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/buffer_head.h>
  10 #include <linux/blkdev.h>
  11 #include <linux/random.h>
  12 #include <linux/iocontext.h>
  13 #include <linux/capability.h>
  14 #include <linux/kthread.h>
  15 #include <linux/math64.h>
  16 #include <asm/div64.h>
  17 #include "ctree.h"
  18 #include "extent_map.h"
  19 #include "disk-io.h"
  20 #include "transaction.h"
  21 #include "print-tree.h"
  22 #include "volumes.h"
  23 #include "async-thread.h"
  24 #include "check-integrity.h"
  25 #include "rcu-string.h"
  26 #include "dev-replace.h"
  27 #include "sysfs.h"
  28
  29 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
  30                                        int scrub_ret);
  31 static void btrfs_dev_replace_update_device_in_mapping_tree(
  32                                                 struct btrfs_fs_info *fs_info,
  33                                                 struct btrfs_device *srcdev,
  34                                                 struct btrfs_device *tgtdev);
  35 static int btrfs_dev_replace_kthread(void *data);
  36
  37 int btrfs_init_dev_replace(struct btrfs_fs_info *fs_info)
  38 {
  39         struct btrfs_key key;
  40         struct btrfs_root *dev_root = fs_info->dev_root;
  41         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
  42         struct extent_buffer *eb;
  43         int slot;
  44         int ret = 0;
  45         struct btrfs_path *path = NULL;
  46         int item_size;
  47         struct btrfs_dev_replace_item *ptr;
  48         u64 src_devid;
  49
  50         path = btrfs_alloc_path();
  51         if (!path) {
  52                 ret = -ENOMEM;
  53                 goto out;
  54         }
  55
  56         key.objectid = 0;
  57         key.type = BTRFS_DEV_REPLACE_KEY;
  58         key.offset = 0;
  59         ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
  60         if (ret) {
  61 no_valid_dev_replace_entry_found:
  62                 ret = 0;
  63                 dev_replace->replace_state =
  64                         BTRFS_DEV_REPLACE_ITEM_STATE_NEVER_STARTED;
  65                 dev_replace->cont_reading_from_srcdev_mode =
  66                     BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_ALWAYS;
  67                 dev_replace->replace_state = 0;
  68                 dev_replace->time_started = 0;
  69                 dev_replace->time_stopped = 0;
  70                 atomic64_set(&dev_replace->num_write_errors, 0);
  71                 atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
  72                 dev_replace->cursor_left = 0;
  73                 dev_replace->committed_cursor_left = 0;
  74                 dev_replace->cursor_left_last_write_of_item = 0;
  75                 dev_replace->cursor_right = 0;
  76                 dev_replace->srcdev = NULL;
  77                 dev_replace->tgtdev = NULL;
  78                 dev_replace->is_valid = 0;
  79                 dev_replace->item_needs_writeback = 0;
  80                 goto out;
  81         }
  82         slot = path->slots[0];
  83         eb = path->nodes[0];
  84         item_size = btrfs_item_size_nr(eb, slot);
  85         ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_replace_item);
  86
  87         if (item_size != sizeof(struct btrfs_dev_replace_item)) {
  88                 btrfs_warn(fs_info,
  89                         "dev_replace entry found has unexpected size, ignore entry");
  90                 goto no_valid_dev_replace_entry_found;
  91         }
  92
  93         src_devid = btrfs_dev_replace_src_devid(eb, ptr);
  94         dev_replace->cont_reading_from_srcdev_mode =
  95                 btrfs_dev_replace_cont_reading_from_srcdev_mode(eb, ptr);
  96         dev_replace->replace_state = btrfs_dev_replace_replace_state(eb, ptr);
  97         dev_replace->time_started = btrfs_dev_replace_time_started(eb, ptr);
  98         dev_replace->time_stopped =
  99                 btrfs_dev_replace_time_stopped(eb, ptr);
 100         atomic64_set(&dev_replace->num_write_errors,
 101                      btrfs_dev_replace_num_write_errors(eb, ptr));
 102         atomic64_set(&dev_replace->num_uncorrectable_read_errors,
 103                      btrfs_dev_replace_num_uncorrectable_read_errors(eb, ptr));
 104         dev_replace->cursor_left = btrfs_dev_replace_cursor_left(eb, ptr);
 105         dev_replace->committed_cursor_left = dev_replace->cursor_left;
 106         dev_replace->cursor_left_last_write_of_item = dev_replace->cursor_left;
 107         dev_replace->cursor_right = btrfs_dev_replace_cursor_right(eb, ptr);
 108         dev_replace->is_valid = 1;
 109
 110         dev_replace->item_needs_writeback = 0;
 111         switch (dev_replace->replace_state) {
 112         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 113         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 114         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 115                 dev_replace->srcdev = NULL;
 116                 dev_replace->tgtdev = NULL;
 117                 break;
 118         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 119         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 120                 dev_replace->srcdev = btrfs_find_device(fs_info, src_devid,
 121                                                         NULL, NULL);
 122                 dev_replace->tgtdev = btrfs_find_device(fs_info,
 123                                                         BTRFS_DEV_REPLACE_DEVID,
 124                                                         NULL, NULL);
 125                 /*
 126                  * allow 'btrfs dev replace_cancel' if src/tgt device is
 127                  * missing
 128                  */
 129                 if (!dev_replace->srcdev &&
 130                     !btrfs_test_opt(fs_info, DEGRADED)) {
 131                         ret = -EIO;
 132                         btrfs_warn(fs_info,
 133                            "cannot mount because device replace operation is ongoing and");
 134                         btrfs_warn(fs_info,
 135                            "srcdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 136                            src_devid);
 137                 }
 138                 if (!dev_replace->tgtdev &&
