| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 4 | */ |
| 5 | |
| 6 | #include <linux/sched.h> |
| 7 | #include <linux/sched/mm.h> |
| 8 | #include <linux/slab.h> |
| 9 | #include <linux/ratelimit.h> |
| 10 | #include <linux/kthread.h> |
| 11 | #include <linux/semaphore.h> |
| 12 | #include <linux/uuid.h> |
| 13 | #include <linux/list_sort.h> |
| 14 | #include <linux/namei.h> |
| 15 | #include "misc.h" |
| 16 | #include "ctree.h" |
| 17 | #include "extent_map.h" |
| 18 | #include "disk-io.h" |
| 19 | #include "transaction.h" |
| 20 | #include "print-tree.h" |
| 21 | #include "volumes.h" |
| 22 | #include "raid56.h" |
| 23 | #include "rcu-string.h" |
| 24 | #include "dev-replace.h" |
| 25 | #include "sysfs.h" |
| 26 | #include "tree-checker.h" |
| 27 | #include "space-info.h" |
| 28 | #include "block-group.h" |
| 29 | #include "discard.h" |
| 30 | #include "zoned.h" |
| 31 | #include "fs.h" |
| 32 | #include "accessors.h" |
| 33 | #include "uuid-tree.h" |
| 34 | #include "ioctl.h" |
| 35 | #include "relocation.h" |
| 36 | #include "scrub.h" |
| 37 | #include "super.h" |
| 38 | #include "raid-stripe-tree.h" |
| 39 | |
| 40 | #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \ |
| 41 | BTRFS_BLOCK_GROUP_RAID10 | \ |
| 42 | BTRFS_BLOCK_GROUP_RAID56_MASK) |
| 43 | |
| 44 | struct btrfs_io_geometry { |
| 45 | u32 stripe_index; |
| 46 | u32 stripe_nr; |
| 47 | int mirror_num; |
| 48 | int num_stripes; |
| 49 | u64 stripe_offset; |
| 50 | u64 raid56_full_stripe_start; |
| 51 | int max_errors; |
| 52 | enum btrfs_map_op op; |
| 53 | }; |
| 54 | |
| 55 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
| 56 | [BTRFS_RAID_RAID10] = { |
| 57 | .sub_stripes = 2, |
| 58 | .dev_stripes = 1, |
| 59 | .devs_max = 0, /* 0 == as many as possible */ |
| 60 | .devs_min = 2, |
| 61 | .tolerated_failures = 1, |
| 62 | .devs_increment = 2, |
| 63 | .ncopies = 2, |
| 64 | .nparity = 0, |
| 65 | .raid_name = "raid10", |
| 66 | .bg_flag = BTRFS_BLOCK_GROUP_RAID10, |
| 67 | .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, |
| 68 | }, |
| 69 | [BTRFS_RAID_RAID1] = { |
| 70 | .sub_stripes = 1, |
| 71 | .dev_stripes = 1, |
| 72 | .devs_max = 2, |
| 73 | .devs_min = 2, |
| 74 | .tolerated_failures = 1, |
| 75 | .devs_increment = 2, |
| 76 | .ncopies = 2, |
| 77 | .nparity = 0, |
| 78 | .raid_name = "raid1", |
| 79 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1, |
| 80 | .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, |
| 81 | }, |
| 82 | [BTRFS_RAID_RAID1C3] = { |
| 83 | .sub_stripes = 1, |
| 84 | .dev_stripes = 1, |
| 85 | .devs_max = 3, |
| 86 | .devs_min = 3, |
| 87 | .tolerated_failures = 2, |
| 88 | .devs_increment = 3, |
| 89 | .ncopies = 3, |
| 90 | .nparity = 0, |
| 91 | .raid_name = "raid1c3", |
| 92 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3, |
| 93 | .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET, |
| 94 | }, |
| 95 | [BTRFS_RAID_RAID1C4] = { |
| 96 | .sub_stripes = 1, |
| 97 | .dev_stripes = 1, |
| 98 | .devs_max = 4, |
| 99 | .devs_min = 4, |
| 100 | .tolerated_failures = 3, |
| 101 | .devs_increment = 4, |
| 102 | .ncopies = 4, |
| 103 | .nparity = 0, |
| 104 | .raid_name = "raid1c4", |
| 105 | .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4, |
| 106 | .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET, |
| 107 | }, |
| 108 | [BTRFS_RAID_DUP] = { |
| 109 | .sub_stripes = 1, |
| 110 | .dev_stripes = 2, |
| 111 | .devs_max = 1, |
| 112 | .devs_min = 1, |
| 113 | .tolerated_failures = 0, |
| 114 | .devs_increment = 1, |
| 115 | .ncopies = 2, |
| 116 | .nparity = 0, |
| 117 | .raid_name = "dup", |
| 118 | .bg_flag = BTRFS_BLOCK_GROUP_DUP, |
| 119 | .mindev_error = 0, |
| 120 | }, |
| 121 | [BTRFS_RAID_RAID0] = { |
| 122 | .sub_stripes = 1, |
| 123 | .dev_stripes = 1, |
| 124 | .devs_max = 0, |
| 125 | .devs_min = 1, |
| 126 | .tolerated_failures = 0, |
| 127 | .devs_increment = 1, |
| 128 | .ncopies = 1, |
| 129 | .nparity = 0, |
| 130 | .raid_name = "raid0", |
| 131 | .bg_flag = BTRFS_BLOCK_GROUP_RAID0, |
| 132 | .mindev_error = 0, |
| 133 | }, |
| 134 | [BTRFS_RAID_SINGLE] = { |
| 135 | .sub_stripes = 1, |
| 136 | .dev_stripes = 1, |
| 137 | .devs_max = 1, |
| 138 | .devs_min = 1, |
| 139 | .tolerated_failures = 0, |
| 140 | .devs_increment = 1, |
| 141 | .ncopies = 1, |
| 142 | .nparity = 0, |
| 143 | .raid_name = "single", |
| 144 | .bg_flag = 0, |
| 145 | .mindev_error = 0, |
| 146 | }, |
| 147 | [BTRFS_RAID_RAID5] = { |
| 148 | .sub_stripes = 1, |
| 149 | .dev_stripes = 1, |
| 150 | .devs_max = 0, |
| 151 | .devs_min = 2, |
| 152 | .tolerated_failures = 1, |
| 153 | .devs_increment = 1, |
| 154 | .ncopies = 1, |
| 155 | .nparity = 1, |
| 156 | .raid_name = "raid5", |
| 157 | .bg_flag = BTRFS_BLOCK_GROUP_RAID5, |
| 158 | .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, |
| 159 | }, |
| 160 | [BTRFS_RAID_RAID6] = { |
| 161 | .sub_stripes = 1, |
| 162 | .dev_stripes = 1, |
| 163 | .devs_max = 0, |
| 164 | .devs_min = 3, |
| 165 | .tolerated_failures = 2, |
| 166 | .devs_increment = 1, |
| 167 | .ncopies = 1, |
| 168 | .nparity = 2, |
| 169 | .raid_name = "raid6", |
| 170 | .bg_flag = BTRFS_BLOCK_GROUP_RAID6, |
| 171 | .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, |
| 172 | }, |
| 173 | }; |
| 174 | |
| 175 | /* |
| 176 | * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which |
| 177 | * can be used as index to access btrfs_raid_array[]. |
| 178 | */ |
| 179 | enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags) |
| 180 | { |
| 181 | const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK); |
| 182 | |
| 183 | if (!profile) |
| 184 | return BTRFS_RAID_SINGLE; |
| 185 | |
| 186 | return BTRFS_BG_FLAG_TO_INDEX(profile); |
| 187 | } |
| 188 | |
| 189 | const char *btrfs_bg_type_to_raid_name(u64 flags) |
| 190 | { |
| 191 | const int index = btrfs_bg_flags_to_raid_index(flags); |
| 192 | |
| 193 | if (index >= BTRFS_NR_RAID_TYPES) |
| 194 | return NULL; |
| 195 | |
| 196 | return btrfs_raid_array[index].raid_name; |
| 197 | } |
| 198 | |
| 199 | int btrfs_nr_parity_stripes(u64 type) |
| 200 | { |
| 201 | enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type); |
| 202 | |
| 203 | return btrfs_raid_array[index].nparity; |
| 204 | } |
| 205 | |
| 206 | /* |
| 207 | * Fill @buf with textual description of @bg_flags, no more than @size_buf |
| 208 | * bytes including terminating null byte. |
| 209 | */ |
| 210 | void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf) |
| 211 | { |
| 212 | int i; |
| 213 | int ret; |
| 214 | char *bp = buf; |
| 215 | u64 flags = bg_flags; |
| 216 | u32 size_bp = size_buf; |
| 217 | |
| 218 | if (!flags) { |
| 219 | strcpy(bp, "NONE"); |
| 220 | return; |
| 221 | } |
| 222 | |
| 223 | #define DESCRIBE_FLAG(flag, desc) \ |
| 224 | do { \ |
| 225 | if (flags & (flag)) { \ |
| 226 | ret = snprintf(bp, size_bp, "%s|", (desc)); \ |
| 227 | if (ret < 0 || ret >= size_bp) \ |
| 228 | goto out_overflow; \ |
| 229 | size_bp -= ret; \ |
| 230 | bp += ret; \ |
| 231 | flags &= ~(flag); \ |
| 232 | } \ |
| 233 | } while (0) |
| 234 | |
| 235 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data"); |
| 236 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system"); |
| 237 | DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata"); |
| 238 | |
| 239 | DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single"); |
| 240 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
| 241 | DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag, |
| 242 | btrfs_raid_array[i].raid_name); |
| 243 | #undef DESCRIBE_FLAG |
| 244 | |
| 245 | if (flags) { |
| 246 | ret = snprintf(bp, size_bp, "0x%llx|", flags); |
| 247 | size_bp -= ret; |
| 248 | } |
| 249 | |
| 250 | if (size_bp < size_buf) |
| 251 | buf[size_buf - size_bp - 1] = '\0'; /* remove last | */ |
| 252 | |
| 253 | /* |
| 254 | * The text is trimmed, it's up to the caller to provide sufficiently |
| 255 | * large buffer |
| 256 | */ |
| 257 | out_overflow:; |
| 258 | } |
| 259 | |
| 260 | static int init_first_rw_device(struct btrfs_trans_handle *trans); |
| 261 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info); |
| 262 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
| 263 | |
| 264 | /* |
| 265 | * Device locking |
| 266 | * ============== |
| 267 | * |
| 268 | * There are several mutexes that protect manipulation of devices and low-level |
| 269 | * structures like chunks but not block groups, extents or files |
| 270 | * |
| 271 | * uuid_mutex (global lock) |
| 272 | * ------------------------ |
| 273 | * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from |
| 274 | * the SCAN_DEV ioctl registration or from mount either implicitly (the first |
| 275 | * device) or requested by the device= mount option |
| 276 | * |
| 277 | * the mutex can be very coarse and can cover long-running operations |
| 278 | * |
| 279 | * protects: updates to fs_devices counters like missing devices, rw devices, |
| 280 | * seeding, structure cloning, opening/closing devices at mount/umount time |
| 281 | * |
| 282 | * global::fs_devs - add, remove, updates to the global list |
| 283 | * |
| 284 | * does not protect: manipulation of the fs_devices::devices list in general |
| 285 | * but in mount context it could be used to exclude list modifications by eg. |
| 286 | * scan ioctl |
| 287 | * |
| 288 | * btrfs_device::name - renames (write side), read is RCU |
| 289 | * |
| 290 | * fs_devices::device_list_mutex (per-fs, with RCU) |
| 291 | * ------------------------------------------------ |
| 292 | * protects updates to fs_devices::devices, ie. adding and deleting |
| 293 | * |
| 294 | * simple list traversal with read-only actions can be done with RCU protection |
| 295 | * |
| 296 | * may be used to exclude some operations from running concurrently without any |
| 297 | * modifications to the list (see write_all_supers) |
| 298 | * |
| 299 | * Is not required at mount and close times, because our device list is |
| 300 | * protected by the uuid_mutex at that point. |
| 301 | * |
| 302 | * balance_mutex |
| 303 | * ------------- |
| 304 | * protects balance structures (status, state) and context accessed from |
| 305 | * several places (internally, ioctl) |
| 306 | * |
| 307 | * chunk_mutex |
| 308 | * ----------- |
| 309 | * protects chunks, adding or removing during allocation, trim or when a new |
| 310 | * device is added/removed. Additionally it also protects post_commit_list of |
| 311 | * individual devices, since they can be added to the transaction's |
| 312 | * post_commit_list only with chunk_mutex held. |
| 313 | * |
| 314 | * cleaner_mutex |
| 315 | * ------------- |
| 316 | * a big lock that is held by the cleaner thread and prevents running subvolume |
| 317 | * cleaning together with relocation or delayed iputs |
| 318 | * |
| 319 | * |
| 320 | * Lock nesting |
| 321 | * ============ |
| 322 | * |
| 323 | * uuid_mutex |
| 324 | * device_list_mutex |
| 325 | * chunk_mutex |
| 326 | * balance_mutex |
| 327 | * |
| 328 | * |
| 329 | * Exclusive operations |
| 330 | * ==================== |
| 331 | * |
| 332 | * Maintains the exclusivity of the following operations that apply to the |
| 333 | * whole filesystem and cannot run in parallel. |
| 334 | * |
| 335 | * - Balance (*) |
| 336 | * - Device add |
| 337 | * - Device remove |
| 338 | * - Device replace (*) |
| 339 | * - Resize |
| 340 | * |
| 341 | * The device operations (as above) can be in one of the following states: |
| 342 | * |
| 343 | * - Running state |
| 344 | * - Paused state |
| 345 | * - Completed state |
| 346 | * |
| 347 | * Only device operations marked with (*) can go into the Paused state for the |
| 348 | * following reasons: |
| 349 | * |
| 350 | * - ioctl (only Balance can be Paused through ioctl) |
| 351 | * - filesystem remounted as read-only |
| 352 | * - filesystem unmounted and mounted as read-only |
| 353 | * - system power-cycle and filesystem mounted as read-only |
| 354 | * - filesystem or device errors leading to forced read-only |
| 355 | * |
| 356 | * The status of exclusive operation is set and cleared atomically. |
| 357 | * During the course of Paused state, fs_info::exclusive_operation remains set. |
| 358 | * A device operation in Paused or Running state can be canceled or resumed |
| 359 | * either by ioctl (Balance only) or when remounted as read-write. |
| 360 | * The exclusive status is cleared when the device operation is canceled or |
| 361 | * completed. |
| 362 | */ |
| 363 | |
| 364 | DEFINE_MUTEX(uuid_mutex); |
| 365 | static LIST_HEAD(fs_uuids); |
| 366 | struct list_head * __attribute_const__ btrfs_get_fs_uuids(void) |
| 367 | { |
| 368 | return &fs_uuids; |
| 369 | } |
| 370 | |
| 371 | /* |
| 372 | * Allocate new btrfs_fs_devices structure identified by a fsid. |
| 373 | * |
| 374 | * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to |
| 375 | * fs_devices::metadata_fsid |
| 376 | * |
| 377 | * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR(). |
| 378 | * The returned struct is not linked onto any lists and can be destroyed with |
| 379 | * kfree() right away. |
| 380 | */ |
| 381 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) |
| 382 | { |
| 383 | struct btrfs_fs_devices *fs_devs; |
| 384 | |
| 385 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
| 386 | if (!fs_devs) |
| 387 | return ERR_PTR(-ENOMEM); |
| 388 | |
| 389 | mutex_init(&fs_devs->device_list_mutex); |
| 390 | |
| 391 | INIT_LIST_HEAD(&fs_devs->devices); |
| 392 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
| 393 | INIT_LIST_HEAD(&fs_devs->fs_list); |
| 394 | INIT_LIST_HEAD(&fs_devs->seed_list); |
| 395 | |
| 396 | if (fsid) { |
| 397 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); |
| 398 | memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE); |
| 399 | } |
| 400 | |
| 401 | return fs_devs; |
| 402 | } |
| 403 | |
| 404 | static void btrfs_free_device(struct btrfs_device *device) |
| 405 | { |
| 406 | WARN_ON(!list_empty(&device->post_commit_list)); |
| 407 | rcu_string_free(device->name); |
| 408 | extent_io_tree_release(&device->alloc_state); |
| 409 | btrfs_destroy_dev_zone_info(device); |
| 410 | kfree(device); |
| 411 | } |
| 412 | |
| 413 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
| 414 | { |
| 415 | struct btrfs_device *device; |
| 416 | |
| 417 | WARN_ON(fs_devices->opened); |
| 418 | while (!list_empty(&fs_devices->devices)) { |
| 419 | device = list_entry(fs_devices->devices.next, |
| 420 | struct btrfs_device, dev_list); |
| 421 | list_del(&device->dev_list); |
| 422 | btrfs_free_device(device); |
| 423 | } |
| 424 | kfree(fs_devices); |
| 425 | } |
| 426 | |
| 427 | void __exit btrfs_cleanup_fs_uuids(void) |
| 428 | { |
| 429 | struct btrfs_fs_devices *fs_devices; |
| 430 | |
| 431 | while (!list_empty(&fs_uuids)) { |
| 432 | fs_devices = list_entry(fs_uuids.next, |
| 433 | struct btrfs_fs_devices, fs_list); |
| 434 | list_del(&fs_devices->fs_list); |
| 435 | free_fs_devices(fs_devices); |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices, |
| 440 | const u8 *fsid, const u8 *metadata_fsid) |
| 441 | { |
| 442 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0) |
| 443 | return false; |
| 444 | |
| 445 | if (!metadata_fsid) |
| 446 | return true; |
| 447 | |
| 448 | if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0) |
| 449 | return false; |
| 450 | |
| 451 | return true; |
| 452 | } |
| 453 | |
| 454 | static noinline struct btrfs_fs_devices *find_fsid( |
| 455 | const u8 *fsid, const u8 *metadata_fsid) |
| 456 | { |
| 457 | struct btrfs_fs_devices *fs_devices; |
| 458 | |
| 459 | ASSERT(fsid); |
| 460 | |
| 461 | /* Handle non-split brain cases */ |
| 462 | list_for_each_entry(fs_devices, &fs_uuids, fs_list) { |
| 463 | if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid)) |
| 464 | return fs_devices; |
| 465 | } |
| 466 | return NULL; |
| 467 | } |
| 468 | |
| 469 | static int |
| 470 | btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder, |
| 471 | int flush, struct bdev_handle **bdev_handle, |
| 472 | struct btrfs_super_block **disk_super) |
| 473 | { |
| 474 | struct block_device *bdev; |
| 475 | int ret; |
| 476 | |
| 477 | *bdev_handle = bdev_open_by_path(device_path, flags, holder, NULL); |
| 478 | |
| 479 | if (IS_ERR(*bdev_handle)) { |
| 480 | ret = PTR_ERR(*bdev_handle); |
| 481 | goto error; |
| 482 | } |
| 483 | bdev = (*bdev_handle)->bdev; |
| 484 | |
| 485 | if (flush) |
| 486 | sync_blockdev(bdev); |
| 487 | ret = set_blocksize(bdev, BTRFS_BDEV_BLOCKSIZE); |
| 488 | if (ret) { |
| 489 | bdev_release(*bdev_handle); |
| 490 | goto error; |
| 491 | } |
| 492 | invalidate_bdev(bdev); |
| 493 | *disk_super = btrfs_read_dev_super(bdev); |
| 494 | if (IS_ERR(*disk_super)) { |
| 495 | ret = PTR_ERR(*disk_super); |
| 496 | bdev_release(*bdev_handle); |
| 497 | goto error; |
| 498 | } |
| 499 | |
| 500 | return 0; |
| 501 | |
| 502 | error: |
| 503 | *bdev_handle = NULL; |
| 504 | return ret; |
| 505 | } |
| 506 | |
| 507 | /* |
| 508 | * Search and remove all stale devices (which are not mounted). When both |
| 509 | * inputs are NULL, it will search and release all stale devices. |
| 510 | * |
| 511 | * @devt: Optional. When provided will it release all unmounted devices |
| 512 | * matching this devt only. |
| 513 | * @skip_device: Optional. Will skip this device when searching for the stale |
| 514 | * devices. |
| 515 | * |
| 516 | * Return: 0 for success or if @devt is 0. |
| 517 | * -EBUSY if @devt is a mounted device. |
| 518 | * -ENOENT if @devt does not match any device in the list. |
| 519 | */ |
| 520 | static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device) |
| 521 | { |
| 522 | struct btrfs_fs_devices *fs_devices, *tmp_fs_devices; |
| 523 | struct btrfs_device *device, *tmp_device; |
| 524 | int ret; |
| 525 | bool freed = false; |
| 526 | |
| 527 | lockdep_assert_held(&uuid_mutex); |
| 528 | |
| 529 | /* Return good status if there is no instance of devt. */ |
| 530 | ret = 0; |
| 531 | list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) { |
| 532 | |
| 533 | mutex_lock(&fs_devices->device_list_mutex); |
| 534 | list_for_each_entry_safe(device, tmp_device, |
| 535 | &fs_devices->devices, dev_list) { |
| 536 | if (skip_device && skip_device == device) |
| 537 | continue; |
| 538 | if (devt && devt != device->devt) |
| 539 | continue; |
| 540 | if (fs_devices->opened) { |
| 541 | if (devt) |
| 542 | ret = -EBUSY; |
| 543 | break; |
| 544 | } |
| 545 | |
| 546 | /* delete the stale device */ |
| 547 | fs_devices->num_devices--; |
| 548 | list_del(&device->dev_list); |
| 549 | btrfs_free_device(device); |
| 550 | |
| 551 | freed = true; |
| 552 | } |
| 553 | mutex_unlock(&fs_devices->device_list_mutex); |
| 554 | |
| 555 | if (fs_devices->num_devices == 0) { |
| 556 | btrfs_sysfs_remove_fsid(fs_devices); |
| 557 | list_del(&fs_devices->fs_list); |
| 558 | free_fs_devices(fs_devices); |
| 559 | } |
| 560 | } |
| 561 | |
| 562 | /* If there is at least one freed device return 0. */ |
| 563 | if (freed) |
| 564 | return 0; |
| 565 | |
| 566 | return ret; |
| 567 | } |
| 568 | |
| 569 | static struct btrfs_fs_devices *find_fsid_by_device( |
| 570 | struct btrfs_super_block *disk_super, |
| 571 | dev_t devt, bool *same_fsid_diff_dev) |
| 572 | { |
| 573 | struct btrfs_fs_devices *fsid_fs_devices; |
| 574 | struct btrfs_fs_devices *devt_fs_devices; |
| 575 | const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
| 576 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); |
| 577 | bool found_by_devt = false; |
| 578 | |
| 579 | /* Find the fs_device by the usual method, if found use it. */ |
| 580 | fsid_fs_devices = find_fsid(disk_super->fsid, |
| 581 | has_metadata_uuid ? disk_super->metadata_uuid : NULL); |
| 582 | |
| 583 | /* The temp_fsid feature is supported only with single device filesystem. */ |
| 584 | if (btrfs_super_num_devices(disk_super) != 1) |
| 585 | return fsid_fs_devices; |
| 586 | |
| 587 | /* |
| 588 | * A seed device is an integral component of the sprout device, which |
| 589 | * functions as a multi-device filesystem. So, temp-fsid feature is |
| 590 | * not supported. |
| 591 | */ |
| 592 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) |
| 593 | return fsid_fs_devices; |
| 594 | |
| 595 | /* Try to find a fs_devices by matching devt. */ |
| 596 | list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) { |
| 597 | struct btrfs_device *device; |
| 598 | |
| 599 | list_for_each_entry(device, &devt_fs_devices->devices, dev_list) { |
| 600 | if (device->devt == devt) { |
| 601 | found_by_devt = true; |
| 602 | break; |
| 603 | } |
| 604 | } |
| 605 | if (found_by_devt) |
| 606 | break; |
| 607 | } |
| 608 | |
| 609 | if (found_by_devt) { |
| 610 | /* Existing device. */ |
| 611 | if (fsid_fs_devices == NULL) { |
| 612 | if (devt_fs_devices->opened == 0) { |
| 613 | /* Stale device. */ |
| 614 | return NULL; |
| 615 | } else { |
| 616 | /* temp_fsid is mounting a subvol. */ |
| 617 | return devt_fs_devices; |
| 618 | } |
| 619 | } else { |
| 620 | /* Regular or temp_fsid device mounting a subvol. */ |
| 621 | return devt_fs_devices; |
| 622 | } |
| 623 | } else { |
| 624 | /* New device. */ |
| 625 | if (fsid_fs_devices == NULL) { |
| 626 | return NULL; |
| 627 | } else { |
| 628 | /* sb::fsid is already used create a new temp_fsid. */ |
| 629 | *same_fsid_diff_dev = true; |
| 630 | return NULL; |
| 631 | } |
| 632 | } |
| 633 | |
| 634 | /* Not reached. */ |
| 635 | } |
| 636 | |
| 637 | /* |
| 638 | * This is only used on mount, and we are protected from competing things |
| 639 | * messing with our fs_devices by the uuid_mutex, thus we do not need the |
| 640 | * fs_devices->device_list_mutex here. |
| 641 | */ |
| 642 | static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices, |
| 643 | struct btrfs_device *device, blk_mode_t flags, |
| 644 | void *holder) |
| 645 | { |
| 646 | struct bdev_handle *bdev_handle; |
| 647 | struct btrfs_super_block *disk_super; |
| 648 | u64 devid; |
| 649 | int ret; |
| 650 | |
| 651 | if (device->bdev) |
| 652 | return -EINVAL; |
| 653 | if (!device->name) |
| 654 | return -EINVAL; |
| 655 | |
| 656 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, |
| 657 | &bdev_handle, &disk_super); |
| 658 | if (ret) |
| 659 | return ret; |
| 660 | |
| 661 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
| 662 | if (devid != device->devid) |
| 663 | goto error_free_page; |
| 664 | |
| 665 | if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE)) |
| 666 | goto error_free_page; |
| 667 | |
| 668 | device->generation = btrfs_super_generation(disk_super); |
| 669 | |
| 670 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
| 671 | if (btrfs_super_incompat_flags(disk_super) & |
| 672 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID) { |
| 673 | pr_err( |
| 674 | "BTRFS: Invalid seeding and uuid-changed device detected\n"); |
| 675 | goto error_free_page; |
| 676 | } |
| 677 | |
| 678 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 679 | fs_devices->seeding = true; |
| 680 | } else { |
| 681 | if (bdev_read_only(bdev_handle->bdev)) |
| 682 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 683 | else |
| 684 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 685 | } |
| 686 | |
| 687 | if (!bdev_nonrot(bdev_handle->bdev)) |
| 688 | fs_devices->rotating = true; |
| 689 | |
| 690 | if (bdev_max_discard_sectors(bdev_handle->bdev)) |
| 691 | fs_devices->discardable = true; |
| 692 | |
| 693 | device->bdev_handle = bdev_handle; |
| 694 | device->bdev = bdev_handle->bdev; |
| 695 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
| 696 | |
| 697 | fs_devices->open_devices++; |
| 698 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
| 699 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
| 700 | fs_devices->rw_devices++; |
| 701 | list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list); |
| 702 | } |
| 703 | btrfs_release_disk_super(disk_super); |
| 704 | |
| 705 | return 0; |
| 706 | |
| 707 | error_free_page: |
| 708 | btrfs_release_disk_super(disk_super); |
| 709 | bdev_release(bdev_handle); |
| 710 | |
| 711 | return -EINVAL; |
| 712 | } |
| 713 | |
| 714 | u8 *btrfs_sb_fsid_ptr(struct btrfs_super_block *sb) |
| 715 | { |
| 716 | bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) & |
| 717 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); |
| 718 | |
| 719 | return has_metadata_uuid ? sb->metadata_uuid : sb->fsid; |
| 720 | } |
| 721 | |
| 722 | /* |
| 723 | * Add new device to list of registered devices |
| 724 | * |
| 725 | * Returns: |
| 726 | * device pointer which was just added or updated when successful |
| 727 | * error pointer when failed |
| 728 | */ |
| 729 | static noinline struct btrfs_device *device_list_add(const char *path, |
| 730 | struct btrfs_super_block *disk_super, |
| 731 | bool *new_device_added) |
| 732 | { |
| 733 | struct btrfs_device *device; |
| 734 | struct btrfs_fs_devices *fs_devices = NULL; |
| 735 | struct rcu_string *name; |
| 736 | u64 found_transid = btrfs_super_generation(disk_super); |
| 737 | u64 devid = btrfs_stack_device_id(&disk_super->dev_item); |
| 738 | dev_t path_devt; |
| 739 | int error; |
| 740 | bool same_fsid_diff_dev = false; |
| 741 | bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) & |
| 742 | BTRFS_FEATURE_INCOMPAT_METADATA_UUID); |
| 743 | |
| 744 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) { |
| 745 | btrfs_err(NULL, |
| 746 | "device %s has incomplete metadata_uuid change, please use btrfstune to complete", |
| 747 | path); |
| 748 | return ERR_PTR(-EAGAIN); |
| 749 | } |
| 750 | |
| 751 | error = lookup_bdev(path, &path_devt); |
| 752 | if (error) { |
| 753 | btrfs_err(NULL, "failed to lookup block device for path %s: %d", |
| 754 | path, error); |
| 755 | return ERR_PTR(error); |
| 756 | } |
| 757 | |
| 758 | fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev); |
| 759 | |
| 760 | if (!fs_devices) { |
| 761 | fs_devices = alloc_fs_devices(disk_super->fsid); |
| 762 | if (IS_ERR(fs_devices)) |
| 763 | return ERR_CAST(fs_devices); |
| 764 | |
| 765 | if (has_metadata_uuid) |
| 766 | memcpy(fs_devices->metadata_uuid, |
| 767 | disk_super->metadata_uuid, BTRFS_FSID_SIZE); |
| 768 | |
| 769 | if (same_fsid_diff_dev) { |
| 770 | generate_random_uuid(fs_devices->fsid); |
| 771 | fs_devices->temp_fsid = true; |
| 772 | pr_info("BTRFS: device %s using temp-fsid %pU\n", |
| 773 | path, fs_devices->fsid); |
| 774 | } |
| 775 | |
| 776 | mutex_lock(&fs_devices->device_list_mutex); |
| 777 | list_add(&fs_devices->fs_list, &fs_uuids); |
| 778 | |
| 779 | device = NULL; |
| 780 | } else { |
| 781 | struct btrfs_dev_lookup_args args = { |
| 782 | .devid = devid, |
| 783 | .uuid = disk_super->dev_item.uuid, |
| 784 | }; |
| 785 | |
| 786 | mutex_lock(&fs_devices->device_list_mutex); |
| 787 | device = btrfs_find_device(fs_devices, &args); |
| 788 | |
| 789 | if (found_transid > fs_devices->latest_generation) { |
| 790 | memcpy(fs_devices->fsid, disk_super->fsid, |
| 791 | BTRFS_FSID_SIZE); |
| 792 | memcpy(fs_devices->metadata_uuid, |
| 793 | btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE); |
| 794 | } |
| 795 | } |
| 796 | |
| 797 | if (!device) { |
| 798 | unsigned int nofs_flag; |
| 799 | |
| 800 | if (fs_devices->opened) { |
| 801 | btrfs_err(NULL, |
| 802 | "device %s belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)", |
| 803 | path, fs_devices->fsid, current->comm, |
| 804 | task_pid_nr(current)); |
| 805 | mutex_unlock(&fs_devices->device_list_mutex); |
| 806 | return ERR_PTR(-EBUSY); |
| 807 | } |
| 808 | |
| 809 | nofs_flag = memalloc_nofs_save(); |
| 810 | device = btrfs_alloc_device(NULL, &devid, |
| 811 | disk_super->dev_item.uuid, path); |
| 812 | memalloc_nofs_restore(nofs_flag); |
| 813 | if (IS_ERR(device)) { |
| 814 | mutex_unlock(&fs_devices->device_list_mutex); |
| 815 | /* we can safely leave the fs_devices entry around */ |
| 816 | return device; |
| 817 | } |
| 818 | |
| 819 | device->devt = path_devt; |
| 820 | |
| 821 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
| 822 | fs_devices->num_devices++; |
| 823 | |
| 824 | device->fs_devices = fs_devices; |
| 825 | *new_device_added = true; |
| 826 | |
| 827 | if (disk_super->label[0]) |
| 828 | pr_info( |
| 829 | "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n", |
| 830 | disk_super->label, devid, found_transid, path, |
| 831 | current->comm, task_pid_nr(current)); |
| 832 | else |
| 833 | pr_info( |
| 834 | "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n", |
| 835 | disk_super->fsid, devid, found_transid, path, |
| 836 | current->comm, task_pid_nr(current)); |
| 837 | |
| 838 | } else if (!device->name || strcmp(device->name->str, path)) { |
| 839 | /* |
| 840 | * When FS is already mounted. |
| 841 | * 1. If you are here and if the device->name is NULL that |
| 842 | * means this device was missing at time of FS mount. |
| 843 | * 2. If you are here and if the device->name is different |
| 844 | * from 'path' that means either |
| 845 | * a. The same device disappeared and reappeared with |
| 846 | * different name. or |
| 847 | * b. The missing-disk-which-was-replaced, has |
| 848 | * reappeared now. |
| 849 | * |
| 850 | * We must allow 1 and 2a above. But 2b would be a spurious |
| 851 | * and unintentional. |
| 852 | * |
| 853 | * Further in case of 1 and 2a above, the disk at 'path' |
| 854 | * would have missed some transaction when it was away and |
| 855 | * in case of 2a the stale bdev has to be updated as well. |
| 856 | * 2b must not be allowed at all time. |
| 857 | */ |
| 858 | |
| 859 | /* |
| 860 | * For now, we do allow update to btrfs_fs_device through the |
| 861 | * btrfs dev scan cli after FS has been mounted. We're still |
| 862 | * tracking a problem where systems fail mount by subvolume id |
| 863 | * when we reject replacement on a mounted FS. |
| 864 | */ |
| 865 | if (!fs_devices->opened && found_transid < device->generation) { |
| 866 | /* |
| 867 | * That is if the FS is _not_ mounted and if you |
| 868 | * are here, that means there is more than one |
| 869 | * disk with same uuid and devid.We keep the one |
| 870 | * with larger generation number or the last-in if |
| 871 | * generation are equal. |
| 872 | */ |
| 873 | mutex_unlock(&fs_devices->device_list_mutex); |
| 874 | btrfs_err(NULL, |
| 875 | "device %s already registered with a higher generation, found %llu expect %llu", |
| 876 | path, found_transid, device->generation); |
| 877 | return ERR_PTR(-EEXIST); |
| 878 | } |
| 879 | |
| 880 | /* |
| 881 | * We are going to replace the device path for a given devid, |
| 882 | * make sure it's the same device if the device is mounted |
| 883 | * |
| 884 | * NOTE: the device->fs_info may not be reliable here so pass |
| 885 | * in a NULL to message helpers instead. This avoids a possible |
| 886 | * use-after-free when the fs_info and fs_info->sb are already |
| 887 | * torn down. |
| 888 | */ |
| 889 | if (device->bdev) { |
| 890 | if (device->devt != path_devt) { |
| 891 | mutex_unlock(&fs_devices->device_list_mutex); |
| 892 | btrfs_warn_in_rcu(NULL, |
| 893 | "duplicate device %s devid %llu generation %llu scanned by %s (%d)", |
| 894 | path, devid, found_transid, |
| 895 | current->comm, |
| 896 | task_pid_nr(current)); |
| 897 | return ERR_PTR(-EEXIST); |
| 898 | } |
| 899 | btrfs_info_in_rcu(NULL, |
| 900 | "devid %llu device path %s changed to %s scanned by %s (%d)", |
| 901 | devid, btrfs_dev_name(device), |
| 902 | path, current->comm, |
| 903 | task_pid_nr(current)); |
| 904 | } |
| 905 | |
| 906 | name = rcu_string_strdup(path, GFP_NOFS); |
| 907 | if (!name) { |
| 908 | mutex_unlock(&fs_devices->device_list_mutex); |
| 909 | return ERR_PTR(-ENOMEM); |
| 910 | } |
| 911 | rcu_string_free(device->name); |
| 912 | rcu_assign_pointer(device->name, name); |
| 913 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
| 914 | fs_devices->missing_devices--; |
| 915 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
| 916 | } |
| 917 | device->devt = path_devt; |
| 918 | } |
| 919 | |
| 920 | /* |
| 921 | * Unmount does not free the btrfs_device struct but would zero |
| 922 | * generation along with most of the other members. So just update |
| 923 | * it back. We need it to pick the disk with largest generation |
| 924 | * (as above). |
| 925 | */ |
| 926 | if (!fs_devices->opened) { |
| 927 | device->generation = found_transid; |
| 928 | fs_devices->latest_generation = max_t(u64, found_transid, |
| 929 | fs_devices->latest_generation); |
| 930 | } |
| 931 | |
| 932 | fs_devices->total_devices = btrfs_super_num_devices(disk_super); |
| 933 | |
| 934 | mutex_unlock(&fs_devices->device_list_mutex); |
| 935 | return device; |
| 936 | } |
| 937 | |
| 938 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
| 939 | { |
| 940 | struct btrfs_fs_devices *fs_devices; |
| 941 | struct btrfs_device *device; |
| 942 | struct btrfs_device *orig_dev; |
| 943 | int ret = 0; |
| 944 | |
| 945 | lockdep_assert_held(&uuid_mutex); |
| 946 | |
| 947 | fs_devices = alloc_fs_devices(orig->fsid); |
| 948 | if (IS_ERR(fs_devices)) |
| 949 | return fs_devices; |
| 950 | |
| 951 | fs_devices->total_devices = orig->total_devices; |
| 952 | |
| 953 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
| 954 | const char *dev_path = NULL; |
| 955 | |
| 956 | /* |
| 957 | * This is ok to do without RCU read locked because we hold the |
| 958 | * uuid mutex so nothing we touch in here is going to disappear. |
| 959 | */ |
| 960 | if (orig_dev->name) |
| 961 | dev_path = orig_dev->name->str; |
| 962 | |
| 963 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
| 964 | orig_dev->uuid, dev_path); |
| 965 | if (IS_ERR(device)) { |
| 966 | ret = PTR_ERR(device); |
| 967 | goto error; |
| 968 | } |
| 969 | |
| 970 | if (orig_dev->zone_info) { |
| 971 | struct btrfs_zoned_device_info *zone_info; |
| 972 | |
| 973 | zone_info = btrfs_clone_dev_zone_info(orig_dev); |
| 974 | if (!zone_info) { |
| 975 | btrfs_free_device(device); |
| 976 | ret = -ENOMEM; |
| 977 | goto error; |
| 978 | } |
| 979 | device->zone_info = zone_info; |
| 980 | } |
| 981 | |
| 982 | list_add(&device->dev_list, &fs_devices->devices); |
| 983 | device->fs_devices = fs_devices; |
| 984 | fs_devices->num_devices++; |
| 985 | } |
| 986 | return fs_devices; |
| 987 | error: |
| 988 | free_fs_devices(fs_devices); |
| 989 | return ERR_PTR(ret); |
| 990 | } |
| 991 | |
| 992 | static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices, |
| 993 | struct btrfs_device **latest_dev) |
| 994 | { |
| 995 | struct btrfs_device *device, *next; |
| 996 | |
| 997 | /* This is the initialized path, it is safe to release the devices. */ |
| 998 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
| 999 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) { |
| 1000 | if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT, |
| 1001 | &device->dev_state) && |
| 1002 | !test_bit(BTRFS_DEV_STATE_MISSING, |
| 1003 | &device->dev_state) && |
| 1004 | (!*latest_dev || |
| 1005 | device->generation > (*latest_dev)->generation)) { |
| 1006 | *latest_dev = device; |
| 1007 | } |
| 1008 | continue; |
| 1009 | } |
| 1010 | |
| 1011 | /* |
| 1012 | * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID, |
| 1013 | * in btrfs_init_dev_replace() so just continue. |
| 1014 | */ |
| 1015 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
| 1016 | continue; |
| 1017 | |
| 1018 | if (device->bdev_handle) { |
| 1019 | bdev_release(device->bdev_handle); |
| 1020 | device->bdev = NULL; |
| 1021 | device->bdev_handle = NULL; |
| 1022 | fs_devices->open_devices--; |
| 1023 | } |
| 1024 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 1025 | list_del_init(&device->dev_alloc_list); |
| 1026 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 1027 | fs_devices->rw_devices--; |
| 1028 | } |
| 1029 | list_del_init(&device->dev_list); |
| 1030 | fs_devices->num_devices--; |
| 1031 | btrfs_free_device(device); |
| 1032 | } |
| 1033 | |
| 1034 | } |
| 1035 | |
| 1036 | /* |
| 1037 | * After we have read the system tree and know devids belonging to this |
| 1038 | * filesystem, remove the device which does not belong there. |
| 1039 | */ |
| 1040 | void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices) |
| 1041 | { |
| 1042 | struct btrfs_device *latest_dev = NULL; |
| 1043 | struct btrfs_fs_devices *seed_dev; |
| 1044 | |
| 1045 | mutex_lock(&uuid_mutex); |
| 1046 | __btrfs_free_extra_devids(fs_devices, &latest_dev); |
| 1047 | |
| 1048 | list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list) |
| 1049 | __btrfs_free_extra_devids(seed_dev, &latest_dev); |
| 1050 | |
| 1051 | fs_devices->latest_dev = latest_dev; |
| 1052 | |
| 1053 | mutex_unlock(&uuid_mutex); |
| 1054 | } |
| 1055 | |
| 1056 | static void btrfs_close_bdev(struct btrfs_device *device) |
| 1057 | { |
| 1058 | if (!device->bdev) |
| 1059 | return; |
| 1060 | |
| 1061 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 1062 | sync_blockdev(device->bdev); |
| 1063 | invalidate_bdev(device->bdev); |
| 1064 | } |
| 1065 | |
| 1066 | bdev_release(device->bdev_handle); |
| 1067 | } |
| 1068 | |
| 1069 | static void btrfs_close_one_device(struct btrfs_device *device) |
| 1070 | { |
| 1071 | struct btrfs_fs_devices *fs_devices = device->fs_devices; |
| 1072 | |
| 1073 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
