o Build dmeventd against multilog

to retrieve which devices got registered with the daemon;
  needs locking changes as well

VERSION

@@ -1 +1 @@
-2.01.09-cvs (2005-03-22)
+2.01.10-cvs (2005-04-04)

WHATS_NEW

@@ -1,5 +1,23 @@
-Version 2.01.09 - 
-=================================
+Version 2.01.10 - 
+================================
+  Tidy lv_segment interface.
+  Initial pv_segment support.
+  vgchange --physicalextentsize
+  Internal snapshot restructuring.
+  Remove unused internal non-persistent snapshot option.
+  Allow offline extension of snapshot volumes.
+  Move from 2-step to 3-step on-disk metadata commit.
+  Scan ramdisks too and allow non-O_DIRECT fallback.
+  Annotate, tidy and extend list.h.
+  Alignment tidying.
+  Make clvmd work around some "bugs" in gulm's node state notifications.
+  Tidy clvmd's SIGHUP handler
+
+Version 2.01.09 - 4th April 2005
+================================
+  Add --ignorelockingfailure to vgmknodes.
+  clvmd: Don't allow user operations to start until the lvm thread is fully up.
+  clvmd-gulm: set KEEPALIVE on sockets.
 
 Version 2.01.08 - 22nd March 2005
 =================================

WHATS_NEW_DM

@@ -1,3 +1,6 @@
+Version 1.01.02 - 
+=============================
+
 Version 1.01.01 - 29 Mar 2005
 =============================
   Update dmsetup man page.

daemons/clvmd/clvmd-comms.h

@@ -40,6 +40,7 @@ struct cluster_ops {
 
 	void (*get_our_csid) (char *csid);
 	void (*add_up_node) (char *csid);
+	void (*reread_config) (void);
 	void (*cluster_closedown) (void);
 
 	int (*sync_lock) (const char *resource, int mode, int flags, int *lockid);

daemons/clvmd/clvmd-gulm.c

@@ -60,8 +60,9 @@ static struct hash_table *node_hash;
 /* hash list of outstanding lock requests */
 static struct hash_table *lock_hash;
 
-/* Copy of the current core state */
-static uint8_t current_corestate;
+/* Copy of the current quorate state */
+static uint8_t gulm_quorate = 0;
+static enum {INIT_NOTDONE, INIT_DONE, INIT_WAITQUORATE} init_state = INIT_NOTDONE;
 
 /* Number of active nodes */
 static int num_nodes;
@@ -135,12 +136,11 @@ static void badsig_handler(int sig)
     exit(0);
 }
 
-static void sighup_handler(int sig)
+static void _reread_config(void)
 {
-    DEBUGLOG("got SIGHUP\n");
-
-    /* Re-read CCS node list */
-    get_all_cluster_nodes();
+        /* Re-read CCS node list */
+	DEBUGLOG("Re-reading CCS config\n");
+	get_all_cluster_nodes();
 }
 
 static int _init_cluster(void)
@@ -243,9 +243,6 @@ static int _init_cluster(void)
     signal(SIGINT, badsig_handler);
     signal(SIGTERM, badsig_handler);
 
-    /* Re-read the node list on SIGHUP */
-    signal(SIGHUP, sighup_handler);
-
     return 0;
 }
 
@@ -312,12 +309,16 @@ static int core_login_reply(void *misc, uint64_t gen, uint32_t error, uint32_t r
    if (error)
        exit(error);
 
-   current_corestate = corestate;
+   /* Get the current core state (for quorum) */
+   lg_core_corestate(gulm_if);
+
    return 0;
 }
 
 static void set_node_state(struct node_info *ninfo, char *csid, uint8_t nodestate)
 {
+    int oldstate = ninfo->state;
+
     if (nodestate == lg_core_Logged_in)
     {
 	/* Don't clobber NODE_CLVMD state */
@@ -339,11 +340,17 @@ static void set_node_state(struct node_info *ninfo, char *csid, uint8_t nodestat
 	    if (ninfo->state != NODE_DOWN)
 		num_nodes--;
 	    ninfo->state = NODE_DOWN;
-	    tcp_remove_client(csid);
 	}
     }
-    DEBUGLOG("set_node_state, '%s' state = %d, num_nodes=%d\n",
-	     ninfo->name, ninfo->state, num_nodes);
+    /* Gulm doesn't always send node DOWN events, so even if this a a node UP we must
+     * assume (ahem) that it prevously went down at some time. So we close
+     * the sockets here to make sure that we don't have any dead connections
+     * to that node.
+     */
+    tcp_remove_client(csid);
+
+    DEBUGLOG("set_node_state, '%s' state = %d (oldstate=%d), num_nodes=%d\n",
+	     ninfo->name, ninfo->state, oldstate, num_nodes);
 }
 
 static struct node_info *add_or_set_node(char *name, struct in6_addr *ip, uint8_t state)
@@ -400,7 +407,16 @@ static int core_nodelist(void *misc, lglcb_t type, char *name, struct in6_addr *
 		char ourcsid[GULM_MAX_CSID_LEN];
 
 		DEBUGLOG("Got Nodelist, stop\n");
-		clvmd_cluster_init_completed();
+		if (gulm_quorate)
+		{
+			clvmd_cluster_init_completed();
+			init_state = INIT_DONE;
+		}
+		else
+		{
+			if (init_state == INIT_NOTDONE)
+				init_state = INIT_WAITQUORATE;
+		}
 
 		/* Mark ourself as up */
 		_get_our_csid(ourcsid);
@@ -418,10 +434,15 @@ static int core_nodelist(void *misc, lglcb_t type, char *name, struct in6_addr *
 
 static int core_statechange(void *misc, uint8_t corestate, uint8_t quorate, struct in6_addr *masterip, char *mastername)
 {
-    DEBUGLOG("CORE Got statechange  corestate:%#x mastername:%s\n",
-	     corestate, mastername);
+    DEBUGLOG("CORE Got statechange. quorate:%d, corestate:%x mastername:%s\n",
+	     quorate, corestate, mastername);
 
-    current_corestate = corestate;
+    gulm_quorate = quorate;
+    if (quorate && init_state == INIT_WAITQUORATE)
+    {
+	    clvmd_cluster_init_completed();
+	    init_state = INIT_DONE;
+    }
     return 0;
 }
 
@@ -474,7 +495,7 @@ static int lock_login_reply(void *misc, uint32_t error, uint8_t which)
 	lock_start_flag = 0;
 	pthread_mutex_unlock(&lock_start_mutex);
     }
-    
+
     return 0;
 }
 
@@ -615,7 +636,11 @@ void gulm_add_up_node(char *csid)
     }
 
     DEBUGLOG("gulm_add_up_node %s\n", ninfo->name);
+
+    if (ninfo->state == NODE_DOWN)
+	    num_nodes++;
     ninfo->state = NODE_CLVMD;
+
     return;
 
 }
@@ -853,12 +878,7 @@ static int _sync_unlock(const char *resource, int lockid)
 
 static int _is_quorate()
 {
-    if (current_corestate == lg_core_Slave ||
-	current_corestate == lg_core_Master ||
-	current_corestate == lg_core_Client)
-	return 1;
-    else
-	return 0;
+	return gulm_quorate;
 }
 
 /* Get all the cluster node names & IPs from CCS and
@@ -962,6 +982,7 @@ static struct cluster_ops _cluster_gulm_ops = {
 	.is_quorate               = _is_quorate,
 	.get_our_csid             = _get_our_csid,
 	.add_up_node              = gulm_add_up_node,
+	.reread_config            = _reread_config,
 	.cluster_closedown        = _cluster_closedown,
 	.sync_lock                = _sync_lock,
 	.sync_unlock              = _sync_unlock,

daemons/clvmd/clvmd.c

@@ -87,8 +87,10 @@ struct lvm_thread_cmd {
 static pthread_t lvm_thread;
 static pthread_mutex_t lvm_thread_mutex;
 static pthread_cond_t lvm_thread_cond;
+static pthread_mutex_t lvm_start_mutex;
 static struct list lvm_cmd_head;
 static volatile sig_atomic_t quit = 0;
+static volatile sig_atomic_t reread_config = 0;
 static int child_pipe[2];
 
 /* Reasons the daemon failed initialisation */
@@ -100,6 +102,7 @@ static int child_pipe[2];
 
 /* Prototypes for code further down */
 static void sigusr2_handler(int sig);
+static void sighup_handler(int sig);
 static void sigterm_handler(int sig);
 static void send_local_reply(struct local_client *client, int status,
 			     int clientid);
@@ -221,8 +224,10 @@ int main(int argc, char *argv[])
 
 	/* Set up signal handlers, USR1 is for cluster change notifications (in cman)
 	   USR2 causes child threads to exit.
+	   HUP causes gulm version to re-read nodes list from CCS.
 	   PIPE should be ignored */
 	signal(SIGUSR2, sigusr2_handler);
+	signal(SIGHUP,  sighup_handler);
 	signal(SIGPIPE, SIG_IGN);
 
 	/* Block SIGUSR2 in the main process */
@@ -234,6 +239,7 @@ int main(int argc, char *argv[])
 	list_init(&lvm_cmd_head);
 	pthread_mutex_init(&lvm_thread_mutex, NULL);
 	pthread_cond_init(&lvm_thread_cond, NULL);
+	pthread_mutex_init(&lvm_start_mutex, NULL);
 	init_lvhash();
 
 	/* Start the cluster interface */
@@ -278,6 +284,7 @@ int main(int argc, char *argv[])
 	        child_init_signal(DFAIL_MALLOC);
 
 	newfd->fd = local_sock;
+	newfd->removeme = 0;
 	newfd->type = LOCAL_RENDEZVOUS;
 	newfd->callback = local_rendezvous_callback;
 	newfd->next = local_client_head.next;
@@ -344,6 +351,7 @@ static int local_rendezvous_callback(struct local_client *thisfd, char *buf,
 		newfd->fd = client_fd;
 		newfd->type = LOCAL_SOCK;
 		newfd->xid = 0;
+		newfd->removeme = 0;
 		newfd->callback = local_sock_callback;
 		newfd->bits.localsock.replies = NULL;
 		newfd->bits.localsock.expected_replies = 0;
@@ -510,13 +518,38 @@ static void main_loop(int local_sock, int cmd_timeout)
 				FD_SET(thisfd->fd, &in);
 		}
 
-		if ((select_status = select(FD_SETSIZE, &in, NULL, NULL, &tv)) > 0) {
+		select_status = select(FD_SETSIZE, &in, NULL, NULL, &tv);
+
+		if (reread_config) {
+			int saved_errno = errno;
+
+			reread_config = 0;
+			if (clops->reread_config)
+				clops->reread_config();
+			errno = saved_errno;
+		}
+
+		if (select_status > 0) {
 			struct local_client *lastfd = NULL;
 			char csid[MAX_CSID_LEN];
 			char buf[max_cluster_message];
 
 			for (thisfd = &local_client_head; thisfd != NULL;
 			     thisfd = thisfd->next) {
+
+				if (thisfd->removeme) {
+					struct local_client *free_fd;
+					lastfd->next = thisfd->next;
+					free_fd = thisfd;
+					thisfd = lastfd;
+
+					DEBUGLOG("removeme set for fd %d\n", free_fd->fd);
+
+					/* Queue cleanup, this also frees the client struct */
+					add_to_lvmqueue(free_fd, NULL, 0, NULL);
+					break;
+				}
+
 				if (FD_ISSET(thisfd->fd, &in)) {
 					struct local_client *newfd;
 					int ret;
@@ -903,6 +936,7 @@ static int read_from_local_sock(struct local_client *thisfd)
 		DEBUGLOG("creating pipe, [%d, %d]\n", comms_pipe[0],
 			 comms_pipe[1]);
 		newfd->fd = comms_pipe[0];
+		newfd->removeme = 0;
 		newfd->type = THREAD_PIPE;
 		newfd->callback = local_pipe_callback;
 		newfd->next = thisfd->next;
@@ -1059,8 +1093,8 @@ void process_remote_command(struct clvm_header *msg, int msglen, int fd,
 	/* Get the node name as we /may/ need it later */
 	clops->name_from_csid(csid, nodename);
 
-	DEBUGLOG("process_remote_command %d for clientid 0x%x on node %s\n",
-		 msg->cmd, msg->clientid, nodename);
+	DEBUGLOG("process_remote_command %d for clientid 0x%x XID %d on node %s\n",
+		 msg->cmd, msg->clientid, msg->xid, nodename);
 
 	/* Is the data to be found in the system LV ? */
 	if (msg->flags & CLVMD_FLAG_SYSTEMLV) {
@@ -1303,6 +1337,11 @@ static void *pre_and_post_thread(void *arg)
 
 	DEBUGLOG("in sub thread: client = %p\n", client);
 
+	/* Don't start until the LVM thread is ready */
+	pthread_mutex_lock(&lvm_start_mutex);
+	pthread_mutex_unlock(&lvm_start_mutex);
+	DEBUGLOG("Sub thread ready for work.\n");
+
 	/* Ignore SIGUSR1 (handled by master process) but enable
 	   SIGUSR2 (kills subthreads) */
 	sigemptyset(&ss);
@@ -1568,9 +1607,10 @@ static int send_message(void *buf, int msglen, char *csid, int fd,
 
 static int process_work_item(struct lvm_thread_cmd *cmd)
 {
-
 	/* If msg is NULL then this is a cleanup request */
 	if (cmd->msg == NULL) {
+		DEBUGLOG("process_work_item: free fd %d\n", cmd->client->fd);
+		close(cmd->client->fd);
 		cmd_client_cleanup(cmd->client);
 		free(cmd->client);
 		return 0;
@@ -1597,6 +1637,9 @@ static void *lvm_thread_fn(void *arg)
 	sigset_t ss;
 	int using_gulm = (int)arg;
 
+	/* Don't let anyone else to do work until we are started */
+	pthread_mutex_lock(&lvm_start_mutex);
+
 	DEBUGLOG("LVM thread function started\n");
 
 	/* Ignore SIGUSR1 & 2 */
@@ -1608,6 +1651,9 @@ static void *lvm_thread_fn(void *arg)
 	/* Initialise the interface to liblvm */
 	init_lvm(using_gulm);
 
+	/* Allow others to get moving */
+	pthread_mutex_unlock(&lvm_start_mutex);
+
 	/* Now wait for some actual work */
 	for (;;) {
 		DEBUGLOG("LVM thread waiting for work\n");
@@ -1625,7 +1671,8 @@ static void *lvm_thread_fn(void *arg)
 			pthread_mutex_unlock(&lvm_thread_mutex);
 
 			process_work_item(cmd);
-			free(cmd->msg);
+			if (cmd->msg)
+				free(cmd->msg);
 			free(cmd);
 
 			pthread_mutex_lock(&lvm_thread_mutex);
@@ -1788,6 +1835,12 @@ static void sigterm_handler(int sig)
 	return;
 }
 
+static void sighup_handler(int sig)
+{
+        DEBUGLOG("got SIGHUP\n");
+	reread_config = 1;
+}
+
 int sync_lock(const char *resource, int mode, int flags, int *lockid)
 {
 	return clops->sync_lock(resource, mode, flags, lockid);

daemons/clvmd/clvmd.h

@@ -86,6 +86,7 @@ struct local_client {
 	struct local_client *next;
 	unsigned short xid;
 	fd_callback_t callback;
+	uint8_t removeme;
 
 	union {
 		struct localsock_bits localsock;
@@ -95,7 +96,7 @@ struct local_client {
 };
 
 #ifdef DEBUG
-#define DEBUGLOG(fmt, args...) fprintf(stderr, "CLVMD[%x]: %ld ", (int)pthread_self(), time(NULL) ); fprintf(stderr, fmt, ## args)
+#define DEBUGLOG(fmt, args...) {time_t P; time(&P); fprintf(stderr, "CLVMD[%x]: %.15s ", (int)pthread_self(), ctime(&P)+4 ); fprintf(stderr, fmt, ## args);}
 #else
 #define DEBUGLOG(fmt, args...)
 #endif

daemons/clvmd/tcp-comms.c

@@ -69,6 +69,7 @@ int init_comms(unsigned short port)
     {
 	int one = 1;
 	setsockopt(listen_fd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof(int));
+	setsockopt(listen_fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(int));
     }
 
     memset(&addr, 0, sizeof(addr)); // Bind to INADDR_ANY
@@ -104,6 +105,7 @@ void tcp_remove_client(char *csid)
     if (client)
     {
 	hash_remove_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+	client->removeme = 1;
     }
 
     /* Look for a mangled one too */
@@ -113,6 +115,7 @@ void tcp_remove_client(char *csid)
     if (client)
     {
 	hash_remove_binary(sock_hash, csid, GULM_MAX_CSID_LEN);
+	client->removeme = 1;
     }
 
     /* Put it back as we found it */
@@ -286,7 +289,7 @@ static int read_from_tcpsock(struct local_client *client, char *buf, int len, ch
     else {
 	    gulm_add_up_node(csid);
 	    /* Send it back to clvmd */
-	    process_message(client, buf, len, csid);
+	    process_message(client, buf, status, csid);
     }
     return status;
 }
@@ -296,6 +299,7 @@ int gulm_connect_csid(char *csid, struct local_client **newclient)
     int fd;
     struct sockaddr_in6 addr;
     int status;
+    int one = 1;
 
     DEBUGLOG("Connecting socket\n");
     fd = socket(PF_INET6, SOCK_STREAM, 0);
@@ -327,6 +331,7 @@ int gulm_connect_csid(char *csid, struct local_client **newclient)
 
     /* Set Close-on-exec */
     fcntl(fd, F_SETFD, 1);
+    setsockopt(fd, SOL_SOCKET, SO_KEEPALIVE, &one, sizeof(int));
 
     status = alloc_client(fd, csid, newclient);
     if (status)

daemons/dmeventd/.exported_symbols

@@ -0,0 +1,3 @@
+process_event
+register_device
+unregister_device

daemons/dmeventd/Makefile.in

@@ -0,0 +1,51 @@
+#
+# Copyright (C) 2001-2004 Sistina Software, Inc. All rights reserved.
+# Copyright (C) 2004 Red Hat, Inc. All rights reserved.
+#
+# This file is part of the device-mapper userspace tools.
+#
+# This copyrighted material is made available to anyone wishing to use,
+# modify, copy, or redistribute it subject to the terms and conditions
+# of the GNU General Public License v.2.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+
+srcdir = @srcdir@
+top_srcdir = @top_srcdir@
+VPATH = @srcdir@
+
+TARGETS = dmevent dmeventd
+INSTALL_TYPE = install_dynamic
+
+SOURCES = noop.c
+CLEAN_TARGETS = dmevent
+
+ifeq ("@LIB_SUFFIX@","dylib")
+  LIB_SHARED = libdmeventnoop.dylib
+else
+  LIB_SHARED = libdmeventnoop.so
+endif
+ 
+LDFLAGS += -ldl -ldevmapper -lpthread
+
+include ../make.tmpl
+
+libdmeventnoop.so: noop.o
+
+dmevent: dmevent.o $(interfacedir)/libdevmapper.$(LIB_SUFFIX) $(top_srcdir)/lib/event/libdmevent.$(LIB_SUFFIX)
+	$(CC) -o $@ dmevent.o $(LDFLAGS) \
+	      -L$(interfacedir) -L$(DESTDIR)/lib -L$(top_srcdir)/lib/event -L$(top_srcdir)/multilog -lmultilog -ldmevent $(LIBS)
+
+dmeventd: dmeventd.o $(interfacedir)/libdevmapper.$(LIB_SUFFIX) $(top_srcdir)/lib/event/libdmevent.$(LIB_SUFFIX)
+	$(CC) -o $@ dmeventd.o $(LDFLAGS) \
+	      -L$(interfacedir) -L$(DESTDIR)/lib -L$(top_srcdir)/lib/event -L$(top_srcdir)/multilog -ldmevent -lmultilog $(LIBS)
+
+install: $(INSTALL_TYPE)
+
+.PHONY: install_dynamic
+
+install_dynamic: dmeventd
+	$(INSTALL) -D $(OWNER) $(GROUP) -m 555 $(STRIP) dmeventd $(sbindir)/dmeventd
+

daemons/dmeventd/dmevent.c

@@ -0,0 +1,137 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of the device-mapper userspace tools.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU Lesser General Public License v.2.1.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "libdevmapper.h"
+#include "libdm-event.h"
+
+#include <errno.h>
+#include <fcntl.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/file.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+#include <unistd.h>
+
+static enum event_type events = ALL_ERRORS; /* All until we can distinguish. */
+static char *default_dso_name = "noop";  /* default DSO is noop */
+static int default_reg = 1;		 /* default action is register */
+
+/* Display help. */
+static void print_usage(char *name)
+{
+	char *cmd = strrchr(name, '/');
+
+	cmd = cmd ? cmd + 1 : name;
+	printf("Usage::\n"
+	       "%s [options] <device>\n"
+	       "\n"
+	       "Options:\n"
+	       "  -d <dso>           Specify the DSO to use.\n"
+	       "  -h                 Print this usage.\n"
+	       "  -l                 List registered devices.\n"
+	       "  -r                 Register for event (default).\n"
+	       "  -u                 Unregister for event.\n"
+	       "\n", cmd);
+}
+
+/* Parse command line arguments. */
+static int parse_argv(int argc, char **argv,
+		      char **dso_name, char **device, int *reg, int *list)
+{
+	int c;
+	const char *options = "d:hlru";
+
+	while ((c = getopt(argc, argv, options)) != -1) {
+		switch (c) {
+		case 'd':
+			*dso_name = optarg;
+			break;
+		case 'h':
+			print_usage(argv[0]);
+			exit(EXIT_SUCCESS);
+		case 'l':
+			*list = 1;
+			break;
+		case 'r':
+			*reg = 1;
+			break;
+		case 'u':
+			*reg = 0;
+			break;
+		default:
+			fprintf(stderr, "Unknown option '%c'.\n"
+				"Try '-h' for help.\n", c);
+
+			return 0;
+		}
+	}
+
+	if (!*list) {
+		if (optind >= argc) {
+			fprintf(stderr, "You need to specify a device.\n");
+			return 0;
+		}
+
+		*device = argv[optind];
+	}
+
+	return 1;
+}
+
+int main(int argc, char **argv)
+{
+	int list = 0, ret, reg = default_reg;
+	char *device, *dso_name = default_dso_name;
+
+	if (!parse_argv(argc, argv, &dso_name, &device, &reg, &list))
+		exit(EXIT_FAILURE);
+	
+	if (list) {
+		do {
+			ret = dm_get_next_registered_device(&dso_name, &device, &events);
+			if (!ret) {
+				printf("%s %s 0x%x\n", dso_name, device, events);
+				free(dso_name);
+				free(device);
+			}
+		} while (ret);
+
+		exit(EXIT_SUCCESS);
+	}
+	if ((ret = reg ? dm_register_for_event(dso_name, device, events) :
+			 dm_unregister_for_event(dso_name, device, events))) {
+		fprintf(stderr, "Failed to %sregister %s: %s\n",
+			reg ? "": "un", device, strerror(-ret));
+
+		exit(EXIT_FAILURE);
+	}
+
+	printf("%s %sregistered successfully.\n", device, reg ? "" : "un");
+
+	exit(EXIT_SUCCESS);
+}
+
+/*
+ * Overrides for Emacs so that we follow Linus's tabbing style.
+ * Emacs will notice this stuff at the end of the file and automatically
+ * adjust the settings for this buffer only.  This must remain at the end
+ * of the file.
+ * ---------------------------------------------------------------------------
+ * Local variables:
+ * c-file-style: "linux"
+ * End:
+ */

