Home | History | Annotate | Line # | Download | only in tools
pvmove.c revision 1.1.1.2
      1 /*	$NetBSD: pvmove.c,v 1.1.1.2 2009/12/02 00:25:54 haad Exp $	*/
      2 
      3 /*
      4  * Copyright (C) 2003-2004 Sistina Software, Inc. All rights reserved.
      5  * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
      6  *
      7  * This file is part of LVM2.
      8  *
      9  * This copyrighted material is made available to anyone wishing to use,
     10  * modify, copy, or redistribute it subject to the terms and conditions
     11  * of the GNU Lesser General Public License v.2.1.
     12  *
     13  * You should have received a copy of the GNU Lesser General Public License
     14  * along with this program; if not, write to the Free Software Foundation,
     15  * Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
     16  */
     17 
     18 #include "tools.h"
     19 #include "polldaemon.h"
     20 #include "display.h"
     21 
     22 #define PVMOVE_FIRST_TIME   0x00000001      /* Called for first time */
     23 
     24 static int _pvmove_target_present(struct cmd_context *cmd, int clustered)
     25 {
     26 	const struct segment_type *segtype;
     27 	unsigned attr = 0;
     28 	int found = 1;
     29 	static int _clustered_found = -1;
     30 
     31 	if (clustered && _clustered_found >= 0)
     32 		return _clustered_found;
     33 
     34 	if (!(segtype = get_segtype_from_string(cmd, "mirror")))
     35 		return_0;
     36 
     37 	if (activation() && segtype->ops->target_present &&
     38 	    !segtype->ops->target_present(cmd, NULL, clustered ? &attr : NULL))
     39 		found = 0;
     40 
     41 	if (activation() && clustered) {
     42 		if (found && (attr & MIRROR_LOG_CLUSTERED))
     43 			_clustered_found = found = 1;
     44 		else
     45 			_clustered_found = found = 0;
     46 	}
     47 
     48 	return found;
     49 }
     50 
     51 static unsigned _pvmove_is_exclusive(struct cmd_context *cmd,
     52 				     struct volume_group *vg)
     53 {
     54 	if (vg_is_clustered(vg))
     55 		if (!_pvmove_target_present(cmd, 1))
     56 			return 1;
     57 
     58 	return 0;
     59 }
     60 
     61 /* Allow /dev/vgname/lvname, vgname/lvname or lvname */
     62 static const char *_extract_lvname(struct cmd_context *cmd, const char *vgname,
     63 				   const char *arg)
     64 {
     65 	const char *lvname;
     66 
     67 	/* Is an lvname supplied directly? */
     68 	if (!strchr(arg, '/'))
     69 		return arg;
     70 
     71 	lvname = skip_dev_dir(cmd, arg, NULL);
     72 	while (*lvname == '/')
     73 		lvname++;
     74 	if (!strchr(lvname, '/')) {
     75 		log_error("--name takes a logical volume name");
     76 		return NULL;
     77 	}
     78 	if (strncmp(vgname, lvname, strlen(vgname)) ||
     79 	    (lvname += strlen(vgname), *lvname != '/')) {
     80 		log_error("Named LV and old PV must be in the same VG");
     81 		return NULL;
     82 	}
     83 	while (*lvname == '/')
     84 		lvname++;
     85 	if (!*lvname) {
     86 		log_error("Incomplete LV name supplied with --name");
     87 		return NULL;
     88 	}
     89 	return lvname;
     90 }
     91 
     92 static struct volume_group *_get_vg(struct cmd_context *cmd, const char *vgname)
     93 {
     94 	dev_close_all();
     95 
     96 	return vg_read_for_update(cmd, vgname, NULL, 0);
     97 }
     98 
     99 /* Create list of PVs for allocation of replacement extents */
    100 static struct dm_list *_get_allocatable_pvs(struct cmd_context *cmd, int argc,
    101 					 char **argv, struct volume_group *vg,
    102 					 struct physical_volume *pv,