 139                     !btrfs_test_opt(fs_info, DEGRADED)) {
 140                         ret = -EIO;
 141                         btrfs_warn(fs_info,
 142                            "cannot mount because device replace operation is ongoing and");
 143                         btrfs_warn(fs_info,
 144                            "tgtdev (devid %llu) is missing, need to run 'btrfs dev scan'?",
 145                                 BTRFS_DEV_REPLACE_DEVID);
 146                 }
 147                 if (dev_replace->tgtdev) {
 148                         if (dev_replace->srcdev) {
 149                                 dev_replace->tgtdev->total_bytes =
 150                                         dev_replace->srcdev->total_bytes;
 151                                 dev_replace->tgtdev->disk_total_bytes =
 152                                         dev_replace->srcdev->disk_total_bytes;
 153                                 dev_replace->tgtdev->commit_total_bytes =
 154                                         dev_replace->srcdev->commit_total_bytes;
 155                                 dev_replace->tgtdev->bytes_used =
 156                                         dev_replace->srcdev->bytes_used;
 157                                 dev_replace->tgtdev->commit_bytes_used =
 158                                         dev_replace->srcdev->commit_bytes_used;
 159                         }
 160                         set_bit(BTRFS_DEV_STATE_REPLACE_TGT,
 161                                 &dev_replace->tgtdev->dev_state);
 162
 163                         WARN_ON(fs_info->fs_devices->rw_devices == 0);
 164                         dev_replace->tgtdev->io_width = fs_info->sectorsize;
 165                         dev_replace->tgtdev->io_align = fs_info->sectorsize;
 166                         dev_replace->tgtdev->sector_size = fs_info->sectorsize;
 167                         dev_replace->tgtdev->fs_info = fs_info;
 168                         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
 169                                 &dev_replace->tgtdev->dev_state);
 170                 }
 171                 break;
 172         }
 173
 174 out:
 175         btrfs_free_path(path);
 176         return ret;
 177 }
 178
 179 /*
 180  * Initialize a new device for device replace target from a given source dev
 181  * and path.
 182  *
 183  * Return 0 and new device in @device_out, otherwise return < 0
 184  */
 185 static int btrfs_init_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
 186                                   const char *device_path,
 187                                   struct btrfs_device *srcdev,
 188                                   struct btrfs_device **device_out)
 189 {
 190         struct btrfs_device *device;
 191         struct block_device *bdev;
 192         struct list_head *devices;
 193         struct rcu_string *name;
 194         u64 devid = BTRFS_DEV_REPLACE_DEVID;
 195         int ret = 0;
 196
 197         *device_out = NULL;
 198         if (fs_info->fs_devices->seeding) {
 199                 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
 200                 return -EINVAL;
 201         }
 202
 203         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
 204                                   fs_info->bdev_holder);
 205         if (IS_ERR(bdev)) {
 206                 btrfs_err(fs_info, "target device %s is invalid!", device_path);
 207                 return PTR_ERR(bdev);
 208         }
 209
 210         filemap_write_and_wait(bdev->bd_inode->i_mapping);
 211
 212         devices = &fs_info->fs_devices->devices;
 213         list_for_each_entry(device, devices, dev_list) {
 214                 if (device->bdev == bdev) {
 215                         btrfs_err(fs_info,
 216                                   "target device is in the filesystem!");
 217                         ret = -EEXIST;
 218                         goto error;
 219                 }
 220         }
 221
 222
 223         if (i_size_read(bdev->bd_inode) <
 224             btrfs_device_get_total_bytes(srcdev)) {
 225                 btrfs_err(fs_info,
 226                           "target device is smaller than source device!");
 227                 ret = -EINVAL;
 228                 goto error;
 229         }
 230
 231
 232         device = btrfs_alloc_device(NULL, &devid, NULL);
 233         if (IS_ERR(device)) {
 234                 ret = PTR_ERR(device);
 235                 goto error;
 236         }
 237
 238         name = rcu_string_strdup(device_path, GFP_KERNEL);
 239         if (!name) {
 240                 btrfs_free_device(device);
 241                 ret = -ENOMEM;
 242                 goto error;
 243         }
 244         rcu_assign_pointer(device->name, name);
 245
 246         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 247         set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
 248         device->generation = 0;
 249         device->io_width = fs_info->sectorsize;
 250         device->io_align = fs_info->sectorsize;
 251         device->sector_size = fs_info->sectorsize;
 252         device->total_bytes = btrfs_device_get_total_bytes(srcdev);
 253         device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
 254         device->bytes_used = btrfs_device_get_bytes_used(srcdev);
 255         device->commit_total_bytes = srcdev->commit_total_bytes;
 256         device->commit_bytes_used = device->bytes_used;
 257         device->fs_info = fs_info;
 258         device->bdev = bdev;
 259         set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
 260         set_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
 261         device->mode = FMODE_EXCL;
 262         device->dev_stats_valid = 1;
 263         set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
 264         device->fs_devices = fs_info->fs_devices;
 265         list_add(&device->dev_list, &fs_info->fs_devices->devices);
 266         fs_info->fs_devices->num_devices++;
 267         fs_info->fs_devices->open_devices++;
 268         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 269
 270         *device_out = device;
 271         return 0;
 272
 273 error:
 274         blkdev_put(bdev, FMODE_EXCL);
 275         return ret;
 276 }
 277
 278 /*
 279  * called from commit_transaction. Writes changed device replace state to
 280  * disk.
