               if (bestfit.rec_len < 2*length) {
+                                       break;
+                               }
+                       }
+               }
+
+               /* move to the next record */
+               last_ptr = rec_ptr;
+               rec_ptr = rec->next;
+       }
+
+       if (bestfit.rec_ptr != 0) {
+               if (rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
+                       goto fail;
+               }
+
+               newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr, 
rec, bestfit.last_ptr);
+               tdb_unlock(tdb, -1, F_WRLCK);
+               return newrec_ptr;
+       }
+
+       /* we didn't find enough space. See if we can expand the
+          database and if we can then try again */
+       if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
+               goto again;
+ fail:
+       tdb_unlock(tdb, -1, F_WRLCK);
+       return 0;
+}
+
+/* initialise a new database with a specified hash size */
+static int tdb_new_database(TDB_CONTEXT *tdb, int hash_size)
+{
+       struct tdb_header *newdb;
+       int size, ret = -1;
+
+       /* We make it up in memory, then write it out if not internal */
+       size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off);
+       if (!(newdb = talloc_zero_size(tdb, size)))
+               return TDB_ERRCODE(TDB_ERR_OOM, -1);
+
+       /* Fill in the header */
+       newdb->version = TDB_VERSION;
+       newdb->hash_size = hash_size;
+       if (tdb->flags & TDB_INTERNAL) {
+               tdb->map_size = size;
+               tdb->map_ptr = (char *)newdb;
+               memcpy(&tdb->header, newdb, sizeof(tdb->header));
+               /* Convert the `ondisk' version if asked. */
+               CONVERT(*newdb);
+               return 0;
+       }
+       if (lseek(tdb->fd, 0, SEEK_SET) == -1)
+               goto fail;
+
+       if (ftruncate(tdb->fd, 0) == -1)
+               goto fail;
+
+       /* This creates an endian-converted header, as if read from disk */
+       CONVERT(*newdb);
+       memcpy(&tdb->header, newdb, sizeof(tdb->header));
+       /* Don't endian-convert the magic food! */
+       memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
+       if (write(tdb->fd, newdb, size) != size)
+               ret = -1;
+       else
+               ret = 0;
+
+  fail:
+       SAFE_FREE(newdb);
+       return ret;
+}
+
+/* Returns 0 on fail.  On success, return offset of record, and fills
+   in rec */
+static tdb_off tdb_find(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash,
+                       struct list_struct *r)
+{
+       tdb_off rec_ptr;
+       
+       /* read in the hash top */
+       if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
+               return 0;
+
+       /* keep looking until we find the right record */
+       while (rec_ptr) {
+               if (rec_read(tdb, rec_ptr, r) == -1)
+                       return 0;
+
+               if (!TDB_DEAD(r) && hash==r->full_hash && 
key.dsize==r->key_len) {
+                       /* a very likely hit - read the key */
+                       int cmp = tdb_key_eq(tdb, rec_ptr + sizeof(*r), key);
+                       if (cmp < 0)
+                               return 0;
+                       else if (cmp > 0)
+                               return rec_ptr;
+               }
+               rec_ptr = r->next;
+       }
+       return TDB_ERRCODE(TDB_ERR_NOEXIST, 0);
+}
+
+/* As tdb_find, but if you succeed, keep the lock */
+static tdb_off tdb_find_lock_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, 
int locktype,
+                            struct list_struct *rec)
+{
+       u32 rec_ptr;
+
+       if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
+               return 0;
+       if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
+               tdb_unlock(tdb, BUCKET(hash), locktype);
+       return rec_ptr;
+}
+
+enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb)
+{
+       return tdb->ecode;
+}
+
+static struct tdb_errname {
+       enum TDB_ERROR ecode; const char *estring;
+} emap[] = { {TDB_SUCCESS, "Success"},
+            {TDB_ERR_CORRUPT, "Corrupt database"},
+            {TDB_ERR_IO, "IO Error"},
+            {TDB_ERR_LOCK, "Locking error"},
+            {TDB_ERR_OOM, "Out of memory"},
+            {TDB_ERR_EXISTS, "Record exists"},
+            {TDB_ERR_NOLOCK, "Lock exists on other keys"},
+            {TDB_ERR_NOEXIST, "Record does not exist"} };
+
+/* Error string for the last tdb error */
+const char *tdb_errorstr(TDB_CONTEXT *tdb)
+{
+       u32 i;
+       for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
+               if (tdb->ecode == emap[i].ecode)
+                       return emap[i].estring;
+       return "Invalid error code";
+}
+
+/* update an entry in place - this only works if the new data size
+   is <= the old data size and the key exists.
+   on failure return -1.
+*/
+
+static int tdb_update_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, TDB_DATA 
dbuf)
+{
+       struct list_struct rec;
+       tdb_off rec_ptr;
+
+       /* find entry */
+       if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
+               return -1;
+
+       /* must be long enough key, data and tailer */
+       if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off)) {
+               tdb->ecode = TDB_SUCCESS; /* Not really an error */
+               return -1;
+       }
+
+       if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+                     dbuf.dptr, dbuf.dsize) == -1)
+               return -1;
+
+       if (dbuf.dsize != rec.data_len) {
+               /* update size */
+               rec.data_len = dbuf.dsize;
+               return rec_write(tdb, rec_ptr, &rec);
+       }
+ 
+       return 0;
+}
+
+/* find an entry in the database given a key */
+/* If an entry doesn't exist tdb_err will be set to
+ * TDB_ERR_NOEXIST. If a key has no data attached
+ * then the TDB_DATA will have zero length but
+ * a non-zero pointer
+ */
+
+TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       tdb_off rec_ptr;
+       struct list_struct rec;
+       TDB_DATA ret;
+       u32 hash;
+
+       /* find which hash bucket it is in */
+       hash = tdb->hash_fn(&key);
+       if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
+               return tdb_null;
+
+       ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+                                 rec.data_len);
+       ret.dsize = rec.data_len;
+       tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+       return ret;
+}
+
+/* check if an entry in the database exists 
+
+   note that 1 is returned if the key is found and 0 is returned if not found
+   this doesn't match the conventions in the rest of this module, but is
+   compatible with gdbm
+*/
+static int tdb_exists_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
+{
+       struct list_struct rec;
+       
+       if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
+               return 0;
+       tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
+       return 1;
+}
+
+int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       u32 hash = tdb->hash_fn(&key);
+       return tdb_exists_hash(tdb, key, hash);
+}
+
+/* record lock stops delete underneath */
+static int lock_record(TDB_CONTEXT *tdb, tdb_off off)
+{
+       return off ? tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
+}
+/*
+  Write locks override our own fcntl readlocks, so check it here.
