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 "disk-io.h"
18 #include "transaction.h"
19 #include "volumes.h"
20 #include "raid56.h"
21 #include "rcu-string.h"
22 #include "dev-replace.h"
23 #include "sysfs.h"
24 #include "tree-checker.h"
25 #include "space-info.h"
26 #include "block-group.h"
27 #include "discard.h"
28 #include "zoned.h"
29 #include "fs.h"
30 #include "accessors.h"
31 #include "uuid-tree.h"
32 #include "ioctl.h"
33 #include "relocation.h"
34 #include "scrub.h"
35 #include "super.h"
36 #include "raid-stripe-tree.h"
37
38 #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
39 BTRFS_BLOCK_GROUP_RAID10 | \
40 BTRFS_BLOCK_GROUP_RAID56_MASK)
41
42 struct btrfs_io_geometry {
43 u32 stripe_index;
44 u32 stripe_nr;
45 int mirror_num;
46 int num_stripes;
47 u64 stripe_offset;
48 u64 raid56_full_stripe_start;
49 int max_errors;
50 enum btrfs_map_op op;
51 };
52
53 const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
54 [BTRFS_RAID_RAID10] = {
55 .sub_stripes = 2,
56 .dev_stripes = 1,
57 .devs_max = 0, /* 0 == as many as possible */
58 .devs_min = 2,
59 .tolerated_failures = 1,
60 .devs_increment = 2,
61 .ncopies = 2,
62 .nparity = 0,
63 .raid_name = "raid10",
64 .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
65 .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
66 },
67 [BTRFS_RAID_RAID1] = {
68 .sub_stripes = 1,
69 .dev_stripes = 1,
70 .devs_max = 2,
71 .devs_min = 2,
72 .tolerated_failures = 1,
73 .devs_increment = 2,
74 .ncopies = 2,
75 .nparity = 0,
76 .raid_name = "raid1",
77 .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
78 .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
79 },
80 [BTRFS_RAID_RAID1C3] = {
81 .sub_stripes = 1,
82 .dev_stripes = 1,
83 .devs_max = 3,
84 .devs_min = 3,
85 .tolerated_failures = 2,
86 .devs_increment = 3,
87 .ncopies = 3,
88 .nparity = 0,
89 .raid_name = "raid1c3",
90 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
91 .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
92 },
93 [BTRFS_RAID_RAID1C4] = {
94 .sub_stripes = 1,
95 .dev_stripes = 1,
96 .devs_max = 4,
97 .devs_min = 4,
98 .tolerated_failures = 3,
99 .devs_increment = 4,
100 .ncopies = 4,
101 .nparity = 0,
102 .raid_name = "raid1c4",
103 .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
104 .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
105 },
106 [BTRFS_RAID_DUP] = {
107 .sub_stripes = 1,
108 .dev_stripes = 2,
109 .devs_max = 1,
110 .devs_min = 1,
111 .tolerated_failures = 0,
112 .devs_increment = 1,
113 .ncopies = 2,
114 .nparity = 0,
115 .raid_name = "dup",
116 .bg_flag = BTRFS_BLOCK_GROUP_DUP,
117 .mindev_error = 0,
118 },
119 [BTRFS_RAID_RAID0] = {
120 .sub_stripes = 1,
121 .dev_stripes = 1,
122 .devs_max = 0,
123 .devs_min = 1,
124 .tolerated_failures = 0,
125 .devs_increment = 1,
126 .ncopies = 1,
127 .nparity = 0,
128 .raid_name = "raid0",
129 .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
130 .mindev_error = 0,
131 },
132 [BTRFS_RAID_SINGLE] = {
133 .sub_stripes = 1,
134 .dev_stripes = 1,
135 .devs_max = 1,
136 .devs_min = 1,
137 .tolerated_failures = 0,
138 .devs_increment = 1,
139 .ncopies = 1,
140 .nparity = 0,
141 .raid_name = "single",
142 .bg_flag = 0,
143 .mindev_error = 0,
144 },
145 [BTRFS_RAID_RAID5] = {
146 .sub_stripes = 1,
147 .dev_stripes = 1,
148 .devs_max = 0,
149 .devs_min = 2,
150 .tolerated_failures = 1,
151 .devs_increment = 1,
152 .ncopies = 1,
153 .nparity = 1,
154 .raid_name = "raid5",
155 .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
156 .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
157 },
158 [BTRFS_RAID_RAID6] = {
159 .sub_stripes = 1,
160 .dev_stripes = 1,
161 .devs_max = 0,
162 .devs_min = 3,
163 .tolerated_failures = 2,
164 .devs_increment = 1,
165 .ncopies = 1,
166 .nparity = 2,
167 .raid_name = "raid6",
168 .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
169 .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
170 },
171 };
172
173 /*
174 * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
175 * can be used as index to access btrfs_raid_array[].