daemons/dmeventd/mktestdevices

@@ -0,0 +1,12 @@
+#!/bin/sh
+#
+# Create test devices for dmeventd
+#
+
+trap "rm -f /tmp/tmp.$$" 0 1 2 3 15
+
+echo "0 1024 zero" > /tmp/tmp.$$
+dmsetup create test /tmp/tmp.$$
+dmsetup create test1 /tmp/tmp.$$
+
+kill -15 $$

daemons/dmeventd/noop.c

@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of the device-mapper userspace tools.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU Lesser General Public License v.2.1.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "libdm-event.h"
+
+#include <stdio.h>
+
+void process_event(char *device, enum event_type event)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s, Event %d\n",
+		__FILE__, __func__, __LINE__, device, event);
+}
+
+int register_device(char *device)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s\n",
+		__FILE__, __func__, __LINE__, device);
+
+	return 1;
+}
+
+int unregister_device(char *device)
+{
+	fprintf(stderr, "[%s] %s(%d) - Device: %s\n",
+		__FILE__, __func__, __LINE__, device);
+
+	return 1;
+}

include/.symlinks

@@ -31,6 +31,7 @@
 ../lib/log/log.h
 ../lib/metadata/lv_alloc.h
 ../lib/metadata/metadata.h
+../lib/metadata/pv_alloc.h
 ../lib/metadata/segtype.h
 ../lib/mm/dbg_malloc.h
 ../lib/mm/memlock.h

lib/Makefile.in

@@ -70,6 +70,7 @@ SOURCES =\
 	metadata/merge.c \
 	metadata/metadata.c \
 	metadata/mirror.c \
+	metadata/pv_manip.c \
 	metadata/pv_map.c \
 	metadata/segtype.c \
 	metadata/snapshot_manip.c \

lib/activate/dev_manager.c

@@ -727,6 +727,25 @@ static int _emit_target_line(struct dev_manager *dm, struct dm_task *dmt,
 	return 1;
 }
 
+int compose_log_line(struct dev_manager *dm, struct lv_segment *seg,
+		     char *params, size_t paramsize, int *pos, int areas,
+		     uint32_t region_size)
+{
+	int tw;
+
+	tw = lvm_snprintf(params, paramsize, "core 1 %u %u ",
+			  region_size, areas);
+
+	if (tw < 0) {
+		stack;
+		return -1;
+	}
+
+	*pos += tw;
+
+	return 1;
+}
+
 int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,
 		       char *params, size_t paramsize, int *pos, int start_area,
 		       int areas)
@@ -875,22 +894,23 @@ static int _populate_snapshot(struct dev_manager *dm,
 {
 	char *origin, *cow;
 	char params[PATH_MAX * 2 + 32];
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct dev_layer *dlo, *dlc;
 	char devbufo[10], devbufc[10];
 	uint64_t size;
 
-	if (!(s = find_cow(dl->lv))) {
+	if (!(snap_seg = find_cow(dl->lv))) {
 		log_error("Couldn't find snapshot for '%s'.", dl->lv->name);
 		return 0;
 	}
 
-	if (!(origin = _build_dlid(dm->mem, s->origin->lvid.s, "real"))) {
+	if (!(origin = _build_dlid(dm->mem, snap_seg->origin->lvid.s,
+				   "real"))) {
 		stack;
 		return 0;
 	}
 
-	if (!(cow = _build_dlid(dm->mem, s->cow->lvid.s, "cow"))) {
+	if (!(cow = _build_dlid(dm->mem, snap_seg->cow->lvid.s, "cow"))) {
 		stack;
 		return 0;
 	}
@@ -909,24 +929,24 @@ static int _populate_snapshot(struct dev_manager *dm,
 	if (!dm_format_dev(devbufo, sizeof(devbufo), dlo->info.major,
 			   dlo->info.minor)) {
 		log_error("Couldn't create origin device parameters for '%s'.",
-			  s->origin->name);
+			  snap_seg->origin->name);
 		return 0;
 	}
 
 	if (!dm_format_dev(devbufc, sizeof(devbufc), dlc->info.major,
 			   dlc->info.minor)) {
 		log_error("Couldn't create cow device parameters for '%s'.",
-			  s->cow->name);
+			  snap_seg->cow->name);
 		return 0;
 	}
 
 	if (lvm_snprintf(params, sizeof(params), "%s %s P %d",
-			 devbufo, devbufc, s->chunk_size) == -1) {
+			 devbufo, devbufc, snap_seg->chunk_size) == -1) {
 		stack;
 		return 0;
 	}
 
-	size = (uint64_t) s->le_count * s->origin->vg->extent_size;
+	size = (uint64_t) snap_seg->len * snap_seg->origin->vg->extent_size;
 
 	log_debug("Adding target: 0 %" PRIu64 " snapshot %s", size, params);
 	if (!dm_task_add_target(dmt, UINT64_C(0), size, "snapshot", params)) {
@@ -1274,7 +1294,7 @@ static int _expand_origin_real(struct dev_manager *dm,
 static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 {
 	struct logical_volume *active;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh;
 
 	/*
@@ -1283,7 +1303,7 @@ static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 	 */
 	list_iterate(sh, &dm->active_list) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && (s->origin == lv))
+		if ((snap_seg = find_cow(active)) && (snap_seg->origin == lv))
 			return _expand_origin_real(dm, lv);
 	}
 
@@ -1294,7 +1314,7 @@ static int _expand_origin(struct dev_manager *dm, struct logical_volume *lv)
 }
 
 static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
-			    struct snapshot *s)
+			    struct lv_segment *snap_seg)
 {
 	/*
 	 * snapshot(org, cow)
@@ -1331,13 +1351,15 @@ static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
 
 	/* add the dependency on the real origin device */
 	if (!str_list_add(dm->mem, &dl->pre_create,
-			  _build_dlid(dm->mem, s->origin->lvid.s, "real"))) {
+			  _build_dlid(dm->mem, snap_seg->origin->lvid.s,
+				      "real"))) {
 		stack;
 		return 0;
 	}
 
 	/* add the dependency on the visible origin device */
-	if (!str_list_add(dm->mem, &dl->pre_suspend, s->origin->lvid.s)) {
+	if (!str_list_add(dm->mem, &dl->pre_suspend,
+			  snap_seg->origin->lvid.s)) {
 		stack;
 		return 0;
 	}
@@ -1350,13 +1372,13 @@ static int _expand_snapshot(struct dev_manager *dm, struct logical_volume *lv,
  */
 static int _expand_lv(struct dev_manager *dm, struct logical_volume *lv)
 {
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 
 	/*
 	 * FIXME: this doesn't cope with recursive snapshots yet.
 	 */
-	if ((s = find_cow(lv)))
-		return _expand_snapshot(dm, lv, s);
+	if ((snap_seg = find_cow(lv)))
+		return _expand_snapshot(dm, lv, snap_seg);
 
 	else if (lv_is_origin(lv))
 		return _expand_origin(dm, lv);
@@ -1445,6 +1467,8 @@ static int _mark_lvs(struct dev_manager *dm, struct list *lvs, int flag)
 
 	list_iterate(lvh, lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
 
 		if (!(dl = _lookup(dm, lv->lvid.s, NULL))) {
 			stack;
@@ -1606,14 +1630,16 @@ static int _create_rec(struct dev_manager *dm, struct dev_layer *dl)
 static int _build_all_layers(struct dev_manager *dm, struct volume_group *vg)
 {
 	struct list *lvh;
-	struct logical_volume *lvt;
+	struct logical_volume *lv;
 
 	/*
 	 * Build layers for complete vg.
 	 */
 	list_iterate(lvh, &vg->lvs) {
-		lvt = list_item(lvh, struct lv_list)->lv;
-		if (!_expand_lv(dm, lvt)) {
+		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
+		if (!_expand_lv(dm, lv)) {
 			stack;
 			return 0;
 		}
@@ -1915,12 +1941,14 @@ static int _add_lvs(struct pool *mem,
 		    struct list *head, struct logical_volume *origin)
 {
 	struct logical_volume *lv;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *lvh;
 
 	list_iterate(lvh, &origin->vg->lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
-		if ((s = find_cow(lv)) && s->origin == origin)
+		if (lv->status & SNAPSHOT)
+			continue;
+		if ((snap_seg = find_cow(lv)) && snap_seg->origin == origin)
 			if (!_add_lv(mem, head, lv))
 				return 0;
 	}
@@ -1945,7 +1973,7 @@ static void _remove_lv(struct list *head, struct logical_volume *lv)
 static int _remove_lvs(struct dev_manager *dm, struct logical_volume *lv)
 {
 	struct logical_volume *active, *old_origin;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh, *active_head;
 
 	active_head = &dm->active_list;
@@ -1953,22 +1981,23 @@ static int _remove_lvs(struct dev_manager *dm, struct logical_volume *lv)
 	/* Remove any snapshots with given origin */
 	list_iterate(sh, active_head) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && s->origin == lv) {
+		if ((snap_seg = find_cow(active)) && snap_seg->origin == lv) {
 			_remove_lv(active_head, active);
 		}
 	}
 
 	_remove_lv(active_head, lv);
 
-	if (!(s = find_cow(lv)))
+	if (!(snap_seg = find_cow(lv)))
 		return 1;
 
-	old_origin = s->origin;
+	old_origin = snap_seg->origin;
 
 	/* Was this the last active snapshot with this origin? */
 	list_iterate(sh, active_head) {
 		active = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(active)) && s->origin == old_origin) {
+		if ((snap_seg = find_cow(active)) &&
+		    snap_seg->origin == old_origin) {
 			return 1;
 		}
 	}
@@ -1980,7 +2009,7 @@ static int _remove_suspended_lvs(struct dev_manager *dm,
 				 struct logical_volume *lv)
 {
 	struct logical_volume *suspended;
-	struct snapshot *s;
+	struct lv_segment *snap_seg;
 	struct list *sh, *suspend_head;
 
 	suspend_head = &dm->suspend_list;
@@ -1988,7 +2017,8 @@ static int _remove_suspended_lvs(struct dev_manager *dm,
 	/* Remove from list any snapshots with given origin */
 	list_iterate(sh, suspend_head) {
 		suspended = list_item(sh, struct lv_list)->lv;
-		if ((s = find_cow(suspended)) && s->origin == lv) {
+		if ((snap_seg = find_cow(suspended)) &&
+		    snap_seg->origin == lv) {
 			_remove_lv(suspend_head, suspended);
 		}
 	}
@@ -2074,6 +2104,8 @@ static int _fill_in_active_list(struct dev_manager *dm, struct volume_group *vg)
 
 	list_iterate(lvh, &vg->lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
+		if (lv->status & SNAPSHOT)
+			continue;
 
 		if (!(dlid = _build_dlid(dm->mem, lv->lvid.s, NULL))) {
 			stack;

lib/activate/targets.h

@@ -19,7 +19,12 @@
 struct dev_manager;
 struct lv_segment;
 
-int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,                                 char *params, size_t paramsize, int *pos,
-		                        int start_area, int areas);
+int compose_areas_line(struct dev_manager *dm, struct lv_segment *seg,
+		       char *params, size_t paramsize, int *pos,
+		       int start_area, int areas);
+
+int compose_log_line(struct dev_manager *dm, struct lv_segment *seg,
+		     char *params, size_t paramsize, int *pos, int areas,
+		     uint32_t region_size);
 
 #endif

lib/cache/lvmcache.h

@@ -36,9 +36,10 @@ struct volume_group;
 struct lvmcache_vginfo {
 	struct list list;	/* Join these vginfos together */
 	struct list infos;	/* List head for lvmcache_infos */
+	const struct format_type *fmt;
 	char *vgname;		/* "" == orphan */
 	char vgid[ID_LEN + 1];
-	const struct format_type *fmt;
+	char _padding[7];
 };
 
 struct lvmcache_info {

lib/commands/toolcontext.h

@@ -32,17 +32,15 @@ struct config_info {
 	int syslog;
 	int activation;
 	int suffix;
-	uint64_t unit_factor;
-	char unit_type;
-	const char *msg_prefix;
-	int cmd_name;		/* Show command name? */
-
 	int archive;		/* should we archive ? */
 	int backup;		/* should we backup ? */
-
+	const char *msg_prefix;
 	struct format_type *fmt;
-
+	uint64_t unit_factor;
+	int cmd_name;		/* Show command name? */
 	mode_t umask;
+	char unit_type;
+	char _padding[1];
 };
 
 struct config_tree;

lib/config/defaults.h

@@ -95,15 +95,18 @@
 #define DEFAULT_VGS_COLS "vg_name,pv_count,lv_count,snap_count,vg_attr,vg_size,vg_free"
 #define DEFAULT_PVS_COLS "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free"
 #define DEFAULT_SEGS_COLS "lv_name,vg_name,lv_attr,stripes,segtype,seg_size"
+#define DEFAULT_PVSEGS_COLS "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,pvseg_start,pvseg_size"
 
 #define DEFAULT_LVS_COLS_VERB "lv_name,vg_name,seg_count,lv_attr,lv_size,lv_major,lv_minor,lv_kernel_major,lv_kernel_minor,origin,snap_percent,move_pv,copy_percent,lv_uuid"
 #define DEFAULT_VGS_COLS_VERB "vg_name,vg_attr,vg_extent_size,pv_count,lv_count,snap_count,vg_size,vg_free,vg_uuid"
 #define DEFAULT_PVS_COLS_VERB "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,dev_size,pv_uuid"
 #define DEFAULT_SEGS_COLS_VERB "lv_name,vg_name,lv_attr,seg_start,seg_size,stripes,segtype,stripesize,chunksize"
+#define DEFAULT_PVSEGS_COLS_VERB "pv_name,vg_name,pv_fmt,pv_attr,pv_size,pv_free,pvseg_start,pvseg_size"
 
 #define DEFAULT_LVS_SORT "vg_name,lv_name"
 #define DEFAULT_VGS_SORT "vg_name"
 #define DEFAULT_PVS_SORT "pv_name"
 #define DEFAULT_SEGS_SORT "vg_name,lv_name,seg_start"
+#define DEFAULT_PVSEGS_SORT "pv_name,pvseg_start"
 
 #endif				/* _LVM_DEFAULTS_H */

lib/datastruct/list.h

@@ -18,17 +18,30 @@
 
 #include <assert.h>
 
+/*
+ * A list consists of a list head plus elements.
+ * Each element has 'next' and 'previous' pointers.
+ * The list head's pointers point to the first and the last element.
+ */
+
 struct list {
 	struct list *n, *p;
 };
 
+/*
+ * Initialise a list before use.
+ * The list head's next and previous pointers point back to itself.
+ */
 #define LIST_INIT(name)	struct list name = { &(name), &(name) }
-
 static inline void list_init(struct list *head)
 {
 	head->n = head->p = head;
 }
 
+/*
+ * Insert an element before 'head'.
+ * If 'head' is the list head, this adds an element to the end of the list.
+ */
 static inline void list_add(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -40,6 +53,10 @@ static inline void list_add(struct list *head, struct list *elem)
 	head->p = elem;
 }
 
+/*
+ * Insert an element after 'head'.
+ * If 'head' is the list head, this adds an element to the front of the list.
+ */
 static inline void list_add_h(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -51,53 +68,127 @@ static inline void list_add_h(struct list *head, struct list *elem)
 	head->n = elem;
 }
 
+/*
+ * Delete an element from its list.
+ * Note that this doesn't change the element itself - it may still be safe
+ * to follow its pointers.
+ */
 static inline void list_del(struct list *elem)
 {
 	elem->n->p = elem->p;
 	elem->p->n = elem->n;
 }
 
+/*
+ * Is the list empty?
+ */
 static inline int list_empty(struct list *head)
 {
 	return head->n == head;
 }
 
+/*
+ * Is this the first element of the list?
+ */
+static inline int list_start(struct list *head, struct list *elem)
+{
+	return elem->p == head;
+}
+
+/*
+ * Is this the last element of the list?
+ */
 static inline int list_end(struct list *head, struct list *elem)
 {
 	return elem->n == head;
 }
 
+/*
+ * Return the previous element of the list, or NULL if we've reached the start.
+ */
+static inline struct list *list_prev(struct list *head, struct list *elem)
+{
+	return (list_start(head, elem) ? NULL : elem->p);
+}
+
+/*
+ * Return the next element of the list, or NULL if we've reached the end.
+ */
 static inline struct list *list_next(struct list *head, struct list *elem)
 {
 	return (list_end(head, elem) ? NULL : elem->n);
 }
 
-#define list_item(v, t) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->list))
+/*
+ * Given the address v of an instance of 'struct list' called 'head' 
+ * contained in a structure of type t, return the containing structure.
+ */
+#define list_struct_base(v, t, head) \
+    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->head))
 
-#define list_struct_base(v, t, h) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->h))
+/*
+ * Given the address v of an instance of 'struct list list' contained in
+ * a structure of type t, return the containing structure.
+ */
+#define list_item(v, t) list_struct_base((v), t, list)
 
-/* Given a known element in a known structure, locate another */
+/*
+ * Given the address v of one known element e in a known structure of type t,
+ * return another element f.
+ */
 #define struct_field(v, t, e, f) \
     (((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->e))->f)
 
-/* Given a known element in a known structure, locate the list head */
+/*
+ * Given the address v of a known element e in a known structure of type t,
+ * return the list head 'list'
+ */
 #define list_head(v, t, e) struct_field(v, t, e, list)
 
+/*
+ * Set v to each element of a list in turn.
+ */
 #define list_iterate(v, head) \
 	for (v = (head)->n; v != head; v = v->n)
 
+/*
+ * Set v to each element in a list in turn, starting from the element 
+ * in front of 'start'.
+ * You can use this to 'unwind' a list_iterate and back out actions on
+ * already-processed elements.
+ * If 'start' is 'head' it walks the list backwards.
+ */
 #define list_uniterate(v, head, start) \
 	for (v = (start)->p; v != head; v = v->p)
 
+/*
+ * A safe way to walk a list and delete and free some elements along
+ * the way.
+ * t must be defined as a temporary variable of the same type as v.
+ */
 #define list_iterate_safe(v, t, head) \
 	for (v = (head)->n, t = v->n; v != head; v = t, t = v->n)
 
-#define list_iterate_items(v, head) \
-	for (v = list_item((head)->n, typeof(*v)); &v->list != (head); \
-	     v = list_item(v->list.n, typeof(*v)))
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The 'struct list' variable within the containing structure is 'field'.
+ */
+#define list_iterate_items_gen(v, head, field) \
+	for (v = list_struct_base((head)->n, typeof(*v), field); \
+	     &v->field != (head); \
+	     v = list_struct_base(v->field.n, typeof(*v), field))
 
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The list should be 'struct list list' within the containing structure.
+ */
+#define list_iterate_items(v, head) list_iterate_items_gen(v, (head), list)
+
+/*
+ * Return the number of elements in a list by walking it.
+ */
 static inline unsigned int list_size(const struct list *head)
 {
 	unsigned int s = 0;

lib/device/dev-io.c

@@ -330,8 +330,13 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 	}
 
 #ifdef O_DIRECT_SUPPORT
-	if (direct)
-		flags |= O_DIRECT;
+	if (direct) {
+		if (!(dev->flags & DEV_O_DIRECT_TESTED))
+			dev->flags |= DEV_O_DIRECT;
+
+		if ((dev->flags & DEV_O_DIRECT))
+			flags |= O_DIRECT;
+	}
 #endif
 
 #ifdef O_NOATIME
@@ -341,6 +346,16 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 #endif
 
 	if ((dev->fd = open(name, flags, 0777)) < 0) {
+#ifdef O_DIRECT_SUPPORT
+		if (direct && !(dev->flags & DEV_O_DIRECT_TESTED)) {
+			flags &= ~O_DIRECT;
+			if ((dev->fd = open(name, flags, 0777)) >= 0) {
+				dev->flags &= ~DEV_O_DIRECT;
+				log_debug("%s: Not using O_DIRECT", name);
+				goto opened;
+			}
+		}
+#endif
 		if (quiet)
 			log_sys_debug("open", name);
 		else
@@ -348,8 +363,14 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 		return 0;
 	}
 
+#ifdef O_DIRECT_SUPPORT
+      opened:
+	if (direct)
+		dev->flags |= DEV_O_DIRECT_TESTED;
+#endif
 	dev->open_count++;
 	dev->flags &= ~DEV_ACCESSED_W;
+
 	if ((flags & O_ACCMODE) == O_RDWR)
 		dev->flags |= DEV_OPENED_RW;
 	else
@@ -373,8 +394,9 @@ int dev_open_flags(struct device *dev, int flags, int direct, int quiet)
 
 	list_add(&_open_devices, &dev->open_list);
 
