Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) STRATO AG 2012.  All rights reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/bio.h>
   8#include <linux/slab.h>
 
   9#include <linux/blkdev.h>
 
 
 
  10#include <linux/kthread.h>
  11#include <linux/math64.h>
  12#include "misc.h"
  13#include "ctree.h"
  14#include "extent_map.h"
  15#include "disk-io.h"
  16#include "transaction.h"
  17#include "print-tree.h"
  18#include "volumes.h"
  19#include "async-thread.h"
  20#include "check-integrity.h"
  21#include "rcu-string.h"
  22#include "dev-replace.h"
  23#include "sysfs.h"
  24
  25static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  26				       int scrub_ret);
  27static void btrfs_dev_replace_update_device_in_mapping_tree(
  28						struct btrfs_fs_info *fs_info,
  29						struct btrfs_device *srcdev,
  30						struct btrfs_device *tgtdev);
 
 
 
 
  31static int btrfs_dev_replace_kthread(void *data);
 
 
  32
  33int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  34{
  35	struct btrfs_key key;
  36	struct btrfs_root *dev_root = fs_info->dev_root;
  37	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  38	struct extent_buffer *eb;
  39	int slot;
  40	int ret = 0;
  41	struct btrfs_path *path = NULL;
  42	int item_size;
  43	struct btrfs_dev_replace_item *ptr;
  44	u64 src_devid;
  45
  46	path = btrfs_alloc_path();
  47	if (!path) {
  48		ret = -ENOMEM;
  49		goto out;
  50	}
  51
  52	key.objectid = 0;
  53	key.type = BTRFS_DEV_REPLACE_KEY;
  54	key.offset = 0;
  55	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  56	if (ret) {
  57no_valid_dev_replace_entry_found:
  58		ret = 0;
  59		dev_replace->replace_state =
  60			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
  61		dev_replace->cont_reading_from_srcdev_mode =
  62		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
 
  63		dev_replace->time_started = 0;
  64		dev_replace->time_stopped = 0;
  65		atomic64_set(&dev_replace->num_write_errors, 0);
  66		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
  67		dev_replace->cursor_left = 0;
  68		dev_replace->committed_cursor_left = 0;
  69		dev_replace->cursor_left_last_write_of_item = 0;
  70		dev_replace->cursor_right = 0;
  71		dev_replace->srcdev = NULL;
  72		dev_replace->tgtdev = NULL;
  73		dev_replace->is_valid = 0;
  74		dev_replace->item_needs_writeback = 0;
  75		goto out;
  76	}
  77	slot = path->slots[0];
  78	eb = path->nodes[0];
  79	item_size = btrfs_item_size_nr(eb, slot);
  80	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
  81
  82	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
  83		btrfs_warn(fs_info,
  84			"dev_replace entry found has unexpected size, ignore entry");
  85		goto no_valid_dev_replace_entry_found;
  86	}
  87
  88	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
  89	dev_replace->cont_reading_from_srcdev_mode =
  90		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
  91	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
  92	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
  93	dev_replace->time_stopped =
  94		btrfs_dev_replace_time_stopped(eb, ptr);
  95	atomic64_set(&dev_replace->num_write_errors,
  96		     btrfs_dev_replace_num_write_errors(eb, ptr));
  97	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
  98		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
  99	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 100	dev_replace->committed_cursor_left = dev_replace->cursor_left;
 101	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 102	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 103	dev_replace->is_valid = 1;
 104
 105	dev_replace->item_needs_writeback = 0;
 106	switch (dev_replace->replace_state) {
 107	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 108	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 109	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 110		dev_replace->srcdev = NULL;
 111		dev_replace->tgtdev = NULL;
 112		break;
 113	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 114	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 115		dev_replace->srcdev = btrfs_find_device(fs_info->fs_devices,
 116						src_devid, NULL, NULL, true);
 117		dev_replace->tgtdev = btrfs_find_device(fs_info->fs_devices,
 118							BTRFS_DEV_REPLACE_DEVID,
 119							NULL, NULL, true);
 120		/*
 121		 * allow 'btrfs dev replace_cancel' if src/tgt device is
 122		 * missing
 123		 */
 124		if (!dev_replace->srcdev &&
 125		    !btrfs_test_opt(fs_info, DEGRADED)) {
 126			ret = -EIO;
 127			btrfs_warn(fs_info,
 128			   "cannot mount because device replace operation is ongoing and");
 129			btrfs_warn(fs_info,
 130			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 131			   src_devid);
 132		}
 133		if (!dev_replace->tgtdev &&
 134		    !btrfs_test_opt(fs_info, DEGRADED)) {
 135			ret = -EIO;
 136			btrfs_warn(fs_info,
 137			   "cannot mount because device replace operation is ongoing and");
 138			btrfs_warn(fs_info,
 139			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 140				BTRFS_DEV_REPLACE_DEVID);
 141		}
 142		if (dev_replace->tgtdev) {
 143			if (dev_replace->srcdev) {
 144				dev_replace->tgtdev->total_bytes =
 145					dev_replace->srcdev->total_bytes;
 146				dev_replace->tgtdev->disk_total_bytes =
 147					dev_replace->srcdev->disk_total_bytes;
 148				dev_replace->tgtdev->commit_total_bytes =
 149					dev_replace->srcdev->commit_total_bytes;
 150				dev_replace->tgtdev->bytes_used =
 151					dev_replace->srcdev->bytes_used;
 152				dev_replace->tgtdev->commit_bytes_used =
 153					dev_replace->srcdev->commit_bytes_used;
 154			}
 155			set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 156				&dev_replace->tgtdev->dev_state);
 157
 158			WARN_ON(fs_info->fs_devices->rw_devices == 0);
 159			dev_replace->tgtdev->io_width = fs_info->sectorsize;
 160			dev_replace->tgtdev->io_align = fs_info->sectorsize;
 161			dev_replace->tgtdev->sector_size = fs_info->sectorsize;
 162			dev_replace->tgtdev->fs_info = fs_info;
 163			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
 164				&dev_replace->tgtdev->dev_state);
 165		}
 166		break;
 167	}
 168
 169out:
 170	btrfs_free_path(path);
 171	return ret;
 172}
 173
 174/*
 175 * Initialize a new device for device replace target from a given source dev
 176 * and path.
 177 *
 178 * Return 0 and new device in @device_out, otherwise return < 0
 179 */
 180static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
 181				  const char *device_path,
 182				  struct btrfs_device *srcdev,
 183				  struct btrfs_device **device_out)
 184{
 185	struct btrfs_device *device;
 186	struct block_device *bdev;
 187	struct list_head *devices;
 188	struct rcu_string *name;
 189	u64 devid = BTRFS_DEV_REPLACE_DEVID;
 190	int ret = 0;
 191
 192	*device_out = NULL;
 193	if (fs_info->fs_devices->seeding) {
 194		btrfs_err(fs_info, "the filesystem is a seed filesystem!");
 195		return -EINVAL;
 196	}
 197
 198	bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
 199				  fs_info->bdev_holder);
 200	if (IS_ERR(bdev)) {
 201		btrfs_err(fs_info, "target device %s is invalid!", device_path);
 202		return PTR_ERR(bdev);
 203	}
 204
 205	sync_blockdev(bdev);
 206
 207	devices = &fs_info->fs_devices->devices;
 208	list_for_each_entry(device, devices, dev_list) {
 209		if (device->bdev == bdev) {
 210			btrfs_err(fs_info,
 211				  "target device is in the filesystem!");
 212			ret = -EEXIST;
 213			goto error;
 214		}
 215	}
 216
 217
 218	if (i_size_read(bdev->bd_inode) <
 219	    btrfs_device_get_total_bytes(srcdev)) {
 220		btrfs_err(fs_info,
 221			  "target device is smaller than source device!");
 222		ret = -EINVAL;
 223		goto error;
 224	}
 225
 226
 227	device = btrfs_alloc_device(NULL, &devid, NULL);
 228	if (IS_ERR(device)) {
 229		ret = PTR_ERR(device);
 230		goto error;
 231	}
 232
 233	name = rcu_string_strdup(device_path, GFP_KERNEL);
 234	if (!name) {
 235		btrfs_free_device(device);
 236		ret = -ENOMEM;
 237		goto error;
 238	}
 239	rcu_assign_pointer(device->name, name);
 240
 241	set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 242	device->generation = 0;
 243	device->io_width = fs_info->sectorsize;
 244	device->io_align = fs_info->sectorsize;
 245	device->sector_size = fs_info->sectorsize;
 246	device->total_bytes = btrfs_device_get_total_bytes(srcdev);
 247	device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
 248	device->bytes_used = btrfs_device_get_bytes_used(srcdev);
 249	device->commit_total_bytes = srcdev->commit_total_bytes;
 250	device->commit_bytes_used = device->bytes_used;
 251	device->fs_info = fs_info;
 252	device->bdev = bdev;
 253	set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 254	set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
 255	device->mode = FMODE_EXCL;
 256	device->dev_stats_valid = 1;
 257	set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
 258	device->fs_devices = fs_info->fs_devices;
 259
 260	mutex_lock(&fs_info->fs_devices->device_list_mutex);
 261	list_add(&device->dev_list, &fs_info->fs_devices->devices);
 262	fs_info->fs_devices->num_devices++;
 263	fs_info->fs_devices->open_devices++;
 264	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 265
 266	*device_out = device;
 267	return 0;
 268
 269error:
 270	blkdev_put(bdev, FMODE_EXCL);
 271	return ret;
 272}
 273
 274/*
 275 * called from commit_transaction. Writes changed device replace state to
 276 * disk.
 277 */
 278int btrfs_run_dev_replace(struct btrfs_trans_handle *trans)
 