    103 					 alloc_policy_t alloc)
    104 {
    105 	struct dm_list *allocatable_pvs, *pvht, *pvh;
    106 	struct pv_list *pvl;
    107 
    108 	if (argc)
    109 		allocatable_pvs = create_pv_list(cmd->mem, vg, argc, argv, 1);
    110 	else
    111 		allocatable_pvs = clone_pv_list(cmd->mem, &vg->pvs);
    112 
    113 	if (!allocatable_pvs)
    114 		return_NULL;
    115 
    116 	dm_list_iterate_safe(pvh, pvht, allocatable_pvs) {
    117 		pvl = dm_list_item(pvh, struct pv_list);
    118 
    119 		/* Don't allocate onto the PV we're clearing! */
    120 		if ((alloc != ALLOC_ANYWHERE) && (pvl->pv->dev == pv_dev(pv))) {
    121 			dm_list_del(&pvl->list);
    122 			continue;
    123 		}
    124 
    125 		/* Remove PV if full */
    126 		if ((pvl->pv->pe_count == pvl->pv->pe_alloc_count))
    127 			dm_list_del(&pvl->list);
    128 	}
    129 
    130 	if (dm_list_empty(allocatable_pvs)) {
    131 		log_error("No extents available for allocation");
    132 		return NULL;
    133 	}
    134 
    135 	return allocatable_pvs;
    136 }
    137 
    138 /*
    139  * Replace any LV segments on given PV with temporary mirror.
    140  * Returns list of LVs changed.
    141  */
    142 static int _insert_pvmove_mirrors(struct cmd_context *cmd,
    143 				  struct logical_volume *lv_mirr,
    144 				  struct dm_list *source_pvl,
    145 				  struct logical_volume *lv,
    146 				  struct dm_list *lvs_changed)
    147 
    148 {
    149 	struct pv_list *pvl;
    150 	uint32_t prev_le_count;
    151 
    152 	/* Only 1 PV may feature in source_pvl */
    153 	pvl = dm_list_item(source_pvl->n, struct pv_list);
    154 
    155 	prev_le_count = lv_mirr->le_count;
    156 	if (!insert_layer_for_segments_on_pv(cmd, lv, lv_mirr, PVMOVE,
    157 					     pvl, lvs_changed))
    158 		return_0;
    159 
    160 	/* check if layer was inserted */
    161 	if (lv_mirr->le_count - prev_le_count) {
    162 		lv->status |= LOCKED;
    163 
    164 		log_verbose("Moving %u extents of logical volume %s/%s",
    165 			    lv_mirr->le_count - prev_le_count,
    166 			    lv->vg->name, lv->name);
    167 	}
    168 
    169 	return 1;
    170 }
    171 
    172 /* Create new LV with mirror segments for the required copies */
    173 static struct logical_volume *_set_up_pvmove_lv(struct cmd_context *cmd,
    174 						struct volume_group *vg,
    175 						struct dm_list *source_pvl,
    176 						const char *lv_name,
    177 						struct dm_list *allocatable_pvs,
    178 						alloc_policy_t alloc,
    179 						struct dm_list **lvs_changed)
    180 {
    181 	struct logical_volume *lv_mirr, *lv;
    182 	struct lv_list *lvl;
    183 	uint32_t log_count = 0;
    184 	int lv_found = 0;
    185 
    186 	/* FIXME Cope with non-contiguous => splitting existing segments */
    187 	if (!(lv_mirr = lv_create_empty("pvmove%d", NULL,
    188 					LVM_READ | LVM_WRITE,
    189 					ALLOC_CONTIGUOUS, vg))) {
    190 		log_error("Creation of temporary pvmove LV failed");
    191 		return NULL;
    192 	}
    193 
    194 	lv_mirr->status |= (PVMOVE | LOCKED);
    195 
    196 	if (!(*lvs_changed = dm_pool_alloc(cmd->mem, sizeof(**lvs_changed)))) {
    197 		log_error("lvs_changed list struct allocation failed");
    198 		return NULL;
    199 	}
    200 
    201 	dm_list_init(*lvs_changed);
    202 
    203 	/* Find segments to be moved and set up mirrors */
    204 	dm_list_iterate_items(lvl, &vg->lvs) {
    205 		lv = lvl->lv;
    206 		if ((lv == lv_mirr))
    207 			continue;
    208 		if (lv_name) {
    209 			if (strcmp(lv->name, lv_name))
    210 				continue;