176 */
177 enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
178 {
179 const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
180
181 if (!profile)
182 return BTRFS_RAID_SINGLE;
183
184 return BTRFS_BG_FLAG_TO_INDEX(profile);
185 }
186
187 const char *btrfs_bg_type_to_raid_name(u64 flags)
188 {
189 const int index = btrfs_bg_flags_to_raid_index(flags);
190
191 if (index >= BTRFS_NR_RAID_TYPES)
192 return NULL;
193
194 return btrfs_raid_array[index].raid_name;
195 }
196
197 int btrfs_nr_parity_stripes(u64 type)
198 {
199 enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
200
201 return btrfs_raid_array[index].nparity;
202 }
203
204 /*
205 * Fill @buf with textual description of @bg_flags, no more than @size_buf
206 * bytes including terminating null byte.
207 */
208 void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
209 {
210 int i;
211 int ret;
212 char *bp = buf;
213 u64 flags = bg_flags;
214 u32 size_bp = size_buf;
215
216 if (!flags) {
217 strcpy(bp, "NONE");
218 return;
219 }
220
221 #define DESCRIBE_FLAG(flag, desc) \
222 do { \
223 if (flags & (flag)) { \
224 ret = snprintf(bp, size_bp, "%s|", (desc)); \
225 if (ret < 0 || ret >= size_bp) \
226 goto out_overflow; \
227 size_bp -= ret; \
228 bp += ret; \
229 flags &= ~(flag); \
230 } \
231 } while (0)
232
233 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
234 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
235 DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
236
237 DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
238 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
239 DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
240 btrfs_raid_array[i].raid_name);
241 #undef DESCRIBE_FLAG
242
243 if (flags) {
244 ret = snprintf(bp, size_bp, "0x%llx|", flags);
245 size_bp -= ret;
246 }
247
248 if (size_bp < size_buf)
249 buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
250
251 /*
252 * The text is trimmed, it's up to the caller to provide sufficiently
253 * large buffer
254 */
255 out_overflow:;
256 }
257
258 static int init_first_rw_device(struct btrfs_trans_handle *trans);
259 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
260 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
261
262 /*
263 * Device locking
264 * ==============
265 *
266 * There are several mutexes that protect manipulation of devices and low-level
267 * structures like chunks but not block groups, extents or files
268 *
269 * uuid_mutex (global lock)
270 * ------------------------
271 * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
272 * the SCAN_DEV ioctl registration or from mount either implicitly (the first
273 * device) or requested by the device= mount option
274 *
275 * the mutex can be very coarse and can cover long-running operations
276 *
277 * protects: updates to fs_devices counters like missing devices, rw devices,
278 * seeding, structure cloning, opening/closing devices at mount/umount time
279 *
280 * global::fs_devs - add, remove, updates to the global list
281 *
282 * does not protect: manipulation of the fs_devices::devices list in general
283 * but in mount context it could be used to exclude list modifications by eg.
284 * scan ioctl
285 *
286 * btrfs_device::name - renames (write side), read is RCU
287 *
288 * fs_devices::device_list_mutex (per-fs, with RCU)
289 * ------------------------------------------------
290 * protects updates to fs_devices::devices, ie. adding and deleting
291 *
292 * simple list traversal with read-only actions can be done with RCU protection
293 *
294 * may be used to exclude some operations from running concurrently without any
295 * modifications to the list (see write_all_supers)
296 *
297 * Is not required at mount and close times, because our device list is
298 * protected by the uuid_mutex at that point.
299 *
300 * balance_mutex
301 * -------------
302 * protects balance structures (status, state) and context accessed from
303 * several places (internally, ioctl)
304 *
305 * chunk_mutex
306 * -----------
307 * protects chunks, adding or removing during allocation, trim or when a new
308 * device is added/removed. Additionally it also protects post_commit_list of
309 * individual devices, since they can be added to the transaction's
310 * post_commit_list only with chunk_mutex held.
311 *
312 * cleaner_mutex
313 * -------------
314 * a big lock that is held by the cleaner thread and prevents running subvolume
315 * cleaning together with relocation or delayed iputs
316 *
317 *
318 * Lock nesting
319 * ============
320 *
321 * uuid_mutex
322 * device_list_mutex
323 * chunk_mutex
324 * balance_mutex
325 *
326 *
327 * Exclusive operations
328 * ====================
329 *
330 * Maintains the exclusivity of the following operations that apply to the
331 * whole filesystem and cannot run in parallel.