-	log_debug("Opened %s %s", dev_name(dev),
-		  dev->flags & DEV_OPENED_RW ? "RW" : "RO");
+	log_debug("Opened %s %s%s", dev_name(dev),
+		  dev->flags & DEV_OPENED_RW ? "RW" : "RO",
+		  dev->flags & DEV_O_DIRECT ? " O_DIRECT" : "");
 
 	return 1;
 }

lib/device/device.h

@@ -23,6 +23,8 @@
 #define DEV_REGULAR		0x00000002	/* Regular file? */
 #define DEV_ALLOCED		0x00000004	/* dbg_malloc used */
 #define DEV_OPENED_RW		0x00000008	/* Opened RW */
+#define DEV_O_DIRECT		0x00000010	/* Use O_DIRECT */
+#define DEV_O_DIRECT_TESTED	0x00000020	/* DEV_O_DIRECT is reliable */
 
 /*
  * All devices in LVM will be represented by one of these.
@@ -41,6 +43,7 @@ struct device {
 	struct list open_list;
 
 	char pvid[ID_LEN + 1];
+	char _padding[7];
 };
 
 struct device_list {
@@ -63,7 +66,7 @@ int dev_get_sectsize(struct device *dev, uint32_t *size);
 /* Use quiet version if device number could change e.g. when opening LV */
 int dev_open(struct device *dev);
 int dev_open_quiet(struct device *dev);
-int dev_open_flags(struct device *dev, int flags, int append, int quiet);
+int dev_open_flags(struct device *dev, int flags, int direct, int quiet);
 int dev_close(struct device *dev);
 int dev_close_immediate(struct device *dev);
 void dev_close_all(void);

lib/display/display.c

@@ -337,10 +337,9 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 		   void *handle)
 {
 	struct lvinfo info;
-	int inkernel, snap_active;
+	int inkernel, snap_active = 0;
 	char uuid[64];
-	struct snapshot *snap = NULL;
-	struct list *slh, *snaplist;
+	struct lv_segment *snap_seg = NULL;
 	float snap_percent;	/* fused, fsize; */
 
 	if (!id_write_format(&lv->lvid.id[1], uuid, sizeof(uuid))) {
@@ -364,27 +363,30 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 	if (lv_is_origin(lv)) {
 		log_print("LV snapshot status     source of");
 
-		snaplist = find_snapshots(lv);
-		list_iterate(slh, snaplist) {
-			snap = list_item(slh, struct snapshot_list)->snapshot;
-			snap_active = lv_snapshot_percent(snap->cow,
-							  &snap_percent);
-			if (!snap_active || snap_percent < 0 ||
-			    snap_percent >= 100) snap_active = 0;
+		list_iterate_items_gen(snap_seg, &lv->snapshot_segs,
+				       origin_list) {
+			if (inkernel &&
+			    (snap_active = lv_snapshot_percent(snap_seg->cow,
+							       &snap_percent)))
+				if (snap_percent < 0 || snap_percent >= 100)
+					snap_active = 0;
 			log_print("                       %s%s/%s [%s]",
 				  lv->vg->cmd->dev_dir, lv->vg->name,
-				  snap->cow->name,
+				  snap_seg->cow->name,
 				  (snap_active > 0) ? "active" : "INACTIVE");
 		}
-		snap = NULL;
-	} else if ((snap = find_cow(lv))) {
-		snap_active = lv_snapshot_percent(lv, &snap_percent);
-		if (!snap_active || snap_percent < 0 || snap_percent >= 100)
-			snap_active = 0;
+		snap_seg = NULL;
+	} else if ((snap_seg = find_cow(lv))) {
+		if (inkernel &&
+		    (snap_active = lv_snapshot_percent(snap_seg->cow,
+						       &snap_percent)))
+			if (snap_percent < 0 || snap_percent >= 100)
+				snap_active = 0;
+
 		log_print("LV snapshot status     %s destination for %s%s/%s",
 			  (snap_active > 0) ? "active" : "INACTIVE",
 			  lv->vg->cmd->dev_dir, lv->vg->name,
-			  snap->origin->name);
+			  snap_seg->origin->name);
 	}
 
 	if (inkernel && info.suspended)
@@ -402,15 +404,24 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 
 	log_print("LV Size                %s",
 		  display_size(cmd,
-			       snap ? snap->origin->size : lv->size,
+			       snap_seg ? snap_seg->origin->size : lv->size,
 			       SIZE_SHORT));
 
 	log_print("Current LE             %u",
-		  snap ? snap->origin->le_count : lv->le_count);
+		  snap_seg ? snap_seg->origin->le_count : lv->le_count);
 
-/********** FIXME allocation
-    log_print("Allocated LE           %u", lv->allocated_le);
-**********/
+	if (snap_seg) {
+		log_print("COW-table size         %s",
+			  display_size(cmd, (uint64_t) lv->size, SIZE_SHORT));
+		log_print("COW-table LE           %u", lv->le_count);
+
+		if (snap_active)
+			log_print("Allocated to snapshot  %.2f%% ", snap_percent);	
+
+		log_print("Snapshot chunk size    %s",
+			  display_size(cmd, (uint64_t) snap_seg->chunk_size,
+				       SIZE_SHORT));
+	}
 
 	log_print("Segments               %u", list_size(&lv->segments));
 
@@ -418,31 +429,6 @@ int lvdisplay_full(struct cmd_context *cmd, struct logical_volume *lv,
 	log_print("Stripe size (KByte)    %u", lv->stripesize / 2);
 ***********/
 
-	if (snap) {
-		if (snap_percent == -1)
-			snap_percent = 100;
-
-		log_print("Snapshot chunk size    %s",
-			  display_size(cmd, (uint64_t) snap->chunk_size,
-				       SIZE_SHORT));
-
-/*
-	size = display_size(lv->size, SIZE_SHORT);
-	sscanf(size, "%f", &fsize);
-	fused = fsize * snap_percent / 100;
-*/
-		log_print("Allocated to snapshot  %.2f%% ",	/* [%.2f/%s]", */
-			  snap_percent);	/*, fused, size); */
-		/* dbg_free(size); */
-	}
-
-/********** FIXME Snapshot
-	size = ???
-	log_print("Allocated to COW-table %s", size);
-	dbg_free(size);
-    }
-******************/
-
 	log_print("Allocation             %s", get_alloc_string(lv->alloc));
 	log_print("Read ahead sectors     %u", lv->read_ahead);
 
@@ -550,7 +536,7 @@ void vgdisplay_full(struct volume_group *vg)
 			  vg->status & SHARED ? "yes" : "no");
 	}
 	log_print("MAX LV                %u", vg->max_lv);
-	log_print("Cur LV                %u", vg->lv_count);
+	log_print("Cur LV                %u", vg->lv_count + vg->snapshot_count);
 	log_print("Open LV               %u", lvs_in_vg_opened(vg));
 /****** FIXME Max LV Size
       log_print ( "MAX LV Size           %s",

lib/filters/filter.c

@@ -70,6 +70,7 @@ static const device_info_t device_info[] = {
 	{"i2o_block", 16},	/* i2o Block Disk */
 	{"iseries/vd", 8},	/* iSeries disks */
 	{"gnbd", 1},		/* Network block device */
+	{"ramdisk", 1},		/* RAM disk */
 	{NULL, 0}
 };
 

lib/format1/format1.c

@@ -148,7 +148,6 @@ static struct volume_group *_build_vg(struct format_instance *fid,
 	vg->seqno = 0;
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	if (!_check_vgs(pvs, &partial))

lib/format1/import-export.c

@@ -26,6 +26,7 @@
 #include "filter.h"
 #include "toolcontext.h"
 #include "segtype.h"
+#include "pv_alloc.h"
 
 #include <time.h>
 
@@ -89,6 +90,13 @@ int import_pv(struct pool *mem, struct device *dev,
 	pv->pe_alloc_count = pvd->pe_allocated;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
+
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
 
 	return 1;
 }
@@ -323,6 +331,8 @@ int import_lv(struct pool *mem, struct logical_volume *lv, struct lv_disk *lvd)
 	lv->size = lvd->lv_size;
 	lv->le_count = lvd->lv_allocated_le;
 
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -495,7 +505,7 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 	       struct physical_volume *pv, const char *dev_dir)
 {
 	int r = 0;
-	struct list *lvh, *sh;
+	struct list *lvh;
 	struct lv_list *ll;
 	struct lvd_list *lvdl;
 	size_t len;
@@ -524,6 +534,9 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 
 	list_iterate(lvh, &vg->lvs) {
 		ll = list_item(lvh, struct lv_list);
+		if (ll->lv->status & SNAPSHOT)
+			continue;
+
 		if (!(lvdl = pool_alloc(dl->mem, sizeof(*lvdl)))) {
 			stack;
 			goto out;
@@ -532,7 +545,6 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 		_export_lv(&lvdl->lvd, vg, ll->lv, dev_dir);
 
 		lv_num = lvnum_from_lvid(&ll->lv->lvid);
-
 		lvdl->lvd.lv_number = lv_num;
 
 		if (!hash_insert(lvd_hash, ll->lv->name, &lvdl->lvd)) {
@@ -545,37 +557,20 @@ int export_lvs(struct disk_list *dl, struct volume_group *vg,
 			goto out;
 		}
 
+		if (lv_is_origin(ll->lv))
+			lvdl->lvd.lv_access |= LV_SNAPSHOT_ORG;
+
+		if (lv_is_cow(ll->lv)) {
+			lvdl->lvd.lv_access |= LV_SNAPSHOT;
+			lvdl->lvd.lv_chunk_size = ll->lv->snapshot->chunk_size;
+			lvdl->lvd.lv_snapshot_minor =
+			    lvnum_from_lvid(&ll->lv->snapshot->origin->lvid);
+		}
+
 		list_add(&dl->lvds, &lvdl->list);
 		dl->pvd.lv_cur++;
 	}
 
-	/*
-	 * Now we need to run through the snapshots, exporting
-	 * the SNAPSHOT_ORG flags etc.
-	 */
-	list_iterate(sh, &vg->snapshots) {
-		struct lv_disk *org, *cow;
-		struct snapshot *s = list_item(sh,
-					       struct snapshot_list)->snapshot;
-
-		if (!(org = hash_lookup(lvd_hash, s->origin->name))) {
-			log_err("Couldn't find snapshot origin '%s'.",
-				s->origin->name);
-			goto out;
-		}
-
-		if (!(cow = hash_lookup(lvd_hash, s->cow->name))) {
-			log_err("Couldn't find snapshot cow store '%s'.",
-				s->cow->name);
-			goto out;
-		}
-
-		org->lv_access |= LV_SNAPSHOT_ORG;
-		cow->lv_access |= LV_SNAPSHOT;
-		cow->lv_snapshot_minor = org->lv_number;
-		cow->lv_chunk_size = s->chunk_size;
-	}
-
 	r = 1;
 
       out:
@@ -646,7 +641,8 @@ int import_snapshots(struct pool *mem, struct volume_group *vg,
 				continue;
 
 			/* insert the snapshot */
-			if (!vg_add_snapshot(org, cow, 1, NULL, org->le_count, 
+			if (!vg_add_snapshot(vg->fid, NULL, org, cow, NULL,
+					     org->le_count, 
 					     lvd->lv_chunk_size)) {
 				log_err("Couldn't add snapshot.");
 				return 0;

lib/format1/import-extents.c

@@ -60,6 +60,8 @@ static struct hash_table *_create_lv_maps(struct pool *mem,
 
 	list_iterate(llh, &vg->lvs) {
 		ll = list_item(llh, struct lv_list);
+		if (ll->lv->status & SNAPSHOT)
+			continue;
 
 		if (!(lvm = pool_alloc(mem, sizeof(*lvm)))) {
 			stack;
@@ -205,58 +207,55 @@ static int _check_maps_are_complete(struct hash_table *maps)
 
 static int _read_linear(struct cmd_context *cmd, struct lv_map *lvm)
 {
-	uint32_t le = 0;
+	uint32_t le = 0, len;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
+
+	if (!(segtype = get_segtype_from_string(cmd, "striped"))) {
+		stack;
+		return 0;
+	}
 
 	while (le < lvm->lv->le_count) {
-		seg = alloc_lv_segment(cmd->mem, 1);
+		len = 0;
 
-		seg->lv = lvm->lv;
-		if (!(seg->segtype = get_segtype_from_string(cmd, "striped"))) {
-			stack;
+		do
+			len++;
+		while ((lvm->map[le + len].pv == lvm->map[le].pv) &&
+			 (lvm->map[le].pv &&
+			  lvm->map[le + len].pe == lvm->map[le].pe + len));
+
+		if (!(seg = alloc_lv_segment(cmd->mem, segtype, lvm->lv, le,
+					     len, 0, 0, 1, len, 0, 0))) {
+			log_error("Failed to allocate linear segment.");
 			return 0;
 		}
 
-		seg->le = le;
-		seg->len = 0;
-		seg->area_len = 0;
-		seg->stripe_size = 0;
-
-		seg->area[0].type = AREA_PV;
-		seg->area[0].u.pv.pv = lvm->map[le].pv;
-		seg->area[0].u.pv.pe = lvm->map[le].pe;
-
-		do {
-			seg->len++;
-			seg->area_len++;
-		} while ((lvm->map[le + seg->len].pv == seg->area[0].u.pv.pv) &&
-			 (seg->area[0].u.pv.pv &&
-			  lvm->map[le + seg->len].pe == seg->area[0].u.pv.pe +
-			  seg->len));
-
-		le += seg->len;
+		set_lv_segment_area_pv(seg, 0, lvm->map[le].pv, lvm->map[le].pe);
 
 		list_add(&lvm->lv->segments, &seg->list);
+
+		le += seg->len;
 	}
 
 	return 1;
 }
 
-static int _check_stripe(struct lv_map *lvm, struct lv_segment *seg,
-			 uint32_t base_le, uint32_t len)
+static int _check_stripe(struct lv_map *lvm, uint32_t area_count,
+			 uint32_t seg_len, uint32_t base_le, uint32_t len)
 {
-	uint32_t le, st;
-
-	le = base_le + seg->len;
+	uint32_t st;
 
 	/*
 	 * Is the next physical extent in every stripe adjacent to the last?
 	 */
-	for (st = 0; st < seg->area_count; st++)
-		if ((lvm->map[le + st * len].pv != seg->area[st].u.pv.pv) ||
-		    (seg->area[st].u.pv.pv &&
-		     lvm->map[le + st * len].pe !=
-		     seg->area[st].u.pv.pe + seg->len)) return 0;
+	for (st = 0; st < area_count; st++)
+		if ((lvm->map[base_le + st * len + seg_len].pv !=
+		     lvm->map[base_le + st * len].pv) ||
+		    (lvm->map[base_le + st * len].pv &&
+		     lvm->map[base_le + st * len + seg_len].pe !=
+		     lvm->map[base_le + st * len].pe + seg_len))
+			return 0;
 
 	return 1;
 }
@@ -264,7 +263,9 @@ static int _check_stripe(struct lv_map *lvm, struct lv_segment *seg,
 static int _read_stripes(struct cmd_context *cmd, struct lv_map *lvm)
 {
 	uint32_t st, le = 0, len;
+	uint32_t area_len;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 
 	/*
 	 * Work out overall striped length
@@ -276,43 +277,42 @@ static int _read_stripes(struct cmd_context *cmd, struct lv_map *lvm)
 	}
 	len = lvm->lv->le_count / lvm->stripes;
 
+	if (!(segtype = get_segtype_from_string(cmd, "striped"))) {
+		stack;
+		return 0;
+	}
+
 	while (le < len) {
-		if (!(seg = alloc_lv_segment(cmd->mem, lvm->stripes))) {
-			stack;
-			return 0;
-		}
-
-		seg->lv = lvm->lv;
-		if (!(seg->segtype = get_segtype_from_string(cmd, "striped"))) {
-			stack;
-			return 0;
-		}
-		seg->stripe_size = lvm->stripe_size;
-		seg->le = seg->area_count * le;
-		seg->len = 1;
-		seg->area_len = 1;
-
-		/*
-		 * Set up start positions of each stripe in this segment
-		 */
-		for (st = 0; st < seg->area_count; st++) {
-			seg->area[st].u.pv.pv = lvm->map[le + st * len].pv;
-			seg->area[st].u.pv.pe = lvm->map[le + st * len].pe;
-		}
+		area_len = 1;
 
 		/* 
 		 * Find how many blocks are contiguous in all stripes
 		 * and so can form part of this segment
 		 */
-		while (_check_stripe(lvm, seg, le, len)) {
-			seg->len++;
-			seg->area_len++;
+		while (_check_stripe(lvm, lvm->stripes,
+				     area_len * lvm->stripes, le, len))
+			area_len++;
+
+		if (!(seg = alloc_lv_segment(cmd->mem, segtype, lvm->lv,
+					     lvm->stripes * le,
+					     lvm->stripes * area_len,
+					     0, lvm->stripe_size, lvm->stripes,
+					     area_len, 0, 0))) {
+			log_error("Failed to allocate striped segment.");
+			return 0;
 		}
 
-		le += seg->len;
-		seg->len *= seg->area_count;
+		/*
+		 * Set up start positions of each stripe in this segment
+		 */
+		for (st = 0; st < seg->area_count; st++)
+			set_lv_segment_area_pv(seg, st,
+					       lvm->map[le + st * len].pv,
+					       lvm->map[le + st * len].pe);
 
 		list_add(&lvm->lv->segments, &seg->list);
+
+		le += seg->len;
 	}
 
 	return 1;

lib/format_pool/format_pool.c

@@ -132,7 +132,6 @@ static struct volume_group *_build_vg_from_pds(struct format_instance
 	vg->system_id = NULL;
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	if (!import_pool_vg(vg, smem, pds)) {

lib/format_pool/import_export.c

@@ -21,6 +21,7 @@
 #include "disk_rep.h"
 #include "sptype_names.h"
 #include "lv_alloc.h"
+#include "pv_alloc.h"
 #include "str_list.h"
 #include "display.h"
 #include "segtype.h"
@@ -82,6 +83,8 @@ int import_pool_lvs(struct volume_group *vg, struct pool *mem, struct list *pls)
 	lv->name = NULL;
 	lv->le_count = 0;
 	lv->read_ahead = 0;
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -112,6 +115,8 @@ int import_pool_lvs(struct volume_group *vg, struct pool *mem, struct list *pls)
 		} else {
 			lv->minor = -1;
 		}
+		lv->snapshot = NULL;
+		list_init(&lv->snapshot_segs);
 		list_init(&lv->segments);
 		list_init(&lv->tags);
 	}
@@ -178,6 +183,13 @@ int import_pool_pv(const struct format_type *fmt, struct pool *mem,
 	pv->pe_alloc_count = pv->pe_count;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
+
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
 
 	return 1;
 }
@@ -200,40 +212,41 @@ static int _add_stripe_seg(struct pool *mem,
 			   uint32_t *le_cur)
 {
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 	int j;
+	uint32_t area_len;
 
-	if (!(seg = alloc_lv_segment(mem, usp->num_devs))) {
-		log_error("Unable to allocate striped lv_segment structure");
-		return 0;
-	}
-	if(usp->striping & (usp->striping - 1)) {
+	if (usp->striping & (usp->striping - 1)) {
 		log_error("Stripe size must be a power of 2");
 		return 0;
 	}
-	seg->stripe_size = usp->striping;
-	seg->status |= 0;
-	seg->le += *le_cur;
+
+	area_len = (usp->devs[0].blocks) / POOL_PE_SIZE;
+
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd,
+						     "striped"))) {
+		stack;
+		return 0;
+	}
+
+	if (!(seg = alloc_lv_segment(mem, segtype, lv, *le_cur, 
+				     area_len * usp->num_devs, 0,
+				     usp->striping, usp->num_devs, area_len,
+				     0, 0))) {
+		log_error("Unable to allocate striped lv_segment structure");
+		return 0;
+	}
+
+	for (j = 0; j < usp->num_devs; j++)
+		set_lv_segment_area_pv(seg, j, usp->devs[j].pv, 0);
 
 	/* add the subpool type to the segment tag list */
 	str_list_add(mem, &seg->tags, _cvt_sptype(usp->type));
 
-	for (j = 0; j < usp->num_devs; j++) {
-		if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd,
-							     "striped"))) {
-			stack;
-			return 0;
-		}
-
-		seg->area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
-		seg->len += seg->area_len;
-		*le_cur += seg->area_len;
-		seg->lv = lv;
-
-		seg->area[j].type = AREA_PV;
-		seg->area[j].u.pv.pv = usp->devs[j].pv;
-		seg->area[j].u.pv.pe = 0;
-	}
 	list_add(&lv->segments, &seg->list);
+
+	*le_cur += seg->len;
+
 	return 1;
 }
 
@@ -242,37 +255,35 @@ static int _add_linear_seg(struct pool *mem,
 			   uint32_t *le_cur)
 {
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 	int j;
+	uint32_t area_len;
+
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
+		stack;
+		return 0;
+	}
 
 	for (j = 0; j < usp->num_devs; j++) {
-		/* linear segments only have 1 data area */
-		if (!(seg = alloc_lv_segment(mem, 1))) {
+		area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
+
+		if (!(seg = alloc_lv_segment(mem, segtype, lv, *le_cur,
+					     area_len, 0, usp->striping,
+					     1, area_len, POOL_PE_SIZE, 0))) {
 			log_error("Unable to allocate linear lv_segment "
 				  "structure");
 			return 0;
 		}
-		seg->stripe_size = usp->striping;
-		seg->le += *le_cur;
-		seg->chunk_size = POOL_PE_SIZE;
-		seg->status |= 0;
-		if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd,
-							     "striped"))) {
-			stack;
-			return 0;
-		}
+
 		/* add the subpool type to the segment tag list */
 		str_list_add(mem, &seg->tags, _cvt_sptype(usp->type));
 