    211 			lv_found = 1;
    212 		}
    213 		if (lv_is_origin(lv) || lv_is_cow(lv)) {
    214 			log_print("Skipping snapshot-related LV %s", lv->name);
    215 			continue;
    216 		}
    217 		if (lv->status & MIRRORED) {
    218 			log_print("Skipping mirror LV %s", lv->name);
    219 			continue;
    220 		}
    221 		if (lv->status & MIRROR_LOG) {
    222 			log_print("Skipping mirror log LV %s", lv->name);
    223 			continue;
    224 		}
    225 		if (lv->status & MIRROR_IMAGE) {
    226 			log_print("Skipping mirror image LV %s", lv->name);
    227 			continue;
    228 		}
    229 		if (lv->status & LOCKED) {
    230 			log_print("Skipping locked LV %s", lv->name);
    231 			continue;
    232 		}
    233 		if (!_insert_pvmove_mirrors(cmd, lv_mirr, source_pvl, lv,
    234 					    *lvs_changed))
    235 			return_NULL;
    236 	}
    237 
    238 	if (lv_name && !lv_found) {
    239 		log_error("Logical volume %s not found.", lv_name);
    240 		return NULL;
    241 	}
    242 
    243 	/* Is temporary mirror empty? */
    244 	if (!lv_mirr->le_count) {
    245 		log_error("No data to move for %s", vg->name);
    246 		return NULL;
    247 	}
    248 
    249 	if (!lv_add_mirrors(cmd, lv_mirr, 1, 1, 0, log_count,
    250 			    allocatable_pvs, alloc, MIRROR_BY_SEG)) {
    251 		log_error("Failed to convert pvmove LV to mirrored");
    252 		return_NULL;
    253 	}
    254 
    255 	if (!split_parent_segments_for_layer(cmd, lv_mirr)) {
    256 		log_error("Failed to split segments being moved");
    257 		return_NULL;
    258 	}
    259 
    260 	return lv_mirr;
    261 }
    262 
    263 static int _activate_lv(struct cmd_context *cmd, struct logical_volume *lv_mirr,
    264 			unsigned exclusive)
    265 {
    266 	if (exclusive)
    267 		return activate_lv_excl(cmd, lv_mirr);
    268 
    269 	return activate_lv(cmd, lv_mirr);
    270 }
    271 
    272 static int _finish_pvmove(struct cmd_context *cmd, struct volume_group *vg,
    273 			  struct logical_volume *lv_mirr,
    274 			  struct dm_list *lvs_changed);
    275 
    276 static int _update_metadata(struct cmd_context *cmd, struct volume_group *vg,
    277 			    struct logical_volume *lv_mirr,
    278 			    struct dm_list *lvs_changed, unsigned flags)
    279 {
    280 	unsigned exclusive = _pvmove_is_exclusive(cmd, vg);
    281 	unsigned first_time = (flags & PVMOVE_FIRST_TIME) ? 1 : 0;
    282 	int r = 0;
    283 
    284 	log_verbose("Updating volume group metadata");
    285 	if (!vg_write(vg)) {
    286 		log_error("ABORTING: Volume group metadata update failed.");
    287 		return 0;
    288 	}
    289 
    290 	/* Suspend lvs_changed */
    291 	if (!suspend_lvs(cmd, lvs_changed))
    292 		goto_out;
    293 
    294 	/* Suspend mirrors on subsequent calls */
    295 	if (!first_time) {
    296 		if (!suspend_lv(cmd, lv_mirr)) {
    297 			resume_lvs(cmd, lvs_changed);
    298 			vg_revert(vg);
    299 			goto_out;
    300 		}
    301 	}
    302 
    303 	/* Commit on-disk metadata */
    304 	if (!vg_commit(vg)) {
    305 		log_error("ABORTING: Volume group metadata update failed.");
    306 		if (!first_time)
    307 			resume_lv(cmd, lv_mirr);
    308 		resume_lvs(cmd, lvs_changed);
    309 		goto out;
    310 	}
    311 
    312 	/* Activate the temporary mirror LV */
    313 	/* Only the first mirror segment gets activated as a mirror */
    314 	/* FIXME: Add option to use a log */
    315 	if (first_time) {
    316 		if (!_activate_lv(cmd, lv_mirr, exclusive)) {
    317 			if (test_mode())
    318 				goto out;
    319 
    320 			/*
    321 			 * Nothing changed yet, try to revert pvmove.