 279{
 280	struct btrfs_fs_info *fs_info = trans->fs_info;
 281	int ret;
 282	struct btrfs_root *dev_root = fs_info->dev_root;
 283	struct btrfs_path *path;
 284	struct btrfs_key key;
 285	struct extent_buffer *eb;
 286	struct btrfs_dev_replace_item *ptr;
 287	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 288
 289	down_read(&dev_replace->rwsem);
 290	if (!dev_replace->is_valid ||
 291	    !dev_replace->item_needs_writeback) {
 292		up_read(&dev_replace->rwsem);
 293		return 0;
 294	}
 295	up_read(&dev_replace->rwsem);
 296
 297	key.objectid = 0;
 298	key.type = BTRFS_DEV_REPLACE_KEY;
 299	key.offset = 0;
 300
 301	path = btrfs_alloc_path();
 302	if (!path) {
 303		ret = -ENOMEM;
 304		goto out;
 305	}
 306	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 307	if (ret < 0) {
 308		btrfs_warn(fs_info,
 309			   "error %d while searching for dev_replace item!",
 310			   ret);
 311		goto out;
 312	}
 313
 314	if (ret == 0 &&
 315	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 316		/*
 317		 * need to delete old one and insert a new one.
 318		 * Since no attempt is made to recover any old state, if the
 319		 * dev_replace state is 'running', the data on the target
 320		 * drive is lost.
 321		 * It would be possible to recover the state: just make sure
 322		 * that the beginning of the item is never changed and always
 323		 * contains all the essential information. Then read this
 324		 * minimal set of information and use it as a base for the
 325		 * new state.
 326		 */
 327		ret = btrfs_del_item(trans, dev_root, path);
 328		if (ret != 0) {
 329			btrfs_warn(fs_info,
 330				   "delete too small dev_replace item failed %d!",
 331				   ret);
 332			goto out;
 333		}
 334		ret = 1;
 335	}
 336
 337	if (ret == 1) {
 338		/* need to insert a new item */
 339		btrfs_release_path(path);
 340		ret = btrfs_insert_empty_item(trans, dev_root, path,
 341					      &key, sizeof(*ptr));
 342		if (ret < 0) {
 343			btrfs_warn(fs_info,
 344				   "insert dev_replace item failed %d!", ret);
 345			goto out;
 346		}
 347	}
 348
 349	eb = path->nodes[0];
 350	ptr = btrfs_item_ptr(eb, path->slots[0],
 351			     struct btrfs_dev_replace_item);
 352
 353	down_write(&dev_replace->rwsem);
 354	if (dev_replace->srcdev)
 355		btrfs_set_dev_replace_src_devid(eb, ptr,
 356			dev_replace->srcdev->devid);
 357	else
 358		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 359	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 360		dev_replace->cont_reading_from_srcdev_mode);
 361	btrfs_set_dev_replace_replace_state(eb, ptr,
 362		dev_replace->replace_state);
 363	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 364	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 365	btrfs_set_dev_replace_num_write_errors(eb, ptr,
 366		atomic64_read(&dev_replace->num_write_errors));
 367	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 368		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 369	dev_replace->cursor_left_last_write_of_item =
 370		dev_replace->cursor_left;
 371	btrfs_set_dev_replace_cursor_left(eb, ptr,
 372		dev_replace->cursor_left_last_write_of_item);
 373	btrfs_set_dev_replace_cursor_right(eb, ptr,
 374		dev_replace->cursor_right);
 375	dev_replace->item_needs_writeback = 0;
 376	up_write(&dev_replace->rwsem);
 377
 378	btrfs_mark_buffer_dirty(eb);
 379
 380out:
 381	btrfs_free_path(path);
 382
 383	return ret;
 384}
 385
 386static char* btrfs_dev_name(struct btrfs_device *device)
 387{
 388	if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 389		return "<missing disk>";
 390	else
 391		return rcu_str_deref(device->name);
 392}
 393
 394static int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 395		const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 396		int read_src)
 397{
 398	struct btrfs_root *root = fs_info->dev_root;
 399	struct btrfs_trans_handle *trans;
 
 400	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 401	int ret;
 402	struct btrfs_device *tgt_device = NULL;
 403	struct btrfs_device *src_device = NULL;
 404
 405	src_device = btrfs_find_device_by_devspec(fs_info, srcdevid,
 406						  srcdev_name);
 407	if (IS_ERR(src_device))
 408		return PTR_ERR(src_device);
 409
 410	if (btrfs_pinned_by_swapfile(fs_info, src_device)) {
 411		btrfs_warn_in_rcu(fs_info,
 412	  "cannot replace device %s (devid %llu) due to active swapfile",
 413			btrfs_dev_name(src_device), src_device->devid);
 414		return -ETXTBSY;
 415	}
 416
 417	/*
 418	 * Here we commit the transaction to make sure commit_total_bytes
 419	 * of all the devices are updated.
 420	 */
 421	trans = btrfs_attach_transaction(root);
 422	if (!IS_ERR(trans)) {
 423		ret = btrfs_commit_transaction(trans);
 424		if (ret)
 425			return ret;
 426	} else if (PTR_ERR(trans) != -ENOENT) {
 427		return PTR_ERR(trans);
 428	}
 429
 430	ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 431					    src_device, &tgt_device);
 432	if (ret)
 433		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 434
 435	down_write(&dev_replace->rwsem);
 436	switch (dev_replace->replace_state) {
 437	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 438	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 439	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 440		break;
 441	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 442	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 443		ASSERT(0);
 444		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 445		up_write(&dev_replace->rwsem);
 446		goto leave;
 447	}
 448
 449	dev_replace->cont_reading_from_srcdev_mode = read_src;
 
 
 450	dev_replace->srcdev = src_device;
 
 451	dev_replace->tgtdev = tgt_device;
 452
 453	btrfs_info_in_rcu(fs_info,
 454		      "dev_replace from %s (devid %llu) to %s started",
 455		      btrfs_dev_name(src_device),
 
 456		      src_device->devid,
 457		      rcu_str_deref(tgt_device->name));
 458
 
 
 
