fs/btrfs/dev-replace.c

   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 "ctree.h"
  13 #include "extent_map.h"
  14 #include "disk-io.h"
  15 #include "transaction.h"
  16 #include "print-tree.h"
  17 #include "volumes.h"
  18 #include "async-thread.h"
  19 #include "check-integrity.h"
  20 #include "rcu-string.h"
  21 #include "dev-replace.h"
  22 #include "sysfs.h"
  23
  24 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  25                                        int scrub_ret);
  26 static void btrfs_dev_replace_update_device_in_mapping_tree(
  27                                                 struct btrfs_fs_info *fs_info,
  28                                                 struct btrfs_device *srcdev,
  29                                                 struct btrfs_device *tgtdev);
  30 static int btrfs_dev_replace_kthread(void *data);
  31
  32 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  33 {
  34         struct btrfs_key key;
  35         struct btrfs_root *dev_root = fs_info->dev_root;
  36         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  37         struct extent_buffer *eb;
  38         int slot;
  39         int ret = 0;
  40         struct btrfs_path *path = NULL;
  41         int item_size;
  42         struct btrfs_dev_replace_item *ptr;
  43         u64 src_devid;
  44
  45         path = btrfs_alloc_path();
  46         if (!path) {
  47                 ret = -ENOMEM;
  48                 goto out;
  49         }
  50
  51         key.objectid = 0;
  52         key.type = BTRFS_DEV_REPLACE_KEY;
  53         key.offset = 0;
  54         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  55         if (ret) {
  56 no_valid_dev_replace_entry_found:
  57                 ret = 0;
  58                 dev_replace->replace_state =
  59                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  60                 dev_replace->cont_reading_from_srcdev_mode =
  61                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  62                 dev_replace->replace_state = 0;
  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, src_devid,
 116                                                         NULL, NULL);
 117                 dev_replace->tgtdev = btrfs_find_device(fs_info,
 118                                                         BTRFS_DEV_REPLACE_DEVID,
 119                                                         NULL, NULL);
 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
 169 out:
 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  */
 180 static 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         filemap_write_and_wait(bdev->bd_inode->i_mapping);
 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         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 242         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 243         device->generation = 0;
 244         device->io_width = fs_info->sectorsize;
 245         device->io_align = fs_info->sectorsize;
 246         device->sector_size = fs_info->sectorsize;
 247         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
 248         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
 249         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
 250         device->commit_total_bytes = srcdev->commit_total_bytes;
 251         device->commit_bytes_used = device->bytes_used;
 252         device->fs_info = fs_info;
 253         device->bdev = bdev;
 254         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 255         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
 256         device->mode = FMODE_EXCL;
 257         device->dev_stats_valid = 1;
 258         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
 259         device->fs_devices = fs_info->fs_devices;
 260         list_add(&device->dev_list, &fs_info->fs_devices->devices);
 261         fs_info->fs_devices->num_devices++;
 262         fs_info->fs_devices->open_devices++;
 263         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 264
 265         *device_out = device;
 266         return 0;
 267
 268 error:
 269         blkdev_put(bdev, FMODE_EXCL);
 270         return ret;
 271 }
 272
 273 /*
 274  * called from commit_transaction. Writes changed device replace state to
 275  * disk.
 276  */
 277 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 278                           struct btrfs_fs_info *fs_info)
 279 {
 280         int ret;
 281         struct btrfs_root *dev_root = fs_info->dev_root;
 282         struct btrfs_path *path;
 283         struct btrfs_key key;
 284         struct extent_buffer *eb;
 285         struct btrfs_dev_replace_item *ptr;
 286         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 287
 288         btrfs_dev_replace_read_lock(dev_replace);
 289         if (!dev_replace->is_valid ||
 290             !dev_replace->item_needs_writeback) {
 291                 btrfs_dev_replace_read_unlock(dev_replace);
 292                 return 0;
 293         }
 294         btrfs_dev_replace_read_unlock(dev_replace);
 295
 296         key.objectid = 0;
 297         key.type = BTRFS_DEV_REPLACE_KEY;
 298         key.offset = 0;
 299
 300         path = btrfs_alloc_path();
 301         if (!path) {
 302                 ret = -ENOMEM;
 303                 goto out;
 304         }
 305         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 306         if (ret < 0) {
 307                 btrfs_warn(fs_info,
 308                            "error %d while searching for dev_replace item!",
 309                            ret);
 310                 goto out;
 311         }
 312
 313         if (ret == 0 &&
 314             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 315                 /*
 316                  * need to delete old one and insert a new one.