-		seg->lv = lv;
-
-		seg->area_len = (usp->devs[j].blocks) / POOL_PE_SIZE;
-		seg->len = seg->area_len;
-		*le_cur += seg->len;
-		seg->area[0].type = AREA_PV;
-		seg->area[0].u.pv.pv = usp->devs[j].pv;
-		seg->area[0].u.pv.pe = 0;
+		set_lv_segment_area_pv(seg, 0, usp->devs[j].pv, 0);
 		list_add(&lv->segments, &seg->list);
+
+		*le_cur += seg->len;
 	}
+
 	return 1;
 }
 
@@ -288,8 +299,11 @@ int import_pool_segments(struct list *lvs, struct pool *mem,
 
 	list_iterate(lvhs, lvs) {
 		lvl = list_item(lvhs, struct lv_list);
-
 		lv = lvl->lv;
+
+		if (lv->status & SNAPSHOT)
+			continue;
+
 		for (i = 0; i < subpools; i++) {
 			if (usp[i].striping) {
 				if (!_add_stripe_seg(mem, &usp[i], lv, &le_cur)) {

lib/format_text/export.c

@@ -477,87 +477,6 @@ static int _count_segments(struct logical_volume *lv)
 	return r;
 }
 
-static int _print_snapshot(struct formatter *f, struct snapshot *snap,
-			   unsigned int count)
-{
-	char buffer[256];
-	struct lv_segment seg;
-
-	f->nl(f);
-
-	outf(f, "snapshot%u {", count);
-	_inc_indent(f);
-
-	if (!id_write_format(&snap->id, buffer, sizeof(buffer))) {
-		stack;
-		return 0;
-	}
-
-	outf(f, "id = \"%s\"", buffer);
-
-	seg.status = LVM_READ | LVM_WRITE | VISIBLE_LV;
-	if (!print_flags(seg.status, LV_FLAGS, buffer, sizeof(buffer))) {
-		stack;
-		return 0;
-	}
-
-	outf(f, "status = %s", buffer);
-	outf(f, "segment_count = 1");
-
-	f->nl(f);
-
-	if (!(seg.segtype = get_segtype_from_string(snap->origin->vg->cmd,
-						    "snapshot"))) {
-		stack;
-		return 0;
-	}
-
-	seg.le = 0;
-	seg.len = snap->le_count;
-	seg.origin = snap->origin;
-	seg.cow = snap->cow;
-	seg.chunk_size = snap->chunk_size;
-
-	/* FIXME Dummy values */
-	list_init(&seg.list);
-	seg.lv = snap->cow;
-	seg.stripe_size = 0;
-	seg.area_count = 0;
-	seg.area_len = 0;
-	seg.extents_copied = 0;
-
-	/* Can't tag a snapshot independently of its origin */
-	list_init(&seg.tags);
-
-	if (!_print_segment(f, snap->origin->vg, 1, &seg)) {
-		stack;
-		return 0;
-	}
-
-	_dec_indent(f);
-	outf(f, "}");
-
-	return 1;
-}
-
-static int _print_snapshots(struct formatter *f, struct volume_group *vg)
-{
-	struct list *sh;
-	struct snapshot *s;
-	unsigned int count = 0;
-
-	list_iterate(sh, &vg->snapshots) {
-		s = list_item(sh, struct snapshot_list)->snapshot;
-
-		if (!_print_snapshot(f, s, count++)) {
-			stack;
-			return 0;
-		}
-	}
-
-	return 1;
-}
-
 static int _print_lvs(struct formatter *f, struct volume_group *vg)
 {
 	struct list *lvh;
@@ -629,11 +548,6 @@ static int _print_lvs(struct formatter *f, struct volume_group *vg)
 		outf(f, "}");
 	}
 
-	if (!_print_snapshots(f, vg)) {
-		stack;
-		return 0;
-	}
-
 	_dec_indent(f);
 	outf(f, "}");
 

lib/format_text/flags.c

@@ -54,6 +54,7 @@ static struct flag _lv_flags[] = {
 	{LOCKED, "LOCKED"},
 	{MIRRORED, NULL},
 	{VIRTUAL, NULL},
+	{SNAPSHOT, NULL},
 	{0, NULL}
 };
 

lib/format_text/format-text.c

@@ -183,27 +183,30 @@ static int _raw_write_mda_header(const struct format_type *fmt,
 
 static struct raw_locn *_find_vg_rlocn(struct device_area *dev_area,
 				       struct mda_header *mdah,
-				       const char *vgname)
+				       const char *vgname,
+				       int precommit)
 {
 	size_t len;
 	char vgnamebuf[NAME_LEN + 2];
 	struct raw_locn *rlocn;
 	struct lvmcache_info *info;
 
-	rlocn = mdah->raw_locns;
+	rlocn = mdah->raw_locns;	/* Slot 0 */
 
+	if (precommit)
+		rlocn++;		/* Slot 1 */
+
+	/* FIXME Loop through rlocns two-at-a-time.  List null-terminated. */
 	/* FIXME Ignore if checksum incorrect!!! */
-	while (rlocn->offset) {
-		if (!dev_read(dev_area->dev, dev_area->start + rlocn->offset,
-			      sizeof(vgnamebuf), vgnamebuf)) {
-			stack;
-			goto error;
-		}
-		if (!strncmp(vgnamebuf, vgname, len = strlen(vgname)) &&
-		    (isspace(vgnamebuf[len]) || vgnamebuf[len] == '{')) {
-			return rlocn;
-		}
-		rlocn++;
+	if (!dev_read(dev_area->dev, dev_area->start + rlocn->offset,
+		      sizeof(vgnamebuf), vgnamebuf)) {
+		stack;
+		goto error;
+	}
+
+	if (!strncmp(vgnamebuf, vgname, len = strlen(vgname)) &&
+	    (isspace(vgnamebuf[len]) || vgnamebuf[len] == '{')) {
+		return rlocn;
 	}
 
       error:
@@ -213,32 +216,41 @@ static struct raw_locn *_find_vg_rlocn(struct device_area *dev_area,
 	return NULL;
 }
 
-static struct raw_locn *_vg_posn(struct format_instance *fid,
-				 struct device_area *dev_area,
-				 const char *vgname)
+/*
+ * Determine offset for uncommitted metadata
+ */
+static uint64_t _next_rlocn_offset(struct raw_locn *rlocn,
+				   struct mda_header *mdah)
 {
+	if (!rlocn)
+		/* Find an empty slot */
+		/* FIXME Assume only one VG per mdah for now */
+		return MDA_HEADER_SIZE;
 
-	struct mda_header *mdah;
-
-	if (!(mdah = _raw_read_mda_header(fid->fmt, dev_area))) {
-		stack;
-		return NULL;
-	}
-
-	return _find_vg_rlocn(dev_area, mdah, vgname);
+	/* Start of free space - round up to next sector; circular */
+	return ((rlocn->offset + rlocn->size +
+		(SECTOR_SIZE - rlocn->size % SECTOR_SIZE) -
+		MDA_HEADER_SIZE) % (mdah->size - MDA_HEADER_SIZE))
+	       + MDA_HEADER_SIZE;
 }
 
 static int _raw_holds_vgname(struct format_instance *fid,
 			     struct device_area *dev_area, const char *vgname)
 {
 	int r = 0;
+	struct mda_header *mdah;
 
 	if (!dev_open(dev_area->dev)) {
 		stack;
 		return 0;
 	}
 
-	if (_vg_posn(fid, dev_area, vgname))
+	if (!(mdah = _raw_read_mda_header(fid->fmt, dev_area))) {
+		stack;
+		return 0;
+	}
+
+	if (_find_vg_rlocn(dev_area, mdah, vgname, 0))
 		r = 1;
 
 	if (!dev_close(dev_area->dev))
@@ -249,7 +261,8 @@ static int _raw_holds_vgname(struct format_instance *fid,
 
 static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 					      const char *vgname,
-					      struct device_area *area)
+					      struct device_area *area,
+					      int precommit)
 {
 	struct volume_group *vg = NULL;
 	struct raw_locn *rlocn;
@@ -268,7 +281,7 @@ static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 		goto out;
 	}
 
-	if (!(rlocn = _vg_posn(fid, area, vgname))) {
+	if (!(rlocn = _find_vg_rlocn(area, mdah, vgname, precommit))) {
 		log_debug("VG %s not found on %s", vgname, dev_name(area->dev));
 		goto out;
 	}
@@ -292,8 +305,9 @@ static struct volume_group *_vg_read_raw_area(struct format_instance *fid,
 		stack;
 		goto out;
 	}
-	log_debug("Read %s metadata (%u) from %s at %" PRIu64 " size %" PRIu64,
-		  vg->name, vg->seqno, dev_name(area->dev),
+	log_debug("Read %s %smetadata (%u) from %s at %" PRIu64 " size %"
+		  PRIu64, vg->name, precommit ? "pre-commit " : "",
+		  vg->seqno, dev_name(area->dev),
 		  area->start + rlocn->offset, rlocn->size);
 
       out:
@@ -309,7 +323,16 @@ static struct volume_group *_vg_read_raw(struct format_instance *fid,
 {
 	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
 
-	return _vg_read_raw_area(fid, vgname, &mdac->area);
+	return _vg_read_raw_area(fid, vgname, &mdac->area, 0);
+}
+
+static struct volume_group *_vg_read_precommit_raw(struct format_instance *fid,
+						   const char *vgname,
+						   struct metadata_area *mda)
+{
+	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
+
+	return _vg_read_raw_area(fid, vgname, &mdac->area, 1);
 }
 
 static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
@@ -349,18 +372,8 @@ static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if ((rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
-		/* Start of free space - round up to next sector; circular */
-		mdac->rlocn.offset =
-		    ((rlocn->offset + rlocn->size +
-		      (SECTOR_SIZE - rlocn->size % SECTOR_SIZE) -
-		      MDA_HEADER_SIZE) % (mdah->size - MDA_HEADER_SIZE))
-		    + MDA_HEADER_SIZE;
-	} else {
-		/* Find an empty slot */
-		/* FIXME Assume only one VG per mdah for now */
-		mdac->rlocn.offset = MDA_HEADER_SIZE;
-	}
+	rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0);
+	mdac->rlocn.offset = _next_rlocn_offset(rlocn, mdah);
 
 	if (!(mdac->rlocn.size = text_vg_export_raw(vg, "", buf, sizeof(buf)))) {
 		log_error("VG %s metadata writing failed", vg->name);
@@ -425,8 +438,10 @@ static int _vg_write_raw(struct format_instance *fid, struct volume_group *vg,
 	return r;
 }
 
-static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
-			  struct metadata_area *mda)
+static int _vg_commit_raw_rlocn(struct format_instance *fid,
+				struct volume_group *vg,
+				struct metadata_area *mda,
+				int precommit)
 {
 	struct mda_context *mdac = (struct mda_context *) mda->metadata_locn;
 	struct mda_header *mdah;
@@ -453,18 +468,23 @@ static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
-		rlocn = &mdah->raw_locns[0];
+	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0))) {
+		mdah->raw_locns[0].offset = 0;
 		mdah->raw_locns[1].offset = 0;
+		mdah->raw_locns[2].offset = 0;
+		rlocn = &mdah->raw_locns[0];
 	}
 
+	if (precommit)
+		rlocn++;
+
 	rlocn->offset = mdac->rlocn.offset;
 	rlocn->size = mdac->rlocn.size;
 	rlocn->checksum = mdac->rlocn.checksum;
 
-	log_debug("Committing %s metadata (%u) to %s header at %" PRIu64,
-		  vg->name, vg->seqno, dev_name(mdac->area.dev),
-		  mdac->area.start);
+	log_debug("%sCommitting %s metadata (%u) to %s header at %" PRIu64,
+		  precommit ? "Pre-" : "", vg->name, vg->seqno,
+		  dev_name(mdac->area.dev), mdac->area.start);
 	if (!_raw_write_mda_header(fid->fmt, mdac->area.dev, mdac->area.start,
 				   mdah)) {
 		log_error("Failed to write metadata area header");
@@ -474,12 +494,25 @@ static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
 	r = 1;
 
       out:
-	if (!dev_close(mdac->area.dev))
+	if (!precommit && !dev_close(mdac->area.dev))
 		stack;
 
 	return r;
 }
 
+static int _vg_commit_raw(struct format_instance *fid, struct volume_group *vg,
+			  struct metadata_area *mda)
+{
+	return _vg_commit_raw_rlocn(fid, vg, mda, 0);
+}
+
+static int _vg_precommit_raw(struct format_instance *fid,
+			     struct volume_group *vg,
+			     struct metadata_area *mda)
+{
+	return _vg_commit_raw_rlocn(fid, vg, mda, 1);
+}
+
 /* Close metadata area devices */
 static int _vg_revert_raw(struct format_instance *fid, struct volume_group *vg,
 			  struct metadata_area *mda)
@@ -525,7 +558,7 @@ static int _vg_remove_raw(struct format_instance *fid, struct volume_group *vg,
 		goto out;
 	}
 
-	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name))) {
+	if (!(rlocn = _find_vg_rlocn(&mdac->area, mdah, vg->name, 0))) {
 		rlocn = &mdah->raw_locns[0];
 		mdah->raw_locns[1].offset = 0;
 	}
@@ -551,13 +584,13 @@ static int _vg_remove_raw(struct format_instance *fid, struct volume_group *vg,
 
 static struct volume_group *_vg_read_file_name(struct format_instance *fid,
 					       const char *vgname,
-					       const char *path_live)
+					       const char *read_path)
 {
 	struct volume_group *vg;
 	time_t when;
 	char *desc;
 
-	if (!(vg = text_vg_import_file(fid, path_live, &when, &desc))) {
+	if (!(vg = text_vg_import_file(fid, read_path, &when, &desc))) {
 		stack;
 		return NULL;
 	}
@@ -570,10 +603,10 @@ static struct volume_group *_vg_read_file_name(struct format_instance *fid,
 	if (vgname && strcmp(vgname, vg->name)) {
 		pool_free(fid->fmt->cmd->mem, vg);
 		log_err("'%s' does not contain volume group '%s'.",
-			path_live, vgname);
+			read_path, vgname);
 		return NULL;
 	} else
-		log_debug("Read volume group %s from %s", vg->name, path_live);
+		log_debug("Read volume group %s from %s", vg->name, read_path);
 
 	return vg;
 }
@@ -587,6 +620,15 @@ static struct volume_group *_vg_read_file(struct format_instance *fid,
 	return _vg_read_file_name(fid, vgname, tc->path_live);
 }
 
+static struct volume_group *_vg_read_precommit_file(struct format_instance *fid,
+						    const char *vgname,
+						    struct metadata_area *mda)
+{
+	struct text_context *tc = (struct text_context *) mda->metadata_locn;
+
+	return _vg_read_file_name(fid, vgname, tc->path_edit);
+}
+
 static int _vg_write_file(struct format_instance *fid, struct volume_group *vg,
 			  struct metadata_area *mda)
 {
@@ -858,7 +900,7 @@ static int _scan_raw(const struct format_type *fmt)
 		if (vgname_from_mda(fmt, &rl->dev_area, vgnamebuf,
 				    sizeof(vgnamebuf))) {
 			if ((vg = _vg_read_raw_area(&fid, vgnamebuf,
-						    &rl->dev_area)))
+						    &rl->dev_area, 0)))
 				lvmcache_update_vg(vg);
 		}
 	}
@@ -1107,34 +1149,6 @@ static int _pv_write(const struct format_type *fmt, struct physical_volume *pv,
 	return 1;
 }
 
-static int _get_pv_from_vg(const struct format_type *fmt, const char *vg_name,
-			   const char *id, struct physical_volume *pv)
-{
-	struct volume_group *vg;
-	struct list *pvh;
-	struct pv_list *pvl;
-	int consistent = 0;
-
-	if (!(vg = vg_read(fmt->cmd, vg_name, &consistent))) {
-		log_error("format_text: _vg_read failed to read VG %s",
-			  vg_name);
-		return 0;
-	}
-
-	if (!consistent)
-		log_error("Warning: Volume group %s is not consistent",
-			  vg_name);
-
-	list_iterate(pvh, &vg->pvs) {
-		pvl = list_item(pvh, struct pv_list);
-		if (id_equal(&pvl->pv->id, (const struct id *) id)) {
-			memcpy(pv, pvl->pv, sizeof(*pv));
-			return 1;
-		}
-	}
-	return 0;
-}
-
 static int _add_raw(struct list *raw_list, struct device_area *dev_area)
 {
 	struct raw_list *rl;
@@ -1183,8 +1197,8 @@ static int _pv_read(const struct format_type *fmt, const char *pv_name,
 
 	/* Have we already cached vgname? */
 	if (info->vginfo && info->vginfo->vgname && *info->vginfo->vgname &&
-	    _get_pv_from_vg(info->fmt, info->vginfo->vgname, info->dev->pvid,
-			    pv)) {
+	    get_pv_from_vg_by_id(info->fmt, info->vginfo->vgname,
+				 info->dev->pvid, pv)) {
 		return 1;
 	}
 
@@ -1194,8 +1208,8 @@ static int _pv_read(const struct format_type *fmt, const char *pv_name,
 
 		if (info->vginfo && info->vginfo->vgname &&
 		    *info->vginfo->vgname &&
-		    _get_pv_from_vg(info->fmt, info->vginfo->vgname,
-				    info->dev->pvid, pv)) {
+		    get_pv_from_vg_by_id(info->fmt, info->vginfo->vgname,
+					 info->dev->pvid, pv)) {
 			return 1;
 		}
 	}
@@ -1280,6 +1294,7 @@ static void _destroy(const struct format_type *fmt)
 
 static struct metadata_area_ops _metadata_text_file_ops = {
 	vg_read:_vg_read_file,
+	vg_read_precommit:_vg_read_precommit_file,
 	vg_write:_vg_write_file,
 	vg_remove:_vg_remove_file,
 	vg_commit:_vg_commit_file
@@ -1294,8 +1309,10 @@ static struct metadata_area_ops _metadata_text_file_backup_ops = {
 
 static struct metadata_area_ops _metadata_text_raw_ops = {
 	vg_read:_vg_read_raw,
+	vg_read_precommit:_vg_read_precommit_raw,
 	vg_write:_vg_write_raw,
 	vg_remove:_vg_remove_raw,
+	vg_precommit:_vg_precommit_raw,
 	vg_commit:_vg_commit_raw,
 	vg_revert:_vg_revert_raw
 };
@@ -1651,7 +1668,8 @@ struct format_type *create_text_format(struct cmd_context *cmd)
 	fmt->ops = &_text_handler;
 	fmt->name = FMT_TEXT_NAME;
 	fmt->alias = FMT_TEXT_ALIAS;
-	fmt->features = FMT_SEGMENTS | FMT_MDAS | FMT_TAGS | FMT_UNLIMITED_VOLS;
+	fmt->features = FMT_SEGMENTS | FMT_MDAS | FMT_TAGS | FMT_PRECOMMIT |
+			FMT_UNLIMITED_VOLS;
 
 	if (!(mda_lists = dbg_malloc(sizeof(struct mda_lists)))) {
 		log_error("Failed to allocate dir_list");

lib/format_text/import_vsn1.c

@@ -22,6 +22,7 @@
 #include "toolcontext.h"
 #include "lvmcache.h"
 #include "lv_alloc.h"
+#include "pv_alloc.h"
 #include "segtype.h"
 #include "text_import.h"
 
@@ -190,6 +191,8 @@ static int _read_pv(struct format_instance *fid, struct pool *mem,
 	}
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	/* Optional tags */
 	if ((cn = find_config_node(pvn, "tags")) &&
@@ -208,6 +211,11 @@ static int _read_pv(struct format_instance *fid, struct pool *mem,
 	pv->pe_alloc_count = 0;
 	pv->fmt = fid->fmt;
 
+	if (!alloc_pv_segment_whole_pv(mem, pv)) {
+		stack;
+		return 0;
+	}
+
 	vg->pv_count++;
 	list_add(&vg->pvs, &pvl->list);
 
@@ -283,19 +291,13 @@ static int _read_segment(struct pool *mem, struct volume_group *vg,
 		return 0;
 	}
 
-	if (!(seg = alloc_lv_segment(mem, area_count))) {
+	if (!(seg = alloc_lv_segment(mem, segtype, lv, start_extent,
+				     extent_count, 0, 0, area_count,
+				     extent_count, 0, 0))) {
 		log_error("Segment allocation failed");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->le = start_extent;
-	seg->len = extent_count;
-	seg->area_len = extent_count;
-	seg->status = 0u;
-	seg->segtype = segtype;
-	seg->extents_copied = 0u;
-
 	if (seg->segtype->ops->text_import &&
 	    !seg->segtype->ops->text_import(seg, sn, pv_hash)) {
 		stack;
@@ -361,19 +363,14 @@ int text_import_areas(struct lv_segment *seg, const struct config_node *sn,
 
 		/* FIXME Cope if LV not yet read in */
 		if ((pv = hash_lookup(pv_hash, cv->v.str))) {
-			seg->area[s].type = AREA_PV;
-			seg->area[s].u.pv.pv = pv;
-			seg->area[s].u.pv.pe = cv->next->v.i;
+			set_lv_segment_area_pv(seg, s, pv, cv->next->v.i);
 			/*
 			 * Adjust extent counts in the pv and vg.
 			 */
 			pv->pe_alloc_count += seg->area_len;
 			seg->lv->vg->free_count -= seg->area_len;
-
 		} else if ((lv1 = find_lv(seg->lv->vg, cv->v.str))) {
-			seg->area[s].type = AREA_LV;
-			seg->area[s].u.lv.lv = lv1;
-			seg->area[s].u.lv.le = cv->next->v.i;
+			set_lv_segment_area_lv(seg, s, lv1, cv->next->v.i);
 		} else {
 			log_error("Couldn't find volume '%s' "
 				  "for segment '%s'.",
@@ -508,6 +505,8 @@ static int _read_lvnames(struct format_instance *fid, struct pool *mem,
 	if (!_read_int32(lvn, "read_ahead", &lv->read_ahead))
 		lv->read_ahead = 0;
 
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
@@ -561,24 +560,29 @@ static int _read_lvsegs(struct format_instance *fid, struct pool *mem,
 
 	lv->size = (uint64_t) lv->le_count * (uint64_t) vg->extent_size;
 