 
 459	/*
 460	 * from now on, the writes to the srcdev are all duplicated to
 461	 * go to the tgtdev as well (refer to btrfs_map_block()).
 462	 */
 463	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 464	dev_replace->time_started = ktime_get_real_seconds();
 465	dev_replace->cursor_left = 0;
 466	dev_replace->committed_cursor_left = 0;
 467	dev_replace->cursor_left_last_write_of_item = 0;
 468	dev_replace->cursor_right = 0;
 469	dev_replace->is_valid = 1;
 470	dev_replace->item_needs_writeback = 1;
 471	atomic64_set(&dev_replace->num_write_errors, 0);
 472	atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 473	up_write(&dev_replace->rwsem);
 474
 475	ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 476	if (ret)
 477		btrfs_err(fs_info, "kobj add dev failed %d", ret);
 478
 479	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 480
 481	/* Commit dev_replace state and reserve 1 item for it. */
 482	trans = btrfs_start_transaction(root, 1);
 483	if (IS_ERR(trans)) {
 484		ret = PTR_ERR(trans);
 485		down_write(&dev_replace->rwsem);
 486		dev_replace->replace_state =
 487			BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 488		dev_replace->srcdev = NULL;
 489		dev_replace->tgtdev = NULL;
 490		up_write(&dev_replace->rwsem);
 491		goto leave;
 492	}
 493
 494	ret = btrfs_commit_transaction(trans);
 495	WARN_ON(ret);
 496
 497	/* the disk copy procedure reuses the scrub code */
 498	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 499			      btrfs_device_get_total_bytes(src_device),
 500			      &dev_replace->scrub_progress, 0, 1);
 501
 502	ret = btrfs_dev_replace_finishing(fs_info, ret);
 503	if (ret == -EINPROGRESS) {
 504		ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 505	} else if (ret != -ECANCELED) {
 506		WARN_ON(ret);
 507	}
 508
 509	return ret;
 510
 511leave:
 512	btrfs_destroy_dev_replace_tgtdev(tgt_device);
 513	return ret;
 514}
 515
 516int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 517			    struct btrfs_ioctl_dev_replace_args *args)
 518{
 519	int ret;
 520
 521	switch (args->start.cont_reading_from_srcdev_mode) {
 522	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 523	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 524		break;
 525	default:
 526		return -EINVAL;
 527	}
 528
 529	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 530	    args->start.tgtdev_name[0] == '\0')
 531		return -EINVAL;
 532
 533	ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 534					args->start.srcdevid,
 535					args->start.srcdev_name,
 536					args->start.cont_reading_from_srcdev_mode);
 537	args->result = ret;
 538	/* don't warn if EINPROGRESS, someone else might be running scrub */
 539	if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS ||
 540	    ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR)
 541		return 0;
 542
 543	return ret;
 544}
 545
 546/*
 547 * blocked until all in-flight bios operations are finished.
 548 */
 549static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 550{
 
 
 
 551	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 552	wait_event(fs_info->dev_replace.replace_wait, !percpu_counter_sum(
 553		   &fs_info->dev_replace.bio_counter));
 
 
 
 
 
 
 554}
 555
 556/*
 557 * we have removed target device, it is safe to allow new bios request.
 558 */
 559static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 560{
 561	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 562	wake_up(&fs_info->dev_replace.replace_wait);
 
 563}
 564
 565static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 566				       int scrub_ret)
 567{
 568	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 569	struct btrfs_device *tgt_device;
 570	struct btrfs_device *src_device;
 571	struct btrfs_root *root = fs_info->tree_root;
 572	u8 uuid_tmp[BTRFS_UUID_SIZE];
 573	struct btrfs_trans_handle *trans;
 574	int ret = 0;
 575
 576	/* don't allow cancel or unmount to disturb the finishing procedure */
 577	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 578
 579	down_read(&dev_replace->rwsem);
 580	/* was the operation canceled, or is it finished? */
 581	if (dev_replace->replace_state !=
 582	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 583		up_read(&dev_replace->rwsem);
 584		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 585		return 0;
 586	}
 587
 588	tgt_device = dev_replace->tgtdev;
 589	src_device = dev_replace->srcdev;
 590	up_read(&dev_replace->rwsem);
 591
 592	/*
 593	 * flush all outstanding I/O and inode extent mappings before the
 594	 * copy operation is declared as being finished
 595	 */
 596	ret = btrfs_start_delalloc_roots(fs_info, -1);
 597	if (ret) {
 598		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 599		return ret;
 600	}
 601	btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 602
 603	/*
 604	 * We have to use this loop approach because at this point src_device
 605	 * has to be available for transaction commit to complete, yet new
 606	 * chunks shouldn't be allocated on the device.
 607	 */
 608	while (1) {
 609		trans = btrfs_start_transaction(root, 0);
 610		if (IS_ERR(trans)) {
 611			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 612			return PTR_ERR(trans);
 613		}
 614		ret = btrfs_commit_transaction(trans);
 615		WARN_ON(ret);
 616
 617		/* Prevent write_all_supers() during the finishing procedure */
 618		mutex_lock(&fs_info->fs_devices->device_list_mutex);
 619		/* Prevent new chunks being allocated on the source device */
 620		mutex_lock(&fs_info->chunk_mutex);
 621
 622		if (!list_empty(&src_device->post_commit_list)) {
 623			mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 624			mutex_unlock(&fs_info->chunk_mutex);
 625		} else {
 626			break;
 627		}
 628	}
 
 
 629
 630	down_write(&dev_replace->rwsem);
 
 
 