| 1074 | device->devid != BTRFS_DEV_REPLACE_DEVID) { |
| 1075 | list_del_init(&device->dev_alloc_list); |
| 1076 | fs_devices->rw_devices--; |
| 1077 | } |
| 1078 | |
| 1079 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) |
| 1080 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
| 1081 | |
| 1082 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
| 1083 | clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
| 1084 | fs_devices->missing_devices--; |
| 1085 | } |
| 1086 | |
| 1087 | btrfs_close_bdev(device); |
| 1088 | if (device->bdev) { |
| 1089 | fs_devices->open_devices--; |
| 1090 | device->bdev = NULL; |
| 1091 | } |
| 1092 | clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 1093 | btrfs_destroy_dev_zone_info(device); |
| 1094 | |
| 1095 | device->fs_info = NULL; |
| 1096 | atomic_set(&device->dev_stats_ccnt, 0); |
| 1097 | extent_io_tree_release(&device->alloc_state); |
| 1098 | |
| 1099 | /* |
| 1100 | * Reset the flush error record. We might have a transient flush error |
| 1101 | * in this mount, and if so we aborted the current transaction and set |
| 1102 | * the fs to an error state, guaranteeing no super blocks can be further |
| 1103 | * committed. However that error might be transient and if we unmount the |
| 1104 | * filesystem and mount it again, we should allow the mount to succeed |
| 1105 | * (btrfs_check_rw_degradable() should not fail) - if after mounting the |
| 1106 | * filesystem again we still get flush errors, then we will again abort |
| 1107 | * any transaction and set the error state, guaranteeing no commits of |
| 1108 | * unsafe super blocks. |
| 1109 | */ |
| 1110 | device->last_flush_error = 0; |
| 1111 | |
| 1112 | /* Verify the device is back in a pristine state */ |
| 1113 | WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state)); |
| 1114 | WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)); |
| 1115 | WARN_ON(!list_empty(&device->dev_alloc_list)); |
| 1116 | WARN_ON(!list_empty(&device->post_commit_list)); |
| 1117 | } |
| 1118 | |
| 1119 | static void close_fs_devices(struct btrfs_fs_devices *fs_devices) |
| 1120 | { |
| 1121 | struct btrfs_device *device, *tmp; |
| 1122 | |
| 1123 | lockdep_assert_held(&uuid_mutex); |
| 1124 | |
| 1125 | if (--fs_devices->opened > 0) |
| 1126 | return; |
| 1127 | |
| 1128 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) |
| 1129 | btrfs_close_one_device(device); |
| 1130 | |
| 1131 | WARN_ON(fs_devices->open_devices); |
| 1132 | WARN_ON(fs_devices->rw_devices); |
| 1133 | fs_devices->opened = 0; |
| 1134 | fs_devices->seeding = false; |
| 1135 | fs_devices->fs_info = NULL; |
| 1136 | } |
| 1137 | |
| 1138 | void btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
| 1139 | { |
| 1140 | LIST_HEAD(list); |
| 1141 | struct btrfs_fs_devices *tmp; |
| 1142 | |
| 1143 | mutex_lock(&uuid_mutex); |
| 1144 | close_fs_devices(fs_devices); |
| 1145 | if (!fs_devices->opened) { |
| 1146 | list_splice_init(&fs_devices->seed_list, &list); |
| 1147 | |
| 1148 | /* |
| 1149 | * If the struct btrfs_fs_devices is not assembled with any |
| 1150 | * other device, it can be re-initialized during the next mount |
| 1151 | * without the needing device-scan step. Therefore, it can be |
| 1152 | * fully freed. |
| 1153 | */ |
| 1154 | if (fs_devices->num_devices == 1) { |
| 1155 | list_del(&fs_devices->fs_list); |
| 1156 | free_fs_devices(fs_devices); |
| 1157 | } |
| 1158 | } |
| 1159 | |
| 1160 | |
| 1161 | list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) { |
| 1162 | close_fs_devices(fs_devices); |
| 1163 | list_del(&fs_devices->seed_list); |
| 1164 | free_fs_devices(fs_devices); |
| 1165 | } |
| 1166 | mutex_unlock(&uuid_mutex); |
| 1167 | } |
| 1168 | |
| 1169 | static int open_fs_devices(struct btrfs_fs_devices *fs_devices, |
| 1170 | blk_mode_t flags, void *holder) |
| 1171 | { |
| 1172 | struct btrfs_device *device; |
| 1173 | struct btrfs_device *latest_dev = NULL; |
| 1174 | struct btrfs_device *tmp_device; |
| 1175 | |
| 1176 | list_for_each_entry_safe(device, tmp_device, &fs_devices->devices, |
| 1177 | dev_list) { |
| 1178 | int ret; |
| 1179 | |
| 1180 | ret = btrfs_open_one_device(fs_devices, device, flags, holder); |
| 1181 | if (ret == 0 && |
| 1182 | (!latest_dev || device->generation > latest_dev->generation)) { |
| 1183 | latest_dev = device; |
| 1184 | } else if (ret == -ENODATA) { |
| 1185 | fs_devices->num_devices--; |
| 1186 | list_del(&device->dev_list); |
| 1187 | btrfs_free_device(device); |
| 1188 | } |
| 1189 | } |
| 1190 | if (fs_devices->open_devices == 0) |
| 1191 | return -EINVAL; |
| 1192 | |
| 1193 | fs_devices->opened = 1; |
| 1194 | fs_devices->latest_dev = latest_dev; |
| 1195 | fs_devices->total_rw_bytes = 0; |
| 1196 | fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR; |
| 1197 | fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
| 1198 | |
| 1199 | return 0; |
| 1200 | } |
| 1201 | |
| 1202 | static int devid_cmp(void *priv, const struct list_head *a, |
| 1203 | const struct list_head *b) |
| 1204 | { |
| 1205 | const struct btrfs_device *dev1, *dev2; |
| 1206 | |
| 1207 | dev1 = list_entry(a, struct btrfs_device, dev_list); |
| 1208 | dev2 = list_entry(b, struct btrfs_device, dev_list); |
| 1209 | |
| 1210 | if (dev1->devid < dev2->devid) |
| 1211 | return -1; |
| 1212 | else if (dev1->devid > dev2->devid) |
| 1213 | return 1; |
| 1214 | return 0; |
| 1215 | } |
| 1216 | |
| 1217 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
| 1218 | blk_mode_t flags, void *holder) |
| 1219 | { |
| 1220 | int ret; |
| 1221 | |
| 1222 | lockdep_assert_held(&uuid_mutex); |
| 1223 | /* |
| 1224 | * The device_list_mutex cannot be taken here in case opening the |
| 1225 | * underlying device takes further locks like open_mutex. |
| 1226 | * |
| 1227 | * We also don't need the lock here as this is called during mount and |
| 1228 | * exclusion is provided by uuid_mutex |
| 1229 | */ |
| 1230 | |
| 1231 | if (fs_devices->opened) { |
| 1232 | fs_devices->opened++; |
| 1233 | ret = 0; |
| 1234 | } else { |
| 1235 | list_sort(NULL, &fs_devices->devices, devid_cmp); |
| 1236 | ret = open_fs_devices(fs_devices, flags, holder); |
| 1237 | } |
| 1238 | |
| 1239 | return ret; |
| 1240 | } |
| 1241 | |
| 1242 | void btrfs_release_disk_super(struct btrfs_super_block *super) |
| 1243 | { |
| 1244 | struct page *page = virt_to_page(super); |
| 1245 | |
| 1246 | put_page(page); |
| 1247 | } |
| 1248 | |
| 1249 | static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev, |
| 1250 | u64 bytenr, u64 bytenr_orig) |
| 1251 | { |
| 1252 | struct btrfs_super_block *disk_super; |
| 1253 | struct page *page; |
| 1254 | void *p; |
| 1255 | pgoff_t index; |
| 1256 | |
| 1257 | /* make sure our super fits in the device */ |
| 1258 | if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev)) |
| 1259 | return ERR_PTR(-EINVAL); |
| 1260 | |
| 1261 | /* make sure our super fits in the page */ |
| 1262 | if (sizeof(*disk_super) > PAGE_SIZE) |
| 1263 | return ERR_PTR(-EINVAL); |
| 1264 | |
| 1265 | /* make sure our super doesn't straddle pages on disk */ |
| 1266 | index = bytenr >> PAGE_SHIFT; |
| 1267 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index) |
| 1268 | return ERR_PTR(-EINVAL); |
| 1269 | |
| 1270 | /* pull in the page with our super */ |
| 1271 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL); |
| 1272 | |
| 1273 | if (IS_ERR(page)) |
| 1274 | return ERR_CAST(page); |
| 1275 | |
| 1276 | p = page_address(page); |
| 1277 | |
| 1278 | /* align our pointer to the offset of the super block */ |
| 1279 | disk_super = p + offset_in_page(bytenr); |
| 1280 | |
| 1281 | if (btrfs_super_bytenr(disk_super) != bytenr_orig || |
| 1282 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { |
| 1283 | btrfs_release_disk_super(p); |
| 1284 | return ERR_PTR(-EINVAL); |
| 1285 | } |
| 1286 | |
| 1287 | if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1]) |
| 1288 | disk_super->label[BTRFS_LABEL_SIZE - 1] = 0; |
| 1289 | |
| 1290 | return disk_super; |
| 1291 | } |
| 1292 | |
| 1293 | int btrfs_forget_devices(dev_t devt) |
| 1294 | { |
| 1295 | int ret; |
| 1296 | |
| 1297 | mutex_lock(&uuid_mutex); |
| 1298 | ret = btrfs_free_stale_devices(devt, NULL); |
| 1299 | mutex_unlock(&uuid_mutex); |
| 1300 | |
| 1301 | return ret; |
| 1302 | } |
| 1303 | |
| 1304 | /* |
| 1305 | * Look for a btrfs signature on a device. This may be called out of the mount path |
| 1306 | * and we are not allowed to call set_blocksize during the scan. The superblock |
| 1307 | * is read via pagecache. |
| 1308 | * |
| 1309 | * With @mount_arg_dev it's a scan during mount time that will always register |
| 1310 | * the device or return an error. Multi-device and seeding devices are registered |
| 1311 | * in both cases. |
| 1312 | */ |
| 1313 | struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags, |
| 1314 | bool mount_arg_dev) |
| 1315 | { |
| 1316 | struct btrfs_super_block *disk_super; |
| 1317 | bool new_device_added = false; |
| 1318 | struct btrfs_device *device = NULL; |
| 1319 | struct bdev_handle *bdev_handle; |
| 1320 | u64 bytenr, bytenr_orig; |
| 1321 | int ret; |
| 1322 | |
| 1323 | lockdep_assert_held(&uuid_mutex); |
| 1324 | |
| 1325 | /* |
| 1326 | * we would like to check all the supers, but that would make |
| 1327 | * a btrfs mount succeed after a mkfs from a different FS. |
| 1328 | * So, we need to add a special mount option to scan for |
| 1329 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead |
| 1330 | */ |
| 1331 | |
| 1332 | /* |
| 1333 | * Avoid an exclusive open here, as the systemd-udev may initiate the |
| 1334 | * device scan which may race with the user's mount or mkfs command, |
| 1335 | * resulting in failure. |
| 1336 | * Since the device scan is solely for reading purposes, there is no |
| 1337 | * need for an exclusive open. Additionally, the devices are read again |
| 1338 | * during the mount process. It is ok to get some inconsistent |
| 1339 | * values temporarily, as the device paths of the fsid are the only |
| 1340 | * required information for assembling the volume. |
| 1341 | */ |
| 1342 | bdev_handle = bdev_open_by_path(path, flags, NULL, NULL); |
| 1343 | if (IS_ERR(bdev_handle)) |
| 1344 | return ERR_CAST(bdev_handle); |
| 1345 | |
| 1346 | bytenr_orig = btrfs_sb_offset(0); |
| 1347 | ret = btrfs_sb_log_location_bdev(bdev_handle->bdev, 0, READ, &bytenr); |
| 1348 | if (ret) { |
| 1349 | device = ERR_PTR(ret); |
| 1350 | goto error_bdev_put; |
| 1351 | } |
| 1352 | |
| 1353 | disk_super = btrfs_read_disk_super(bdev_handle->bdev, bytenr, |
| 1354 | bytenr_orig); |
| 1355 | if (IS_ERR(disk_super)) { |
| 1356 | device = ERR_CAST(disk_super); |
| 1357 | goto error_bdev_put; |
| 1358 | } |
| 1359 | |
| 1360 | if (!mount_arg_dev && btrfs_super_num_devices(disk_super) == 1 && |
| 1361 | !(btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)) { |
| 1362 | dev_t devt; |
| 1363 | |
| 1364 | ret = lookup_bdev(path, &devt); |
| 1365 | if (ret) |
| 1366 | btrfs_warn(NULL, "lookup bdev failed for path %s: %d", |
| 1367 | path, ret); |
| 1368 | else |
| 1369 | btrfs_free_stale_devices(devt, NULL); |
| 1370 | |
| 1371 | pr_debug("BTRFS: skip registering single non-seed device %s\n", path); |
| 1372 | device = NULL; |
| 1373 | goto free_disk_super; |
| 1374 | } |
| 1375 | |
| 1376 | device = device_list_add(path, disk_super, &new_device_added); |
| 1377 | if (!IS_ERR(device) && new_device_added) |
| 1378 | btrfs_free_stale_devices(device->devt, device); |
| 1379 | |
| 1380 | free_disk_super: |
| 1381 | btrfs_release_disk_super(disk_super); |
| 1382 | |
| 1383 | error_bdev_put: |
| 1384 | bdev_release(bdev_handle); |
| 1385 | |
| 1386 | return device; |
| 1387 | } |
| 1388 | |
| 1389 | /* |
| 1390 | * Try to find a chunk that intersects [start, start + len] range and when one |
| 1391 | * such is found, record the end of it in *start |
| 1392 | */ |
| 1393 | static bool contains_pending_extent(struct btrfs_device *device, u64 *start, |
| 1394 | u64 len) |
| 1395 | { |
| 1396 | u64 physical_start, physical_end; |
| 1397 | |
| 1398 | lockdep_assert_held(&device->fs_info->chunk_mutex); |
| 1399 | |
| 1400 | if (find_first_extent_bit(&device->alloc_state, *start, |
| 1401 | &physical_start, &physical_end, |
| 1402 | CHUNK_ALLOCATED, NULL)) { |
| 1403 | |
| 1404 | if (in_range(physical_start, *start, len) || |
| 1405 | in_range(*start, physical_start, |
| 1406 | physical_end - physical_start)) { |
| 1407 | *start = physical_end + 1; |
| 1408 | return true; |
| 1409 | } |
| 1410 | } |
| 1411 | return false; |
| 1412 | } |
| 1413 | |
| 1414 | static u64 dev_extent_search_start(struct btrfs_device *device) |
| 1415 | { |
| 1416 | switch (device->fs_devices->chunk_alloc_policy) { |
| 1417 | case BTRFS_CHUNK_ALLOC_REGULAR: |
| 1418 | return BTRFS_DEVICE_RANGE_RESERVED; |
| 1419 | case BTRFS_CHUNK_ALLOC_ZONED: |
| 1420 | /* |
| 1421 | * We don't care about the starting region like regular |
| 1422 | * allocator, because we anyway use/reserve the first two zones |
| 1423 | * for superblock logging. |
| 1424 | */ |
| 1425 | return 0; |
| 1426 | default: |
| 1427 | BUG(); |
| 1428 | } |
| 1429 | } |
| 1430 | |
| 1431 | static bool dev_extent_hole_check_zoned(struct btrfs_device *device, |
| 1432 | u64 *hole_start, u64 *hole_size, |
| 1433 | u64 num_bytes) |
| 1434 | { |
| 1435 | u64 zone_size = device->zone_info->zone_size; |
| 1436 | u64 pos; |
| 1437 | int ret; |
| 1438 | bool changed = false; |
| 1439 | |
| 1440 | ASSERT(IS_ALIGNED(*hole_start, zone_size)); |
| 1441 | |
| 1442 | while (*hole_size > 0) { |
| 1443 | pos = btrfs_find_allocatable_zones(device, *hole_start, |
| 1444 | *hole_start + *hole_size, |
| 1445 | num_bytes); |
| 1446 | if (pos != *hole_start) { |
| 1447 | *hole_size = *hole_start + *hole_size - pos; |
| 1448 | *hole_start = pos; |
| 1449 | changed = true; |
| 1450 | if (*hole_size < num_bytes) |
| 1451 | break; |
| 1452 | } |
| 1453 | |
| 1454 | ret = btrfs_ensure_empty_zones(device, pos, num_bytes); |
| 1455 | |
| 1456 | /* Range is ensured to be empty */ |
| 1457 | if (!ret) |
| 1458 | return changed; |
| 1459 | |
| 1460 | /* Given hole range was invalid (outside of device) */ |
| 1461 | if (ret == -ERANGE) { |
| 1462 | *hole_start += *hole_size; |
| 1463 | *hole_size = 0; |
| 1464 | return true; |
| 1465 | } |
| 1466 | |
| 1467 | *hole_start += zone_size; |
| 1468 | *hole_size -= zone_size; |
| 1469 | changed = true; |
| 1470 | } |
| 1471 | |
| 1472 | return changed; |
| 1473 | } |
| 1474 | |
| 1475 | /* |
| 1476 | * Check if specified hole is suitable for allocation. |
| 1477 | * |
| 1478 | * @device: the device which we have the hole |
| 1479 | * @hole_start: starting position of the hole |
| 1480 | * @hole_size: the size of the hole |
| 1481 | * @num_bytes: the size of the free space that we need |
| 1482 | * |
| 1483 | * This function may modify @hole_start and @hole_size to reflect the suitable |
| 1484 | * position for allocation. Returns 1 if hole position is updated, 0 otherwise. |
| 1485 | */ |
| 1486 | static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start, |
| 1487 | u64 *hole_size, u64 num_bytes) |
| 1488 | { |
| 1489 | bool changed = false; |
| 1490 | u64 hole_end = *hole_start + *hole_size; |
| 1491 | |
| 1492 | for (;;) { |
| 1493 | /* |
| 1494 | * Check before we set max_hole_start, otherwise we could end up |
| 1495 | * sending back this offset anyway. |
| 1496 | */ |
| 1497 | if (contains_pending_extent(device, hole_start, *hole_size)) { |
| 1498 | if (hole_end >= *hole_start) |
| 1499 | *hole_size = hole_end - *hole_start; |
| 1500 | else |
| 1501 | *hole_size = 0; |
| 1502 | changed = true; |
| 1503 | } |
| 1504 | |
| 1505 | switch (device->fs_devices->chunk_alloc_policy) { |
| 1506 | case BTRFS_CHUNK_ALLOC_REGULAR: |
| 1507 | /* No extra check */ |
| 1508 | break; |
| 1509 | case BTRFS_CHUNK_ALLOC_ZONED: |
| 1510 | if (dev_extent_hole_check_zoned(device, hole_start, |
| 1511 | hole_size, num_bytes)) { |
| 1512 | changed = true; |
| 1513 | /* |
| 1514 | * The changed hole can contain pending extent. |
| 1515 | * Loop again to check that. |
| 1516 | */ |
| 1517 | continue; |
| 1518 | } |
| 1519 | break; |
| 1520 | default: |
| 1521 | BUG(); |
| 1522 | } |
| 1523 | |
| 1524 | break; |
| 1525 | } |
| 1526 | |
| 1527 | return changed; |
| 1528 | } |
| 1529 | |
| 1530 | /* |
| 1531 | * Find free space in the specified device. |
| 1532 | * |
| 1533 | * @device: the device which we search the free space in |
| 1534 | * @num_bytes: the size of the free space that we need |
| 1535 | * @search_start: the position from which to begin the search |
| 1536 | * @start: store the start of the free space. |
| 1537 | * @len: the size of the free space. that we find, or the size |
| 1538 | * of the max free space if we don't find suitable free space |
| 1539 | * |
| 1540 | * This does a pretty simple search, the expectation is that it is called very |
| 1541 | * infrequently and that a given device has a small number of extents. |
| 1542 | * |
| 1543 | * @start is used to store the start of the free space if we find. But if we |
| 1544 | * don't find suitable free space, it will be used to store the start position |
| 1545 | * of the max free space. |
| 1546 | * |
| 1547 | * @len is used to store the size of the free space that we find. |
| 1548 | * But if we don't find suitable free space, it is used to store the size of |
| 1549 | * the max free space. |
| 1550 | * |
| 1551 | * NOTE: This function will search *commit* root of device tree, and does extra |
| 1552 | * check to ensure dev extents are not double allocated. |
| 1553 | * This makes the function safe to allocate dev extents but may not report |
| 1554 | * correct usable device space, as device extent freed in current transaction |
| 1555 | * is not reported as available. |
| 1556 | */ |
| 1557 | static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
| 1558 | u64 *start, u64 *len) |
| 1559 | { |
| 1560 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 1561 | struct btrfs_root *root = fs_info->dev_root; |
| 1562 | struct btrfs_key key; |
| 1563 | struct btrfs_dev_extent *dev_extent; |
| 1564 | struct btrfs_path *path; |
| 1565 | u64 search_start; |
| 1566 | u64 hole_size; |
| 1567 | u64 max_hole_start; |
| 1568 | u64 max_hole_size = 0; |
| 1569 | u64 extent_end; |
| 1570 | u64 search_end = device->total_bytes; |
| 1571 | int ret; |
| 1572 | int slot; |
| 1573 | struct extent_buffer *l; |
| 1574 | |
| 1575 | search_start = dev_extent_search_start(device); |
| 1576 | max_hole_start = search_start; |
| 1577 | |
| 1578 | WARN_ON(device->zone_info && |
| 1579 | !IS_ALIGNED(num_bytes, device->zone_info->zone_size)); |
| 1580 | |
| 1581 | path = btrfs_alloc_path(); |
| 1582 | if (!path) { |
| 1583 | ret = -ENOMEM; |
| 1584 | goto out; |
| 1585 | } |
| 1586 | again: |
| 1587 | if (search_start >= search_end || |
| 1588 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
| 1589 | ret = -ENOSPC; |
| 1590 | goto out; |
| 1591 | } |
| 1592 | |
| 1593 | path->reada = READA_FORWARD; |
| 1594 | path->search_commit_root = 1; |
| 1595 | path->skip_locking = 1; |
| 1596 | |
| 1597 | key.objectid = device->devid; |
| 1598 | key.offset = search_start; |
| 1599 | key.type = BTRFS_DEV_EXTENT_KEY; |
| 1600 | |
| 1601 | ret = btrfs_search_backwards(root, &key, path); |
| 1602 | if (ret < 0) |
| 1603 | goto out; |
| 1604 | |
| 1605 | while (search_start < search_end) { |
| 1606 | l = path->nodes[0]; |
| 1607 | slot = path->slots[0]; |
| 1608 | if (slot >= btrfs_header_nritems(l)) { |
| 1609 | ret = btrfs_next_leaf(root, path); |
| 1610 | if (ret == 0) |
| 1611 | continue; |
| 1612 | if (ret < 0) |
| 1613 | goto out; |
| 1614 | |
| 1615 | break; |
| 1616 | } |
| 1617 | btrfs_item_key_to_cpu(l, &key, slot); |
| 1618 | |
| 1619 | if (key.objectid < device->devid) |
| 1620 | goto next; |
| 1621 | |
| 1622 | if (key.objectid > device->devid) |
| 1623 | break; |
| 1624 | |
| 1625 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
| 1626 | goto next; |
| 1627 | |
| 1628 | if (key.offset > search_end) |
| 1629 | break; |
| 1630 | |
| 1631 | if (key.offset > search_start) { |
| 1632 | hole_size = key.offset - search_start; |
| 1633 | dev_extent_hole_check(device, &search_start, &hole_size, |
| 1634 | num_bytes); |
| 1635 | |
| 1636 | if (hole_size > max_hole_size) { |
| 1637 | max_hole_start = search_start; |
| 1638 | max_hole_size = hole_size; |
| 1639 | } |
| 1640 | |
| 1641 | /* |
| 1642 | * If this free space is greater than which we need, |
| 1643 | * it must be the max free space that we have found |
| 1644 | * until now, so max_hole_start must point to the start |
| 1645 | * of this free space and the length of this free space |
| 1646 | * is stored in max_hole_size. Thus, we return |
| 1647 | * max_hole_start and max_hole_size and go back to the |
| 1648 | * caller. |
| 1649 | */ |
| 1650 | if (hole_size >= num_bytes) { |
| 1651 | ret = 0; |
| 1652 | goto out; |
| 1653 | } |
| 1654 | } |
| 1655 | |
| 1656 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
| 1657 | extent_end = key.offset + btrfs_dev_extent_length(l, |
| 1658 | dev_extent); |
| 1659 | if (extent_end > search_start) |
| 1660 | search_start = extent_end; |
| 1661 | next: |
| 1662 | path->slots[0]++; |
| 1663 | cond_resched(); |
| 1664 | } |
| 1665 | |
| 1666 | /* |
| 1667 | * At this point, search_start should be the end of |
| 1668 | * allocated dev extents, and when shrinking the device, |
| 1669 | * search_end may be smaller than search_start. |
| 1670 | */ |
| 1671 | if (search_end > search_start) { |
| 1672 | hole_size = search_end - search_start; |
| 1673 | if (dev_extent_hole_check(device, &search_start, &hole_size, |
| 1674 | num_bytes)) { |
| 1675 | btrfs_release_path(path); |
| 1676 | goto again; |
| 1677 | } |
| 1678 | |
| 1679 | if (hole_size > max_hole_size) { |
| 1680 | max_hole_start = search_start; |
| 1681 | max_hole_size = hole_size; |
| 1682 | } |
| 1683 | } |
| 1684 | |
| 1685 | /* See above. */ |
| 1686 | if (max_hole_size < num_bytes) |
| 1687 | ret = -ENOSPC; |
| 1688 | else |
| 1689 | ret = 0; |
| 1690 | |
| 1691 | ASSERT(max_hole_start + max_hole_size <= search_end); |
| 1692 | out: |
| 1693 | btrfs_free_path(path); |
| 1694 | *start = max_hole_start; |
| 1695 | if (len) |
| 1696 | *len = max_hole_size; |
| 1697 | return ret; |
| 1698 | } |
| 1699 | |
| 1700 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
| 1701 | struct btrfs_device *device, |
| 1702 | u64 start, u64 *dev_extent_len) |
| 1703 | { |
| 1704 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 1705 | struct btrfs_root *root = fs_info->dev_root; |
| 1706 | int ret; |
| 1707 | struct btrfs_path *path; |
| 1708 | struct btrfs_key key; |
| 1709 | struct btrfs_key found_key; |
| 1710 | struct extent_buffer *leaf = NULL; |
| 1711 | struct btrfs_dev_extent *extent = NULL; |
| 1712 | |
| 1713 | path = btrfs_alloc_path(); |
| 1714 | if (!path) |
| 1715 | return -ENOMEM; |
| 1716 | |
| 1717 | key.objectid = device->devid; |
| 1718 | key.offset = start; |
| 1719 | key.type = BTRFS_DEV_EXTENT_KEY; |
| 1720 | again: |
| 1721 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1722 | if (ret > 0) { |
| 1723 | ret = btrfs_previous_item(root, path, key.objectid, |
| 1724 | BTRFS_DEV_EXTENT_KEY); |
| 1725 | if (ret) |
| 1726 | goto out; |
| 1727 | leaf = path->nodes[0]; |
| 1728 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 1729 | extent = btrfs_item_ptr(leaf, path->slots[0], |
| 1730 | struct btrfs_dev_extent); |
| 1731 | BUG_ON(found_key.offset > start || found_key.offset + |
| 1732 | btrfs_dev_extent_length(leaf, extent) < start); |
| 1733 | key = found_key; |
| 1734 | btrfs_release_path(path); |
| 1735 | goto again; |
| 1736 | } else if (ret == 0) { |
| 1737 | leaf = path->nodes[0]; |
| 1738 | extent = btrfs_item_ptr(leaf, path->slots[0], |
| 1739 | struct btrfs_dev_extent); |
| 1740 | } else { |
| 1741 | goto out; |
| 1742 | } |
| 1743 | |
| 1744 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
| 1745 | |
| 1746 | ret = btrfs_del_item(trans, root, path); |
| 1747 | if (ret == 0) |
| 1748 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
| 1749 | out: |
| 1750 | btrfs_free_path(path); |
| 1751 | return ret; |
| 1752 | } |
| 1753 | |
| 1754 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
| 1755 | { |
| 1756 | struct rb_node *n; |
| 1757 | u64 ret = 0; |
| 1758 | |
| 1759 | read_lock(&fs_info->mapping_tree_lock); |
| 1760 | n = rb_last(&fs_info->mapping_tree.rb_root); |
| 1761 | if (n) { |
| 1762 | struct btrfs_chunk_map *map; |
| 1763 | |
| 1764 | map = rb_entry(n, struct btrfs_chunk_map, rb_node); |
| 1765 | ret = map->start + map->chunk_len; |
| 1766 | } |
| 1767 | read_unlock(&fs_info->mapping_tree_lock); |
| 1768 | |
| 1769 | return ret; |
| 1770 | } |
| 1771 | |
| 1772 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
| 1773 | u64 *devid_ret) |
| 1774 | { |
| 1775 | int ret; |
| 1776 | struct btrfs_key key; |
| 1777 | struct btrfs_key found_key; |
| 1778 | struct btrfs_path *path; |
| 1779 | |
| 1780 | path = btrfs_alloc_path(); |
| 1781 | if (!path) |
| 1782 | return -ENOMEM; |
| 1783 | |
| 1784 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 1785 | key.type = BTRFS_DEV_ITEM_KEY; |
| 1786 | key.offset = (u64)-1; |
| 1787 | |
| 1788 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
| 1789 | if (ret < 0) |
| 1790 | goto error; |
| 1791 | |
| 1792 | if (ret == 0) { |
| 1793 | /* Corruption */ |
| 1794 | btrfs_err(fs_info, "corrupted chunk tree devid -1 matched"); |
| 1795 | ret = -EUCLEAN; |
| 1796 | goto error; |
| 1797 | } |
| 1798 | |
| 1799 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
| 1800 | BTRFS_DEV_ITEMS_OBJECTID, |
| 1801 | BTRFS_DEV_ITEM_KEY); |
| 1802 | if (ret) { |
| 1803 | *devid_ret = 1; |
| 1804 | } else { |
| 1805 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
| 1806 | path->slots[0]); |
| 1807 | *devid_ret = found_key.offset + 1; |
| 1808 | } |
| 1809 | ret = 0; |
| 1810 | error: |
| 1811 | btrfs_free_path(path); |
| 1812 | return ret; |
| 1813 | } |
| 1814 | |
| 1815 | /* |
| 1816 | * the device information is stored in the chunk root |
| 1817 | * the btrfs_device struct should be fully filled in |
| 1818 | */ |
| 1819 | static int btrfs_add_dev_item(struct btrfs_trans_handle *trans, |
| 1820 | struct btrfs_device *device) |
| 1821 | { |
| 1822 | int ret; |
| 1823 | struct btrfs_path *path; |
| 1824 | struct btrfs_dev_item *dev_item; |
| 1825 | struct extent_buffer *leaf; |
| 1826 | struct btrfs_key key; |
| 1827 | unsigned long ptr; |
| 1828 | |
| 1829 | path = btrfs_alloc_path(); |
| 1830 | if (!path) |
| 1831 | return -ENOMEM; |
| 1832 | |
| 1833 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 1834 | key.type = BTRFS_DEV_ITEM_KEY; |
| 1835 | key.offset = device->devid; |
| 1836 | |
| 1837 | btrfs_reserve_chunk_metadata(trans, true); |
| 1838 | ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path, |
| 1839 | &key, sizeof(*dev_item)); |
| 1840 | btrfs_trans_release_chunk_metadata(trans); |
| 1841 | if (ret) |
| 1842 | goto out; |
| 1843 | |
| 1844 | leaf = path->nodes[0]; |
| 1845 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); |
| 1846 | |
| 1847 | btrfs_set_device_id(leaf, dev_item, device->devid); |
| 1848 | btrfs_set_device_generation(leaf, dev_item, 0); |
| 1849 | btrfs_set_device_type(leaf, dev_item, device->type); |
| 1850 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); |
| 1851 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); |
| 1852 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); |
| 1853 | btrfs_set_device_total_bytes(leaf, dev_item, |
| 1854 | btrfs_device_get_disk_total_bytes(device)); |
| 1855 | btrfs_set_device_bytes_used(leaf, dev_item, |
| 1856 | btrfs_device_get_bytes_used(device)); |
| 1857 | btrfs_set_device_group(leaf, dev_item, 0); |
| 1858 | btrfs_set_device_seek_speed(leaf, dev_item, 0); |
| 1859 | btrfs_set_device_bandwidth(leaf, dev_item, 0); |
| 1860 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
| 1861 | |
| 1862 | ptr = btrfs_device_uuid(dev_item); |
| 1863 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
| 1864 | ptr = btrfs_device_fsid(dev_item); |
| 1865 | write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid, |
| 1866 | ptr, BTRFS_FSID_SIZE); |
| 1867 | btrfs_mark_buffer_dirty(trans, leaf); |
| 1868 | |
| 1869 | ret = 0; |
| 1870 | out: |
| 1871 | btrfs_free_path(path); |
| 1872 | return ret; |
| 1873 | } |
| 1874 | |
| 1875 | /* |
| 1876 | * Function to update ctime/mtime for a given device path. |
| 1877 | * Mainly used for ctime/mtime based probe like libblkid. |
| 1878 | * |
| 1879 | * We don't care about errors here, this is just to be kind to userspace. |
| 1880 | */ |
| 1881 | static void update_dev_time(const char *device_path) |
| 1882 | { |
| 1883 | struct path path; |
| 1884 | int ret; |
| 1885 | |
| 1886 | ret = kern_path(device_path, LOOKUP_FOLLOW, &path); |
| 1887 | if (ret) |
| 1888 | return; |
| 1889 | |
| 1890 | inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION); |
| 1891 | path_put(&path); |
| 1892 | } |
| 1893 | |
| 1894 | static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans, |
| 1895 | struct btrfs_device *device) |
| 1896 | { |
| 1897 | struct btrfs_root *root = device->fs_info->chunk_root; |
| 1898 | int ret; |
| 1899 | struct btrfs_path *path; |
| 1900 | struct btrfs_key key; |
| 1901 | |
| 1902 | path = btrfs_alloc_path(); |
| 1903 | if (!path) |
| 1904 | return -ENOMEM; |
| 1905 | |
| 1906 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 1907 | key.type = BTRFS_DEV_ITEM_KEY; |
| 1908 | key.offset = device->devid; |
| 1909 | |
| 1910 | btrfs_reserve_chunk_metadata(trans, false); |
| 1911 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1912 | btrfs_trans_release_chunk_metadata(trans); |
| 1913 | if (ret) { |
| 1914 | if (ret > 0) |
| 1915 | ret = -ENOENT; |
| 1916 | goto out; |
| 1917 | } |
| 1918 | |
| 1919 | ret = btrfs_del_item(trans, root, path); |
| 1920 | out: |
| 1921 | btrfs_free_path(path); |
| 1922 | return ret; |
| 1923 | } |
| 1924 | |
| 1925 | /* |
| 1926 | * Verify that @num_devices satisfies the RAID profile constraints in the whole |
| 1927 | * filesystem. It's up to the caller to adjust that number regarding eg. device |
| 1928 | * replace. |
| 1929 | */ |
| 1930 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, |
| 1931 | u64 num_devices) |
| 1932 | { |
| 1933 | u64 all_avail; |
| 1934 | unsigned seq; |
| 1935 | int i; |
| 1936 | |
| 1937 | do { |
| 1938 | seq = read_seqbegin(&fs_info->profiles_lock); |
| 1939 | |
| 1940 | all_avail = fs_info->avail_data_alloc_bits | |
| 1941 | fs_info->avail_system_alloc_bits | |
| 1942 | fs_info->avail_metadata_alloc_bits; |
| 1943 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
| 1944 | |
| 1945 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
| 1946 | if (!(all_avail & btrfs_raid_array[i].bg_flag)) |
| 1947 | continue; |
| 1948 | |
| 1949 | if (num_devices < btrfs_raid_array[i].devs_min) |
| 1950 | return btrfs_raid_array[i].mindev_error; |
| 1951 | } |
| 1952 | |
| 1953 | return 0; |
| 1954 | } |
| 1955 | |
| 1956 | static struct btrfs_device * btrfs_find_next_active_device( |
| 1957 | struct btrfs_fs_devices *fs_devs, struct btrfs_device *device) |
| 1958 | { |
| 1959 | struct btrfs_device *next_device; |
| 1960 | |
| 1961 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { |
| 1962 | if (next_device != device && |
| 1963 | !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state) |
| 1964 | && next_device->bdev) |
| 1965 | return next_device; |
| 1966 | } |
| 1967 | |
| 1968 | return NULL; |
| 1969 | } |
| 1970 | |
| 1971 | /* |
| 1972 | * Helper function to check if the given device is part of s_bdev / latest_dev |
| 1973 | * and replace it with the provided or the next active device, in the context |
| 1974 | * where this function called, there should be always be another device (or |
| 1975 | * this_dev) which is active. |
| 1976 | */ |
| 1977 | void __cold btrfs_assign_next_active_device(struct btrfs_device *device, |
| 1978 | struct btrfs_device *next_device) |
| 1979 | { |
| 1980 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 1981 | |
| 1982 | if (!next_device) |
| 1983 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, |
| 1984 | device); |
| 1985 | ASSERT(next_device); |
| 1986 | |
| 1987 | if (fs_info->sb->s_bdev && |
| 1988 | (fs_info->sb->s_bdev == device->bdev)) |
| 1989 | fs_info->sb->s_bdev = next_device->bdev; |
| 1990 | |
| 1991 | if (fs_info->fs_devices->latest_dev->bdev == device->bdev) |
| 1992 | fs_info->fs_devices->latest_dev = next_device; |
| 1993 | } |
| 1994 | |
| 1995 | /* |
| 1996 | * Return btrfs_fs_devices::num_devices excluding the device that's being |
| 1997 | * currently replaced. |
| 1998 | */ |
| 1999 | static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info) |
| 2000 | { |
| 2001 | u64 num_devices = fs_info->fs_devices->num_devices; |
| 2002 | |
| 2003 | down_read(&fs_info->dev_replace.rwsem); |
| 2004 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
| 2005 | ASSERT(num_devices > 1); |
| 2006 | num_devices--; |
| 2007 | } |
| 2008 | up_read(&fs_info->dev_replace.rwsem); |
| 2009 | |
| 2010 | return num_devices; |
| 2011 | } |
| 2012 | |
| 2013 | static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info, |
| 2014 | struct block_device *bdev, int copy_num) |
| 2015 | { |
| 2016 | struct btrfs_super_block *disk_super; |
| 2017 | const size_t len = sizeof(disk_super->magic); |
| 2018 | const u64 bytenr = btrfs_sb_offset(copy_num); |
| 2019 | int ret; |
| 2020 | |
| 2021 | disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr); |
| 2022 | if (IS_ERR(disk_super)) |
| 2023 | return; |
| 2024 | |
| 2025 | memset(&disk_super->magic, 0, len); |
| 2026 | folio_mark_dirty(virt_to_folio(disk_super)); |
| 2027 | btrfs_release_disk_super(disk_super); |
| 2028 | |
| 2029 | ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1); |
| 2030 | if (ret) |
| 2031 | btrfs_warn(fs_info, "error clearing superblock number %d (%d)", |
| 2032 | copy_num, ret); |
| 2033 | } |
| 2034 | |
| 2035 | void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, |
| 2036 | struct block_device *bdev, |
| 2037 | const char *device_path) |
| 2038 | { |
| 2039 | int copy_num; |
| 2040 | |
| 2041 | if (!bdev) |
| 2042 | return; |
| 2043 | |
| 2044 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) { |
| 2045 | if (bdev_is_zoned(bdev)) |
| 2046 | btrfs_reset_sb_log_zones(bdev, copy_num); |
| 2047 | else |
| 2048 | btrfs_scratch_superblock(fs_info, bdev, copy_num); |
| 2049 | } |
| 2050 | |
| 2051 | /* Notify udev that device has changed */ |
| 2052 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
| 2053 | |
| 2054 | /* Update ctime/mtime for device path for libblkid */ |
| 2055 | update_dev_time(device_path); |
| 2056 | } |
| 2057 | |
| 2058 | int btrfs_rm_device(struct btrfs_fs_info *fs_info, |
| 2059 | struct btrfs_dev_lookup_args *args, |
| 2060 | struct bdev_handle **bdev_handle) |
| 2061 | { |
| 2062 | struct btrfs_trans_handle *trans; |
| 2063 | struct btrfs_device *device; |
| 2064 | struct btrfs_fs_devices *cur_devices; |
| 2065 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 2066 | u64 num_devices; |
| 2067 | int ret = 0; |
| 2068 | |
| 2069 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
| 2070 | btrfs_err(fs_info, "device remove not supported on extent tree v2 yet"); |
| 2071 | return -EINVAL; |
| 2072 | } |
| 2073 | |
| 2074 | /* |
| 2075 | * The device list in fs_devices is accessed without locks (neither |
| 2076 | * uuid_mutex nor device_list_mutex) as it won't change on a mounted |
| 2077 | * filesystem and another device rm cannot run. |
| 2078 | */ |
| 2079 | num_devices = btrfs_num_devices(fs_info); |
| 2080 | |
| 2081 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
| 2082 | if (ret) |
| 2083 | return ret; |
| 2084 | |
| 2085 | device = btrfs_find_device(fs_info->fs_devices, args); |
| 2086 | if (!device) { |
| 2087 | if (args->missing) |
| 2088 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
| 2089 | else |
| 2090 | ret = -ENOENT; |
| 2091 | return ret; |
| 2092 | } |
| 2093 | |
| 2094 | if (btrfs_pinned_by_swapfile(fs_info, device)) { |
| 2095 | btrfs_warn_in_rcu(fs_info, |
| 2096 | "cannot remove device %s (devid %llu) due to active swapfile", |
| 2097 | btrfs_dev_name(device), device->devid); |
| 2098 | return -ETXTBSY; |
| 2099 | } |
| 2100 | |
| 2101 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
| 2102 | return BTRFS_ERROR_DEV_TGT_REPLACE; |
| 2103 | |
| 2104 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
| 2105 | fs_info->fs_devices->rw_devices == 1) |
| 2106 | return BTRFS_ERROR_DEV_ONLY_WRITABLE; |
| 2107 | |
| 2108 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 2109 | mutex_lock(&fs_info->chunk_mutex); |
| 2110 | list_del_init(&device->dev_alloc_list); |
| 2111 | device->fs_devices->rw_devices--; |
| 2112 | mutex_unlock(&fs_info->chunk_mutex); |
| 2113 | } |
| 2114 | |
| 2115 | ret = btrfs_shrink_device(device, 0); |
| 2116 | if (ret) |
| 2117 | goto error_undo; |
| 2118 | |
| 2119 | trans = btrfs_start_transaction(fs_info->chunk_root, 0); |
| 2120 | if (IS_ERR(trans)) { |
| 2121 | ret = PTR_ERR(trans); |
| 2122 | goto error_undo; |
| 2123 | } |
| 2124 | |
| 2125 | ret = btrfs_rm_dev_item(trans, device); |