332 *
333 * - Balance (*)
334 * - Device add
335 * - Device remove
336 * - Device replace (*)
337 * - Resize
338 *
339 * The device operations (as above) can be in one of the following states:
340 *
341 * - Running state
342 * - Paused state
343 * - Completed state
344 *
345 * Only device operations marked with (*) can go into the Paused state for the
346 * following reasons:
347 *
348 * - ioctl (only Balance can be Paused through ioctl)
349 * - filesystem remounted as read-only
350 * - filesystem unmounted and mounted as read-only
351 * - system power-cycle and filesystem mounted as read-only
352 * - filesystem or device errors leading to forced read-only
353 *
354 * The status of exclusive operation is set and cleared atomically.
355 * During the course of Paused state, fs_info::exclusive_operation remains set.
356 * A device operation in Paused or Running state can be canceled or resumed
357 * either by ioctl (Balance only) or when remounted as read-write.
358 * The exclusive status is cleared when the device operation is canceled or
359 * completed.
360 */
361
362 DEFINE_MUTEX(uuid_mutex);
363 static LIST_HEAD(fs_uuids);
364 struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
365 {
366 return &fs_uuids;
367 }
368
369 /*
370 * Allocate new btrfs_fs_devices structure identified by a fsid.
371 *
372 * @fsid: if not NULL, copy the UUID to fs_devices::fsid and to
373 * fs_devices::metadata_fsid
374 *
375 * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
376 * The returned struct is not linked onto any lists and can be destroyed with
377 * kfree() right away.
378 */
379 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
380 {
381 struct btrfs_fs_devices *fs_devs;
382
383 fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
384 if (!fs_devs)
385 return ERR_PTR(-ENOMEM);
386
387 mutex_init(&fs_devs->device_list_mutex);
388
389 INIT_LIST_HEAD(&fs_devs->devices);
390 INIT_LIST_HEAD(&fs_devs->alloc_list);
391 INIT_LIST_HEAD(&fs_devs->fs_list);
392 INIT_LIST_HEAD(&fs_devs->seed_list);
393
394 if (fsid) {
395 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
396 memcpy(fs_devs->metadata_uuid, fsid, BTRFS_FSID_SIZE);
397 }
398
399 return fs_devs;
400 }
401
402 static void btrfs_free_device(struct btrfs_device *device)
403 {
404 WARN_ON(!list_empty(&device->post_commit_list));
405 rcu_string_free(device->name);
406 extent_io_tree_release(&device->alloc_state);
407 btrfs_destroy_dev_zone_info(device);
408 kfree(device);
409 }
410
411 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
412 {
413 struct btrfs_device *device;
414
415 WARN_ON(fs_devices->opened);
416 while (!list_empty(&fs_devices->devices)) {
417 device = list_entry(fs_devices->devices.next,
418 struct btrfs_device, dev_list);
419 list_del(&device->dev_list);
420 btrfs_free_device(device);
421 }
422 kfree(fs_devices);
423 }
424
425 void __exit btrfs_cleanup_fs_uuids(void)
426 {
427 struct btrfs_fs_devices *fs_devices;
428
429 while (!list_empty(&fs_uuids)) {
430 fs_devices = list_entry(fs_uuids.next,
431 struct btrfs_fs_devices, fs_list);
432 list_del(&fs_devices->fs_list);
433 free_fs_devices(fs_devices);
434 }
435 }
436
437 static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
438 const u8 *fsid, const u8 *metadata_fsid)
439 {
440 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
441 return false;
442
443 if (!metadata_fsid)
444 return true;
445
446 if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
447 return false;
448
449 return true;
450 }
451
452 static noinline struct btrfs_fs_devices *find_fsid(
453 const u8 *fsid, const u8 *metadata_fsid)
454 {
455 struct btrfs_fs_devices *fs_devices;
456
457 ASSERT(fsid);
458
459 /* Handle non-split brain cases */
460 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
461 if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
462 return fs_devices;
463 }
464 return NULL;
465 }
466
467 static int
468 btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
469 int flush, struct file **bdev_file,
470 struct btrfs_super_block **disk_super)
471 {
472 struct block_device *bdev;
473 int ret;
474
475 *bdev_file = bdev_file_open_by_path(device_path, flags, holder, NULL);
476
477 if (IS_ERR(*bdev_file)) {
478 ret = PTR_ERR(*bdev_file);
479 goto error;
480 }
481 bdev = file_bdev(*bdev_file);
482
483 if (flush)
484 sync_blockdev(bdev);
485 if (holder) {
486 ret = set_blocksize(*bdev_file, BTRFS_BDEV_BLOCKSIZE);
487 if (ret) {
488 fput(*bdev_file);
489 goto error;
490 }
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 fput(*bdev_file);
497 goto error;
498 }
499
500 return 0;
501
502 error:
503 *disk_super = NULL;
504 *bdev_file = NULL;
505 return ret;
506 }
507
508 /*
509 * Search and remove all stale devices (which are not mounted). When both
510 * inputs are NULL, it will search and release all stale devices.