 631	dev_replace->replace_state =
 632		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 633			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 634	dev_replace->tgtdev = NULL;
 635	dev_replace->srcdev = NULL;
 636	dev_replace->time_stopped = ktime_get_real_seconds();
 637	dev_replace->item_needs_writeback = 1;
 638
 639	/* replace old device with new one in mapping tree */
 640	if (!scrub_ret) {
 641		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 642								src_device,
 643								tgt_device);
 644	} else {
 645		if (scrub_ret != -ECANCELED)
 646			btrfs_err_in_rcu(fs_info,
 647				 "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 648				 btrfs_dev_name(src_device),
 649				 src_device->devid,
 650				 rcu_str_deref(tgt_device->name), scrub_ret);
 651		up_write(&dev_replace->rwsem);
 652		mutex_unlock(&fs_info->chunk_mutex);
 653		mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 654		btrfs_rm_dev_replace_blocked(fs_info);
 655		if (tgt_device)
 656			btrfs_destroy_dev_replace_tgtdev(tgt_device);
 657		btrfs_rm_dev_replace_unblocked(fs_info);
 658		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 659
 660		return scrub_ret;
 661	}
 662
 663	btrfs_info_in_rcu(fs_info,
 664			  "dev_replace from %s (devid %llu) to %s finished",
 665			  btrfs_dev_name(src_device),
 666			  src_device->devid,
 667			  rcu_str_deref(tgt_device->name));
 668	clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 
 669	tgt_device->devid = src_device->devid;
 670	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 
 671	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 672	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 673	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 674	btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 675	btrfs_device_set_disk_total_bytes(tgt_device,
 676					  src_device->disk_total_bytes);
 677	btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 678	tgt_device->commit_bytes_used = src_device->bytes_used;
 679
 680	btrfs_assign_next_active_device(src_device, tgt_device);
 681
 682	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 683	fs_info->fs_devices->rw_devices++;
 684
 685	up_write(&dev_replace->rwsem);
 686	btrfs_rm_dev_replace_blocked(fs_info);
 687
 688	btrfs_rm_dev_replace_remove_srcdev(src_device);
 689
 690	btrfs_rm_dev_replace_unblocked(fs_info);
 691
 692	/*
 693	 * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 694	 * update on-disk dev stats value during commit transaction
 695	 */
 696	atomic_inc(&tgt_device->dev_stats_ccnt);
 697
 698	/*
 699	 * this is again a consistent state where no dev_replace procedure
 700	 * is running, the target device is part of the filesystem, the
 701	 * source device is not part of the filesystem anymore and its 1st
 702	 * superblock is scratched out so that it is no longer marked to
 703	 * belong to this filesystem.
 704	 */
 705	mutex_unlock(&fs_info->chunk_mutex);
 706	mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 707
 708	/* replace the sysfs entry */
 709	btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 710	btrfs_rm_dev_replace_free_srcdev(src_device);
 711
 712	/* write back the superblocks */
 713	trans = btrfs_start_transaction(root, 0);
 714	if (!IS_ERR(trans))
 715		btrfs_commit_transaction(trans);
 716
 717	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 718
 719	return 0;
 720}
 721
 722static void btrfs_dev_replace_update_device_in_mapping_tree(
 723						struct btrfs_fs_info *fs_info,
 724						struct btrfs_device *srcdev,
 725						struct btrfs_device *tgtdev)
 726{
 727	struct extent_map_tree *em_tree = &fs_info->mapping_tree;
 728	struct extent_map *em;
 729	struct map_lookup *map;
 730	u64 start = 0;
 731	int i;
 732
 733	write_lock(&em_tree->lock);
 734	do {
 735		em = lookup_extent_mapping(em_tree, start, (u64)-1);
 736		if (!em)
 737			break;
 738		map = em->map_lookup;
 739		for (i = 0; i < map->num_stripes; i++)
 740			if (srcdev == map->stripes[i].dev)
 741				map->stripes[i].dev = tgtdev;
 742		start = em->start + em->len;
 743		free_extent_map(em);
 744	} while (start);
 745	write_unlock(&em_tree->lock);
 746}
 747
 748/*
 749 * Read progress of device replace status according to the state and last
 750 * stored position. The value format is the same as for
 751 * btrfs_dev_replace::progress_1000
 752 */
 753static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 754{
 755	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 756	u64 ret = 0;
 757
 758	switch (dev_replace->replace_state) {
 759	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 760	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 761		ret = 0;
 762		break;
 763	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 764		ret = 1000;
 765		break;
 766	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 767	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 768		ret = div64_u64(dev_replace->cursor_left,
 769				div_u64(btrfs_device_get_total_bytes(
 770						dev_replace->srcdev), 1000));
 771		break;
 772	}
 773
 774	return ret;
 775}
 776
 777void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 778			      struct btrfs_ioctl_dev_replace_args *args)
 779{
 780	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 781
 782	down_read(&dev_replace->rwsem);
 783	/* even if !dev_replace_is_valid, the values are good enough for
 784	 * the replace_status ioctl */
 785	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 786	args->status.replace_state = dev_replace->replace_state;
 787	args->status.time_started = dev_replace->time_started;
 788	args->status.time_stopped = dev_replace->time_stopped;
 789	args->status.num_write_errors =
 790		atomic64_read(&dev_replace->num_write_errors);
 791	args->status.num_uncorrectable_read_errors =
 792		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 793	args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 794	up_read(&dev_replace->rwsem);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 795}
 796
 797int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 798{
 799	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 800	struct btrfs_device *tgt_device = NULL;
 801	struct btrfs_device *src_device = NULL;
 802	struct btrfs_trans_handle *trans;
 803	struct btrfs_root *root = fs_info->tree_root;
 804	int result;
 805	int ret;
 806
 807	if (sb_rdonly(fs_info->sb))
 808		return -EROFS;
 809
 810	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 811	down_write(&dev_replace->rwsem);
 812	switch (dev_replace->replace_state) {
 813	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 814	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 815	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 816		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 817		up_write(&dev_replace->rwsem);
 818		break;
 819	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 820		tgt_device = dev_replace->tgtdev;
 821		src_device = dev_replace->srcdev;
 822		up_write(&dev_replace->rwsem);
 823		ret = btrfs_scrub_cancel(fs_info);
 824		if (ret < 0) {
 825			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 826		} else {
 827			result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 828			/*
 829			 * btrfs_dev_replace_finishing() will handle the
 830			 * cleanup part
 831			 */
 832			btrfs_info_in_rcu(fs_info,
 833				"dev_replace from %s (devid %llu) to %s canceled",
 834				btrfs_dev_name(src_device), src_device->devid,
 835				btrfs_dev_name(tgt_device));
 836		}
 837		break;
 838	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 839		/*
 840		 * Scrub doing the replace isn't running so we need to do the
 841		 * cleanup step of btrfs_dev_replace_finishing() here
 842		 */
 843		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 844		tgt_device = dev_replace->tgtdev;
 845		src_device = dev_replace->srcdev;
 846		dev_replace->tgtdev = NULL;
 847		dev_replace->srcdev = NULL;
 848		dev_replace->replace_state =
 849				BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 850		dev_replace->time_stopped = ktime_get_real_seconds();
 851		dev_replace->item_needs_writeback = 1;
 852
 853		up_write(&dev_replace->rwsem);
 854
 855		/* Scrub for replace must not be running in suspended state */
 856		ret = btrfs_scrub_cancel(fs_info);
 857		ASSERT(ret != -ENOTCONN);
 858
 859		trans = btrfs_start_transaction(root, 0);
 860		if (IS_ERR(trans)) {
 861			mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 862			return PTR_ERR(trans);
 863		}
 864		ret = btrfs_commit_transaction(trans);
 865		WARN_ON(ret);
 866
 867		btrfs_info_in_rcu(fs_info,
 868		"suspended dev_replace from %s (devid %llu) to %s canceled",
 869			btrfs_dev_name(src_device), src_device->devid,
 870			btrfs_dev_name(tgt_device));
 871
 872		if (tgt_device)
 873			btrfs_destroy_dev_replace_tgtdev(tgt_device);
 874		break;
 875	default:
 876		up_write(&dev_replace->rwsem);
 877		result = -EINVAL;
 878	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 879
 
 880	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 881	return result;
 882}
 883
 884void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 885{
 886	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 887
 888	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 889	down_write(&dev_replace->rwsem);
 890
 891	switch (dev_replace->replace_state) {
 892	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 893	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 894	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 895	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 896		break;
 897	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 898		dev_replace->replace_state =
 899			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 900		dev_replace->time_stopped = ktime_get_real_seconds();
 901		dev_replace->item_needs_writeback = 1;
 902		btrfs_info(fs_info, "suspending dev_replace for unmount");
 903		break;
 904	}
 905
 906	up_write(&dev_replace->rwsem);
 907	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 908}
 909
 910/* resume dev_replace procedure that was interrupted by unmount */
 911int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 912{
 913	struct task_struct *task;
 914	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 915
 916	down_write(&dev_replace->rwsem);
 917
 918	switch (dev_replace->replace_state) {
 919	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 920	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 921	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 922		up_write(&dev_replace->rwsem);
 923		return 0;
 924	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 925		break;
 926	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 927		dev_replace->replace_state =
 928			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 929		break;
 930	}
 931	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 
 932		btrfs_info(fs_info,
 933			   "cannot continue dev_replace, tgtdev is missing");
 934		btrfs_info(fs_info,
 935			   "you may cancel the operation after 'mount -o degraded'");
 936		dev_replace->replace_state =
 937					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 938		up_write(&dev_replace->rwsem);
 939		return 0;
 940	}
 941	up_write(&dev_replace->rwsem);
 942
 943	/*
 944	 * This could collide with a paused balance, but the exclusive op logic
 945	 * should never allow both to start and pause. We don't want to allow
 946	 * dev-replace to start anyway.
 947	 */
 948	if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
 949		down_write(&dev_replace->rwsem);
 950		dev_replace->replace_state =
 951					BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 952		up_write(&dev_replace->rwsem);
 953		btrfs_info(fs_info,
 954		"cannot resume dev-replace, other exclusive operation running");
 955		return 0;
 956	}
 