+  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+  an error to fail to get the lock here.
+*/
+ 
+static int write_lock_record(TDB_CONTEXT *tdb, tdb_off off)
+{
+       struct tdb_traverse_lock *i;
+       for (i = &tdb->travlocks; i; i = i->next)
+               if (i->off == off)
+                       return -1;
+       return tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
+}
+
+/*
+  Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
+  an error to fail to get the lock here.
+*/
+
+static int write_unlock_record(TDB_CONTEXT *tdb, tdb_off off)
+{
+       return tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
+}
+/* fcntl locks don't stack: avoid unlocking someone else's */
+static int unlock_record(TDB_CONTEXT *tdb, tdb_off off)
+{
+       struct tdb_traverse_lock *i;
+       u32 count = 0;
+
+       if (off == 0)
+               return 0;
+       for (i = &tdb->travlocks; i; i = i->next)
+               if (i->off == off)
+                       count++;
+       return (count == 1 ? tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
+}
+
+/* actually delete an entry in the database given the offset */
+static int do_delete(TDB_CONTEXT *tdb, tdb_off rec_ptr, struct list_struct*rec)
+{
+       tdb_off last_ptr, i;
+       struct list_struct lastrec;
+
+       if (tdb->read_only) return -1;
+
+       if (write_lock_record(tdb, rec_ptr) == -1) {
+               /* Someone traversing here: mark it as dead */
+               rec->magic = TDB_DEAD_MAGIC;
+               return rec_write(tdb, rec_ptr, rec);
+       }
+       if (write_unlock_record(tdb, rec_ptr) != 0)
+               return -1;
+
+       /* find previous record in hash chain */
+       if (ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
+               return -1;
+       for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
+               if (rec_read(tdb, i, &lastrec) == -1)
+                       return -1;
+
+       /* unlink it: next ptr is at start of record. */
+       if (last_ptr == 0)
+               last_ptr = TDB_HASH_TOP(rec->full_hash);
+       if (ofs_write(tdb, last_ptr, &rec->next) == -1)
+               return -1;
+
+       /* recover the space */
+       if (tdb_free(tdb, rec_ptr, rec) == -1)
+               return -1;
+       return 0;
+}
+
+/* Uses traverse lock: 0 = finish, -1 = error, other = record offset */
+static int tdb_next_lock(TDB_CONTEXT *tdb, struct tdb_traverse_lock *tlock,
+                        struct list_struct *rec)
+{
+       int want_next = (tlock->off != 0);
+
+       /* Lock each chain from the start one. */
+       for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
+
+               /* this is an optimisation for the common case where
+                  the hash chain is empty, which is particularly
+                  common for the use of tdb with ldb, where large
+                  hashes are used. In that case we spend most of our
+                  time in tdb_brlock(), locking empty hash chains.
+
+                  To avoid this, we do an unlocked pre-check to see
+                  if the hash chain is empty before starting to look
+                  inside it. If it is empty then we can avoid that
+                  hash chain. If it isn't empty then we can't believe
+                  the value we get back, as we read it without a
+                  lock, so instead we get the lock and re-fetch the
+                  value below.
+
+                  Notice that not doing this optimisation on the
+                  first hash chain is critical. We must guarantee
+                  that we have done at least one fcntl lock at the
+                  start of a search to guarantee that memory is
+                  coherent on SMP systems. If records are added by
+                  others during the search then thats OK, and we
+                  could possibly miss those with this trick, but we
+                  could miss them anyway without this trick, so the
+                  semantics don't change.
+
+                  With a non-indexed ldb search this trick gains us a
+                  factor of around 80 in speed on a linux 2.6.x
+                  system (testing using ldbtest).