511 *
512 * @devt: Optional. When provided will it release all unmounted devices
513 * matching this devt only.
514 * @skip_device: Optional. Will skip this device when searching for the stale
515 * devices.
516 *
517 * Return: 0 for success or if @devt is 0.
518 * -EBUSY if @devt is a mounted device.
519 * -ENOENT if @devt does not match any device in the list.
520 */
521 static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
522 {
523 struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
524 struct btrfs_device *device, *tmp_device;
525 int ret;
526 bool freed = false;
527
528 lockdep_assert_held(&uuid_mutex);
529
530 /* Return good status if there is no instance of devt. */
531 ret = 0;
532 list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
533
534 mutex_lock(&fs_devices->device_list_mutex);
535 list_for_each_entry_safe(device, tmp_device,
536 &fs_devices->devices, dev_list) {
537 if (skip_device && skip_device == device)
538 continue;
539 if (devt && devt != device->devt)
540 continue;
541 if (fs_devices->opened) {
542 if (devt)
543 ret = -EBUSY;
544 break;
545 }
546
547 /* delete the stale device */
548 fs_devices->num_devices--;
549 list_del(&device->dev_list);
550 btrfs_free_device(device);
551
552 freed = true;
553 }
554 mutex_unlock(&fs_devices->device_list_mutex);
555
556 if (fs_devices->num_devices == 0) {
557 btrfs_sysfs_remove_fsid(fs_devices);
558 list_del(&fs_devices->fs_list);
559 free_fs_devices(fs_devices);
560 }
561 }
562
563 /* If there is at least one freed device return 0. */
564 if (freed)
565 return 0;
566
567 return ret;
568 }
569
570 static struct btrfs_fs_devices *find_fsid_by_device(
571 struct btrfs_super_block *disk_super,
572 dev_t devt, bool *same_fsid_diff_dev)
573 {
574 struct btrfs_fs_devices *fsid_fs_devices;
575 struct btrfs_fs_devices *devt_fs_devices;
576 const bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
577 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
578 bool found_by_devt = false;
579
580 /* Find the fs_device by the usual method, if found use it. */
581 fsid_fs_devices = find_fsid(disk_super->fsid,
582 has_metadata_uuid ? disk_super->metadata_uuid : NULL);
583
584 /* The temp_fsid feature is supported only with single device filesystem. */
585 if (btrfs_super_num_devices(disk_super) != 1)
586 return fsid_fs_devices;
587
588 /*
589 * A seed device is an integral component of the sprout device, which
590 * functions as a multi-device filesystem. So, temp-fsid feature is
591 * not supported.
592 */
593 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING)
594 return fsid_fs_devices;
595
596 /* Try to find a fs_devices by matching devt. */
597 list_for_each_entry(devt_fs_devices, &fs_uuids, fs_list) {
598 struct btrfs_device *device;
599
600 list_for_each_entry(device, &devt_fs_devices->devices, dev_list) {
601 if (device->devt == devt) {
602 found_by_devt = true;
603 break;
604 }
605 }
606 if (found_by_devt)
607 break;
608 }
609
610 if (found_by_devt) {
611 /* Existing device. */
612 if (fsid_fs_devices == NULL) {
613 if (devt_fs_devices->opened == 0) {
614 /* Stale device. */
615 return NULL;
616 } else {
617 /* temp_fsid is mounting a subvol. */
618 return devt_fs_devices;
619 }
620 } else {
621 /* Regular or temp_fsid device mounting a subvol. */
622 return devt_fs_devices;
623 }
624 } else {
625 /* New device. */
626 if (fsid_fs_devices == NULL) {
627 return NULL;
628 } else {
629 /* sb::fsid is already used create a new temp_fsid. */
630 *same_fsid_diff_dev = true;
631 return NULL;
632 }
633 }
634
635 /* Not reached. */
636 }
637
638 /*
639 * This is only used on mount, and we are protected from competing things
640 * messing with our fs_devices by the uuid_mutex, thus we do not need the
641 * fs_devices->device_list_mutex here.