    322 			 */
    323 			log_error("Temporary pvmove mirror activation failed.");
    324 			if (!_finish_pvmove(cmd, vg, lv_mirr, lvs_changed))
    325 				log_error("ABORTING: Restoring original configuration "
    326 					  "before pvmove failed. Run pvmove --abort.");
    327 			goto out;
    328 		}
    329 	} else if (!resume_lv(cmd, lv_mirr)) {
    330 		log_error("Unable to reactivate logical volume \"%s\"",
    331 			  lv_mirr->name);
    332 		resume_lvs(cmd, lvs_changed);
    333 		goto out;
    334 	}
    335 
    336 	/* Unsuspend LVs */
    337 	if (!resume_lvs(cmd, lvs_changed)) {
    338 		log_error("Unable to resume logical volumes");
    339 		goto out;
    340 	}
    341 
    342 	r = 1;
    343 out:
    344 	backup(vg);
    345 	return r;
    346 }
    347 
    348 static int _set_up_pvmove(struct cmd_context *cmd, const char *pv_name,
    349 			  int argc, char **argv)
    350 {
    351 	const char *lv_name = NULL;
    352 	char *pv_name_arg;
    353 	struct volume_group *vg;
    354 	struct dm_list *source_pvl;
    355 	struct dm_list *allocatable_pvs;
    356 	alloc_policy_t alloc;
    357 	struct dm_list *lvs_changed;
    358 	struct physical_volume *pv;
    359 	struct logical_volume *lv_mirr;
    360 	unsigned first_time = 1;
    361 	unsigned exclusive;
    362 	int r = ECMD_FAILED;
    363 
    364 	pv_name_arg = argv[0];
    365 	argc--;
    366 	argv++;
    367 
    368 	/* Find PV (in VG) */
    369 	if (!(pv = find_pv_by_name(cmd, pv_name))) {
    370 		stack;
    371 		return EINVALID_CMD_LINE;
    372 	}
    373 
    374 	if (arg_count(cmd, name_ARG)) {
    375 		if (!(lv_name = _extract_lvname(cmd, pv_vg_name(pv),
    376 						arg_value(cmd, name_ARG)))) {
    377 			stack;
    378 			return EINVALID_CMD_LINE;
    379 		}
    380 
    381 		if (!validate_name(lv_name)) {
    382 			log_error("Logical volume name %s is invalid", lv_name);
    383 			return EINVALID_CMD_LINE;
    384 		}
    385 	}
    386 
    387 	/* Read VG */
    388 	log_verbose("Finding volume group \"%s\"", pv_vg_name(pv));
    389 
    390 	vg = _get_vg(cmd, pv_vg_name(pv));
    391 	if (vg_read_error(vg)) {
    392 		vg_release(vg);
    393 		stack;
    394 		return ECMD_FAILED;
    395 	}
    396 
    397 	exclusive = _pvmove_is_exclusive(cmd, vg);
    398 
    399 	if ((lv_mirr = find_pvmove_lv(vg, pv_dev(pv), PVMOVE))) {
    400 		log_print("Detected pvmove in progress for %s", pv_name);
    401 		if (argc || lv_name)
    402 			log_error("Ignoring remaining command line arguments");
    403 
    404 		if (!(lvs_changed = lvs_using_lv(cmd, vg, lv_mirr))) {
    405 			log_error("ABORTING: Failed to generate list of moving LVs");
    406 			goto out;
    407 		}
    408 
    409 		/* Ensure mirror LV is active */
    410 		if (!_activate_lv(cmd, lv_mirr, exclusive)) {
    411 			log_error("ABORTING: Temporary mirror activation failed.");
    412 			goto out;
    413 		}
    414 
    415 		first_time = 0;
    416 	} else {
    417 		/* Determine PE ranges to be moved */
    418 		if (!(source_pvl = create_pv_list(cmd->mem, vg, 1,
    419 						  &pv_name_arg, 0)))
    420 			goto_out;
    421 