| 2126 | if (ret) { |
| 2127 | /* Any error in dev item removal is critical */ |
| 2128 | btrfs_crit(fs_info, |
| 2129 | "failed to remove device item for devid %llu: %d", |
| 2130 | device->devid, ret); |
| 2131 | btrfs_abort_transaction(trans, ret); |
| 2132 | btrfs_end_transaction(trans); |
| 2133 | return ret; |
| 2134 | } |
| 2135 | |
| 2136 | clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
| 2137 | btrfs_scrub_cancel_dev(device); |
| 2138 | |
| 2139 | /* |
| 2140 | * the device list mutex makes sure that we don't change |
| 2141 | * the device list while someone else is writing out all |
| 2142 | * the device supers. Whoever is writing all supers, should |
| 2143 | * lock the device list mutex before getting the number of |
| 2144 | * devices in the super block (super_copy). Conversely, |
| 2145 | * whoever updates the number of devices in the super block |
| 2146 | * (super_copy) should hold the device list mutex. |
| 2147 | */ |
| 2148 | |
| 2149 | /* |
| 2150 | * In normal cases the cur_devices == fs_devices. But in case |
| 2151 | * of deleting a seed device, the cur_devices should point to |
| 2152 | * its own fs_devices listed under the fs_devices->seed_list. |
| 2153 | */ |
| 2154 | cur_devices = device->fs_devices; |
| 2155 | mutex_lock(&fs_devices->device_list_mutex); |
| 2156 | list_del_rcu(&device->dev_list); |
| 2157 | |
| 2158 | cur_devices->num_devices--; |
| 2159 | cur_devices->total_devices--; |
| 2160 | /* Update total_devices of the parent fs_devices if it's seed */ |
| 2161 | if (cur_devices != fs_devices) |
| 2162 | fs_devices->total_devices--; |
| 2163 | |
| 2164 | if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) |
| 2165 | cur_devices->missing_devices--; |
| 2166 | |
| 2167 | btrfs_assign_next_active_device(device, NULL); |
| 2168 | |
| 2169 | if (device->bdev_handle) { |
| 2170 | cur_devices->open_devices--; |
| 2171 | /* remove sysfs entry */ |
| 2172 | btrfs_sysfs_remove_device(device); |
| 2173 | } |
| 2174 | |
| 2175 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
| 2176 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); |
| 2177 | mutex_unlock(&fs_devices->device_list_mutex); |
| 2178 | |
| 2179 | /* |
| 2180 | * At this point, the device is zero sized and detached from the |
| 2181 | * devices list. All that's left is to zero out the old supers and |
| 2182 | * free the device. |
| 2183 | * |
| 2184 | * We cannot call btrfs_close_bdev() here because we're holding the sb |
| 2185 | * write lock, and bdev_release() will pull in the ->open_mutex on |
| 2186 | * the block device and it's dependencies. Instead just flush the |
| 2187 | * device and let the caller do the final bdev_release. |
| 2188 | */ |
| 2189 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 2190 | btrfs_scratch_superblocks(fs_info, device->bdev, |
| 2191 | device->name->str); |
| 2192 | if (device->bdev) { |
| 2193 | sync_blockdev(device->bdev); |
| 2194 | invalidate_bdev(device->bdev); |
| 2195 | } |
| 2196 | } |
| 2197 | |
| 2198 | *bdev_handle = device->bdev_handle; |
| 2199 | synchronize_rcu(); |
| 2200 | btrfs_free_device(device); |
| 2201 | |
| 2202 | /* |
| 2203 | * This can happen if cur_devices is the private seed devices list. We |
| 2204 | * cannot call close_fs_devices() here because it expects the uuid_mutex |
| 2205 | * to be held, but in fact we don't need that for the private |
| 2206 | * seed_devices, we can simply decrement cur_devices->opened and then |
| 2207 | * remove it from our list and free the fs_devices. |
| 2208 | */ |
| 2209 | if (cur_devices->num_devices == 0) { |
| 2210 | list_del_init(&cur_devices->seed_list); |
| 2211 | ASSERT(cur_devices->opened == 1); |
| 2212 | cur_devices->opened--; |
| 2213 | free_fs_devices(cur_devices); |
| 2214 | } |
| 2215 | |
| 2216 | ret = btrfs_commit_transaction(trans); |
| 2217 | |
| 2218 | return ret; |
| 2219 | |
| 2220 | error_undo: |
| 2221 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 2222 | mutex_lock(&fs_info->chunk_mutex); |
| 2223 | list_add(&device->dev_alloc_list, |
| 2224 | &fs_devices->alloc_list); |
| 2225 | device->fs_devices->rw_devices++; |
| 2226 | mutex_unlock(&fs_info->chunk_mutex); |
| 2227 | } |
| 2228 | return ret; |
| 2229 | } |
| 2230 | |
| 2231 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev) |
| 2232 | { |
| 2233 | struct btrfs_fs_devices *fs_devices; |
| 2234 | |
| 2235 | lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex); |
| 2236 | |
| 2237 | /* |
| 2238 | * in case of fs with no seed, srcdev->fs_devices will point |
| 2239 | * to fs_devices of fs_info. However when the dev being replaced is |
| 2240 | * a seed dev it will point to the seed's local fs_devices. In short |
| 2241 | * srcdev will have its correct fs_devices in both the cases. |
| 2242 | */ |
| 2243 | fs_devices = srcdev->fs_devices; |
| 2244 | |
| 2245 | list_del_rcu(&srcdev->dev_list); |
| 2246 | list_del(&srcdev->dev_alloc_list); |
| 2247 | fs_devices->num_devices--; |
| 2248 | if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state)) |
| 2249 | fs_devices->missing_devices--; |
| 2250 | |
| 2251 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state)) |
| 2252 | fs_devices->rw_devices--; |
| 2253 | |
| 2254 | if (srcdev->bdev) |
| 2255 | fs_devices->open_devices--; |
| 2256 | } |
| 2257 | |
| 2258 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev) |
| 2259 | { |
| 2260 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; |
| 2261 | |
| 2262 | mutex_lock(&uuid_mutex); |
| 2263 | |
| 2264 | btrfs_close_bdev(srcdev); |
| 2265 | synchronize_rcu(); |
| 2266 | btrfs_free_device(srcdev); |
| 2267 | |
| 2268 | /* if this is no devs we rather delete the fs_devices */ |
| 2269 | if (!fs_devices->num_devices) { |
| 2270 | /* |
| 2271 | * On a mounted FS, num_devices can't be zero unless it's a |
| 2272 | * seed. In case of a seed device being replaced, the replace |
| 2273 | * target added to the sprout FS, so there will be no more |
| 2274 | * device left under the seed FS. |
| 2275 | */ |
| 2276 | ASSERT(fs_devices->seeding); |
| 2277 | |
| 2278 | list_del_init(&fs_devices->seed_list); |
| 2279 | close_fs_devices(fs_devices); |
| 2280 | free_fs_devices(fs_devices); |
| 2281 | } |
| 2282 | mutex_unlock(&uuid_mutex); |
| 2283 | } |
| 2284 | |
| 2285 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev) |
| 2286 | { |
| 2287 | struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices; |
| 2288 | |
| 2289 | mutex_lock(&fs_devices->device_list_mutex); |
| 2290 | |
| 2291 | btrfs_sysfs_remove_device(tgtdev); |
| 2292 | |
| 2293 | if (tgtdev->bdev) |
| 2294 | fs_devices->open_devices--; |
| 2295 | |
| 2296 | fs_devices->num_devices--; |
| 2297 | |
| 2298 | btrfs_assign_next_active_device(tgtdev, NULL); |
| 2299 | |
| 2300 | list_del_rcu(&tgtdev->dev_list); |
| 2301 | |
| 2302 | mutex_unlock(&fs_devices->device_list_mutex); |
| 2303 | |
| 2304 | btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev, |
| 2305 | tgtdev->name->str); |
| 2306 | |
| 2307 | btrfs_close_bdev(tgtdev); |
| 2308 | synchronize_rcu(); |
| 2309 | btrfs_free_device(tgtdev); |
| 2310 | } |
| 2311 | |
| 2312 | /* |
| 2313 | * Populate args from device at path. |
| 2314 | * |
| 2315 | * @fs_info: the filesystem |
| 2316 | * @args: the args to populate |
| 2317 | * @path: the path to the device |
| 2318 | * |
| 2319 | * This will read the super block of the device at @path and populate @args with |
| 2320 | * the devid, fsid, and uuid. This is meant to be used for ioctls that need to |
| 2321 | * lookup a device to operate on, but need to do it before we take any locks. |
| 2322 | * This properly handles the special case of "missing" that a user may pass in, |
| 2323 | * and does some basic sanity checks. The caller must make sure that @path is |
| 2324 | * properly NUL terminated before calling in, and must call |
| 2325 | * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and |
| 2326 | * uuid buffers. |
| 2327 | * |
| 2328 | * Return: 0 for success, -errno for failure |
| 2329 | */ |
| 2330 | int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info, |
| 2331 | struct btrfs_dev_lookup_args *args, |
| 2332 | const char *path) |
| 2333 | { |
| 2334 | struct btrfs_super_block *disk_super; |
| 2335 | struct bdev_handle *bdev_handle; |
| 2336 | int ret; |
| 2337 | |
| 2338 | if (!path || !path[0]) |
| 2339 | return -EINVAL; |
| 2340 | if (!strcmp(path, "missing")) { |
| 2341 | args->missing = true; |
| 2342 | return 0; |
| 2343 | } |
| 2344 | |
| 2345 | args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL); |
| 2346 | args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL); |
| 2347 | if (!args->uuid || !args->fsid) { |
| 2348 | btrfs_put_dev_args_from_path(args); |
| 2349 | return -ENOMEM; |
| 2350 | } |
| 2351 | |
| 2352 | ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0, |
| 2353 | &bdev_handle, &disk_super); |
| 2354 | if (ret) { |
| 2355 | btrfs_put_dev_args_from_path(args); |
| 2356 | return ret; |
| 2357 | } |
| 2358 | |
| 2359 | args->devid = btrfs_stack_device_id(&disk_super->dev_item); |
| 2360 | memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE); |
| 2361 | if (btrfs_fs_incompat(fs_info, METADATA_UUID)) |
| 2362 | memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE); |
| 2363 | else |
| 2364 | memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE); |
| 2365 | btrfs_release_disk_super(disk_super); |
| 2366 | bdev_release(bdev_handle); |
| 2367 | return 0; |
| 2368 | } |
| 2369 | |
| 2370 | /* |
| 2371 | * Only use this jointly with btrfs_get_dev_args_from_path() because we will |
| 2372 | * allocate our ->uuid and ->fsid pointers, everybody else uses local variables |
| 2373 | * that don't need to be freed. |
| 2374 | */ |
| 2375 | void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args) |
| 2376 | { |
| 2377 | kfree(args->uuid); |
| 2378 | kfree(args->fsid); |
| 2379 | args->uuid = NULL; |
| 2380 | args->fsid = NULL; |
| 2381 | } |
| 2382 | |
| 2383 | struct btrfs_device *btrfs_find_device_by_devspec( |
| 2384 | struct btrfs_fs_info *fs_info, u64 devid, |
| 2385 | const char *device_path) |
| 2386 | { |
| 2387 | BTRFS_DEV_LOOKUP_ARGS(args); |
| 2388 | struct btrfs_device *device; |
| 2389 | int ret; |
| 2390 | |
| 2391 | if (devid) { |
| 2392 | args.devid = devid; |
| 2393 | device = btrfs_find_device(fs_info->fs_devices, &args); |
| 2394 | if (!device) |
| 2395 | return ERR_PTR(-ENOENT); |
| 2396 | return device; |
| 2397 | } |
| 2398 | |
| 2399 | ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path); |
| 2400 | if (ret) |
| 2401 | return ERR_PTR(ret); |
| 2402 | device = btrfs_find_device(fs_info->fs_devices, &args); |
| 2403 | btrfs_put_dev_args_from_path(&args); |
| 2404 | if (!device) |
| 2405 | return ERR_PTR(-ENOENT); |
| 2406 | return device; |
| 2407 | } |
| 2408 | |
| 2409 | static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info) |
| 2410 | { |
| 2411 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 2412 | struct btrfs_fs_devices *old_devices; |
| 2413 | struct btrfs_fs_devices *seed_devices; |
| 2414 | |
| 2415 | lockdep_assert_held(&uuid_mutex); |
| 2416 | if (!fs_devices->seeding) |
| 2417 | return ERR_PTR(-EINVAL); |
| 2418 | |
| 2419 | /* |
| 2420 | * Private copy of the seed devices, anchored at |
| 2421 | * fs_info->fs_devices->seed_list |
| 2422 | */ |
| 2423 | seed_devices = alloc_fs_devices(NULL); |
| 2424 | if (IS_ERR(seed_devices)) |
| 2425 | return seed_devices; |
| 2426 | |
| 2427 | /* |
| 2428 | * It's necessary to retain a copy of the original seed fs_devices in |
| 2429 | * fs_uuids so that filesystems which have been seeded can successfully |
| 2430 | * reference the seed device from open_seed_devices. This also supports |
| 2431 | * multiple fs seed. |
| 2432 | */ |
| 2433 | old_devices = clone_fs_devices(fs_devices); |
| 2434 | if (IS_ERR(old_devices)) { |
| 2435 | kfree(seed_devices); |
| 2436 | return old_devices; |
| 2437 | } |
| 2438 | |
| 2439 | list_add(&old_devices->fs_list, &fs_uuids); |
| 2440 | |
| 2441 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
| 2442 | seed_devices->opened = 1; |
| 2443 | INIT_LIST_HEAD(&seed_devices->devices); |
| 2444 | INIT_LIST_HEAD(&seed_devices->alloc_list); |
| 2445 | mutex_init(&seed_devices->device_list_mutex); |
| 2446 | |
| 2447 | return seed_devices; |
| 2448 | } |
| 2449 | |
| 2450 | /* |
| 2451 | * Splice seed devices into the sprout fs_devices. |
| 2452 | * Generate a new fsid for the sprouted read-write filesystem. |
| 2453 | */ |
| 2454 | static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info, |
| 2455 | struct btrfs_fs_devices *seed_devices) |
| 2456 | { |
| 2457 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 2458 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
| 2459 | struct btrfs_device *device; |
| 2460 | u64 super_flags; |
| 2461 | |
| 2462 | /* |
| 2463 | * We are updating the fsid, the thread leading to device_list_add() |
| 2464 | * could race, so uuid_mutex is needed. |
| 2465 | */ |
| 2466 | lockdep_assert_held(&uuid_mutex); |
| 2467 | |
| 2468 | /* |
| 2469 | * The threads listed below may traverse dev_list but can do that without |
| 2470 | * device_list_mutex: |
| 2471 | * - All device ops and balance - as we are in btrfs_exclop_start. |
| 2472 | * - Various dev_list readers - are using RCU. |
| 2473 | * - btrfs_ioctl_fitrim() - is using RCU. |
| 2474 | * |
| 2475 | * For-read threads as below are using device_list_mutex: |
| 2476 | * - Readonly scrub btrfs_scrub_dev() |
| 2477 | * - Readonly scrub btrfs_scrub_progress() |
| 2478 | * - btrfs_get_dev_stats() |
| 2479 | */ |
| 2480 | lockdep_assert_held(&fs_devices->device_list_mutex); |
| 2481 | |
| 2482 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
| 2483 | synchronize_rcu); |
| 2484 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
| 2485 | device->fs_devices = seed_devices; |
| 2486 | |
| 2487 | fs_devices->seeding = false; |
| 2488 | fs_devices->num_devices = 0; |
| 2489 | fs_devices->open_devices = 0; |
| 2490 | fs_devices->missing_devices = 0; |
| 2491 | fs_devices->rotating = false; |
| 2492 | list_add(&seed_devices->seed_list, &fs_devices->seed_list); |
| 2493 | |
| 2494 | generate_random_uuid(fs_devices->fsid); |
| 2495 | memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE); |
| 2496 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
| 2497 | |
| 2498 | super_flags = btrfs_super_flags(disk_super) & |
| 2499 | ~BTRFS_SUPER_FLAG_SEEDING; |
| 2500 | btrfs_set_super_flags(disk_super, super_flags); |
| 2501 | } |
| 2502 | |
| 2503 | /* |
| 2504 | * Store the expected generation for seed devices in device items. |
| 2505 | */ |
| 2506 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans) |
| 2507 | { |
| 2508 | BTRFS_DEV_LOOKUP_ARGS(args); |
| 2509 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2510 | struct btrfs_root *root = fs_info->chunk_root; |
| 2511 | struct btrfs_path *path; |
| 2512 | struct extent_buffer *leaf; |
| 2513 | struct btrfs_dev_item *dev_item; |
| 2514 | struct btrfs_device *device; |
| 2515 | struct btrfs_key key; |
| 2516 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
| 2517 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
| 2518 | int ret; |
| 2519 | |
| 2520 | path = btrfs_alloc_path(); |
| 2521 | if (!path) |
| 2522 | return -ENOMEM; |
| 2523 | |
| 2524 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 2525 | key.offset = 0; |
| 2526 | key.type = BTRFS_DEV_ITEM_KEY; |
| 2527 | |
| 2528 | while (1) { |
| 2529 | btrfs_reserve_chunk_metadata(trans, false); |
| 2530 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
| 2531 | btrfs_trans_release_chunk_metadata(trans); |
| 2532 | if (ret < 0) |
| 2533 | goto error; |
| 2534 | |
| 2535 | leaf = path->nodes[0]; |
| 2536 | next_slot: |
| 2537 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
| 2538 | ret = btrfs_next_leaf(root, path); |
| 2539 | if (ret > 0) |
| 2540 | break; |
| 2541 | if (ret < 0) |
| 2542 | goto error; |
| 2543 | leaf = path->nodes[0]; |
| 2544 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2545 | btrfs_release_path(path); |
| 2546 | continue; |
| 2547 | } |
| 2548 | |
| 2549 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2550 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || |
| 2551 | key.type != BTRFS_DEV_ITEM_KEY) |
| 2552 | break; |
| 2553 | |
| 2554 | dev_item = btrfs_item_ptr(leaf, path->slots[0], |
| 2555 | struct btrfs_dev_item); |
| 2556 | args.devid = btrfs_device_id(leaf, dev_item); |
| 2557 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
| 2558 | BTRFS_UUID_SIZE); |
| 2559 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
| 2560 | BTRFS_FSID_SIZE); |
| 2561 | args.uuid = dev_uuid; |
| 2562 | args.fsid = fs_uuid; |
| 2563 | device = btrfs_find_device(fs_info->fs_devices, &args); |
| 2564 | BUG_ON(!device); /* Logic error */ |
| 2565 | |
| 2566 | if (device->fs_devices->seeding) { |
| 2567 | btrfs_set_device_generation(leaf, dev_item, |
| 2568 | device->generation); |
| 2569 | btrfs_mark_buffer_dirty(trans, leaf); |
| 2570 | } |
| 2571 | |
| 2572 | path->slots[0]++; |
| 2573 | goto next_slot; |
| 2574 | } |
| 2575 | ret = 0; |
| 2576 | error: |
| 2577 | btrfs_free_path(path); |
| 2578 | return ret; |
| 2579 | } |
| 2580 | |
| 2581 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path) |
| 2582 | { |
| 2583 | struct btrfs_root *root = fs_info->dev_root; |
| 2584 | struct btrfs_trans_handle *trans; |
| 2585 | struct btrfs_device *device; |
| 2586 | struct bdev_handle *bdev_handle; |
| 2587 | struct super_block *sb = fs_info->sb; |
| 2588 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 2589 | struct btrfs_fs_devices *seed_devices = NULL; |
| 2590 | u64 orig_super_total_bytes; |
| 2591 | u64 orig_super_num_devices; |
| 2592 | int ret = 0; |
| 2593 | bool seeding_dev = false; |
| 2594 | bool locked = false; |
| 2595 | |
| 2596 | if (sb_rdonly(sb) && !fs_devices->seeding) |
| 2597 | return -EROFS; |
| 2598 | |
| 2599 | bdev_handle = bdev_open_by_path(device_path, BLK_OPEN_WRITE, |
| 2600 | fs_info->bdev_holder, NULL); |
| 2601 | if (IS_ERR(bdev_handle)) |
| 2602 | return PTR_ERR(bdev_handle); |
| 2603 | |
| 2604 | if (!btrfs_check_device_zone_type(fs_info, bdev_handle->bdev)) { |
| 2605 | ret = -EINVAL; |
| 2606 | goto error; |
| 2607 | } |
| 2608 | |
| 2609 | if (fs_devices->seeding) { |
| 2610 | seeding_dev = true; |
| 2611 | down_write(&sb->s_umount); |
| 2612 | mutex_lock(&uuid_mutex); |
| 2613 | locked = true; |
| 2614 | } |
| 2615 | |
| 2616 | sync_blockdev(bdev_handle->bdev); |
| 2617 | |
| 2618 | rcu_read_lock(); |
| 2619 | list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) { |
| 2620 | if (device->bdev == bdev_handle->bdev) { |
| 2621 | ret = -EEXIST; |
| 2622 | rcu_read_unlock(); |
| 2623 | goto error; |
| 2624 | } |
| 2625 | } |
| 2626 | rcu_read_unlock(); |
| 2627 | |
| 2628 | device = btrfs_alloc_device(fs_info, NULL, NULL, device_path); |
| 2629 | if (IS_ERR(device)) { |
| 2630 | /* we can safely leave the fs_devices entry around */ |
| 2631 | ret = PTR_ERR(device); |
| 2632 | goto error; |
| 2633 | } |
| 2634 | |
| 2635 | device->fs_info = fs_info; |
| 2636 | device->bdev_handle = bdev_handle; |
| 2637 | device->bdev = bdev_handle->bdev; |
| 2638 | ret = lookup_bdev(device_path, &device->devt); |
| 2639 | if (ret) |
| 2640 | goto error_free_device; |
| 2641 | |
| 2642 | ret = btrfs_get_dev_zone_info(device, false); |
| 2643 | if (ret) |
| 2644 | goto error_free_device; |
| 2645 | |
| 2646 | trans = btrfs_start_transaction(root, 0); |
| 2647 | if (IS_ERR(trans)) { |
| 2648 | ret = PTR_ERR(trans); |
| 2649 | goto error_free_zone; |
| 2650 | } |
| 2651 | |
| 2652 | set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state); |
| 2653 | device->generation = trans->transid; |
| 2654 | device->io_width = fs_info->sectorsize; |
| 2655 | device->io_align = fs_info->sectorsize; |
| 2656 | device->sector_size = fs_info->sectorsize; |
| 2657 | device->total_bytes = |
| 2658 | round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize); |
| 2659 | device->disk_total_bytes = device->total_bytes; |
| 2660 | device->commit_total_bytes = device->total_bytes; |
| 2661 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
| 2662 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
| 2663 | device->dev_stats_valid = 1; |
| 2664 | set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE); |
| 2665 | |
| 2666 | if (seeding_dev) { |
| 2667 | btrfs_clear_sb_rdonly(sb); |
| 2668 | |
| 2669 | /* GFP_KERNEL allocation must not be under device_list_mutex */ |
| 2670 | seed_devices = btrfs_init_sprout(fs_info); |
| 2671 | if (IS_ERR(seed_devices)) { |
| 2672 | ret = PTR_ERR(seed_devices); |
| 2673 | btrfs_abort_transaction(trans, ret); |
| 2674 | goto error_trans; |
| 2675 | } |
| 2676 | } |
| 2677 | |
| 2678 | mutex_lock(&fs_devices->device_list_mutex); |
| 2679 | if (seeding_dev) { |
| 2680 | btrfs_setup_sprout(fs_info, seed_devices); |
| 2681 | btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev, |
| 2682 | device); |
| 2683 | } |
| 2684 | |
| 2685 | device->fs_devices = fs_devices; |
| 2686 | |
| 2687 | mutex_lock(&fs_info->chunk_mutex); |
| 2688 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
| 2689 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); |
| 2690 | fs_devices->num_devices++; |
| 2691 | fs_devices->open_devices++; |
| 2692 | fs_devices->rw_devices++; |
| 2693 | fs_devices->total_devices++; |
| 2694 | fs_devices->total_rw_bytes += device->total_bytes; |
| 2695 | |
| 2696 | atomic64_add(device->total_bytes, &fs_info->free_chunk_space); |
| 2697 | |
| 2698 | if (!bdev_nonrot(device->bdev)) |
| 2699 | fs_devices->rotating = true; |
| 2700 | |
| 2701 | orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
| 2702 | btrfs_set_super_total_bytes(fs_info->super_copy, |
| 2703 | round_down(orig_super_total_bytes + device->total_bytes, |
| 2704 | fs_info->sectorsize)); |
| 2705 | |
| 2706 | orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy); |
| 2707 | btrfs_set_super_num_devices(fs_info->super_copy, |
| 2708 | orig_super_num_devices + 1); |
| 2709 | |
| 2710 | /* |
| 2711 | * we've got more storage, clear any full flags on the space |
| 2712 | * infos |
| 2713 | */ |
| 2714 | btrfs_clear_space_info_full(fs_info); |
| 2715 | |
| 2716 | mutex_unlock(&fs_info->chunk_mutex); |
| 2717 | |
| 2718 | /* Add sysfs device entry */ |
| 2719 | btrfs_sysfs_add_device(device); |
| 2720 | |
| 2721 | mutex_unlock(&fs_devices->device_list_mutex); |
| 2722 | |
| 2723 | if (seeding_dev) { |
| 2724 | mutex_lock(&fs_info->chunk_mutex); |
| 2725 | ret = init_first_rw_device(trans); |
| 2726 | mutex_unlock(&fs_info->chunk_mutex); |
| 2727 | if (ret) { |
| 2728 | btrfs_abort_transaction(trans, ret); |
| 2729 | goto error_sysfs; |
| 2730 | } |
| 2731 | } |
| 2732 | |
| 2733 | ret = btrfs_add_dev_item(trans, device); |
| 2734 | if (ret) { |
| 2735 | btrfs_abort_transaction(trans, ret); |
| 2736 | goto error_sysfs; |
| 2737 | } |
| 2738 | |
| 2739 | if (seeding_dev) { |
| 2740 | ret = btrfs_finish_sprout(trans); |
| 2741 | if (ret) { |
| 2742 | btrfs_abort_transaction(trans, ret); |
| 2743 | goto error_sysfs; |
| 2744 | } |
| 2745 | |
| 2746 | /* |
| 2747 | * fs_devices now represents the newly sprouted filesystem and |
| 2748 | * its fsid has been changed by btrfs_sprout_splice(). |
| 2749 | */ |
| 2750 | btrfs_sysfs_update_sprout_fsid(fs_devices); |
| 2751 | } |
| 2752 | |
| 2753 | ret = btrfs_commit_transaction(trans); |
| 2754 | |
| 2755 | if (seeding_dev) { |
| 2756 | mutex_unlock(&uuid_mutex); |
| 2757 | up_write(&sb->s_umount); |
| 2758 | locked = false; |
| 2759 | |
| 2760 | if (ret) /* transaction commit */ |
| 2761 | return ret; |
| 2762 | |
| 2763 | ret = btrfs_relocate_sys_chunks(fs_info); |
| 2764 | if (ret < 0) |
| 2765 | btrfs_handle_fs_error(fs_info, ret, |
| 2766 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
| 2767 | trans = btrfs_attach_transaction(root); |
| 2768 | if (IS_ERR(trans)) { |
| 2769 | if (PTR_ERR(trans) == -ENOENT) |
| 2770 | return 0; |
| 2771 | ret = PTR_ERR(trans); |
| 2772 | trans = NULL; |
| 2773 | goto error_sysfs; |
| 2774 | } |
| 2775 | ret = btrfs_commit_transaction(trans); |
| 2776 | } |
| 2777 | |
| 2778 | /* |
| 2779 | * Now that we have written a new super block to this device, check all |
| 2780 | * other fs_devices list if device_path alienates any other scanned |
| 2781 | * device. |
| 2782 | * We can ignore the return value as it typically returns -EINVAL and |
| 2783 | * only succeeds if the device was an alien. |
| 2784 | */ |
| 2785 | btrfs_forget_devices(device->devt); |
| 2786 | |
| 2787 | /* Update ctime/mtime for blkid or udev */ |
| 2788 | update_dev_time(device_path); |
| 2789 | |
| 2790 | return ret; |
| 2791 | |
| 2792 | error_sysfs: |
| 2793 | btrfs_sysfs_remove_device(device); |
| 2794 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
| 2795 | mutex_lock(&fs_info->chunk_mutex); |
| 2796 | list_del_rcu(&device->dev_list); |
| 2797 | list_del(&device->dev_alloc_list); |
| 2798 | fs_info->fs_devices->num_devices--; |
| 2799 | fs_info->fs_devices->open_devices--; |
| 2800 | fs_info->fs_devices->rw_devices--; |
| 2801 | fs_info->fs_devices->total_devices--; |
| 2802 | fs_info->fs_devices->total_rw_bytes -= device->total_bytes; |
| 2803 | atomic64_sub(device->total_bytes, &fs_info->free_chunk_space); |
| 2804 | btrfs_set_super_total_bytes(fs_info->super_copy, |
| 2805 | orig_super_total_bytes); |
| 2806 | btrfs_set_super_num_devices(fs_info->super_copy, |
| 2807 | orig_super_num_devices); |
| 2808 | mutex_unlock(&fs_info->chunk_mutex); |
| 2809 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
| 2810 | error_trans: |
| 2811 | if (seeding_dev) |
| 2812 | btrfs_set_sb_rdonly(sb); |
| 2813 | if (trans) |
| 2814 | btrfs_end_transaction(trans); |
| 2815 | error_free_zone: |
| 2816 | btrfs_destroy_dev_zone_info(device); |
| 2817 | error_free_device: |
| 2818 | btrfs_free_device(device); |
| 2819 | error: |
| 2820 | bdev_release(bdev_handle); |
| 2821 | if (locked) { |
| 2822 | mutex_unlock(&uuid_mutex); |
| 2823 | up_write(&sb->s_umount); |
| 2824 | } |
| 2825 | return ret; |
| 2826 | } |
| 2827 | |
| 2828 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
| 2829 | struct btrfs_device *device) |
| 2830 | { |
| 2831 | int ret; |
| 2832 | struct btrfs_path *path; |
| 2833 | struct btrfs_root *root = device->fs_info->chunk_root; |
| 2834 | struct btrfs_dev_item *dev_item; |
| 2835 | struct extent_buffer *leaf; |
| 2836 | struct btrfs_key key; |
| 2837 | |
| 2838 | path = btrfs_alloc_path(); |
| 2839 | if (!path) |
| 2840 | return -ENOMEM; |
| 2841 | |
| 2842 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 2843 | key.type = BTRFS_DEV_ITEM_KEY; |
| 2844 | key.offset = device->devid; |
| 2845 | |
| 2846 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); |
| 2847 | if (ret < 0) |
| 2848 | goto out; |
| 2849 | |
| 2850 | if (ret > 0) { |
| 2851 | ret = -ENOENT; |
| 2852 | goto out; |
| 2853 | } |
| 2854 | |
| 2855 | leaf = path->nodes[0]; |
| 2856 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); |
| 2857 | |
| 2858 | btrfs_set_device_id(leaf, dev_item, device->devid); |
| 2859 | btrfs_set_device_type(leaf, dev_item, device->type); |
| 2860 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); |
| 2861 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); |
| 2862 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); |
| 2863 | btrfs_set_device_total_bytes(leaf, dev_item, |
| 2864 | btrfs_device_get_disk_total_bytes(device)); |
| 2865 | btrfs_set_device_bytes_used(leaf, dev_item, |
| 2866 | btrfs_device_get_bytes_used(device)); |
| 2867 | btrfs_mark_buffer_dirty(trans, leaf); |
| 2868 | |
| 2869 | out: |
| 2870 | btrfs_free_path(path); |
| 2871 | return ret; |
| 2872 | } |
| 2873 | |
| 2874 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
| 2875 | struct btrfs_device *device, u64 new_size) |
| 2876 | { |
| 2877 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 2878 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
| 2879 | u64 old_total; |
| 2880 | u64 diff; |
| 2881 | int ret; |
| 2882 | |
| 2883 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
| 2884 | return -EACCES; |
| 2885 | |
| 2886 | new_size = round_down(new_size, fs_info->sectorsize); |
| 2887 | |
| 2888 | mutex_lock(&fs_info->chunk_mutex); |
| 2889 | old_total = btrfs_super_total_bytes(super_copy); |
| 2890 | diff = round_down(new_size - device->total_bytes, fs_info->sectorsize); |
| 2891 | |
| 2892 | if (new_size <= device->total_bytes || |
| 2893 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
| 2894 | mutex_unlock(&fs_info->chunk_mutex); |
| 2895 | return -EINVAL; |
| 2896 | } |
| 2897 | |
| 2898 | btrfs_set_super_total_bytes(super_copy, |
| 2899 | round_down(old_total + diff, fs_info->sectorsize)); |
| 2900 | device->fs_devices->total_rw_bytes += diff; |
| 2901 | atomic64_add(diff, &fs_info->free_chunk_space); |
| 2902 | |
| 2903 | btrfs_device_set_total_bytes(device, new_size); |
| 2904 | btrfs_device_set_disk_total_bytes(device, new_size); |
| 2905 | btrfs_clear_space_info_full(device->fs_info); |
| 2906 | if (list_empty(&device->post_commit_list)) |
| 2907 | list_add_tail(&device->post_commit_list, |
| 2908 | &trans->transaction->dev_update_list); |
| 2909 | mutex_unlock(&fs_info->chunk_mutex); |
| 2910 | |
| 2911 | btrfs_reserve_chunk_metadata(trans, false); |
| 2912 | ret = btrfs_update_device(trans, device); |
| 2913 | btrfs_trans_release_chunk_metadata(trans); |
| 2914 | |
| 2915 | return ret; |
| 2916 | } |
| 2917 | |
| 2918 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
| 2919 | { |
| 2920 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2921 | struct btrfs_root *root = fs_info->chunk_root; |
| 2922 | int ret; |
| 2923 | struct btrfs_path *path; |
| 2924 | struct btrfs_key key; |
| 2925 | |
| 2926 | path = btrfs_alloc_path(); |
| 2927 | if (!path) |
| 2928 | return -ENOMEM; |
| 2929 | |
| 2930 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
| 2931 | key.offset = chunk_offset; |
| 2932 | key.type = BTRFS_CHUNK_ITEM_KEY; |
| 2933 | |
| 2934 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 2935 | if (ret < 0) |
| 2936 | goto out; |
| 2937 | else if (ret > 0) { /* Logic error or corruption */ |
| 2938 | btrfs_handle_fs_error(fs_info, -ENOENT, |
| 2939 | "Failed lookup while freeing chunk."); |
| 2940 | ret = -ENOENT; |
| 2941 | goto out; |
| 2942 | } |
| 2943 | |
| 2944 | ret = btrfs_del_item(trans, root, path); |
| 2945 | if (ret < 0) |
| 2946 | btrfs_handle_fs_error(fs_info, ret, |
| 2947 | "Failed to delete chunk item."); |
| 2948 | out: |
| 2949 | btrfs_free_path(path); |
| 2950 | return ret; |
| 2951 | } |
| 2952 | |
| 2953 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
| 2954 | { |
| 2955 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
| 2956 | struct btrfs_disk_key *disk_key; |
| 2957 | struct btrfs_chunk *chunk; |
| 2958 | u8 *ptr; |
| 2959 | int ret = 0; |
| 2960 | u32 num_stripes; |
| 2961 | u32 array_size; |
| 2962 | u32 len = 0; |
| 2963 | u32 cur; |
| 2964 | struct btrfs_key key; |
| 2965 | |
| 2966 | lockdep_assert_held(&fs_info->chunk_mutex); |
| 2967 | array_size = btrfs_super_sys_array_size(super_copy); |
| 2968 | |
| 2969 | ptr = super_copy->sys_chunk_array; |
| 2970 | cur = 0; |
| 2971 | |
| 2972 | while (cur < array_size) { |
| 2973 | disk_key = (struct btrfs_disk_key *)ptr; |
| 2974 | btrfs_disk_key_to_cpu(&key, disk_key); |
| 2975 | |
| 2976 | len = sizeof(*disk_key); |
| 2977 | |
| 2978 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
| 2979 | chunk = (struct btrfs_chunk *)(ptr + len); |
| 2980 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); |
| 2981 | len += btrfs_chunk_item_size(num_stripes); |
| 2982 | } else { |
| 2983 | ret = -EIO; |
| 2984 | break; |
| 2985 | } |
| 2986 | if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID && |
| 2987 | key.offset == chunk_offset) { |
| 2988 | memmove(ptr, ptr + len, array_size - (cur + len)); |
| 2989 | array_size -= len; |
| 2990 | btrfs_set_super_sys_array_size(super_copy, array_size); |
| 2991 | } else { |
| 2992 | ptr += len; |
| 2993 | cur += len; |
| 2994 | } |
| 2995 | } |
| 2996 | return ret; |
| 2997 | } |
| 2998 | |
| 2999 | struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info, |
| 3000 | u64 logical, u64 length) |
| 3001 | { |
| 3002 | struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node; |
| 3003 | struct rb_node *prev = NULL; |
| 3004 | struct rb_node *orig_prev; |
| 3005 | struct btrfs_chunk_map *map; |
| 3006 | struct btrfs_chunk_map *prev_map = NULL; |
| 3007 | |
| 3008 | while (node) { |
| 3009 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); |
| 3010 | prev = node; |
| 3011 | prev_map = map; |
| 3012 | |
| 3013 | if (logical < map->start) { |
| 3014 | node = node->rb_left; |
| 3015 | } else if (logical >= map->start + map->chunk_len) { |
| 3016 | node = node->rb_right; |
| 3017 | } else { |
| 3018 | refcount_inc(&map->refs); |
| 3019 | return map; |
| 3020 | } |
| 3021 | } |
| 3022 | |
| 3023 | if (!prev) |
| 3024 | return NULL; |
| 3025 | |
| 3026 | orig_prev = prev; |
| 3027 | while (prev && logical >= prev_map->start + prev_map->chunk_len) { |
| 3028 | prev = rb_next(prev); |
| 3029 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); |
| 3030 | } |
| 3031 | |
| 3032 | if (!prev) { |
| 3033 | prev = orig_prev; |
| 3034 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); |
| 3035 | while (prev && logical < prev_map->start) { |
| 3036 | prev = rb_prev(prev); |
| 3037 | prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node); |
| 3038 | } |
| 3039 | } |
| 3040 | |
| 3041 | if (prev) { |
| 3042 | u64 end = logical + length; |
| 3043 | |
| 3044 | /* |
| 3045 | * Caller can pass a U64_MAX length when it wants to get any |
| 3046 | * chunk starting at an offset of 'logical' or higher, so deal |
| 3047 | * with underflow by resetting the end offset to U64_MAX. |
| 3048 | */ |
| 3049 | if (end < logical) |
| 3050 | end = U64_MAX; |
| 3051 | |
| 3052 | if (end > prev_map->start && |
| 3053 | logical < prev_map->start + prev_map->chunk_len) { |
| 3054 | refcount_inc(&prev_map->refs); |
| 3055 | return prev_map; |
| 3056 | } |
| 3057 | } |
| 3058 | |
| 3059 | return NULL; |
| 3060 | } |
| 3061 | |
| 3062 | struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info, |
| 3063 | u64 logical, u64 length) |
| 3064 | { |
| 3065 | struct btrfs_chunk_map *map; |
| 3066 | |
| 3067 | read_lock(&fs_info->mapping_tree_lock); |
| 3068 | map = btrfs_find_chunk_map_nolock(fs_info, logical, length); |
| 3069 | read_unlock(&fs_info->mapping_tree_lock); |
| 3070 | |
| 3071 | return map; |
| 3072 | } |
| 3073 | |
| 3074 | /* |
| 3075 | * Find the mapping containing the given logical extent. |
| 3076 | * |
| 3077 | * @logical: Logical block offset in bytes. |
| 3078 | * @length: Length of extent in bytes. |
| 3079 | * |
| 3080 | * Return: Chunk mapping or ERR_PTR. |
| 3081 | */ |
| 3082 | struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info, |
| 3083 | u64 logical, u64 length) |
| 3084 | { |
| 3085 | struct btrfs_chunk_map *map; |
| 3086 | |
| 3087 | map = btrfs_find_chunk_map(fs_info, logical, length); |
| 3088 | |
| 3089 | if (unlikely(!map)) { |
| 3090 | read_unlock(&fs_info->mapping_tree_lock); |
| 3091 | btrfs_crit(fs_info, |
| 3092 | "unable to find chunk map for logical %llu length %llu", |
| 3093 | logical, length); |
| 3094 | return ERR_PTR(-EINVAL); |
| 3095 | } |
| 3096 | |
| 3097 | if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) { |
| 3098 | read_unlock(&fs_info->mapping_tree_lock); |
| 3099 | btrfs_crit(fs_info, |
| 3100 | "found a bad chunk map, wanted %llu-%llu, found %llu-%llu", |
| 3101 | logical, logical + length, map->start, |
| 3102 | map->start + map->chunk_len); |
| 3103 | btrfs_free_chunk_map(map); |
| 3104 | return ERR_PTR(-EINVAL); |
| 3105 | } |
| 3106 | |
| 3107 | /* Callers are responsible for dropping the reference. */ |
| 3108 | return map; |
| 3109 | } |
| 3110 | |
| 3111 | static int remove_chunk_item(struct btrfs_trans_handle *trans, |
| 3112 | struct btrfs_chunk_map *map, u64 chunk_offset) |
| 3113 | { |
| 3114 | int i; |
| 3115 | |
| 3116 | /* |
| 3117 | * Removing chunk items and updating the device items in the chunks btree |
| 3118 | * requires holding the chunk_mutex. |
| 3119 | * See the comment at btrfs_chunk_alloc() for the details. |
| 3120 | */ |
| 3121 | lockdep_assert_held(&trans->fs_info->chunk_mutex); |
| 3122 | |
| 3123 | for (i = 0; i < map->num_stripes; i++) { |
| 3124 | int ret; |
| 3125 | |
| 3126 | ret = btrfs_update_device(trans, map->stripes[i].dev); |
| 3127 | if (ret) |
| 3128 | return ret; |
| 3129 | } |
| 3130 | |
| 3131 | return btrfs_free_chunk(trans, chunk_offset); |
| 3132 | } |
| 3133 | |
| 3134 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset) |
| 3135 | { |
| 3136 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 3137 | struct btrfs_chunk_map *map; |
| 3138 | u64 dev_extent_len = 0; |
| 3139 | int i, ret = 0; |
| 3140 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 3141 | |
| 3142 | map = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
| 3143 | if (IS_ERR(map)) { |
| 3144 | /* |
| 3145 | * This is a logic error, but we don't want to just rely on the |
| 3146 | * user having built with ASSERT enabled, so if ASSERT doesn't |
| 3147 | * do anything we still error out. |