-	/* Skip this for now for snapshots */
-	if (!(lv->status & SNAPSHOT)) {
-		lv->minor = -1;
-		if ((lv->status & FIXED_MINOR) &&
-		    !_read_int32(lvn, "minor", &lv->minor)) {
-			log_error("Couldn't read minor number for logical "
-				  "volume %s.", lv->name);
-			return 0;
-		}
-		lv->major = -1;
-		if ((lv->status & FIXED_MINOR) &&
-		    !_read_int32(lvn, "major", &lv->major)) {
-			log_error("Couldn't read major number for logical "
-				  "volume %s.", lv->name);
-		}
-	} else {
+	/*
+	 * FIXME We now have 2 LVs for each snapshot. The real one was
+	 * created by vg_add_snapshot from the segment text_import.
+	 */
+	if (lv->status & SNAPSHOT) {
 		vg->lv_count--;
 		list_del(&lvl->list);
+		return 1;
+	}
+
+	lv->minor = -1;
+	if ((lv->status & FIXED_MINOR) &&
+	    !_read_int32(lvn, "minor", &lv->minor)) {
+		log_error("Couldn't read minor number for logical "
+			  "volume %s.", lv->name);
+		return 0;
+	}
+
+	lv->major = -1;
+	if ((lv->status & FIXED_MINOR) &&
+	    !_read_int32(lvn, "major", &lv->major)) {
+		log_error("Couldn't read major number for logical "
+			  "volume %s.", lv->name);
 	}
 
 	return 1;
@@ -736,7 +740,6 @@ static struct volume_group *_read_vg(struct format_instance *fid,
 	}
 
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 
 	/* Optional tags */

lib/metadata/lv_alloc.h

@@ -16,5 +16,23 @@
 #ifndef _LVM_LV_ALLOC_H
 #include "pool.h"
 
-struct lv_segment *alloc_lv_segment(struct pool *mem, uint32_t num_areas);
+struct lv_segment *alloc_lv_segment(struct pool *mem,
+				    struct segment_type *segtype,
+				    struct logical_volume *lv,
+				    uint32_t le, uint32_t len,
+				    uint32_t status,
+				    uint32_t stripe_size,
+				    uint32_t area_count,
+				    uint32_t area_len,
+				    uint32_t chunk_size,
+				    uint32_t extents_copied);
+
+struct lv_segment *alloc_snapshot_seg(struct logical_volume *lv,
+				      uint32_t allocated);
+
+void set_lv_segment_area_pv(struct lv_segment *seg, uint32_t area_num,
+			    struct physical_volume *pv, uint32_t pe);
+void set_lv_segment_area_lv(struct lv_segment *seg, uint32_t area_num,
+			    struct logical_volume *lv, uint32_t le);
+
 #endif

lib/metadata/lv_manip.c

@@ -61,22 +61,57 @@ static void _put_extents(struct lv_segment *seg)
 	}
 }
 
-struct lv_segment *alloc_lv_segment(struct pool *mem, uint32_t num_areas)
+struct lv_segment *alloc_lv_segment(struct pool *mem,
+				    struct segment_type *segtype,
+				    struct logical_volume *lv,
+				    uint32_t le, uint32_t len,
+				    uint32_t status,
+				    uint32_t stripe_size,
+				    uint32_t area_count,
+				    uint32_t area_len,
+				    uint32_t chunk_size,
+				    uint32_t extents_copied)
 {
 	struct lv_segment *seg;
-	uint32_t len = sizeof(*seg) + (num_areas * sizeof(seg->area[0]));
+	uint32_t sz = sizeof(*seg) + (area_count * sizeof(seg->area[0]));
 
-	if (!(seg = pool_zalloc(mem, len))) {
+	if (!(seg = pool_zalloc(mem, sz))) {
 		stack;
 		return NULL;
 	}
 
-	seg->area_count = num_areas;
+	seg->segtype = segtype;
+	seg->lv = lv;
+	seg->le = le;
+	seg->len = len;
+	seg->status = status;
+	seg->stripe_size = stripe_size;
+	seg->area_count = area_count;
+	seg->area_len = area_len;
+	seg->chunk_size = chunk_size;
+	seg->extents_copied = extents_copied;
 	list_init(&seg->tags);
 
 	return seg;
 }
 
+void set_lv_segment_area_pv(struct lv_segment *seg, uint32_t area_num,
+			    struct physical_volume *pv, uint32_t pe)
+{
+	seg->area[area_num].type = AREA_PV;
+	seg->area[area_num].u.pv.pv = pv;
+	seg->area[area_num].u.pv.pe = pe;
+	seg->area[area_num].u.pv.pvseg = NULL;
+}
+
+void set_lv_segment_area_lv(struct lv_segment *seg, uint32_t area_num,
+			    struct logical_volume *lv, uint32_t le)
+{
+	seg->area[area_num].type = AREA_LV;
+	seg->area[area_num].u.lv.lv = lv;
+	seg->area[area_num].u.lv.le = le;
+}
+
 static int _alloc_parallel_area(struct logical_volume *lv, uint32_t area_count,
 				uint32_t stripe_size,
 				struct segment_type *segtype,
@@ -98,24 +133,17 @@ static int _alloc_parallel_area(struct logical_volume *lv, uint32_t area_count,
 	if (smallest < area_len)
 		area_len = smallest;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, area_count))) {
-		log_err("Couldn't allocate new parallel segment.");
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     area_len * (striped ? area_count : 1),
+				     0u, stripe_size, area_count, area_len,
+				     0u, 0u))) {
+		log_error("Couldn't allocate new parallel segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->le = *ix;
-	seg->len = area_len * (striped ? area_count : 1);
-	seg->area_len = area_len;
-	seg->stripe_size = stripe_size;
-	seg->extents_copied = 0u;
-
 	for (s = 0; s < area_count; s++) {
 		struct pv_area *pva = areas[s];
-		seg->area[s].type = AREA_PV;
-		seg->area[s].u.pv.pv = pva->map->pvl->pv;
-		seg->area[s].u.pv.pe = pva->start;
+		set_lv_segment_area_pv(seg, s, pva->map->pvl->pv, pva->start);
 		consume_pv_area(pva, area_len);
 	}
 
@@ -221,34 +249,31 @@ static int _alloc_linear_area(struct logical_volume *lv, uint32_t *ix,
 {
 	uint32_t count, remaining;
 	struct lv_segment *seg;
+	struct segment_type *segtype;
 
 	count = pva->count;
 	remaining = lv->le_count - *ix;
 	if (count > remaining)
 		count = remaining;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 1))) {
-		log_err("Couldn't allocate new stripe segment.");
-		return 0;
-	}
-
-	seg->lv = lv;
-	if (!(seg->segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
+	if (!(segtype = get_segtype_from_string(lv->vg->cmd, "striped"))) {
 		stack;
 		return 0;
 	}
-	seg->le = *ix;
-	seg->len = count;
-	seg->area_len = count;
-	seg->stripe_size = 0;
-	seg->area[0].type = AREA_PV;
-	seg->area[0].u.pv.pv = map->pvl->pv;
-	seg->area[0].u.pv.pe = pva->start;
+
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     count, 0, 0, 1, count, 0, 0))) {
+		log_error("Couldn't allocate new stripe segment.");
+		return 0;
+	}
+
+	set_lv_segment_area_pv(seg, 0, map->pvl->pv, pva->start);
 
 	list_add(&lv->segments, &seg->list);
 
-	consume_pv_area(pva, count);
-	*ix += count;
+	consume_pv_area(pva, seg->len);
+	*ix += seg->len;
+
 	return 1;
 }
 
@@ -266,30 +291,21 @@ static int _alloc_mirrored_area(struct logical_volume *lv, uint32_t *ix,
 	if (count > remaining)
 		count = remaining;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 2))) {
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, *ix,
+				     count, 0, 0, 2, count, 0, 0))) {
 		log_err("Couldn't allocate new mirrored segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->le = *ix;
-	seg->status = 0u;
-	seg->len = count;
-	seg->area_len = count;
-	seg->stripe_size = 0;
-	seg->extents_copied = 0u;
 	/* FIXME Remove AREA_PV restriction here? */
-	seg->area[0].type = AREA_PV;
-	seg->area[0].u.pv.pv = mirrored_pv;
-	seg->area[0].u.pv.pe = mirrored_pe;
-	seg->area[1].type = AREA_PV;
-	seg->area[1].u.pv.pv = map->pvl->pv;
-	seg->area[1].u.pv.pe = pva->start;
+	set_lv_segment_area_pv(seg, 0, mirrored_pv, mirrored_pe);
+	set_lv_segment_area_pv(seg, 1, map->pvl->pv, pva->start);
+
 	list_add(&lv->segments, &seg->list);
 
-	consume_pv_area(pva, count);
-	*ix += count;
+	consume_pv_area(pva, seg->len);
+	*ix += seg->len;
+
 	return 1;
 }
 
@@ -421,25 +437,44 @@ static int _alloc_virtual(struct logical_volume *lv,
 {
 	struct lv_segment *seg;
 
-	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, 0))) {
-		log_err("Couldn't allocate new zero segment.");
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, allocated,
+				     lv->le_count - allocated, 0, 0, 0,
+				     lv->le_count - allocated, 0, 0))) {
+		log_error("Couldn't allocate new zero segment.");
 		return 0;
 	}
 
-	seg->lv = lv;
-	seg->segtype = segtype;
-	seg->status = 0u;
-	seg->le = allocated;
-	seg->len = lv->le_count - allocated;
-	seg->area_len = seg->len;
-	seg->stripe_size = 0;
-	seg->extents_copied = 0u;
 	list_add(&lv->segments, &seg->list);
 	lv->status |= VIRTUAL;
 
 	return 1;
 }
 
+struct lv_segment *alloc_snapshot_seg(struct logical_volume *lv,
+				      uint32_t allocated)
+{
+	struct lv_segment *seg;
+	struct segment_type *segtype;
+
+	segtype = get_segtype_from_string(lv->vg->cmd, "snapshot");
+	if (!segtype) {
+		log_error("Failed to find snapshot segtype");
+		return NULL;
+	}
+
+	if (!(seg = alloc_lv_segment(lv->vg->cmd->mem, segtype, lv, allocated,
+				     lv->le_count - allocated, 0, 0, 0,
+				     lv->le_count - allocated, 0, 0))) {
+		log_error("Couldn't allocate new snapshot segment.");
+		return NULL;
+	}
+
+	list_add(&lv->segments, &seg->list);
+	lv->status |= VIRTUAL;
+
+	return seg;
+}
+
 /*
  * Chooses a correct allocation policy.
  */
@@ -543,8 +578,10 @@ static char *_generate_lv_name(struct volume_group *vg, const char *format,
 struct logical_volume *lv_create_empty(struct format_instance *fi,
 				       const char *name,
 				       const char *name_format,
+				       union lvid *lvid,
 				       uint32_t status,
 				       alloc_policy_t alloc,
+				       int import,
 				       struct volume_group *vg)
 {
 	struct cmd_context *cmd = vg->cmd;
@@ -565,7 +602,8 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 		return NULL;
 	}
 
-	log_verbose("Creating logical volume %s", name);
+	if (!import)
+		log_verbose("Creating logical volume %s", name);
 
 	if (!(ll = pool_zalloc(cmd->mem, sizeof(*ll))) ||
 	    !(ll->lv = pool_zalloc(cmd->mem, sizeof(*ll->lv)))) {
@@ -592,9 +630,14 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 	lv->minor = -1;
 	lv->size = UINT64_C(0);
 	lv->le_count = 0;
+	lv->snapshot = NULL;
+	list_init(&lv->snapshot_segs);
 	list_init(&lv->segments);
 	list_init(&lv->tags);
 
+	if (lvid)
+		lv->lvid = *lvid;
+
 	if (fi->fmt->ops->lv_setup && !fi->fmt->ops->lv_setup(fi, lv)) {
 		stack;
 		if (ll)
@@ -602,7 +645,9 @@ struct logical_volume *lv_create_empty(struct format_instance *fi,
 		return NULL;
 	}
 
-	vg->lv_count++;
+	if (!import)
+		vg->lv_count++;
+
 	list_add(&vg->lvs, &ll->list);
 
 	return lv;

lib/metadata/merge.c

@@ -98,11 +98,17 @@ static int _lv_split_segment(struct logical_volume *lv, struct lv_segment *seg,
 	}
 
 	/* Clone the existing segment */
-	if (!(split_seg = alloc_lv_segment(lv->vg->cmd->mem, seg->area_count))) {
-		log_error("Couldn't allocate new LV segment.");
+	if (!(split_seg = alloc_lv_segment(lv->vg->cmd->mem, seg->segtype,
+					   seg->lv, seg->le, seg->len,
+					   seg->status, seg->stripe_size,
+					   seg->area_count, seg->area_len,
+					   seg->chunk_size,
+					   seg->extents_copied))) {
+		log_error("Couldn't allocate cloned LV segment.");
 		return 0;
 	}
 
+	/* FIXME Avoid the memcpy */
 	len = sizeof(*seg) + (seg->area_count * sizeof(seg->area[0]));
 	memcpy(split_seg, seg, len);
 

lib/metadata/metadata.c

@@ -21,6 +21,8 @@
 #include "lvm-string.h"
 #include "lvmcache.h"
 #include "memlock.h"
+#include "str_list.h"
+#include "pv_alloc.h"
 
 static int _add_pv_to_vg(struct format_instance *fid, struct volume_group *vg,
 			 const char *pv_name)
@@ -109,6 +111,57 @@ static int _add_pv_to_vg(struct format_instance *fid, struct volume_group *vg,
 	return 1;
 }
 
+static int _copy_pv(struct physical_volume *pv_to,
+		    struct physical_volume *pv_from)
+{
+	memcpy(pv_to, pv_from, sizeof(*pv_to));
+
+	if (!str_list_dup(pv_to->fmt->cmd->mem, &pv_to->tags, &pv_from->tags)) {
+		log_error("PV tags duplication failed");
+		return 0;
+	}
+
+	if (!peg_dup(pv_to->fmt->cmd->mem, &pv_to->segments,
+		     &pv_to->free_segments, &pv_from->segments)) {
+		stack;
+		return 0;
+	}
+
+	return 1;
+}
+
+int get_pv_from_vg_by_id(const struct format_type *fmt, const char *vg_name,
+			 const char *id, struct physical_volume *pv)
+{
+	struct volume_group *vg;
+	struct list *pvh;
+	struct pv_list *pvl;
+	int consistent = 0;
+
+	if (!(vg = vg_read(fmt->cmd, vg_name, &consistent))) {
+		log_error("get_pv_from_vg_by_id: vg_read failed to read VG %s",
+			  vg_name);
+		return 0;
+	}
+
+	if (!consistent)
+		log_error("Warning: Volume group %s is not consistent",
+			  vg_name);
+
+	list_iterate(pvh, &vg->pvs) {
+		pvl = list_item(pvh, struct pv_list);
+		if (id_equal(&pvl->pv->id, (const struct id *) id)) {
+			if (!_copy_pv(pv, pvl->pv)) {
+				stack;
+				return 0;
+			}
+			return 1;
+		}
+	}
+
+	return 0;
+}
+
 int vg_rename(struct cmd_context *cmd, struct volume_group *vg,
 	      const char *new_name)
 {
@@ -221,7 +274,6 @@ struct volume_group *vg_create(struct cmd_context *cmd, const char *vg_name,
 	list_init(&vg->lvs);
 
 	vg->snapshot_count = 0;
-	list_init(&vg->snapshots);
 
 	list_init(&vg->tags);
 
@@ -249,6 +301,153 @@ struct volume_group *vg_create(struct cmd_context *cmd, const char *vg_name,
 	return NULL;
 }
 
+static int _recalc_extents(uint32_t *extents, const char *desc1,
+			   const char *desc2, uint32_t old_size,
+			   uint32_t new_size)
+{
+	uint64_t size = (uint64_t) old_size * (*extents);
+
+	if (size % new_size) {
+		log_error("New size %" PRIu64 " for %s%s not an exact number "
+			  "of new extents.", size, desc1, desc2);
+		return 0;
+	}
+
+	size /= new_size;
+
+	if (size > UINT32_MAX) {
+		log_error("New extent count %" PRIu64 " for %s%s exceeds "
+			  "32 bits.", size, desc1, desc2);
+		return 0;
+	}
+
+	*extents = (uint32_t) size;
+
+	return 1;
+}
+
+int vg_change_pesize(struct cmd_context *cmd, struct volume_group *vg,
+		     uint32_t new_size)
+{
+	uint32_t old_size = vg->extent_size;
+	struct pv_list *pvl;
+	struct lv_list *lvl;
+	struct physical_volume *pv;
+	struct logical_volume *lv;
+	struct lv_segment *seg;
+	uint32_t s;
+
+	vg->extent_size = new_size;
+
+	if (vg->fid->fmt->ops->vg_setup &&
+	    !vg->fid->fmt->ops->vg_setup(vg->fid, vg)) {
+		stack;
+		return 0;
+	}
+
+	if (!_recalc_extents(&vg->extent_count, vg->name, "", old_size,
+			     new_size)) {
+		stack;
+		return 0;
+	}
+
+	if (!_recalc_extents(&vg->free_count, vg->name, " free space",
+			     old_size, new_size)) {
+		stack;
+		return 0;
+	}
+
+	/* foreach PV */
+	list_iterate_items(pvl, &vg->pvs) {
+		pv = pvl->pv;
+
+		pv->pe_size = new_size;
+		if (!_recalc_extents(&pv->pe_count, dev_name(pv->dev), "",
+				     old_size, new_size)) {
+			stack;
+			return 0;
+		}
+
+		if (!_recalc_extents(&pv->pe_alloc_count, dev_name(pv->dev),
+				     " allocated space", old_size, new_size)) {
+			stack;
+			return 0;
+		}
+	}
+
+	/* foreach LV */
+	list_iterate_items(lvl, &vg->lvs) {
+		lv = lvl->lv;
+
+		if (!_recalc_extents(&lv->le_count, lv->name, "", old_size,
+				     new_size)) {
+			stack;
+			return 0;
+		}
+
+		list_iterate_items(seg, &lv->segments) {
+			if (!_recalc_extents(&seg->le, lv->name,
+					     " segment start", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->len, lv->name,
+					     " segment length", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->area_len, lv->name,
+					     " area length", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			if (!_recalc_extents(&seg->extents_copied, lv->name,
+					     " extents moved", old_size,
+					     new_size)) {
+				stack;
+				return 0;
+			}
+
+			/* foreach area */
+			for (s = 0; s < seg->area_count; s++) {
+				switch (seg->area[s].type) {
+				case AREA_PV:
+					if (!_recalc_extents
+					    (&seg->area[s].u.pv.pe, lv->name,
+					     " area start", old_size,
+					     new_size)) {
+						stack;
+						return 0;
+					}
+					break;
+				case AREA_LV:
+					if (!_recalc_extents
+					    (&seg->area[s].u.lv.le, lv->name,
+					     " area start", old_size,
+					     new_size)) {
+						stack;
+						return 0;
+					}
+					break;
+				default:
+					log_error("Unrecognised segment type "
+						  "%u", seg->area[s].type);
+					return 0;
+				}
+			}
+		}
+
+	}
+
+	return 1;
+}
+
 /* Sizes in sectors */
 struct physical_volume *pv_create(const struct format_type *fmt,
 				  struct device *dev,
@@ -311,6 +510,8 @@ struct physical_volume *pv_create(const struct format_type *fmt,
 	pv->fmt = fmt;
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	if (!fmt->ops->pv_setup(fmt, pe_start, existing_extent_count,
 				existing_extent_size,
@@ -507,6 +708,7 @@ int vg_write(struct volume_group *vg)
 			/* Revert */
 			list_uniterate(mdah2, &vg->fid->metadata_areas, mdah) {
 				mda = list_item(mdah2, struct metadata_area);
+
 				if (mda->ops->vg_revert &&
 				    !mda->ops->vg_revert(vg->fid, vg, mda)) {
 					stack;
@@ -528,6 +730,24 @@ int vg_write(struct volume_group *vg)
 		}
 	}
 
+	/* Now pre-commit each copy of the new metadata */
+	list_iterate(mdah, &vg->fid->metadata_areas) {
+		mda = list_item(mdah, struct metadata_area);
+		if (mda->ops->vg_precommit &&
+		    !mda->ops->vg_precommit(vg->fid, vg, mda)) {
+			stack;
+			/* Revert */
+			list_iterate(mdah2, &vg->fid->metadata_areas) {
+				mda = list_item(mdah2, struct metadata_area);
+				if (mda->ops->vg_revert &&
+				    !mda->ops->vg_revert(vg->fid, vg, mda)) {
+					stack;
+				}
+			}
+			return 0;
+		}
+	}
+
 	return 1;
 }
 
@@ -598,7 +818,6 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd)
 	}
 	list_init(&vg->pvs);
 	list_init(&vg->lvs);
-	list_init(&vg->snapshots);
 	list_init(&vg->tags);
 	vg->cmd = cmd;
 	if (!(vg->name = pool_strdup(cmd->mem, ORPHAN))) {
@@ -630,8 +849,9 @@ static struct volume_group *_vg_read_orphans(struct cmd_context *cmd)
  * and take appropriate action if it isn't (e.g. abort; get write lock 
  * and call vg_read again).
  */
-struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
-			     int *consistent)
+static struct volume_group *_vg_read(struct cmd_context *cmd,
+				     const char *vgname,
+				     int *consistent, int precommitted)
 {
 	struct format_instance *fid;
 	const struct format_type *fmt;
@@ -641,6 +861,11 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	int inconsistent = 0;
 
 	if (!*vgname) {
+		if (precommitted) {
+			log_error("Internal error: vg_read requires vgname "
+				  "with pre-commit.");
+			return NULL;
+		}
 		*consistent = 1;
 		return _vg_read_orphans(cmd);
 	}
@@ -662,6 +887,12 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 		}
 	}
 
