pvchange: add importclone optiondev-dct-vgimportclone-1

This is a native implementation of vgimportclone.

pvchange --importclone PV...

All cloned PVs from the VG must be imported together.

6 files changed, 464 insertions, 4 deletions

diff --git a/lib/cache/lvmcache.c b/lib/cache/lvmcache.c
index 0934b2068..16152e24a 100644
--- a/lib/cache/lvmcache.c
+++ b/lib/cache/lvmcache.c
@@ -88,6 +88,46 @@ static int _vg_global_lock_held = 0;	/* Global lock held when cache wiped? */
 static int _found_duplicate_pvs = 0;	/* If we never see a duplicate PV we can skip checking for them later. */
 static int _suppress_lock_ordering = 0;
 
+static int _importing_clones = 0;
+static DM_LIST_INIT(_import_clone_devs);
+
+int lvmcache_add_import_clone_dev(struct cmd_context *cmd, struct device *dev)
+{
+	struct device_list *devl;
+
+	if (!(devl = dm_pool_alloc(cmd->mem, sizeof(*devl)))) {
+		log_error("device_list element allocation failed");
+		return 0;
+	}
+	devl->dev = dev;
+
+	dm_list_add(&_import_clone_devs, &devl->list);
+	_importing_clones = 1;
+	return 1;
+}
+
+int lvmcache_is_import_clone_dev(struct device *dev)
+{
+	struct device_list *devl;
+
+	dm_list_iterate_items(devl, &_import_clone_devs) {
+		if (devl->dev == dev)
+			return 1;
+	}
+	return 0;
+}
+
+void lvmcache_clear_import_clone_devs(void)
+{
+	dm_list_init(&_import_clone_devs);
+	_importing_clones = 0;
+}
+
+int lvmcache_importing_clones(void)
+{
+	return _importing_clones;
+}
+
 int lvmcache_init(void)
 {
 	/*
diff --git a/lib/cache/lvmcache.h b/lib/cache/lvmcache.h
index 4fb74dbac..404015071 100644
--- a/lib/cache/lvmcache.h
+++ b/lib/cache/lvmcache.h
@@ -213,4 +213,9 @@ int lvmcache_dev_is_unchosen_duplicate(struct device *dev);
 
 void lvmcache_remove_unchosen_duplicate(struct device *dev);
 
+int lvmcache_add_import_clone_dev(struct cmd_context *cmd, struct device *dev);
+int lvmcache_is_import_clone_dev(struct device *dev);
+void lvmcache_clear_import_clone_devs(void);
+int lvmcache_importing_clones(void);
+
 #endif
diff --git a/lib/filters/filter-usable.c b/lib/filters/filter-usable.c
index 4ee2e9df8..d521fa448 100644
--- a/lib/filters/filter-usable.c
+++ b/lib/filters/filter-usable.c
@@ -15,6 +15,7 @@
 #include "lib.h"
 #include "filter.h"
 #include "activate.h" /* device_is_usable */
+#include "lvmcache.h"
 #ifdef UDEV_SYNC_SUPPORT
 #include <libudev.h>
 #include "dev-ext-udev-constants.h"
@@ -158,6 +159,11 @@ static int _passes_usable_filter(struct dev_filter *f, struct device *dev)
 		}
 	}
 
+	if (lvmcache_importing_clones() && !lvmcache_is_import_clone_dev(dev)) {
+		log_debug_devs("%s: Skipping non-import device.", dev_name(dev));
+		r = 0;
+	}
+
 	return r;
 }
 
diff --git a/tools/args.h b/tools/args.h
index ebddd2ec4..156234a03 100644
--- a/tools/args.h
+++ b/tools/args.h
@@ -49,6 +49,7 @@ arg(ignorelockingfailure_ARG, '\0', "ignorelockingfailure", NULL, 0)
 arg(ignoremonitoring_ARG, '\0', "ignoremonitoring", NULL, 0)
 arg(ignoreskippedcluster_ARG, '\0', "ignoreskippedcluster", NULL, 0)
 arg(ignoreunsupported_ARG, '\0', "ignoreunsupported", NULL, 0)
+arg(importclone_ARG, '\0', "importclone", NULL, 0)
 arg(labelsector_ARG, '\0', "labelsector", int_arg, 0)
 arg(lockopt_ARG, '\0', "lockopt", string_arg, 0)
 arg(lockstart_ARG, '\0', "lockstart", NULL, 0)
diff --git a/tools/commands.h b/tools/commands.h
index 7b32835ea..993853a7d 100644
--- a/tools/commands.h
+++ b/tools/commands.h
@@ -739,7 +739,7 @@ xx(pvchange,
 