| 3148 | */ |
| 3149 | ASSERT(0); |
| 3150 | return PTR_ERR(map); |
| 3151 | } |
| 3152 | |
| 3153 | /* |
| 3154 | * First delete the device extent items from the devices btree. |
| 3155 | * We take the device_list_mutex to avoid racing with the finishing phase |
| 3156 | * of a device replace operation. See the comment below before acquiring |
| 3157 | * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex |
| 3158 | * because that can result in a deadlock when deleting the device extent |
| 3159 | * items from the devices btree - COWing an extent buffer from the btree |
| 3160 | * may result in allocating a new metadata chunk, which would attempt to |
| 3161 | * lock again fs_info->chunk_mutex. |
| 3162 | */ |
| 3163 | mutex_lock(&fs_devices->device_list_mutex); |
| 3164 | for (i = 0; i < map->num_stripes; i++) { |
| 3165 | struct btrfs_device *device = map->stripes[i].dev; |
| 3166 | ret = btrfs_free_dev_extent(trans, device, |
| 3167 | map->stripes[i].physical, |
| 3168 | &dev_extent_len); |
| 3169 | if (ret) { |
| 3170 | mutex_unlock(&fs_devices->device_list_mutex); |
| 3171 | btrfs_abort_transaction(trans, ret); |
| 3172 | goto out; |
| 3173 | } |
| 3174 | |
| 3175 | if (device->bytes_used > 0) { |
| 3176 | mutex_lock(&fs_info->chunk_mutex); |
| 3177 | btrfs_device_set_bytes_used(device, |
| 3178 | device->bytes_used - dev_extent_len); |
| 3179 | atomic64_add(dev_extent_len, &fs_info->free_chunk_space); |
| 3180 | btrfs_clear_space_info_full(fs_info); |
| 3181 | mutex_unlock(&fs_info->chunk_mutex); |
| 3182 | } |
| 3183 | } |
| 3184 | mutex_unlock(&fs_devices->device_list_mutex); |
| 3185 | |
| 3186 | /* |
| 3187 | * We acquire fs_info->chunk_mutex for 2 reasons: |
| 3188 | * |
| 3189 | * 1) Just like with the first phase of the chunk allocation, we must |
| 3190 | * reserve system space, do all chunk btree updates and deletions, and |
| 3191 | * update the system chunk array in the superblock while holding this |
| 3192 | * mutex. This is for similar reasons as explained on the comment at |
| 3193 | * the top of btrfs_chunk_alloc(); |
| 3194 | * |
| 3195 | * 2) Prevent races with the final phase of a device replace operation |
| 3196 | * that replaces the device object associated with the map's stripes, |
| 3197 | * because the device object's id can change at any time during that |
| 3198 | * final phase of the device replace operation |
| 3199 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the |
| 3200 | * replaced device and then see it with an ID of |
| 3201 | * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating |
| 3202 | * the device item, which does not exists on the chunk btree. |
| 3203 | * The finishing phase of device replace acquires both the |
| 3204 | * device_list_mutex and the chunk_mutex, in that order, so we are |
| 3205 | * safe by just acquiring the chunk_mutex. |
| 3206 | */ |
| 3207 | trans->removing_chunk = true; |
| 3208 | mutex_lock(&fs_info->chunk_mutex); |
| 3209 | |
| 3210 | check_system_chunk(trans, map->type); |
| 3211 | |
| 3212 | ret = remove_chunk_item(trans, map, chunk_offset); |
| 3213 | /* |
| 3214 | * Normally we should not get -ENOSPC since we reserved space before |
| 3215 | * through the call to check_system_chunk(). |
| 3216 | * |
| 3217 | * Despite our system space_info having enough free space, we may not |
| 3218 | * be able to allocate extents from its block groups, because all have |
| 3219 | * an incompatible profile, which will force us to allocate a new system |
| 3220 | * block group with the right profile, or right after we called |
| 3221 | * check_system_space() above, a scrub turned the only system block group |
| 3222 | * with enough free space into RO mode. |
| 3223 | * This is explained with more detail at do_chunk_alloc(). |
| 3224 | * |
| 3225 | * So if we get -ENOSPC, allocate a new system chunk and retry once. |
| 3226 | */ |
| 3227 | if (ret == -ENOSPC) { |
| 3228 | const u64 sys_flags = btrfs_system_alloc_profile(fs_info); |
| 3229 | struct btrfs_block_group *sys_bg; |
| 3230 | |
| 3231 | sys_bg = btrfs_create_chunk(trans, sys_flags); |
| 3232 | if (IS_ERR(sys_bg)) { |
| 3233 | ret = PTR_ERR(sys_bg); |
| 3234 | btrfs_abort_transaction(trans, ret); |
| 3235 | goto out; |
| 3236 | } |
| 3237 | |
| 3238 | ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg); |
| 3239 | if (ret) { |
| 3240 | btrfs_abort_transaction(trans, ret); |
| 3241 | goto out; |
| 3242 | } |
| 3243 | |
| 3244 | ret = remove_chunk_item(trans, map, chunk_offset); |
| 3245 | if (ret) { |
| 3246 | btrfs_abort_transaction(trans, ret); |
| 3247 | goto out; |
| 3248 | } |
| 3249 | } else if (ret) { |
| 3250 | btrfs_abort_transaction(trans, ret); |
| 3251 | goto out; |
| 3252 | } |
| 3253 | |
| 3254 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len); |
| 3255 | |
| 3256 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
| 3257 | ret = btrfs_del_sys_chunk(fs_info, chunk_offset); |
| 3258 | if (ret) { |
| 3259 | btrfs_abort_transaction(trans, ret); |
| 3260 | goto out; |
| 3261 | } |
| 3262 | } |
| 3263 | |
| 3264 | mutex_unlock(&fs_info->chunk_mutex); |
| 3265 | trans->removing_chunk = false; |
| 3266 | |
| 3267 | /* |
| 3268 | * We are done with chunk btree updates and deletions, so release the |
| 3269 | * system space we previously reserved (with check_system_chunk()). |
| 3270 | */ |
| 3271 | btrfs_trans_release_chunk_metadata(trans); |
| 3272 | |
| 3273 | ret = btrfs_remove_block_group(trans, map); |
| 3274 | if (ret) { |
| 3275 | btrfs_abort_transaction(trans, ret); |
| 3276 | goto out; |
| 3277 | } |
| 3278 | |
| 3279 | out: |
| 3280 | if (trans->removing_chunk) { |
| 3281 | mutex_unlock(&fs_info->chunk_mutex); |
| 3282 | trans->removing_chunk = false; |
| 3283 | } |
| 3284 | /* once for us */ |
| 3285 | btrfs_free_chunk_map(map); |
| 3286 | return ret; |
| 3287 | } |
| 3288 | |
| 3289 | int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
| 3290 | { |
| 3291 | struct btrfs_root *root = fs_info->chunk_root; |
| 3292 | struct btrfs_trans_handle *trans; |
| 3293 | struct btrfs_block_group *block_group; |
| 3294 | u64 length; |
| 3295 | int ret; |
| 3296 | |
| 3297 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
| 3298 | btrfs_err(fs_info, |
| 3299 | "relocate: not supported on extent tree v2 yet"); |
| 3300 | return -EINVAL; |
| 3301 | } |
| 3302 | |
| 3303 | /* |
| 3304 | * Prevent races with automatic removal of unused block groups. |
| 3305 | * After we relocate and before we remove the chunk with offset |
| 3306 | * chunk_offset, automatic removal of the block group can kick in, |
| 3307 | * resulting in a failure when calling btrfs_remove_chunk() below. |
| 3308 | * |
| 3309 | * Make sure to acquire this mutex before doing a tree search (dev |
| 3310 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might |
| 3311 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after |
| 3312 | * we release the path used to search the chunk/dev tree and before |
| 3313 | * the current task acquires this mutex and calls us. |
| 3314 | */ |
| 3315 | lockdep_assert_held(&fs_info->reclaim_bgs_lock); |
| 3316 | |
| 3317 | /* step one, relocate all the extents inside this chunk */ |
| 3318 | btrfs_scrub_pause(fs_info); |
| 3319 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
| 3320 | btrfs_scrub_continue(fs_info); |
| 3321 | if (ret) { |
| 3322 | /* |
| 3323 | * If we had a transaction abort, stop all running scrubs. |
| 3324 | * See transaction.c:cleanup_transaction() why we do it here. |
| 3325 | */ |
| 3326 | if (BTRFS_FS_ERROR(fs_info)) |
| 3327 | btrfs_scrub_cancel(fs_info); |
| 3328 | return ret; |
| 3329 | } |
| 3330 | |
| 3331 | block_group = btrfs_lookup_block_group(fs_info, chunk_offset); |
| 3332 | if (!block_group) |
| 3333 | return -ENOENT; |
| 3334 | btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group); |
| 3335 | length = block_group->length; |
| 3336 | btrfs_put_block_group(block_group); |
| 3337 | |
| 3338 | /* |
| 3339 | * On a zoned file system, discard the whole block group, this will |
| 3340 | * trigger a REQ_OP_ZONE_RESET operation on the device zone. If |
| 3341 | * resetting the zone fails, don't treat it as a fatal problem from the |
| 3342 | * filesystem's point of view. |
| 3343 | */ |
| 3344 | if (btrfs_is_zoned(fs_info)) { |
| 3345 | ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL); |
| 3346 | if (ret) |
| 3347 | btrfs_info(fs_info, |
| 3348 | "failed to reset zone %llu after relocation", |
| 3349 | chunk_offset); |
| 3350 | } |
| 3351 | |
| 3352 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
| 3353 | chunk_offset); |
| 3354 | if (IS_ERR(trans)) { |
| 3355 | ret = PTR_ERR(trans); |
| 3356 | btrfs_handle_fs_error(root->fs_info, ret, NULL); |
| 3357 | return ret; |
| 3358 | } |
| 3359 | |
| 3360 | /* |
| 3361 | * step two, delete the device extents and the |
| 3362 | * chunk tree entries |
| 3363 | */ |
| 3364 | ret = btrfs_remove_chunk(trans, chunk_offset); |
| 3365 | btrfs_end_transaction(trans); |
| 3366 | return ret; |
| 3367 | } |
| 3368 | |
| 3369 | static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info) |
| 3370 | { |
| 3371 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
| 3372 | struct btrfs_path *path; |
| 3373 | struct extent_buffer *leaf; |
| 3374 | struct btrfs_chunk *chunk; |
| 3375 | struct btrfs_key key; |
| 3376 | struct btrfs_key found_key; |
| 3377 | u64 chunk_type; |
| 3378 | bool retried = false; |
| 3379 | int failed = 0; |
| 3380 | int ret; |
| 3381 | |
| 3382 | path = btrfs_alloc_path(); |
| 3383 | if (!path) |
| 3384 | return -ENOMEM; |
| 3385 | |
| 3386 | again: |
| 3387 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
| 3388 | key.offset = (u64)-1; |
| 3389 | key.type = BTRFS_CHUNK_ITEM_KEY; |
| 3390 | |
| 3391 | while (1) { |
| 3392 | mutex_lock(&fs_info->reclaim_bgs_lock); |
| 3393 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
| 3394 | if (ret < 0) { |
| 3395 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3396 | goto error; |
| 3397 | } |
| 3398 | BUG_ON(ret == 0); /* Corruption */ |
| 3399 | |
| 3400 | ret = btrfs_previous_item(chunk_root, path, key.objectid, |
| 3401 | key.type); |
| 3402 | if (ret) |
| 3403 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3404 | if (ret < 0) |
| 3405 | goto error; |
| 3406 | if (ret > 0) |
| 3407 | break; |
| 3408 | |
| 3409 | leaf = path->nodes[0]; |
| 3410 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 3411 | |
| 3412 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
| 3413 | struct btrfs_chunk); |
| 3414 | chunk_type = btrfs_chunk_type(leaf, chunk); |
| 3415 | btrfs_release_path(path); |
| 3416 | |
| 3417 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
| 3418 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
| 3419 | if (ret == -ENOSPC) |
| 3420 | failed++; |
| 3421 | else |
| 3422 | BUG_ON(ret); |
| 3423 | } |
| 3424 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3425 | |
| 3426 | if (found_key.offset == 0) |
| 3427 | break; |
| 3428 | key.offset = found_key.offset - 1; |
| 3429 | } |
| 3430 | ret = 0; |
| 3431 | if (failed && !retried) { |
| 3432 | failed = 0; |
| 3433 | retried = true; |
| 3434 | goto again; |
| 3435 | } else if (WARN_ON(failed && retried)) { |
| 3436 | ret = -ENOSPC; |
| 3437 | } |
| 3438 | error: |
| 3439 | btrfs_free_path(path); |
| 3440 | return ret; |
| 3441 | } |
| 3442 | |
| 3443 | /* |
| 3444 | * return 1 : allocate a data chunk successfully, |
| 3445 | * return <0: errors during allocating a data chunk, |
| 3446 | * return 0 : no need to allocate a data chunk. |
| 3447 | */ |
| 3448 | static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info, |
| 3449 | u64 chunk_offset) |
| 3450 | { |
| 3451 | struct btrfs_block_group *cache; |
| 3452 | u64 bytes_used; |
| 3453 | u64 chunk_type; |
| 3454 | |
| 3455 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); |
| 3456 | ASSERT(cache); |
| 3457 | chunk_type = cache->flags; |
| 3458 | btrfs_put_block_group(cache); |
| 3459 | |
| 3460 | if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA)) |
| 3461 | return 0; |
| 3462 | |
| 3463 | spin_lock(&fs_info->data_sinfo->lock); |
| 3464 | bytes_used = fs_info->data_sinfo->bytes_used; |
| 3465 | spin_unlock(&fs_info->data_sinfo->lock); |
| 3466 | |
| 3467 | if (!bytes_used) { |
| 3468 | struct btrfs_trans_handle *trans; |
| 3469 | int ret; |
| 3470 | |
| 3471 | trans = btrfs_join_transaction(fs_info->tree_root); |
| 3472 | if (IS_ERR(trans)) |
| 3473 | return PTR_ERR(trans); |
| 3474 | |
| 3475 | ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA); |
| 3476 | btrfs_end_transaction(trans); |
| 3477 | if (ret < 0) |
| 3478 | return ret; |
| 3479 | return 1; |
| 3480 | } |
| 3481 | |
| 3482 | return 0; |
| 3483 | } |
| 3484 | |
| 3485 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
| 3486 | struct btrfs_balance_control *bctl) |
| 3487 | { |
| 3488 | struct btrfs_root *root = fs_info->tree_root; |
| 3489 | struct btrfs_trans_handle *trans; |
| 3490 | struct btrfs_balance_item *item; |
| 3491 | struct btrfs_disk_balance_args disk_bargs; |
| 3492 | struct btrfs_path *path; |
| 3493 | struct extent_buffer *leaf; |
| 3494 | struct btrfs_key key; |
| 3495 | int ret, err; |
| 3496 | |
| 3497 | path = btrfs_alloc_path(); |
| 3498 | if (!path) |
| 3499 | return -ENOMEM; |
| 3500 | |
| 3501 | trans = btrfs_start_transaction(root, 0); |
| 3502 | if (IS_ERR(trans)) { |
| 3503 | btrfs_free_path(path); |
| 3504 | return PTR_ERR(trans); |
| 3505 | } |
| 3506 | |
| 3507 | key.objectid = BTRFS_BALANCE_OBJECTID; |
| 3508 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
| 3509 | key.offset = 0; |
| 3510 | |
| 3511 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
| 3512 | sizeof(*item)); |
| 3513 | if (ret) |
| 3514 | goto out; |
| 3515 | |
| 3516 | leaf = path->nodes[0]; |
| 3517 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); |
| 3518 | |
| 3519 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
| 3520 | |
| 3521 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); |
| 3522 | btrfs_set_balance_data(leaf, item, &disk_bargs); |
| 3523 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); |
| 3524 | btrfs_set_balance_meta(leaf, item, &disk_bargs); |
| 3525 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); |
| 3526 | btrfs_set_balance_sys(leaf, item, &disk_bargs); |
| 3527 | |
| 3528 | btrfs_set_balance_flags(leaf, item, bctl->flags); |
| 3529 | |
| 3530 | btrfs_mark_buffer_dirty(trans, leaf); |
| 3531 | out: |
| 3532 | btrfs_free_path(path); |
| 3533 | err = btrfs_commit_transaction(trans); |
| 3534 | if (err && !ret) |
| 3535 | ret = err; |
| 3536 | return ret; |
| 3537 | } |
| 3538 | |
| 3539 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
| 3540 | { |
| 3541 | struct btrfs_root *root = fs_info->tree_root; |
| 3542 | struct btrfs_trans_handle *trans; |
| 3543 | struct btrfs_path *path; |
| 3544 | struct btrfs_key key; |
| 3545 | int ret, err; |
| 3546 | |
| 3547 | path = btrfs_alloc_path(); |
| 3548 | if (!path) |
| 3549 | return -ENOMEM; |
| 3550 | |
| 3551 | trans = btrfs_start_transaction_fallback_global_rsv(root, 0); |
| 3552 | if (IS_ERR(trans)) { |
| 3553 | btrfs_free_path(path); |
| 3554 | return PTR_ERR(trans); |
| 3555 | } |
| 3556 | |
| 3557 | key.objectid = BTRFS_BALANCE_OBJECTID; |
| 3558 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
| 3559 | key.offset = 0; |
| 3560 | |
| 3561 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 3562 | if (ret < 0) |
| 3563 | goto out; |
| 3564 | if (ret > 0) { |
| 3565 | ret = -ENOENT; |
| 3566 | goto out; |
| 3567 | } |
| 3568 | |
| 3569 | ret = btrfs_del_item(trans, root, path); |
| 3570 | out: |
| 3571 | btrfs_free_path(path); |
| 3572 | err = btrfs_commit_transaction(trans); |
| 3573 | if (err && !ret) |
| 3574 | ret = err; |
| 3575 | return ret; |
| 3576 | } |
| 3577 | |
| 3578 | /* |
| 3579 | * This is a heuristic used to reduce the number of chunks balanced on |
| 3580 | * resume after balance was interrupted. |
| 3581 | */ |
| 3582 | static void update_balance_args(struct btrfs_balance_control *bctl) |
| 3583 | { |
| 3584 | /* |
| 3585 | * Turn on soft mode for chunk types that were being converted. |
| 3586 | */ |
| 3587 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) |
| 3588 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; |
| 3589 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) |
| 3590 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; |
| 3591 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) |
| 3592 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; |
| 3593 | |
| 3594 | /* |
| 3595 | * Turn on usage filter if is not already used. The idea is |
| 3596 | * that chunks that we have already balanced should be |
| 3597 | * reasonably full. Don't do it for chunks that are being |
| 3598 | * converted - that will keep us from relocating unconverted |
| 3599 | * (albeit full) chunks. |
| 3600 | */ |
| 3601 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && |
| 3602 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
| 3603 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
| 3604 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; |
| 3605 | bctl->data.usage = 90; |
| 3606 | } |
| 3607 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && |
| 3608 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
| 3609 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
| 3610 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; |
| 3611 | bctl->sys.usage = 90; |
| 3612 | } |
| 3613 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && |
| 3614 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
| 3615 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
| 3616 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; |
| 3617 | bctl->meta.usage = 90; |
| 3618 | } |
| 3619 | } |
| 3620 | |
| 3621 | /* |
| 3622 | * Clear the balance status in fs_info and delete the balance item from disk. |
| 3623 | */ |
| 3624 | static void reset_balance_state(struct btrfs_fs_info *fs_info) |
| 3625 | { |
| 3626 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 3627 | int ret; |
| 3628 | |
| 3629 | BUG_ON(!fs_info->balance_ctl); |
| 3630 | |
| 3631 | spin_lock(&fs_info->balance_lock); |
| 3632 | fs_info->balance_ctl = NULL; |
| 3633 | spin_unlock(&fs_info->balance_lock); |
| 3634 | |
| 3635 | kfree(bctl); |
| 3636 | ret = del_balance_item(fs_info); |
| 3637 | if (ret) |
| 3638 | btrfs_handle_fs_error(fs_info, ret, NULL); |
| 3639 | } |
| 3640 | |
| 3641 | /* |
| 3642 | * Balance filters. Return 1 if chunk should be filtered out |
| 3643 | * (should not be balanced). |
| 3644 | */ |
| 3645 | static int chunk_profiles_filter(u64 chunk_type, |
| 3646 | struct btrfs_balance_args *bargs) |
| 3647 | { |
| 3648 | chunk_type = chunk_to_extended(chunk_type) & |
| 3649 | BTRFS_EXTENDED_PROFILE_MASK; |
| 3650 | |
| 3651 | if (bargs->profiles & chunk_type) |
| 3652 | return 0; |
| 3653 | |
| 3654 | return 1; |
| 3655 | } |
| 3656 | |
| 3657 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
| 3658 | struct btrfs_balance_args *bargs) |
| 3659 | { |
| 3660 | struct btrfs_block_group *cache; |
| 3661 | u64 chunk_used; |
| 3662 | u64 user_thresh_min; |
| 3663 | u64 user_thresh_max; |
| 3664 | int ret = 1; |
| 3665 | |
| 3666 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); |
| 3667 | chunk_used = cache->used; |
| 3668 | |
| 3669 | if (bargs->usage_min == 0) |
| 3670 | user_thresh_min = 0; |
| 3671 | else |
| 3672 | user_thresh_min = mult_perc(cache->length, bargs->usage_min); |
| 3673 | |
| 3674 | if (bargs->usage_max == 0) |
| 3675 | user_thresh_max = 1; |
| 3676 | else if (bargs->usage_max > 100) |
| 3677 | user_thresh_max = cache->length; |
| 3678 | else |
| 3679 | user_thresh_max = mult_perc(cache->length, bargs->usage_max); |
| 3680 | |
| 3681 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) |
| 3682 | ret = 0; |
| 3683 | |
| 3684 | btrfs_put_block_group(cache); |
| 3685 | return ret; |
| 3686 | } |
| 3687 | |
| 3688 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
| 3689 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
| 3690 | { |
| 3691 | struct btrfs_block_group *cache; |
| 3692 | u64 chunk_used, user_thresh; |
| 3693 | int ret = 1; |
| 3694 | |
| 3695 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); |
| 3696 | chunk_used = cache->used; |
| 3697 | |
| 3698 | if (bargs->usage_min == 0) |
| 3699 | user_thresh = 1; |
| 3700 | else if (bargs->usage > 100) |
| 3701 | user_thresh = cache->length; |
| 3702 | else |
| 3703 | user_thresh = mult_perc(cache->length, bargs->usage); |
| 3704 | |
| 3705 | if (chunk_used < user_thresh) |
| 3706 | ret = 0; |
| 3707 | |
| 3708 | btrfs_put_block_group(cache); |
| 3709 | return ret; |
| 3710 | } |
| 3711 | |
| 3712 | static int chunk_devid_filter(struct extent_buffer *leaf, |
| 3713 | struct btrfs_chunk *chunk, |
| 3714 | struct btrfs_balance_args *bargs) |
| 3715 | { |
| 3716 | struct btrfs_stripe *stripe; |
| 3717 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
| 3718 | int i; |
| 3719 | |
| 3720 | for (i = 0; i < num_stripes; i++) { |
| 3721 | stripe = btrfs_stripe_nr(chunk, i); |
| 3722 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) |
| 3723 | return 0; |
| 3724 | } |
| 3725 | |
| 3726 | return 1; |
| 3727 | } |
| 3728 | |
| 3729 | static u64 calc_data_stripes(u64 type, int num_stripes) |
| 3730 | { |
| 3731 | const int index = btrfs_bg_flags_to_raid_index(type); |
| 3732 | const int ncopies = btrfs_raid_array[index].ncopies; |
| 3733 | const int nparity = btrfs_raid_array[index].nparity; |
| 3734 | |
| 3735 | return (num_stripes - nparity) / ncopies; |
| 3736 | } |
| 3737 | |
| 3738 | /* [pstart, pend) */ |
| 3739 | static int chunk_drange_filter(struct extent_buffer *leaf, |
| 3740 | struct btrfs_chunk *chunk, |
| 3741 | struct btrfs_balance_args *bargs) |
| 3742 | { |
| 3743 | struct btrfs_stripe *stripe; |
| 3744 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
| 3745 | u64 stripe_offset; |
| 3746 | u64 stripe_length; |
| 3747 | u64 type; |
| 3748 | int factor; |
| 3749 | int i; |
| 3750 | |
| 3751 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) |
| 3752 | return 0; |
| 3753 | |
| 3754 | type = btrfs_chunk_type(leaf, chunk); |
| 3755 | factor = calc_data_stripes(type, num_stripes); |
| 3756 | |
| 3757 | for (i = 0; i < num_stripes; i++) { |
| 3758 | stripe = btrfs_stripe_nr(chunk, i); |
| 3759 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) |
| 3760 | continue; |
| 3761 | |
| 3762 | stripe_offset = btrfs_stripe_offset(leaf, stripe); |
| 3763 | stripe_length = btrfs_chunk_length(leaf, chunk); |
| 3764 | stripe_length = div_u64(stripe_length, factor); |
| 3765 | |
| 3766 | if (stripe_offset < bargs->pend && |
| 3767 | stripe_offset + stripe_length > bargs->pstart) |
| 3768 | return 0; |
| 3769 | } |
| 3770 | |
| 3771 | return 1; |
| 3772 | } |
| 3773 | |
| 3774 | /* [vstart, vend) */ |
| 3775 | static int chunk_vrange_filter(struct extent_buffer *leaf, |
| 3776 | struct btrfs_chunk *chunk, |
| 3777 | u64 chunk_offset, |
| 3778 | struct btrfs_balance_args *bargs) |
| 3779 | { |
| 3780 | if (chunk_offset < bargs->vend && |
| 3781 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) |
| 3782 | /* at least part of the chunk is inside this vrange */ |
| 3783 | return 0; |
| 3784 | |
| 3785 | return 1; |
| 3786 | } |
| 3787 | |
| 3788 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
| 3789 | struct btrfs_chunk *chunk, |
| 3790 | struct btrfs_balance_args *bargs) |
| 3791 | { |
| 3792 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
| 3793 | |
| 3794 | if (bargs->stripes_min <= num_stripes |
| 3795 | && num_stripes <= bargs->stripes_max) |
| 3796 | return 0; |
| 3797 | |
| 3798 | return 1; |
| 3799 | } |
| 3800 | |
| 3801 | static int chunk_soft_convert_filter(u64 chunk_type, |
| 3802 | struct btrfs_balance_args *bargs) |
| 3803 | { |
| 3804 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
| 3805 | return 0; |
| 3806 | |
| 3807 | chunk_type = chunk_to_extended(chunk_type) & |
| 3808 | BTRFS_EXTENDED_PROFILE_MASK; |
| 3809 | |
| 3810 | if (bargs->target == chunk_type) |
| 3811 | return 1; |
| 3812 | |
| 3813 | return 0; |
| 3814 | } |
| 3815 | |
| 3816 | static int should_balance_chunk(struct extent_buffer *leaf, |
| 3817 | struct btrfs_chunk *chunk, u64 chunk_offset) |
| 3818 | { |
| 3819 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
| 3820 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 3821 | struct btrfs_balance_args *bargs = NULL; |
| 3822 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); |
| 3823 | |
| 3824 | /* type filter */ |
| 3825 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & |
| 3826 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { |
| 3827 | return 0; |
| 3828 | } |
| 3829 | |
| 3830 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) |
| 3831 | bargs = &bctl->data; |
| 3832 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) |
| 3833 | bargs = &bctl->sys; |
| 3834 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) |
| 3835 | bargs = &bctl->meta; |
| 3836 | |
| 3837 | /* profiles filter */ |
| 3838 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && |
| 3839 | chunk_profiles_filter(chunk_type, bargs)) { |
| 3840 | return 0; |
| 3841 | } |
| 3842 | |
| 3843 | /* usage filter */ |
| 3844 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && |
| 3845 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
| 3846 | return 0; |
| 3847 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
| 3848 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
| 3849 | return 0; |
| 3850 | } |
| 3851 | |
| 3852 | /* devid filter */ |
| 3853 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && |
| 3854 | chunk_devid_filter(leaf, chunk, bargs)) { |
| 3855 | return 0; |
| 3856 | } |
| 3857 | |
| 3858 | /* drange filter, makes sense only with devid filter */ |
| 3859 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && |
| 3860 | chunk_drange_filter(leaf, chunk, bargs)) { |
| 3861 | return 0; |
| 3862 | } |
| 3863 | |
| 3864 | /* vrange filter */ |
| 3865 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && |
| 3866 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { |
| 3867 | return 0; |
| 3868 | } |
| 3869 | |
| 3870 | /* stripes filter */ |
| 3871 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && |
| 3872 | chunk_stripes_range_filter(leaf, chunk, bargs)) { |
| 3873 | return 0; |
| 3874 | } |
| 3875 | |
| 3876 | /* soft profile changing mode */ |
| 3877 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && |
| 3878 | chunk_soft_convert_filter(chunk_type, bargs)) { |
| 3879 | return 0; |
| 3880 | } |
| 3881 | |
| 3882 | /* |
| 3883 | * limited by count, must be the last filter |
| 3884 | */ |
| 3885 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { |
| 3886 | if (bargs->limit == 0) |
| 3887 | return 0; |
| 3888 | else |
| 3889 | bargs->limit--; |
| 3890 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
| 3891 | /* |
| 3892 | * Same logic as the 'limit' filter; the minimum cannot be |
| 3893 | * determined here because we do not have the global information |
| 3894 | * about the count of all chunks that satisfy the filters. |
| 3895 | */ |
| 3896 | if (bargs->limit_max == 0) |
| 3897 | return 0; |
| 3898 | else |
| 3899 | bargs->limit_max--; |
| 3900 | } |
| 3901 | |
| 3902 | return 1; |
| 3903 | } |
| 3904 | |
| 3905 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
| 3906 | { |
| 3907 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 3908 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
| 3909 | u64 chunk_type; |
| 3910 | struct btrfs_chunk *chunk; |
| 3911 | struct btrfs_path *path = NULL; |
| 3912 | struct btrfs_key key; |
| 3913 | struct btrfs_key found_key; |
| 3914 | struct extent_buffer *leaf; |
| 3915 | int slot; |
| 3916 | int ret; |
| 3917 | int enospc_errors = 0; |
| 3918 | bool counting = true; |
| 3919 | /* The single value limit and min/max limits use the same bytes in the */ |
| 3920 | u64 limit_data = bctl->data.limit; |
| 3921 | u64 limit_meta = bctl->meta.limit; |
| 3922 | u64 limit_sys = bctl->sys.limit; |
| 3923 | u32 count_data = 0; |
| 3924 | u32 count_meta = 0; |
| 3925 | u32 count_sys = 0; |
| 3926 | int chunk_reserved = 0; |
| 3927 | |
| 3928 | path = btrfs_alloc_path(); |
| 3929 | if (!path) { |
| 3930 | ret = -ENOMEM; |
| 3931 | goto error; |
| 3932 | } |
| 3933 | |
| 3934 | /* zero out stat counters */ |
| 3935 | spin_lock(&fs_info->balance_lock); |
| 3936 | memset(&bctl->stat, 0, sizeof(bctl->stat)); |
| 3937 | spin_unlock(&fs_info->balance_lock); |
| 3938 | again: |
| 3939 | if (!counting) { |
| 3940 | /* |
| 3941 | * The single value limit and min/max limits use the same bytes |
| 3942 | * in the |
| 3943 | */ |
| 3944 | bctl->data.limit = limit_data; |
| 3945 | bctl->meta.limit = limit_meta; |
| 3946 | bctl->sys.limit = limit_sys; |
| 3947 | } |
| 3948 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
| 3949 | key.offset = (u64)-1; |
| 3950 | key.type = BTRFS_CHUNK_ITEM_KEY; |
| 3951 | |
| 3952 | while (1) { |
| 3953 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
| 3954 | atomic_read(&fs_info->balance_cancel_req)) { |
| 3955 | ret = -ECANCELED; |
| 3956 | goto error; |
| 3957 | } |
| 3958 | |
| 3959 | mutex_lock(&fs_info->reclaim_bgs_lock); |
| 3960 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
| 3961 | if (ret < 0) { |
| 3962 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3963 | goto error; |
| 3964 | } |
| 3965 | |
| 3966 | /* |
| 3967 | * this shouldn't happen, it means the last relocate |
| 3968 | * failed |
| 3969 | */ |
| 3970 | if (ret == 0) |
| 3971 | BUG(); /* FIXME break ? */ |
| 3972 | |
| 3973 | ret = btrfs_previous_item(chunk_root, path, 0, |
| 3974 | BTRFS_CHUNK_ITEM_KEY); |
| 3975 | if (ret) { |
| 3976 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3977 | ret = 0; |
| 3978 | break; |
| 3979 | } |
| 3980 | |
| 3981 | leaf = path->nodes[0]; |
| 3982 | slot = path->slots[0]; |
| 3983 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
| 3984 | |
| 3985 | if (found_key.objectid != key.objectid) { |
| 3986 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 3987 | break; |
| 3988 | } |
| 3989 | |
| 3990 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
| 3991 | chunk_type = btrfs_chunk_type(leaf, chunk); |
| 3992 | |
| 3993 | if (!counting) { |
| 3994 | spin_lock(&fs_info->balance_lock); |
| 3995 | bctl->stat.considered++; |
| 3996 | spin_unlock(&fs_info->balance_lock); |
| 3997 | } |
| 3998 | |
| 3999 | ret = should_balance_chunk(leaf, chunk, found_key.offset); |
| 4000 | |
| 4001 | btrfs_release_path(path); |
| 4002 | if (!ret) { |
| 4003 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4004 | goto loop; |
| 4005 | } |
| 4006 | |
| 4007 | if (counting) { |
| 4008 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4009 | spin_lock(&fs_info->balance_lock); |
| 4010 | bctl->stat.expected++; |
| 4011 | spin_unlock(&fs_info->balance_lock); |
| 4012 | |
| 4013 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) |
| 4014 | count_data++; |
| 4015 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) |
| 4016 | count_sys++; |
| 4017 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) |
| 4018 | count_meta++; |
| 4019 | |
| 4020 | goto loop; |
| 4021 | } |
| 4022 | |
| 4023 | /* |
| 4024 | * Apply limit_min filter, no need to check if the LIMITS |
| 4025 | * filter is used, limit_min is 0 by default |
| 4026 | */ |
| 4027 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && |
| 4028 | count_data < bctl->data.limit_min) |
| 4029 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && |
| 4030 | count_meta < bctl->meta.limit_min) |
| 4031 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && |
| 4032 | count_sys < bctl->sys.limit_min)) { |
| 4033 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4034 | goto loop; |
| 4035 | } |
| 4036 | |
| 4037 | if (!chunk_reserved) { |
| 4038 | /* |
| 4039 | * We may be relocating the only data chunk we have, |
| 4040 | * which could potentially end up with losing data's |
| 4041 | * raid profile, so lets allocate an empty one in |
| 4042 | * advance. |
| 4043 | */ |
| 4044 | ret = btrfs_may_alloc_data_chunk(fs_info, |
| 4045 | found_key.offset); |
| 4046 | if (ret < 0) { |
| 4047 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4048 | goto error; |
| 4049 | } else if (ret == 1) { |
| 4050 | chunk_reserved = 1; |
| 4051 | } |
| 4052 | } |
| 4053 | |
| 4054 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
| 4055 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4056 | if (ret == -ENOSPC) { |
| 4057 | enospc_errors++; |
| 4058 | } else if (ret == -ETXTBSY) { |
| 4059 | btrfs_info(fs_info, |
| 4060 | "skipping relocation of block group %llu due to active swapfile", |
| 4061 | found_key.offset); |
| 4062 | ret = 0; |
| 4063 | } else if (ret) { |
| 4064 | goto error; |
| 4065 | } else { |
| 4066 | spin_lock(&fs_info->balance_lock); |
| 4067 | bctl->stat.completed++; |
| 4068 | spin_unlock(&fs_info->balance_lock); |
| 4069 | } |
| 4070 | loop: |
| 4071 | if (found_key.offset == 0) |
| 4072 | break; |
| 4073 | key.offset = found_key.offset - 1; |
| 4074 | } |
| 4075 | |
| 4076 | if (counting) { |
| 4077 | btrfs_release_path(path); |
| 4078 | counting = false; |
| 4079 | goto again; |
| 4080 | } |
| 4081 | error: |
| 4082 | btrfs_free_path(path); |
| 4083 | if (enospc_errors) { |
| 4084 | btrfs_info(fs_info, "%d enospc errors during balance", |
| 4085 | enospc_errors); |
| 4086 | if (!ret) |
| 4087 | ret = -ENOSPC; |
| 4088 | } |
| 4089 | |
| 4090 | return ret; |
| 4091 | } |
| 4092 | |
| 4093 | /* |
| 4094 | * See if a given profile is valid and reduced. |
| 4095 | * |
| 4096 | * @flags: profile to validate |
| 4097 | * @extended: if true @flags is treated as an extended profile |
| 4098 | */ |
| 4099 | static int alloc_profile_is_valid(u64 flags, int extended) |
| 4100 | { |
| 4101 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : |
| 4102 | BTRFS_BLOCK_GROUP_PROFILE_MASK); |
| 4103 | |
| 4104 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; |
| 4105 | |
| 4106 | /* 1) check that all other bits are zeroed */ |
| 4107 | if (flags & ~mask) |
| 4108 | return 0; |
| 4109 | |
| 4110 | /* 2) see if profile is reduced */ |
| 4111 | if (flags == 0) |
| 4112 | return !extended; /* "0" is valid for usual profiles */ |
| 4113 | |
| 4114 | return has_single_bit_set(flags); |
| 4115 | } |
| 4116 | |
| 4117 | /* |
| 4118 | * Validate target profile against allowed profiles and return true if it's OK. |
| 4119 | * Otherwise print the error message and return false. |
| 4120 | */ |
| 4121 | static inline int validate_convert_profile(struct btrfs_fs_info *fs_info, |
| 4122 | const struct btrfs_balance_args *bargs, |
| 4123 | u64 allowed, const char *type) |
| 4124 | { |
| 4125 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) |
| 4126 | return true; |
| 4127 | |
| 4128 | /* Profile is valid and does not have bits outside of the allowed set */ |
| 4129 | if (alloc_profile_is_valid(bargs->target, 1) && |
| 4130 | (bargs->target & ~allowed) == 0) |
| 4131 | return true; |
| 4132 | |
| 4133 | btrfs_err(fs_info, "balance: invalid convert %s profile %s", |
| 4134 | type, btrfs_bg_type_to_raid_name(bargs->target)); |
| 4135 | return false; |
| 4136 | } |
| 4137 | |
| 4138 | /* |
| 4139 | * Fill @buf with textual description of balance filter flags @bargs, up to |
| 4140 | * @size_buf including the terminating null. The output may be trimmed if it |
| 4141 | * does not fit into the provided buffer. |
| 4142 | */ |
| 4143 | static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf, |
| 4144 | u32 size_buf) |
| 4145 | { |
| 4146 | int ret; |
| 4147 | u32 size_bp = size_buf; |
| 4148 | char *bp = buf; |
| 4149 | u64 flags = bargs->flags; |
| 4150 | char tmp_buf[128] = {'\0'}; |
| 4151 | |
| 4152 | if (!flags) |
| 4153 | return; |
| 4154 | |
| 4155 | #define CHECK_APPEND_NOARG(a) \ |
| 4156 | do { \ |
| 4157 | ret = snprintf(bp, size_bp, (a)); \ |
| 4158 | if (ret < 0 || ret >= size_bp) \ |
| 4159 | goto out_overflow; \ |
| 4160 | size_bp -= ret; \ |
| 4161 | bp += ret; \ |
| 4162 | } while (0) |
| 4163 | |
| 4164 | #define CHECK_APPEND_1ARG(a, v1) \ |
| 4165 | do { \ |
| 4166 | ret = snprintf(bp, size_bp, (a), (v1)); \ |
| 4167 | if (ret < 0 || ret >= size_bp) \ |
| 4168 | goto out_overflow; \ |
| 4169 | size_bp -= ret; \ |
| 4170 | bp += ret; \ |
| 4171 | } while (0) |
| 4172 | |
| 4173 | #define CHECK_APPEND_2ARG(a, v1, v2) \ |
| 4174 | do { \ |
| 4175 | ret = snprintf(bp, size_bp, (a), (v1), (v2)); \ |
| 4176 | if (ret < 0 || ret >= size_bp) \ |
| 4177 | goto out_overflow; \ |
| 4178 | size_bp -= ret; \ |
| 4179 | bp += ret; \ |
| 4180 | } while (0) |
| 4181 | |
| 4182 | if (flags & BTRFS_BALANCE_ARGS_CONVERT) |
| 4183 | CHECK_APPEND_1ARG("convert=%s,", |
| 4184 | btrfs_bg_type_to_raid_name(bargs->target)); |
| 4185 | |
| 4186 | if (flags & BTRFS_BALANCE_ARGS_SOFT) |
| 4187 | CHECK_APPEND_NOARG("soft,"); |
| 4188 | |
| 4189 | if (flags & BTRFS_BALANCE_ARGS_PROFILES) { |
| 4190 | btrfs_describe_block_groups(bargs->profiles, tmp_buf, |
| 4191 | sizeof(tmp_buf)); |
| 4192 | CHECK_APPEND_1ARG("profiles=%s,", tmp_buf); |
| 4193 | } |
| 4194 | |
| 4195 | if (flags & BTRFS_BALANCE_ARGS_USAGE) |
| 4196 | CHECK_APPEND_1ARG("usage=%llu,", bargs->usage); |
| 4197 | |