 957
 
 
 958	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 959	return PTR_ERR_OR_ZERO(task);
 960}
 961
 962static int btrfs_dev_replace_kthread(void *data)
 963{
 964	struct btrfs_fs_info *fs_info = data;
 965	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 
 966	u64 progress;
 967	int ret;
 968
 969	progress = btrfs_dev_replace_progress(fs_info);
 970	progress = div_u64(progress, 10);
 971	btrfs_info_in_rcu(fs_info,
 972		"continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 973		btrfs_dev_name(dev_replace->srcdev),
 974		dev_replace->srcdev->devid,
 975		btrfs_dev_name(dev_replace->tgtdev),
 976		(unsigned int)progress);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 977
 978	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 979			      dev_replace->committed_cursor_left,
 980			      btrfs_device_get_total_bytes(dev_replace->srcdev),
 981			      &dev_replace->scrub_progress, 0, 1);
 982	ret = btrfs_dev_replace_finishing(fs_info, ret);
 983	WARN_ON(ret && ret != -ECANCELED);
 984
 985	clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 986	return 0;
 987}
 988
 989int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 990{
 991	if (!dev_replace->is_valid)
 992		return 0;
 993
 994	switch (dev_replace->replace_state) {
 995	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 996	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 997	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 998		return 0;
 999	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
1000	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
1001		/*
1002		 * return true even if tgtdev is missing (this is
1003		 * something that can happen if the dev_replace
1004		 * procedure is suspended by an umount and then
1005		 * the tgtdev is missing (or "btrfs dev scan") was
1006		 * not called and the filesystem is remounted
1007		 * in degraded state. This does not stop the
1008		 * dev_replace procedure. It needs to be canceled
1009		 * manually if the cancellation is wanted.
1010		 */
1011		break;
1012	}
1013	return 1;
1014}
1015
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1016void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1017{
1018	percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1019}
1020
1021void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1022{
1023	percpu_counter_sub(&fs_info->dev_replace.bio_counter, amount);
1024	cond_wake_up_nomb(&fs_info->dev_replace.replace_wait);
 
 
1025}
1026
1027void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1028{
1029	while (1) {
1030		percpu_counter_inc(&fs_info->dev_replace.bio_counter);
1031		if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1032				     &fs_info->fs_state)))
1033			break;
1034
1035		btrfs_bio_counter_dec(fs_info);
1036		wait_event(fs_info->dev_replace.replace_wait,
1037			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1038				     &fs_info->fs_state));
 
1039	}
 
1040}
v3.15
 
  1/*
  2 * Copyright (C) STRATO AG 2012.  All rights reserved.
  3 *
  4 * This program is free software; you can redistribute it and/or
  5 * modify it under the terms of the GNU General Public
  6 * License v2 as published by the Free Software Foundation.
  7 *
  8 * This program is distributed in the hope that it will be useful,
  9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 11 * General Public License for more details.
 12 *
 13 * You should have received a copy of the GNU General Public
 14 * License along with this program; if not, write to the
 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
 16 * Boston, MA 021110-1307, USA.
 17 */
 
 18#include <linux/sched.h>
 19#include <linux/bio.h>
 20#include <linux/slab.h>
 21#include <linux/buffer_head.h>
 22#include <linux/blkdev.h>
 23#include <linux/random.h>
 24#include <linux/iocontext.h>
 25#include <linux/capability.h>
 26#include <linux/kthread.h>
 27#include <linux/math64.h>
 28#include <asm/div64.h>
 29#include "ctree.h"
 30#include "extent_map.h"
 31#include "disk-io.h"
 32#include "transaction.h"
 33#include "print-tree.h"
 34#include "volumes.h"
 35#include "async-thread.h"
 36#include "check-integrity.h"
 37#include "rcu-string.h"
 38#include "dev-replace.h"
 
 39
 40static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 41				       int scrub_ret);
 42static void btrfs_dev_replace_update_device_in_mapping_tree(
 43						struct btrfs_fs_info *fs_info,
 44						struct btrfs_device *srcdev,
 45						struct btrfs_device *tgtdev);
 46static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
 47					 char *srcdev_name,
 48					 struct btrfs_device **device);
 49static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info);
 50static int btrfs_dev_replace_kthread(void *data);
 51static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info);
 52
 53
 54int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
 55{
 56	struct btrfs_key key;
 57	struct btrfs_root *dev_root = fs_info->dev_root;
 58	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 59	struct extent_buffer *eb;
 60	int slot;
 61	int ret = 0;
 62	struct btrfs_path *path = NULL;
 63	int item_size;
 64	struct btrfs_dev_replace_item *ptr;
 65	u64 src_devid;
 66
 67	path = btrfs_alloc_path();
 68	if (!path) {
 69		ret = -ENOMEM;
 70		goto out;
 71	}
 72
 73	key.objectid = 0;
 74	key.type = BTRFS_DEV_REPLACE_KEY;
 75	key.offset = 0;
 76	ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
 77	if (ret) {
 78no_valid_dev_replace_entry_found:
 79		ret = 0;
 80		dev_replace->replace_state =
 81			BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
 82		dev_replace->cont_reading_from_srcdev_mode =
 83		    BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
 84		dev_replace->replace_state = 0;
 85		dev_replace->time_started = 0;
 86		dev_replace->time_stopped = 0;
 87		atomic64_set(&dev_replace->num_write_errors, 0);
 88		atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 89		dev_replace->cursor_left = 0;
 90		dev_replace->committed_cursor_left = 0;
 91		dev_replace->cursor_left_last_write_of_item = 0;
 92		dev_replace->cursor_right = 0;
 93		dev_replace->srcdev = NULL;
 94		dev_replace->tgtdev = NULL;
 95		dev_replace->is_valid = 0;
 96		dev_replace->item_needs_writeback = 0;
 97		goto out;
 98	}
 99	slot = path->slots[0];
100	eb = path->nodes[0];
101	item_size = btrfs_item_size_nr(eb, slot);
102	ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
103
104	if (item_size != sizeof(struct btrfs_dev_replace_item)) {
105		btrfs_warn(fs_info,
106			"dev_replace entry found has unexpected size, ignore entry");
107		goto no_valid_dev_replace_entry_found;
108	}
109
110	src_devid = btrfs_dev_replace_src_devid(eb, ptr);
111	dev_replace->cont_reading_from_srcdev_mode =
112		btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
113	dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
114	dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
115	dev_replace->time_stopped =
116		btrfs_dev_replace_time_stopped(eb, ptr);
117	atomic64_set(&dev_replace->num_write_errors,
118		     btrfs_dev_replace_num_write_errors(eb, ptr));
119	atomic64_set(&dev_replace->num_uncorrectable_read_errors,
120		     btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
121	dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
122	dev_replace->committed_cursor_left = dev_replace->cursor_left;
123	dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
124	dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
125	dev_replace->is_valid = 1;
126
127	dev_replace->item_needs_writeback = 0;
128	switch (dev_replace->replace_state) {
129	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
130	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
131	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
132		dev_replace->srcdev = NULL;
133		dev_replace->tgtdev = NULL;
134		break;
135	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
136	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
137		dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
138							NULL, NULL);
139		dev_replace->tgtdev = btrfs_find_device(fs_info,
140							BTRFS_DEV_REPLACE_DEVID,
141							NULL, NULL);
142		/*
143		 * allow 'btrfs dev replace_cancel' if src/tgt device is
144		 * missing
145		 */
146		if (!dev_replace->srcdev &&
147		    !btrfs_test_opt(dev_root, DEGRADED)) {
148			ret = -EIO;
149			btrfs_warn(fs_info,
150			   "cannot mount because device replace operation is ongoing and");
151			btrfs_warn(fs_info,
152			   "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
153			   src_devid);
154		}
155		if (!dev_replace->tgtdev &&
156		    !btrfs_test_opt(dev_root, DEGRADED)) {
157			ret = -EIO;
158			btrfs_warn(fs_info,
159			   "cannot mount because device replace operation is ongoing and");
160			btrfs_warn(fs_info,
161			   "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
162				BTRFS_DEV_REPLACE_DEVID);
163		}
164		if (dev_replace->tgtdev) {
165			if (dev_replace->srcdev) {
166				dev_replace->tgtdev->total_bytes =
167					dev_replace->srcdev->total_bytes;
168				dev_replace->tgtdev->disk_total_bytes =
169					dev_replace->srcdev->disk_total_bytes;
 
 
170				dev_replace->tgtdev->bytes_used =
171					dev_replace->srcdev->bytes_used;
 
 
172			}
173			dev_replace->tgtdev->is_tgtdev_for_dev_replace = 1;
174			btrfs_init_dev_replace_tgtdev_for_resume(fs_info,
175				dev_replace->tgtdev);
 
 
 
 
 
 
 
176		}
177		break;
178	}
179
180out:
181	if (path)
182		btrfs_free_path(path);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
183	return ret;
184}
185
186/*
187 * called from commit_transaction. Writes changed device replace state to
188 * disk.
189 */
190int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
191			  struct btrfs_fs_info *fs_info)
192{
 