+                */
+               if (!tlock->off && tlock->hash != 0) {
+                       u32 off;
+                       if (tdb->map_ptr) {
+                               for (;tlock->hash < 
tdb->header.hash_size;tlock->hash++) {
+                                       if (0 != *(u32 
*)(TDB_HASH_TOP(tlock->hash) + (unsigned char *)tdb->map_ptr)) {
+                                               break;
+                                       }
+                               }
+                               if (tlock->hash == tdb->header.hash_size) {
+                                       continue;
+                               }
+                       } else {
+                               if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash), 
&off) == 0 &&
+                                   off == 0) {
+                                       continue;
+                               }
+                       }
+               }
+
+               if (tdb_lock(tdb, tlock->hash, F_WRLCK) == -1)
+                       return -1;
+
+               /* No previous record?  Start at top of chain. */
+               if (!tlock->off) {
+                       if (ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
+                                    &tlock->off) == -1)
+                               goto fail;
+               } else {
+                       /* Otherwise unlock the previous record. */
+                       if (unlock_record(tdb, tlock->off) != 0)
+                               goto fail;
+               }
+
+               if (want_next) {
+                       /* We have offset of old record: grab next */
+                       if (rec_read(tdb, tlock->off, rec) == -1)
+                               goto fail;
+                       tlock->off = rec->next;
+               }
+
+               /* Iterate through chain */
+               while( tlock->off) {
+                       tdb_off current;
+                       if (rec_read(tdb, tlock->off, rec) == -1)
+                               goto fail;
+
+                       /* Detect infinite loops. From "Shlomi Yaakobovich" 
<Shlomi@xxxxxxxxxx>. */
+                       if (tlock->off == rec->next) {
+                               TDB_LOG((tdb, 0, "tdb_next_lock: loop 
detected.\n"));
+                               goto fail;
+                       }
+
+                       if (!TDB_DEAD(rec)) {
+                               /* Woohoo: we found one! */
+                               if (lock_record(tdb, tlock->off) != 0)
+                                       goto fail;
+                               return tlock->off;
+                       }
+
+                       /* Try to clean dead ones from old traverses */
+                       current = tlock->off;
+                       tlock->off = rec->next;
+                       if (!tdb->read_only && 
+                           do_delete(tdb, current, rec) != 0)
+                               goto fail;
+               }
+               tdb_unlock(tdb, tlock->hash, F_WRLCK);
+               want_next = 0;
+       }
+       /* We finished iteration without finding anything */
+       return TDB_ERRCODE(TDB_SUCCESS, 0);
+
+ fail:
+       tlock->off = 0;
+       if (tdb_unlock(tdb, tlock->hash, F_WRLCK) != 0)
+               TDB_LOG((tdb, 0, "tdb_next_lock: On error unlock failed!\n"));
+       return -1;
+}
+
+/* traverse the entire database - calling fn(tdb, key, data) on each element.
+   return -1 on error or the record count traversed
+   if fn is NULL then it is not called
+   a non-zero return value from fn() indicates that the traversal should stop
+  */
+int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *private)
+{
+       TDB_DATA key, dbuf;
+       struct list_struct rec;
+       struct tdb_traverse_lock tl = { NULL, 0, 0 };
+       int ret, count = 0;
+
+       /* This was in the initializaton, above, but the IRIX compiler
+        * did not like it.  crh
+        */
+       tl.next = tdb->travlocks.next;
+
+       /* fcntl locks don't stack: beware traverse inside traverse */
+       tdb->travlocks.next = &tl;
+
+       /* tdb_next_lock places locks on the record returned, and its chain */
+       while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
+               count++;
+               /* now read the full record */
+               key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec), 
+                                         rec.key_len + rec.data_len);
+               if (!key.dptr) {
+                       ret = -1;
+                       if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0)
+                               goto out;
+                       if (unlock_record(tdb, tl.off) != 0)
+                               TDB_LOG((tdb, 0, "tdb_traverse: key.dptr == 
NULL and unlock_record failed!\n"));
+                       goto out;
+               }
+               key.dsize = rec.key_len;
+               dbuf.dptr = key.dptr + rec.key_len;
+               dbuf.dsize = rec.data_len;
+
+               /* Drop chain lock, call out */
+               if (tdb_unlock(tdb, tl.hash, F_WRLCK) != 0) {
+                       ret = -1;
+                       goto out;
+               }
+               if (fn && fn(tdb, key, dbuf, private)) {
+                       /* They want us to terminate traversal */
+                       ret = count;
+                       if (unlock_record(tdb, tl.off) != 0) {
+                               TDB_LOG((tdb, 0, "tdb_traverse: unlock_record 
failed!\n"));;
+                               ret = -1;
+                       }
+                       tdb->travlocks.next = tl.next;
+                       SAFE_FREE(key.dptr);
+                       return count;
+               }
+               SAFE_FREE(key.dptr);
+       }
+out:
+       tdb->travlocks.next = tl.next;
+       if (ret < 0)
+               return -1;
+       else
+               return count;
+}
+
+/* find the first entry in the database and return its key */
+TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
+{
+       TDB_DATA key;
+       struct list_struct rec;
+
+       /* release any old lock */
+       if (unlock_record(tdb, tdb->travlocks.off) != 0)
+               return tdb_null;
+       tdb->travlocks.off = tdb->travlocks.hash = 0;
+
+       if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
+               return tdb_null;
+       /* now read the key */
+       key.dsize = rec.key_len;
+       key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
+       if (tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK) != 0)
+               TDB_LOG((tdb, 0, "tdb_firstkey: error occurred while 
tdb_unlocking!\n"));
+       return key;
+}
+