642 */
643 static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
644 struct btrfs_device *device, blk_mode_t flags,
645 void *holder)
646 {
647 struct file *bdev_file;
648 struct btrfs_super_block *disk_super;
649 u64 devid;
650 int ret;
651
652 if (device->bdev)
653 return -EINVAL;
654 if (!device->name)
655 return -EINVAL;
656
657 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
658 &bdev_file, &disk_super);
659 if (ret)
660 return ret;
661
662 devid = btrfs_stack_device_id(&disk_super->dev_item);
663 if (devid != device->devid)
664 goto error_free_page;
665
666 if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
667 goto error_free_page;
668
669 device->generation = btrfs_super_generation(disk_super);
670
671 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
672 if (btrfs_super_incompat_flags(disk_super) &
673 BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
674 pr_err(
675 "BTRFS: Invalid seeding and uuid-changed device detected\n");
676 goto error_free_page;
677 }
678
679 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
680 fs_devices->seeding = true;
681 } else {
682 if (bdev_read_only(file_bdev(bdev_file)))
683 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
684 else
685 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
686 }
687
688 if (!bdev_nonrot(file_bdev(bdev_file)))
689 fs_devices->rotating = true;
690
691 if (bdev_max_discard_sectors(file_bdev(bdev_file)))
692 fs_devices->discardable = true;
693
694 device->bdev_file = bdev_file;
695 device->bdev = file_bdev(bdev_file);
696 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
697
698 if (device->devt != device->bdev->bd_dev) {
699 btrfs_warn(NULL,
700 "device %s maj:min changed from %d:%d to %d:%d",
701 device->name->str, MAJOR(device->devt),
702 MINOR(device->devt), MAJOR(device->bdev->bd_dev),
703 MINOR(device->bdev->bd_dev));
704
705 device->devt = device->bdev->bd_dev;
706 }
707
708 fs_devices->open_devices++;
709 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
710 device->devid != BTRFS_DEV_REPLACE_DEVID) {
711 fs_devices->rw_devices++;
712 list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
713 }
714 btrfs_release_disk_super(disk_super);
715
716 return 0;
717
718 error_free_page:
719 btrfs_release_disk_super(disk_super);
720 fput(bdev_file);
721
722 return -EINVAL;
723 }
724
725 const u8 *btrfs_sb_fsid_ptr(const struct btrfs_super_block *sb)
726 {
727 bool has_metadata_uuid = (btrfs_super_incompat_flags(sb) &
728 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
729
730 return has_metadata_uuid ? sb->metadata_uuid : sb->fsid;
731 }
732
733 /*
734 * Add new device to list of registered devices
735 *
736 * Returns:
737 * device pointer which was just added or updated when successful
738 * error pointer when failed
739 */
740 static noinline struct btrfs_device *device_list_add(const char *path,
741 struct btrfs_super_block *disk_super,
742 bool *new_device_added)
743 {
744 struct btrfs_device *device;
745 struct btrfs_fs_devices *fs_devices = NULL;
746 struct rcu_string *name;
747 u64 found_transid = btrfs_super_generation(disk_super);
748 u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
749 dev_t path_devt;
750 int error;
751 bool same_fsid_diff_dev = false;
752 bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
753 BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
754
755 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_CHANGING_FSID_V2) {
756 btrfs_err(NULL,
757 "device %s has incomplete metadata_uuid change, please use btrfstune to complete",
758 path);
759 return ERR_PTR(-EAGAIN);
760 }
761
762 error = lookup_bdev(path, &path_devt);
763 if (error) {
764 btrfs_err(NULL, "failed to lookup block device for path %s: %d",
765 path, error);
766 return ERR_PTR(error);
767 }
768
769 fs_devices = find_fsid_by_device(disk_super, path_devt, &same_fsid_diff_dev);
770
771 if (!fs_devices) {
772 fs_devices = alloc_fs_devices(disk_super->fsid);
773 if (IS_ERR(fs_devices))
774 return ERR_CAST(fs_devices);
775
776 if (has_metadata_uuid)
777 memcpy(fs_devices->metadata_uuid,
778 disk_super->metadata_uuid, BTRFS_FSID_SIZE);
779
780 if (same_fsid_diff_dev) {