 317                  * Since no attempt is made to recover any old state, if the
 318                  * dev_replace state is 'running', the data on the target
 319                  * drive is lost.
 320                  * It would be possible to recover the state: just make sure
 321                  * that the beginning of the item is never changed and always
 322                  * contains all the essential information. Then read this
 323                  * minimal set of information and use it as a base for the
 324                  * new state.
 325                  */
 326                 ret = btrfs_del_item(trans, dev_root, path);
 327                 if (ret != 0) {
 328                         btrfs_warn(fs_info,
 329                                    "delete too small dev_replace item failed %d!",
 330                                    ret);
 331                         goto out;
 332                 }
 333                 ret = 1;
 334         }
 335
 336         if (ret == 1) {
 337                 /* need to insert a new item */
 338                 btrfs_release_path(path);
 339                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 340                                               &key, sizeof(*ptr));
 341                 if (ret < 0) {
 342                         btrfs_warn(fs_info,
 343                                    "insert dev_replace item failed %d!", ret);
 344                         goto out;
 345                 }
 346         }
 347
 348         eb = path->nodes[0];
 349         ptr = btrfs_item_ptr(eb, path->slots[0],
 350                              struct btrfs_dev_replace_item);
 351
 352         btrfs_dev_replace_write_lock(dev_replace);
 353         if (dev_replace->srcdev)
 354                 btrfs_set_dev_replace_src_devid(eb, ptr,
 355                         dev_replace->srcdev->devid);
 356         else
 357                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 358         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 359                 dev_replace->cont_reading_from_srcdev_mode);
 360         btrfs_set_dev_replace_replace_state(eb, ptr,
 361                 dev_replace->replace_state);
 362         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 363         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 364         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 365                 atomic64_read(&dev_replace->num_write_errors));
 366         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 367                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 368         dev_replace->cursor_left_last_write_of_item =
 369                 dev_replace->cursor_left;
 370         btrfs_set_dev_replace_cursor_left(eb, ptr,
 371                 dev_replace->cursor_left_last_write_of_item);
 372         btrfs_set_dev_replace_cursor_right(eb, ptr,
 373                 dev_replace->cursor_right);
 374         dev_replace->item_needs_writeback = 0;
 375         btrfs_dev_replace_write_unlock(dev_replace);
 376
 377         btrfs_mark_buffer_dirty(eb);
 378
 379 out:
 380         btrfs_free_path(path);
 381
 382         return ret;
 383 }
 384
 385 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 386 {
 387         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 388
 389         dev_replace->committed_cursor_left =
 390                 dev_replace->cursor_left_last_write_of_item;
 391 }
 392
 393 static char* btrfs_dev_name(struct btrfs_device *device)
 394 {
 395         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 396                 return "<missing disk>";
 397         else
 398                 return rcu_str_deref(device->name);
 399 }
 400
 401 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 402                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 403                 int read_src)
 404 {
 405         struct btrfs_root *root = fs_info->dev_root;
 406         struct btrfs_trans_handle *trans;
 407         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 408         int ret;
 409         struct btrfs_device *tgt_device = NULL;
 410         struct btrfs_device *src_device = NULL;
 411
 412         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 413                                             srcdev_name, &src_device);
 414         if (ret)
 415                 return ret;
 416
 417         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 418                                             src_device, &tgt_device);
 419         if (ret)
 420                 return ret;
 421
 422         /*
 423          * Here we commit the transaction to make sure commit_total_bytes
 424          * of all the devices are updated.