+	if (precommitted && !(fmt->features & FMT_PRECOMMIT)) {
+		log_error("Internal error: %s doesn't support "
+			  "pre-commit", fmt->name);
+		return NULL;
+	}
+
 	/* create format instance with appropriate metadata area */
 	if (!(fid = fmt->ops->create_instance(fmt, vgname, NULL))) {
 		log_error("Failed to create format instance");
@@ -671,7 +902,10 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	/* Ensure contents of all metadata areas match - else do recovery */
 	list_iterate(mdah, &fid->metadata_areas) {
 		mda = list_item(mdah, struct metadata_area);
-		if (!(vg = mda->ops->vg_read(fid, vgname, mda))) {
+		if ((precommitted &&
+		     !(vg = mda->ops->vg_read_precommit(fid, vgname, mda))) ||
+		    (!precommitted &&
+		     !(vg = mda->ops->vg_read(fid, vgname, mda)))) {
 			inconsistent = 1;
 			continue;
 		}
@@ -697,6 +931,12 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 			return NULL;
 		}
 
+		if (precommitted && !(fmt->features & FMT_PRECOMMIT)) {
+			log_error("Internal error: %s doesn't support "
+				  "pre-commit", fmt->name);
+			return NULL;
+		}
+
 		/* create format instance with appropriate metadata area */
 		if (!(fid = fmt->ops->create_instance(fmt, vgname, NULL))) {
 			log_error("Failed to create format instance");
@@ -706,7 +946,11 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 		/* Ensure contents of all metadata areas match - else recover */
 		list_iterate(mdah, &fid->metadata_areas) {
 			mda = list_item(mdah, struct metadata_area);
-			if (!(vg = mda->ops->vg_read(fid, vgname, mda))) {
+			if ((precommitted &&
+			     !(vg = mda->ops->vg_read_precommit(fid, vgname,
+								mda))) ||
+			    (!precommitted &&
+			     !(vg = mda->ops->vg_read(fid, vgname, mda)))) {
 				inconsistent = 1;
 				continue;
 			}
@@ -732,6 +976,12 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	lvmcache_update_vg(correct_vg);
 
 	if (inconsistent) {
+		if (precommitted) {
+			log_error("Inconsistent pre-commit metadata copies "
+				  "for volume group %s", vgname);
+			return NULL;
+		}
+
 		if (!*consistent)
 			return correct_vg;
 
@@ -769,6 +1019,19 @@ struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
 	return correct_vg;
 }
 
+struct volume_group *vg_read(struct cmd_context *cmd, const char *vgname,
+			     int *consistent)
+{
+	return _vg_read(cmd, vgname, consistent, 0);
+}
+
+struct volume_group *vg_read_precommitted(struct cmd_context *cmd,
+					  const char *vgname,
+					  int *consistent)
+{
+	return _vg_read(cmd, vgname, consistent, 1);
+}
+
 /* This is only called by lv_from_lvid, which is only called from 
  * activate.c so we know the appropriate VG lock is already held and 
  * the vg_read is therefore safe.
@@ -868,14 +1131,14 @@ struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
 
 	if (!(dev = dev_cache_get(pv_name, cmd->filter))) {
 		stack;
-		return 0;
+		return NULL;
 	}
 
 	if (!(label_read(dev, &label))) {
 		if (warnings)
 			log_error("No physical volume label read from %s",
 				  pv_name);
-		return 0;
+		return NULL;
 	}
 
 	info = (struct lvmcache_info *) label->info;
@@ -884,22 +1147,29 @@ struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
 
 	if (!(pv = pool_zalloc(cmd->mem, sizeof(*pv)))) {
 		log_error("pv allocation for '%s' failed", pv_name);
-		return 0;
+		return NULL;
 	}
 
 	list_init(&pv->tags);
+	list_init(&pv->segments);
+	list_init(&pv->free_segments);
 
 	/* FIXME Move more common code up here */
 	if (!(info->fmt->ops->pv_read(info->fmt, pv_name, pv, mdas))) {
 		log_error("Failed to read existing physical volume '%s'",
 			  pv_name);
-		return 0;
+		return NULL;
 	}
 
 	if (!pv->size)
 		return NULL;
-	else
-		return pv;
+	
+	if (!alloc_pv_segment_whole_pv(cmd->mem, pv)) {
+		stack;
+		return NULL;
+	}
+
+	return pv;
 }
 
 /* May return empty list */

lib/metadata/metadata.h

@@ -50,7 +50,7 @@
 #define VISIBLE_LV		0x00000040	/* LV */
 #define FIXED_MINOR		0x00000080	/* LV */
 /* FIXME Remove when metadata restructuring is completed */
-#define SNAPSHOT		0x00001000	/* LV - tmp internal use only */
+#define SNAPSHOT		0x00001000	/* LV - internal use only */
 #define PVMOVE			0x00002000	/* VG LV SEG */
 #define LOCKED			0x00004000	/* LV */
 #define MIRRORED		0x00008000	/* LV - internal use only */
@@ -68,6 +68,7 @@
 #define FMT_UNLIMITED_VOLS	0x00000008	/* Unlimited PVs/LVs? */
 #define FMT_RESTRICTED_LVIDS	0x00000010	/* LVID <= 255 */
 #define FMT_ORPHAN_ALLOCATABLE	0x00000020	/* Orphan PV allocatable? */
+#define FMT_PRECOMMIT		0x00000040	/* Supports pre-commit? */
 
 typedef enum {
 	ALLOC_INVALID,
@@ -98,6 +99,20 @@ struct format_type {
 	void *private;
 };
 
+struct pv_segment {
+	struct list list;	/* Member of pv->segments: ordered list
+				 * covering entire data area on this PV */
+
+	struct physical_volume *pv;
+	uint32_t pe;
+	uint32_t len;
+
+	struct lv_segment *lvseg;	/* NULL if free space */
+	uint32_t lv_area;	/* Index to area in LV segment */
+
+	struct list freelist;	/* Member of pv->free_segments */
+};
+
 struct physical_volume {
 	struct id id;
 	struct device *dev;
@@ -113,6 +128,8 @@ struct physical_volume {
 	uint32_t pe_count;
 	uint32_t pe_alloc_count;
 
+	struct list segments;	/* Ordered pv_segments covering complete PV */
+	struct list free_segments;	/* Free pv_segments for this PV */
 	struct list tags;
 };
 
@@ -124,6 +141,9 @@ struct metadata_area_ops {
 	struct volume_group *(*vg_read) (struct format_instance * fi,
 					 const char *vg_name,
 					 struct metadata_area * mda);
+	struct volume_group *(*vg_read_precommit) (struct format_instance * fi,
+					 const char *vg_name,
+					 struct metadata_area * mda);
 	/*
 	 * Write out complete VG metadata.  You must ensure internal
 	 * consistency before calling. eg. PEs can't refer to PVs not
@@ -140,6 +160,9 @@ struct metadata_area_ops {
 	 */
 	int (*vg_write) (struct format_instance * fid, struct volume_group * vg,
 			 struct metadata_area * mda);
+	int (*vg_precommit) (struct format_instance * fid,
+			     struct volume_group * vg,
+			     struct metadata_area * mda);
 	int (*vg_commit) (struct format_instance * fid,
 			  struct volume_group * vg, struct metadata_area * mda);
 	int (*vg_revert) (struct format_instance * fid,
@@ -184,11 +207,8 @@ struct volume_group {
 
 	/* logical volumes */
 	uint32_t lv_count;
-	struct list lvs;
-
-	/* snapshots */
 	uint32_t snapshot_count;
-	struct list snapshots;
+	struct list lvs;
 
 	struct list tags;
 };
@@ -210,7 +230,8 @@ struct lv_segment {
 	uint32_t area_len;
 	struct logical_volume *origin;
 	struct logical_volume *cow;
-	uint32_t chunk_size;
+	struct list origin_list;
+	uint32_t chunk_size;	/* In sectors */
 	uint32_t extents_copied;
 
 	struct list tags;
@@ -222,6 +243,7 @@ struct lv_segment {
 			struct {
 				struct physical_volume *pv;
 				uint32_t pe;
+				struct pv_segment *pvseg;
 			} pv;
 			struct {
 				struct logical_volume *lv;
@@ -246,21 +268,14 @@ struct logical_volume {
 	uint64_t size;
 	uint32_t le_count;
 
+	uint32_t origin_count;
+	struct list snapshot_segs;
+	struct lv_segment *snapshot;
+
 	struct list segments;
 	struct list tags;
 };
 
-struct snapshot {
-	struct id id;
-
-	int persistent;		/* boolean */
-	uint32_t chunk_size;	/* in 512 byte sectors */
-	uint32_t le_count;
-
-	struct logical_volume *origin;
-	struct logical_volume *cow;
-};
-
 struct name_list {
 	struct list list;
 	char *name;
@@ -284,17 +299,16 @@ struct lv_list {
 	struct logical_volume *lv;
 };
 
-struct snapshot_list {
-	struct list list;
-
-	struct snapshot *snapshot;
-};
-
 struct mda_list {
 	struct list list;
 	struct device_area mda;
 };
 
+struct peg_list {
+	struct list list;
+	struct pv_segment *peg;
+};
+
 /*
  * Ownership of objects passes to caller.
  */
@@ -374,6 +388,9 @@ int vg_commit(struct volume_group *vg);
 int vg_revert(struct volume_group *vg);
 struct volume_group *vg_read(struct cmd_context *cmd, const char *vg_name,
 			     int *consistent);
+struct volume_group *vg_read_precommitted(struct cmd_context *cmd,
+					  const char *vg_name,
+					  int *consistent);
 struct volume_group *vg_read_by_vgid(struct cmd_context *cmd, const char *vgid);
 struct physical_volume *pv_read(struct cmd_context *cmd, const char *pv_name,
 				struct list *mdas, uint64_t *label_sector,
@@ -407,13 +424,17 @@ int vg_rename(struct cmd_context *cmd, struct volume_group *vg,
 	      const char *new_name);
 int vg_extend(struct format_instance *fi, struct volume_group *vg,
 	      int pv_count, char **pv_names);
+int vg_change_pesize(struct cmd_context *cmd, struct volume_group *vg,
+		     uint32_t new_extent_size);
 
 /* Manipulate LVs */
 struct logical_volume *lv_create_empty(struct format_instance *fi,
 				       const char *name,
 				       const char *name_format,
+				       union lvid *lvid,
 				       uint32_t status,
 				       alloc_policy_t alloc,
+				       int import,
 				       struct volume_group *vg);
 
 int lv_reduce(struct format_instance *fi,
@@ -440,6 +461,8 @@ struct physical_volume *pv_find(struct volume_group *vg, const char *pv_name);
 struct pv_list *find_pv_in_vg(struct volume_group *vg, const char *pv_name);
 struct physical_volume *find_pv_in_vg_by_uuid(struct volume_group *vg,
 					      struct id *id);
+int get_pv_from_vg_by_id(const struct format_type *fmt, const char *vg_name,
+			 const char *id, struct physical_volume *pv);
 
 /* Find an LV within a given VG */
 struct lv_list *find_lv_in_vg(struct volume_group *vg, const char *lv_name);
@@ -493,12 +516,11 @@ int lv_is_cow(const struct logical_volume *lv);
 
 int pv_is_in_vg(struct volume_group *vg, struct physical_volume *pv);
 
-struct snapshot *find_cow(const struct logical_volume *lv);
-struct snapshot *find_origin(const struct logical_volume *lv);
-struct list *find_snapshots(const struct logical_volume *lv);
+struct lv_segment *find_cow(const struct logical_volume *lv);
 
-int vg_add_snapshot(struct logical_volume *origin, struct logical_volume *cow,
-		    int persistent, struct id *id, uint32_t extent_count,
+int vg_add_snapshot(struct format_instance *fid, const char *name,
+		    struct logical_volume *origin, struct logical_volume *cow,
+		    union lvid *lvid, uint32_t extent_count,
 		    uint32_t chunk_size);
 
 int vg_remove_snapshot(struct volume_group *vg, struct logical_volume *cow);

lib/metadata/mirror.c

@@ -19,6 +19,7 @@
 #include "segtype.h"
 #include "display.h"
 #include "activate.h"
+#include "lv_alloc.h"
 
 /* 
  * Replace any LV segments on given PV with temporary mirror.
@@ -153,9 +154,7 @@ int insert_pvmove_mirrors(struct cmd_context *cmd,
 						  "temporary LV for pvmove.");
 					return 0;
 				}
-				seg->area[s].type = AREA_LV;
-				seg->area[s].u.lv.lv = lv_mirr;
-				seg->area[s].u.lv.le = start_le;
+				set_lv_segment_area_lv(seg, s, lv_mirr, start_le);
 	
 				extent_count += seg->area_len;
 	
@@ -225,9 +224,9 @@ int remove_pvmove_mirrors(struct volume_group *vg,
 				else
 					c = 0;
 
-				seg->area[s].type = AREA_PV;
-				seg->area[s].u.pv.pv = mir_seg->area[c].u.pv.pv;
-				seg->area[s].u.pv.pe = mir_seg->area[c].u.pv.pe;
+				set_lv_segment_area_pv(seg, s,
+						       mir_seg->area[c].u.pv.pv,
+						       mir_seg->area[c].u.pv.pe);
 
 				/* Replace mirror with old area */
 				if (!

lib/metadata/pv_alloc.h

@@ -0,0 +1,22 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef _LVM_PV_ALLOC_H
+#include "pool.h"
+
+int alloc_pv_segment_whole_pv(struct pool *mem, struct physical_volume *pv);
+int peg_dup(struct pool *mem, struct list *peg_new, struct list *peg_free_new,
+	    struct list *peg_old);
+
+#endif

lib/metadata/pv_manip.c

@@ -0,0 +1,83 @@
+/*
+ * Copyright (C) 2003 Sistina Software, Inc. All rights reserved.
+ * Copyright (C) 2004 - 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of LVM2.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License v.2.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#include "lib.h"
+#include "pool.h"
+#include "metadata.h"
+#include "pv_alloc.h"
+
+static struct pv_segment *_alloc_pv_segment(struct pool *mem,
+					    struct physical_volume *pv,
+					    uint32_t pe, uint32_t len)
+{
+	struct pv_segment *peg;
+
+	if (!(peg = pool_zalloc(mem, sizeof(*peg)))) {
+		log_error("pv_segment allocation failed");
+		return NULL;
+	}
+
+	peg->pv = pv;
+	peg->pe = pe;
+	peg->len = len;
+        peg->lvseg = NULL;
+	peg->lv_area = 0;
+
+	list_init(&peg->list);
+	list_init(&peg->freelist);
+
+	return peg;
+}
+
+int alloc_pv_segment_whole_pv(struct pool *mem, struct physical_volume *pv)
+{
+        struct pv_segment *peg;
+
+	/* FIXME Cope with holes in PVs */
+        if (!(peg = _alloc_pv_segment(mem, pv, 0, pv->pe_count))) {
+                stack;
+                return 0;
+        }
+
+        list_add(&pv->segments, &peg->list);
+        list_add(&pv->free_segments, &peg->freelist);
+
+        return 1;
+}
+
+int peg_dup(struct pool *mem, struct list *peg_new, struct list *peg_free_new,
+	    struct list *peg_old)
+{
+	struct pv_segment *peg, *pego;
+
+	list_init(peg_new);
+	list_init(peg_free_new);
+
+	list_iterate_items(pego, peg_old) {
+		if (!(peg = _alloc_pv_segment(mem, pego->pv, pego->pe,
+					      pego->len))) {
+			stack;
+			return 0;
+		} 
+		peg->lvseg = pego->lvseg;
+		peg->lv_area = pego->lv_area;
+		list_add(peg_new, &peg->list);
+		if (!peg->lvseg)
+			list_add(peg_free_new, &peg->freelist);
+	}
+
+	return 1;
+}
+

lib/metadata/snapshot_manip.c

@@ -16,101 +16,31 @@
 #include "lib.h"
 #include "metadata.h"
 #include "toolcontext.h"
+#include "lv_alloc.h"
 
 int lv_is_origin(const struct logical_volume *lv)
 {
-	struct list *slh;
-	struct snapshot *s;
-
-	list_iterate(slh, &lv->vg->snapshots) {
-		s = list_item(slh, struct snapshot_list)->snapshot;
-		if (s->origin == lv)
-			return 1;
-	}
-
-	return 0;
+	return lv->origin_count ? 1 : 0;
 }
 
 int lv_is_cow(const struct logical_volume *lv)
 {
-	struct list *slh;
-	struct snapshot *s;
-
-	list_iterate(slh, &lv->vg->snapshots) {
-		s = list_item(slh, struct snapshot_list)->snapshot;
-		if (s->cow == lv)
-			return 1;
-	}
-
-	return 0;
+	return lv->snapshot ? 1 : 0;
 }
 
-struct snapshot *find_origin(const struct logical_volume *lv)
+/* Given a cow LV, return the snapshot lv_segment that uses it */
+struct lv_segment *find_cow(const struct logical_volume *lv)
 {
-	struct list *slh;
-	struct snapshot *s;
-
-	list_iterate(slh, &lv->vg->snapshots) {
-		s = list_item(slh, struct snapshot_list)->snapshot;
-		if (s->origin == lv)
-			return s;
-	}
-
-	return NULL;
+	return lv->snapshot;
 }
 
-struct snapshot *find_cow(const struct logical_volume *lv)
+int vg_add_snapshot(struct format_instance *fid, const char *name,
+		    struct logical_volume *origin,
+		    struct logical_volume *cow, union lvid *lvid,
+		    uint32_t extent_count, uint32_t chunk_size)
 {
-	struct list *slh;
-	struct snapshot *s;
-
-	list_iterate(slh, &lv->vg->snapshots) {
-		s = list_item(slh, struct snapshot_list)->snapshot;
-		if (s->cow == lv)
-			return s;
-	}
-
-	return NULL;
-}
-
-struct list *find_snapshots(const struct logical_volume *lv)
-{
-	struct list *slh;
-	struct list *snaplist;
-	struct snapshot *s;
-	struct snapshot_list *newsl;
-	struct pool *mem = lv->vg->cmd->mem;
-
-	if (!(snaplist = pool_alloc(mem, sizeof(*snaplist)))) {
-		log_error("snapshot name list allocation failed");
-		return NULL;
-	}
-
-	list_init(snaplist);
-
-	list_iterate(slh, &lv->vg->snapshots) {
-		s = list_item(slh, struct snapshot_list)->snapshot;
-		if (!(s->origin == lv))
-			continue;
-		if (!(newsl = pool_alloc(mem, sizeof(*newsl)))) {
-			log_error("snapshot_list structure allocation failed");
-			pool_free(mem, snaplist);
-			return NULL;
-		}
-		newsl->snapshot = s;
-		list_add(snaplist, &newsl->list);
-	}
-
-	return snaplist;
-}
-
-int vg_add_snapshot(struct logical_volume *origin, struct logical_volume *cow,
-		    int persistent, struct id *id, uint32_t extent_count,
-		    uint32_t chunk_size)
-{
-	struct snapshot *s;
-	struct snapshot_list *sl;
-	struct pool *mem = origin->vg->cmd->mem;
+	struct logical_volume *snap;
+	struct lv_segment *seg;
 
 	/*
 	 * Is the cow device already being used ?
@@ -120,55 +50,53 @@ int vg_add_snapshot(struct logical_volume *origin, struct logical_volume *cow,
 		return 0;
 	}
 
-	if (!(s = pool_alloc(mem, sizeof(*s)))) {
+	if (!(snap = lv_create_empty(fid, name, name ? NULL : "snapshot%d",
+				     lvid, LVM_READ | LVM_WRITE | VISIBLE_LV,
+				     ALLOC_INHERIT, 1, origin->vg))) {
 		stack;
 		return 0;
 	}
 
-	s->persistent = persistent;
-	s->chunk_size = chunk_size;
-	s->le_count = extent_count;
-	s->origin = origin;
-	s->cow = cow;
+	snap->le_count = extent_count;
 
-	if (id)
-		s->id = *id;
-	else if (!id_create(&s->id)) {
-		log_error("Random UUID creation failed for snapshot %s.",
-			  cow->name);
-		return 0;
-	}
-
-	if (!(sl = pool_alloc(mem, sizeof(*sl)))) {
+	if (!(seg = alloc_snapshot_seg(snap, 0))) {
 		stack;
-		pool_free(mem, s);
 		return 0;
 	}
 
+	seg->chunk_size = chunk_size;
+	seg->origin = origin;
+	seg->cow = cow;
+	seg->lv->status |= SNAPSHOT;
+
+	origin->origin_count++;
+	origin->vg->snapshot_count++;
+	origin->vg->lv_count--;
+	cow->snapshot = seg;
+
 	cow->status &= ~VISIBLE_LV;
-	sl->snapshot = s;
-	list_add(&origin->vg->snapshots, &sl->list);
-	origin->vg->snapshot_count++;
+
+	list_add(&origin->snapshot_segs, &seg->origin_list);
 
 	return 1;
 }
 
 int vg_remove_snapshot(struct volume_group *vg, struct logical_volume *cow)
 {
-	struct list *slh;
-	struct snapshot_list *sl;
+	list_del(&cow->snapshot->origin_list);
+	cow->snapshot->origin->origin_count--;
 
-	list_iterate(slh, &vg->snapshots) {
-		sl = list_item(slh, struct snapshot_list);
-
-		if (sl->snapshot->cow == cow) {
-			list_del(slh);
-			vg->snapshot_count--;
-			return 1;
-		}
+	if (!lv_remove(vg, cow->snapshot->lv)) {
+		log_error("Failed to remove internal snapshot LV %s",
+			  cow->snapshot->lv->name);
+		return 0;
 	}
 
-	/* fail */
-	log_err("Asked to remove an unknown snapshot.");
-	return 0;
+	cow->snapshot = NULL;
+
+	vg->snapshot_count--;
+	vg->lv_count++;
+	cow->status |= VISIBLE_LV;
+
+	return 1;
 }

lib/mirror/mirrored.c

@@ -131,6 +131,7 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 	int areas = seg->area_count;
 	int start_area = 0u;
 	uint32_t region_size, region_max;
+	int ret;
 
 	if (!*target_state)
 		*target_state = _init_target(mem, cft);
@@ -166,10 +167,10 @@ static int _compose_target_line(struct dev_manager *dm, struct pool *mem,
 				    region_size);
 		}
 