 281  */
 282 int btrfs_run_dev_replace(struct btrfs_trans_handle *trans,
 283                           struct btrfs_fs_info *fs_info)
 284 {
 285         int ret;
 286         struct btrfs_root *dev_root = fs_info->dev_root;
 287         struct btrfs_path *path;
 288         struct btrfs_key key;
 289         struct extent_buffer *eb;
 290         struct btrfs_dev_replace_item *ptr;
 291         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 292
 293         btrfs_dev_replace_read_lock(dev_replace);
 294         if (!dev_replace->is_valid ||
 295             !dev_replace->item_needs_writeback) {
 296                 btrfs_dev_replace_read_unlock(dev_replace);
 297                 return 0;
 298         }
 299         btrfs_dev_replace_read_unlock(dev_replace);
 300
 301         key.objectid = 0;
 302         key.type = BTRFS_DEV_REPLACE_KEY;
 303         key.offset = 0;
 304
 305         path = btrfs_alloc_path();
 306         if (!path) {
 307                 ret = -ENOMEM;
 308                 goto out;
 309         }
 310         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
 311         if (ret < 0) {
 312                 btrfs_warn(fs_info,
 313                            "error %d while searching for dev_replace item!",
 314                            ret);
 315                 goto out;
 316         }
 317
 318         if (ret == 0 &&
 319             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
 320                 /*
 321                  * need to delete old one and insert a new one.
 322                  * Since no attempt is made to recover any old state, if the
 323                  * dev_replace state is 'running', the data on the target
 324                  * drive is lost.
 325                  * It would be possible to recover the state: just make sure
 326                  * that the beginning of the item is never changed and always
 327                  * contains all the essential information. Then read this
 328                  * minimal set of information and use it as a base for the
 329                  * new state.
 330                  */
 331                 ret = btrfs_del_item(trans, dev_root, path);
 332                 if (ret != 0) {
 333                         btrfs_warn(fs_info,
 334                                    "delete too small dev_replace item failed %d!",
 335                                    ret);
 336                         goto out;
 337                 }
 338                 ret = 1;
 339         }
 340
 341         if (ret == 1) {
 342                 /* need to insert a new item */
 343                 btrfs_release_path(path);
 344                 ret = btrfs_insert_empty_item(trans, dev_root, path,
 345                                               &key, sizeof(*ptr));
 346                 if (ret < 0) {
 347                         btrfs_warn(fs_info,
 348                                    "insert dev_replace item failed %d!", ret);
 349                         goto out;
 350                 }
 351         }
 352
 353         eb = path->nodes[0];
 354         ptr = btrfs_item_ptr(eb, path->slots[0],
 355                              struct btrfs_dev_replace_item);
 356
 357         btrfs_dev_replace_write_lock(dev_replace);
 358         if (dev_replace->srcdev)
 359                 btrfs_set_dev_replace_src_devid(eb, ptr,
 360                         dev_replace->srcdev->devid);
 361         else
 362                 btrfs_set_dev_replace_src_devid(eb, ptr, (u64)-1);
 363         btrfs_set_dev_replace_cont_reading_from_srcdev_mode(eb, ptr,
 364                 dev_replace->cont_reading_from_srcdev_mode);
 365         btrfs_set_dev_replace_replace_state(eb, ptr,
 366                 dev_replace->replace_state);
 367         btrfs_set_dev_replace_time_started(eb, ptr, dev_replace->time_started);
 368         btrfs_set_dev_replace_time_stopped(eb, ptr, dev_replace->time_stopped);
 369         btrfs_set_dev_replace_num_write_errors(eb, ptr,
 370                 atomic64_read(&dev_replace->num_write_errors));
 371         btrfs_set_dev_replace_num_uncorrectable_read_errors(eb, ptr,
 372                 atomic64_read(&dev_replace->num_uncorrectable_read_errors));
 373         dev_replace->cursor_left_last_write_of_item =
 374                 dev_replace->cursor_left;
 375         btrfs_set_dev_replace_cursor_left(eb, ptr,
 376                 dev_replace->cursor_left_last_write_of_item);
 377         btrfs_set_dev_replace_cursor_right(eb, ptr,
 378                 dev_replace->cursor_right);
 379         dev_replace->item_needs_writeback = 0;
 380         btrfs_dev_replace_write_unlock(dev_replace);
 381
 382         btrfs_mark_buffer_dirty(eb);
 383
 384 out:
 385         btrfs_free_path(path);
 386
 387         return ret;
 388 }
 389
 390 void btrfs_after_dev_replace_commit(struct btrfs_fs_info *fs_info)
 391 {
 392         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 393
 394         dev_replace->committed_cursor_left =
 395                 dev_replace->cursor_left_last_write_of_item;
 396 }
 397
 398 static char* btrfs_dev_name(struct btrfs_device *device)
 399 {
 400         if (!device || test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
 401                 return "<missing disk>";
 402         else
 403                 return rcu_str_deref(device->name);
 404 }
 405
 406 int btrfs_dev_replace_start(struct btrfs_fs_info *fs_info,
 407                 const char *tgtdev_name, u64 srcdevid, const char *srcdev_name,
 408                 int read_src)
 409 {
 410         struct btrfs_root *root = fs_info->dev_root;
 411         struct btrfs_trans_handle *trans;
 412         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 413         int ret;
 414         struct btrfs_device *tgt_device = NULL;
 415         struct btrfs_device *src_device = NULL;