    all_ARG, allocatable_ARG, allocation_ARG, autobackup_ARG, deltag_ARG,
    addtag_ARG, force_ARG, ignoreskippedcluster_ARG, metadataignore_ARG,
-   select_ARG, test_ARG, uuid_ARG)
+   select_ARG, test_ARG, uuid_ARG, importclone_ARG)
 
 xx(pvresize,
    "Resize physical volume(s)",
diff --git a/tools/pvchange.c b/tools/pvchange.c
index e100b80c5..0d90a50d0 100644
--- a/tools/pvchange.c
+++ b/tools/pvchange.c
@@ -14,12 +14,408 @@
  */
 
 #include "tools.h"
+#include "lvmcache.h"
+#include "lvmetad-client.h"
 
 struct pvchange_params {
 	unsigned done;
 	unsigned total;
+
+	int import_vg;
+	int found_args;
+	struct dm_list arg_import;
+	const char *base_vgname;
+	const char *old_vgname;
+	const char *new_vgname;
+};
+
+struct pvchange_device {
+	struct dm_list list;
+	struct device *dev;
+	unsigned found_in_vg : 1;
 };
 
+static int _pvchange_import_check_pv_single(struct cmd_context *cmd, struct volume_group *vg,
+					    struct physical_volume *pv, struct processing_handle *handle)
+{
+	struct pvchange_params *pp = (struct pvchange_params *) handle->custom_handle;
+	struct pvchange_device *pd;
+
+	pp->total++;
+
+	if (vg && is_orphan_vg(vg->name)) {
+		log_error("Cannot importclone on orphan PV %s.", dev_name(pv->dev));
+		return ECMD_FAILED;
+	}
+
+	if (!(pd = dm_pool_zalloc(cmd->mem, sizeof(*pd)))) {
+		log_error("alloc failed.");
+		return ECMD_FAILED;
+	}
+
+	pd->dev = pv->dev;
+	dm_list_add(&pp->arg_import, &pd->list);
+
+	log_debug("pvchange dev %s VG %s found to import",
+		  dev_name(pd->dev), vg ? vg->name : "<none>");
+
+	pp->found_args++;
+
+	return ECMD_PROCESSED;
+}
+
+static int _pvchange_import_check_vg_single(struct cmd_context *cmd, const char *vg_name,
+		                            struct volume_group *vg, struct processing_handle *handle)
+{
+	char uuid[64] __attribute__((aligned(8)));
+	struct pvchange_params *pp = (struct pvchange_params *) handle->custom_handle;
+	struct pvchange_device *pd;
+	struct pv_list *pvl;
+	int found;
+
+	if (vg_is_exported(vg) && !pp->import_vg) {
+		log_error("VG %s is exported, use the --importvg option.", vg->name);
+		goto fail;
+	}
+
+	if (vg_status(vg) & PARTIAL_VG) {
+		log_error("VG %s is partial, it must be complete.", vg->name);
+		goto fail;
+	}
+
+	/*
+	 * FIXME: This active LV check is not smart enough to distinguish
+	 * between LVs that are active in the original VG vs the cloned VG
+	 * that's being imported.  So, even when the cloned devices being
+	 * imported do not have any active LVs, you still need to deactivate
+	 * LVs in the original VG that is not being imported.  Using the
+	 * DEV_USED_FOR_LV feature is probably what needs to be done here
+	 * instead.
+	 */
+
+	if (lvs_in_vg_activated(vg)) {
+		log_error("VG %s has active LVs, deactivate first.", vg->name);
+		goto fail;
+	}
+
+	/*
+	 * The arg_import list must match the PVs in VG.
+	 */
+
+	dm_list_iterate_items(pvl, &vg->pvs) {
+		found = 0;
+
+		dm_list_iterate_items(pd, &pp->arg_import) {
+			if (pvl->pv->dev != pd->dev)
+				continue;
+			pd->found_in_vg = 1;
+			found = 1;
+			break;
+		}
+
+		if (!found) {
+			if (!id_write_format(&pvl->pv->id, uuid, sizeof(uuid)))
+				goto fail;
+
+			/* all PVs in the VG must be imported together, pvl is missing from args. */
+			log_error("PV with UUID %s is part of VG %s, but is not included in the devices to import.",
+				   uuid, vg->name);
+			log_error("All PVs in the VG must be imported together.");
+			goto fail;
+		}
+	}
+
+	dm_list_iterate_items(pd, &pp->arg_import) {
+		if (!pd->found_in_vg) {
+			/* device arg is not in the VG. */
+			log_error("Device %s was not found in VG %s.", dev_name(pd->dev), vg->name);
+			log_error("The devices to import must match the devices in the VG.");
+			goto fail;
+		}
+	}
+
+	return ECMD_PROCESSED;
+fail:
+	return ECMD_FAILED;
+}
+
+/*
+ * FIXME: It's ugly that we have to write the VG once to change each PV ID, and
+ * then write the VG again later to change the other things.  The metadata
+ * writing functions don't seem to support writing everything at once.  It
+ * should be possible to change the PV UUIDs in _pvchange_import_vg_single.