-		if ((*pos = lvm_snprintf(params, paramsize, "core 1 %u %u ",
-					 region_size, areas)) < 0) {
+		if ((ret = compose_log_line(dm, seg, params, paramsize, pos,
+					    areas, region_size)) <= 0) {
 			stack;
-			return -1;
+			return ret;
 		}
 	}
 

lib/report/columns.h

@@ -71,4 +71,7 @@ FIELD(SEGS, seg, NUM, "Start", list, 5, segstart, "seg_start")
 FIELD(SEGS, seg, NUM, "SSize", list, 5, segsize, "seg_size")
 FIELD(SEGS, seg, STR, "Seg Tags", tags, 8, tags, "seg_tags")
 FIELD(SEGS, seg, STR, "Devices", list, 5, devices, "devices")
+
+FIELD(PVSEGS, pvseg, NUM, "Start", pe, 5, uint32, "pvseg_start")
+FIELD(PVSEGS, pvseg, NUM, "SSize", len, 5, uint32, "pvseg_size")
 /* *INDENT-ON* */

lib/report/report.c

@@ -31,6 +31,7 @@ static union {
 	struct logical_volume _lv;
 	struct volume_group _vg;
 	struct lv_segment _seg;
+	struct pv_segment _pvseg;
 } _dummy;
 
 /*
@@ -324,7 +325,7 @@ static int _lvstatus_disp(struct report_handle *rh, struct field *field,
 	const struct logical_volume *lv = (const struct logical_volume *) data;
 	struct lvinfo info;
 	char *repstr;
-	struct snapshot *snap;
+	struct lv_segment *snap_seg;
 	float snap_percent;
 
 	if (!(repstr = pool_zalloc(rh->mem, 7))) {
@@ -373,8 +374,8 @@ static int _lvstatus_disp(struct report_handle *rh, struct field *field,
 			repstr[5] = '-';
 
 		/* Snapshot dropped? */
-		if ((snap = find_cow(lv)) &&
-		    (!lv_snapshot_percent(snap->cow, &snap_percent) ||
+		if ((snap_seg = find_cow(lv)) &&
+		    (!lv_snapshot_percent(snap_seg->cow, &snap_percent) ||
 		     snap_percent < 0 || snap_percent >= 100)) {
 			repstr[0] = toupper(repstr[0]);
 			if (info.suspended)
@@ -478,10 +479,10 @@ static int _origin_disp(struct report_handle *rh, struct field *field,
 			const void *data)
 {
 	const struct logical_volume *lv = (const struct logical_volume *) data;
-	struct snapshot *snap;
+	struct lv_segment *snap_seg;
 
-	if ((snap = find_cow(lv)))
-		return _string_disp(rh, field, &snap->origin->name);
+	if ((snap_seg = find_cow(lv)))
+		return _string_disp(rh, field, &snap_seg->origin->name);
 
 	field->report_string = "";
 	field->sort_value = (const void *) field->report_string;
@@ -762,7 +763,7 @@ static int _snpercent_disp(struct report_handle *rh, struct field *field,
 			   const void *data)
 {
 	const struct logical_volume *lv = (const struct logical_volume *) data;
-	struct snapshot *snap;
+	struct lv_segment *snap_seg;
 	struct lvinfo info;
 	float snap_percent;
 	uint64_t *sortval;
@@ -773,15 +774,16 @@ static int _snpercent_disp(struct report_handle *rh, struct field *field,
 		return 0;
 	}
 
-	if (!(snap = find_cow(lv)) ||
-	    (lv_info(snap->cow, &info, 0) && !info.exists)) {
+	if (!(snap_seg = find_cow(lv)) ||
+	    (lv_info(snap_seg->cow, &info, 0) && !info.exists)) {
 		field->report_string = "";
 		*sortval = UINT64_C(0);
 		field->sort_value = sortval;
 		return 1;
 	}
 
-	if (!lv_snapshot_percent(snap->cow, &snap_percent) || snap_percent < 0) {
+	if (!lv_snapshot_percent(snap_seg->cow, &snap_percent)
+	    || snap_percent < 0) {
 		field->report_string = "100.00";
 		*sortval = UINT64_C(100);
 		field->sort_value = sortval;
@@ -855,9 +857,9 @@ static int _copypercent_disp(struct report_handle *rh, struct field *field,
 
 static struct {
 	report_type_t type;
-	const char id[30];
+	const char id[32];
 	off_t offset;
-	const char heading[30];
+	const char heading[32];
 	int width;
 	uint32_t flags;
 	field_report_fn report_fn;
@@ -1077,6 +1079,9 @@ void *report_init(struct cmd_context *cmd, const char *format, const char *keys,
 	case SEGS:
 		rh->field_prefix = "seg_";
 		break;
+	case PVSEGS:
+		rh->field_prefix = "pvseg_";
+		break;
 	default:
 		rh->field_prefix = "";
 	}
@@ -1096,6 +1101,8 @@ void *report_init(struct cmd_context *cmd, const char *format, const char *keys,
 	/* Ensure options selected are compatible */
 	if (rh->type & SEGS)
 		rh->type |= LVS;
+	if (rh->type & PVSEGS)
+		rh->type |= PVS;
 	if ((rh->type & LVS) && (rh->type & PVS)) {
 		log_error("Can't report LV and PV fields at the same time");
 		return NULL;
@@ -1106,6 +1113,8 @@ void *report_init(struct cmd_context *cmd, const char *format, const char *keys,
 		*report_type = SEGS;
 	else if (rh->type & LVS)
 		*report_type = LVS;
+	else if (rh->type & PVSEGS)
+		*report_type = PVSEGS;
 	else if (rh->type & PVS)
 		*report_type = PVS;
 
@@ -1126,7 +1135,7 @@ void report_free(void *handle)
  */
 int report_object(void *handle, struct volume_group *vg,
 		  struct logical_volume *lv, struct physical_volume *pv,
-		  struct lv_segment *seg)
+		  struct lv_segment *seg, struct pv_segment *pvseg)
 {
 	struct report_handle *rh = handle;
 	struct list *fh;
@@ -1186,6 +1195,9 @@ int report_object(void *handle, struct volume_group *vg,
 			break;
 		case SEGS:
 			data = (void *) seg + _fields[fp->field_num].offset;
+			break;
+		case PVSEGS:
+			data = (void *) pvseg + _fields[fp->field_num].offset;
 		}
 
 		if (skip) {

lib/report/report.h

@@ -18,7 +18,7 @@
 
 #include "metadata.h"
 
-typedef enum { LVS = 1, PVS = 2, VGS = 4, SEGS = 8 } report_type_t;
+typedef enum { LVS = 1, PVS = 2, VGS = 4, SEGS = 8, PVSEGS = 16 } report_type_t;
 
 struct field;
 struct report_handle;
@@ -32,7 +32,7 @@ void *report_init(struct cmd_context *cmd, const char *format, const char *keys,
 void report_free(void *handle);
 int report_object(void *handle, struct volume_group *vg,
 		  struct logical_volume *lv, struct physical_volume *pv,
-		  struct lv_segment *seg);
+		  struct lv_segment *seg, struct pv_segment *pvseg);
 int report_output(void *handle);
 
 #endif

lib/snapshot/snapshot.c

@@ -70,8 +70,8 @@ static int _text_import(struct lv_segment *seg, const struct config_node *sn,
 		return 0;
 	}
 
-	if (!vg_add_snapshot(org, cow, 1, &seg->lv->lvid.id[1], seg->len,
-			     chunk_size)) {
+	if (!vg_add_snapshot(seg->lv->vg->fid, seg->lv->name, org, cow,
+			     &seg->lv->lvid, seg->len, chunk_size)) {
 		stack;
 		return 0;
 	}

lib/uuid/uuid.h

@@ -25,11 +25,11 @@ struct id {
 
 /*
  * Unique logical volume identifier
- * With format1 this is VG uuid + LV uuid + '\0'
+ * With format1 this is VG uuid + LV uuid + '\0' + padding
  */
 union lvid {
 	struct id id[2];
-	char s[2 * sizeof(struct id) + 1];
+	char s[2 * sizeof(struct id) + 1 + 7];
 };
 
 int lvid_from_lvnum(union lvid *lvid, struct id *vgid, uint32_t lv_num);

libdm/Makefile.in

@@ -17,6 +17,14 @@ top_srcdir = @top_srcdir@
 VPATH = @srcdir@
 interface = @interface@
 
+ifeq ("@DMEVENTD@", "yes")
+   SUBDIRS += event
+endif
+
+ifeq ($(MAKECMDGOALS),distclean)
+  SUBDIRS += event
+endif
+
 SOURCES = libdm-common.c libdm-file.c $(interface)/libdm-iface.c
 
 INCLUDES = -I$(interface)

libdm/datastruct/list.h

@@ -18,17 +18,30 @@
 
 #include <assert.h>
 
+/*
+ * A list consists of a list head plus elements.
+ * Each element has 'next' and 'previous' pointers.
+ * The list head's pointers point to the first and the last element.
+ */
+
 struct list {
 	struct list *n, *p;
 };
 
+/*
+ * Initialise a list before use.
+ * The list head's next and previous pointers point back to itself.
+ */
 #define LIST_INIT(name)	struct list name = { &(name), &(name) }
-
 static inline void list_init(struct list *head)
 {
 	head->n = head->p = head;
 }
 
+/*
+ * Insert an element before 'head'.
+ * If 'head' is the list head, this adds an element to the end of the list.
+ */
 static inline void list_add(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -40,6 +53,10 @@ static inline void list_add(struct list *head, struct list *elem)
 	head->p = elem;
 }
 
+/*
+ * Insert an element after 'head'.
+ * If 'head' is the list head, this adds an element to the front of the list.
+ */
 static inline void list_add_h(struct list *head, struct list *elem)
 {
 	assert(head->n);
@@ -51,53 +68,127 @@ static inline void list_add_h(struct list *head, struct list *elem)
 	head->n = elem;
 }
 
+/*
+ * Delete an element from its list.
+ * Note that this doesn't change the element itself - it may still be safe
+ * to follow its pointers.
+ */
 static inline void list_del(struct list *elem)
 {
 	elem->n->p = elem->p;
 	elem->p->n = elem->n;
 }
 
+/*
+ * Is the list empty?
+ */
 static inline int list_empty(struct list *head)
 {
 	return head->n == head;
 }
 
+/*
+ * Is this the first element of the list?
+ */
+static inline int list_start(struct list *head, struct list *elem)
+{
+	return elem->p == head;
+}
+
+/*
+ * Is this the last element of the list?
+ */
 static inline int list_end(struct list *head, struct list *elem)
 {
 	return elem->n == head;
 }
 
+/*
+ * Return the previous element of the list, or NULL if we've reached the start.
+ */
+static inline struct list *list_prev(struct list *head, struct list *elem)
+{
+	return (list_start(head, elem) ? NULL : elem->p);
+}
+
+/*
+ * Return the next element of the list, or NULL if we've reached the end.
+ */
 static inline struct list *list_next(struct list *head, struct list *elem)
 {
 	return (list_end(head, elem) ? NULL : elem->n);
 }
 
-#define list_item(v, t) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->list))
+/*
+ * Given the address v of an instance of 'struct list' called 'head' 
+ * contained in a structure of type t, return the containing structure.
+ */
+#define list_struct_base(v, t, head) \
+    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->head))
 
-#define list_struct_base(v, t, h) \
-    ((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->h))
+/*
+ * Given the address v of an instance of 'struct list list' contained in
+ * a structure of type t, return the containing structure.
+ */
+#define list_item(v, t) list_struct_base((v), t, list)
 
-/* Given a known element in a known structure, locate another */
+/*
+ * Given the address v of one known element e in a known structure of type t,
+ * return another element f.
+ */
 #define struct_field(v, t, e, f) \
     (((t *)((uintptr_t)(v) - (uintptr_t)&((t *) 0)->e))->f)
 
-/* Given a known element in a known structure, locate the list head */
+/*
+ * Given the address v of a known element e in a known structure of type t,
+ * return the list head 'list'
+ */
 #define list_head(v, t, e) struct_field(v, t, e, list)
 
+/*
+ * Set v to each element of a list in turn.
+ */
 #define list_iterate(v, head) \
 	for (v = (head)->n; v != head; v = v->n)
 
+/*
+ * Set v to each element in a list in turn, starting from the element 
+ * in front of 'start'.
+ * You can use this to 'unwind' a list_iterate and back out actions on
+ * already-processed elements.
+ * If 'start' is 'head' it walks the list backwards.
+ */
 #define list_uniterate(v, head, start) \
 	for (v = (start)->p; v != head; v = v->p)
 
+/*
+ * A safe way to walk a list and delete and free some elements along
+ * the way.
+ * t must be defined as a temporary variable of the same type as v.
+ */
 #define list_iterate_safe(v, t, head) \
 	for (v = (head)->n, t = v->n; v != head; v = t, t = v->n)
 
-#define list_iterate_items(v, head) \
-	for (v = list_item((head)->n, typeof(*v)); &v->list != (head); \
-	     v = list_item(v->list.n, typeof(*v)))
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The 'struct list' variable within the containing structure is 'field'.
+ */
+#define list_iterate_items_gen(v, head, field) \
+	for (v = list_struct_base((head)->n, typeof(*v), field); \
+	     &v->field != (head); \
+	     v = list_struct_base(v->field.n, typeof(*v), field))
 
+/*
+ * Walk a list, setting 'v' in turn to the containing structure of each item.
+ * The containing structure should be the same type as 'v'.
+ * The list should be 'struct list list' within the containing structure.
+ */
+#define list_iterate_items(v, head) list_iterate_items_gen(v, (head), list)
+
+/*
+ * Return the number of elements in a list by walking it.
+ */
 static inline unsigned int list_size(const struct list *head)
 {
 	unsigned int s = 0;

libdm/libdm-event.h

@@ -0,0 +1,83 @@
+/*
+ * Copyright (C) 2005 Red Hat, Inc. All rights reserved.
+ *
+ * This file is part of the device-mapper userspace tools.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU Lesser General Public License v.2.1.
+ *
+ * You should have received a copy of the GNU Lesser General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+
+#ifndef LIB_DMEVENT_H
+#define LIB_DMEVENT_H
+
+#include "list.h"
+
+#include <stdint.h>
+
+#define	DAEMON		"/sbin/dmeventd"
+#define LOCKFILE	"/var/lock/dmeventd"
+#define	FIFO_CLIENT	"/var/run/dmeventd-client"
+#define	FIFO_SERVER	"/var/run/dmeventd-server"
+#define PIDFILE		"/var/run/dmeventd.pid"
+
+/* Commands for the daemon passed in the message below. */
+enum dmeventd_command {
+	CMD_ACTIVE = 1,
+	CMD_REGISTER_FOR_EVENT,
+	CMD_UNREGISTER_FOR_EVENT,
+	CMD_GET_NEXT_REGISTERED_DEVICE,
+};
+
+/* Message passed between client and daemon. */
+struct daemon_message {
+	union {
+		unsigned int cmd;
+		int	 status;
+	} opcode;
+	char msg[252];
+} __attribute__((packed));
+
+/* Fifos for client/daemon communication. */
+struct fifos
+{
+	int client;
+	int server;
+	char *client_path;
+	char *server_path;
+};
+
+/* Event type definitions. */
+enum event_type {
+	SINGLE		= 0x01, /* Report multiple errors just once. */
+	MULTI		= 0x02, /* Report all of them. */
+	SECTOR_ERROR	= 0x04, /* Failure on a particular sector. */
+	DEVICE_ERROR	= 0x08, /* Device failure. */
+	PATH_ERROR	= 0x10, /* Failure on an io path. */
+	ADAPTOR_ERROR	= 0x20, /* Failure off a host adaptor. */
+	SYNC_STATUS	= 0x40, /* Mirror synchronization completed/failed. */
+};
+#define	ALL_ERRORS (SECTOR_ERROR | DEVICE_ERROR | PATH_ERROR | ADAPTOR_ERROR)
+
+int dm_register_for_event(char *dso_name, char *device, enum event_type events);
+int dm_unregister_for_event(char *dso_name, char *device,
+			    enum event_type events);
+int dm_get_next_registered_device(char **dso_name, char **device,
+			    enum event_type *events);
+
+#endif
+
+/*
+ * Overrides for Emacs so that we follow Linus's tabbing style.
+ * Emacs will notice this stuff at the end of the file and automatically
+ * adjust the settings for this buffer only.  This must remain at the end
+ * of the file.
+ * ---------------------------------------------------------------------------
+ * Local variables:
+ * c-file-style: "linux"
+ * End:
+ */

tools/commands.h

@@ -491,6 +491,7 @@ xx(pvs,
    "\t[--nosuffix]\n"
    "\t[-o|--options [+]Field[,Field]]\n"
    "\t[-O|--sort [+|-]key1[,[+|-]key2[,...]]]\n"
+   "\t[--segments]\n"
    "\t[--separator Separator]\n"
    "\t[--unbuffered]\n"
    "\t[--units hsbkmgtHKMGT]\n"
@@ -499,8 +500,8 @@ xx(pvs,
    "\t[PhysicalVolume [PhysicalVolume...]]\n",
    
    aligned_ARG, all_ARG, ignorelockingfailure_ARG, noheadings_ARG,
-   nolocking_ARG, nosuffix_ARG, options_ARG, separator_ARG, sort_ARG,
-   unbuffered_ARG, units_ARG)
+   nolocking_ARG, nosuffix_ARG, options_ARG, segments_ARG, separator_ARG,
+   sort_ARG, unbuffered_ARG, units_ARG)
 
 xx(pvscan,
    "List all physical volumes",
@@ -569,13 +570,15 @@ xx(vgchange,
    "\t -c|--clustered {y|n} |" "\n"
    "\t -x|--resizeable {y|n} |" "\n"
    "\t -l|--logicalvolume MaxLogicalVolumes |" "\n"
+   "\t -s|--physicalextentsize PhysicalExtentSize[kKmMgGtT] |" "\n"
    "\t --addtag Tag |\n"
    "\t --deltag Tag}\n"
    "\t[VolumeGroupName...]\n",
 
    addtag_ARG, alloc_ARG, allocation_ARG, autobackup_ARG, available_ARG,
    clustered_ARG, deltag_ARG, ignorelockingfailure_ARG, logicalvolume_ARG,
-   partial_ARG, resizeable_ARG, resizable_ARG, test_ARG, uuid_ARG)
+   partial_ARG, physicalextentsize_ARG, resizeable_ARG, resizable_ARG,
+   test_ARG, uuid_ARG)
 
 xx(vgck,
    "Check the consistency of volume group(s)",
@@ -717,9 +720,12 @@ xx(vgmknodes,
    "vgmknodes\n"
    "\t[-d|--debug]\n"
    "\t[-h|--help]\n"
+   "\t[--ignorelockingfailure]\n"
    "\t[-v|--verbose]\n"
    "\t[--version]" "\n"
-   "\t[VolumeGroupName...]\n" )
+   "\t[VolumeGroupName...]\n",
+
+   ignorelockingfailure_ARG)
 
 xx(vgreduce,
    "Remove physical volume(s) from a volume group",

tools/lvcreate.c

@@ -512,8 +512,8 @@ static int _lvcreate(struct cmd_context *cmd, struct lvcreate_params *lp)
 		return 0;
 	}
 
-	if (!(lv = lv_create_empty(vg->fid, lp->lv_name, "lvol%d",
-				   status, lp->alloc, vg))) {
+	if (!(lv = lv_create_empty(vg->fid, lp->lv_name, "lvol%d", NULL,
+				   status, lp->alloc, 0, vg))) {
 		stack;
 		return 0;
 	}
@@ -609,8 +609,8 @@ static int _lvcreate(struct cmd_context *cmd, struct lvcreate_params *lp)
 			return 0;
 		}
 
-		if (!vg_add_snapshot(org, lv, 1, NULL, org->le_count,
-				     lp->chunk_size)) {
+		if (!vg_add_snapshot(vg->fid, NULL, org, lv, NULL,
+				     org->le_count, lp->chunk_size)) {
 			log_err("Couldn't create snapshot.");
 			return 0;
 		}

tools/lvmcmdline.c

@@ -54,7 +54,7 @@ extern char *optarg;
  */
 struct arg the_args[ARG_COUNT + 1] = {
 
-#define arg(a, b, c, d) {b, "--" c, d, 0, NULL, 0, 0, INT64_C(0), UINT64_C(0), 0, NULL},
+#define arg(a, b, c, d) {b, "", "--" c, d, 0, NULL, 0, 0, INT64_C(0), UINT64_C(0), 0, NULL},
 #include "args.h"
 #undef arg
 

tools/lvresize.c

@@ -115,7 +115,7 @@ static int _lvresize(struct cmd_context *cmd, struct lvresize_params *lp)
 {
 	struct volume_group *vg;
 	struct logical_volume *lv;
-	struct snapshot *snap;
+	struct lv_segment *snap_seg;
 	struct lvinfo info;
 	uint32_t stripesize_extents = 0;
 	uint32_t seg_stripes = 0, seg_stripesize = 0, seg_size = 0;
@@ -180,11 +180,6 @@ static int _lvresize(struct cmd_context *cmd, struct lvresize_params *lp)
 		return ECMD_FAILED;
 	}
 
-	if (lv_is_origin(lv)) {
-		log_error("Snapshot origin volumes cannot be resized yet.");
-		return ECMD_FAILED;
-	}
-
 	alloc = (alloc_policy_t) arg_uint_value(cmd, alloc_ARG, lv->alloc);
 
 	if (lp->size) {
@@ -352,6 +347,23 @@ static int _lvresize(struct cmd_context *cmd, struct lvresize_params *lp)
 			lp->resize = LV_EXTEND;
 	}
 
+
+	if (lv_is_origin(lv)) {
+		if (lp->resize == LV_REDUCE) {
+			log_error("Snapshot origin volumes cannot be reduced "
+				  "in size yet.");
+			return ECMD_FAILED;
+		}
+
+		memset(&info, 0, sizeof(info));
+
+		if (lv_info(lv, &info, 0) && info.exists) {
+			log_error("Snapshot origin volumes can be resized "
+				  "only while inactive: try lvchange -an");
+			return ECMD_FAILED;
+		}
+	}
+
 	if (lp->resize == LV_REDUCE) {
 		if (lp->argc)
 			log_print("Ignoring PVs on command line when reducing");
@@ -460,8 +472,8 @@ static int _lvresize(struct cmd_context *cmd, struct lvresize_params *lp)
 	backup(vg);
 