 425          */
 426         trans = btrfs_attach_transaction(root);
 427         if (!IS_ERR(trans)) {
 428                 ret = btrfs_commit_transaction(trans);
 429                 if (ret)
 430                         return ret;
 431         } else if (PTR_ERR(trans) != -ENOENT) {
 432                 return PTR_ERR(trans);
 433         }
 434
 435         btrfs_dev_replace_write_lock(dev_replace);
 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                 goto leave;
 446         }
 447
 448         dev_replace->cont_reading_from_srcdev_mode = read_src;
 449         WARN_ON(!src_device);
 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         btrfs_dev_replace_write_unlock(dev_replace);
 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         /* force writing the updated state information to disk */
 482         trans = btrfs_start_transaction(root, 0);
 483         if (IS_ERR(trans)) {
 484                 ret = PTR_ERR(trans);
 485                 btrfs_dev_replace_write_lock(dev_replace);
 486                 dev_replace->replace_state =
 487                         BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED;
 488                 dev_replace->srcdev = NULL;
 489                 dev_replace->tgtdev = NULL;
 490                 goto leave;
 491         }
 492
 493         ret = btrfs_commit_transaction(trans);
 494         WARN_ON(ret);
 495
 496         /* the disk copy procedure reuses the scrub code */
 497         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 498                               btrfs_device_get_total_bytes(src_device),
 499                               &dev_replace->scrub_progress, 0, 1);
 500
 501         ret = btrfs_dev_replace_finishing(fs_info, ret);
 502         if (ret == -EINPROGRESS) {
 503                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 504         } else {
 505                 WARN_ON(ret);
 506         }
 507
 508         return ret;
 509
 510 leave:
 511         btrfs_dev_replace_write_unlock(dev_replace);
 512         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 513         return ret;
 514 }
 515
 516 int 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 = 0;
 541
 542         return ret;
 543 }
 544
 545 /*
 546  * blocked until all in-flight bios operations are finished.
 547  */
 548 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 549 {
 550         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 551         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 552                    &fs_info->bio_counter));
 553 }
 554
 555 /*
 556  * we have removed target device, it is safe to allow new bios request.
 557  */
 558 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 559 {
 560         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 561         wake_up(&fs_info->replace_wait);
 562 }
 563
 564 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 565                                        int scrub_ret)
 566 {
 567         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 568         struct btrfs_device *tgt_device;
 569         struct btrfs_device *src_device;
 570         struct btrfs_root *root = fs_info->tree_root;
 571         u8 uuid_tmp[BTRFS_UUID_SIZE];
 572         struct btrfs_trans_handle *trans;
 573         int ret = 0;
 574
 575         /* don't allow cancel or unmount to disturb the finishing procedure */
 576         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 577
 578         btrfs_dev_replace_read_lock(dev_replace);
 579         /* was the operation canceled, or is it finished? */
 580         if (dev_replace->replace_state !=
 581             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 582                 btrfs_dev_replace_read_unlock(dev_replace);
 583                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 584                 return 0;
 585         }
 586
 587         tgt_device = dev_replace->tgtdev;
 588         src_device = dev_replace->srcdev;
 589         btrfs_dev_replace_read_unlock(dev_replace);
 590
 591         /*
 592          * flush all outstanding I/O and inode extent mappings before the
 593          * copy operation is declared as being finished
 594          */
 595         ret = btrfs_start_delalloc_roots(fs_info, -1);
 596         if (ret) {
 597                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 598                 return ret;
 599         }
 600         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 601
 602         trans = btrfs_start_transaction(root, 0);
 603         if (IS_ERR(trans)) {
 604                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 605                 return PTR_ERR(trans);
 606         }
 607         ret = btrfs_commit_transaction(trans);
 608         WARN_ON(ret);
 609
 610         /* keep away write_all_supers() during the finishing procedure */
 611         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 612         mutex_lock(&fs_info->chunk_mutex);
 613         btrfs_dev_replace_write_lock(dev_replace);
 614         dev_replace->replace_state =
 615                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 616                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 617         dev_replace->tgtdev = NULL;
 618         dev_replace->srcdev = NULL;
 619         dev_replace->time_stopped = ktime_get_real_seconds();
 620         dev_replace->item_needs_writeback = 1;
 621
 622         /* replace old device with new one in mapping tree */
 623         if (!scrub_ret) {
 624                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 625                                                                 src_device,
 626                                                                 tgt_device);
 627         } else {
 628                 btrfs_err_in_rcu(fs_info,
 629                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 630                                  btrfs_dev_name(src_device),
 631                                  src_device->devid,