    422 		alloc = arg_uint_value(cmd, alloc_ARG, ALLOC_INHERIT);
    423 		if (alloc == ALLOC_INHERIT)
    424 			alloc = vg->alloc;
    425 
    426 		/* Get PVs we can use for allocation */
    427 		if (!(allocatable_pvs = _get_allocatable_pvs(cmd, argc, argv,
    428 							     vg, pv, alloc)))
    429 			goto_out;
    430 
    431 		if (!archive(vg))
    432 			goto_out;
    433 
    434 		if (!(lv_mirr = _set_up_pvmove_lv(cmd, vg, source_pvl, lv_name,
    435 						  allocatable_pvs, alloc,
    436 						  &lvs_changed)))
    437 			goto_out;
    438 	}
    439 
    440 	/* Lock lvs_changed and activate (with old metadata) */
    441 	if (!activate_lvs(cmd, lvs_changed, exclusive))
    442 		goto_out;
    443 
    444 	/* FIXME Presence of a mirror once set PVMOVE - now remove associated logic */
    445 	/* init_pvmove(1); */
    446 	/* vg->status |= PVMOVE; */
    447 
    448 	if (first_time) {
    449 		if (!_update_metadata
    450 		    (cmd, vg, lv_mirr, lvs_changed, PVMOVE_FIRST_TIME))
    451 			goto_out;
    452 	}
    453 
    454 	/* LVs are all in status LOCKED */
    455 	r = ECMD_PROCESSED;
    456 out:
    457 	unlock_and_release_vg(cmd, vg, pv_vg_name(pv));
    458 	return r;
    459 }
    460 
    461 static int _finish_pvmove(struct cmd_context *cmd, struct volume_group *vg,
    462 			  struct logical_volume *lv_mirr,
    463 			  struct dm_list *lvs_changed)
    464 {
    465 	int r = 1;
    466 	struct dm_list lvs_completed;
    467 	struct lv_list *lvl;
    468 
    469 	/* Update metadata to remove mirror segments and break dependencies */
    470 	dm_list_init(&lvs_completed);
    471 	if (!lv_remove_mirrors(cmd, lv_mirr, 1, 0, NULL, PVMOVE) ||
    472 	    !remove_layers_for_segments_all(cmd, lv_mirr, PVMOVE,
    473 					    &lvs_completed)) {
    474 		log_error("ABORTING: Removal of temporary mirror failed");
    475 		return 0;
    476 	}
    477 
    478 	dm_list_iterate_items(lvl, &lvs_completed)
    479 		/* FIXME Assumes only one pvmove at a time! */
    480 		lvl->lv->status &= ~LOCKED;
    481 
    482 	/* Store metadata without dependencies on mirror segments */
    483 	if (!vg_write(vg)) {
    484 		log_error("ABORTING: Failed to write new data locations "
    485 			  "to disk.");
    486 		return 0;
    487 	}
    488 
    489 	/* Suspend LVs changed */
    490 	if (!suspend_lvs(cmd, lvs_changed)) {
    491 		log_error("Locking LVs to remove temporary mirror failed");
    492 		r = 0;
    493 	}
    494 
    495 	/* Suspend mirror LV to flush pending I/O */
    496 	if (!suspend_lv(cmd, lv_mirr)) {
    497 		log_error("Suspension of temporary mirror LV failed");
    498 		r = 0;
    499 	}
    500 
    501 	/* Store metadata without dependencies on mirror segments */
    502 	if (!vg_commit(vg)) {
    503 		log_error("ABORTING: Failed to write new data locations "
    504 			  "to disk.");
    505 		vg_revert(vg);
    506 		resume_lv(cmd, lv_mirr);
    507 		resume_lvs(cmd, lvs_changed);
    508 		return 0;
    509 	}
    510 
    511 	/* Release mirror LV.  (No pending I/O because it's been suspended.) */
    512 	if (!resume_lv(cmd, lv_mirr)) {
    513 		log_error("Unable to reactivate logical volume \"%s\"",