781 generate_random_uuid(fs_devices->fsid);
782 fs_devices->temp_fsid = true;
783 pr_info("BTRFS: device %s (%d:%d) using temp-fsid %pU\n",
784 path, MAJOR(path_devt), MINOR(path_devt),
785 fs_devices->fsid);
786 }
787
788 mutex_lock(&fs_devices->device_list_mutex);
789 list_add(&fs_devices->fs_list, &fs_uuids);
790
791 device = NULL;
792 } else {
793 struct btrfs_dev_lookup_args args = {
794 .devid = devid,
795 .uuid = disk_super->dev_item.uuid,
796 };
797
798 mutex_lock(&fs_devices->device_list_mutex);
799 device = btrfs_find_device(fs_devices, &args);
800
801 if (found_transid > fs_devices->latest_generation) {
802 memcpy(fs_devices->fsid, disk_super->fsid,
803 BTRFS_FSID_SIZE);
804 memcpy(fs_devices->metadata_uuid,
805 btrfs_sb_fsid_ptr(disk_super), BTRFS_FSID_SIZE);
806 }
807 }
808
809 if (!device) {
810 unsigned int nofs_flag;
811
812 if (fs_devices->opened) {
813 btrfs_err(NULL,
814 "device %s (%d:%d) belongs to fsid %pU, and the fs is already mounted, scanned by %s (%d)",
815 path, MAJOR(path_devt), MINOR(path_devt),
816 fs_devices->fsid, current->comm,
817 task_pid_nr(current));
818 mutex_unlock(&fs_devices->device_list_mutex);
819 return ERR_PTR(-EBUSY);
820 }
821
822 nofs_flag = memalloc_nofs_save();
823 device = btrfs_alloc_device(NULL, &devid,
824 disk_super->dev_item.uuid, path);
825 memalloc_nofs_restore(nofs_flag);
826 if (IS_ERR(device)) {
827 mutex_unlock(&fs_devices->device_list_mutex);
828 /* we can safely leave the fs_devices entry around */
829 return device;
830 }
831
832 device->devt = path_devt;
833
834 list_add_rcu(&device->dev_list, &fs_devices->devices);
835 fs_devices->num_devices++;
836
837 device->fs_devices = fs_devices;
838 *new_device_added = true;
839
840 if (disk_super->label[0])
841 pr_info(
842 "BTRFS: device label %s devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
843 disk_super->label, devid, found_transid, path,
844 MAJOR(path_devt), MINOR(path_devt),
845 current->comm, task_pid_nr(current));
846 else
847 pr_info(
848 "BTRFS: device fsid %pU devid %llu transid %llu %s (%d:%d) scanned by %s (%d)\n",
849 disk_super->fsid, devid, found_transid, path,
850 MAJOR(path_devt), MINOR(path_devt),
851 current->comm, task_pid_nr(current));
852
853 } else if (!device->name || strcmp(device->name->str, path)) {
854 /*
855 * When FS is already mounted.
856 * 1. If you are here and if the device->name is NULL that
857 * means this device was missing at time of FS mount.
858 * 2. If you are here and if the device->name is different
859 * from 'path' that means either
860 * a. The same device disappeared and reappeared with
861 * different name. or
862 * b. The missing-disk-which-was-replaced, has
863 * reappeared now.
864 *
865 * We must allow 1 and 2a above. But 2b would be a spurious
866 * and unintentional.
867 *
868 * Further in case of 1 and 2a above, the disk at 'path'
869 * would have missed some transaction when it was away and
870 * in case of 2a the stale bdev has to be updated as well.
871 * 2b must not be allowed at all time.
872 */
873
874 /*
875 * For now, we do allow update to btrfs_fs_device through the
876 * btrfs dev scan cli after FS has been mounted. We're still
877 * tracking a problem where systems fail mount by subvolume id
878 * when we reject replacement on a mounted FS.
879 */
880 if (!fs_devices->opened && found_transid < device->generation) {
881 /*
882 * That is if the FS is _not_ mounted and if you
883 * are here, that means there is more than one
884 * disk with same uuid and devid.We keep the one
885 * with larger generation number or the last-in if
886 * generation are equal.
887 */
888 mutex_unlock(&fs_devices->device_list_mutex);
889 btrfs_err(NULL,
890 "device %s already registered with a higher generation, found %llu expect %llu",
891 path, found_transid, device->generation);
892 return ERR_PTR(-EEXIST);
893 }
894
895 /*
896 * We are going to replace the device path for a given devid,
897 * make sure it's the same device if the device is mounted
898 *
899 * NOTE: the device->fs_info may not be reliable here so pass
900 * in a NULL to message helpers instead. This avoids a possible
901 * use-after-free when the fs_info and fs_info->sb are already
902 * torn down.