+/* find the next entry in the database, returning its key */
+TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
+{
+       u32 oldhash;
+       TDB_DATA key = tdb_null;
+       struct list_struct rec;
+       char *k = NULL;
+
+       /* Is locked key the old key?  If so, traverse will be reliable. */
+       if (tdb->travlocks.off) {
+               if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
+                       return tdb_null;
+               if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
+                   || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
+                                           rec.key_len))
+                   || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
+                       /* No, it wasn't: unlock it and start from scratch */
+                       if (unlock_record(tdb, tdb->travlocks.off) != 0)
+                               return tdb_null;
+                       if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
+                               return tdb_null;
+                       tdb->travlocks.off = 0;
+               }
+
+               SAFE_FREE(k);
+       }
+
+       if (!tdb->travlocks.off) {
+               /* No previous element: do normal find, and lock record */
+               tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, 
tdb->hash_fn(&oldkey), F_WRLCK, &rec);
+               if (!tdb->travlocks.off)
+                       return tdb_null;
+               tdb->travlocks.hash = BUCKET(rec.full_hash);
+               if (lock_record(tdb, tdb->travlocks.off) != 0) {
+                       TDB_LOG((tdb, 0, "tdb_nextkey: lock_record failed 
(%s)!\n", strerror(errno)));
+                       return tdb_null;
+               }
+       }
+       oldhash = tdb->travlocks.hash;
+
+       /* Grab next record: locks chain and returned record,
+          unlocks old record */
+       if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
+               key.dsize = rec.key_len;
+               key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
+                                         key.dsize);
+               /* Unlock the chain of this new record */
+               if (tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK) != 0)
+                       TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock 
failed!\n"));
+       }
+       /* Unlock the chain of old record */
+       if (tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK) != 0)
+               TDB_LOG((tdb, 0, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
+       return key;
+}
+
+/* delete an entry in the database given a key */
+static int tdb_delete_hash(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash)
+{
+       tdb_off rec_ptr;
+       struct list_struct rec;
+       int ret;
+
+       if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK, &rec)))
+               return -1;
+       ret = do_delete(tdb, rec_ptr, &rec);
+       if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
+               TDB_LOG((tdb, 0, "tdb_delete: WARNING tdb_unlock failed!\n"));
+       return ret;
+}
+
+int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       u32 hash = tdb->hash_fn(&key);
+       return tdb_delete_hash(tdb, key, hash);
+}
+
+/* store an element in the database, replacing any existing element
+   with the same key 
+
+   return 0 on success, -1 on failure
+*/
+int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
+{
+       struct list_struct rec;
+       u32 hash;
+       tdb_off rec_ptr;
+       char *p = NULL;
+       int ret = 0;
+
+       /* find which hash bucket it is in */
+       hash = tdb->hash_fn(&key);
+       if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+               return -1;
+
+       /* check for it existing, on insert. */
+       if (flag == TDB_INSERT) {
+               if (tdb_exists_hash(tdb, key, hash)) {
+                       tdb->ecode = TDB_ERR_EXISTS;
+                       goto fail;
+               }
+       } else {
+               /* first try in-place update, on modify or replace. */
+               if (tdb_update_hash(tdb, key, hash, dbuf) == 0)
+                       goto out;
+               if (tdb->ecode == TDB_ERR_NOEXIST &&
+                   flag == TDB_MODIFY) {
+                       /* if the record doesn't exist and we are in TDB_MODIFY 
mode then
+                        we should fail the store */
+                       goto fail;
+               }
+       }
+       /* reset the error code potentially set by the tdb_update() */
+       tdb->ecode = TDB_SUCCESS;
+
+       /* delete any existing record - if it doesn't exist we don't
+           care.  Doing this first reduces fragmentation, and avoids
+           coalescing with `allocated' block before it's updated. */
+       if (flag != TDB_INSERT)
+               tdb_delete_hash(tdb, key, hash);
+
+       /* Copy key+value *before* allocating free space in case malloc
+          fails and we are left with a dead spot in the tdb. */
+
+       if (!(p = (char *)talloc_size(tdb, key.dsize + dbuf.dsize))) {
+               tdb->ecode = TDB_ERR_OOM;
+               goto fail;
+       }
+
+       memcpy(p, key.dptr, key.dsize);
+       if (dbuf.dsize)
+               memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
+
+       /* we have to allocate some space */
+       if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
+               goto fail;
+
+       /* Read hash top into next ptr */
+       if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
+               goto fail;
+
+       rec.key_len = key.dsize;
+       rec.data_len = dbuf.dsize;
+       rec.full_hash = hash;
+       rec.magic = TDB_MAGIC;
+
+       /* write out and point the top of the hash chain at it */
+       if (rec_write(tdb, rec_ptr, &rec) == -1
+           || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
+           || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
+               /* Need to tdb_unallocate() here */
+               goto fail;
+       }
+ out:
+       SAFE_FREE(p); 
+       tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+       return ret;
+fail:
+       ret = -1;
+       goto out;
+}
+
+/* Attempt to append data to an entry in place - this only works if the new 
data size
+   is <= the old data size and the key exists.
+   on failure return -1. Record must be locked before calling.
+*/
+static int tdb_append_inplace(TDB_CONTEXT *tdb, TDB_DATA key, u32 hash, 
TDB_DATA new_dbuf)
+{
+       struct list_struct rec;
+       tdb_off rec_ptr;
+
+       /* find entry */
+       if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
+               return -1;
+
+       /* Append of 0 is always ok. */
+       if (new_dbuf.dsize == 0)
+               return 0;
+