+ */
+
+static int _pvchange_import_pv_single(struct cmd_context *cmd, struct volume_group *vg,
+				      struct physical_volume *pv, struct processing_handle *handle)
+{
+	struct pvchange_params *pp = (struct pvchange_params *) handle->custom_handle;
+	struct pvchange_device *pd;
+	int found = 0;
+
+	/* Check again that this PV is one we are importing. */
+	dm_list_iterate_items(pd, &pp->arg_import) {
+		if (pd->dev == pv->dev) {
+			found = 1;
+			break;
+		}
+	}
+
+	if (!found) {
+		log_error("PV device %s does not match any command arg.", dev_name(pv->dev));
+		return ECMD_FAILED;
+	}
+
+	if (!archive(vg))
+		return_ECMD_FAILED;
+
+	log_debug("pvchange import writing new uuid on dev %s.", dev_name(pv->dev));
+
+	/* Low level pv_write code needs old_id to be set! */
+	memcpy(&pv->old_id, &pv->id, sizeof(pv->id));
+
+	if (!id_create(&pv->id)) {
+		log_error("Failed to generate new random UUID for %s.", dev_name(pv->dev));
+		return ECMD_FAILED;
+	}
+
+	if (!pv_write(cmd, pv, 1)) {
+		log_error("Failed to write PV for %s.", dev_name(pv->dev));
+		return ECMD_FAILED;
+	}
+
+	if (!vg_write(vg) || !vg_commit(vg)) {
+		log_error("Failed to change VG for %s.", dev_name(pv->dev));
+		return ECMD_FAILED;
+	}
+
+	return ECMD_PROCESSED;
+}
+
+static int _pvchange_import_vg_single(struct cmd_context *cmd, const char *vg_name,
+		                      struct volume_group *vg, struct processing_handle *handle)
+{
+	struct pvchange_params *pp = (struct pvchange_params *) handle->custom_handle;
+	struct pv_list *pvl, *new_pvl;
+	struct lv_list *lvl;
+
+	if (pp->import_vg)
+		vg->status &= ~EXPORTED_VG;
+
+	if (!id_create(&vg->id))
+		goto_bad;
+
+	/* Low level vg_write code needs old_name to be set! */
+	vg->old_name = vg->name;
+
+	if (!(vg->name = dm_pool_strdup(vg->vgmem, pp->new_vgname)))
+		goto_bad;
+
+	dm_list_iterate_items(pvl, &vg->pvs) {
+		if (!(new_pvl = dm_pool_zalloc(vg->vgmem, sizeof(*new_pvl))))
+			goto_bad;
+
+		new_pvl->pv = pvl->pv;
+
+		if (!(pvl->pv->vg_name = dm_pool_strdup(vg->vgmem, pp->new_vgname)))
+			goto_bad;
+
+		/*
+		 * FIXME: It would be nice to write new PV uuids here, but the
+		 * metadata writing functions don't seem to be able to do this,
+		 * so we have a separate loop that writes new PV IDs earlier.
+		 */
+		/*
+		memcpy(&new_pvl->pv->old_id, &new_pvl->pv->id, sizeof(new_pvl->pv->id));
+		id_create(&new_pvl->pv->id);
+		dm_list_add(&vg->pv_write_list, &new_pvl->list);
+		*/
+	}
+
+	dm_list_iterate_items(lvl, &vg->lvs)
+		memcpy(&lvl->lv->lvid, &vg->id, sizeof(vg->id));
+
+	if (!vg_write(vg) || !vg_commit(vg))
+		goto_bad;
+
+	return ECMD_PROCESSED;
+bad:
+	return ECMD_FAILED;
+}
+
+static int _pvchange_importclone(struct cmd_context *cmd, struct processing_handle *handle,
+				 int argc, char **argv)
+{
+	struct pvchange_params *pp = (struct pvchange_params *) handle->custom_handle;
+	struct dm_list vgnameids_on_system;     /* vgnameid_list */
+	struct vgnameid_list *vgnl;
+	struct pvchange_device *pd;
+	struct lvmcache_info *info;
+	const char *vgname;
+	char base_vgname[NAME_LEN] = { 0 };
+	char tmp_vgname[NAME_LEN] = { 0 };
+	unsigned int vgname_count;
+	int ret;
+
+	dm_list_init(&vgnameids_on_system);
+	dm_list_init(&pp->arg_import);
+
+	/*
+	 * Importing clones is by definition dealing with duplicate PVs, so
+	 * disable lvmetad if it's not already. If there are duplicate PVs to
+	 * import, it may already be disabled.
+	 */
+	lvmetad_set_disabled(cmd, LVMETAD_DISABLE_REASON_DUPLICATES);
+
+	if (!lock_vol(cmd, VG_GLOBAL, LCK_VG_WRITE, NULL)) {
+		log_error("Unable to obtain global lock.");
+		return ECMD_FAILED;
+	}
+
+	if (!lockd_gl(cmd, "ex", 0))
+		goto_bad;
+	cmd->lockd_gl_disable = 1;
+
+	/*
+	 * Find the devices being imported which are named on the command line.
+	 * They may be in the list of unchosen duplicates.
+	 */
+
+	log_debug("Finding devices to import.");
+	cmd->command->flags |= ENABLE_DUPLICATE_DEVS;
+	ret = process_each_pv(cmd, argc, argv, NULL, 0, READ_ALLOW_EXPORTED, handle, _pvchange_import_check_pv_single);
+	if (ret != ECMD_PROCESSED)