193	int ret;
194	struct btrfs_root *dev_root = fs_info->dev_root;
195	struct btrfs_path *path;
196	struct btrfs_key key;
197	struct extent_buffer *eb;
198	struct btrfs_dev_replace_item *ptr;
199	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
200
201	btrfs_dev_replace_lock(dev_replace);
202	if (!dev_replace->is_valid ||
203	    !dev_replace->item_needs_writeback) {
204		btrfs_dev_replace_unlock(dev_replace);
205		return 0;
206	}
207	btrfs_dev_replace_unlock(dev_replace);
208
209	key.objectid = 0;
210	key.type = BTRFS_DEV_REPLACE_KEY;
211	key.offset = 0;
212
213	path = btrfs_alloc_path();
214	if (!path) {
215		ret = -ENOMEM;
216		goto out;
217	}
218	ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
219	if (ret < 0) {
220		btrfs_warn(fs_info, "error %d while searching for dev_replace item!",
221			ret);
 
222		goto out;
223	}
224
225	if (ret == 0 &&
226	    btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
227		/*
228		 * need to delete old one and insert a new one.
229		 * Since no attempt is made to recover any old state, if the
230		 * dev_replace state is 'running', the data on the target
231		 * drive is lost.
232		 * It would be possible to recover the state: just make sure
233		 * that the beginning of the item is never changed and always
234		 * contains all the essential information. Then read this
235		 * minimal set of information and use it as a base for the
236		 * new state.
237		 */
238		ret = btrfs_del_item(trans, dev_root, path);
239		if (ret != 0) {
240			btrfs_warn(fs_info, "delete too small dev_replace item failed %d!",
241				ret);
 
242			goto out;
243		}
244		ret = 1;
245	}
246
247	if (ret == 1) {
248		/* need to insert a new item */
249		btrfs_release_path(path);
250		ret = btrfs_insert_empty_item(trans, dev_root, path,
251					      &key, sizeof(*ptr));
252		if (ret < 0) {
253			btrfs_warn(fs_info, "insert dev_replace item failed %d!",
254				ret);
255			goto out;
256		}
257	}
258
259	eb = path->nodes[0];
260	ptr = btrfs_item_ptr(eb, path->slots[0],
261			     struct btrfs_dev_replace_item);
262
263	btrfs_dev_replace_lock(dev_replace);
264	if (dev_replace->srcdev)
265		btrfs_set_dev_replace_src_devid(eb, ptr,
266			dev_replace->srcdev->devid);
267	else
268		btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
269	btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
270		dev_replace->cont_reading_from_srcdev_mode);
271	btrfs_set_dev_replace_replace_state(eb, ptr,
272		dev_replace->replace_state);
273	btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
274	btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
275	btrfs_set_dev_replace_num_write_errors(eb, ptr,
276		atomic64_read(&dev_replace->num_write_errors));
277	btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
278		atomic64_read(&dev_replace->num_uncorrectable_read_errors));
279	dev_replace->cursor_left_last_write_of_item =
280		dev_replace->cursor_left;
281	btrfs_set_dev_replace_cursor_left(eb, ptr,
282		dev_replace->cursor_left_last_write_of_item);
283	btrfs_set_dev_replace_cursor_right(eb, ptr,
284		dev_replace->cursor_right);
285	dev_replace->item_needs_writeback = 0;
286	btrfs_dev_replace_unlock(dev_replace);
287
288	btrfs_mark_buffer_dirty(eb);
289
290out:
291	btrfs_free_path(path);
292
293	return ret;
294}
295
296void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
297{
298	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
299
300	dev_replace->committed_cursor_left =
301		dev_replace->cursor_left_last_write_of_item;
302}
303
304int btrfs_dev_replace_start(struct btrfs_root *root,
305			    struct btrfs_ioctl_dev_replace_args *args)
 
306{
 
307	struct btrfs_trans_handle *trans;
308	struct btrfs_fs_info *fs_info = root->fs_info;
309	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
310	int ret;
311	struct btrfs_device *tgt_device = NULL;
312	struct btrfs_device *src_device = NULL;
313
314	if (btrfs_fs_incompat(fs_info, RAID56)) {
315		btrfs_warn(fs_info, "dev_replace cannot yet handle RAID5/RAID6");
316		return -EINVAL;
 
 
 
 
 
 
 
317	}
318
319	switch (args->start.cont_reading_from_srcdev_mode) {
320	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
321	case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
322		break;
323	default:
324		return -EINVAL;
 
 
 
 
 
325	}
326
327	if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
328	    args->start.tgtdev_name[0] == '\0')
329		return -EINVAL;
330
331	mutex_lock(&fs_info->volume_mutex);
332	ret = btrfs_init_dev_replace_tgtdev(root, args->start.tgtdev_name,
333					    &tgt_device);
334	if (ret) {
335		btrfs_err(fs_info, "target device %s is invalid!",
336		       args->start.tgtdev_name);
337		mutex_unlock(&fs_info->volume_mutex);
338		return -EINVAL;
339	}
340
341	ret = btrfs_dev_replace_find_srcdev(root, args->start.srcdevid,
342					    args->start.srcdev_name,
343					    &src_device);
344	mutex_unlock(&fs_info->volume_mutex);
345	if (ret) {
346		ret = -EINVAL;
347		goto leave_no_lock;
348	}
349
350	if (tgt_device->total_bytes < src_device->total_bytes) {
351		btrfs_err(fs_info, "target device is smaller than source device!");
352		ret = -EINVAL;
353		goto leave_no_lock;
354	}
355
356	btrfs_dev_replace_lock(dev_replace);
357	switch (dev_replace->replace_state) {
358	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
359	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
360	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
361		break;
362	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
363	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
364		args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 
 
365		goto leave;
366	}
367
368	dev_replace->cont_reading_from_srcdev_mode =
369		args->start.cont_reading_from_srcdev_mode;
370	WARN_ON(!src_device);
371	dev_replace->srcdev = src_device;
372	WARN_ON(!tgt_device);
373	dev_replace->tgtdev = tgt_device;
374
375	printk_in_rcu(KERN_INFO
376		      "BTRFS: dev_replace from %s (devid %llu) to %s started\n",
377		      src_device->missing ? "<missing disk>" :
378		        rcu_str_deref(src_device->name),
379		      src_device->devid,
380		      rcu_str_deref(tgt_device->name));
381
382	tgt_device->total_bytes = src_device->total_bytes;
383	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
384	tgt_device->bytes_used = src_device->bytes_used;
385
386	/*
387	 * from now on, the writes to the srcdev are all duplicated to
388	 * go to the tgtdev as well (refer to btrfs_map_block()).
389	 */
390	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
391	dev_replace->time_started = get_seconds();
392	dev_replace->cursor_left = 0;
393	dev_replace->committed_cursor_left = 0;
394	dev_replace->cursor_left_last_write_of_item = 0;
395	dev_replace->cursor_right = 0;
396	dev_replace->is_valid = 1;
397	dev_replace->item_needs_writeback = 1;
398	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
399	btrfs_dev_replace_unlock(dev_replace);
 
 
 
 
 
400
401	btrfs_wait_ordered_roots(root->fs_info, -1);
402
403	/* force writing the updated state information to disk */
404	trans = btrfs_start_transaction(root, 0);
405	if (IS_ERR(trans)) {
406		ret = PTR_ERR(trans);
407		btrfs_dev_replace_lock(dev_replace);
 
 
 
 
 
408		goto leave;
409	}
410
411	ret = btrfs_commit_transaction(trans, root);
412	WARN_ON(ret);
413
414	/* the disk copy procedure reuses the scrub code */
415	ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
416			      src_device->total_bytes,
417			      &dev_replace->scrub_progress, 0, 1);
418
419	ret = btrfs_dev_replace_finishing(root->fs_info, ret);
420	WARN_ON(ret);
 
 
 