| 4198 | if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) |
| 4199 | CHECK_APPEND_2ARG("usage=%u..%u,", |
| 4200 | bargs->usage_min, bargs->usage_max); |
| 4201 | |
| 4202 | if (flags & BTRFS_BALANCE_ARGS_DEVID) |
| 4203 | CHECK_APPEND_1ARG("devid=%llu,", bargs->devid); |
| 4204 | |
| 4205 | if (flags & BTRFS_BALANCE_ARGS_DRANGE) |
| 4206 | CHECK_APPEND_2ARG("drange=%llu..%llu,", |
| 4207 | bargs->pstart, bargs->pend); |
| 4208 | |
| 4209 | if (flags & BTRFS_BALANCE_ARGS_VRANGE) |
| 4210 | CHECK_APPEND_2ARG("vrange=%llu..%llu,", |
| 4211 | bargs->vstart, bargs->vend); |
| 4212 | |
| 4213 | if (flags & BTRFS_BALANCE_ARGS_LIMIT) |
| 4214 | CHECK_APPEND_1ARG("limit=%llu,", bargs->limit); |
| 4215 | |
| 4216 | if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE) |
| 4217 | CHECK_APPEND_2ARG("limit=%u..%u,", |
| 4218 | bargs->limit_min, bargs->limit_max); |
| 4219 | |
| 4220 | if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) |
| 4221 | CHECK_APPEND_2ARG("stripes=%u..%u,", |
| 4222 | bargs->stripes_min, bargs->stripes_max); |
| 4223 | |
| 4224 | #undef CHECK_APPEND_2ARG |
| 4225 | #undef CHECK_APPEND_1ARG |
| 4226 | #undef CHECK_APPEND_NOARG |
| 4227 | |
| 4228 | out_overflow: |
| 4229 | |
| 4230 | if (size_bp < size_buf) |
| 4231 | buf[size_buf - size_bp - 1] = '\0'; /* remove last , */ |
| 4232 | else |
| 4233 | buf[0] = '\0'; |
| 4234 | } |
| 4235 | |
| 4236 | static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info) |
| 4237 | { |
| 4238 | u32 size_buf = 1024; |
| 4239 | char tmp_buf[192] = {'\0'}; |
| 4240 | char *buf; |
| 4241 | char *bp; |
| 4242 | u32 size_bp = size_buf; |
| 4243 | int ret; |
| 4244 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 4245 | |
| 4246 | buf = kzalloc(size_buf, GFP_KERNEL); |
| 4247 | if (!buf) |
| 4248 | return; |
| 4249 | |
| 4250 | bp = buf; |
| 4251 | |
| 4252 | #define CHECK_APPEND_1ARG(a, v1) \ |
| 4253 | do { \ |
| 4254 | ret = snprintf(bp, size_bp, (a), (v1)); \ |
| 4255 | if (ret < 0 || ret >= size_bp) \ |
| 4256 | goto out_overflow; \ |
| 4257 | size_bp -= ret; \ |
| 4258 | bp += ret; \ |
| 4259 | } while (0) |
| 4260 | |
| 4261 | if (bctl->flags & BTRFS_BALANCE_FORCE) |
| 4262 | CHECK_APPEND_1ARG("%s", "-f "); |
| 4263 | |
| 4264 | if (bctl->flags & BTRFS_BALANCE_DATA) { |
| 4265 | describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf)); |
| 4266 | CHECK_APPEND_1ARG("-d%s ", tmp_buf); |
| 4267 | } |
| 4268 | |
| 4269 | if (bctl->flags & BTRFS_BALANCE_METADATA) { |
| 4270 | describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf)); |
| 4271 | CHECK_APPEND_1ARG("-m%s ", tmp_buf); |
| 4272 | } |
| 4273 | |
| 4274 | if (bctl->flags & BTRFS_BALANCE_SYSTEM) { |
| 4275 | describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf)); |
| 4276 | CHECK_APPEND_1ARG("-s%s ", tmp_buf); |
| 4277 | } |
| 4278 | |
| 4279 | #undef CHECK_APPEND_1ARG |
| 4280 | |
| 4281 | out_overflow: |
| 4282 | |
| 4283 | if (size_bp < size_buf) |
| 4284 | buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */ |
| 4285 | btrfs_info(fs_info, "balance: %s %s", |
| 4286 | (bctl->flags & BTRFS_BALANCE_RESUME) ? |
| 4287 | "resume" : "start", buf); |
| 4288 | |
| 4289 | kfree(buf); |
| 4290 | } |
| 4291 | |
| 4292 | /* |
| 4293 | * Should be called with balance mutexe held |
| 4294 | */ |
| 4295 | int btrfs_balance(struct btrfs_fs_info *fs_info, |
| 4296 | struct btrfs_balance_control *bctl, |
| 4297 | struct btrfs_ioctl_balance_args *bargs) |
| 4298 | { |
| 4299 | u64 meta_target, data_target; |
| 4300 | u64 allowed; |
| 4301 | int mixed = 0; |
| 4302 | int ret; |
| 4303 | u64 num_devices; |
| 4304 | unsigned seq; |
| 4305 | bool reducing_redundancy; |
| 4306 | bool paused = false; |
| 4307 | int i; |
| 4308 | |
| 4309 | if (btrfs_fs_closing(fs_info) || |
| 4310 | atomic_read(&fs_info->balance_pause_req) || |
| 4311 | btrfs_should_cancel_balance(fs_info)) { |
| 4312 | ret = -EINVAL; |
| 4313 | goto out; |
| 4314 | } |
| 4315 | |
| 4316 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
| 4317 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) |
| 4318 | mixed = 1; |
| 4319 | |
| 4320 | /* |
| 4321 | * In case of mixed groups both data and meta should be picked, |
| 4322 | * and identical options should be given for both of them. |
| 4323 | */ |
| 4324 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
| 4325 | if (mixed && (bctl->flags & allowed)) { |
| 4326 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
| 4327 | !(bctl->flags & BTRFS_BALANCE_METADATA) || |
| 4328 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { |
| 4329 | btrfs_err(fs_info, |
| 4330 | "balance: mixed groups data and metadata options must be the same"); |
| 4331 | ret = -EINVAL; |
| 4332 | goto out; |
| 4333 | } |
| 4334 | } |
| 4335 | |
| 4336 | /* |
| 4337 | * rw_devices will not change at the moment, device add/delete/replace |
| 4338 | * are exclusive |
| 4339 | */ |
| 4340 | num_devices = fs_info->fs_devices->rw_devices; |
| 4341 | |
| 4342 | /* |
| 4343 | * SINGLE profile on-disk has no profile bit, but in-memory we have a |
| 4344 | * special bit for it, to make it easier to distinguish. Thus we need |
| 4345 | * to set it manually, or balance would refuse the profile. |
| 4346 | */ |
| 4347 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
| 4348 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) |
| 4349 | if (num_devices >= btrfs_raid_array[i].devs_min) |
| 4350 | allowed |= btrfs_raid_array[i].bg_flag; |
| 4351 | |
| 4352 | if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") || |
| 4353 | !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") || |
| 4354 | !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) { |
| 4355 | ret = -EINVAL; |
| 4356 | goto out; |
| 4357 | } |
| 4358 | |
| 4359 | /* |
| 4360 | * Allow to reduce metadata or system integrity only if force set for |
| 4361 | * profiles with redundancy (copies, parity) |
| 4362 | */ |
| 4363 | allowed = 0; |
| 4364 | for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) { |
| 4365 | if (btrfs_raid_array[i].ncopies >= 2 || |
| 4366 | btrfs_raid_array[i].tolerated_failures >= 1) |
| 4367 | allowed |= btrfs_raid_array[i].bg_flag; |
| 4368 | } |
| 4369 | do { |
| 4370 | seq = read_seqbegin(&fs_info->profiles_lock); |
| 4371 | |
| 4372 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
| 4373 | (fs_info->avail_system_alloc_bits & allowed) && |
| 4374 | !(bctl->sys.target & allowed)) || |
| 4375 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
| 4376 | (fs_info->avail_metadata_alloc_bits & allowed) && |
| 4377 | !(bctl->meta.target & allowed))) |
| 4378 | reducing_redundancy = true; |
| 4379 | else |
| 4380 | reducing_redundancy = false; |
| 4381 | |
| 4382 | /* if we're not converting, the target field is uninitialized */ |
| 4383 | meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ? |
| 4384 | bctl->meta.target : fs_info->avail_metadata_alloc_bits; |
| 4385 | data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ? |
| 4386 | bctl->data.target : fs_info->avail_data_alloc_bits; |
| 4387 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
| 4388 | |
| 4389 | if (reducing_redundancy) { |
| 4390 | if (bctl->flags & BTRFS_BALANCE_FORCE) { |
| 4391 | btrfs_info(fs_info, |
| 4392 | "balance: force reducing metadata redundancy"); |
| 4393 | } else { |
| 4394 | btrfs_err(fs_info, |
| 4395 | "balance: reduces metadata redundancy, use --force if you want this"); |
| 4396 | ret = -EINVAL; |
| 4397 | goto out; |
| 4398 | } |
| 4399 | } |
| 4400 | |
| 4401 | if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) < |
| 4402 | btrfs_get_num_tolerated_disk_barrier_failures(data_target)) { |
| 4403 | btrfs_warn(fs_info, |
| 4404 | "balance: metadata profile %s has lower redundancy than data profile %s", |
| 4405 | btrfs_bg_type_to_raid_name(meta_target), |
| 4406 | btrfs_bg_type_to_raid_name(data_target)); |
| 4407 | } |
| 4408 | |
| 4409 | ret = insert_balance_item(fs_info, bctl); |
| 4410 | if (ret && ret != -EEXIST) |
| 4411 | goto out; |
| 4412 | |
| 4413 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
| 4414 | BUG_ON(ret == -EEXIST); |
| 4415 | BUG_ON(fs_info->balance_ctl); |
| 4416 | spin_lock(&fs_info->balance_lock); |
| 4417 | fs_info->balance_ctl = bctl; |
| 4418 | spin_unlock(&fs_info->balance_lock); |
| 4419 | } else { |
| 4420 | BUG_ON(ret != -EEXIST); |
| 4421 | spin_lock(&fs_info->balance_lock); |
| 4422 | update_balance_args(bctl); |
| 4423 | spin_unlock(&fs_info->balance_lock); |
| 4424 | } |
| 4425 | |
| 4426 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
| 4427 | set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
| 4428 | describe_balance_start_or_resume(fs_info); |
| 4429 | mutex_unlock(&fs_info->balance_mutex); |
| 4430 | |
| 4431 | ret = __btrfs_balance(fs_info); |
| 4432 | |
| 4433 | mutex_lock(&fs_info->balance_mutex); |
| 4434 | if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) { |
| 4435 | btrfs_info(fs_info, "balance: paused"); |
| 4436 | btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED); |
| 4437 | paused = true; |
| 4438 | } |
| 4439 | /* |
| 4440 | * Balance can be canceled by: |
| 4441 | * |
| 4442 | * - Regular cancel request |
| 4443 | * Then ret == -ECANCELED and balance_cancel_req > 0 |
| 4444 | * |
| 4445 | * - Fatal signal to "btrfs" process |
| 4446 | * Either the signal caught by wait_reserve_ticket() and callers |
| 4447 | * got -EINTR, or caught by btrfs_should_cancel_balance() and |
| 4448 | * got -ECANCELED. |
| 4449 | * Either way, in this case balance_cancel_req = 0, and |
| 4450 | * ret == -EINTR or ret == -ECANCELED. |
| 4451 | * |
| 4452 | * So here we only check the return value to catch canceled balance. |
| 4453 | */ |
| 4454 | else if (ret == -ECANCELED || ret == -EINTR) |
| 4455 | btrfs_info(fs_info, "balance: canceled"); |
| 4456 | else |
| 4457 | btrfs_info(fs_info, "balance: ended with status: %d", ret); |
| 4458 | |
| 4459 | clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags); |
| 4460 | |
| 4461 | if (bargs) { |
| 4462 | memset(bargs, 0, sizeof(*bargs)); |
| 4463 | btrfs_update_ioctl_balance_args(fs_info, bargs); |
| 4464 | } |
| 4465 | |
| 4466 | /* We didn't pause, we can clean everything up. */ |
| 4467 | if (!paused) { |
| 4468 | reset_balance_state(fs_info); |
| 4469 | btrfs_exclop_finish(fs_info); |
| 4470 | } |
| 4471 | |
| 4472 | wake_up(&fs_info->balance_wait_q); |
| 4473 | |
| 4474 | return ret; |
| 4475 | out: |
| 4476 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
| 4477 | reset_balance_state(fs_info); |
| 4478 | else |
| 4479 | kfree(bctl); |
| 4480 | btrfs_exclop_finish(fs_info); |
| 4481 | |
| 4482 | return ret; |
| 4483 | } |
| 4484 | |
| 4485 | static int balance_kthread(void *data) |
| 4486 | { |
| 4487 | struct btrfs_fs_info *fs_info = data; |
| 4488 | int ret = 0; |
| 4489 | |
| 4490 | sb_start_write(fs_info->sb); |
| 4491 | mutex_lock(&fs_info->balance_mutex); |
| 4492 | if (fs_info->balance_ctl) |
| 4493 | ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL); |
| 4494 | mutex_unlock(&fs_info->balance_mutex); |
| 4495 | sb_end_write(fs_info->sb); |
| 4496 | |
| 4497 | return ret; |
| 4498 | } |
| 4499 | |
| 4500 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
| 4501 | { |
| 4502 | struct task_struct *tsk; |
| 4503 | |
| 4504 | mutex_lock(&fs_info->balance_mutex); |
| 4505 | if (!fs_info->balance_ctl) { |
| 4506 | mutex_unlock(&fs_info->balance_mutex); |
| 4507 | return 0; |
| 4508 | } |
| 4509 | mutex_unlock(&fs_info->balance_mutex); |
| 4510 | |
| 4511 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
| 4512 | btrfs_info(fs_info, "balance: resume skipped"); |
| 4513 | return 0; |
| 4514 | } |
| 4515 | |
| 4516 | spin_lock(&fs_info->super_lock); |
| 4517 | ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED); |
| 4518 | fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE; |
| 4519 | spin_unlock(&fs_info->super_lock); |
| 4520 | /* |
| 4521 | * A ro->rw remount sequence should continue with the paused balance |
| 4522 | * regardless of who pauses it, system or the user as of now, so set |
| 4523 | * the resume flag. |
| 4524 | */ |
| 4525 | spin_lock(&fs_info->balance_lock); |
| 4526 | fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME; |
| 4527 | spin_unlock(&fs_info->balance_lock); |
| 4528 | |
| 4529 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); |
| 4530 | return PTR_ERR_OR_ZERO(tsk); |
| 4531 | } |
| 4532 | |
| 4533 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
| 4534 | { |
| 4535 | struct btrfs_balance_control *bctl; |
| 4536 | struct btrfs_balance_item *item; |
| 4537 | struct btrfs_disk_balance_args disk_bargs; |
| 4538 | struct btrfs_path *path; |
| 4539 | struct extent_buffer *leaf; |
| 4540 | struct btrfs_key key; |
| 4541 | int ret; |
| 4542 | |
| 4543 | path = btrfs_alloc_path(); |
| 4544 | if (!path) |
| 4545 | return -ENOMEM; |
| 4546 | |
| 4547 | key.objectid = BTRFS_BALANCE_OBJECTID; |
| 4548 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
| 4549 | key.offset = 0; |
| 4550 | |
| 4551 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
| 4552 | if (ret < 0) |
| 4553 | goto out; |
| 4554 | if (ret > 0) { /* ret = -ENOENT; */ |
| 4555 | ret = 0; |
| 4556 | goto out; |
| 4557 | } |
| 4558 | |
| 4559 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); |
| 4560 | if (!bctl) { |
| 4561 | ret = -ENOMEM; |
| 4562 | goto out; |
| 4563 | } |
| 4564 | |
| 4565 | leaf = path->nodes[0]; |
| 4566 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); |
| 4567 | |
| 4568 | bctl->flags = btrfs_balance_flags(leaf, item); |
| 4569 | bctl->flags |= BTRFS_BALANCE_RESUME; |
| 4570 | |
| 4571 | btrfs_balance_data(leaf, item, &disk_bargs); |
| 4572 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); |
| 4573 | btrfs_balance_meta(leaf, item, &disk_bargs); |
| 4574 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); |
| 4575 | btrfs_balance_sys(leaf, item, &disk_bargs); |
| 4576 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); |
| 4577 | |
| 4578 | /* |
| 4579 | * This should never happen, as the paused balance state is recovered |
| 4580 | * during mount without any chance of other exclusive ops to collide. |
| 4581 | * |
| 4582 | * This gives the exclusive op status to balance and keeps in paused |
| 4583 | * state until user intervention (cancel or umount). If the ownership |
| 4584 | * cannot be assigned, show a message but do not fail. The balance |
| 4585 | * is in a paused state and must have fs_info::balance_ctl properly |
| 4586 | * set up. |
| 4587 | */ |
| 4588 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED)) |
| 4589 | btrfs_warn(fs_info, |
| 4590 | "balance: cannot set exclusive op status, resume manually"); |
| 4591 | |
| 4592 | btrfs_release_path(path); |
| 4593 | |
| 4594 | mutex_lock(&fs_info->balance_mutex); |
| 4595 | BUG_ON(fs_info->balance_ctl); |
| 4596 | spin_lock(&fs_info->balance_lock); |
| 4597 | fs_info->balance_ctl = bctl; |
| 4598 | spin_unlock(&fs_info->balance_lock); |
| 4599 | mutex_unlock(&fs_info->balance_mutex); |
| 4600 | out: |
| 4601 | btrfs_free_path(path); |
| 4602 | return ret; |
| 4603 | } |
| 4604 | |
| 4605 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
| 4606 | { |
| 4607 | int ret = 0; |
| 4608 | |
| 4609 | mutex_lock(&fs_info->balance_mutex); |
| 4610 | if (!fs_info->balance_ctl) { |
| 4611 | mutex_unlock(&fs_info->balance_mutex); |
| 4612 | return -ENOTCONN; |
| 4613 | } |
| 4614 | |
| 4615 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
| 4616 | atomic_inc(&fs_info->balance_pause_req); |
| 4617 | mutex_unlock(&fs_info->balance_mutex); |
| 4618 | |
| 4619 | wait_event(fs_info->balance_wait_q, |
| 4620 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
| 4621 | |
| 4622 | mutex_lock(&fs_info->balance_mutex); |
| 4623 | /* we are good with balance_ctl ripped off from under us */ |
| 4624 | BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
| 4625 | atomic_dec(&fs_info->balance_pause_req); |
| 4626 | } else { |
| 4627 | ret = -ENOTCONN; |
| 4628 | } |
| 4629 | |
| 4630 | mutex_unlock(&fs_info->balance_mutex); |
| 4631 | return ret; |
| 4632 | } |
| 4633 | |
| 4634 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
| 4635 | { |
| 4636 | mutex_lock(&fs_info->balance_mutex); |
| 4637 | if (!fs_info->balance_ctl) { |
| 4638 | mutex_unlock(&fs_info->balance_mutex); |
| 4639 | return -ENOTCONN; |
| 4640 | } |
| 4641 | |
| 4642 | /* |
| 4643 | * A paused balance with the item stored on disk can be resumed at |
| 4644 | * mount time if the mount is read-write. Otherwise it's still paused |
| 4645 | * and we must not allow cancelling as it deletes the item. |
| 4646 | */ |
| 4647 | if (sb_rdonly(fs_info->sb)) { |
| 4648 | mutex_unlock(&fs_info->balance_mutex); |
| 4649 | return -EROFS; |
| 4650 | } |
| 4651 | |
| 4652 | atomic_inc(&fs_info->balance_cancel_req); |
| 4653 | /* |
| 4654 | * if we are running just wait and return, balance item is |
| 4655 | * deleted in btrfs_balance in this case |
| 4656 | */ |
| 4657 | if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) { |
| 4658 | mutex_unlock(&fs_info->balance_mutex); |
| 4659 | wait_event(fs_info->balance_wait_q, |
| 4660 | !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
| 4661 | mutex_lock(&fs_info->balance_mutex); |
| 4662 | } else { |
| 4663 | mutex_unlock(&fs_info->balance_mutex); |
| 4664 | /* |
| 4665 | * Lock released to allow other waiters to continue, we'll |
| 4666 | * reexamine the status again. |
| 4667 | */ |
| 4668 | mutex_lock(&fs_info->balance_mutex); |
| 4669 | |
| 4670 | if (fs_info->balance_ctl) { |
| 4671 | reset_balance_state(fs_info); |
| 4672 | btrfs_exclop_finish(fs_info); |
| 4673 | btrfs_info(fs_info, "balance: canceled"); |
| 4674 | } |
| 4675 | } |
| 4676 | |
| 4677 | ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)); |
| 4678 | atomic_dec(&fs_info->balance_cancel_req); |
| 4679 | mutex_unlock(&fs_info->balance_mutex); |
| 4680 | return 0; |
| 4681 | } |
| 4682 | |
| 4683 | int btrfs_uuid_scan_kthread(void *data) |
| 4684 | { |
| 4685 | struct btrfs_fs_info *fs_info = data; |
| 4686 | struct btrfs_root *root = fs_info->tree_root; |
| 4687 | struct btrfs_key key; |
| 4688 | struct btrfs_path *path = NULL; |
| 4689 | int ret = 0; |
| 4690 | struct extent_buffer *eb; |
| 4691 | int slot; |
| 4692 | struct btrfs_root_item root_item; |
| 4693 | u32 item_size; |
| 4694 | struct btrfs_trans_handle *trans = NULL; |
| 4695 | bool closing = false; |
| 4696 | |
| 4697 | path = btrfs_alloc_path(); |
| 4698 | if (!path) { |
| 4699 | ret = -ENOMEM; |
| 4700 | goto out; |
| 4701 | } |
| 4702 | |
| 4703 | key.objectid = 0; |
| 4704 | key.type = BTRFS_ROOT_ITEM_KEY; |
| 4705 | key.offset = 0; |
| 4706 | |
| 4707 | while (1) { |
| 4708 | if (btrfs_fs_closing(fs_info)) { |
| 4709 | closing = true; |
| 4710 | break; |
| 4711 | } |
| 4712 | ret = btrfs_search_forward(root, &key, path, |
| 4713 | BTRFS_OLDEST_GENERATION); |
| 4714 | if (ret) { |
| 4715 | if (ret > 0) |
| 4716 | ret = 0; |
| 4717 | break; |
| 4718 | } |
| 4719 | |
| 4720 | if (key.type != BTRFS_ROOT_ITEM_KEY || |
| 4721 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && |
| 4722 | key.objectid != BTRFS_FS_TREE_OBJECTID) || |
| 4723 | key.objectid > BTRFS_LAST_FREE_OBJECTID) |
| 4724 | goto skip; |
| 4725 | |
| 4726 | eb = path->nodes[0]; |
| 4727 | slot = path->slots[0]; |
| 4728 | item_size = btrfs_item_size(eb, slot); |
| 4729 | if (item_size < sizeof(root_item)) |
| 4730 | goto skip; |
| 4731 | |
| 4732 | read_extent_buffer(eb, &root_item, |
| 4733 | btrfs_item_ptr_offset(eb, slot), |
| 4734 | (int)sizeof(root_item)); |
| 4735 | if (btrfs_root_refs(&root_item) == 0) |
| 4736 | goto skip; |
| 4737 | |
| 4738 | if (!btrfs_is_empty_uuid(root_item.uuid) || |
| 4739 | !btrfs_is_empty_uuid(root_item.received_uuid)) { |
| 4740 | if (trans) |
| 4741 | goto update_tree; |
| 4742 | |
| 4743 | btrfs_release_path(path); |
| 4744 | /* |
| 4745 | * 1 - subvol uuid item |
| 4746 | * 1 - received_subvol uuid item |
| 4747 | */ |
| 4748 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); |
| 4749 | if (IS_ERR(trans)) { |
| 4750 | ret = PTR_ERR(trans); |
| 4751 | break; |
| 4752 | } |
| 4753 | continue; |
| 4754 | } else { |
| 4755 | goto skip; |
| 4756 | } |
| 4757 | update_tree: |
| 4758 | btrfs_release_path(path); |
| 4759 | if (!btrfs_is_empty_uuid(root_item.uuid)) { |
| 4760 | ret = btrfs_uuid_tree_add(trans, root_item.uuid, |
| 4761 | BTRFS_UUID_KEY_SUBVOL, |
| 4762 | key.objectid); |
| 4763 | if (ret < 0) { |
| 4764 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
| 4765 | ret); |
| 4766 | break; |
| 4767 | } |
| 4768 | } |
| 4769 | |
| 4770 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { |
| 4771 | ret = btrfs_uuid_tree_add(trans, |
| 4772 | root_item.received_uuid, |
| 4773 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
| 4774 | key.objectid); |
| 4775 | if (ret < 0) { |
| 4776 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
| 4777 | ret); |
| 4778 | break; |
| 4779 | } |
| 4780 | } |
| 4781 | |
| 4782 | skip: |
| 4783 | btrfs_release_path(path); |
| 4784 | if (trans) { |
| 4785 | ret = btrfs_end_transaction(trans); |
| 4786 | trans = NULL; |
| 4787 | if (ret) |
| 4788 | break; |
| 4789 | } |
| 4790 | |
| 4791 | if (key.offset < (u64)-1) { |
| 4792 | key.offset++; |
| 4793 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { |
| 4794 | key.offset = 0; |
| 4795 | key.type = BTRFS_ROOT_ITEM_KEY; |
| 4796 | } else if (key.objectid < (u64)-1) { |
| 4797 | key.offset = 0; |
| 4798 | key.type = BTRFS_ROOT_ITEM_KEY; |
| 4799 | key.objectid++; |
| 4800 | } else { |
| 4801 | break; |
| 4802 | } |
| 4803 | cond_resched(); |
| 4804 | } |
| 4805 | |
| 4806 | out: |
| 4807 | btrfs_free_path(path); |
| 4808 | if (trans && !IS_ERR(trans)) |
| 4809 | btrfs_end_transaction(trans); |
| 4810 | if (ret) |
| 4811 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
| 4812 | else if (!closing) |
| 4813 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
| 4814 | up(&fs_info->uuid_tree_rescan_sem); |
| 4815 | return 0; |
| 4816 | } |
| 4817 | |
| 4818 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
| 4819 | { |
| 4820 | struct btrfs_trans_handle *trans; |
| 4821 | struct btrfs_root *tree_root = fs_info->tree_root; |
| 4822 | struct btrfs_root *uuid_root; |
| 4823 | struct task_struct *task; |
| 4824 | int ret; |
| 4825 | |
| 4826 | /* |
| 4827 | * 1 - root node |
| 4828 | * 1 - root item |
| 4829 | */ |
| 4830 | trans = btrfs_start_transaction(tree_root, 2); |
| 4831 | if (IS_ERR(trans)) |
| 4832 | return PTR_ERR(trans); |
| 4833 | |
| 4834 | uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID); |
| 4835 | if (IS_ERR(uuid_root)) { |
| 4836 | ret = PTR_ERR(uuid_root); |
| 4837 | btrfs_abort_transaction(trans, ret); |
| 4838 | btrfs_end_transaction(trans); |
| 4839 | return ret; |
| 4840 | } |
| 4841 | |
| 4842 | fs_info->uuid_root = uuid_root; |
| 4843 | |
| 4844 | ret = btrfs_commit_transaction(trans); |
| 4845 | if (ret) |
| 4846 | return ret; |
| 4847 | |
| 4848 | down(&fs_info->uuid_tree_rescan_sem); |
| 4849 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); |
| 4850 | if (IS_ERR(task)) { |
| 4851 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
| 4852 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
| 4853 | up(&fs_info->uuid_tree_rescan_sem); |
| 4854 | return PTR_ERR(task); |
| 4855 | } |
| 4856 | |
| 4857 | return 0; |
| 4858 | } |
| 4859 | |
| 4860 | /* |
| 4861 | * shrinking a device means finding all of the device extents past |
| 4862 | * the new size, and then following the back refs to the chunks. |
| 4863 | * The chunk relocation code actually frees the device extent |
| 4864 | */ |
| 4865 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) |
| 4866 | { |
| 4867 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 4868 | struct btrfs_root *root = fs_info->dev_root; |
| 4869 | struct btrfs_trans_handle *trans; |
| 4870 | struct btrfs_dev_extent *dev_extent = NULL; |
| 4871 | struct btrfs_path *path; |
| 4872 | u64 length; |
| 4873 | u64 chunk_offset; |
| 4874 | int ret; |
| 4875 | int slot; |
| 4876 | int failed = 0; |
| 4877 | bool retried = false; |
| 4878 | struct extent_buffer *l; |
| 4879 | struct btrfs_key key; |
| 4880 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
| 4881 | u64 old_total = btrfs_super_total_bytes(super_copy); |
| 4882 | u64 old_size = btrfs_device_get_total_bytes(device); |
| 4883 | u64 diff; |
| 4884 | u64 start; |
| 4885 | u64 free_diff = 0; |
| 4886 | |
| 4887 | new_size = round_down(new_size, fs_info->sectorsize); |
| 4888 | start = new_size; |
| 4889 | diff = round_down(old_size - new_size, fs_info->sectorsize); |
| 4890 | |
| 4891 | if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
| 4892 | return -EINVAL; |
| 4893 | |
| 4894 | path = btrfs_alloc_path(); |
| 4895 | if (!path) |
| 4896 | return -ENOMEM; |
| 4897 | |
| 4898 | path->reada = READA_BACK; |
| 4899 | |
| 4900 | trans = btrfs_start_transaction(root, 0); |
| 4901 | if (IS_ERR(trans)) { |
| 4902 | btrfs_free_path(path); |
| 4903 | return PTR_ERR(trans); |
| 4904 | } |
| 4905 | |
| 4906 | mutex_lock(&fs_info->chunk_mutex); |
| 4907 | |
| 4908 | btrfs_device_set_total_bytes(device, new_size); |
| 4909 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 4910 | device->fs_devices->total_rw_bytes -= diff; |
| 4911 | |
| 4912 | /* |
| 4913 | * The new free_chunk_space is new_size - used, so we have to |
| 4914 | * subtract the delta of the old free_chunk_space which included |
| 4915 | * old_size - used. If used > new_size then just subtract this |
| 4916 | * entire device's free space. |
| 4917 | */ |
| 4918 | if (device->bytes_used < new_size) |
| 4919 | free_diff = (old_size - device->bytes_used) - |
| 4920 | (new_size - device->bytes_used); |
| 4921 | else |
| 4922 | free_diff = old_size - device->bytes_used; |
| 4923 | atomic64_sub(free_diff, &fs_info->free_chunk_space); |
| 4924 | } |
| 4925 | |
| 4926 | /* |
| 4927 | * Once the device's size has been set to the new size, ensure all |
| 4928 | * in-memory chunks are synced to disk so that the loop below sees them |
| 4929 | * and relocates them accordingly. |
| 4930 | */ |
| 4931 | if (contains_pending_extent(device, &start, diff)) { |
| 4932 | mutex_unlock(&fs_info->chunk_mutex); |
| 4933 | ret = btrfs_commit_transaction(trans); |
| 4934 | if (ret) |
| 4935 | goto done; |
| 4936 | } else { |
| 4937 | mutex_unlock(&fs_info->chunk_mutex); |
| 4938 | btrfs_end_transaction(trans); |
| 4939 | } |
| 4940 | |
| 4941 | again: |
| 4942 | key.objectid = device->devid; |
| 4943 | key.offset = (u64)-1; |
| 4944 | key.type = BTRFS_DEV_EXTENT_KEY; |
| 4945 | |
| 4946 | do { |
| 4947 | mutex_lock(&fs_info->reclaim_bgs_lock); |
| 4948 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 4949 | if (ret < 0) { |
| 4950 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4951 | goto done; |
| 4952 | } |
| 4953 | |
| 4954 | ret = btrfs_previous_item(root, path, 0, key.type); |
| 4955 | if (ret) { |
| 4956 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4957 | if (ret < 0) |
| 4958 | goto done; |
| 4959 | ret = 0; |
| 4960 | btrfs_release_path(path); |
| 4961 | break; |
| 4962 | } |
| 4963 | |
| 4964 | l = path->nodes[0]; |
| 4965 | slot = path->slots[0]; |
| 4966 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); |
| 4967 | |
| 4968 | if (key.objectid != device->devid) { |
| 4969 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4970 | btrfs_release_path(path); |
| 4971 | break; |
| 4972 | } |
| 4973 | |
| 4974 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
| 4975 | length = btrfs_dev_extent_length(l, dev_extent); |
| 4976 | |
| 4977 | if (key.offset + length <= new_size) { |
| 4978 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4979 | btrfs_release_path(path); |
| 4980 | break; |
| 4981 | } |
| 4982 | |
| 4983 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
| 4984 | btrfs_release_path(path); |
| 4985 | |
| 4986 | /* |
| 4987 | * We may be relocating the only data chunk we have, |
| 4988 | * which could potentially end up with losing data's |
| 4989 | * raid profile, so lets allocate an empty one in |
| 4990 | * advance. |
| 4991 | */ |
| 4992 | ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset); |
| 4993 | if (ret < 0) { |
| 4994 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 4995 | goto done; |
| 4996 | } |
| 4997 | |
| 4998 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
| 4999 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 5000 | if (ret == -ENOSPC) { |
| 5001 | failed++; |
| 5002 | } else if (ret) { |
| 5003 | if (ret == -ETXTBSY) { |
| 5004 | btrfs_warn(fs_info, |
| 5005 | "could not shrink block group %llu due to active swapfile", |
| 5006 | chunk_offset); |
| 5007 | } |
| 5008 | goto done; |
| 5009 | } |
| 5010 | } while (key.offset-- > 0); |
| 5011 | |
| 5012 | if (failed && !retried) { |
| 5013 | failed = 0; |
| 5014 | retried = true; |
| 5015 | goto again; |
| 5016 | } else if (failed && retried) { |
| 5017 | ret = -ENOSPC; |
| 5018 | goto done; |
| 5019 | } |
| 5020 | |
| 5021 | /* Shrinking succeeded, else we would be at "done". */ |
| 5022 | trans = btrfs_start_transaction(root, 0); |
| 5023 | if (IS_ERR(trans)) { |
| 5024 | ret = PTR_ERR(trans); |
| 5025 | goto done; |
| 5026 | } |
| 5027 | |
| 5028 | mutex_lock(&fs_info->chunk_mutex); |
| 5029 | /* Clear all state bits beyond the shrunk device size */ |
| 5030 | clear_extent_bits(&device->alloc_state, new_size, (u64)-1, |
| 5031 | CHUNK_STATE_MASK); |
| 5032 | |
| 5033 | btrfs_device_set_disk_total_bytes(device, new_size); |
| 5034 | if (list_empty(&device->post_commit_list)) |
| 5035 | list_add_tail(&device->post_commit_list, |
| 5036 | &trans->transaction->dev_update_list); |
| 5037 | |
| 5038 | WARN_ON(diff > old_total); |
| 5039 | btrfs_set_super_total_bytes(super_copy, |
| 5040 | round_down(old_total - diff, fs_info->sectorsize)); |
| 5041 | mutex_unlock(&fs_info->chunk_mutex); |
| 5042 | |
| 5043 | btrfs_reserve_chunk_metadata(trans, false); |
| 5044 | /* Now btrfs_update_device() will change the on-disk size. */ |
| 5045 | ret = btrfs_update_device(trans, device); |
| 5046 | btrfs_trans_release_chunk_metadata(trans); |
| 5047 | if (ret < 0) { |
| 5048 | btrfs_abort_transaction(trans, ret); |
| 5049 | btrfs_end_transaction(trans); |
| 5050 | } else { |
| 5051 | ret = btrfs_commit_transaction(trans); |
| 5052 | } |
| 5053 | done: |
| 5054 | btrfs_free_path(path); |
| 5055 | if (ret) { |
| 5056 | mutex_lock(&fs_info->chunk_mutex); |
| 5057 | btrfs_device_set_total_bytes(device, old_size); |
| 5058 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 5059 | device->fs_devices->total_rw_bytes += diff; |
| 5060 | atomic64_add(free_diff, &fs_info->free_chunk_space); |
| 5061 | } |
| 5062 | mutex_unlock(&fs_info->chunk_mutex); |
| 5063 | } |
| 5064 | return ret; |
| 5065 | } |
| 5066 | |
| 5067 | static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info, |
| 5068 | struct btrfs_key *key, |
| 5069 | struct btrfs_chunk *chunk, int item_size) |
| 5070 | { |
| 5071 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
| 5072 | struct btrfs_disk_key disk_key; |
| 5073 | u32 array_size; |
| 5074 | u8 *ptr; |
| 5075 | |
| 5076 | lockdep_assert_held(&fs_info->chunk_mutex); |
| 5077 | |
| 5078 | array_size = btrfs_super_sys_array_size(super_copy); |
| 5079 | if (array_size + item_size + sizeof(disk_key) |
| 5080 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) |
| 5081 | return -EFBIG; |
| 5082 | |
| 5083 | ptr = super_copy->sys_chunk_array + array_size; |
| 5084 | btrfs_cpu_key_to_disk(&disk_key, key); |
| 5085 | memcpy(ptr, &disk_key, sizeof(disk_key)); |
| 5086 | ptr += sizeof(disk_key); |
| 5087 | memcpy(ptr, chunk, item_size); |
| 5088 | item_size += sizeof(disk_key); |
| 5089 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); |
| 5090 | |
| 5091 | return 0; |
| 5092 | } |
| 5093 | |
| 5094 | /* |
| 5095 | * sort the devices in descending order by max_avail, total_avail |
| 5096 | */ |
| 5097 | static int btrfs_cmp_device_info(const void *a, const void *b) |
| 5098 | { |
| 5099 | const struct btrfs_device_info *di_a = a; |
| 5100 | const struct btrfs_device_info *di_b = b; |
| 5101 | |
| 5102 | if (di_a->max_avail > di_b->max_avail) |
| 5103 | return -1; |
| 5104 | if (di_a->max_avail < di_b->max_avail) |
| 5105 | return 1; |
| 5106 | if (di_a->total_avail > di_b->total_avail) |
| 5107 | return -1; |
| 5108 | if (di_a->total_avail < di_b->total_avail) |
| 5109 | return 1; |
| 5110 | return 0; |
| 5111 | } |
| 5112 | |
| 5113 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) |
| 5114 | { |
| 5115 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
| 5116 | return; |
| 5117 | |
| 5118 | btrfs_set_fs_incompat(info, RAID56); |
| 5119 | } |
| 5120 | |
| 5121 | static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type) |
| 5122 | { |
| 5123 | if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4))) |
| 5124 | return; |
| 5125 | |
| 5126 | btrfs_set_fs_incompat(info, RAID1C34); |
| 5127 | } |
| 5128 | |
| 5129 | /* |
| 5130 | * Structure used internally for btrfs_create_chunk() function. |
| 5131 | * Wraps needed parameters. |
| 5132 | */ |
| 5133 | struct alloc_chunk_ctl { |
| 5134 | u64 start; |
| 5135 | u64 type; |
| 5136 | /* Total number of stripes to allocate */ |
| 5137 | int num_stripes; |
| 5138 | /* sub_stripes info for map */ |
| 5139 | int sub_stripes; |
| 5140 | /* Stripes per device */ |
| 5141 | int dev_stripes; |
| 5142 | /* Maximum number of devices to use */ |
| 5143 | int devs_max; |
| 5144 | /* Minimum number of devices to use */ |
| 5145 | int devs_min; |
| 5146 | /* ndevs has to be a multiple of this */ |
| 5147 | int devs_increment; |
| 5148 | /* Number of copies */ |
| 5149 | int ncopies; |
| 5150 | /* Number of stripes worth of bytes to store parity information */ |
| 5151 | int nparity; |
| 5152 | u64 max_stripe_size; |
| 5153 | u64 max_chunk_size; |
| 5154 | u64 dev_extent_min; |
| 5155 | u64 stripe_size; |
| 5156 | u64 chunk_size; |
| 5157 | int ndevs; |
| 5158 | }; |
| 5159 | |
| 5160 | static void init_alloc_chunk_ctl_policy_regular( |
| 5161 | struct btrfs_fs_devices *fs_devices, |
| 5162 | struct alloc_chunk_ctl *ctl) |
| 5163 | { |
| 5164 | struct btrfs_space_info *space_info; |
| 5165 | |
| 5166 | space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type); |
| 5167 | ASSERT(space_info); |
| 5168 | |
| 5169 | ctl->max_chunk_size = READ_ONCE(space_info->chunk_size); |
| 5170 | ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G); |
| 5171 | |
| 5172 | if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM) |
| 5173 | ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK); |
| 5174 | |
| 5175 | /* We don't want a chunk larger than 10% of writable space */ |
| 5176 | ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10), |
| 5177 | ctl->max_chunk_size); |
| 5178 | ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes); |
| 5179 | } |
| 5180 | |
| 5181 | static void init_alloc_chunk_ctl_policy_zoned( |
| 5182 | struct btrfs_fs_devices *fs_devices, |
| 5183 | struct alloc_chunk_ctl *ctl) |
| 5184 | { |
| 5185 | u64 zone_size = fs_devices->fs_info->zone_size; |
| 5186 | u64 limit; |
| 5187 | int min_num_stripes = ctl->devs_min * ctl->dev_stripes; |
| 5188 | int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies; |
| 5189 | u64 min_chunk_size = min_data_stripes * zone_size; |
| 5190 | u64 type = ctl->type; |
| 5191 | |
| 5192 | ctl->max_stripe_size = zone_size; |
| 5193 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
| 5194 | ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE, |
| 5195 | zone_size); |
| 5196 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
| 5197 | ctl->max_chunk_size = ctl->max_stripe_size; |
| 5198 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
| 5199 | ctl->max_chunk_size = 2 * ctl->max_stripe_size; |
| 5200 | ctl->devs_max = min_t(int, ctl->devs_max, |
| 5201 | BTRFS_MAX_DEVS_SYS_CHUNK); |
| 5202 | } else { |
| 5203 | BUG(); |
| 5204 | } |
| 5205 | |
| 5206 | /* We don't want a chunk larger than 10% of writable space */ |
| 5207 | limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10), |
| 5208 | zone_size), |
| 5209 | min_chunk_size); |
| 5210 | ctl->max_chunk_size = min(limit, ctl->max_chunk_size); |
| 5211 | ctl->dev_extent_min = zone_size * ctl->dev_stripes; |
| 5212 | } |
| 5213 | |
| 5214 | static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices, |
| 5215 | struct alloc_chunk_ctl *ctl) |
| 5216 | { |
| 5217 | int index = btrfs_bg_flags_to_raid_index(ctl->type); |
| 5218 | |
| 5219 | ctl->sub_stripes = btrfs_raid_array[index].sub_stripes; |
| 5220 | ctl->dev_stripes = btrfs_raid_array[index].dev_stripes; |
| 5221 | ctl->devs_max = btrfs_raid_array[index].devs_max; |
| 5222 | if (!ctl->devs_max) |
| 5223 | ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info); |
| 5224 | ctl->devs_min = btrfs_raid_array[index].devs_min; |
| 5225 | ctl->devs_increment = btrfs_raid_array[index].devs_increment; |
| 5226 | ctl->ncopies = btrfs_raid_array[index].ncopies; |
| 5227 | ctl->nparity = btrfs_raid_array[index].nparity; |
| 5228 | ctl->ndevs = 0; |
| 5229 | |
| 5230 | switch (fs_devices->chunk_alloc_policy) { |
| 5231 | case BTRFS_CHUNK_ALLOC_REGULAR: |
| 5232 | init_alloc_chunk_ctl_policy_regular(fs_devices, ctl); |
| 5233 | break; |
| 5234 | case BTRFS_CHUNK_ALLOC_ZONED: |