+       /* must be long enough for key, old data + new data and tailer */
+       if (rec.rec_len < key.dsize + rec.data_len + new_dbuf.dsize + 
sizeof(tdb_off)) {
+               /* No room. */
+               tdb->ecode = TDB_SUCCESS; /* Not really an error */
+               return -1;
+       }
+
+       if (tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len + rec.data_len,
+                     new_dbuf.dptr, new_dbuf.dsize) == -1)
+               return -1;
+
+       /* update size */
+       rec.data_len += new_dbuf.dsize;
+       return rec_write(tdb, rec_ptr, &rec);
+}
+
+/* Append to an entry. Create if not exist. */
+
+int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+{
+       struct list_struct rec;
+       u32 hash;
+       tdb_off rec_ptr;
+       char *p = NULL;
+       int ret = 0;
+       size_t new_data_size = 0;
+
+       /* find which hash bucket it is in */
+       hash = tdb->hash_fn(&key);
+       if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
+               return -1;
+
+       /* first try in-place. */
+       if (tdb_append_inplace(tdb, key, hash, new_dbuf) == 0)
+               goto out;
+
+       /* reset the error code potentially set by the tdb_append_inplace() */
+       tdb->ecode = TDB_SUCCESS;
+
+       /* find entry */
+       if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
+               if (tdb->ecode != TDB_ERR_NOEXIST)
+                       goto fail;
+
+               /* Not found - create. */
+
+               ret = tdb_store(tdb, key, new_dbuf, TDB_INSERT);
+               goto out;
+       }
+
+       new_data_size = rec.data_len + new_dbuf.dsize;
+
+       /* Copy key+old_value+value *before* allocating free space in case 
malloc
+          fails and we are left with a dead spot in the tdb. */
+
+       if (!(p = (char *)talloc_size(tdb, key.dsize + new_data_size))) {
+               tdb->ecode = TDB_ERR_OOM;
+               goto fail;
+       }
+
+       /* Copy the key in place. */
+       memcpy(p, key.dptr, key.dsize);
+
+       /* Now read the old data into place. */
+       if (rec.data_len &&
+               tdb_read(tdb, rec_ptr + sizeof(rec) + rec.key_len, p + 
key.dsize, rec.data_len, 0) == -1)
+                       goto fail;
+
+       /* Finally append the new data. */
+       if (new_dbuf.dsize)
+               memcpy(p+key.dsize+rec.data_len, new_dbuf.dptr, new_dbuf.dsize);
+
+       /* delete any existing record - if it doesn't exist we don't
+           care.  Doing this first reduces fragmentation, and avoids
+           coalescing with `allocated' block before it's updated. */
+
+       tdb_delete_hash(tdb, key, hash);
+
+       if (!(rec_ptr = tdb_allocate(tdb, key.dsize + new_data_size, &rec)))
+               goto fail;
+
+       /* Read hash top into next ptr */
+       if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
+               goto fail;
+
+       rec.key_len = key.dsize;
+       rec.data_len = new_data_size;
+       rec.full_hash = hash;
+       rec.magic = TDB_MAGIC;
+
+       /* write out and point the top of the hash chain at it */
+       if (rec_write(tdb, rec_ptr, &rec) == -1
+           || tdb_write(tdb, rec_ptr+sizeof(rec), p, 
key.dsize+new_data_size)==-1
+           || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
+               /* Need to tdb_unallocate() here */
+               goto fail;
+       }
+
+ out:
+       SAFE_FREE(p); 
+       tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
+       return ret;
+
+fail:
+       ret = -1;
+       goto out;
+}
+
+static int tdb_already_open(dev_t device,
+                           ino_t ino)
+{
+       TDB_CONTEXT *i;
+       
+       for (i = tdbs; i; i = i->next) {
+               if (i->device == device && i->inode == ino) {
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
+/* open the database, creating it if necessary 
+
+   The open_flags and mode are passed straight to the open call on the
+   database file. A flags value of O_WRONLY is invalid. The hash size
+   is advisory, use zero for a default value.
+
+   Return is NULL on error, in which case errno is also set.  Don't 
+   try to call tdb_error or tdb_errname, just do strerror(errno).
+
+   @param name may be NULL for internal databases. */
+TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
+                     int open_flags, mode_t mode)
+{
+       return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, 
NULL);
+}
+
+/* a default logging function */
+static void null_log_fn(TDB_CONTEXT *tdb __attribute__((unused)),
+                       int level __attribute__((unused)),
+                       const char *fmt __attribute__((unused)), ...)
+{
+}
+
+
+TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+                        int open_flags, mode_t mode,
+                        tdb_log_func log_fn,
+                        tdb_hash_func hash_fn)
+{
+       TDB_CONTEXT *tdb;
+       struct stat st;
+       int rev = 0, locked = 0;
+       uint8_t *vp;
+       u32 vertest;
+
+       if (!(tdb = talloc_zero(name, TDB_CONTEXT))) {
+               /* Can't log this */
+               errno = ENOMEM;
+               goto fail;
+       }
+       tdb->fd = -1;
+       tdb->name = NULL;
+       tdb->map_ptr = NULL;
+       tdb->flags = tdb_flags;
+       tdb->open_flags = open_flags;
+       tdb->log_fn = log_fn?log_fn:null_log_fn;
+       tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
+
+       if ((open_flags & O_ACCMODE) == O_WRONLY) {
+               TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
+                        name));
+               errno = EINVAL;
+               goto fail;
+       }
+       
+       if (hash_size == 0)
+               hash_size = DEFAULT_HASH_SIZE;
+       if ((open_flags & O_ACCMODE) == O_RDONLY) {
+               tdb->read_only = 1;
+               /* read only databases don't do locking or clear if first */
+               tdb->flags |= TDB_NOLOCK;
+               tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+       }
+
+       /* internal databases don't mmap or lock, and start off cleared */
+       if (tdb->flags & TDB_INTERNAL) {
+               tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
+               tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+               if (tdb_new_database(tdb, hash_size) != 0) {
+                       TDB_LOG((tdb, 0, "tdb_open_ex: tdb_new_database 
failed!"));
+                       goto fail;
+               }
+               goto internal;
+       }
+
+       if ((tdb->fd = open(name, open_flags, mode)) == -1) {
+               TDB_LOG((tdb, 5, "tdb_open_ex: could not open file %s: %s\n",
+                        name, strerror(errno)));
+               goto fail;      /* errno set by open(2) */
+       }
+