    514 			  lv_mirr->name);
    515 		r = 0;
    516 	}
    517 
    518 	/* Unsuspend LVs */
    519 	resume_lvs(cmd, lvs_changed);
    520 
    521 	/* Deactivate mirror LV */
    522 	if (!deactivate_lv(cmd, lv_mirr)) {
    523 		log_error("ABORTING: Unable to deactivate temporary logical "
    524 			  "volume \"%s\"", lv_mirr->name);
    525 		r = 0;
    526 	}
    527 
    528 	log_verbose("Removing temporary pvmove LV");
    529 	if (!lv_remove(lv_mirr)) {
    530 		log_error("ABORTING: Removal of temporary pvmove LV failed");
    531 		return 0;
    532 	}
    533 
    534 	/* Store it on disks */
    535 	log_verbose("Writing out final volume group after pvmove");
    536 	if (!vg_write(vg) || !vg_commit(vg)) {
    537 		log_error("ABORTING: Failed to write new data locations "
    538 			  "to disk.");
    539 		return 0;
    540 	}
    541 
    542 	/* FIXME backup positioning */
    543 	backup(vg);
    544 
    545 	return r;
    546 }
    547 
    548 static struct volume_group *_get_move_vg(struct cmd_context *cmd,
    549 					 const char *name, const char *uuid)
    550 {
    551 	struct physical_volume *pv;
    552 
    553 	/* Reread all metadata in case it got changed */
    554 	if (!(pv = find_pv_by_name(cmd, name))) {
    555 		log_error("ABORTING: Can't reread PV %s", name);
    556 		/* What more could we do here? */
    557 		return NULL;
    558 	}
    559 
    560 	return _get_vg(cmd, pv_vg_name(pv));
    561 }
    562 
    563 static struct poll_functions _pvmove_fns = {
    564 	.get_copy_name_from_lv = get_pvmove_pvname_from_lv_mirr,
    565 	.get_copy_vg = _get_move_vg,
    566 	.get_copy_lv = find_pvmove_lv_from_pvname,
    567 	.poll_progress = poll_mirror_progress,
    568 	.update_metadata = _update_metadata,
    569 	.finish_copy = _finish_pvmove,
    570 };
    571 
    572 int pvmove_poll(struct cmd_context *cmd, const char *pv_name,
    573 		unsigned background)
    574 {
    575 	return poll_daemon(cmd, pv_name, NULL, background, PVMOVE, &_pvmove_fns,
    576 			   "Moved");
    577 }
    578 
    579 int pvmove(struct cmd_context *cmd, int argc, char **argv)
    580 {
    581 	char *pv_name = NULL;
    582 	char *colon;
    583 	int ret;
    584 
    585 	/* dm raid1 target must be present in every case */
    586 	if (!_pvmove_target_present(cmd, 0)) {
    587 		log_error("Required device-mapper target(s) not "
    588 			  "detected in your kernel");
    589 		return ECMD_FAILED;
    590 	}
    591 
    592 	if (argc) {
    593 		pv_name = argv[0];
    594 
    595 		/* Drop any PE lists from PV name */
    596 		if ((colon = strchr(pv_name, ':'))) {
    597 			if (!(pv_name = dm_pool_strndup(cmd->mem, pv_name,
    598 						     (unsigned) (colon -
    599 								 pv_name)))) {
    600 				log_error("Failed to clone PV name");
    601 				return ECMD_FAILED;
    602 			}
    603 		}
    604 
    605 		if (!arg_count(cmd, abort_ARG) &&
    606 		    (ret = _set_up_pvmove(cmd, pv_name, argc, argv)) !=
    607 		    ECMD_PROCESSED) {
    608 			stack;
    609 			return ret;
    610 		}
    611 	}
    612 
    613 	return pvmove_poll(cmd, pv_name, arg_is_set(cmd, background_ARG));
    614 }
    615