| 5235 | init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl); |
| 5236 | break; |
| 5237 | default: |
| 5238 | BUG(); |
| 5239 | } |
| 5240 | } |
| 5241 | |
| 5242 | static int gather_device_info(struct btrfs_fs_devices *fs_devices, |
| 5243 | struct alloc_chunk_ctl *ctl, |
| 5244 | struct btrfs_device_info *devices_info) |
| 5245 | { |
| 5246 | struct btrfs_fs_info *info = fs_devices->fs_info; |
| 5247 | struct btrfs_device *device; |
| 5248 | u64 total_avail; |
| 5249 | u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes; |
| 5250 | int ret; |
| 5251 | int ndevs = 0; |
| 5252 | u64 max_avail; |
| 5253 | u64 dev_offset; |
| 5254 | |
| 5255 | /* |
| 5256 | * in the first pass through the devices list, we gather information |
| 5257 | * about the available holes on each device. |
| 5258 | */ |
| 5259 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
| 5260 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) { |
| 5261 | WARN(1, KERN_ERR |
| 5262 | "BTRFS: read-only device in alloc_list\n"); |
| 5263 | continue; |
| 5264 | } |
| 5265 | |
| 5266 | if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
| 5267 | &device->dev_state) || |
| 5268 | test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) |
| 5269 | continue; |
| 5270 | |
| 5271 | if (device->total_bytes > device->bytes_used) |
| 5272 | total_avail = device->total_bytes - device->bytes_used; |
| 5273 | else |
| 5274 | total_avail = 0; |
| 5275 | |
| 5276 | /* If there is no space on this device, skip it. */ |
| 5277 | if (total_avail < ctl->dev_extent_min) |
| 5278 | continue; |
| 5279 | |
| 5280 | ret = find_free_dev_extent(device, dev_extent_want, &dev_offset, |
| 5281 | &max_avail); |
| 5282 | if (ret && ret != -ENOSPC) |
| 5283 | return ret; |
| 5284 | |
| 5285 | if (ret == 0) |
| 5286 | max_avail = dev_extent_want; |
| 5287 | |
| 5288 | if (max_avail < ctl->dev_extent_min) { |
| 5289 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
| 5290 | btrfs_debug(info, |
| 5291 | "%s: devid %llu has no free space, have=%llu want=%llu", |
| 5292 | __func__, device->devid, max_avail, |
| 5293 | ctl->dev_extent_min); |
| 5294 | continue; |
| 5295 | } |
| 5296 | |
| 5297 | if (ndevs == fs_devices->rw_devices) { |
| 5298 | WARN(1, "%s: found more than %llu devices\n", |
| 5299 | __func__, fs_devices->rw_devices); |
| 5300 | break; |
| 5301 | } |
| 5302 | devices_info[ndevs].dev_offset = dev_offset; |
| 5303 | devices_info[ndevs].max_avail = max_avail; |
| 5304 | devices_info[ndevs].total_avail = total_avail; |
| 5305 | devices_info[ndevs].dev = device; |
| 5306 | ++ndevs; |
| 5307 | } |
| 5308 | ctl->ndevs = ndevs; |
| 5309 | |
| 5310 | /* |
| 5311 | * now sort the devices by hole size / available space |
| 5312 | */ |
| 5313 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), |
| 5314 | btrfs_cmp_device_info, NULL); |
| 5315 | |
| 5316 | return 0; |
| 5317 | } |
| 5318 | |
| 5319 | static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl, |
| 5320 | struct btrfs_device_info *devices_info) |
| 5321 | { |
| 5322 | /* Number of stripes that count for block group size */ |
| 5323 | int data_stripes; |
| 5324 | |
| 5325 | /* |
| 5326 | * The primary goal is to maximize the number of stripes, so use as |
| 5327 | * many devices as possible, even if the stripes are not maximum sized. |
| 5328 | * |
| 5329 | * The DUP profile stores more than one stripe per device, the |
| 5330 | * max_avail is the total size so we have to adjust. |
| 5331 | */ |
| 5332 | ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail, |
| 5333 | ctl->dev_stripes); |
| 5334 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; |
| 5335 | |
| 5336 | /* This will have to be fixed for RAID1 and RAID10 over more drives */ |
| 5337 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; |
| 5338 | |
| 5339 | /* |
| 5340 | * Use the number of data stripes to figure out how big this chunk is |
| 5341 | * really going to be in terms of logical address space, and compare |
| 5342 | * that answer with the max chunk size. If it's higher, we try to |
| 5343 | * reduce stripe_size. |
| 5344 | */ |
| 5345 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { |
| 5346 | /* |
| 5347 | * Reduce stripe_size, round it up to a 16MB boundary again and |
| 5348 | * then use it, unless it ends up being even bigger than the |
| 5349 | * previous value we had already. |
| 5350 | */ |
| 5351 | ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size, |
| 5352 | data_stripes), SZ_16M), |
| 5353 | ctl->stripe_size); |
| 5354 | } |
| 5355 | |
| 5356 | /* Stripe size should not go beyond 1G. */ |
| 5357 | ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G); |
| 5358 | |
| 5359 | /* Align to BTRFS_STRIPE_LEN */ |
| 5360 | ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN); |
| 5361 | ctl->chunk_size = ctl->stripe_size * data_stripes; |
| 5362 | |
| 5363 | return 0; |
| 5364 | } |
| 5365 | |
| 5366 | static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl, |
| 5367 | struct btrfs_device_info *devices_info) |
| 5368 | { |
| 5369 | u64 zone_size = devices_info[0].dev->zone_info->zone_size; |
| 5370 | /* Number of stripes that count for block group size */ |
| 5371 | int data_stripes; |
| 5372 | |
| 5373 | /* |
| 5374 | * It should hold because: |
| 5375 | * dev_extent_min == dev_extent_want == zone_size * dev_stripes |
| 5376 | */ |
| 5377 | ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min); |
| 5378 | |
| 5379 | ctl->stripe_size = zone_size; |
| 5380 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; |
| 5381 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; |
| 5382 | |
| 5383 | /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */ |
| 5384 | if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) { |
| 5385 | ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies, |
| 5386 | ctl->stripe_size) + ctl->nparity, |
| 5387 | ctl->dev_stripes); |
| 5388 | ctl->num_stripes = ctl->ndevs * ctl->dev_stripes; |
| 5389 | data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies; |
| 5390 | ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size); |
| 5391 | } |
| 5392 | |
| 5393 | ctl->chunk_size = ctl->stripe_size * data_stripes; |
| 5394 | |
| 5395 | return 0; |
| 5396 | } |
| 5397 | |
| 5398 | static int decide_stripe_size(struct btrfs_fs_devices *fs_devices, |
| 5399 | struct alloc_chunk_ctl *ctl, |
| 5400 | struct btrfs_device_info *devices_info) |
| 5401 | { |
| 5402 | struct btrfs_fs_info *info = fs_devices->fs_info; |
| 5403 | |
| 5404 | /* |
| 5405 | * Round down to number of usable stripes, devs_increment can be any |
| 5406 | * number so we can't use round_down() that requires power of 2, while |
| 5407 | * rounddown is safe. |
| 5408 | */ |
| 5409 | ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment); |
| 5410 | |
| 5411 | if (ctl->ndevs < ctl->devs_min) { |
| 5412 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) { |
| 5413 | btrfs_debug(info, |
| 5414 | "%s: not enough devices with free space: have=%d minimum required=%d", |
| 5415 | __func__, ctl->ndevs, ctl->devs_min); |
| 5416 | } |
| 5417 | return -ENOSPC; |
| 5418 | } |
| 5419 | |
| 5420 | ctl->ndevs = min(ctl->ndevs, ctl->devs_max); |
| 5421 | |
| 5422 | switch (fs_devices->chunk_alloc_policy) { |
| 5423 | case BTRFS_CHUNK_ALLOC_REGULAR: |
| 5424 | return decide_stripe_size_regular(ctl, devices_info); |
| 5425 | case BTRFS_CHUNK_ALLOC_ZONED: |
| 5426 | return decide_stripe_size_zoned(ctl, devices_info); |
| 5427 | default: |
| 5428 | BUG(); |
| 5429 | } |
| 5430 | } |
| 5431 | |
| 5432 | static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits) |
| 5433 | { |
| 5434 | for (int i = 0; i < map->num_stripes; i++) { |
| 5435 | struct btrfs_io_stripe *stripe = &map->stripes[i]; |
| 5436 | struct btrfs_device *device = stripe->dev; |
| 5437 | |
| 5438 | set_extent_bit(&device->alloc_state, stripe->physical, |
| 5439 | stripe->physical + map->stripe_size - 1, |
| 5440 | bits | EXTENT_NOWAIT, NULL); |
| 5441 | } |
| 5442 | } |
| 5443 | |
| 5444 | static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits) |
| 5445 | { |
| 5446 | for (int i = 0; i < map->num_stripes; i++) { |
| 5447 | struct btrfs_io_stripe *stripe = &map->stripes[i]; |
| 5448 | struct btrfs_device *device = stripe->dev; |
| 5449 | |
| 5450 | __clear_extent_bit(&device->alloc_state, stripe->physical, |
| 5451 | stripe->physical + map->stripe_size - 1, |
| 5452 | bits | EXTENT_NOWAIT, |
| 5453 | NULL, NULL); |
| 5454 | } |
| 5455 | } |
| 5456 | |
| 5457 | void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map) |
| 5458 | { |
| 5459 | write_lock(&fs_info->mapping_tree_lock); |
| 5460 | rb_erase_cached(&map->rb_node, &fs_info->mapping_tree); |
| 5461 | RB_CLEAR_NODE(&map->rb_node); |
| 5462 | chunk_map_device_clear_bits(map, CHUNK_ALLOCATED); |
| 5463 | write_unlock(&fs_info->mapping_tree_lock); |
| 5464 | |
| 5465 | /* Once for the tree reference. */ |
| 5466 | btrfs_free_chunk_map(map); |
| 5467 | } |
| 5468 | |
| 5469 | EXPORT_FOR_TESTS |
| 5470 | int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map) |
| 5471 | { |
| 5472 | struct rb_node **p; |
| 5473 | struct rb_node *parent = NULL; |
| 5474 | bool leftmost = true; |
| 5475 | |
| 5476 | write_lock(&fs_info->mapping_tree_lock); |
| 5477 | p = &fs_info->mapping_tree.rb_root.rb_node; |
| 5478 | while (*p) { |
| 5479 | struct btrfs_chunk_map *entry; |
| 5480 | |
| 5481 | parent = *p; |
| 5482 | entry = rb_entry(parent, struct btrfs_chunk_map, rb_node); |
| 5483 | |
| 5484 | if (map->start < entry->start) { |
| 5485 | p = &(*p)->rb_left; |
| 5486 | } else if (map->start > entry->start) { |
| 5487 | p = &(*p)->rb_right; |
| 5488 | leftmost = false; |
| 5489 | } else { |
| 5490 | write_unlock(&fs_info->mapping_tree_lock); |
| 5491 | return -EEXIST; |
| 5492 | } |
| 5493 | } |
| 5494 | rb_link_node(&map->rb_node, parent, p); |
| 5495 | rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost); |
| 5496 | chunk_map_device_set_bits(map, CHUNK_ALLOCATED); |
| 5497 | chunk_map_device_clear_bits(map, CHUNK_TRIMMED); |
| 5498 | write_unlock(&fs_info->mapping_tree_lock); |
| 5499 | |
| 5500 | return 0; |
| 5501 | } |
| 5502 | |
| 5503 | EXPORT_FOR_TESTS |
| 5504 | struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp) |
| 5505 | { |
| 5506 | struct btrfs_chunk_map *map; |
| 5507 | |
| 5508 | map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp); |
| 5509 | if (!map) |
| 5510 | return NULL; |
| 5511 | |
| 5512 | refcount_set(&map->refs, 1); |
| 5513 | RB_CLEAR_NODE(&map->rb_node); |
| 5514 | |
| 5515 | return map; |
| 5516 | } |
| 5517 | |
| 5518 | struct btrfs_chunk_map *btrfs_clone_chunk_map(struct btrfs_chunk_map *map, gfp_t gfp) |
| 5519 | { |
| 5520 | const int size = btrfs_chunk_map_size(map->num_stripes); |
| 5521 | struct btrfs_chunk_map *clone; |
| 5522 | |
| 5523 | clone = kmemdup(map, size, gfp); |
| 5524 | if (!clone) |
| 5525 | return NULL; |
| 5526 | |
| 5527 | refcount_set(&clone->refs, 1); |
| 5528 | RB_CLEAR_NODE(&clone->rb_node); |
| 5529 | |
| 5530 | return clone; |
| 5531 | } |
| 5532 | |
| 5533 | static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans, |
| 5534 | struct alloc_chunk_ctl *ctl, |
| 5535 | struct btrfs_device_info *devices_info) |
| 5536 | { |
| 5537 | struct btrfs_fs_info *info = trans->fs_info; |
| 5538 | struct btrfs_chunk_map *map; |
| 5539 | struct btrfs_block_group *block_group; |
| 5540 | u64 start = ctl->start; |
| 5541 | u64 type = ctl->type; |
| 5542 | int ret; |
| 5543 | int i; |
| 5544 | int j; |
| 5545 | |
| 5546 | map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS); |
| 5547 | if (!map) |
| 5548 | return ERR_PTR(-ENOMEM); |
| 5549 | |
| 5550 | map->start = start; |
| 5551 | map->chunk_len = ctl->chunk_size; |
| 5552 | map->stripe_size = ctl->stripe_size; |
| 5553 | map->type = type; |
| 5554 | map->io_align = BTRFS_STRIPE_LEN; |
| 5555 | map->io_width = BTRFS_STRIPE_LEN; |
| 5556 | map->sub_stripes = ctl->sub_stripes; |
| 5557 | map->num_stripes = ctl->num_stripes; |
| 5558 | |
| 5559 | for (i = 0; i < ctl->ndevs; ++i) { |
| 5560 | for (j = 0; j < ctl->dev_stripes; ++j) { |
| 5561 | int s = i * ctl->dev_stripes + j; |
| 5562 | map->stripes[s].dev = devices_info[i].dev; |
| 5563 | map->stripes[s].physical = devices_info[i].dev_offset + |
| 5564 | j * ctl->stripe_size; |
| 5565 | } |
| 5566 | } |
| 5567 | |
| 5568 | trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size); |
| 5569 | |
| 5570 | ret = btrfs_add_chunk_map(info, map); |
| 5571 | if (ret) { |
| 5572 | btrfs_free_chunk_map(map); |
| 5573 | return ERR_PTR(ret); |
| 5574 | } |
| 5575 | |
| 5576 | block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size); |
| 5577 | if (IS_ERR(block_group)) { |
| 5578 | btrfs_remove_chunk_map(info, map); |
| 5579 | return block_group; |
| 5580 | } |
| 5581 | |
| 5582 | for (int i = 0; i < map->num_stripes; i++) { |
| 5583 | struct btrfs_device *dev = map->stripes[i].dev; |
| 5584 | |
| 5585 | btrfs_device_set_bytes_used(dev, |
| 5586 | dev->bytes_used + ctl->stripe_size); |
| 5587 | if (list_empty(&dev->post_commit_list)) |
| 5588 | list_add_tail(&dev->post_commit_list, |
| 5589 | &trans->transaction->dev_update_list); |
| 5590 | } |
| 5591 | |
| 5592 | atomic64_sub(ctl->stripe_size * map->num_stripes, |
| 5593 | &info->free_chunk_space); |
| 5594 | |
| 5595 | check_raid56_incompat_flag(info, type); |
| 5596 | check_raid1c34_incompat_flag(info, type); |
| 5597 | |
| 5598 | return block_group; |
| 5599 | } |
| 5600 | |
| 5601 | struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans, |
| 5602 | u64 type) |
| 5603 | { |
| 5604 | struct btrfs_fs_info *info = trans->fs_info; |
| 5605 | struct btrfs_fs_devices *fs_devices = info->fs_devices; |
| 5606 | struct btrfs_device_info *devices_info = NULL; |
| 5607 | struct alloc_chunk_ctl ctl; |
| 5608 | struct btrfs_block_group *block_group; |
| 5609 | int ret; |
| 5610 | |
| 5611 | lockdep_assert_held(&info->chunk_mutex); |
| 5612 | |
| 5613 | if (!alloc_profile_is_valid(type, 0)) { |
| 5614 | ASSERT(0); |
| 5615 | return ERR_PTR(-EINVAL); |
| 5616 | } |
| 5617 | |
| 5618 | if (list_empty(&fs_devices->alloc_list)) { |
| 5619 | if (btrfs_test_opt(info, ENOSPC_DEBUG)) |
| 5620 | btrfs_debug(info, "%s: no writable device", __func__); |
| 5621 | return ERR_PTR(-ENOSPC); |
| 5622 | } |
| 5623 | |
| 5624 | if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { |
| 5625 | btrfs_err(info, "invalid chunk type 0x%llx requested", type); |
| 5626 | ASSERT(0); |
| 5627 | return ERR_PTR(-EINVAL); |
| 5628 | } |
| 5629 | |
| 5630 | ctl.start = find_next_chunk(info); |
| 5631 | ctl.type = type; |
| 5632 | init_alloc_chunk_ctl(fs_devices, &ctl); |
| 5633 | |
| 5634 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
| 5635 | GFP_NOFS); |
| 5636 | if (!devices_info) |
| 5637 | return ERR_PTR(-ENOMEM); |
| 5638 | |
| 5639 | ret = gather_device_info(fs_devices, &ctl, devices_info); |
| 5640 | if (ret < 0) { |
| 5641 | block_group = ERR_PTR(ret); |
| 5642 | goto out; |
| 5643 | } |
| 5644 | |
| 5645 | ret = decide_stripe_size(fs_devices, &ctl, devices_info); |
| 5646 | if (ret < 0) { |
| 5647 | block_group = ERR_PTR(ret); |
| 5648 | goto out; |
| 5649 | } |
| 5650 | |
| 5651 | block_group = create_chunk(trans, &ctl, devices_info); |
| 5652 | |
| 5653 | out: |
| 5654 | kfree(devices_info); |
| 5655 | return block_group; |
| 5656 | } |
| 5657 | |
| 5658 | /* |
| 5659 | * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the |
| 5660 | * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system |
| 5661 | * chunks. |
| 5662 | * |
| 5663 | * See the comment at btrfs_chunk_alloc() for details about the chunk allocation |
| 5664 | * phases. |
| 5665 | */ |
| 5666 | int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans, |
| 5667 | struct btrfs_block_group *bg) |
| 5668 | { |
| 5669 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 5670 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
| 5671 | struct btrfs_key key; |
| 5672 | struct btrfs_chunk *chunk; |
| 5673 | struct btrfs_stripe *stripe; |
| 5674 | struct btrfs_chunk_map *map; |
| 5675 | size_t item_size; |
| 5676 | int i; |
| 5677 | int ret; |
| 5678 | |
| 5679 | /* |
| 5680 | * We take the chunk_mutex for 2 reasons: |
| 5681 | * |
| 5682 | * 1) Updates and insertions in the chunk btree must be done while holding |
| 5683 | * the chunk_mutex, as well as updating the system chunk array in the |
| 5684 | * superblock. See the comment on top of btrfs_chunk_alloc() for the |
| 5685 | * details; |
| 5686 | * |
| 5687 | * 2) To prevent races with the final phase of a device replace operation |
| 5688 | * that replaces the device object associated with the map's stripes, |
| 5689 | * because the device object's id can change at any time during that |
| 5690 | * final phase of the device replace operation |
| 5691 | * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the |
| 5692 | * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID, |
| 5693 | * which would cause a failure when updating the device item, which does |
| 5694 | * not exists, or persisting a stripe of the chunk item with such ID. |
| 5695 | * Here we can't use the device_list_mutex because our caller already |
| 5696 | * has locked the chunk_mutex, and the final phase of device replace |
| 5697 | * acquires both mutexes - first the device_list_mutex and then the |
| 5698 | * chunk_mutex. Using any of those two mutexes protects us from a |
| 5699 | * concurrent device replace. |
| 5700 | */ |
| 5701 | lockdep_assert_held(&fs_info->chunk_mutex); |
| 5702 | |
| 5703 | map = btrfs_get_chunk_map(fs_info, bg->start, bg->length); |
| 5704 | if (IS_ERR(map)) { |
| 5705 | ret = PTR_ERR(map); |
| 5706 | btrfs_abort_transaction(trans, ret); |
| 5707 | return ret; |
| 5708 | } |
| 5709 | |
| 5710 | item_size = btrfs_chunk_item_size(map->num_stripes); |
| 5711 | |
| 5712 | chunk = kzalloc(item_size, GFP_NOFS); |
| 5713 | if (!chunk) { |
| 5714 | ret = -ENOMEM; |
| 5715 | btrfs_abort_transaction(trans, ret); |
| 5716 | goto out; |
| 5717 | } |
| 5718 | |
| 5719 | for (i = 0; i < map->num_stripes; i++) { |
| 5720 | struct btrfs_device *device = map->stripes[i].dev; |
| 5721 | |
| 5722 | ret = btrfs_update_device(trans, device); |
| 5723 | if (ret) |
| 5724 | goto out; |
| 5725 | } |
| 5726 | |
| 5727 | stripe = &chunk->stripe; |
| 5728 | for (i = 0; i < map->num_stripes; i++) { |
| 5729 | struct btrfs_device *device = map->stripes[i].dev; |
| 5730 | const u64 dev_offset = map->stripes[i].physical; |
| 5731 | |
| 5732 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
| 5733 | btrfs_set_stack_stripe_offset(stripe, dev_offset); |
| 5734 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); |
| 5735 | stripe++; |
| 5736 | } |
| 5737 | |
| 5738 | btrfs_set_stack_chunk_length(chunk, bg->length); |
| 5739 | btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID); |
| 5740 | btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN); |
| 5741 | btrfs_set_stack_chunk_type(chunk, map->type); |
| 5742 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); |
| 5743 | btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN); |
| 5744 | btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN); |
| 5745 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
| 5746 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
| 5747 | |
| 5748 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
| 5749 | key.type = BTRFS_CHUNK_ITEM_KEY; |
| 5750 | key.offset = bg->start; |
| 5751 | |
| 5752 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
| 5753 | if (ret) |
| 5754 | goto out; |
| 5755 | |
| 5756 | set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags); |
| 5757 | |
| 5758 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
| 5759 | ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size); |
| 5760 | if (ret) |
| 5761 | goto out; |
| 5762 | } |
| 5763 | |
| 5764 | out: |
| 5765 | kfree(chunk); |
| 5766 | btrfs_free_chunk_map(map); |
| 5767 | return ret; |
| 5768 | } |
| 5769 | |
| 5770 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans) |
| 5771 | { |
| 5772 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 5773 | u64 alloc_profile; |
| 5774 | struct btrfs_block_group *meta_bg; |
| 5775 | struct btrfs_block_group *sys_bg; |
| 5776 | |
| 5777 | /* |
| 5778 | * When adding a new device for sprouting, the seed device is read-only |
| 5779 | * so we must first allocate a metadata and a system chunk. But before |
| 5780 | * adding the block group items to the extent, device and chunk btrees, |
| 5781 | * we must first: |
| 5782 | * |
| 5783 | * 1) Create both chunks without doing any changes to the btrees, as |
| 5784 | * otherwise we would get -ENOSPC since the block groups from the |
| 5785 | * seed device are read-only; |
| 5786 | * |
| 5787 | * 2) Add the device item for the new sprout device - finishing the setup |
| 5788 | * of a new block group requires updating the device item in the chunk |
| 5789 | * btree, so it must exist when we attempt to do it. The previous step |
| 5790 | * ensures this does not fail with -ENOSPC. |
| 5791 | * |
| 5792 | * After that we can add the block group items to their btrees: |
| 5793 | * update existing device item in the chunk btree, add a new block group |
| 5794 | * item to the extent btree, add a new chunk item to the chunk btree and |
| 5795 | * finally add the new device extent items to the devices btree. |
| 5796 | */ |
| 5797 | |
| 5798 | alloc_profile = btrfs_metadata_alloc_profile(fs_info); |
| 5799 | meta_bg = btrfs_create_chunk(trans, alloc_profile); |
| 5800 | if (IS_ERR(meta_bg)) |
| 5801 | return PTR_ERR(meta_bg); |
| 5802 | |
| 5803 | alloc_profile = btrfs_system_alloc_profile(fs_info); |
| 5804 | sys_bg = btrfs_create_chunk(trans, alloc_profile); |
| 5805 | if (IS_ERR(sys_bg)) |
| 5806 | return PTR_ERR(sys_bg); |
| 5807 | |
| 5808 | return 0; |
| 5809 | } |
| 5810 | |
| 5811 | static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map) |
| 5812 | { |
| 5813 | const int index = btrfs_bg_flags_to_raid_index(map->type); |
| 5814 | |
| 5815 | return btrfs_raid_array[index].tolerated_failures; |
| 5816 | } |
| 5817 | |
| 5818 | bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
| 5819 | { |
| 5820 | struct btrfs_chunk_map *map; |
| 5821 | int miss_ndevs = 0; |
| 5822 | int i; |
| 5823 | bool ret = true; |
| 5824 | |
| 5825 | map = btrfs_get_chunk_map(fs_info, chunk_offset, 1); |
| 5826 | if (IS_ERR(map)) |
| 5827 | return false; |
| 5828 | |
| 5829 | for (i = 0; i < map->num_stripes; i++) { |
| 5830 | if (test_bit(BTRFS_DEV_STATE_MISSING, |
| 5831 | &map->stripes[i].dev->dev_state)) { |
| 5832 | miss_ndevs++; |
| 5833 | continue; |
| 5834 | } |
| 5835 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, |
| 5836 | &map->stripes[i].dev->dev_state)) { |
| 5837 | ret = false; |
| 5838 | goto end; |
| 5839 | } |
| 5840 | } |
| 5841 | |
| 5842 | /* |
| 5843 | * If the number of missing devices is larger than max errors, we can |
| 5844 | * not write the data into that chunk successfully. |
| 5845 | */ |
| 5846 | if (miss_ndevs > btrfs_chunk_max_errors(map)) |
| 5847 | ret = false; |
| 5848 | end: |
| 5849 | btrfs_free_chunk_map(map); |
| 5850 | return ret; |
| 5851 | } |
| 5852 | |
| 5853 | void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info) |
| 5854 | { |
| 5855 | write_lock(&fs_info->mapping_tree_lock); |
| 5856 | while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) { |
| 5857 | struct btrfs_chunk_map *map; |
| 5858 | struct rb_node *node; |
| 5859 | |
| 5860 | node = rb_first_cached(&fs_info->mapping_tree); |
| 5861 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); |
| 5862 | rb_erase_cached(&map->rb_node, &fs_info->mapping_tree); |
| 5863 | RB_CLEAR_NODE(&map->rb_node); |
| 5864 | chunk_map_device_clear_bits(map, CHUNK_ALLOCATED); |
| 5865 | /* Once for the tree ref. */ |
| 5866 | btrfs_free_chunk_map(map); |
| 5867 | cond_resched_rwlock_write(&fs_info->mapping_tree_lock); |
| 5868 | } |
| 5869 | write_unlock(&fs_info->mapping_tree_lock); |
| 5870 | } |
| 5871 | |
| 5872 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
| 5873 | { |
| 5874 | struct btrfs_chunk_map *map; |
| 5875 | enum btrfs_raid_types index; |
| 5876 | int ret = 1; |
| 5877 | |
| 5878 | map = btrfs_get_chunk_map(fs_info, logical, len); |
| 5879 | if (IS_ERR(map)) |
| 5880 | /* |
| 5881 | * We could return errors for these cases, but that could get |
| 5882 | * ugly and we'd probably do the same thing which is just not do |
| 5883 | * anything else and exit, so return 1 so the callers don't try |
| 5884 | * to use other copies. |
| 5885 | */ |
| 5886 | return 1; |
| 5887 | |
| 5888 | index = btrfs_bg_flags_to_raid_index(map->type); |
| 5889 | |
| 5890 | /* Non-RAID56, use their ncopies from btrfs_raid_array. */ |
| 5891 | if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
| 5892 | ret = btrfs_raid_array[index].ncopies; |
| 5893 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
| 5894 | ret = 2; |
| 5895 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) |
| 5896 | /* |
| 5897 | * There could be two corrupted data stripes, we need |
| 5898 | * to loop retry in order to rebuild the correct data. |
| 5899 | * |
| 5900 | * Fail a stripe at a time on every retry except the |
| 5901 | * stripe under reconstruction. |
| 5902 | */ |
| 5903 | ret = map->num_stripes; |
| 5904 | btrfs_free_chunk_map(map); |
| 5905 | return ret; |
| 5906 | } |
| 5907 | |
| 5908 | unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info, |
| 5909 | u64 logical) |
| 5910 | { |
| 5911 | struct btrfs_chunk_map *map; |
| 5912 | unsigned long len = fs_info->sectorsize; |
| 5913 | |
| 5914 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
| 5915 | return len; |
| 5916 | |
| 5917 | map = btrfs_get_chunk_map(fs_info, logical, len); |
| 5918 | |
| 5919 | if (!WARN_ON(IS_ERR(map))) { |
| 5920 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
| 5921 | len = btrfs_stripe_nr_to_offset(nr_data_stripes(map)); |
| 5922 | btrfs_free_chunk_map(map); |
| 5923 | } |
| 5924 | return len; |
| 5925 | } |
| 5926 | |
| 5927 | int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
| 5928 | { |
| 5929 | struct btrfs_chunk_map *map; |
| 5930 | int ret = 0; |
| 5931 | |
| 5932 | if (!btrfs_fs_incompat(fs_info, RAID56)) |
| 5933 | return 0; |
| 5934 | |
| 5935 | map = btrfs_get_chunk_map(fs_info, logical, len); |
| 5936 | |
| 5937 | if (!WARN_ON(IS_ERR(map))) { |
| 5938 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
| 5939 | ret = 1; |
| 5940 | btrfs_free_chunk_map(map); |
| 5941 | } |
| 5942 | return ret; |
| 5943 | } |
| 5944 | |
| 5945 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
| 5946 | struct btrfs_chunk_map *map, int first, |
| 5947 | int dev_replace_is_ongoing) |
| 5948 | { |
| 5949 | int i; |
| 5950 | int num_stripes; |
| 5951 | int preferred_mirror; |
| 5952 | int tolerance; |
| 5953 | struct btrfs_device *srcdev; |
| 5954 | |
| 5955 | ASSERT((map->type & |
| 5956 | (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10))); |
| 5957 | |
| 5958 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
| 5959 | num_stripes = map->sub_stripes; |
| 5960 | else |
| 5961 | num_stripes = map->num_stripes; |
| 5962 | |
| 5963 | switch (fs_info->fs_devices->read_policy) { |
| 5964 | default: |
| 5965 | /* Shouldn't happen, just warn and use pid instead of failing */ |
| 5966 | btrfs_warn_rl(fs_info, |
| 5967 | "unknown read_policy type %u, reset to pid", |
| 5968 | fs_info->fs_devices->read_policy); |
| 5969 | fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID; |
| 5970 | fallthrough; |
| 5971 | case BTRFS_READ_POLICY_PID: |
| 5972 | preferred_mirror = first + (current->pid % num_stripes); |
| 5973 | break; |
| 5974 | } |
| 5975 | |
| 5976 | if (dev_replace_is_ongoing && |
| 5977 | fs_info->dev_replace.cont_reading_from_srcdev_mode == |
| 5978 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) |
| 5979 | srcdev = fs_info->dev_replace.srcdev; |
| 5980 | else |
| 5981 | srcdev = NULL; |
| 5982 | |
| 5983 | /* |
| 5984 | * try to avoid the drive that is the source drive for a |
| 5985 | * dev-replace procedure, only choose it if no other non-missing |
| 5986 | * mirror is available |
| 5987 | */ |
| 5988 | for (tolerance = 0; tolerance < 2; tolerance++) { |
| 5989 | if (map->stripes[preferred_mirror].dev->bdev && |
| 5990 | (tolerance || map->stripes[preferred_mirror].dev != srcdev)) |
| 5991 | return preferred_mirror; |
| 5992 | for (i = first; i < first + num_stripes; i++) { |
| 5993 | if (map->stripes[i].dev->bdev && |
| 5994 | (tolerance || map->stripes[i].dev != srcdev)) |
| 5995 | return i; |
| 5996 | } |
| 5997 | } |
| 5998 | |
| 5999 | /* we couldn't find one that doesn't fail. Just return something |
| 6000 | * and the io error handling code will clean up eventually |
| 6001 | */ |
| 6002 | return preferred_mirror; |
| 6003 | } |
| 6004 | |
| 6005 | static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info, |
| 6006 | u64 logical, |
| 6007 | u16 total_stripes) |
| 6008 | { |
| 6009 | struct btrfs_io_context *bioc; |
| 6010 | |
| 6011 | bioc = kzalloc( |
| 6012 | /* The size of btrfs_io_context */ |
| 6013 | sizeof(struct btrfs_io_context) + |
| 6014 | /* Plus the variable array for the stripes */ |
| 6015 | sizeof(struct btrfs_io_stripe) * (total_stripes), |
| 6016 | GFP_NOFS); |
| 6017 | |
| 6018 | if (!bioc) |
| 6019 | return NULL; |
| 6020 | |
| 6021 | refcount_set(&bioc->refs, 1); |
| 6022 | |
| 6023 | bioc->fs_info = fs_info; |
| 6024 | bioc->replace_stripe_src = -1; |
| 6025 | bioc->full_stripe_logical = (u64)-1; |
| 6026 | bioc->logical = logical; |
| 6027 | |
| 6028 | return bioc; |
| 6029 | } |
| 6030 | |
| 6031 | void btrfs_get_bioc(struct btrfs_io_context *bioc) |
| 6032 | { |
| 6033 | WARN_ON(!refcount_read(&bioc->refs)); |
| 6034 | refcount_inc(&bioc->refs); |
| 6035 | } |
| 6036 | |
| 6037 | void btrfs_put_bioc(struct btrfs_io_context *bioc) |
| 6038 | { |
| 6039 | if (!bioc) |
| 6040 | return; |
| 6041 | if (refcount_dec_and_test(&bioc->refs)) |
| 6042 | kfree(bioc); |
| 6043 | } |
| 6044 | |
| 6045 | /* |
| 6046 | * Please note that, discard won't be sent to target device of device |
| 6047 | * replace. |
| 6048 | */ |
| 6049 | struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info, |
| 6050 | u64 logical, u64 *length_ret, |
| 6051 | u32 *num_stripes) |
| 6052 | { |
| 6053 | struct btrfs_chunk_map *map; |
| 6054 | struct btrfs_discard_stripe *stripes; |
| 6055 | u64 length = *length_ret; |
| 6056 | u64 offset; |
| 6057 | u32 stripe_nr; |
| 6058 | u32 stripe_nr_end; |
| 6059 | u32 stripe_cnt; |
| 6060 | u64 stripe_end_offset; |
| 6061 | u64 stripe_offset; |
| 6062 | u32 stripe_index; |
| 6063 | u32 factor = 0; |
| 6064 | u32 sub_stripes = 0; |
| 6065 | u32 stripes_per_dev = 0; |
| 6066 | u32 remaining_stripes = 0; |
| 6067 | u32 last_stripe = 0; |
| 6068 | int ret; |
| 6069 | int i; |
| 6070 | |
| 6071 | map = btrfs_get_chunk_map(fs_info, logical, length); |
| 6072 | if (IS_ERR(map)) |
| 6073 | return ERR_CAST(map); |
| 6074 | |
| 6075 | /* we don't discard raid56 yet */ |
| 6076 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
| 6077 | ret = -EOPNOTSUPP; |
| 6078 | goto out_free_map; |
| 6079 | } |
| 6080 | |
| 6081 | offset = logical - map->start; |
| 6082 | length = min_t(u64, map->start + map->chunk_len - logical, length); |
| 6083 | *length_ret = length; |
| 6084 | |
| 6085 | /* |
| 6086 | * stripe_nr counts the total number of stripes we have to stride |
| 6087 | * to get to this block |
| 6088 | */ |
| 6089 | stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; |
| 6090 | |
| 6091 | /* stripe_offset is the offset of this block in its stripe */ |
| 6092 | stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr); |
| 6093 | |
| 6094 | stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >> |
| 6095 | BTRFS_STRIPE_LEN_SHIFT; |
| 6096 | stripe_cnt = stripe_nr_end - stripe_nr; |
| 6097 | stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) - |
| 6098 | (offset + length); |
| 6099 | /* |
| 6100 | * after this, stripe_nr is the number of stripes on this |
| 6101 | * device we have to walk to find the data, and stripe_index is |
| 6102 | * the number of our device in the stripe array |
| 6103 | */ |
| 6104 | *num_stripes = 1; |
| 6105 | stripe_index = 0; |
| 6106 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
| 6107 | BTRFS_BLOCK_GROUP_RAID10)) { |
| 6108 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
| 6109 | sub_stripes = 1; |
| 6110 | else |
| 6111 | sub_stripes = map->sub_stripes; |
| 6112 | |
| 6113 | factor = map->num_stripes / sub_stripes; |
| 6114 | *num_stripes = min_t(u64, map->num_stripes, |
| 6115 | sub_stripes * stripe_cnt); |
| 6116 | stripe_index = stripe_nr % factor; |
| 6117 | stripe_nr /= factor; |
| 6118 | stripe_index *= sub_stripes; |
| 6119 | |
| 6120 | remaining_stripes = stripe_cnt % factor; |
| 6121 | stripes_per_dev = stripe_cnt / factor; |
| 6122 | last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes; |
| 6123 | } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
| 6124 | BTRFS_BLOCK_GROUP_DUP)) { |
| 6125 | *num_stripes = map->num_stripes; |
| 6126 | } else { |
| 6127 | stripe_index = stripe_nr % map->num_stripes; |
| 6128 | stripe_nr /= map->num_stripes; |
| 6129 | } |
| 6130 | |
| 6131 | stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS); |
| 6132 | if (!stripes) { |
| 6133 | ret = -ENOMEM; |
| 6134 | goto out_free_map; |
| 6135 | } |
| 6136 | |
| 6137 | for (i = 0; i < *num_stripes; i++) { |
| 6138 | stripes[i].physical = |
| 6139 | map->stripes[stripe_index].physical + |
| 6140 | stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr); |
| 6141 | stripes[i].dev = map->stripes[stripe_index].dev; |
| 6142 | |
| 6143 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
| 6144 | BTRFS_BLOCK_GROUP_RAID10)) { |
| 6145 | stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev); |
| 6146 | |
| 6147 | if (i / sub_stripes < remaining_stripes) |
| 6148 | stripes[i].length += BTRFS_STRIPE_LEN; |
| 6149 | |
| 6150 | /* |
| 6151 | * Special for the first stripe and |
| 6152 | * the last stripe: |
| 6153 | * |
| 6154 | * |-------|...|-------| |
| 6155 | * |----------| |
| 6156 | * off end_off |
| 6157 | */ |
| 6158 | if (i < sub_stripes) |
| 6159 | stripes[i].length -= stripe_offset; |
| 6160 | |
| 6161 | if (stripe_index >= last_stripe && |
| 6162 | stripe_index <= (last_stripe + |
| 6163 | sub_stripes - 1)) |
| 6164 | stripes[i].length -= stripe_end_offset; |
| 6165 | |
| 6166 | if (i == sub_stripes - 1) |
| 6167 | stripe_offset = 0; |
| 6168 | } else { |
| 6169 | stripes[i].length = length; |
| 6170 | } |
| 6171 | |
| 6172 | stripe_index++; |
| 6173 | if (stripe_index == map->num_stripes) { |
| 6174 | stripe_index = 0; |
| 6175 | stripe_nr++; |
| 6176 | } |
| 6177 | } |
| 6178 | |
| 6179 | btrfs_free_chunk_map(map); |
| 6180 | return stripes; |
| 6181 | out_free_map: |
| 6182 | btrfs_free_chunk_map(map); |
| 6183 | return ERR_PTR(ret); |
| 6184 | } |
| 6185 | |
| 6186 | static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical) |
| 6187 | { |
| 6188 | struct btrfs_block_group *cache; |
| 6189 | bool ret; |
| 6190 | |
| 6191 | /* Non zoned filesystem does not use "to_copy" flag */ |
| 6192 | if (!btrfs_is_zoned(fs_info)) |
| 6193 | return false; |
| 6194 | |
| 6195 | cache = btrfs_lookup_block_group(fs_info, logical); |
| 6196 | |
| 6197 | ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags); |
| 6198 | |
| 6199 | btrfs_put_block_group(cache); |
| 6200 | return ret; |
| 6201 | } |
| 6202 | |
| 6203 | static void handle_ops_on_dev_replace(enum btrfs_map_op op, |
| 6204 | struct btrfs_io_context *bioc, |
| 6205 | struct btrfs_dev_replace *dev_replace, |
| 6206 | u64 logical, |
| 6207 | int *num_stripes_ret, int *max_errors_ret) |
| 6208 | { |
| 6209 | u64 srcdev_devid = dev_replace->srcdev->devid; |
| 6210 | /* |
| 6211 | * At this stage, num_stripes is still the real number of stripes, |
| 6212 | * excluding the duplicated stripes. |
| 6213 | */ |
| 6214 | int num_stripes = *num_stripes_ret; |
| 6215 | int nr_extra_stripes = 0; |
| 6216 | int max_errors = *max_errors_ret; |
| 6217 | int i; |
| 6218 | |
| 6219 | /* |
| 6220 | * A block group which has "to_copy" set will eventually be copied by |
| 6221 | * the dev-replace process. We can avoid cloning IO here. |
| 6222 | */ |
| 6223 | if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical)) |
| 6224 | return; |
| 6225 | |
| 6226 | /* |
| 6227 | * Duplicate the write operations while the dev-replace procedure is |
| 6228 | * running. Since the copying of the old disk to the new disk takes |
| 6229 | * place at run time while the filesystem is mounted writable, the |
| 6230 | * regular write operations to the old disk have to be duplicated to go |
| 6231 | * to the new disk as well. |
| 6232 | * |
| 6233 | * Note that device->missing is handled by the caller, and that the |
| 6234 | * write to the old disk is already set up in the stripes array. |
| 6235 | */ |
| 6236 | for (i = 0; i < num_stripes; i++) { |
| 6237 | struct btrfs_io_stripe *old = &bioc->stripes[i]; |
| 6238 | struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes]; |
| 6239 | |
| 6240 | if (old->dev->devid != srcdev_devid) |
| 6241 | continue; |
| 6242 | |
| 6243 | new->physical = old->physical; |
| 6244 | new->dev = dev_replace->tgtdev; |