 
421
422	return 0;
423
424leave:
425	dev_replace->srcdev = NULL;
426	dev_replace->tgtdev = NULL;
427	btrfs_dev_replace_unlock(dev_replace);
428leave_no_lock:
429	if (tgt_device)
430		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
431	return ret;
432}
433
434/*
435 * blocked until all flighting bios are finished.
436 */
437static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
438{
439	s64 writers;
440	DEFINE_WAIT(wait);
441
442	set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
443	do {
444		prepare_to_wait(&fs_info->replace_wait, &wait,
445				TASK_UNINTERRUPTIBLE);
446		writers = percpu_counter_sum(&fs_info->bio_counter);
447		if (writers)
448			schedule();
449		finish_wait(&fs_info->replace_wait, &wait);
450	} while (writers);
451}
452
453/*
454 * we have removed target device, it is safe to allow new bios request.
455 */
456static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
457{
458	clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
459	if (waitqueue_active(&fs_info->replace_wait))
460		wake_up(&fs_info->replace_wait);
461}
462
463static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
464				       int scrub_ret)
465{
466	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
467	struct btrfs_device *tgt_device;
468	struct btrfs_device *src_device;
469	struct btrfs_root *root = fs_info->tree_root;
470	u8 uuid_tmp[BTRFS_UUID_SIZE];
471	struct btrfs_trans_handle *trans;
472	int ret = 0;
473
474	/* don't allow cancel or unmount to disturb the finishing procedure */
475	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
476
477	btrfs_dev_replace_lock(dev_replace);
478	/* was the operation canceled, or is it finished? */
479	if (dev_replace->replace_state !=
480	    BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
481		btrfs_dev_replace_unlock(dev_replace);
482		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
483		return 0;
484	}
485
486	tgt_device = dev_replace->tgtdev;
487	src_device = dev_replace->srcdev;
488	btrfs_dev_replace_unlock(dev_replace);
489
490	/*
491	 * flush all outstanding I/O and inode extent mappings before the
492	 * copy operation is declared as being finished
493	 */
494	ret = btrfs_start_delalloc_roots(root->fs_info, 0, -1);
495	if (ret) {
496		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
497		return ret;
498	}
499	btrfs_wait_ordered_roots(root->fs_info, -1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
500
501	trans = btrfs_start_transaction(root, 0);
502	if (IS_ERR(trans)) {
503		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
504		return PTR_ERR(trans);
 
 
 
 
 
 
 
505	}
506	ret = btrfs_commit_transaction(trans, root);
507	WARN_ON(ret);
508
509	/* keep away write_all_supers() during the finishing procedure */
510	mutex_lock(&root->fs_info->chunk_mutex);
511	mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
512	btrfs_dev_replace_lock(dev_replace);
513	dev_replace->replace_state =
514		scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
515			  : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
516	dev_replace->tgtdev = NULL;
517	dev_replace->srcdev = NULL;
518	dev_replace->time_stopped = get_seconds();
519	dev_replace->item_needs_writeback = 1;
520
521	/* replace old device with new one in mapping tree */
522	if (!scrub_ret) {
523		btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
524								src_device,
525								tgt_device);
526	} else {
527		printk_in_rcu(KERN_ERR
528			      "BTRFS: btrfs_scrub_dev(%s, %llu, %s) failed %d\n",
529			      src_device->missing ? "<missing disk>" :
530			        rcu_str_deref(src_device->name),
531			      src_device->devid,
532			      rcu_str_deref(tgt_device->name), scrub_ret);
533		btrfs_dev_replace_unlock(dev_replace);
534		mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
535		mutex_unlock(&root->fs_info->chunk_mutex);
 
536		if (tgt_device)
537			btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 
538		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
539
540		return 0;
541	}
542
543	printk_in_rcu(KERN_INFO
544		      "BTRFS: dev_replace from %s (devid %llu) to %s) finished\n",
545		      src_device->missing ? "<missing disk>" :
546		        rcu_str_deref(src_device->name),
547		      src_device->devid,
548		      rcu_str_deref(tgt_device->name));
549	tgt_device->is_tgtdev_for_dev_replace = 0;
550	tgt_device->devid = src_device->devid;
551	src_device->devid = BTRFS_DEV_REPLACE_DEVID;
552	tgt_device->bytes_used = src_device->bytes_used;
553	memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
554	memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
555	memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
556	tgt_device->total_bytes = src_device->total_bytes;
557	tgt_device->disk_total_bytes = src_device->disk_total_bytes;
558	tgt_device->bytes_used = src_device->bytes_used;
559	if (fs_info->sb->s_bdev == src_device->bdev)
560		fs_info->sb->s_bdev = tgt_device->bdev;
561	if (fs_info->fs_devices->latest_bdev == src_device->bdev)
562		fs_info->fs_devices->latest_bdev = tgt_device->bdev;
 
563	list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 
564
 
565	btrfs_rm_dev_replace_blocked(fs_info);
566
567	btrfs_rm_dev_replace_srcdev(fs_info, src_device);
568
569	btrfs_rm_dev_replace_unblocked(fs_info);
570
571	/*
 
 
 
 
 
 
572	 * this is again a consistent state where no dev_replace procedure
573	 * is running, the target device is part of the filesystem, the
574	 * source device is not part of the filesystem anymore and its 1st
575	 * superblock is scratched out so that it is no longer marked to
576	 * belong to this filesystem.
577	 */
578	btrfs_dev_replace_unlock(dev_replace);
579	mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
580	mutex_unlock(&root->fs_info->chunk_mutex);
 
 
 
581
582	/* write back the superblocks */
583	trans = btrfs_start_transaction(root, 0);
584	if (!IS_ERR(trans))
585		btrfs_commit_transaction(trans, root);
586
587	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
588
589	return 0;
590}
591
592static void btrfs_dev_replace_update_device_in_mapping_tree(
593						struct btrfs_fs_info *fs_info,
594						struct btrfs_device *srcdev,
595						struct btrfs_device *tgtdev)
596{
597	struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
598	struct extent_map *em;
599	struct map_lookup *map;
600	u64 start = 0;
601	int i;
602
603	write_lock(&em_tree->lock);
604	do {
605		em = lookup_extent_mapping(em_tree, start, (u64)-1);
606		if (!em)
607			break;
608		map = (struct map_lookup *)em->bdev;
609		for (i = 0; i < map->num_stripes; i++)
610			if (srcdev == map->stripes[i].dev)
611				map->stripes[i].dev = tgtdev;
612		start = em->start + em->len;
613		free_extent_map(em);
614	} while (start);
615	write_unlock(&em_tree->lock);
616}
617
618static int btrfs_dev_replace_find_srcdev(struct btrfs_root *root, u64 srcdevid,
619					 char *srcdev_name,
620					 struct btrfs_device **device)
 
 
 
621{
622	int ret;
 
623
624	if (srcdevid) {
 
 
625		ret = 0;
626		*device = btrfs_find_device(root->fs_info, srcdevid, NULL,
627					    NULL);
628		if (!*device)
629			ret = -ENOENT;
630	} else {
631		ret = btrfs_find_device_missing_or_by_path(root, srcdev_name,
632							   device);
 
 
 
633	}
 
634	return ret;
635}
636
637void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
638			      struct btrfs_ioctl_dev_replace_args *args)
639{
640	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
641
642	btrfs_dev_replace_lock(dev_replace);
643	/* even if !dev_replace_is_valid, the values are good enough for
644	 * the replace_status ioctl */
645	args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
646	args->status.replace_state = dev_replace->replace_state;
647	args->status.time_started = dev_replace->time_started;
648	args->status.time_stopped = dev_replace->time_stopped;
649	args->status.num_write_errors =
650		atomic64_read(&dev_replace->num_write_errors);
651	args->status.num_uncorrectable_read_errors =
652		atomic64_read(&dev_replace->num_uncorrectable_read_errors);
653	switch (dev_replace->replace_state) {
654	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
655	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
656		args->status.progress_1000 = 0;
657		break;
658	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
659		args->status.progress_1000 = 1000;
660		break;
661	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
662	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
663		args->status.progress_1000 = div64_u64(dev_replace->cursor_left,
664			div64_u64(dev_replace->srcdev->total_bytes, 1000));
665		break;
666	}
667	btrfs_dev_replace_unlock(dev_replace);
668}
669
670int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info,
671			     struct btrfs_ioctl_dev_replace_args *args)
672{
673	args->result = __btrfs_dev_replace_cancel(fs_info);
674	return 0;
675}
676
677static u64 __btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
678{
679	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
680	struct btrfs_device *tgt_device = NULL;
 