+		goto_bad;
+
+	if (pp->found_args != argc) {
+		log_error("Failed to find all devices.");
+		goto_bad;
+	}
+
+	/*
+	 * Find the VG name of the PVs being imported, save as old_vgname.
+	 * N.B. If pd->dev is a duplicate, then it may not match info->dev.
+	 * TODO: also verify VGID matches for all.
+	 */
+
+	dm_list_iterate_items(pd, &pp->arg_import) {
+		if (!(info = lvmcache_info_from_pvid(pd->dev->pvid, 0))) {
+			log_error("Failed to find PVID for device %s in lvmcache.", dev_name(pd->dev));
+			goto_bad;
+		}
+
+		if (!(vgname = lvmcache_vgname_from_info(info))) {
+			log_error("Failed to find VG name for device %s in lvmcache.", dev_name(pd->dev));
+			goto_bad;
+		}
+
+		if (!pp->old_vgname) {
+			if (!(pp->old_vgname = dm_pool_strdup(cmd->mem, vgname)))
+				goto_bad;
+		} else {
+			if (strcmp(pp->old_vgname, vgname)) {
+				log_error("Devices must be from the same VG.");
+				goto_bad;
+			}
+		}
+	}
+
+	/*
+	 * Pick a new VG name, save as new_vgname.  The new name begins with
+	 * the basevgname or old_vgname, plus a $i suffix, if necessary, to
+	 * make it unique.  This requires comparing the old_vgname with all the
+	 * VG names on the system.
+	 */
+
+	if (pp->base_vgname) {
+		snprintf(base_vgname, sizeof(base_vgname) - 1, "%s", pp->base_vgname);
+		memcpy(tmp_vgname, base_vgname, NAME_LEN);
+		vgname_count = 0;
+	} else {
+		snprintf(base_vgname, sizeof(base_vgname) - 1, "%s", pp->old_vgname);
+		snprintf(tmp_vgname, sizeof(tmp_vgname) - 1, "%s1", pp->old_vgname);
+		vgname_count = 1;
+	}
+
+	if (!get_vgnameids(cmd, &vgnameids_on_system, NULL, 0))
+		goto_bad;
+
+retry_name:
+	dm_list_iterate_items(vgnl, &vgnameids_on_system) {
+		if (!strcmp(vgnl->vg_name, tmp_vgname)) {
+			vgname_count++;
+			snprintf(tmp_vgname, sizeof(tmp_vgname) - 1, "%s%u", base_vgname, vgname_count);
+			goto retry_name;
+		}
+	}
+
+	if (!(pp->new_vgname = dm_pool_strdup(cmd->mem, tmp_vgname))) {
+		log_error("strdup failed.");
+		goto_bad;
+	}
+	log_debug("Using new VG name %s.", pp->new_vgname);
+
+	/*
+	 * Create a device filter so that we are only working with the devices
+	 * in arg_import.  With the original devs hidden (that arg_import were
+	 * cloned from), we can read and write the cloned PVs and VG without
+	 * touching the original PVs/VG.
+	 */
+
+	dm_list_iterate_items(pd, &pp->arg_import)
+		lvmcache_add_import_clone_dev(cmd, pd->dev);
+	lvmcache_destroy(cmd, 1, 0);
+	dev_cache_full_scan(cmd->full_filter);
+
+	/*
+	 * Read old_vgname, and check that vg->pvs matches the list of devices
+	 * being imported in arg_import, plus some other VG checks.
+	 */
+
+	log_debug("Checking VG to import.");
+	ret = process_each_vg(cmd, 0, NULL, pp->old_vgname, 0, handle, _pvchange_import_check_vg_single);
+	if (ret != ECMD_PROCESSED)
+		goto_bad;
+
+	/*
+	 * Change the PV uuid of each device.
+	 */
+
+	log_debug("Changing PVs.");
+	ret = process_each_pv(cmd, argc, argv, pp->old_vgname, 0, READ_FOR_UPDATE | READ_ALLOW_EXPORTED, handle, _pvchange_import_pv_single);
+	if (ret != ECMD_PROCESSED)
+		goto_bad;
+
+	/*
+	 * Change the VG name and uuid.
+	 */
+
+	log_debug("Changing VG %s to %s.", pp->old_vgname, pp->new_vgname);
+
+	/* We don't care if the new name comes before the old in lock order. */
+	lvmcache_lock_ordering(0);
+
+	if (!lock_vol(cmd, pp->new_vgname, LCK_VG_WRITE, NULL)) {
+		log_error("Can't get lock for new VG name %s", pp->new_vgname);
+		lvmcache_lock_ordering(1);
+		goto bad;
+	}
+
+	ret = process_each_vg(cmd, 0, NULL, pp->old_vgname, READ_FOR_UPDATE | READ_ALLOW_EXPORTED, handle, _pvchange_import_vg_single);
+	if (ret != ECMD_PROCESSED)
+		goto_bad;
+
+	unlock_vg(cmd, pp->new_vgname);
+	unlock_vg(cmd, VG_GLOBAL);
+
+	pp->done = pp->found_args;
+	return ECMD_PROCESSED;
+
+bad:
+	unlock_vg(cmd, VG_GLOBAL);
+	return ECMD_FAILED;
+}
+
 static int _pvchange_single(struct cmd_context *cmd, struct volume_group *vg,
 			    struct physical_volume *pv, struct processing_handle *handle)
 {
@@ -207,8 +603,11 @@ int pvchange(struct cmd_context *cmd, int argc, char **argv)
 	struct processing_handle *handle = NULL;
 	int ret;
 