+       /* ensure there is only one process initialising at once */
+       if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
+               TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: 
%s\n",
+                        name, strerror(errno)));
+               goto fail;      /* errno set by tdb_brlock */
+       }
+
+       /* we need to zero database if we are the only one with it open */
+       if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
+               (locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 
0))) {
+               open_flags |= O_CREAT;
+               if (ftruncate(tdb->fd, 0) == -1) {
+                       TDB_LOG((tdb, 0, "tdb_open_ex: "
+                                "failed to truncate %s: %s\n",
+                                name, strerror(errno)));
+                       goto fail; /* errno set by ftruncate */
+               }
+       }
+
+       if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != 
sizeof(tdb->header)
+           || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
+           || (tdb->header.version != TDB_VERSION
+               && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
+               /* its not a valid database - possibly initialise it */
+               if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) 
== -1) {
+                       errno = EIO; /* ie bad format or something */
+                       goto fail;
+               }
+               rev = (tdb->flags & TDB_CONVERT);
+       }
+       vp = (uint8_t *)&tdb->header.version;
+       vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
+                 (((u32)vp[2]) << 8) | (u32)vp[3];
+       tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
+       if (!rev)
+               tdb->flags &= ~TDB_CONVERT;
+       else {
+               tdb->flags |= TDB_CONVERT;
+               convert(&tdb->header, sizeof(tdb->header));
+       }
+       if (fstat(tdb->fd, &st) == -1)
+               goto fail;
+
+       /* Is it already in the open list?  If so, fail. */
+       if (tdb_already_open(st.st_dev, st.st_ino)) {
+               TDB_LOG((tdb, 2, "tdb_open_ex: "
+                        "%s (%d,%d) is already open in this process\n",
+                        name, (int)st.st_dev, (int)st.st_ino));
+               errno = EBUSY;
+               goto fail;
+       }
+
+       if (!(tdb->name = (char *)talloc_strdup(tdb, name))) {
+               errno = ENOMEM;
+               goto fail;
+       }
+
+       tdb->map_size = st.st_size;
+       tdb->device = st.st_dev;
+       tdb->inode = st.st_ino;
+       tdb->locked = talloc_zero_array(tdb, struct tdb_lock_type,
+                                       tdb->header.hash_size+1);
+       if (!tdb->locked) {
+               TDB_LOG((tdb, 2, "tdb_open_ex: "
+                        "failed to allocate lock structure for %s\n",
+                        name));
+               errno = ENOMEM;
+               goto fail;
+       }
+       tdb_mmap(tdb);
+       if (locked) {
+               if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
+                       TDB_LOG((tdb, 0, "tdb_open_ex: "
+                                "failed to take ACTIVE_LOCK on %s: %s\n",
+                                name, strerror(errno)));
+                       goto fail;
+               }
+
+       }
+
+       /* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
+          we didn't get the initial exclusive lock as we need to let all other
+          users know we're using it. */
+
+       if (tdb_flags & TDB_CLEAR_IF_FIRST) {
+       /* leave this lock in place to indicate it's in use */
+       if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
+               goto fail;
+       }
+
+
+ internal:
+       /* Internal (memory-only) databases skip all the code above to
+        * do with disk files, and resume here by releasing their
+        * global lock and hooking into the active list. */
+       if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
+               goto fail;
+       tdb->next = tdbs;
+       tdbs = tdb;
+       return tdb;
+
+ fail:
+       { int save_errno = errno;
+
+       if (!tdb)
+               return NULL;
+       
+       if (tdb->map_ptr) {
+               if (tdb->flags & TDB_INTERNAL)
+                       SAFE_FREE(tdb->map_ptr);
+               else
+                       tdb_munmap(tdb);
+       }
+       SAFE_FREE(tdb->name);
+       if (tdb->fd != -1)
+               if (close(tdb->fd) != 0)
+                       TDB_LOG((tdb, 5, "tdb_open_ex: failed to close tdb->fd 
on error!\n"));
+       SAFE_FREE(tdb->locked);
+       SAFE_FREE(tdb);
+       errno = save_errno;
+       return NULL;
+       }
+}
+
+/**
+ * Close a database.
+ *
+ * @returns -1 for error; 0 for success.
+ **/
+int tdb_close(TDB_CONTEXT *tdb)
+{
+       TDB_CONTEXT **i;
+       int ret = 0;
+
+       if (tdb->map_ptr) {
+               if (tdb->flags & TDB_INTERNAL)
+                       SAFE_FREE(tdb->map_ptr);
+               else
+                       tdb_munmap(tdb);
+       }
+       SAFE_FREE(tdb->name);
+       if (tdb->fd != -1)
+               ret = close(tdb->fd);
+       SAFE_FREE(tdb->locked);
+
+       /* Remove from contexts list */
+       for (i = &tdbs; *i; i = &(*i)->next) {
+               if (*i == tdb) {
+                       *i = tdb->next;
+                       break;
+               }
+       }
+
+       memset(tdb, 0, sizeof(*tdb));
+       SAFE_FREE(tdb);
+
+       return ret;
+}
+
+/* lock/unlock entire database */
+int tdb_lockall(TDB_CONTEXT *tdb)
+{
+       u32 i;
+
+       /* There are no locks on read-only dbs */
+       if (tdb->read_only)
+               return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+       for (i = 0; i < tdb->header.hash_size; i++) 
+               if (tdb_lock(tdb, i, F_WRLCK))
+                       break;
+
+       /* If error, release locks we have... */
+       if (i < tdb->header.hash_size) {
+               u32 j;
+
+               for ( j = 0; j < i; j++)
+                       tdb_unlock(tdb, j, F_WRLCK);
+               return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
+       }
+
+       return 0;
+}
+void tdb_unlockall(TDB_CONTEXT *tdb)
+{
+       u32 i;
+       for (i=0; i < tdb->header.hash_size; i++)
+               tdb_unlock(tdb, i, F_WRLCK);
+}
+
+/* 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(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
+}
+
+int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       return tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key)
+{
+       return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
+}
+
+
+/* register a loging function */
+void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , 
const char *, ...))