| 6245 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
| 6246 | bioc->replace_stripe_src = i; |
| 6247 | nr_extra_stripes++; |
| 6248 | } |
| 6249 | |
| 6250 | /* We can only have at most 2 extra nr_stripes (for DUP). */ |
| 6251 | ASSERT(nr_extra_stripes <= 2); |
| 6252 | /* |
| 6253 | * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for |
| 6254 | * replace. |
| 6255 | * If we have 2 extra stripes, only choose the one with smaller physical. |
| 6256 | */ |
| 6257 | if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) { |
| 6258 | struct btrfs_io_stripe *first = &bioc->stripes[num_stripes]; |
| 6259 | struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1]; |
| 6260 | |
| 6261 | /* Only DUP can have two extra stripes. */ |
| 6262 | ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP); |
| 6263 | |
| 6264 | /* |
| 6265 | * Swap the last stripe stripes and reduce @nr_extra_stripes. |
| 6266 | * The extra stripe would still be there, but won't be accessed. |
| 6267 | */ |
| 6268 | if (first->physical > second->physical) { |
| 6269 | swap(second->physical, first->physical); |
| 6270 | swap(second->dev, first->dev); |
| 6271 | nr_extra_stripes--; |
| 6272 | } |
| 6273 | } |
| 6274 | |
| 6275 | *num_stripes_ret = num_stripes + nr_extra_stripes; |
| 6276 | *max_errors_ret = max_errors + nr_extra_stripes; |
| 6277 | bioc->replace_nr_stripes = nr_extra_stripes; |
| 6278 | } |
| 6279 | |
| 6280 | static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, enum btrfs_map_op op, |
| 6281 | u64 offset, u32 *stripe_nr, u64 *stripe_offset, |
| 6282 | u64 *full_stripe_start) |
| 6283 | { |
| 6284 | /* |
| 6285 | * Stripe_nr is the stripe where this block falls. stripe_offset is |
| 6286 | * the offset of this block in its stripe. |
| 6287 | */ |
| 6288 | *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK; |
| 6289 | *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT; |
| 6290 | ASSERT(*stripe_offset < U32_MAX); |
| 6291 | |
| 6292 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
| 6293 | unsigned long full_stripe_len = |
| 6294 | btrfs_stripe_nr_to_offset(nr_data_stripes(map)); |
| 6295 | |
| 6296 | /* |
| 6297 | * For full stripe start, we use previously calculated |
| 6298 | * @stripe_nr. Align it to nr_data_stripes, then multiply with |
| 6299 | * STRIPE_LEN. |
| 6300 | * |
| 6301 | * By this we can avoid u64 division completely. And we have |
| 6302 | * to go rounddown(), not round_down(), as nr_data_stripes is |
| 6303 | * not ensured to be power of 2. |
| 6304 | */ |
| 6305 | *full_stripe_start = |
| 6306 | btrfs_stripe_nr_to_offset( |
| 6307 | rounddown(*stripe_nr, nr_data_stripes(map))); |
| 6308 | |
| 6309 | ASSERT(*full_stripe_start + full_stripe_len > offset); |
| 6310 | ASSERT(*full_stripe_start <= offset); |
| 6311 | /* |
| 6312 | * For writes to RAID56, allow to write a full stripe set, but |
| 6313 | * no straddling of stripe sets. |
| 6314 | */ |
| 6315 | if (op == BTRFS_MAP_WRITE) |
| 6316 | return full_stripe_len - (offset - *full_stripe_start); |
| 6317 | } |
| 6318 | |
| 6319 | /* |
| 6320 | * For other RAID types and for RAID56 reads, allow a single stripe (on |
| 6321 | * a single disk). |
| 6322 | */ |
| 6323 | if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK) |
| 6324 | return BTRFS_STRIPE_LEN - *stripe_offset; |
| 6325 | return U64_MAX; |
| 6326 | } |
| 6327 | |
| 6328 | static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical, |
| 6329 | u64 *length, struct btrfs_io_stripe *dst, |
| 6330 | struct btrfs_chunk_map *map, |
| 6331 | struct btrfs_io_geometry *io_geom) |
| 6332 | { |
| 6333 | dst->dev = map->stripes[io_geom->stripe_index].dev; |
| 6334 | |
| 6335 | if (io_geom->op == BTRFS_MAP_READ && |
| 6336 | btrfs_need_stripe_tree_update(fs_info, map->type)) |
| 6337 | return btrfs_get_raid_extent_offset(fs_info, logical, length, |
| 6338 | map->type, |
| 6339 | io_geom->stripe_index, dst); |
| 6340 | |
| 6341 | dst->physical = map->stripes[io_geom->stripe_index].physical + |
| 6342 | io_geom->stripe_offset + |
| 6343 | btrfs_stripe_nr_to_offset(io_geom->stripe_nr); |
| 6344 | return 0; |
| 6345 | } |
| 6346 | |
| 6347 | static bool is_single_device_io(struct btrfs_fs_info *fs_info, |
| 6348 | const struct btrfs_io_stripe *smap, |
| 6349 | const struct btrfs_chunk_map *map, |
| 6350 | int num_alloc_stripes, |
| 6351 | enum btrfs_map_op op, int mirror_num) |
| 6352 | { |
| 6353 | if (!smap) |
| 6354 | return false; |
| 6355 | |
| 6356 | if (num_alloc_stripes != 1) |
| 6357 | return false; |
| 6358 | |
| 6359 | if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ) |
| 6360 | return false; |
| 6361 | |
| 6362 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) |
| 6363 | return false; |
| 6364 | |
| 6365 | return true; |
| 6366 | } |
| 6367 | |
| 6368 | static void map_blocks_raid0(const struct btrfs_chunk_map *map, |
| 6369 | struct btrfs_io_geometry *io_geom) |
| 6370 | { |
| 6371 | io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes; |
| 6372 | io_geom->stripe_nr /= map->num_stripes; |
| 6373 | if (io_geom->op == BTRFS_MAP_READ) |
| 6374 | io_geom->mirror_num = 1; |
| 6375 | } |
| 6376 | |
| 6377 | static void map_blocks_raid1(struct btrfs_fs_info *fs_info, |
| 6378 | struct btrfs_chunk_map *map, |
| 6379 | struct btrfs_io_geometry *io_geom, |
| 6380 | bool dev_replace_is_ongoing) |
| 6381 | { |
| 6382 | if (io_geom->op != BTRFS_MAP_READ) { |
| 6383 | io_geom->num_stripes = map->num_stripes; |
| 6384 | return; |
| 6385 | } |
| 6386 | |
| 6387 | if (io_geom->mirror_num) { |
| 6388 | io_geom->stripe_index = io_geom->mirror_num - 1; |
| 6389 | return; |
| 6390 | } |
| 6391 | |
| 6392 | io_geom->stripe_index = find_live_mirror(fs_info, map, 0, |
| 6393 | dev_replace_is_ongoing); |
| 6394 | io_geom->mirror_num = io_geom->stripe_index + 1; |
| 6395 | } |
| 6396 | |
| 6397 | static void map_blocks_dup(const struct btrfs_chunk_map *map, |
| 6398 | struct btrfs_io_geometry *io_geom) |
| 6399 | { |
| 6400 | if (io_geom->op != BTRFS_MAP_READ) { |
| 6401 | io_geom->num_stripes = map->num_stripes; |
| 6402 | return; |
| 6403 | } |
| 6404 | |
| 6405 | if (io_geom->mirror_num) { |
| 6406 | io_geom->stripe_index = io_geom->mirror_num - 1; |
| 6407 | return; |
| 6408 | } |
| 6409 | |
| 6410 | io_geom->mirror_num = 1; |
| 6411 | } |
| 6412 | |
| 6413 | static void map_blocks_raid10(struct btrfs_fs_info *fs_info, |
| 6414 | struct btrfs_chunk_map *map, |
| 6415 | struct btrfs_io_geometry *io_geom, |
| 6416 | bool dev_replace_is_ongoing) |
| 6417 | { |
| 6418 | u32 factor = map->num_stripes / map->sub_stripes; |
| 6419 | int old_stripe_index; |
| 6420 | |
| 6421 | io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes; |
| 6422 | io_geom->stripe_nr /= factor; |
| 6423 | |
| 6424 | if (io_geom->op != BTRFS_MAP_READ) { |
| 6425 | io_geom->num_stripes = map->sub_stripes; |
| 6426 | return; |
| 6427 | } |
| 6428 | |
| 6429 | if (io_geom->mirror_num) { |
| 6430 | io_geom->stripe_index += io_geom->mirror_num - 1; |
| 6431 | return; |
| 6432 | } |
| 6433 | |
| 6434 | old_stripe_index = io_geom->stripe_index; |
| 6435 | io_geom->stripe_index = find_live_mirror(fs_info, map, |
| 6436 | io_geom->stripe_index, |
| 6437 | dev_replace_is_ongoing); |
| 6438 | io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1; |
| 6439 | } |
| 6440 | |
| 6441 | static void map_blocks_raid56_write(struct btrfs_chunk_map *map, |
| 6442 | struct btrfs_io_geometry *io_geom, |
| 6443 | u64 logical, u64 *length) |
| 6444 | { |
| 6445 | int data_stripes = nr_data_stripes(map); |
| 6446 | |
| 6447 | /* |
| 6448 | * Needs full stripe mapping. |
| 6449 | * |
| 6450 | * Push stripe_nr back to the start of the full stripe For those cases |
| 6451 | * needing a full stripe, @stripe_nr is the full stripe number. |
| 6452 | * |
| 6453 | * Originally we go raid56_full_stripe_start / full_stripe_len, but |
| 6454 | * that can be expensive. Here we just divide @stripe_nr with |
| 6455 | * @data_stripes. |
| 6456 | */ |
| 6457 | io_geom->stripe_nr /= data_stripes; |
| 6458 | |
| 6459 | /* RAID[56] write or recovery. Return all stripes */ |
| 6460 | io_geom->num_stripes = map->num_stripes; |
| 6461 | io_geom->max_errors = btrfs_chunk_max_errors(map); |
| 6462 | |
| 6463 | /* Return the length to the full stripe end. */ |
| 6464 | *length = min(logical + *length, |
| 6465 | io_geom->raid56_full_stripe_start + map->start + |
| 6466 | btrfs_stripe_nr_to_offset(data_stripes)) - |
| 6467 | logical; |
| 6468 | io_geom->stripe_index = 0; |
| 6469 | io_geom->stripe_offset = 0; |
| 6470 | } |
| 6471 | |
| 6472 | static void map_blocks_raid56_read(struct btrfs_chunk_map *map, |
| 6473 | struct btrfs_io_geometry *io_geom) |
| 6474 | { |
| 6475 | int data_stripes = nr_data_stripes(map); |
| 6476 | |
| 6477 | ASSERT(io_geom->mirror_num <= 1); |
| 6478 | /* Just grab the data stripe directly. */ |
| 6479 | io_geom->stripe_index = io_geom->stripe_nr % data_stripes; |
| 6480 | io_geom->stripe_nr /= data_stripes; |
| 6481 | |
| 6482 | /* We distribute the parity blocks across stripes. */ |
| 6483 | io_geom->stripe_index = |
| 6484 | (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes; |
| 6485 | |
| 6486 | if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1) |
| 6487 | io_geom->mirror_num = 1; |
| 6488 | } |
| 6489 | |
| 6490 | static void map_blocks_single(const struct btrfs_chunk_map *map, |
| 6491 | struct btrfs_io_geometry *io_geom) |
| 6492 | { |
| 6493 | io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes; |
| 6494 | io_geom->stripe_nr /= map->num_stripes; |
| 6495 | io_geom->mirror_num = io_geom->stripe_index + 1; |
| 6496 | } |
| 6497 | |
| 6498 | /* |
| 6499 | * Map one logical range to one or more physical ranges. |
| 6500 | * |
| 6501 | * @length: (Mandatory) mapped length of this run. |
| 6502 | * One logical range can be split into different segments |
| 6503 | * due to factors like zones and RAID0/5/6/10 stripe |
| 6504 | * boundaries. |
| 6505 | * |
| 6506 | * @bioc_ret: (Mandatory) returned btrfs_io_context structure. |
| 6507 | * which has one or more physical ranges (btrfs_io_stripe) |
| 6508 | * recorded inside. |
| 6509 | * Caller should call btrfs_put_bioc() to free it after use. |
| 6510 | * |
| 6511 | * @smap: (Optional) single physical range optimization. |
| 6512 | * If the map request can be fulfilled by one single |
| 6513 | * physical range, and this is parameter is not NULL, |
| 6514 | * then @bioc_ret would be NULL, and @smap would be |
| 6515 | * updated. |
| 6516 | * |
| 6517 | * @mirror_num_ret: (Mandatory) returned mirror number if the original |
| 6518 | * value is 0. |
| 6519 | * |
| 6520 | * Mirror number 0 means to choose any live mirrors. |
| 6521 | * |
| 6522 | * For non-RAID56 profiles, non-zero mirror_num means |
| 6523 | * the Nth mirror. (e.g. mirror_num 1 means the first |
| 6524 | * copy). |
| 6525 | * |
| 6526 | * For RAID56 profile, mirror 1 means rebuild from P and |
| 6527 | * the remaining data stripes. |
| 6528 | * |
| 6529 | * For RAID6 profile, mirror > 2 means mark another |
| 6530 | * data/P stripe error and rebuild from the remaining |
| 6531 | * stripes.. |
| 6532 | */ |
| 6533 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
| 6534 | u64 logical, u64 *length, |
| 6535 | struct btrfs_io_context **bioc_ret, |
| 6536 | struct btrfs_io_stripe *smap, int *mirror_num_ret) |
| 6537 | { |
| 6538 | struct btrfs_chunk_map *map; |
| 6539 | struct btrfs_io_geometry io_geom = { 0 }; |
| 6540 | u64 map_offset; |
| 6541 | int i; |
| 6542 | int ret = 0; |
| 6543 | int num_copies; |
| 6544 | struct btrfs_io_context *bioc = NULL; |
| 6545 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
| 6546 | int dev_replace_is_ongoing = 0; |
| 6547 | u16 num_alloc_stripes; |
| 6548 | u64 max_len; |
| 6549 | |
| 6550 | ASSERT(bioc_ret); |
| 6551 | |
| 6552 | io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0); |
| 6553 | io_geom.num_stripes = 1; |
| 6554 | io_geom.stripe_index = 0; |
| 6555 | io_geom.op = op; |
| 6556 | |
| 6557 | num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize); |
| 6558 | if (io_geom.mirror_num > num_copies) |
| 6559 | return -EINVAL; |
| 6560 | |
| 6561 | map = btrfs_get_chunk_map(fs_info, logical, *length); |
| 6562 | if (IS_ERR(map)) |
| 6563 | return PTR_ERR(map); |
| 6564 | |
| 6565 | map_offset = logical - map->start; |
| 6566 | io_geom.raid56_full_stripe_start = (u64)-1; |
| 6567 | max_len = btrfs_max_io_len(map, io_geom.op, map_offset, &io_geom.stripe_nr, |
| 6568 | &io_geom.stripe_offset, |
| 6569 | &io_geom.raid56_full_stripe_start); |
| 6570 | *length = min_t(u64, map->chunk_len - map_offset, max_len); |
| 6571 | |
| 6572 | down_read(&dev_replace->rwsem); |
| 6573 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
| 6574 | /* |
| 6575 | * Hold the semaphore for read during the whole operation, write is |
| 6576 | * requested at commit time but must wait. |
| 6577 | */ |
| 6578 | if (!dev_replace_is_ongoing) |
| 6579 | up_read(&dev_replace->rwsem); |
| 6580 | |
| 6581 | switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
| 6582 | case BTRFS_BLOCK_GROUP_RAID0: |
| 6583 | map_blocks_raid0(map, &io_geom); |
| 6584 | break; |
| 6585 | case BTRFS_BLOCK_GROUP_RAID1: |
| 6586 | case BTRFS_BLOCK_GROUP_RAID1C3: |
| 6587 | case BTRFS_BLOCK_GROUP_RAID1C4: |
| 6588 | map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing); |
| 6589 | break; |
| 6590 | case BTRFS_BLOCK_GROUP_DUP: |
| 6591 | map_blocks_dup(map, &io_geom); |
| 6592 | break; |
| 6593 | case BTRFS_BLOCK_GROUP_RAID10: |
| 6594 | map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing); |
| 6595 | break; |
| 6596 | case BTRFS_BLOCK_GROUP_RAID5: |
| 6597 | case BTRFS_BLOCK_GROUP_RAID6: |
| 6598 | if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1) |
| 6599 | map_blocks_raid56_write(map, &io_geom, logical, length); |
| 6600 | else |
| 6601 | map_blocks_raid56_read(map, &io_geom); |
| 6602 | break; |
| 6603 | default: |
| 6604 | /* |
| 6605 | * After this, stripe_nr is the number of stripes on this |
| 6606 | * device we have to walk to find the data, and stripe_index is |
| 6607 | * the number of our device in the stripe array |
| 6608 | */ |
| 6609 | map_blocks_single(map, &io_geom); |
| 6610 | break; |
| 6611 | } |
| 6612 | if (io_geom.stripe_index >= map->num_stripes) { |
| 6613 | btrfs_crit(fs_info, |
| 6614 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", |
| 6615 | io_geom.stripe_index, map->num_stripes); |
| 6616 | ret = -EINVAL; |
| 6617 | goto out; |
| 6618 | } |
| 6619 | |
| 6620 | num_alloc_stripes = io_geom.num_stripes; |
| 6621 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
| 6622 | op != BTRFS_MAP_READ) |
| 6623 | /* |
| 6624 | * For replace case, we need to add extra stripes for extra |
| 6625 | * duplicated stripes. |
| 6626 | * |
| 6627 | * For both WRITE and GET_READ_MIRRORS, we may have at most |
| 6628 | * 2 more stripes (DUP types, otherwise 1). |
| 6629 | */ |
| 6630 | num_alloc_stripes += 2; |
| 6631 | |
| 6632 | /* |
| 6633 | * If this I/O maps to a single device, try to return the device and |
| 6634 | * physical block information on the stack instead of allocating an |
| 6635 | * I/O context structure. |
| 6636 | */ |
| 6637 | if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op, |
| 6638 | io_geom.mirror_num)) { |
| 6639 | ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom); |
| 6640 | if (mirror_num_ret) |
| 6641 | *mirror_num_ret = io_geom.mirror_num; |
| 6642 | *bioc_ret = NULL; |
| 6643 | goto out; |
| 6644 | } |
| 6645 | |
| 6646 | bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes); |
| 6647 | if (!bioc) { |
| 6648 | ret = -ENOMEM; |
| 6649 | goto out; |
| 6650 | } |
| 6651 | bioc->map_type = map->type; |
| 6652 | |
| 6653 | /* |
| 6654 | * For RAID56 full map, we need to make sure the stripes[] follows the |
| 6655 | * rule that data stripes are all ordered, then followed with P and Q |
| 6656 | * (if we have). |
| 6657 | * |
| 6658 | * It's still mostly the same as other profiles, just with extra rotation. |
| 6659 | */ |
| 6660 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
| 6661 | (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) { |
| 6662 | /* |
| 6663 | * For RAID56 @stripe_nr is already the number of full stripes |
| 6664 | * before us, which is also the rotation value (needs to modulo |
| 6665 | * with num_stripes). |
| 6666 | * |
| 6667 | * In this case, we just add @stripe_nr with @i, then do the |
| 6668 | * modulo, to reduce one modulo call. |
| 6669 | */ |
| 6670 | bioc->full_stripe_logical = map->start + |
| 6671 | btrfs_stripe_nr_to_offset(io_geom.stripe_nr * |
| 6672 | nr_data_stripes(map)); |
| 6673 | for (int i = 0; i < io_geom.num_stripes; i++) { |
| 6674 | struct btrfs_io_stripe *dst = &bioc->stripes[i]; |
| 6675 | u32 stripe_index; |
| 6676 | |
| 6677 | stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes; |
| 6678 | dst->dev = map->stripes[stripe_index].dev; |
| 6679 | dst->physical = |
| 6680 | map->stripes[stripe_index].physical + |
| 6681 | io_geom.stripe_offset + |
| 6682 | btrfs_stripe_nr_to_offset(io_geom.stripe_nr); |
| 6683 | } |
| 6684 | } else { |
| 6685 | /* |
| 6686 | * For all other non-RAID56 profiles, just copy the target |
| 6687 | * stripe into the bioc. |
| 6688 | */ |
| 6689 | for (i = 0; i < io_geom.num_stripes; i++) { |
| 6690 | ret = set_io_stripe(fs_info, logical, length, |
| 6691 | &bioc->stripes[i], map, &io_geom); |
| 6692 | if (ret < 0) |
| 6693 | break; |
| 6694 | io_geom.stripe_index++; |
| 6695 | } |
| 6696 | } |
| 6697 | |
| 6698 | if (ret) { |
| 6699 | *bioc_ret = NULL; |
| 6700 | btrfs_put_bioc(bioc); |
| 6701 | goto out; |
| 6702 | } |
| 6703 | |
| 6704 | if (op != BTRFS_MAP_READ) |
| 6705 | io_geom.max_errors = btrfs_chunk_max_errors(map); |
| 6706 | |
| 6707 | if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL && |
| 6708 | op != BTRFS_MAP_READ) { |
| 6709 | handle_ops_on_dev_replace(op, bioc, dev_replace, logical, |
| 6710 | &io_geom.num_stripes, &io_geom.max_errors); |
| 6711 | } |
| 6712 | |
| 6713 | *bioc_ret = bioc; |
| 6714 | bioc->num_stripes = io_geom.num_stripes; |
| 6715 | bioc->max_errors = io_geom.max_errors; |
| 6716 | bioc->mirror_num = io_geom.mirror_num; |
| 6717 | |
| 6718 | out: |
| 6719 | if (dev_replace_is_ongoing) { |
| 6720 | lockdep_assert_held(&dev_replace->rwsem); |
| 6721 | /* Unlock and let waiting writers proceed */ |
| 6722 | up_read(&dev_replace->rwsem); |
| 6723 | } |
| 6724 | btrfs_free_chunk_map(map); |
| 6725 | return ret; |
| 6726 | } |
| 6727 | |
| 6728 | static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args, |
| 6729 | const struct btrfs_fs_devices *fs_devices) |
| 6730 | { |
| 6731 | if (args->fsid == NULL) |
| 6732 | return true; |
| 6733 | if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0) |
| 6734 | return true; |
| 6735 | return false; |
| 6736 | } |
| 6737 | |
| 6738 | static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args, |
| 6739 | const struct btrfs_device *device) |
| 6740 | { |
| 6741 | if (args->missing) { |
| 6742 | if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) && |
| 6743 | !device->bdev) |
| 6744 | return true; |
| 6745 | return false; |
| 6746 | } |
| 6747 | |
| 6748 | if (device->devid != args->devid) |
| 6749 | return false; |
| 6750 | if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0) |
| 6751 | return false; |
| 6752 | return true; |
| 6753 | } |
| 6754 | |
| 6755 | /* |
| 6756 | * Find a device specified by @devid or @uuid in the list of @fs_devices, or |
| 6757 | * return NULL. |
| 6758 | * |
| 6759 | * If devid and uuid are both specified, the match must be exact, otherwise |
| 6760 | * only devid is used. |
| 6761 | */ |
| 6762 | struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices, |
| 6763 | const struct btrfs_dev_lookup_args *args) |
| 6764 | { |
| 6765 | struct btrfs_device *device; |
| 6766 | struct btrfs_fs_devices *seed_devs; |
| 6767 | |
| 6768 | if (dev_args_match_fs_devices(args, fs_devices)) { |
| 6769 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
| 6770 | if (dev_args_match_device(args, device)) |
| 6771 | return device; |
| 6772 | } |
| 6773 | } |
| 6774 | |
| 6775 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
| 6776 | if (!dev_args_match_fs_devices(args, seed_devs)) |
| 6777 | continue; |
| 6778 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
| 6779 | if (dev_args_match_device(args, device)) |
| 6780 | return device; |
| 6781 | } |
| 6782 | } |
| 6783 | |
| 6784 | return NULL; |
| 6785 | } |
| 6786 | |
| 6787 | static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices, |
| 6788 | u64 devid, u8 *dev_uuid) |
| 6789 | { |
| 6790 | struct btrfs_device *device; |
| 6791 | unsigned int nofs_flag; |
| 6792 | |
| 6793 | /* |
| 6794 | * We call this under the chunk_mutex, so we want to use NOFS for this |
| 6795 | * allocation, however we don't want to change btrfs_alloc_device() to |
| 6796 | * always do NOFS because we use it in a lot of other GFP_KERNEL safe |
| 6797 | * places. |
| 6798 | */ |
| 6799 | |
| 6800 | nofs_flag = memalloc_nofs_save(); |
| 6801 | device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL); |
| 6802 | memalloc_nofs_restore(nofs_flag); |
| 6803 | if (IS_ERR(device)) |
| 6804 | return device; |
| 6805 | |
| 6806 | list_add(&device->dev_list, &fs_devices->devices); |
| 6807 | device->fs_devices = fs_devices; |
| 6808 | fs_devices->num_devices++; |
| 6809 | |
| 6810 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
| 6811 | fs_devices->missing_devices++; |
| 6812 | |
| 6813 | return device; |
| 6814 | } |
| 6815 | |
| 6816 | /* |
| 6817 | * Allocate new device struct, set up devid and UUID. |
| 6818 | * |
| 6819 | * @fs_info: used only for generating a new devid, can be NULL if |
| 6820 | * devid is provided (i.e. @devid != NULL). |
| 6821 | * @devid: a pointer to devid for this device. If NULL a new devid |
| 6822 | * is generated. |
| 6823 | * @uuid: a pointer to UUID for this device. If NULL a new UUID |
| 6824 | * is generated. |
| 6825 | * @path: a pointer to device path if available, NULL otherwise. |
| 6826 | * |
| 6827 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() |
| 6828 | * on error. Returned struct is not linked onto any lists and must be |
| 6829 | * destroyed with btrfs_free_device. |
| 6830 | */ |
| 6831 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, |
| 6832 | const u64 *devid, const u8 *uuid, |
| 6833 | const char *path) |
| 6834 | { |
| 6835 | struct btrfs_device *dev; |
| 6836 | u64 tmp; |
| 6837 | |
| 6838 | if (WARN_ON(!devid && !fs_info)) |
| 6839 | return ERR_PTR(-EINVAL); |
| 6840 | |
| 6841 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
| 6842 | if (!dev) |
| 6843 | return ERR_PTR(-ENOMEM); |
| 6844 | |
| 6845 | INIT_LIST_HEAD(&dev->dev_list); |
| 6846 | INIT_LIST_HEAD(&dev->dev_alloc_list); |
| 6847 | INIT_LIST_HEAD(&dev->post_commit_list); |
| 6848 | |
| 6849 | atomic_set(&dev->dev_stats_ccnt, 0); |
| 6850 | btrfs_device_data_ordered_init(dev); |
| 6851 | extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE); |
| 6852 | |
| 6853 | if (devid) |
| 6854 | tmp = *devid; |
| 6855 | else { |
| 6856 | int ret; |
| 6857 | |
| 6858 | ret = find_next_devid(fs_info, &tmp); |
| 6859 | if (ret) { |
| 6860 | btrfs_free_device(dev); |
| 6861 | return ERR_PTR(ret); |
| 6862 | } |
| 6863 | } |
| 6864 | dev->devid = tmp; |
| 6865 | |
| 6866 | if (uuid) |
| 6867 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); |
| 6868 | else |
| 6869 | generate_random_uuid(dev->uuid); |
| 6870 | |
| 6871 | if (path) { |
| 6872 | struct rcu_string *name; |
| 6873 | |
| 6874 | name = rcu_string_strdup(path, GFP_KERNEL); |
| 6875 | if (!name) { |
| 6876 | btrfs_free_device(dev); |
| 6877 | return ERR_PTR(-ENOMEM); |
| 6878 | } |
| 6879 | rcu_assign_pointer(dev->name, name); |
| 6880 | } |
| 6881 | |
| 6882 | return dev; |
| 6883 | } |
| 6884 | |
| 6885 | static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info, |
| 6886 | u64 devid, u8 *uuid, bool error) |
| 6887 | { |
| 6888 | if (error) |
| 6889 | btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing", |
| 6890 | devid, uuid); |
| 6891 | else |
| 6892 | btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing", |
| 6893 | devid, uuid); |
| 6894 | } |
| 6895 | |
| 6896 | u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map) |
| 6897 | { |
| 6898 | const int data_stripes = calc_data_stripes(map->type, map->num_stripes); |
| 6899 | |
| 6900 | return div_u64(map->chunk_len, data_stripes); |
| 6901 | } |
| 6902 | |
| 6903 | #if BITS_PER_LONG == 32 |
| 6904 | /* |
| 6905 | * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE |
| 6906 | * can't be accessed on 32bit systems. |
| 6907 | * |
| 6908 | * This function do mount time check to reject the fs if it already has |
| 6909 | * metadata chunk beyond that limit. |
| 6910 | */ |
| 6911 | static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info, |
| 6912 | u64 logical, u64 length, u64 type) |
| 6913 | { |
| 6914 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) |
| 6915 | return 0; |
| 6916 | |
| 6917 | if (logical + length < MAX_LFS_FILESIZE) |
| 6918 | return 0; |
| 6919 | |
| 6920 | btrfs_err_32bit_limit(fs_info); |
| 6921 | return -EOVERFLOW; |
| 6922 | } |
| 6923 | |
| 6924 | /* |
| 6925 | * This is to give early warning for any metadata chunk reaching |
| 6926 | * BTRFS_32BIT_EARLY_WARN_THRESHOLD. |
| 6927 | * Although we can still access the metadata, it's not going to be possible |
| 6928 | * once the limit is reached. |
| 6929 | */ |
| 6930 | static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info, |
| 6931 | u64 logical, u64 length, u64 type) |
| 6932 | { |
| 6933 | if (!(type & BTRFS_BLOCK_GROUP_METADATA)) |
| 6934 | return; |
| 6935 | |
| 6936 | if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD) |
| 6937 | return; |
| 6938 | |
| 6939 | btrfs_warn_32bit_limit(fs_info); |
| 6940 | } |
| 6941 | #endif |
| 6942 | |
| 6943 | static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info, |
| 6944 | u64 devid, u8 *uuid) |
| 6945 | { |
| 6946 | struct btrfs_device *dev; |
| 6947 | |
| 6948 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
| 6949 | btrfs_report_missing_device(fs_info, devid, uuid, true); |
| 6950 | return ERR_PTR(-ENOENT); |
| 6951 | } |
| 6952 | |
| 6953 | dev = add_missing_dev(fs_info->fs_devices, devid, uuid); |
| 6954 | if (IS_ERR(dev)) { |
| 6955 | btrfs_err(fs_info, "failed to init missing device %llu: %ld", |
| 6956 | devid, PTR_ERR(dev)); |
| 6957 | return dev; |
| 6958 | } |
| 6959 | btrfs_report_missing_device(fs_info, devid, uuid, false); |
| 6960 | |
| 6961 | return dev; |
| 6962 | } |
| 6963 | |
| 6964 | static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf, |
| 6965 | struct btrfs_chunk *chunk) |
| 6966 | { |
| 6967 | BTRFS_DEV_LOOKUP_ARGS(args); |
| 6968 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
| 6969 | struct btrfs_chunk_map *map; |
| 6970 | u64 logical; |
| 6971 | u64 length; |
| 6972 | u64 devid; |
| 6973 | u64 type; |
| 6974 | u8 uuid[BTRFS_UUID_SIZE]; |
| 6975 | int index; |
| 6976 | int num_stripes; |
| 6977 | int ret; |
| 6978 | int i; |
| 6979 | |
| 6980 | logical = key->offset; |
| 6981 | length = btrfs_chunk_length(leaf, chunk); |
| 6982 | type = btrfs_chunk_type(leaf, chunk); |
| 6983 | index = btrfs_bg_flags_to_raid_index(type); |
| 6984 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
| 6985 | |
| 6986 | #if BITS_PER_LONG == 32 |
| 6987 | ret = check_32bit_meta_chunk(fs_info, logical, length, type); |
| 6988 | if (ret < 0) |
| 6989 | return ret; |
| 6990 | warn_32bit_meta_chunk(fs_info, logical, length, type); |
| 6991 | #endif |
| 6992 | |
| 6993 | /* |
| 6994 | * Only need to verify chunk item if we're reading from sys chunk array, |
| 6995 | * as chunk item in tree block is already verified by tree-checker. |
| 6996 | */ |
| 6997 | if (leaf->start == BTRFS_SUPER_INFO_OFFSET) { |
| 6998 | ret = btrfs_check_chunk_valid(leaf, chunk, logical); |
| 6999 | if (ret) |
| 7000 | return ret; |
| 7001 | } |
| 7002 | |
| 7003 | map = btrfs_find_chunk_map(fs_info, logical, 1); |
| 7004 | |
| 7005 | /* already mapped? */ |
| 7006 | if (map && map->start <= logical && map->start + map->chunk_len > logical) { |
| 7007 | btrfs_free_chunk_map(map); |
| 7008 | return 0; |
| 7009 | } else if (map) { |
| 7010 | btrfs_free_chunk_map(map); |
| 7011 | } |
| 7012 | |
| 7013 | map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS); |
| 7014 | if (!map) |
| 7015 | return -ENOMEM; |
| 7016 | |
| 7017 | map->start = logical; |
| 7018 | map->chunk_len = length; |
| 7019 | map->num_stripes = num_stripes; |
| 7020 | map->io_width = btrfs_chunk_io_width(leaf, chunk); |
| 7021 | map->io_align = btrfs_chunk_io_align(leaf, chunk); |
| 7022 | map->type = type; |
| 7023 | /* |
| 7024 | * We can't use the sub_stripes value, as for profiles other than |
| 7025 | * RAID10, they may have 0 as sub_stripes for filesystems created by |
| 7026 | * older mkfs (<v5.4). |
| 7027 | * In that case, it can cause divide-by-zero errors later. |
| 7028 | * Since currently sub_stripes is fixed for each profile, let's |
| 7029 | * use the trusted value instead. |
| 7030 | */ |
| 7031 | map->sub_stripes = btrfs_raid_array[index].sub_stripes; |
| 7032 | map->verified_stripes = 0; |
| 7033 | map->stripe_size = btrfs_calc_stripe_length(map); |
| 7034 | for (i = 0; i < num_stripes; i++) { |
| 7035 | map->stripes[i].physical = |
| 7036 | btrfs_stripe_offset_nr(leaf, chunk, i); |
| 7037 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); |
| 7038 | args.devid = devid; |
| 7039 | read_extent_buffer(leaf, uuid, (unsigned long) |
| 7040 | btrfs_stripe_dev_uuid_nr(chunk, i), |
| 7041 | BTRFS_UUID_SIZE); |
| 7042 | args.uuid = uuid; |
| 7043 | map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args); |
| 7044 | if (!map->stripes[i].dev) { |
| 7045 | map->stripes[i].dev = handle_missing_device(fs_info, |
| 7046 | devid, uuid); |
| 7047 | if (IS_ERR(map->stripes[i].dev)) { |
| 7048 | ret = PTR_ERR(map->stripes[i].dev); |
| 7049 | btrfs_free_chunk_map(map); |
| 7050 | return ret; |
| 7051 | } |
| 7052 | } |
| 7053 | |
| 7054 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, |
| 7055 | &(map->stripes[i].dev->dev_state)); |
| 7056 | } |
| 7057 | |
| 7058 | ret = btrfs_add_chunk_map(fs_info, map); |
| 7059 | if (ret < 0) { |
| 7060 | btrfs_err(fs_info, |
| 7061 | "failed to add chunk map, start=%llu len=%llu: %d", |
| 7062 | map->start, map->chunk_len, ret); |
| 7063 | } |
| 7064 | |
| 7065 | return ret; |
| 7066 | } |
| 7067 | |
| 7068 | static void fill_device_from_item(struct extent_buffer *leaf, |
| 7069 | struct btrfs_dev_item *dev_item, |
| 7070 | struct btrfs_device *device) |
| 7071 | { |
| 7072 | unsigned long ptr; |
| 7073 | |
| 7074 | device->devid = btrfs_device_id(leaf, dev_item); |
| 7075 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
| 7076 | device->total_bytes = device->disk_total_bytes; |
| 7077 | device->commit_total_bytes = device->disk_total_bytes; |
| 7078 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
| 7079 | device->commit_bytes_used = device->bytes_used; |
| 7080 | device->type = btrfs_device_type(leaf, dev_item); |
| 7081 | device->io_align = btrfs_device_io_align(leaf, dev_item); |
| 7082 | device->io_width = btrfs_device_io_width(leaf, dev_item); |
| 7083 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); |
| 7084 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
| 7085 | clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state); |
| 7086 | |
| 7087 | ptr = btrfs_device_uuid(dev_item); |
| 7088 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
| 7089 | } |
| 7090 | |
| 7091 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info, |
| 7092 | u8 *fsid) |
| 7093 | { |
| 7094 | struct btrfs_fs_devices *fs_devices; |
| 7095 | int ret; |
| 7096 | |
| 7097 | lockdep_assert_held(&uuid_mutex); |
| 7098 | ASSERT(fsid); |
| 7099 | |
| 7100 | /* This will match only for multi-device seed fs */ |
| 7101 | list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list) |
| 7102 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE)) |
| 7103 | return fs_devices; |
| 7104 | |
| 7105 | |
| 7106 | fs_devices = find_fsid(fsid, NULL); |
| 7107 | if (!fs_devices) { |
| 7108 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
| 7109 | return ERR_PTR(-ENOENT); |
| 7110 | |
| 7111 | fs_devices = alloc_fs_devices(fsid); |
| 7112 | if (IS_ERR(fs_devices)) |
| 7113 | return fs_devices; |
| 7114 | |
| 7115 | fs_devices->seeding = true; |
| 7116 | fs_devices->opened = 1; |
| 7117 | return fs_devices; |
| 7118 | } |
| 7119 | |
| 7120 | /* |
| 7121 | * Upon first call for a seed fs fsid, just create a private copy of the |
| 7122 | * respective fs_devices and anchor it at fs_info->fs_devices->seed_list |
| 7123 | */ |
| 7124 | fs_devices = clone_fs_devices(fs_devices); |
| 7125 | if (IS_ERR(fs_devices)) |
| 7126 | return fs_devices; |
| 7127 | |
| 7128 | ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder); |
| 7129 | if (ret) { |
| 7130 | free_fs_devices(fs_devices); |
| 7131 | return ERR_PTR(ret); |
| 7132 | } |
| 7133 | |
| 7134 | if (!fs_devices->seeding) { |
| 7135 | close_fs_devices(fs_devices); |
| 7136 | free_fs_devices(fs_devices); |
| 7137 | return ERR_PTR(-EINVAL); |
| 7138 | } |
| 7139 | |
| 7140 | list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list); |
| 7141 | |
| 7142 | return fs_devices; |
| 7143 | } |
| 7144 | |
| 7145 | static int read_one_dev(struct extent_buffer *leaf, |
| 7146 | struct btrfs_dev_item *dev_item) |
| 7147 | { |
| 7148 | BTRFS_DEV_LOOKUP_ARGS(args); |
| 7149 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
| 7150 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 7151 | struct btrfs_device *device; |
| 7152 | u64 devid; |
| 7153 | int ret; |
| 7154 | u8 fs_uuid[BTRFS_FSID_SIZE]; |
| 7155 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
| 7156 | |
| 7157 | devid = btrfs_device_id(leaf, dev_item); |
| 7158 | args.devid = devid; |
| 7159 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
| 7160 | BTRFS_UUID_SIZE); |
| 7161 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
| 7162 | BTRFS_FSID_SIZE); |
| 7163 | args.uuid = dev_uuid; |
| 7164 | args.fsid = fs_uuid; |
| 7165 | |
| 7166 | if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) { |
| 7167 | fs_devices = open_seed_devices(fs_info, fs_uuid); |
| 7168 | if (IS_ERR(fs_devices)) |
| 7169 | return PTR_ERR(fs_devices); |
| 7170 | } |
| 7171 | |
| 7172 | device = btrfs_find_device(fs_info->fs_devices, &args); |
| 7173 | if (!device) { |
| 7174 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
| 7175 | btrfs_report_missing_device(fs_info, devid, |
| 7176 | dev_uuid, true); |
| 7177 | return -ENOENT; |
| 7178 | } |
| 7179 | |
| 7180 | device = add_missing_dev(fs_devices, devid, dev_uuid); |
| 7181 | if (IS_ERR(device)) { |
| 7182 | btrfs_err(fs_info, |
| 7183 | "failed to add missing dev %llu: %ld", |
| 7184 | devid, PTR_ERR(device)); |
| 7185 | return PTR_ERR(device); |
| 7186 | } |
| 7187 | btrfs_report_missing_device(fs_info, devid, dev_uuid, false); |
| 7188 | } else { |
| 7189 | if (!device->bdev) { |
| 7190 | if (!btrfs_test_opt(fs_info, DEGRADED)) { |
| 7191 | btrfs_report_missing_device(fs_info, |
| 7192 | devid, dev_uuid, true); |
| 7193 | return -ENOENT; |
| 7194 | } |
| 7195 | btrfs_report_missing_device(fs_info, devid, |
| 7196 | dev_uuid, false); |
| 7197 | } |
| 7198 | |
| 7199 | if (!device->bdev && |
| 7200 | !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) { |
| 7201 | /* |
| 7202 | * this happens when a device that was properly setup |
| 7203 | * in the device info lists suddenly goes bad. |
| 7204 | * device->bdev is NULL, and so we have to set |
| 7205 | * device->missing to one here |
| 7206 | */ |
| 7207 | device->fs_devices->missing_devices++; |
| 7208 | set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state); |
| 7209 | } |
| 7210 | |
| 7211 | /* Move the device to its own fs_devices */ |
| 7212 | if (device->fs_devices != fs_devices) { |
| 7213 | ASSERT(test_bit(BTRFS_DEV_STATE_MISSING, |
| 7214 | &device->dev_state)); |
| 7215 | |
| 7216 | list_move(&device->dev_list, &fs_devices->devices); |
| 7217 | device->fs_devices->num_devices--; |
| 7218 | fs_devices->num_devices++; |
| 7219 | |
| 7220 | device->fs_devices->missing_devices--; |
| 7221 | fs_devices->missing_devices++; |
| 7222 | |
| 7223 | device->fs_devices = fs_devices; |
| 7224 | } |
| 7225 | } |
| 7226 | |
| 7227 | if (device->fs_devices != fs_info->fs_devices) { |
| 7228 | BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)); |
| 7229 | if (device->generation != |
| 7230 | btrfs_device_generation(leaf, dev_item)) |
| 7231 | return -EINVAL; |
| 7232 | } |
| 7233 | |
| 7234 | fill_device_from_item(leaf, dev_item, device); |
| 7235 | if (device->bdev) { |
| 7236 | u64 max_total_bytes = bdev_nr_bytes(device->bdev); |
| 7237 | |
| 7238 | if (device->total_bytes > max_total_bytes) { |
| 7239 | btrfs_err(fs_info, |
| 7240 | "device total_bytes should be at most %llu but found %llu", |
| 7241 | max_total_bytes, device->total_bytes); |
| 7242 | return -EINVAL; |
| 7243 | } |
| 7244 | } |
| 7245 | set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state); |
| 7246 | if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) && |
| 7247 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