 416
 417         /* the disk copy procedure reuses the scrub code */
 418         mutex_lock(&fs_info->volume_mutex);
 419         ret = btrfs_find_device_by_devspec(fs_info, srcdevid,
 420                                             srcdev_name, &src_device);
 421         if (ret) {
 422                 mutex_unlock(&fs_info->volume_mutex);
 423                 return ret;
 424         }
 425
 426         ret = btrfs_init_dev_replace_tgtdev(fs_info, tgtdev_name,
 427                                             src_device, &tgt_device);
 428         mutex_unlock(&fs_info->volume_mutex);
 429         if (ret)
 430                 return ret;
 431
 432         /*
 433          * Here we commit the transaction to make sure commit_total_bytes
 434          * of all the devices are updated.
 435          */
 436         trans = btrfs_attach_transaction(root);
 437         if (!IS_ERR(trans)) {
 438                 ret = btrfs_commit_transaction(trans);
 439                 if (ret)
 440                         return ret;
 441         } else if (PTR_ERR(trans) != -ENOENT) {
 442                 return PTR_ERR(trans);
 443         }
 444
 445         btrfs_dev_replace_write_lock(dev_replace);
 446         switch (dev_replace->replace_state) {
 447         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 448         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 449         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 450                 break;
 451         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 452         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 453                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_ALREADY_STARTED;
 454                 goto leave;
 455         }
 456
 457         dev_replace->cont_reading_from_srcdev_mode = read_src;
 458         WARN_ON(!src_device);
 459         dev_replace->srcdev = src_device;
 460         dev_replace->tgtdev = tgt_device;
 461
 462         btrfs_info_in_rcu(fs_info,
 463                       "dev_replace from %s (devid %llu) to %s started",
 464                       btrfs_dev_name(src_device),
 465                       src_device->devid,
 466                       rcu_str_deref(tgt_device->name));
 467
 468         /*
 469          * from now on, the writes to the srcdev are all duplicated to
 470          * go to the tgtdev as well (refer to btrfs_map_block()).
 471          */
 472         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 473         dev_replace->time_started = get_seconds();
 474         dev_replace->cursor_left = 0;
 475         dev_replace->committed_cursor_left = 0;
 476         dev_replace->cursor_left_last_write_of_item = 0;
 477         dev_replace->cursor_right = 0;
 478         dev_replace->is_valid = 1;
 479         dev_replace->item_needs_writeback = 1;
 480         atomic64_set(&dev_replace->num_write_errors, 0);
 481         atomic64_set(&dev_replace->num_uncorrectable_read_errors, 0);
 482         btrfs_dev_replace_write_unlock(dev_replace);
 483
 484         ret = btrfs_sysfs_add_device_link(tgt_device->fs_devices, tgt_device);
 485         if (ret)
 486                 btrfs_err(fs_info, "kobj add dev failed %d", ret);
 487
 488         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 489
 490         /* force writing the updated state information to disk */
 491         trans = btrfs_start_transaction(root, 0);
 492         if (IS_ERR(trans)) {
 493                 ret = PTR_ERR(trans);
 494                 btrfs_dev_replace_write_lock(dev_replace);
 495                 goto leave;
 496         }
 497
 498         ret = btrfs_commit_transaction(trans);
 499         WARN_ON(ret);
 500
 501         /* the disk copy procedure reuses the scrub code */
 502         ret = btrfs_scrub_dev(fs_info, src_device->devid, 0,
 503                               btrfs_device_get_total_bytes(src_device),
 504                               &dev_replace->scrub_progress, 0, 1);
 505
 506         ret = btrfs_dev_replace_finishing(fs_info, ret);
 507         if (ret == -EINPROGRESS) {
 508                 ret = BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS;
 509         } else {
 510                 WARN_ON(ret);
 511         }
 512
 513         return ret;
 514
 515 leave:
 516         dev_replace->srcdev = NULL;
 517         dev_replace->tgtdev = NULL;
 518         btrfs_dev_replace_write_unlock(dev_replace);
 519         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 520         return ret;
 521 }
 522
 523 int btrfs_dev_replace_by_ioctl(struct btrfs_fs_info *fs_info,
 524                             struct btrfs_ioctl_dev_replace_args *args)
 525 {
 526         int ret;
 527
 528         switch (args->start.cont_reading_from_srcdev_mode) {
 529         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_ALWAYS:
 530         case BTRFS_IOCTL_DEV_REPLACE_CONT_READING_FROM_SRCDEV_MODE_AVOID:
 531                 break;
 532         default:
 533                 return -EINVAL;
 534         }
 535
 536         if ((args->start.srcdevid == 0 && args->start.srcdev_name[0] == '\0') ||
 537             args->start.tgtdev_name[0] == '\0')
 538                 return -EINVAL;
 539
 540         ret = btrfs_dev_replace_start(fs_info, args->start.tgtdev_name,
 541                                         args->start.srcdevid,
 542                                         args->start.srcdev_name,
 543                                         args->start.cont_reading_from_srcdev_mode);
 544         args->result = ret;
 545         /* don't warn if EINPROGRESS, someone else might be running scrub */
 546         if (ret == BTRFS_IOCTL_DEV_REPLACE_RESULT_SCRUB_INPROGRESS)
 547                 ret = 0;
 548
 549         return ret;
 550 }
 551
 552 /*
 553  * blocked until all in-flight bios operations are finished.