+{
+       tdb->log_fn = fn?fn:null_log_fn;
+}
+
+
+/* reopen a tdb - this can be used after a fork to ensure that we have an 
independent
+   seek pointer from our parent and to re-establish locks */
+int tdb_reopen(TDB_CONTEXT *tdb)
+{
+       struct stat st;
+
+       if (tdb->flags & TDB_INTERNAL)
+               return 0; /* Nothing to do. */
+       if (tdb_munmap(tdb) != 0) {
+               TDB_LOG((tdb, 0, "tdb_reopen: munmap failed (%s)\n", 
strerror(errno)));
+               goto fail;
+       }
+       if (close(tdb->fd) != 0)
+               TDB_LOG((tdb, 0, "tdb_reopen: WARNING closing tdb->fd 
failed!\n"));
+       tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
+       if (tdb->fd == -1) {
+               TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", 
strerror(errno)));
+               goto fail;
+       }
+       if (fstat(tdb->fd, &st) != 0) {
+               TDB_LOG((tdb, 0, "tdb_reopen: fstat failed (%s)\n", 
strerror(errno)));
+               goto fail;
+       }
+       if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
+               TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
+               goto fail;
+       }
+       tdb_mmap(tdb);
+       if ((tdb->flags & TDB_CLEAR_IF_FIRST) && (tdb_brlock(tdb, ACTIVE_LOCK, 
F_RDLCK, F_SETLKW, 0) == -1)) {
+               TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
+               goto fail;
+       }
+
+       return 0;
+
+fail:
+       tdb_close(tdb);
+       return -1;
+}
+
+/* Not general: only works if single writer. */
+TDB_CONTEXT *tdb_copy(TDB_CONTEXT *tdb, const char *outfile)
+{
+       int fd, saved_errno;
+       TDB_CONTEXT *copy;
+
+       fd = open(outfile, O_TRUNC|O_CREAT|O_WRONLY, 0640);
+       if (fd < 0)
+               return NULL;
+       if (tdb->map_ptr) {
+               if (write(fd,tdb->map_ptr,tdb->map_size) != (int)tdb->map_size)
+                       goto fail;
+       } else {
+               char buf[65536];
+               int r;
+
+               lseek(tdb->fd, 0, SEEK_SET);
+               while ((r = read(tdb->fd, buf, sizeof(buf))) > 0) {
+                       if (write(fd, buf, r) != r)
+                               goto fail;
+               }
+               if (r < 0)
+                       goto fail;
+       }
+       copy = tdb_open(outfile, 0, 0, O_RDWR, 0);
+       if (!copy)
+               goto fail;
+       close(fd);
+       return copy;
+
+fail:
+       saved_errno = errno;
+       close(fd);
+       unlink(outfile);
+       errno = saved_errno;
+       return NULL;
+}
+
+/* reopen all tdb's */
+int tdb_reopen_all(void)
+{
+       TDB_CONTEXT *tdb;
+
+       for (tdb=tdbs; tdb; tdb = tdb->next) {
+               /* Ensure no clear-if-first. */
+               tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+               if (tdb_reopen(tdb) != 0)
+                       return -1;
+       }
+
+       return 0;
+}
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/tdb.h
--- /dev/null   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/tdb.h      Fri Sep 23 13:25:01 2005
@@ -0,0 +1,157 @@
+#ifndef __TDB_H__
+#define __TDB_H__
+
+/* 
+   Unix SMB/CIFS implementation.
+
+   trivial database library
+
+   Copyright (C) Andrew Tridgell 1999-2004
+   
+     ** NOTE! The following LGPL license applies to the tdb
+     ** library. This does NOT imply that all of Samba is released
+     ** under the LGPL
+   
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the Free Software
+   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+*/
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+
+/* flags to tdb_store() */
+#define TDB_REPLACE 1
+#define TDB_INSERT 2
+#define TDB_MODIFY 3
+
+/* flags for tdb_open() */
+#define TDB_DEFAULT 0 /* just a readability place holder */
+#define TDB_CLEAR_IF_FIRST 1
+#define TDB_INTERNAL 2 /* don't store on disk */
+#define TDB_NOLOCK   4 /* don't do any locking */
+#define TDB_NOMMAP   8 /* don't use mmap */
+#define TDB_CONVERT 16 /* convert endian (internal use) */
+#define TDB_BIGENDIAN 32 /* header is big-endian (internal use) */
+
+#define TDB_ERRCODE(code, ret) ((tdb->ecode = (code)), ret)
+
+/* error codes */
+enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK, 
+               TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOLOCK, 
TDB_ERR_LOCK_TIMEOUT,
+               TDB_ERR_NOEXIST};
+
+#ifndef u32
+#define u32 unsigned
+#endif
+
+typedef struct TDB_DATA {
+       char *dptr;
+       size_t dsize;
+} TDB_DATA;
+
+typedef u32 tdb_len;
+typedef u32 tdb_off;
+
+/* this is stored at the front of every database */
+struct tdb_header {
+       char magic_food[32]; /* for /etc/magic */
+       u32 version; /* version of the code */
+       u32 hash_size; /* number of hash entries */
+       tdb_off rwlocks;
+       tdb_off reserved[31];
+};
+
+struct tdb_lock_type {
+       u32 count;
+       u32 ltype;
+};
+
+struct tdb_traverse_lock {
+       struct tdb_traverse_lock *next;
+       u32 off;
+       u32 hash;
+};
+
+#ifndef PRINTF_ATTRIBUTE
+#define PRINTF_ATTRIBUTE(a,b)
+#endif
+
+/* this is the context structure that is returned from a db open */
+typedef struct tdb_context {
+       char *name; /* the name of the database */
+       void *map_ptr; /* where it is currently mapped */
+       int fd; /* open file descriptor for the database */
+       tdb_len map_size; /* how much space has been mapped */
+       int read_only; /* opened read-only */
+       struct tdb_lock_type *locked; /* array of chain locks */
+       enum TDB_ERROR ecode; /* error code for last tdb error */
+       struct tdb_header header; /* a cached copy of the header */
+       u32 flags; /* the flags passed to tdb_open */
+       struct tdb_traverse_lock travlocks; /* current traversal locks */
+       struct tdb_context *next; /* all tdbs to avoid multiple opens */
+       dev_t device;   /* uniquely identifies this tdb */
+       ino_t inode;    /* uniquely identifies this tdb */
+       void (*log_fn)(struct tdb_context *tdb, int level, const char *, ...) 