 632                                  rcu_str_deref(tgt_device->name), scrub_ret);
 633                 btrfs_dev_replace_write_unlock(dev_replace);
 634                 mutex_unlock(&fs_info->chunk_mutex);
 635                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 636                 btrfs_rm_dev_replace_blocked(fs_info);
 637                 if (tgt_device)
 638                         btrfs_destroy_dev_replace_tgtdev(tgt_device);
 639                 btrfs_rm_dev_replace_unblocked(fs_info);
 640                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 641
 642                 return scrub_ret;
 643         }
 644
 645         btrfs_info_in_rcu(fs_info,
 646                           "dev_replace from %s (devid %llu) to %s finished",
 647                           btrfs_dev_name(src_device),
 648                           src_device->devid,
 649                           rcu_str_deref(tgt_device->name));
 650         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 651         tgt_device->devid = src_device->devid;
 652         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 653         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 654         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 655         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 656         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 657         btrfs_device_set_disk_total_bytes(tgt_device,
 658                                           src_device->disk_total_bytes);
 659         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 660         ASSERT(list_empty(&src_device->resized_list));
 661         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 662         tgt_device->commit_bytes_used = src_device->bytes_used;
 663
 664         btrfs_assign_next_active_device(src_device, tgt_device);
 665
 666         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 667         fs_info->fs_devices->rw_devices++;
 668
 669         btrfs_dev_replace_write_unlock(dev_replace);
 670
 671         btrfs_rm_dev_replace_blocked(fs_info);
 672
 673         btrfs_rm_dev_replace_remove_srcdev(src_device);
 674
 675         btrfs_rm_dev_replace_unblocked(fs_info);
 676
 677         /*
 678          * Increment dev_stats_ccnt so that btrfs_run_dev_stats() will
 679          * update on-disk dev stats value during commit transaction
 680          */
 681         atomic_inc(&tgt_device->dev_stats_ccnt);
 682
 683         /*
 684          * this is again a consistent state where no dev_replace procedure
 685          * is running, the target device is part of the filesystem, the
 686          * source device is not part of the filesystem anymore and its 1st
 687          * superblock is scratched out so that it is no longer marked to
 688          * belong to this filesystem.
 689          */
 690         mutex_unlock(&fs_info->chunk_mutex);
 691         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 692
 693         /* replace the sysfs entry */
 694         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 695         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 696
 697         /* write back the superblocks */
 698         trans = btrfs_start_transaction(root, 0);
 699         if (!IS_ERR(trans))
 700                 btrfs_commit_transaction(trans);
 701
 702         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 703
 704         return 0;
 705 }
 706
 707 static void btrfs_dev_replace_update_device_in_mapping_tree(
 708                                                 struct btrfs_fs_info *fs_info,
 709                                                 struct btrfs_device *srcdev,
 710                                                 struct btrfs_device *tgtdev)
 711 {
 712         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 713         struct extent_map *em;
 714         struct map_lookup *map;
 715         u64 start = 0;
 716         int i;
 717
 718         write_lock(&em_tree->lock);
 719         do {
 720                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 721                 if (!em)
 722                         break;
 723                 map = em->map_lookup;
 724                 for (i = 0; i < map->num_stripes; i++)
 725                         if (srcdev == map->stripes[i].dev)
 726                                 map->stripes[i].dev = tgtdev;
 727                 start = em->start + em->len;
 728                 free_extent_map(em);
 729         } while (start);
 730         write_unlock(&em_tree->lock);
 731 }
 732
 733 /*
 734  * Read progress of device replace status according to the state and last
 735  * stored position. The value format is the same as for
 736  * btrfs_dev_replace::progress_1000
 737  */
 738 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 739 {
 740         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 741         u64 ret = 0;
 742
 743         switch (dev_replace->replace_state) {
 744         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 745         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 746                 ret = 0;
 747                 break;
 748         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 749                 ret = 1000;
 750                 break;
 751         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 752         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 753                 ret = div64_u64(dev_replace->cursor_left,
 754                                 div_u64(btrfs_device_get_total_bytes(
 755                                                 dev_replace->srcdev), 1000));
 756                 break;
 757         }
 758
 759         return ret;
 760 }
 761
 762 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 763                               struct btrfs_ioctl_dev_replace_args *args)
 764 {
 765         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 766
 767         btrfs_dev_replace_read_lock(dev_replace);
 768         /* even if !dev_replace_is_valid, the values are good enough for