 554  */
 555 static void btrfs_rm_dev_replace_blocked(struct btrfs_fs_info *fs_info)
 556 {
 557         set_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 558         wait_event(fs_info->replace_wait, !percpu_counter_sum(
 559                    &fs_info->bio_counter));
 560 }
 561
 562 /*
 563  * we have removed target device, it is safe to allow new bios request.
 564  */
 565 static void btrfs_rm_dev_replace_unblocked(struct btrfs_fs_info *fs_info)
 566 {
 567         clear_bit(BTRFS_FS_STATE_DEV_REPLACING, &fs_info->fs_state);
 568         wake_up(&fs_info->replace_wait);
 569 }
 570
 571 static int btrfs_dev_replace_finishing(struct btrfs_fs_info *fs_info,
 572                                        int scrub_ret)
 573 {
 574         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 575         struct btrfs_device *tgt_device;
 576         struct btrfs_device *src_device;
 577         struct btrfs_root *root = fs_info->tree_root;
 578         u8 uuid_tmp[BTRFS_UUID_SIZE];
 579         struct btrfs_trans_handle *trans;
 580         int ret = 0;
 581
 582         /* don't allow cancel or unmount to disturb the finishing procedure */
 583         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 584
 585         btrfs_dev_replace_read_lock(dev_replace);
 586         /* was the operation canceled, or is it finished? */
 587         if (dev_replace->replace_state !=
 588             BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED) {
 589                 btrfs_dev_replace_read_unlock(dev_replace);
 590                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 591                 return 0;
 592         }
 593
 594         tgt_device = dev_replace->tgtdev;
 595         src_device = dev_replace->srcdev;
 596         btrfs_dev_replace_read_unlock(dev_replace);
 597
 598         /*
 599          * flush all outstanding I/O and inode extent mappings before the
 600          * copy operation is declared as being finished
 601          */
 602         ret = btrfs_start_delalloc_roots(fs_info, 0, -1);
 603         if (ret) {
 604                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 605                 return ret;
 606         }
 607         btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
 608
 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         mutex_lock(&uuid_mutex);
 618         /* keep away write_all_supers() during the finishing procedure */
 619         mutex_lock(&fs_info->fs_devices->device_list_mutex);
 620         mutex_lock(&fs_info->chunk_mutex);
 621         btrfs_dev_replace_write_lock(dev_replace);
 622         dev_replace->replace_state =
 623                 scrub_ret ? BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED
 624                           : BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED;
 625         dev_replace->tgtdev = NULL;
 626         dev_replace->srcdev = NULL;
 627         dev_replace->time_stopped = get_seconds();
 628         dev_replace->item_needs_writeback = 1;
 629
 630         /* replace old device with new one in mapping tree */
 631         if (!scrub_ret) {
 632                 btrfs_dev_replace_update_device_in_mapping_tree(fs_info,
 633                                                                 src_device,
 634                                                                 tgt_device);
 635         } else {
 636                 btrfs_err_in_rcu(fs_info,
 637                                  "btrfs_scrub_dev(%s, %llu, %s) failed %d",
 638                                  btrfs_dev_name(src_device),
 639                                  src_device->devid,
 640                                  rcu_str_deref(tgt_device->name), scrub_ret);
 641                 btrfs_dev_replace_write_unlock(dev_replace);
 642                 mutex_unlock(&fs_info->chunk_mutex);
 643                 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 644                 mutex_unlock(&uuid_mutex);
 645                 btrfs_rm_dev_replace_blocked(fs_info);
 646                 if (tgt_device)
 647                         btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 648                 btrfs_rm_dev_replace_unblocked(fs_info);
 649                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 650
 651                 return scrub_ret;
 652         }
 653
 654         btrfs_info_in_rcu(fs_info,
 655                           "dev_replace from %s (devid %llu) to %s finished",
 656                           btrfs_dev_name(src_device),
 657                           src_device->devid,
 658                           rcu_str_deref(tgt_device->name));
 659         clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &tgt_device->dev_state);
 660         tgt_device->devid = src_device->devid;
 661         src_device->devid = BTRFS_DEV_REPLACE_DEVID;
 662         memcpy(uuid_tmp, tgt_device->uuid, sizeof(uuid_tmp));
 663         memcpy(tgt_device->uuid, src_device->uuid, sizeof(tgt_device->uuid));
 664         memcpy(src_device->uuid, uuid_tmp, sizeof(src_device->uuid));
 665         btrfs_device_set_total_bytes(tgt_device, src_device->total_bytes);
 666         btrfs_device_set_disk_total_bytes(tgt_device,
 667                                           src_device->disk_total_bytes);
 668         btrfs_device_set_bytes_used(tgt_device, src_device->bytes_used);