| 7248 | device->fs_devices->total_rw_bytes += device->total_bytes; |
| 7249 | atomic64_add(device->total_bytes - device->bytes_used, |
| 7250 | &fs_info->free_chunk_space); |
| 7251 | } |
| 7252 | ret = 0; |
| 7253 | return ret; |
| 7254 | } |
| 7255 | |
| 7256 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
| 7257 | { |
| 7258 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
| 7259 | struct extent_buffer *sb; |
| 7260 | struct btrfs_disk_key *disk_key; |
| 7261 | struct btrfs_chunk *chunk; |
| 7262 | u8 *array_ptr; |
| 7263 | unsigned long sb_array_offset; |
| 7264 | int ret = 0; |
| 7265 | u32 num_stripes; |
| 7266 | u32 array_size; |
| 7267 | u32 len = 0; |
| 7268 | u32 cur_offset; |
| 7269 | u64 type; |
| 7270 | struct btrfs_key key; |
| 7271 | |
| 7272 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
| 7273 | |
| 7274 | /* |
| 7275 | * We allocated a dummy extent, just to use extent buffer accessors. |
| 7276 | * There will be unused space after BTRFS_SUPER_INFO_SIZE, but |
| 7277 | * that's fine, we will not go beyond system chunk array anyway. |
| 7278 | */ |
| 7279 | sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET); |
| 7280 | if (!sb) |
| 7281 | return -ENOMEM; |
| 7282 | set_extent_buffer_uptodate(sb); |
| 7283 | |
| 7284 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
| 7285 | array_size = btrfs_super_sys_array_size(super_copy); |
| 7286 | |
| 7287 | array_ptr = super_copy->sys_chunk_array; |
| 7288 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); |
| 7289 | cur_offset = 0; |
| 7290 | |
| 7291 | while (cur_offset < array_size) { |
| 7292 | disk_key = (struct btrfs_disk_key *)array_ptr; |
| 7293 | len = sizeof(*disk_key); |
| 7294 | if (cur_offset + len > array_size) |
| 7295 | goto out_short_read; |
| 7296 | |
| 7297 | btrfs_disk_key_to_cpu(&key, disk_key); |
| 7298 | |
| 7299 | array_ptr += len; |
| 7300 | sb_array_offset += len; |
| 7301 | cur_offset += len; |
| 7302 | |
| 7303 | if (key.type != BTRFS_CHUNK_ITEM_KEY) { |
| 7304 | btrfs_err(fs_info, |
| 7305 | "unexpected item type %u in sys_array at offset %u", |
| 7306 | (u32)key.type, cur_offset); |
| 7307 | ret = -EIO; |
| 7308 | break; |
| 7309 | } |
| 7310 | |
| 7311 | chunk = (struct btrfs_chunk *)sb_array_offset; |
| 7312 | /* |
| 7313 | * At least one btrfs_chunk with one stripe must be present, |
| 7314 | * exact stripe count check comes afterwards |
| 7315 | */ |
| 7316 | len = btrfs_chunk_item_size(1); |
| 7317 | if (cur_offset + len > array_size) |
| 7318 | goto out_short_read; |
| 7319 | |
| 7320 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
| 7321 | if (!num_stripes) { |
| 7322 | btrfs_err(fs_info, |
| 7323 | "invalid number of stripes %u in sys_array at offset %u", |
| 7324 | num_stripes, cur_offset); |
| 7325 | ret = -EIO; |
| 7326 | break; |
| 7327 | } |
| 7328 | |
| 7329 | type = btrfs_chunk_type(sb, chunk); |
| 7330 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { |
| 7331 | btrfs_err(fs_info, |
| 7332 | "invalid chunk type %llu in sys_array at offset %u", |
| 7333 | type, cur_offset); |
| 7334 | ret = -EIO; |
| 7335 | break; |
| 7336 | } |
| 7337 | |
| 7338 | len = btrfs_chunk_item_size(num_stripes); |
| 7339 | if (cur_offset + len > array_size) |
| 7340 | goto out_short_read; |
| 7341 | |
| 7342 | ret = read_one_chunk(&key, sb, chunk); |
| 7343 | if (ret) |
| 7344 | break; |
| 7345 | |
| 7346 | array_ptr += len; |
| 7347 | sb_array_offset += len; |
| 7348 | cur_offset += len; |
| 7349 | } |
| 7350 | clear_extent_buffer_uptodate(sb); |
| 7351 | free_extent_buffer_stale(sb); |
| 7352 | return ret; |
| 7353 | |
| 7354 | out_short_read: |
| 7355 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
| 7356 | len, cur_offset); |
| 7357 | clear_extent_buffer_uptodate(sb); |
| 7358 | free_extent_buffer_stale(sb); |
| 7359 | return -EIO; |
| 7360 | } |
| 7361 | |
| 7362 | /* |
| 7363 | * Check if all chunks in the fs are OK for read-write degraded mount |
| 7364 | * |
| 7365 | * If the @failing_dev is specified, it's accounted as missing. |
| 7366 | * |
| 7367 | * Return true if all chunks meet the minimal RW mount requirements. |
| 7368 | * Return false if any chunk doesn't meet the minimal RW mount requirements. |
| 7369 | */ |
| 7370 | bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info, |
| 7371 | struct btrfs_device *failing_dev) |
| 7372 | { |
| 7373 | struct btrfs_chunk_map *map; |
| 7374 | u64 next_start; |
| 7375 | bool ret = true; |
| 7376 | |
| 7377 | map = btrfs_find_chunk_map(fs_info, 0, U64_MAX); |
| 7378 | /* No chunk at all? Return false anyway */ |
| 7379 | if (!map) { |
| 7380 | ret = false; |
| 7381 | goto out; |
| 7382 | } |
| 7383 | while (map) { |
| 7384 | int missing = 0; |
| 7385 | int max_tolerated; |
| 7386 | int i; |
| 7387 | |
| 7388 | max_tolerated = |
| 7389 | btrfs_get_num_tolerated_disk_barrier_failures( |
| 7390 | map->type); |
| 7391 | for (i = 0; i < map->num_stripes; i++) { |
| 7392 | struct btrfs_device *dev = map->stripes[i].dev; |
| 7393 | |
| 7394 | if (!dev || !dev->bdev || |
| 7395 | test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) || |
| 7396 | dev->last_flush_error) |
| 7397 | missing++; |
| 7398 | else if (failing_dev && failing_dev == dev) |
| 7399 | missing++; |
| 7400 | } |
| 7401 | if (missing > max_tolerated) { |
| 7402 | if (!failing_dev) |
| 7403 | btrfs_warn(fs_info, |
| 7404 | "chunk %llu missing %d devices, max tolerance is %d for writable mount", |
| 7405 | map->start, missing, max_tolerated); |
| 7406 | btrfs_free_chunk_map(map); |
| 7407 | ret = false; |
| 7408 | goto out; |
| 7409 | } |
| 7410 | next_start = map->start + map->chunk_len; |
| 7411 | btrfs_free_chunk_map(map); |
| 7412 | |
| 7413 | map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start); |
| 7414 | } |
| 7415 | out: |
| 7416 | return ret; |
| 7417 | } |
| 7418 | |
| 7419 | static void readahead_tree_node_children(struct extent_buffer *node) |
| 7420 | { |
| 7421 | int i; |
| 7422 | const int nr_items = btrfs_header_nritems(node); |
| 7423 | |
| 7424 | for (i = 0; i < nr_items; i++) |
| 7425 | btrfs_readahead_node_child(node, i); |
| 7426 | } |
| 7427 | |
| 7428 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
| 7429 | { |
| 7430 | struct btrfs_root *root = fs_info->chunk_root; |
| 7431 | struct btrfs_path *path; |
| 7432 | struct extent_buffer *leaf; |
| 7433 | struct btrfs_key key; |
| 7434 | struct btrfs_key found_key; |
| 7435 | int ret; |
| 7436 | int slot; |
| 7437 | int iter_ret = 0; |
| 7438 | u64 total_dev = 0; |
| 7439 | u64 last_ra_node = 0; |
| 7440 | |
| 7441 | path = btrfs_alloc_path(); |
| 7442 | if (!path) |
| 7443 | return -ENOMEM; |
| 7444 | |
| 7445 | /* |
| 7446 | * uuid_mutex is needed only if we are mounting a sprout FS |
| 7447 | * otherwise we don't need it. |
| 7448 | */ |
| 7449 | mutex_lock(&uuid_mutex); |
| 7450 | |
| 7451 | /* |
| 7452 | * It is possible for mount and umount to race in such a way that |
| 7453 | * we execute this code path, but open_fs_devices failed to clear |
| 7454 | * total_rw_bytes. We certainly want it cleared before reading the |
| 7455 | * device items, so clear it here. |
| 7456 | */ |
| 7457 | fs_info->fs_devices->total_rw_bytes = 0; |
| 7458 | |
| 7459 | /* |
| 7460 | * Lockdep complains about possible circular locking dependency between |
| 7461 | * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores |
| 7462 | * used for freeze procection of a fs (struct super_block.s_writers), |
| 7463 | * which we take when starting a transaction, and extent buffers of the |
| 7464 | * chunk tree if we call read_one_dev() while holding a lock on an |
| 7465 | * extent buffer of the chunk tree. Since we are mounting the filesystem |
| 7466 | * and at this point there can't be any concurrent task modifying the |
| 7467 | * chunk tree, to keep it simple, just skip locking on the chunk tree. |
| 7468 | */ |
| 7469 | ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags)); |
| 7470 | path->skip_locking = 1; |
| 7471 | |
| 7472 | /* |
| 7473 | * Read all device items, and then all the chunk items. All |
| 7474 | * device items are found before any chunk item (their object id |
| 7475 | * is smaller than the lowest possible object id for a chunk |
| 7476 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). |
| 7477 | */ |
| 7478 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
| 7479 | key.offset = 0; |
| 7480 | key.type = 0; |
| 7481 | btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) { |
| 7482 | struct extent_buffer *node = path->nodes[1]; |
| 7483 | |
| 7484 | leaf = path->nodes[0]; |
| 7485 | slot = path->slots[0]; |
| 7486 | |
| 7487 | if (node) { |
| 7488 | if (last_ra_node != node->start) { |
| 7489 | readahead_tree_node_children(node); |
| 7490 | last_ra_node = node->start; |
| 7491 | } |
| 7492 | } |
| 7493 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
| 7494 | struct btrfs_dev_item *dev_item; |
| 7495 | dev_item = btrfs_item_ptr(leaf, slot, |
| 7496 | struct btrfs_dev_item); |
| 7497 | ret = read_one_dev(leaf, dev_item); |
| 7498 | if (ret) |
| 7499 | goto error; |
| 7500 | total_dev++; |
| 7501 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
| 7502 | struct btrfs_chunk *chunk; |
| 7503 | |
| 7504 | /* |
| 7505 | * We are only called at mount time, so no need to take |
| 7506 | * fs_info->chunk_mutex. Plus, to avoid lockdep warnings, |
| 7507 | * we always lock first fs_info->chunk_mutex before |
| 7508 | * acquiring any locks on the chunk tree. This is a |
| 7509 | * requirement for chunk allocation, see the comment on |
| 7510 | * top of btrfs_chunk_alloc() for details. |
| 7511 | */ |
| 7512 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
| 7513 | ret = read_one_chunk(&found_key, leaf, chunk); |
| 7514 | if (ret) |
| 7515 | goto error; |
| 7516 | } |
| 7517 | } |
| 7518 | /* Catch error found during iteration */ |
| 7519 | if (iter_ret < 0) { |
| 7520 | ret = iter_ret; |
| 7521 | goto error; |
| 7522 | } |
| 7523 | |
| 7524 | /* |
| 7525 | * After loading chunk tree, we've got all device information, |
| 7526 | * do another round of validation checks. |
| 7527 | */ |
| 7528 | if (total_dev != fs_info->fs_devices->total_devices) { |
| 7529 | btrfs_warn(fs_info, |
| 7530 | "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit", |
| 7531 | btrfs_super_num_devices(fs_info->super_copy), |
| 7532 | total_dev); |
| 7533 | fs_info->fs_devices->total_devices = total_dev; |
| 7534 | btrfs_set_super_num_devices(fs_info->super_copy, total_dev); |
| 7535 | } |
| 7536 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
| 7537 | fs_info->fs_devices->total_rw_bytes) { |
| 7538 | btrfs_err(fs_info, |
| 7539 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
| 7540 | btrfs_super_total_bytes(fs_info->super_copy), |
| 7541 | fs_info->fs_devices->total_rw_bytes); |
| 7542 | ret = -EINVAL; |
| 7543 | goto error; |
| 7544 | } |
| 7545 | ret = 0; |
| 7546 | error: |
| 7547 | mutex_unlock(&uuid_mutex); |
| 7548 | |
| 7549 | btrfs_free_path(path); |
| 7550 | return ret; |
| 7551 | } |
| 7552 | |
| 7553 | int btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
| 7554 | { |
| 7555 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
| 7556 | struct btrfs_device *device; |
| 7557 | int ret = 0; |
| 7558 | |
| 7559 | fs_devices->fs_info = fs_info; |
| 7560 | |
| 7561 | mutex_lock(&fs_devices->device_list_mutex); |
| 7562 | list_for_each_entry(device, &fs_devices->devices, dev_list) |
| 7563 | device->fs_info = fs_info; |
| 7564 | |
| 7565 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
| 7566 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
| 7567 | device->fs_info = fs_info; |
| 7568 | ret = btrfs_get_dev_zone_info(device, false); |
| 7569 | if (ret) |
| 7570 | break; |
| 7571 | } |
| 7572 | |
| 7573 | seed_devs->fs_info = fs_info; |
| 7574 | } |
| 7575 | mutex_unlock(&fs_devices->device_list_mutex); |
| 7576 | |
| 7577 | return ret; |
| 7578 | } |
| 7579 | |
| 7580 | static u64 btrfs_dev_stats_value(const struct extent_buffer *eb, |
| 7581 | const struct btrfs_dev_stats_item *ptr, |
| 7582 | int index) |
| 7583 | { |
| 7584 | u64 val; |
| 7585 | |
| 7586 | read_extent_buffer(eb, &val, |
| 7587 | offsetof(struct btrfs_dev_stats_item, values) + |
| 7588 | ((unsigned long)ptr) + (index * sizeof(u64)), |
| 7589 | sizeof(val)); |
| 7590 | return val; |
| 7591 | } |
| 7592 | |
| 7593 | static void btrfs_set_dev_stats_value(struct extent_buffer *eb, |
| 7594 | struct btrfs_dev_stats_item *ptr, |
| 7595 | int index, u64 val) |
| 7596 | { |
| 7597 | write_extent_buffer(eb, &val, |
| 7598 | offsetof(struct btrfs_dev_stats_item, values) + |
| 7599 | ((unsigned long)ptr) + (index * sizeof(u64)), |
| 7600 | sizeof(val)); |
| 7601 | } |
| 7602 | |
| 7603 | static int btrfs_device_init_dev_stats(struct btrfs_device *device, |
| 7604 | struct btrfs_path *path) |
| 7605 | { |
| 7606 | struct btrfs_dev_stats_item *ptr; |
| 7607 | struct extent_buffer *eb; |
| 7608 | struct btrfs_key key; |
| 7609 | int item_size; |
| 7610 | int i, ret, slot; |
| 7611 | |
| 7612 | if (!device->fs_info->dev_root) |
| 7613 | return 0; |
| 7614 | |
| 7615 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
| 7616 | key.type = BTRFS_PERSISTENT_ITEM_KEY; |
| 7617 | key.offset = device->devid; |
| 7618 | ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0); |
| 7619 | if (ret) { |
| 7620 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
| 7621 | btrfs_dev_stat_set(device, i, 0); |
| 7622 | device->dev_stats_valid = 1; |
| 7623 | btrfs_release_path(path); |
| 7624 | return ret < 0 ? ret : 0; |
| 7625 | } |
| 7626 | slot = path->slots[0]; |
| 7627 | eb = path->nodes[0]; |
| 7628 | item_size = btrfs_item_size(eb, slot); |
| 7629 | |
| 7630 | ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item); |
| 7631 | |
| 7632 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
| 7633 | if (item_size >= (1 + i) * sizeof(__le64)) |
| 7634 | btrfs_dev_stat_set(device, i, |
| 7635 | btrfs_dev_stats_value(eb, ptr, i)); |
| 7636 | else |
| 7637 | btrfs_dev_stat_set(device, i, 0); |
| 7638 | } |
| 7639 | |
| 7640 | device->dev_stats_valid = 1; |
| 7641 | btrfs_dev_stat_print_on_load(device); |
| 7642 | btrfs_release_path(path); |
| 7643 | |
| 7644 | return 0; |
| 7645 | } |
| 7646 | |
| 7647 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) |
| 7648 | { |
| 7649 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs; |
| 7650 | struct btrfs_device *device; |
| 7651 | struct btrfs_path *path = NULL; |
| 7652 | int ret = 0; |
| 7653 | |
| 7654 | path = btrfs_alloc_path(); |
| 7655 | if (!path) |
| 7656 | return -ENOMEM; |
| 7657 | |
| 7658 | mutex_lock(&fs_devices->device_list_mutex); |
| 7659 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
| 7660 | ret = btrfs_device_init_dev_stats(device, path); |
| 7661 | if (ret) |
| 7662 | goto out; |
| 7663 | } |
| 7664 | list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) { |
| 7665 | list_for_each_entry(device, &seed_devs->devices, dev_list) { |
| 7666 | ret = btrfs_device_init_dev_stats(device, path); |
| 7667 | if (ret) |
| 7668 | goto out; |
| 7669 | } |
| 7670 | } |
| 7671 | out: |
| 7672 | mutex_unlock(&fs_devices->device_list_mutex); |
| 7673 | |
| 7674 | btrfs_free_path(path); |
| 7675 | return ret; |
| 7676 | } |
| 7677 | |
| 7678 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, |
| 7679 | struct btrfs_device *device) |
| 7680 | { |
| 7681 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 7682 | struct btrfs_root *dev_root = fs_info->dev_root; |
| 7683 | struct btrfs_path *path; |
| 7684 | struct btrfs_key key; |
| 7685 | struct extent_buffer *eb; |
| 7686 | struct btrfs_dev_stats_item *ptr; |
| 7687 | int ret; |
| 7688 | int i; |
| 7689 | |
| 7690 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
| 7691 | key.type = BTRFS_PERSISTENT_ITEM_KEY; |
| 7692 | key.offset = device->devid; |
| 7693 | |
| 7694 | path = btrfs_alloc_path(); |
| 7695 | if (!path) |
| 7696 | return -ENOMEM; |
| 7697 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); |
| 7698 | if (ret < 0) { |
| 7699 | btrfs_warn_in_rcu(fs_info, |
| 7700 | "error %d while searching for dev_stats item for device %s", |
| 7701 | ret, btrfs_dev_name(device)); |
| 7702 | goto out; |
| 7703 | } |
| 7704 | |
| 7705 | if (ret == 0 && |
| 7706 | btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { |
| 7707 | /* need to delete old one and insert a new one */ |
| 7708 | ret = btrfs_del_item(trans, dev_root, path); |
| 7709 | if (ret != 0) { |
| 7710 | btrfs_warn_in_rcu(fs_info, |
| 7711 | "delete too small dev_stats item for device %s failed %d", |
| 7712 | btrfs_dev_name(device), ret); |
| 7713 | goto out; |
| 7714 | } |
| 7715 | ret = 1; |
| 7716 | } |
| 7717 | |
| 7718 | if (ret == 1) { |
| 7719 | /* need to insert a new item */ |
| 7720 | btrfs_release_path(path); |
| 7721 | ret = btrfs_insert_empty_item(trans, dev_root, path, |
| 7722 | &key, sizeof(*ptr)); |
| 7723 | if (ret < 0) { |
| 7724 | btrfs_warn_in_rcu(fs_info, |
| 7725 | "insert dev_stats item for device %s failed %d", |
| 7726 | btrfs_dev_name(device), ret); |
| 7727 | goto out; |
| 7728 | } |
| 7729 | } |
| 7730 | |
| 7731 | eb = path->nodes[0]; |
| 7732 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); |
| 7733 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
| 7734 | btrfs_set_dev_stats_value(eb, ptr, i, |
| 7735 | btrfs_dev_stat_read(device, i)); |
| 7736 | btrfs_mark_buffer_dirty(trans, eb); |
| 7737 | |
| 7738 | out: |
| 7739 | btrfs_free_path(path); |
| 7740 | return ret; |
| 7741 | } |
| 7742 | |
| 7743 | /* |
| 7744 | * called from commit_transaction. Writes all changed device stats to disk. |
| 7745 | */ |
| 7746 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans) |
| 7747 | { |
| 7748 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 7749 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 7750 | struct btrfs_device *device; |
| 7751 | int stats_cnt; |
| 7752 | int ret = 0; |
| 7753 | |
| 7754 | mutex_lock(&fs_devices->device_list_mutex); |
| 7755 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
| 7756 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
| 7757 | if (!device->dev_stats_valid || stats_cnt == 0) |
| 7758 | continue; |
| 7759 | |
| 7760 | |
| 7761 | /* |
| 7762 | * There is a LOAD-LOAD control dependency between the value of |
| 7763 | * dev_stats_ccnt and updating the on-disk values which requires |
| 7764 | * reading the in-memory counters. Such control dependencies |
| 7765 | * require explicit read memory barriers. |
| 7766 | * |
| 7767 | * This memory barriers pairs with smp_mb__before_atomic in |
| 7768 | * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full |
| 7769 | * barrier implied by atomic_xchg in |
| 7770 | * btrfs_dev_stats_read_and_reset |
| 7771 | */ |
| 7772 | smp_rmb(); |
| 7773 | |
| 7774 | ret = update_dev_stat_item(trans, device); |
| 7775 | if (!ret) |
| 7776 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
| 7777 | } |
| 7778 | mutex_unlock(&fs_devices->device_list_mutex); |
| 7779 | |
| 7780 | return ret; |
| 7781 | } |
| 7782 | |
| 7783 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
| 7784 | { |
| 7785 | btrfs_dev_stat_inc(dev, index); |
| 7786 | |
| 7787 | if (!dev->dev_stats_valid) |
| 7788 | return; |
| 7789 | btrfs_err_rl_in_rcu(dev->fs_info, |
| 7790 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
| 7791 | btrfs_dev_name(dev), |
| 7792 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
| 7793 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), |
| 7794 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), |
| 7795 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
| 7796 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); |
| 7797 | } |
| 7798 | |
| 7799 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
| 7800 | { |
| 7801 | int i; |
| 7802 | |
| 7803 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
| 7804 | if (btrfs_dev_stat_read(dev, i) != 0) |
| 7805 | break; |
| 7806 | if (i == BTRFS_DEV_STAT_VALUES_MAX) |
| 7807 | return; /* all values == 0, suppress message */ |
| 7808 | |
| 7809 | btrfs_info_in_rcu(dev->fs_info, |
| 7810 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
| 7811 | btrfs_dev_name(dev), |
| 7812 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
| 7813 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), |
| 7814 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), |
| 7815 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
| 7816 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); |
| 7817 | } |
| 7818 | |
| 7819 | int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info, |
| 7820 | struct btrfs_ioctl_get_dev_stats *stats) |
| 7821 | { |
| 7822 | BTRFS_DEV_LOOKUP_ARGS(args); |
| 7823 | struct btrfs_device *dev; |
| 7824 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 7825 | int i; |
| 7826 | |
| 7827 | mutex_lock(&fs_devices->device_list_mutex); |
| 7828 | args.devid = stats->devid; |
| 7829 | dev = btrfs_find_device(fs_info->fs_devices, &args); |
| 7830 | mutex_unlock(&fs_devices->device_list_mutex); |
| 7831 | |
| 7832 | if (!dev) { |
| 7833 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
| 7834 | return -ENODEV; |
| 7835 | } else if (!dev->dev_stats_valid) { |
| 7836 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
| 7837 | return -ENODEV; |
| 7838 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
| 7839 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
| 7840 | if (stats->nr_items > i) |
| 7841 | stats->values[i] = |
| 7842 | btrfs_dev_stat_read_and_reset(dev, i); |
| 7843 | else |
| 7844 | btrfs_dev_stat_set(dev, i, 0); |
| 7845 | } |
| 7846 | btrfs_info(fs_info, "device stats zeroed by %s (%d)", |
| 7847 | current->comm, task_pid_nr(current)); |
| 7848 | } else { |
| 7849 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) |
| 7850 | if (stats->nr_items > i) |
| 7851 | stats->values[i] = btrfs_dev_stat_read(dev, i); |
| 7852 | } |
| 7853 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) |
| 7854 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; |
| 7855 | return 0; |
| 7856 | } |
| 7857 | |
| 7858 | /* |
| 7859 | * Update the size and bytes used for each device where it changed. This is |
| 7860 | * delayed since we would otherwise get errors while writing out the |
| 7861 | * superblocks. |
| 7862 | * |
| 7863 | * Must be invoked during transaction commit. |
| 7864 | */ |
| 7865 | void btrfs_commit_device_sizes(struct btrfs_transaction *trans) |
| 7866 | { |
| 7867 | struct btrfs_device *curr, *next; |
| 7868 | |
| 7869 | ASSERT(trans->state == TRANS_STATE_COMMIT_DOING); |
| 7870 | |
| 7871 | if (list_empty(&trans->dev_update_list)) |
| 7872 | return; |
| 7873 | |
| 7874 | /* |
| 7875 | * We don't need the device_list_mutex here. This list is owned by the |
| 7876 | * transaction and the transaction must complete before the device is |
| 7877 | * released. |
| 7878 | */ |
| 7879 | mutex_lock(&trans->fs_info->chunk_mutex); |
| 7880 | list_for_each_entry_safe(curr, next, &trans->dev_update_list, |
| 7881 | post_commit_list) { |
| 7882 | list_del_init(&curr->post_commit_list); |
| 7883 | curr->commit_total_bytes = curr->disk_total_bytes; |
| 7884 | curr->commit_bytes_used = curr->bytes_used; |
| 7885 | } |
| 7886 | mutex_unlock(&trans->fs_info->chunk_mutex); |
| 7887 | } |
| 7888 | |
| 7889 | /* |
| 7890 | * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10. |
| 7891 | */ |
| 7892 | int btrfs_bg_type_to_factor(u64 flags) |
| 7893 | { |
| 7894 | const int index = btrfs_bg_flags_to_raid_index(flags); |
| 7895 | |
| 7896 | return btrfs_raid_array[index].ncopies; |
| 7897 | } |
| 7898 | |
| 7899 | |
| 7900 | |
| 7901 | static int verify_one_dev_extent(struct btrfs_fs_info *fs_info, |
| 7902 | u64 chunk_offset, u64 devid, |
| 7903 | u64 physical_offset, u64 physical_len) |
| 7904 | { |
| 7905 | struct btrfs_dev_lookup_args args = { .devid = devid }; |
| 7906 | struct btrfs_chunk_map *map; |
| 7907 | struct btrfs_device *dev; |
| 7908 | u64 stripe_len; |
| 7909 | bool found = false; |
| 7910 | int ret = 0; |
| 7911 | int i; |
| 7912 | |
| 7913 | map = btrfs_find_chunk_map(fs_info, chunk_offset, 1); |
| 7914 | if (!map) { |
| 7915 | btrfs_err(fs_info, |
| 7916 | "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk", |
| 7917 | physical_offset, devid); |
| 7918 | ret = -EUCLEAN; |
| 7919 | goto out; |
| 7920 | } |
| 7921 | |
| 7922 | stripe_len = btrfs_calc_stripe_length(map); |
| 7923 | if (physical_len != stripe_len) { |
| 7924 | btrfs_err(fs_info, |
| 7925 | "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu", |
| 7926 | physical_offset, devid, map->start, physical_len, |
| 7927 | stripe_len); |
| 7928 | ret = -EUCLEAN; |
| 7929 | goto out; |
| 7930 | } |
| 7931 | |
| 7932 | /* |
| 7933 | * Very old mkfs.btrfs (before v4.1) will not respect the reserved |
| 7934 | * space. Although kernel can handle it without problem, better to warn |
| 7935 | * the users. |
| 7936 | */ |
| 7937 | if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED) |
| 7938 | btrfs_warn(fs_info, |
| 7939 | "devid %llu physical %llu len %llu inside the reserved space", |
| 7940 | devid, physical_offset, physical_len); |
| 7941 | |
| 7942 | for (i = 0; i < map->num_stripes; i++) { |
| 7943 | if (map->stripes[i].dev->devid == devid && |
| 7944 | map->stripes[i].physical == physical_offset) { |
| 7945 | found = true; |
| 7946 | if (map->verified_stripes >= map->num_stripes) { |
| 7947 | btrfs_err(fs_info, |
| 7948 | "too many dev extents for chunk %llu found", |
| 7949 | map->start); |
| 7950 | ret = -EUCLEAN; |
| 7951 | goto out; |
| 7952 | } |
| 7953 | map->verified_stripes++; |
| 7954 | break; |
| 7955 | } |
| 7956 | } |
| 7957 | if (!found) { |
| 7958 | btrfs_err(fs_info, |
| 7959 | "dev extent physical offset %llu devid %llu has no corresponding chunk", |
| 7960 | physical_offset, devid); |
| 7961 | ret = -EUCLEAN; |
| 7962 | } |
| 7963 | |
| 7964 | /* Make sure no dev extent is beyond device boundary */ |
| 7965 | dev = btrfs_find_device(fs_info->fs_devices, &args); |
| 7966 | if (!dev) { |
| 7967 | btrfs_err(fs_info, "failed to find devid %llu", devid); |
| 7968 | ret = -EUCLEAN; |
| 7969 | goto out; |
| 7970 | } |
| 7971 | |
| 7972 | if (physical_offset + physical_len > dev->disk_total_bytes) { |
| 7973 | btrfs_err(fs_info, |
| 7974 | "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu", |
| 7975 | devid, physical_offset, physical_len, |
| 7976 | dev->disk_total_bytes); |
| 7977 | ret = -EUCLEAN; |
| 7978 | goto out; |
| 7979 | } |
| 7980 | |
| 7981 | if (dev->zone_info) { |
| 7982 | u64 zone_size = dev->zone_info->zone_size; |
| 7983 | |
| 7984 | if (!IS_ALIGNED(physical_offset, zone_size) || |
| 7985 | !IS_ALIGNED(physical_len, zone_size)) { |
| 7986 | btrfs_err(fs_info, |
| 7987 | "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone", |
| 7988 | devid, physical_offset, physical_len); |
| 7989 | ret = -EUCLEAN; |
| 7990 | goto out; |
| 7991 | } |
| 7992 | } |
| 7993 | |
| 7994 | out: |
| 7995 | btrfs_free_chunk_map(map); |
| 7996 | return ret; |
| 7997 | } |
| 7998 | |
| 7999 | static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info) |
| 8000 | { |
| 8001 | struct rb_node *node; |
| 8002 | int ret = 0; |
| 8003 | |
| 8004 | read_lock(&fs_info->mapping_tree_lock); |
| 8005 | for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) { |
| 8006 | struct btrfs_chunk_map *map; |
| 8007 | |
| 8008 | map = rb_entry(node, struct btrfs_chunk_map, rb_node); |
| 8009 | if (map->num_stripes != map->verified_stripes) { |
| 8010 | btrfs_err(fs_info, |
| 8011 | "chunk %llu has missing dev extent, have %d expect %d", |
| 8012 | map->start, map->verified_stripes, map->num_stripes); |
| 8013 | ret = -EUCLEAN; |
| 8014 | goto out; |
| 8015 | } |
| 8016 | } |
| 8017 | out: |
| 8018 | read_unlock(&fs_info->mapping_tree_lock); |
| 8019 | return ret; |
| 8020 | } |
| 8021 | |
| 8022 | /* |
| 8023 | * Ensure that all dev extents are mapped to correct chunk, otherwise |
| 8024 | * later chunk allocation/free would cause unexpected behavior. |
| 8025 | * |
| 8026 | * NOTE: This will iterate through the whole device tree, which should be of |
| 8027 | * the same size level as the chunk tree. This slightly increases mount time. |
| 8028 | */ |
| 8029 | int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info) |
| 8030 | { |
| 8031 | struct btrfs_path *path; |
| 8032 | struct btrfs_root *root = fs_info->dev_root; |
| 8033 | struct btrfs_key key; |
| 8034 | u64 prev_devid = 0; |
| 8035 | u64 prev_dev_ext_end = 0; |
| 8036 | int ret = 0; |
| 8037 | |
| 8038 | /* |
| 8039 | * We don't have a dev_root because we mounted with ignorebadroots and |
| 8040 | * failed to load the root, so we want to skip the verification in this |
| 8041 | * case for sure. |
| 8042 | * |
| 8043 | * However if the dev root is fine, but the tree itself is corrupted |
| 8044 | * we'd still fail to mount. This verification is only to make sure |
| 8045 | * writes can happen safely, so instead just bypass this check |
| 8046 | * completely in the case of IGNOREBADROOTS. |
| 8047 | */ |
| 8048 | if (btrfs_test_opt(fs_info, IGNOREBADROOTS)) |
| 8049 | return 0; |
| 8050 | |
| 8051 | key.objectid = 1; |
| 8052 | key.type = BTRFS_DEV_EXTENT_KEY; |
| 8053 | key.offset = 0; |
| 8054 | |
| 8055 | path = btrfs_alloc_path(); |
| 8056 | if (!path) |
| 8057 | return -ENOMEM; |
| 8058 | |
| 8059 | path->reada = READA_FORWARD; |
| 8060 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 8061 | if (ret < 0) |
| 8062 | goto out; |
| 8063 | |
| 8064 | if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { |
| 8065 | ret = btrfs_next_leaf(root, path); |
| 8066 | if (ret < 0) |
| 8067 | goto out; |
| 8068 | /* No dev extents at all? Not good */ |
| 8069 | if (ret > 0) { |
| 8070 | ret = -EUCLEAN; |
| 8071 | goto out; |
| 8072 | } |
| 8073 | } |
| 8074 | while (1) { |
| 8075 | struct extent_buffer *leaf = path->nodes[0]; |
| 8076 | struct btrfs_dev_extent *dext; |
| 8077 | int slot = path->slots[0]; |
| 8078 | u64 chunk_offset; |
| 8079 | u64 physical_offset; |
| 8080 | u64 physical_len; |
| 8081 | u64 devid; |
| 8082 | |
| 8083 | btrfs_item_key_to_cpu(leaf, &key, slot); |
| 8084 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
| 8085 | break; |
| 8086 | devid = key.objectid; |
| 8087 | physical_offset = key.offset; |
| 8088 | |
| 8089 | dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent); |
| 8090 | chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext); |
| 8091 | physical_len = btrfs_dev_extent_length(leaf, dext); |
| 8092 | |
| 8093 | /* Check if this dev extent overlaps with the previous one */ |
| 8094 | if (devid == prev_devid && physical_offset < prev_dev_ext_end) { |
| 8095 | btrfs_err(fs_info, |
| 8096 | "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu", |
| 8097 | devid, physical_offset, prev_dev_ext_end); |
| 8098 | ret = -EUCLEAN; |
| 8099 | goto out; |
| 8100 | } |
| 8101 | |
| 8102 | ret = verify_one_dev_extent(fs_info, chunk_offset, devid, |
| 8103 | physical_offset, physical_len); |
| 8104 | if (ret < 0) |
| 8105 | goto out; |
| 8106 | prev_devid = devid; |
| 8107 | prev_dev_ext_end = physical_offset + physical_len; |
| 8108 | |
| 8109 | ret = btrfs_next_item(root, path); |
| 8110 | if (ret < 0) |
| 8111 | goto out; |
| 8112 | if (ret > 0) { |
| 8113 | ret = 0; |
| 8114 | break; |
| 8115 | } |
| 8116 | } |
| 8117 | |
| 8118 | /* Ensure all chunks have corresponding dev extents */ |
| 8119 | ret = verify_chunk_dev_extent_mapping(fs_info); |
| 8120 | out: |
| 8121 | btrfs_free_path(path); |
| 8122 | return ret; |
| 8123 | } |
| 8124 | |
| 8125 | /* |
| 8126 | * Check whether the given block group or device is pinned by any inode being |
| 8127 | * used as a swapfile. |
| 8128 | */ |
| 8129 | bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr) |
| 8130 | { |
| 8131 | struct btrfs_swapfile_pin *sp; |
| 8132 | struct rb_node *node; |
| 8133 | |
| 8134 | spin_lock(&fs_info->swapfile_pins_lock); |
| 8135 | node = fs_info->swapfile_pins.rb_node; |
| 8136 | while (node) { |
| 8137 | sp = rb_entry(node, struct btrfs_swapfile_pin, node); |
| 8138 | if (ptr < sp->ptr) |
| 8139 | node = node->rb_left; |
| 8140 | else if (ptr > sp->ptr) |
| 8141 | node = node->rb_right; |
| 8142 | else |
| 8143 | break; |
| 8144 | } |
| 8145 | spin_unlock(&fs_info->swapfile_pins_lock); |
| 8146 | return node != NULL; |
| 8147 | } |
| 8148 | |
| 8149 | static int relocating_repair_kthread(void *data) |
| 8150 | { |
| 8151 | struct btrfs_block_group *cache = data; |
| 8152 | struct btrfs_fs_info *fs_info = cache->fs_info; |
| 8153 | u64 target; |
| 8154 | int ret = 0; |
| 8155 | |
| 8156 | target = cache->start; |
| 8157 | btrfs_put_block_group(cache); |
| 8158 | |
| 8159 | sb_start_write(fs_info->sb); |
| 8160 | if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) { |
| 8161 | btrfs_info(fs_info, |
| 8162 | "zoned: skip relocating block group %llu to repair: EBUSY", |
| 8163 | target); |
| 8164 | sb_end_write(fs_info->sb); |
| 8165 | return -EBUSY; |
| 8166 | } |
| 8167 | |
| 8168 | mutex_lock(&fs_info->reclaim_bgs_lock); |
| 8169 | |
| 8170 | /* Ensure block group still exists */ |
| 8171 | cache = btrfs_lookup_block_group(fs_info, target); |
| 8172 | if (!cache) |
| 8173 | goto out; |
| 8174 | |
| 8175 | if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) |
| 8176 | goto out; |
| 8177 | |
| 8178 | ret = btrfs_may_alloc_data_chunk(fs_info, target); |
| 8179 | if (ret < 0) |
| 8180 | goto out; |
| 8181 | |
| 8182 | btrfs_info(fs_info, |
| 8183 | "zoned: relocating block group %llu to repair IO failure", |
| 8184 | target); |
| 8185 | ret = btrfs_relocate_chunk(fs_info, target); |
| 8186 | |
| 8187 | out: |
| 8188 | if (cache) |
| 8189 | btrfs_put_block_group(cache); |
| 8190 | mutex_unlock(&fs_info->reclaim_bgs_lock); |
| 8191 | btrfs_exclop_finish(fs_info); |
| 8192 | sb_end_write(fs_info->sb); |
| 8193 | |
| 8194 | return ret; |
| 8195 | } |
| 8196 | |
| 8197 | bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical) |
| 8198 | { |
| 8199 | struct btrfs_block_group *cache; |
| 8200 | |
| 8201 | if (!btrfs_is_zoned(fs_info)) |
| 8202 | return false; |
| 8203 | |
| 8204 | /* Do not attempt to repair in degraded state */ |
| 8205 | if (btrfs_test_opt(fs_info, DEGRADED)) |
| 8206 | return true; |
| 8207 | |
| 8208 | cache = btrfs_lookup_block_group(fs_info, logical); |
| 8209 | if (!cache) |
| 8210 | return true; |
| 8211 | |
| 8212 | if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) { |
| 8213 | btrfs_put_block_group(cache); |
| 8214 | return true; |
| 8215 | } |
| 8216 | |
| 8217 | kthread_run(relocating_repair_kthread, cache, |
| 8218 | "btrfs-relocating-repair"); |
| 8219 | |
| 8220 | return true; |
| 8221 | } |
| 8222 | |
| 8223 | static void map_raid56_repair_block(struct btrfs_io_context *bioc, |
| 8224 | struct btrfs_io_stripe *smap, |
| 8225 | u64 logical) |
| 8226 | { |
| 8227 | int data_stripes = nr_bioc_data_stripes(bioc); |
| 8228 | int i; |
| 8229 | |
| 8230 | for (i = 0; i < data_stripes; i++) { |
| 8231 | u64 stripe_start = bioc->full_stripe_logical + |
| 8232 | btrfs_stripe_nr_to_offset(i); |
| 8233 | |
| 8234 | if (logical >= stripe_start && |
| 8235 | logical < stripe_start + BTRFS_STRIPE_LEN) |
| 8236 | break; |
| 8237 | } |
| 8238 | ASSERT(i < data_stripes); |
| 8239 | smap->dev = bioc->stripes[i].dev; |
| 8240 | smap->physical = bioc->stripes[i].physical + |
| 8241 | ((logical - bioc->full_stripe_logical) & |
| 8242 | BTRFS_STRIPE_LEN_MASK); |
| 8243 | } |
| 8244 | |
| 8245 | /* |
| 8246 | * Map a repair write into a single device. |
| 8247 | * |
| 8248 | * A repair write is triggered by read time repair or scrub, which would only |
| 8249 | * update the contents of a single device. |
| 8250 | * Not update any other mirrors nor go through RMW path. |
| 8251 | * |
| 8252 | * Callers should ensure: |
| 8253 | * |
| 8254 | * - Call btrfs_bio_counter_inc_blocked() first |
| 8255 | * - The range does not cross stripe boundary |
| 8256 | * - Has a valid @mirror_num passed in. |
| 8257 | */ |
| 8258 | int btrfs_map_repair_block(struct btrfs_fs_info *fs_info, |
| 8259 | struct btrfs_io_stripe *smap, u64 logical, |
| 8260 | u32 length, int mirror_num) |
| 8261 | { |
| 8262 | struct btrfs_io_context *bioc = NULL; |
| 8263 | u64 map_length = length; |
| 8264 | int mirror_ret = mirror_num; |
| 8265 | int ret; |
| 8266 | |
| 8267 | ASSERT(mirror_num > 0); |
| 8268 | |
| 8269 | ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length, |
| 8270 | &bioc, smap, &mirror_ret); |
| 8271 | if (ret < 0) |
| 8272 | return ret; |
| 8273 | |
| 8274 | /* The map range should not cross stripe boundary. */ |
| 8275 | ASSERT(map_length >= length); |
| 8276 | |
| 8277 | /* Already mapped to single stripe. */ |
| 8278 | if (!bioc) |
| 8279 | goto out; |
| 8280 | |
| 8281 | /* Map the RAID56 multi-stripe writes to a single one. */ |
| 8282 | if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
| 8283 | map_raid56_repair_block(bioc, smap, logical); |
| 8284 | goto out; |
| 8285 | } |
| 8286 | |
| 8287 | ASSERT(mirror_num <= bioc->num_stripes); |
| 8288 | smap->dev = bioc->stripes[mirror_num - 1].dev; |
| 8289 | smap->physical = bioc->stripes[mirror_num - 1].physical; |
| 8290 | out: |
| 8291 | btrfs_put_bioc(bioc); |
| 8292 | ASSERT(smap->dev); |
| 8293 | return 0; |
| 8294 | } |