 769          * the replace_status ioctl */
 770         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 771         args->status.replace_state = dev_replace->replace_state;
 772         args->status.time_started = dev_replace->time_started;
 773         args->status.time_stopped = dev_replace->time_stopped;
 774         args->status.num_write_errors =
 775                 atomic64_read(&dev_replace->num_write_errors);
 776         args->status.num_uncorrectable_read_errors =
 777                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 778         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 779         btrfs_dev_replace_read_unlock(dev_replace);
 780 }
 781
 782 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 783 {
 784         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 785         struct btrfs_device *tgt_device = NULL;
 786         struct btrfs_device *src_device = NULL;
 787         struct btrfs_trans_handle *trans;
 788         struct btrfs_root *root = fs_info->tree_root;
 789         int result;
 790         int ret;
 791
 792         if (sb_rdonly(fs_info->sb))
 793                 return -EROFS;
 794
 795         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 796         btrfs_dev_replace_write_lock(dev_replace);
 797         switch (dev_replace->replace_state) {
 798         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 799         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 800         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 801                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 802                 btrfs_dev_replace_write_unlock(dev_replace);
 803                 goto leave;
 804         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 805         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 806                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 807                 tgt_device = dev_replace->tgtdev;
 808                 src_device = dev_replace->srcdev;
 809                 dev_replace->tgtdev = NULL;
 810                 dev_replace->srcdev = NULL;
 811                 break;
 812         }
 813         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 814         dev_replace->time_stopped = ktime_get_real_seconds();
 815         dev_replace->item_needs_writeback = 1;
 816         btrfs_dev_replace_write_unlock(dev_replace);
 817         btrfs_scrub_cancel(fs_info);
 818
 819         trans = btrfs_start_transaction(root, 0);
 820         if (IS_ERR(trans)) {
 821                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 822                 return PTR_ERR(trans);
 823         }
 824         ret = btrfs_commit_transaction(trans);
 825         WARN_ON(ret);
 826
 827         btrfs_info_in_rcu(fs_info,
 828                 "dev_replace from %s (devid %llu) to %s canceled",
 829                 btrfs_dev_name(src_device), src_device->devid,
 830                 btrfs_dev_name(tgt_device));
 831
 832         if (tgt_device)
 833                 btrfs_destroy_dev_replace_tgtdev(tgt_device);
 834
 835 leave:
 836         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 837         return result;
 838 }
 839
 840 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 841 {
 842         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 843
 844         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 845         btrfs_dev_replace_write_lock(dev_replace);
 846         switch (dev_replace->replace_state) {
 847         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 848         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 849         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 850         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 851                 break;
 852         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 853                 dev_replace->replace_state =
 854                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 855                 dev_replace->time_stopped = ktime_get_real_seconds();
 856                 dev_replace->item_needs_writeback = 1;
 857                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 858                 break;
 859         }
 860
 861         btrfs_dev_replace_write_unlock(dev_replace);
 862         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 863 }
 864
 865 /* resume dev_replace procedure that was interrupted by unmount */
 866 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 867 {
 868         struct task_struct *task;
 869         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 870
 871         btrfs_dev_replace_write_lock(dev_replace);
 872         switch (dev_replace->replace_state) {
 873         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 874         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 875         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 876                 btrfs_dev_replace_write_unlock(dev_replace);
 877                 return 0;
 878         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 879                 break;
 880         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 881                 dev_replace->replace_state =
 882                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 883                 break;
 884         }
 885         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 886                 btrfs_info(fs_info,
 887                            "cannot continue dev_replace, tgtdev is missing");
 888                 btrfs_info(fs_info,
 889                            "you may cancel the operation after 'mount -o degraded'");
 890                 btrfs_dev_replace_write_unlock(dev_replace);
 891                 return 0;
 892         }
 893         btrfs_dev_replace_write_unlock(dev_replace);
 894
 895         /*
 896          * This could collide with a paused balance, but the exclusive op logic
 897          * should never allow both to start and pause. We don't want to allow
 898          * dev-replace to start anyway.