PRINTF_ATTRIBUTE(3,4); /* logging function */
+       u32 (*hash_fn)(TDB_DATA *key);
+       int open_flags; /* flags used in the open - needed by reopen */
+} TDB_CONTEXT;
+
+typedef int (*tdb_traverse_func)(TDB_CONTEXT *, TDB_DATA, TDB_DATA, void *);
+typedef void (*tdb_log_func)(TDB_CONTEXT *, int , const char *, ...);
+typedef u32 (*tdb_hash_func)(TDB_DATA *key);
+
+TDB_CONTEXT *tdb_open(const char *name, int hash_size, int tdb_flags,
+                     int open_flags, mode_t mode);
+TDB_CONTEXT *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+                        int open_flags, mode_t mode,
+                        tdb_log_func log_fn,
+                        tdb_hash_func hash_fn);
+
+int tdb_reopen(TDB_CONTEXT *tdb);
+int tdb_reopen_all(void);
+void tdb_logging_function(TDB_CONTEXT *tdb, tdb_log_func);
+enum TDB_ERROR tdb_error(TDB_CONTEXT *tdb);
+const char *tdb_errorstr(TDB_CONTEXT *tdb);
+TDB_DATA tdb_fetch(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
+int tdb_append(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA new_dbuf);
+int tdb_close(TDB_CONTEXT *tdb);
+TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb);
+TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_traverse(TDB_CONTEXT *tdb, tdb_traverse_func fn, void *);
+int tdb_exists(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_lockall(TDB_CONTEXT *tdb);
+void tdb_unlockall(TDB_CONTEXT *tdb);
+
+/* Low level locking functions: use with care */
+int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_chainlock_read(TDB_CONTEXT *tdb, TDB_DATA key);
+int tdb_chainunlock_read(TDB_CONTEXT *tdb, TDB_DATA key);
+TDB_CONTEXT *tdb_copy(TDB_CONTEXT *tdb, const char *outfile);
+
+/* Debug functions. Not used in production. */
+void tdb_dump_all(TDB_CONTEXT *tdb);
+int tdb_printfreelist(TDB_CONTEXT *tdb);
+
+extern TDB_DATA tdb_null;
+
+#ifdef  __cplusplus
+}
+#endif
+
+#endif /* tdb.h */
diff -r 6aef7d1062bb -r 76af1a1df67c tools/xenstore/xs_tdb_dump.c
--- /dev/null   Fri Sep 23 13:24:58 2005
+++ b/tools/xenstore/xs_tdb_dump.c      Fri Sep 23 13:25:01 2005
@@ -0,0 +1,81 @@
+/* Simple program to dump out all records of TDB */
+#include <stdint.h>
+#include <stdlib.h>
+#include <fcntl.h>
+#include <stdio.h>
+
+#include "xs_lib.h"
+#include "tdb.h"
+#include "talloc.h"
+#include "utils.h"
+
+struct record_hdr {
+       u32 num_perms;
+       u32 datalen;
+       u32 childlen;
+       struct xs_permissions perms[0];
+};
+
+static u32 total_size(struct record_hdr *hdr)
+{
+       return sizeof(*hdr) + hdr->num_perms * sizeof(struct xs_permissions) 
+               + hdr->datalen + hdr->childlen;
+}
+
+static char perm_to_char(enum xs_perm_type perm)
+{
+       return perm == XS_PERM_READ ? 'r' :
+               perm == XS_PERM_WRITE ? 'w' :
+               perm == XS_PERM_NONE ? '-' :
+               perm == (XS_PERM_READ|XS_PERM_WRITE) ? 'b' :
+               '?';
+}
+
+int main(int argc, char *argv[])
+{
+       TDB_DATA key;
+       TDB_CONTEXT *tdb;
+
+       if (argc != 2)
+               barf("Usage: xs_tdb_dump <tdbfile>");
+
+       tdb = tdb_open(talloc_strdup(NULL, argv[1]), 0, 0, O_RDONLY, 0);
+       if (!tdb)
+               barf_perror("Could not open %s", argv[1]);
+
+       key = tdb_firstkey(tdb);
+       while (key.dptr) {
+               TDB_DATA data;
+               struct record_hdr *hdr;
+
+               data = tdb_fetch(tdb, key);
+               hdr = (void *)data.dptr;
+               if (data.dsize < sizeof(*hdr))
+                       fprintf(stderr, "%.*s: BAD truncated\n",
+                               key.dsize, key.dptr);
+               else if (data.dsize != total_size(hdr))
+                       fprintf(stderr, "%.*s: BAD length %i for %i/%i/%i 
(%i)\n",
+                               key.dsize, key.dptr, data.dsize,
+                               hdr->num_perms, hdr->datalen,
+                               hdr->childlen, total_size(hdr));
+               else {
+                       unsigned int i;
+                       char *p;
+
+                       printf("%.*s: ", key.dsize, key.dptr);
+                       for (i = 0; i < hdr->num_perms; i++)
+                               printf("%s%c%i",
+                                      i == 0 ? "" : ",",
+                                      perm_to_char(hdr->perms[i].perms),
+                                      hdr->perms[i].id);
+                       p = (void *)&hdr->perms[hdr->num_perms];
+                       printf(" %.*s\n", hdr->datalen, p);
+                       p += hdr->datalen;
+                       for (i = 0; i < hdr->childlen; i += strlen(p+i)+1)
+                               printf("\t-> %s\n", p+i);
+               }
+               key = tdb_nextkey(tdb, key);
+       }
+       return 0;
+}
+

_______________________________________________
Xen-changelog mailing list
Xen-changelog@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-changelog