681	struct btrfs_trans_handle *trans;
682	struct btrfs_root *root = fs_info->tree_root;
683	u64 result;
684	int ret;
685
686	if (fs_info->sb->s_flags & MS_RDONLY)
687		return -EROFS;
688
689	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
690	btrfs_dev_replace_lock(dev_replace);
691	switch (dev_replace->replace_state) {
692	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
693	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
694	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
695		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
696		btrfs_dev_replace_unlock(dev_replace);
697		goto leave;
698	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
699	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 
 
 
 
700		result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
701		tgt_device = dev_replace->tgtdev;
 
702		dev_replace->tgtdev = NULL;
703		dev_replace->srcdev = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
704		break;
 
 
 
705	}
706	dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
707	dev_replace->time_stopped = get_seconds();
708	dev_replace->item_needs_writeback = 1;
709	btrfs_dev_replace_unlock(dev_replace);
710	btrfs_scrub_cancel(fs_info);
711
712	trans = btrfs_start_transaction(root, 0);
713	if (IS_ERR(trans)) {
714		mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
715		return PTR_ERR(trans);
716	}
717	ret = btrfs_commit_transaction(trans, root);
718	WARN_ON(ret);
719	if (tgt_device)
720		btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
721
722leave:
723	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
724	return result;
725}
726
727void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
728{
729	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
730
731	mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
732	btrfs_dev_replace_lock(dev_replace);
 
733	switch (dev_replace->replace_state) {
734	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
735	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
736	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
737	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
738		break;
739	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
740		dev_replace->replace_state =
741			BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
742		dev_replace->time_stopped = get_seconds();
743		dev_replace->item_needs_writeback = 1;
744		btrfs_info(fs_info, "suspending dev_replace for unmount");
745		break;
746	}
747
748	btrfs_dev_replace_unlock(dev_replace);
749	mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
750}
751
752/* resume dev_replace procedure that was interrupted by unmount */
753int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
754{
755	struct task_struct *task;
756	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
757
758	btrfs_dev_replace_lock(dev_replace);
 
759	switch (dev_replace->replace_state) {
760	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
761	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
762	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
763		btrfs_dev_replace_unlock(dev_replace);
764		return 0;
765	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
766		break;
767	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
768		dev_replace->replace_state =
769			BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
770		break;
771	}
772	if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
773		btrfs_info(fs_info, "cannot continue dev_replace, tgtdev is missing");
774		btrfs_info(fs_info,
775			"you may cancel the operation after 'mount -o degraded'");
776		btrfs_dev_replace_unlock(dev_replace);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
777		return 0;
778	}
779	btrfs_dev_replace_unlock(dev_replace);
780
781	WARN_ON(atomic_xchg(
782		&fs_info->mutually_exclusive_operation_running, 1));
783	task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
784	return PTR_ERR_OR_ZERO(task);
785}
786
787static int btrfs_dev_replace_kthread(void *data)
788{
789	struct btrfs_fs_info *fs_info = data;
790	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
791	struct btrfs_ioctl_dev_replace_args *status_args;
792	u64 progress;
 
793
794	status_args = kzalloc(sizeof(*status_args), GFP_NOFS);
795	if (status_args) {
796		btrfs_dev_replace_status(fs_info, status_args);
797		progress = status_args->status.progress_1000;
798		kfree(status_args);
799		do_div(progress, 10);
800		printk_in_rcu(KERN_INFO
801			"BTRFS: continuing dev_replace from %s (devid %llu) to %s @%u%%\n",
802			dev_replace->srcdev->missing ? "<missing disk>" :
803			rcu_str_deref(dev_replace->srcdev->name),
804			dev_replace->srcdev->devid,
805			dev_replace->tgtdev ?
806			rcu_str_deref(dev_replace->tgtdev->name) :
807			"<missing target disk>",
808			(unsigned int)progress);
809	}
810	btrfs_dev_replace_continue_on_mount(fs_info);
811	atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
812
813	return 0;
814}
815
816static int btrfs_dev_replace_continue_on_mount(struct btrfs_fs_info *fs_info)
817{
818	struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
819	int ret;
820
821	ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
822			      dev_replace->committed_cursor_left,
823			      dev_replace->srcdev->total_bytes,
824			      &dev_replace->scrub_progress, 0, 1);
825	ret = btrfs_dev_replace_finishing(fs_info, ret);
826	WARN_ON(ret);
 
 
827	return 0;
828}
829
830int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
831{
832	if (!dev_replace->is_valid)
833		return 0;
834
835	switch (dev_replace->replace_state) {
836	case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
837	case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
838	case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
839		return 0;
840	case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
841	case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
842		/*
843		 * return true even if tgtdev is missing (this is
844		 * something that can happen if the dev_replace
845		 * procedure is suspended by an umount and then
846		 * the tgtdev is missing (or "btrfs dev scan") was
847		 * not called and the the filesystem is remounted
848		 * in degraded state. This does not stop the
849		 * dev_replace procedure. It needs to be canceled
850		 * manually if the cancelation is wanted.
851		 */
852		break;
853	}
854	return 1;
855}
856
857void btrfs_dev_replace_lock(struct btrfs_dev_replace *dev_replace)
858{
859	/* the beginning is just an optimization for the typical case */
860	if (atomic_read(&dev_replace->nesting_level) == 0) {
861acquire_lock:
862		/* this is not a nested case where the same thread
863		 * is trying to acqurire the same lock twice */
864		mutex_lock(&dev_replace->lock);
865		mutex_lock(&dev_replace->lock_management_lock);
866		dev_replace->lock_owner = current->pid;
867		atomic_inc(&dev_replace->nesting_level);
868		mutex_unlock(&dev_replace->lock_management_lock);
869		return;
870	}
871
872	mutex_lock(&dev_replace->lock_management_lock);
873	if (atomic_read(&dev_replace->nesting_level) > 0 &&
874	    dev_replace->lock_owner == current->pid) {
875		WARN_ON(!mutex_is_locked(&dev_replace->lock));
876		atomic_inc(&dev_replace->nesting_level);
877		mutex_unlock(&dev_replace->lock_management_lock);
878		return;
879	}
880
881	mutex_unlock(&dev_replace->lock_management_lock);
882	goto acquire_lock;
883}
884
885void btrfs_dev_replace_unlock(struct btrfs_dev_replace *dev_replace)
886{
887	WARN_ON(!mutex_is_locked(&dev_replace->lock));
888	mutex_lock(&dev_replace->lock_management_lock);
889	WARN_ON(atomic_read(&dev_replace->nesting_level) < 1);
890	WARN_ON(dev_replace->lock_owner != current->pid);
891	atomic_dec(&dev_replace->nesting_level);
892	if (atomic_read(&dev_replace->nesting_level) == 0) {
893		dev_replace->lock_owner = 0;
894		mutex_unlock(&dev_replace->lock_management_lock);
895		mutex_unlock(&dev_replace->lock);
896	} else {
897		mutex_unlock(&dev_replace->lock_management_lock);
898	}
899}
900
901void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
902{
903	percpu_counter_inc(&fs_info->bio_counter);
904}
905
906void btrfs_bio_counter_dec(struct btrfs_fs_info *fs_info)
907{
908	percpu_counter_dec(&fs_info->bio_counter);
909
910	if (waitqueue_active(&fs_info->replace_wait))
911		wake_up(&fs_info->replace_wait);
912}
913
914void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
915{
916	DEFINE_WAIT(wait);
917again:
918	percpu_counter_inc(&fs_info->bio_counter);
919	if (test_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state)) {
 
 
920		btrfs_bio_counter_dec(fs_info);
921		wait_event(fs_info->replace_wait,
922			   !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
923				     &fs_info->fs_state));
924		goto again;
925	}
926
927}