903 */
904 if (device->bdev) {
905 if (device->devt != path_devt) {
906 mutex_unlock(&fs_devices->device_list_mutex);
907 btrfs_warn_in_rcu(NULL,
908 "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
909 path, devid, found_transid,
910 current->comm,
911 task_pid_nr(current));
912 return ERR_PTR(-EEXIST);
913 }
914 btrfs_info_in_rcu(NULL,
915 "devid %llu device path %s changed to %s scanned by %s (%d)",
916 devid, btrfs_dev_name(device),
917 path, current->comm,
918 task_pid_nr(current));
919 }
920
921 name = rcu_string_strdup(path, GFP_NOFS);
922 if (!name) {
923 mutex_unlock(&fs_devices->device_list_mutex);
924 return ERR_PTR(-ENOMEM);
925 }
926 rcu_string_free(device->name);
927 rcu_assign_pointer(device->name, name);
928 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
929 fs_devices->missing_devices--;
930 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
931 }
932 device->devt = path_devt;
933 }
934
935 /*
936 * Unmount does not free the btrfs_device struct but would zero
937 * generation along with most of the other members. So just update
938 * it back. We need it to pick the disk with largest generation
939 * (as above).
940 */
941 if (!fs_devices->opened) {
942 device->generation = found_transid;
943 fs_devices->latest_generation = max_t(u64, found_transid,
944 fs_devices->latest_generation);
945 }
946
947 fs_devices->total_devices = btrfs_super_num_devices(disk_super);
948
949 mutex_unlock(&fs_devices->device_list_mutex);
950 return device;
951 }
952
953 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
954 {
955 struct btrfs_fs_devices *fs_devices;
956 struct btrfs_device *device;
957 struct btrfs_device *orig_dev;
958 int ret = 0;
959
960 lockdep_assert_held(&uuid_mutex);
961
962 fs_devices = alloc_fs_devices(orig->fsid);
963 if (IS_ERR(fs_devices))
964 return fs_devices;
965
966 fs_devices->total_devices = orig->total_devices;
967
968 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
969 const char *dev_path = NULL;
970
971 /*
972 * This is ok to do without RCU read locked because we hold the
973 * uuid mutex so nothing we touch in here is going to disappear.
974 */
975 if (orig_dev->name)
976 dev_path = orig_dev->name->str;
977
978 device = btrfs_alloc_device(NULL, &orig_dev->devid,
979 orig_dev->uuid, dev_path);
980 if (IS_ERR(device)) {
981 ret = PTR_ERR(device);
982 goto error;
983 }
984
985 if (orig_dev->zone_info) {
986 struct btrfs_zoned_device_info *zone_info;
987
988 zone_info = btrfs_clone_dev_zone_info(orig_dev);
989 if (!zone_info) {
990 btrfs_free_device(device);
991 ret = -ENOMEM;
992 goto error;
993 }
994 device->zone_info = zone_info;
995 }
996
997 list_add(&device->dev_list, &fs_devices->devices);
998 device->fs_devices = fs_devices;
999 fs_devices->num_devices++;
1000 }
1001 return fs_devices;
1002 error:
1003 free_fs_devices(fs_devices);
1004 return ERR_PTR(ret);
1005 }
1006
1007 static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1008 struct btrfs_device **latest_dev)
1009 {
1010 struct btrfs_device *device, *next;
1011
1012 /* This is the initialized path, it is safe to release the devices. */
1013 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1014 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1015 if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1016 &device->dev_state) &&
1017 !test_bit(BTRFS_DEV_STATE_MISSING,
1018 &device->dev_state) &&
1019 (!*latest_dev ||
1020 device->generation > (*latest_dev)->generation)) {
1021 *latest_dev = device;
1022 }
1023 continue;
1024 }
1025
1026 /*
1027 * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1028 * in btrfs_init_dev_replace() so just continue.
1029 */
1030 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1031 continue;
1032
1033 if (device->bdev_file) {
1034 fput(device->bdev_file);
1035 device->bdev = NULL;
1036 device->bdev_file = NULL;
1037 fs_devices->open_devices--;
1038 }
1039 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1040 list_del_init(&device->dev_alloc_list);
1041 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1042 fs_devices->rw_devices--;
1043 }
1044 list_del_init(&device->dev_list);
1045 fs_devices->num_devices--;
1046 btrfs_free_device(device);
1047 }
1048
1049 }
1050
1051 /*
1052 * After we have read the system tree and know devids belonging to this
1053 * filesystem, remove the device which does not belong there.