 	/* If snapshot, must suspend all associated devices */
-	if ((snap = find_cow(lv)))
-		lock_lvid = snap->origin->lvid.s;
+	if ((snap_seg = find_cow(lv)))
+		lock_lvid = snap_seg->origin->lvid.s;
 	else
 		lock_lvid = lv->lvid.s;
 

tools/polldaemon.c

@@ -72,6 +72,8 @@ static int _check_mirror_status(struct cmd_context *cmd,
 	float segment_percent = 0.0, overall_percent = 0.0;
 	uint32_t event_nr = 0;
 
+void *x;
+
 	/* By default, caller should not retry */
 	*finished = 1;
 
@@ -97,6 +99,9 @@ static int _check_mirror_status(struct cmd_context *cmd,
 	else
 		log_verbose("%s: Moved: %.1f%%", name, overall_percent);
 
+x = pool_alloc(cmd->mem, 1);
+pool_free(cmd->mem, x);
+
 	if (segment_percent < 100.0) {
 		/* The only case the caller *should* try again later */
 		*finished = 0;

tools/pvmove.c

@@ -141,9 +141,9 @@ static struct logical_volume *_set_up_pvmove_lv(struct cmd_context *cmd,
 
 	/* FIXME Cope with non-contiguous => splitting existing segments */
 	/* FIXME Pass 'alloc' down to lv_extend */
-	if (!(lv_mirr = lv_create_empty(vg->fid, NULL, "pvmove%d",
+	if (!(lv_mirr = lv_create_empty(vg->fid, NULL, "pvmove%d", NULL,
 					LVM_READ | LVM_WRITE,
-					ALLOC_CONTIGUOUS, vg))) {
+					ALLOC_CONTIGUOUS, 0, vg))) {
 		log_error("Creation of temporary pvmove LV failed");
 		return NULL;
 	}
@@ -160,7 +160,8 @@ static struct logical_volume *_set_up_pvmove_lv(struct cmd_context *cmd,
 	/* Find segments to be moved and set up mirrors */
 	list_iterate_items(lvl, &vg->lvs) {
 		lv = lvl->lv;
-		if ((lv == lv_mirr) || (lv_name && strcmp(lv->name, lv_name)))
+		if ((lv == lv_mirr) ||
+		    (lv_name && strcmp(lv->name, lv_name)))
 			continue;
 		if (lv_is_origin(lv) || lv_is_cow(lv)) {
 			log_print("Skipping snapshot-related LV %s", lv->name);

tools/reporter.c

@@ -24,7 +24,7 @@ static int _vgs_single(struct cmd_context *cmd, const char *vg_name,
 		return ECMD_FAILED;
 	}
 
-	if (!report_object(handle, vg, NULL, NULL, NULL))
+	if (!report_object(handle, vg, NULL, NULL, NULL, NULL))
 		return ECMD_FAILED;
 
 	return ECMD_PROCESSED;
@@ -33,7 +33,7 @@ static int _vgs_single(struct cmd_context *cmd, const char *vg_name,
 static int _lvs_single(struct cmd_context *cmd, struct logical_volume *lv,
 		       void *handle)
 {
-	if (!report_object(handle, lv->vg, lv, NULL, NULL))
+	if (!report_object(handle, lv->vg, lv, NULL, NULL, NULL))
 		return ECMD_FAILED;
 
 	return ECMD_PROCESSED;
@@ -42,18 +42,50 @@ static int _lvs_single(struct cmd_context *cmd, struct logical_volume *lv,
 static int _segs_single(struct cmd_context *cmd, struct lv_segment *seg,
 			void *handle)
 {
-	if (!report_object(handle, seg->lv->vg, seg->lv, NULL, seg))
+	if (!report_object(handle, seg->lv->vg, seg->lv, NULL, seg, NULL))
 		return ECMD_FAILED;
 
 	return ECMD_PROCESSED;
 }
 
+static int _pvsegs_sub_single(struct cmd_context *cmd, struct volume_group *vg,
+			      struct pv_segment *pvseg, void *handle)
+{
+	int consistent = 0;
+	struct physical_volume *pv = pvseg->pv;
+	int ret = ECMD_PROCESSED;
+
+	if (!lock_vol(cmd, pv->vg_name, LCK_VG_READ)) {
+		log_error("Can't lock %s: skipping", pv->vg_name);
+		return ECMD_FAILED;
+	}
+
+	if (!(vg = vg_read(cmd, pv->vg_name, &consistent))) {
+		log_error("Can't read %s: skipping", pv->vg_name);
+		unlock_vg(cmd, pv->vg_name);
+		return ECMD_FAILED;
+	}
+
+	if (!report_object(handle, vg, NULL, pv, NULL, pvseg))
+		ret = ECMD_FAILED;
+
+	unlock_vg(cmd, pv->vg_name);
+	return ret;
+}
+
 static int _lvsegs_single(struct cmd_context *cmd, struct logical_volume *lv,
 			  void *handle)
 {
 	return process_each_segment_in_lv(cmd, lv, handle, _segs_single);
 }
 
+static int _pvsegs_single(struct cmd_context *cmd, struct volume_group *vg,
+			  struct physical_volume *pv, void *handle)
+{
+	return process_each_segment_in_pv(cmd, vg, pv, handle,
+					  _pvsegs_sub_single);
+}
+
 static int _pvs_single(struct cmd_context *cmd, struct volume_group *vg,
 		       struct physical_volume *pv, void *handle)
 {
@@ -73,7 +105,7 @@ static int _pvs_single(struct cmd_context *cmd, struct volume_group *vg,
 		}
 	}
 
-	if (!report_object(handle, vg, NULL, pv, NULL))
+	if (!report_object(handle, vg, NULL, pv, NULL, NULL))
 		ret = ECMD_FAILED;
 
 	if (pv->vg_name)
@@ -151,6 +183,18 @@ static int _report(struct cmd_context *cmd, int argc, char **argv,
 						  "report/segs_cols_verbose",
 						  DEFAULT_SEGS_COLS_VERB);
 		break;
+	case PVSEGS:
+		keys = find_config_str(cmd->cft->root, "report/pvsegs_sort",
+				       DEFAULT_PVSEGS_SORT);
+		if (!arg_count(cmd, verbose_ARG))
+			options = find_config_str(cmd->cft->root,
+						  "report/pvsegs_cols",
+						  DEFAULT_PVSEGS_COLS);
+		else
+			options = find_config_str(cmd->cft->root,
+						  "report/pvsegs_cols_verbose",
+						  DEFAULT_PVSEGS_COLS_VERB);
+		break;
 	}
 
 	/* If -o supplied use it, else use default for report_type */
@@ -208,6 +252,10 @@ static int _report(struct cmd_context *cmd, int argc, char **argv,
 		r = process_each_lv(cmd, argc, argv, LCK_VG_READ, report_handle,
 				    &_lvsegs_single);
 		break;
+	case PVSEGS:
+		r = process_each_pv(cmd, argc, argv, NULL, report_handle,
+				    &_pvsegs_single);
+		break;
 	}
 
 	report_output(report_handle);
@@ -235,5 +283,12 @@ int vgs(struct cmd_context *cmd, int argc, char **argv)
 
 int pvs(struct cmd_context *cmd, int argc, char **argv)
 {
-	return _report(cmd, argc, argv, PVS);
+	report_type_t type;
+
+	if (arg_count(cmd, segments_ARG))
+		type = PVSEGS;
+	else
+		type = PVS;
+
+	return _report(cmd, argc, argv, type);
 }

tools/toollib.c

@@ -60,6 +60,9 @@ int process_each_lv_in_vg(struct cmd_context *cmd, struct volume_group *vg,
 	}
 
 	list_iterate_items(lvl, &vg->lvs) {
+		if (lvl->lv->status & SNAPSHOT)
+			continue;
+
 		/* Should we process this LV? */
 		if (process_all)
 			process_lv = 1;
@@ -285,6 +288,28 @@ int process_each_lv(struct cmd_context *cmd, int argc, char **argv,
 	return ret_max;
 }
 
+int process_each_segment_in_pv(struct cmd_context *cmd,
+			       struct volume_group *vg,
+			       struct physical_volume *pv,
+			       void *handle,
+			       int (*process_single) (struct cmd_context * cmd,
+						      struct volume_group * vg,
+						      struct pv_segment * pvseg,
+						      void *handle))
+{
+	struct pv_segment *pvseg;
+	int ret_max = 0;
+	int ret;
+
+	list_iterate_items(pvseg, &pv->segments) {
+		ret = process_single(cmd, vg, pvseg, handle);
+		if (ret > ret_max)
+			ret_max = ret;
+	}
+
+	return ret_max;
+}
+
 int process_each_segment_in_lv(struct cmd_context *cmd,
 			       struct logical_volume *lv,
 			       void *handle,
@@ -472,6 +497,8 @@ static int _process_all_devs(struct cmd_context *cmd, void *handle,
 		if (!(pv = pv_read(cmd, dev_name(dev), NULL, NULL, 0))) {
 			memset(&pv_dummy, 0, sizeof(pv_dummy));
 			list_init(&pv_dummy.tags);
+			list_init(&pv_dummy.segments);
+			list_init(&pv_dummy.free_segments);
 			pv_dummy.dev = dev;
 			pv_dummy.fmt = NULL;
 			pv = &pv_dummy;

tools/toollib.h

@@ -40,6 +40,14 @@ int process_each_pv(struct cmd_context *cmd, int argc, char **argv,
 					   struct volume_group * vg,
 					   struct physical_volume * pv,
 					   void *handle));
+int process_each_segment_in_pv(struct cmd_context *cmd,
+			       struct volume_group *vg,
+			       struct physical_volume *pv,
+			       void *handle,
+			       int (*process_single) (struct cmd_context * cmd,
+						      struct volume_group * vg,
+						      struct pv_segment * pvseg,
+						      void *handle));
 
 int process_each_lv(struct cmd_context *cmd, int argc, char **argv,
 		    int lock_type, void *handle,

tools/tools.h

@@ -89,7 +89,9 @@ enum {
 /* a global table of possible arguments */
 struct arg {
 	const char short_arg;
+	char _padding[7];
 	const char *long_arg;
+
 	int (*fn) (struct cmd_context * cmd, struct arg * a);
 
 	unsigned int count;

tools/vgchange.c

@@ -27,7 +27,7 @@ static int _activate_lvs_in_vg(struct cmd_context *cmd,
 		lv = lvl->lv;
 
 		/* Only request activation of snapshot origin devices */
-		if (lv_is_cow(lv))
+		if ((lv->status & SNAPSHOT) || lv_is_cow(lv))
 			continue;
 
 		/* Can't deactive a pvmove LV */
@@ -251,6 +251,64 @@ static int _vgchange_logicalvolume(struct cmd_context *cmd,
 	return ECMD_PROCESSED;
 }
 
+static int _vgchange_pesize(struct cmd_context *cmd, struct volume_group *vg)
+{
+	uint32_t extent_size;
+
+	if (!(vg->status & RESIZEABLE_VG)) {
+		log_error("Volume group \"%s\" must be resizeable "
+			  "to change PE size", vg->name);
+		return ECMD_FAILED;
+	}
+
+	if (arg_sign_value(cmd, physicalextentsize_ARG, 0) == SIGN_MINUS) {
+		log_error("Physical extent size may not be negative");
+		return EINVALID_CMD_LINE;
+	}
+
+	extent_size = arg_uint_value(cmd, physicalextentsize_ARG, 0) * 2;
+	if (!extent_size) {
+		log_error("Physical extent size may not be zero");
+		return EINVALID_CMD_LINE;
+	}
+
+	if (extent_size == vg->extent_size) {
+		log_error("Physical extent size of VG %s is already %s",
+			  vg->name, display_size(cmd, extent_size, SIZE_SHORT));
+		return ECMD_PROCESSED;
+	}
+
+	if (extent_size & (extent_size - 1)) {
+		log_error("Physical extent size must be a power of 2.");
+		return EINVALID_CMD_LINE;
+	}
+
+	if (extent_size > vg->extent_size) {
+		if ((uint64_t) vg->extent_size * vg->extent_count % extent_size) {
+			/* FIXME Adjust used PV sizes instead */
+			log_error("New extent size is not a perfect fit");
+			return EINVALID_CMD_LINE;
+		}
+	}
+
+	if (!archive(vg))
+		return ECMD_FAILED;
+
+	if (!vg_change_pesize(cmd, vg, extent_size)) {
+		stack;
+		return ECMD_FAILED;
+	}
+
+	if (!vg_write(vg) || !vg_commit(vg))
+		return ECMD_FAILED;
+
+	backup(vg);
+
+	log_print("Volume group \"%s\" successfully changed", vg->name);
+
+	return ECMD_PROCESSED;
+}
+
 static int _vgchange_tag(struct cmd_context *cmd, struct volume_group *vg,
 			 int arg)
 {
@@ -369,6 +427,9 @@ static int vgchange_single(struct cmd_context *cmd, const char *vg_name,
 	else if (arg_count(cmd, deltag_ARG))
 		r = _vgchange_tag(cmd, vg, deltag_ARG);
 
+	else if (arg_count(cmd, physicalextentsize_ARG))
+		r = _vgchange_pesize(cmd, vg);
+
 	else if (arg_count(cmd, uuid_ARG))
 		r = _vgchange_uuid(cmd, vg);
 
@@ -387,9 +448,10 @@ int vgchange(struct cmd_context *cmd, int argc, char **argv)
 	    (arg_count(cmd, available_ARG) + arg_count(cmd, logicalvolume_ARG) +
 	     arg_count(cmd, resizeable_ARG) + arg_count(cmd, deltag_ARG) +
 	     arg_count(cmd, addtag_ARG) + arg_count(cmd, uuid_ARG) +
+	     arg_count(cmd, physicalextentsize_ARG) +
 	     arg_count(cmd, clustered_ARG) + arg_count(cmd, alloc_ARG))) {
-		log_error("One of -a, -c, -l, -x, --alloc, --addtag, --deltag "
-			  "or --uuid required");
+		log_error("One of -a, -c, -l, -s, -x, --uuid, --alloc, --addtag or "
+			  "--deltag required");
 		return EINVALID_CMD_LINE;
 	}
 
@@ -397,8 +459,9 @@ int vgchange(struct cmd_context *cmd, int argc, char **argv)
 	if (arg_count(cmd, available_ARG) + arg_count(cmd, logicalvolume_ARG) +
 	    arg_count(cmd, resizeable_ARG) + arg_count(cmd, deltag_ARG) +
 	    arg_count(cmd, addtag_ARG) + arg_count(cmd, alloc_ARG) +
-	    arg_count(cmd, uuid_ARG) + arg_count(cmd, clustered_ARG) > 1) {
-		log_error("Only one of -a, -c, -l, -x, --uuid, --alloc, "
+	    arg_count(cmd, uuid_ARG) + arg_count(cmd, clustered_ARG) +
+	    arg_count(cmd, physicalextentsize_ARG) > 1) {
+		log_error("Only one of -a, -c, -l, -s, -x, --uuid, --alloc, "
 			  "--addtag or --deltag allowed");
 		return EINVALID_CMD_LINE;
 	}

tools/vgconvert.c

@@ -92,6 +92,8 @@ static int vgconvert_single(struct cmd_context *cmd, const char *vg_name,
 	if (cmd->fmt->features & FMT_RESTRICTED_LVIDS) {
 		list_iterate_items(lvl, &vg->lvs) {
 			lv = lvl->lv;
+			if (lv->status & SNAPSHOT)
+				continue;
 			if (lvnum_from_lvid(&lv->lvid) < MAX_RESTRICTED_LVS)
 				continue;
 			if (lv_info(lv, &info, 0) && info.exists) {

tools/vgreduce.c

@@ -49,9 +49,9 @@ static int _remove_pv(struct volume_group *vg, struct pv_list *pvl)
 static int _remove_lv(struct cmd_context *cmd, struct logical_volume *lv,
 		      int *list_unsafe)
 {
-	struct snapshot *snap;
-	struct snapshot_list *snl;
-	struct list *snaplist;
+	struct lv_segment *snap_seg;
+	struct list *snh, *snht;
+	struct logical_volume *cow;
 
 	log_verbose("%s/%s has missing extents: removing (including "
 		    "dependencies)", lv->vg->name, lv->name);
@@ -65,36 +65,34 @@ static int _remove_lv(struct cmd_context *cmd, struct logical_volume *lv,
 			log_error("Failed to deactivate LV %s", lv->name);
 			return 0;
 		}
-	} else if ((snap = find_cow(lv))) {
+	} else if ((snap_seg = find_cow(lv))) {
 		log_verbose("Deactivating (if active) logical volume %s "
-			    "(origin of %s)", snap->origin->name, lv->name);
+			    "(origin of %s)", snap_seg->origin->name, lv->name);
 
-		if (!deactivate_lv(cmd, snap->origin->lvid.s)) {
+		if (!deactivate_lv(cmd, snap_seg->origin->lvid.s)) {
 			log_error("Failed to deactivate LV %s",
-				  snap->origin->name);
+				  snap_seg->origin->name);
 			return 0;
 		}
 
 		/* Use the origin LV */
-		lv = snap->origin;
+		lv = snap_seg->origin;
 	}
 
 	/* Remove snapshot dependencies */
-	if (!(snaplist = find_snapshots(lv))) {
-		stack;
-		return 0;
-	}
-	/* List may be empty */
-	list_iterate_items(snl, snaplist) {
+	list_iterate_safe(snh, snht, &lv->snapshot_segs) {
+		snap_seg = list_struct_base(snh, struct lv_segment,
+					    origin_list);
+		cow = snap_seg->cow;
+
 		*list_unsafe = 1;	/* May remove caller's lvht! */
-		snap = snl->snapshot;
-		if (!vg_remove_snapshot(lv->vg, snap->cow)) {
+		if (!vg_remove_snapshot(lv->vg, cow)) {
 			stack;
 			return 0;
 		}
-		log_verbose("Removing LV %s from VG %s", snap->cow->name,
+		log_verbose("Removing LV %s from VG %s", cow->name,
 			    lv->vg->name);
-		if (!lv_remove(lv->vg, snap->cow)) {
+		if (!lv_remove(lv->vg, cow)) {
 			stack;
 			return 0;
 		}

tools/vgsplit.c

@@ -44,6 +44,19 @@ static int _move_pv(struct volume_group *vg_from, struct volume_group *vg_to,
 	return 1;
 }
 
+/* FIXME Why not (lv->vg == vg) ? */
+static int _lv_is_in_vg(struct volume_group *vg, struct logical_volume *lv)
+{
+	struct lv_list *lvl;
+
+	list_iterate_items(lvl, &vg->lvs)
+		if (lv == lvl->lv)
+			 return 1;
+
+	return 0;
+}
+
+
 static int _move_lvs(struct volume_group *vg_from, struct volume_group *vg_to)
 {
 	struct list *lvh, *lvht;
@@ -56,6 +69,9 @@ static int _move_lvs(struct volume_group *vg_from, struct volume_group *vg_to)
 	list_iterate_safe(lvh, lvht, &vg_from->lvs) {
 		lv = list_item(lvh, struct lv_list)->lv;
 
+		if ((lv->status & SNAPSHOT))
+			continue;
+
 		/* Ensure all the PVs used by this LV remain in the same */
 		/* VG as each other */
 		vg_with = NULL;
@@ -89,8 +105,9 @@ static int _move_lvs(struct volume_group *vg_from, struct volume_group *vg_to)
 					  dev_name(pv->dev));
 				return 0;
 			}
-		}
 
+		}
+			
 		if (vg_with == vg_from)
 			continue;
 
@@ -107,39 +124,38 @@ static int _move_lvs(struct volume_group *vg_from, struct volume_group *vg_to)
 	return 1;
 }
 
-static int _lv_is_in_vg(struct volume_group *vg, struct logical_volume *lv)
-{
-	struct lv_list *lvl;
-
-	list_iterate_items(lvl, &vg->lvs)
-		if (lv == lvl->lv)
-			 return 1;
-
-	return 0;
-}
-
 static int _move_snapshots(struct volume_group *vg_from,
 			   struct volume_group *vg_to)
 {
-	struct list *slh, *slth;
-	struct snapshot *snap;
-	int cow_from, origin_from;
+	struct list *lvh, *lvht;
+	struct logical_volume *lv;
+	struct lv_segment *seg;
+	int cow_from = 0;
+	int origin_from = 0;
 
-	list_iterate_safe(slh, slth, &vg_from->snapshots) {
-		snap = list_item(slh, struct snapshot_list)->snapshot;
-		cow_from = _lv_is_in_vg(vg_from, snap->cow);
-		origin_from = _lv_is_in_vg(vg_from, snap->origin);
-		if (cow_from && origin_from)
-			return 1;
-		if ((!cow_from && origin_from) || (cow_from && !origin_from)) {
-			log_error("Snapshot %s split", snap->cow->name);
-			return 0;
+	list_iterate_safe(lvh, lvht, &vg_from->lvs) {
+		lv = list_item(lvh, struct lv_list)->lv;
+
+		if (!(lv->status & SNAPSHOT))
+			continue;
+
+		list_iterate_items(seg, &lv->segments) {
+			cow_from = _lv_is_in_vg(vg_from, seg->cow);
+			origin_from = _lv_is_in_vg(vg_from, seg->origin);
 		}
+		if (cow_from && origin_from)
+			continue;
+		if ((!cow_from && origin_from) || (cow_from && !origin_from)) {
+			log_error("Snapshot %s split", seg->cow->name);
+			return 0;
+		}	
+
+		/* Move this snapshot */
+		list_del(lvh);
+		list_add(&vg_to->lvs, lvh);
+
 		vg_from->snapshot_count--;
 		vg_to->snapshot_count++;
-
-		list_del(slh);
-		list_add(&vg_to->snapshots, slh);
 	}
 
 	return 1;