 669         ASSERT(list_empty(&src_device->resized_list));
 670         tgt_device->commit_total_bytes = src_device->commit_total_bytes;
 671         tgt_device->commit_bytes_used = src_device->bytes_used;
 672
 673         btrfs_assign_next_active_device(fs_info, src_device, tgt_device);
 674
 675         list_add(&tgt_device->dev_alloc_list, &fs_info->fs_devices->alloc_list);
 676         fs_info->fs_devices->rw_devices++;
 677
 678         btrfs_dev_replace_write_unlock(dev_replace);
 679
 680         btrfs_rm_dev_replace_blocked(fs_info);
 681
 682         btrfs_rm_dev_replace_remove_srcdev(fs_info, src_device);
 683
 684         btrfs_rm_dev_replace_unblocked(fs_info);
 685
 686         /*
 687          * this is again a consistent state where no dev_replace procedure
 688          * is running, the target device is part of the filesystem, the
 689          * source device is not part of the filesystem anymore and its 1st
 690          * superblock is scratched out so that it is no longer marked to
 691          * belong to this filesystem.
 692          */
 693         mutex_unlock(&fs_info->chunk_mutex);
 694         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
 695         mutex_unlock(&uuid_mutex);
 696
 697         /* replace the sysfs entry */
 698         btrfs_sysfs_rm_device_link(fs_info->fs_devices, src_device);
 699         btrfs_rm_dev_replace_free_srcdev(fs_info, src_device);
 700
 701         /* write back the superblocks */
 702         trans = btrfs_start_transaction(root, 0);
 703         if (!IS_ERR(trans))
 704                 btrfs_commit_transaction(trans);
 705
 706         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 707
 708         return 0;
 709 }
 710
 711 static void btrfs_dev_replace_update_device_in_mapping_tree(
 712                                                 struct btrfs_fs_info *fs_info,
 713                                                 struct btrfs_device *srcdev,
 714                                                 struct btrfs_device *tgtdev)
 715 {
 716         struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
 717         struct extent_map *em;
 718         struct map_lookup *map;
 719         u64 start = 0;
 720         int i;
 721
 722         write_lock(&em_tree->lock);
 723         do {
 724                 em = lookup_extent_mapping(em_tree, start, (u64)-1);
 725                 if (!em)
 726                         break;
 727                 map = em->map_lookup;
 728                 for (i = 0; i < map->num_stripes; i++)
 729                         if (srcdev == map->stripes[i].dev)
 730                                 map->stripes[i].dev = tgtdev;
 731                 start = em->start + em->len;
 732                 free_extent_map(em);
 733         } while (start);
 734         write_unlock(&em_tree->lock);
 735 }
 736
 737 /*
 738  * Read progress of device replace status according to the state and last
 739  * stored position. The value format is the same as for
 740  * btrfs_dev_replace::progress_1000
 741  */
 742 static u64 btrfs_dev_replace_progress(struct btrfs_fs_info *fs_info)
 743 {
 744         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 745         u64 ret = 0;
 746
 747         switch (dev_replace->replace_state) {
 748         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 749         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 750                 ret = 0;
 751                 break;
 752         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 753                 ret = 1000;
 754                 break;
 755         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 756         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 757                 ret = div64_u64(dev_replace->cursor_left,
 758                                 div_u64(btrfs_device_get_total_bytes(
 759                                                 dev_replace->srcdev), 1000));
 760                 break;
 761         }
 762
 763         return ret;
 764 }
 765
 766 void btrfs_dev_replace_status(struct btrfs_fs_info *fs_info,
 767                               struct btrfs_ioctl_dev_replace_args *args)
 768 {
 769         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 770
 771         btrfs_dev_replace_read_lock(dev_replace);
 772         /* even if !dev_replace_is_valid, the values are good enough for
 773          * the replace_status ioctl */
 774         args->result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 775         args->status.replace_state = dev_replace->replace_state;
 776         args->status.time_started = dev_replace->time_started;
 777         args->status.time_stopped = dev_replace->time_stopped;
 778         args->status.num_write_errors =
 779                 atomic64_read(&dev_replace->num_write_errors);
 780         args->status.num_uncorrectable_read_errors =
 781                 atomic64_read(&dev_replace->num_uncorrectable_read_errors);
 782         args->status.progress_1000 = btrfs_dev_replace_progress(fs_info);
 783         btrfs_dev_replace_read_unlock(dev_replace);
 784 }
 785
 786 int btrfs_dev_replace_cancel(struct btrfs_fs_info *fs_info)
 787 {
 788         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 789         struct btrfs_device *tgt_device = NULL;