-	if (!(arg_count(cmd, allocatable_ARG) + arg_is_set(cmd, addtag_ARG) +
-	    arg_is_set(cmd, deltag_ARG) + arg_count(cmd, uuid_ARG) +
+	if (!(arg_count(cmd, allocatable_ARG) +
+	    arg_is_set(cmd, addtag_ARG) +
+	    arg_is_set(cmd, deltag_ARG) +
+	    arg_is_set(cmd, importclone_ARG) +
+	    arg_count(cmd, uuid_ARG) +
 	    arg_count(cmd, metadataignore_ARG))) {
 		log_error("Please give one or more of -x, -uuid, "
 			  "--addtag, --deltag or --metadataignore");
@@ -236,6 +635,12 @@ int pvchange(struct cmd_context *cmd, int argc, char **argv)
 		goto out;
 	}
 
+	if (arg_is_set(cmd, importclone_ARG) && (!argc || arg_count(cmd, all_ARG))) {
+		log_error("Option --importclone requires PV args.");
+		ret = EINVALID_CMD_LINE;
+		goto out;
+	}
+
 	if (!argc) {
 		/*
 		 * Take the global lock here so the lvmcache remains
@@ -252,7 +657,10 @@ int pvchange(struct cmd_context *cmd, int argc, char **argv)
 
 	set_pv_notify(cmd);
 
-	ret = process_each_pv(cmd, argc, argv, NULL, 0, READ_FOR_UPDATE | READ_ALLOW_EXPORTED, handle, _pvchange_single);
+	if (arg_is_set(cmd, importclone_ARG))
+		ret = _pvchange_importclone(cmd, handle, argc, argv);
+	else
+		ret = process_each_pv(cmd, argc, argv, NULL, 0, READ_FOR_UPDATE | READ_ALLOW_EXPORTED, handle, _pvchange_single);
 
 	if (!argc)
 		unlock_vg(cmd, VG_GLOBAL);