 899          */
 900         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
 901                 btrfs_info(fs_info,
 902                 "cannot resume dev-replace, other exclusive operation running");
 903                 return 0;
 904         }
 905
 906         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 907         return PTR_ERR_OR_ZERO(task);
 908 }
 909
 910 static int btrfs_dev_replace_kthread(void *data)
 911 {
 912         struct btrfs_fs_info *fs_info = data;
 913         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 914         u64 progress;
 915         int ret;
 916
 917         progress = btrfs_dev_replace_progress(fs_info);
 918         progress = div_u64(progress, 10);
 919         btrfs_info_in_rcu(fs_info,
 920                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 921                 btrfs_dev_name(dev_replace->srcdev),
 922                 dev_replace->srcdev->devid,
 923                 btrfs_dev_name(dev_replace->tgtdev),
 924                 (unsigned int)progress);
 925
 926         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 927                               dev_replace->committed_cursor_left,
 928                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 929                               &dev_replace->scrub_progress, 0, 1);
 930         ret = btrfs_dev_replace_finishing(fs_info, ret);
 931         WARN_ON(ret);
 932
 933         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 934         return 0;
 935 }
 936
 937 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 938 {
 939         if (!dev_replace->is_valid)
 940                 return 0;
 941
 942         switch (dev_replace->replace_state) {
 943         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 944         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 945         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 946                 return 0;
 947         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 948         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 949                 /*
 950                  * return true even if tgtdev is missing (this is
 951                  * something that can happen if the dev_replace
 952                  * procedure is suspended by an umount and then
 953                  * the tgtdev is missing (or "btrfs dev scan") was
 954                  * not called and the the filesystem is remounted
 955                  * in degraded state. This does not stop the
 956                  * dev_replace procedure. It needs to be canceled
 957                  * manually if the cancellation is wanted.
 958                  */
 959                 break;
 960         }
 961         return 1;
 962 }
 963
 964 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
 965 {
 966         read_lock(&dev_replace->lock);
 967         atomic_inc(&dev_replace->read_locks);
 968 }
 969
 970 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
 971 {
 972         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 973         atomic_dec(&dev_replace->read_locks);
 974         read_unlock(&dev_replace->lock);
 975 }
 976
 977 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
 978 {
 979 again:
 980         wait_event(dev_replace->read_lock_wq,
 981                    atomic_read(&dev_replace->blocking_readers) == 0);
 982         write_lock(&dev_replace->lock);
 983         if (atomic_read(&dev_replace->blocking_readers)) {
 984                 write_unlock(&dev_replace->lock);
 985                 goto again;
 986         }
 987 }
 988
 989 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
 990 {
 991         ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 992         write_unlock(&dev_replace->lock);
 993 }
 994
 995 /* inc blocking cnt and release read lock */
 996 void btrfs_dev_replace_set_lock_blocking(
 997                                         struct btrfs_dev_replace *dev_replace)
 998 {
 999         /* only set blocking for read lock */
1000         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1001         atomic_inc(&dev_replace->blocking_readers);
1002         read_unlock(&dev_replace->lock);
1003 }
1004
1005 /* acquire read lock and dec blocking cnt */
1006 void btrfs_dev_replace_clear_lock_blocking(
1007                                         struct btrfs_dev_replace *dev_replace)
1008 {
1009         /* only set blocking for read lock */
1010         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1011         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
1012         read_lock(&dev_replace->lock);
1013         /* Barrier implied by atomic_dec_and_test */
1014         if (atomic_dec_and_test(&dev_replace->blocking_readers))
1015                 cond_wake_up_nomb(&dev_replace->read_lock_wq);
1016 }
1017
1018 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1019 {
1020         percpu_counter_inc(&fs_info->bio_counter);
1021 }
1022
1023 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1024 {
1025         percpu_counter_sub(&fs_info->bio_counter, amount);
1026         cond_wake_up_nomb(&fs_info->replace_wait);
1027 }
1028
1029 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1030 {
1031         while (1) {
1032                 percpu_counter_inc(&fs_info->bio_counter);
1033                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1034                                      &fs_info->fs_state)))
1035                         break;
1036
1037                 btrfs_bio_counter_dec(fs_info);
1038                 wait_event(fs_info->replace_wait,
1039                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1040                                      &fs_info->fs_state));
1041         }
1042 }