 790         struct btrfs_device *src_device = NULL;
 791         struct btrfs_trans_handle *trans;
 792         struct btrfs_root *root = fs_info->tree_root;
 793         int result;
 794         int ret;
 795
 796         if (sb_rdonly(fs_info->sb))
 797                 return -EROFS;
 798
 799         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 800         btrfs_dev_replace_write_lock(dev_replace);
 801         switch (dev_replace->replace_state) {
 802         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 803         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 804         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 805                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NOT_STARTED;
 806                 btrfs_dev_replace_write_unlock(dev_replace);
 807                 goto leave;
 808         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 809         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 810                 result = BTRFS_IOCTL_DEV_REPLACE_RESULT_NO_ERROR;
 811                 tgt_device = dev_replace->tgtdev;
 812                 src_device = dev_replace->srcdev;
 813                 dev_replace->tgtdev = NULL;
 814                 dev_replace->srcdev = NULL;
 815                 break;
 816         }
 817         dev_replace->replace_state = BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED;
 818         dev_replace->time_stopped = get_seconds();
 819         dev_replace->item_needs_writeback = 1;
 820         btrfs_dev_replace_write_unlock(dev_replace);
 821         btrfs_scrub_cancel(fs_info);
 822
 823         trans = btrfs_start_transaction(root, 0);
 824         if (IS_ERR(trans)) {
 825                 mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 826                 return PTR_ERR(trans);
 827         }
 828         ret = btrfs_commit_transaction(trans);
 829         WARN_ON(ret);
 830
 831         btrfs_info_in_rcu(fs_info,
 832                 "dev_replace from %s (devid %llu) to %s canceled",
 833                 btrfs_dev_name(src_device), src_device->devid,
 834                 btrfs_dev_name(tgt_device));
 835
 836         if (tgt_device)
 837                 btrfs_destroy_dev_replace_tgtdev(fs_info, tgt_device);
 838
 839 leave:
 840         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 841         return result;
 842 }
 843
 844 void btrfs_dev_replace_suspend_for_unmount(struct btrfs_fs_info *fs_info)
 845 {
 846         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 847
 848         mutex_lock(&dev_replace->lock_finishing_cancel_unmount);
 849         btrfs_dev_replace_write_lock(dev_replace);
 850         switch (dev_replace->replace_state) {
 851         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 852         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 853         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 854         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 855                 break;
 856         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 857                 dev_replace->replace_state =
 858                         BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED;
 859                 dev_replace->time_stopped = get_seconds();
 860                 dev_replace->item_needs_writeback = 1;
 861                 btrfs_info(fs_info, "suspending dev_replace for unmount");
 862                 break;
 863         }
 864
 865         btrfs_dev_replace_write_unlock(dev_replace);
 866         mutex_unlock(&dev_replace->lock_finishing_cancel_unmount);
 867 }
 868
 869 /* resume dev_replace procedure that was interrupted by unmount */
 870 int btrfs_resume_dev_replace_async(struct btrfs_fs_info *fs_info)
 871 {
 872         struct task_struct *task;
 873         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 874
 875         btrfs_dev_replace_write_lock(dev_replace);
 876         switch (dev_replace->replace_state) {
 877         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 878         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 879         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 880                 btrfs_dev_replace_write_unlock(dev_replace);
 881                 return 0;
 882         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 883                 break;
 884         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 885                 dev_replace->replace_state =
 886                         BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED;
 887                 break;
 888         }
 889         if (!dev_replace->tgtdev || !dev_replace->tgtdev->bdev) {
 890                 btrfs_info(fs_info,
 891                            "cannot continue dev_replace, tgtdev is missing");
 892                 btrfs_info(fs_info,
 893                            "you may cancel the operation after 'mount -o degraded'");
 894                 btrfs_dev_replace_write_unlock(dev_replace);
 895                 return 0;
 896         }
 897         btrfs_dev_replace_write_unlock(dev_replace);
 898
 899         /*
 900          * This could collide with a paused balance, but the exclusive op logic
 901          * should never allow both to start and pause. We don't want to allow
 902          * dev-replace to start anyway.
 903          */
 904         if (test_and_set_bit(BTRFS_FS_EXCL_OP, &fs_info->flags)) {
 905                 btrfs_info(fs_info,
 906                 "cannot resume dev-replace, other exclusive operation running");
 907                 return 0;
 908         }
 909
 910         task = kthread_run(btrfs_dev_replace_kthread, fs_info, "btrfs-devrepl");
 911         return PTR_ERR_OR_ZERO(task);
 912 }
 913
 914 static int btrfs_dev_replace_kthread(void *data)
 915 {
 916         struct btrfs_fs_info *fs_info = data;
 917         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
 918         u64 progress;
 919         int ret;
 920
 921         progress = btrfs_dev_replace_progress(fs_info);
 922         progress = div_u64(progress, 10);
 923         btrfs_info_in_rcu(fs_info,
 924                 "continuing dev_replace from %s (devid %llu) to target %s @%u%%",
 925                 btrfs_dev_name(dev_replace->srcdev),
 926                 dev_replace->srcdev->devid,
 927                 btrfs_dev_name(dev_replace->tgtdev),
 928                 (unsigned int)progress);
 929
 930         ret = btrfs_scrub_dev(fs_info, dev_replace->srcdev->devid,
 931                               dev_replace->committed_cursor_left,
 932                               btrfs_device_get_total_bytes(dev_replace->srcdev),
 933                               &dev_replace->scrub_progress, 0, 1);
 934         ret = btrfs_dev_replace_finishing(fs_info, ret);
 935         WARN_ON(ret);
 936
 937         clear_bit(BTRFS_FS_EXCL_OP, &fs_info->flags);
 938         return 0;
 939 }
 940
 941 int btrfs_dev_replace_is_ongoing(struct btrfs_dev_replace *dev_replace)
 942 {
 943         if (!dev_replace->is_valid)
 944                 return 0;
 945
 946         switch (dev_replace->replace_state) {
 947         case BTRFS_IOCTL_DEV_REPLACE_STATE_NEVER_STARTED:
 948         case BTRFS_IOCTL_DEV_REPLACE_STATE_FINISHED:
 949         case BTRFS_IOCTL_DEV_REPLACE_STATE_CANCELED:
 950                 return 0;
 951         case BTRFS_IOCTL_DEV_REPLACE_STATE_STARTED:
 952         case BTRFS_IOCTL_DEV_REPLACE_STATE_SUSPENDED:
 953                 /*
 954                  * return true even if tgtdev is missing (this is
 955                  * something that can happen if the dev_replace
 956                  * procedure is suspended by an umount and then
 957                  * the tgtdev is missing (or "btrfs dev scan") was
 958                  * not called and the the filesystem is remounted
 959                  * in degraded state. This does not stop the
 960                  * dev_replace procedure. It needs to be canceled
 961                  * manually if the cancellation is wanted.
 962                  */
 963                 break;
 964         }
 965         return 1;
 966 }
 967
 968 void btrfs_dev_replace_read_lock(struct btrfs_dev_replace *dev_replace)
 969 {
 970         read_lock(&dev_replace->lock);
 971         atomic_inc(&dev_replace->read_locks);
 972 }
 973
 974 void btrfs_dev_replace_read_unlock(struct btrfs_dev_replace *dev_replace)
 975 {
 976         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
 977         atomic_dec(&dev_replace->read_locks);
 978         read_unlock(&dev_replace->lock);
 979 }
 980
 981 void btrfs_dev_replace_write_lock(struct btrfs_dev_replace *dev_replace)
 982 {
 983 again:
 984         wait_event(dev_replace->read_lock_wq,
 985                    atomic_read(&dev_replace->blocking_readers) == 0);
 986         write_lock(&dev_replace->lock);
 987         if (atomic_read(&dev_replace->blocking_readers)) {
 988                 write_unlock(&dev_replace->lock);
 989                 goto again;
 990         }
 991 }
 992
 993 void btrfs_dev_replace_write_unlock(struct btrfs_dev_replace *dev_replace)
 994 {
 995         ASSERT(atomic_read(&dev_replace->blocking_readers) == 0);
 996         write_unlock(&dev_replace->lock);
 997 }
 998
 999 /* inc blocking cnt and release read lock */
1000 void btrfs_dev_replace_set_lock_blocking(
1001                                         struct btrfs_dev_replace *dev_replace)
1002 {
1003         /* only set blocking for read lock */
1004         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1005         atomic_inc(&dev_replace->blocking_readers);
1006         read_unlock(&dev_replace->lock);
1007 }
1008
1009 /* acquire read lock and dec blocking cnt */
1010 void btrfs_dev_replace_clear_lock_blocking(
1011                                         struct btrfs_dev_replace *dev_replace)
1012 {
1013         /* only set blocking for read lock */
1014         ASSERT(atomic_read(&dev_replace->read_locks) > 0);
1015         ASSERT(atomic_read(&dev_replace->blocking_readers) > 0);
1016         read_lock(&dev_replace->lock);
1017         if (atomic_dec_and_test(&dev_replace->blocking_readers) &&
1018             waitqueue_active(&dev_replace->read_lock_wq))
1019                 wake_up(&dev_replace->read_lock_wq);
1020 }
1021
1022 void btrfs_bio_counter_inc_noblocked(struct btrfs_fs_info *fs_info)
1023 {
1024         percpu_counter_inc(&fs_info->bio_counter);
1025 }
1026
1027 void btrfs_bio_counter_sub(struct btrfs_fs_info *fs_info, s64 amount)
1028 {
1029         percpu_counter_sub(&fs_info->bio_counter, amount);
1030
1031         if (waitqueue_active(&fs_info->replace_wait))
1032                 wake_up(&fs_info->replace_wait);
1033 }
1034
1035 void btrfs_bio_counter_inc_blocked(struct btrfs_fs_info *fs_info)
1036 {
1037         while (1) {
1038                 percpu_counter_inc(&fs_info->bio_counter);
1039                 if (likely(!test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1040                                      &fs_info->fs_state)))
1041                         break;
1042
1043                 btrfs_bio_counter_dec(fs_info);
1044                 wait_event(fs_info->replace_wait,
1045                            !test_bit(BTRFS_FS_STATE_DEV_REPLACING,
1046                                      &fs_info->fs_state));
1047         }
1048 }