1054 */
1055 void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1056 {
1057 struct btrfs_device *latest_dev = NULL;
1058 struct btrfs_fs_devices *seed_dev;
1059
1060 mutex_lock(&uuid_mutex);
1061 __btrfs_free_extra_devids(fs_devices, &latest_dev);
1062
1063 list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1064 __btrfs_free_extra_devids(seed_dev, &latest_dev);
1065
1066 fs_devices->latest_dev = latest_dev;
1067
1068 mutex_unlock(&uuid_mutex);
1069 }
1070
1071 static void btrfs_close_bdev(struct btrfs_device *device)
1072 {
1073 if (!device->bdev)
1074 return;
1075
1076 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1077 sync_blockdev(device->bdev);
1078 invalidate_bdev(device->bdev);
1079 }
1080
1081 fput(device->bdev_file);
1082 }
1083
1084 static void btrfs_close_one_device(struct btrfs_device *device)
1085 {
1086 struct btrfs_fs_devices *fs_devices = device->fs_devices;
1087
1088 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1089 device->devid != BTRFS_DEV_REPLACE_DEVID) {
1090 list_del_init(&device->dev_alloc_list);
1091 fs_devices->rw_devices--;
1092 }
1093
1094 if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1095 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1096
1097 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1098 clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1099 fs_devices->missing_devices--;
1100 }
1101
1102 btrfs_close_bdev(device);
1103 if (device->bdev) {
1104 fs_devices->open_devices--;
1105 device->bdev = NULL;
1106 }
1107 clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1108 btrfs_destroy_dev_zone_info(device);
1109
1110 device->fs_info = NULL;
1111 atomic_set(&device->dev_stats_ccnt, 0);
1112 extent_io_tree_release(&device->alloc_state);
1113
1114 /*
1115 * Reset the flush error record. We might have a transient flush error
1116 * in this mount, and if so we aborted the current transaction and set
1117 * the fs to an error state, guaranteeing no super blocks can be further
1118 * committed. However that error might be transient and if we unmount the
1119 * filesystem and mount it again, we should allow the mount to succeed
1120 * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1121 * filesystem again we still get flush errors, then we will again abort
1122 * any transaction and set the error state, guaranteeing no commits of
1123 * unsafe super blocks.
1124 */
1125 device->last_flush_error = 0;
1126
1127 /* Verify the device is back in a pristine state */
1128 WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1129 WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1130 WARN_ON(!list_empty(&device->dev_alloc_list));
1131 WARN_ON(!list_empty(&device->post_commit_list));
1132 }
1133
1134 static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1135 {
1136 struct btrfs_device *device, *tmp;
1137
1138 lockdep_assert_held(&uuid_mutex);
1139
1140 if (--fs_devices->opened > 0)
1141 return;
1142
1143 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1144 btrfs_close_one_device(device);
1145
1146 WARN_ON(fs_devices->open_devices);
1147 WARN_ON(fs_devices->rw_devices);
1148 fs_devices->opened = 0;
1149 fs_devices->seeding = false;
1150 fs_devices->fs_info = NULL;
1151 }
1152
1153 void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1154 {
1155 LIST_HEAD(list);
1156 struct btrfs_fs_devices *tmp;
1157
1158 mutex_lock(&uuid_mutex);
1159 close_fs_devices(fs_devices);
1160 if (!fs_devices->opened) {
1161 list_splice_init(&fs_devices->seed_list, &list);
1162
1163 /*
1164 * If the struct btrfs_fs_devices is not assembled with any
1165 * other device, it can be re-initialized during the next mount
1166 * without the needing device-scan step. Therefore, it can be
1167 * fully freed.
1168 */
1169 if (fs_devices->num_devices == 1) {
1170 list_del(&fs_devices->fs_list);
1171 free_fs_devices(fs_devices);
1172 }
1173 }
1174
1175
1176 list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1177 close_fs_devices(fs_devices);
1178 list_del(&fs_devices->seed_list);
1179 free_fs_devices(fs_devices);
1180 }
1181 mutex_unlock(&uuid_mutex);
1182 }
1183
1184 static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1185 blk_mode_t flags, void *holder)
1186 {
1187 struct btrfs_device *device;
1188 struct btrfs_device *latest_dev = NULL;
1189 struct btrfs_device *tmp_device;
1190 int ret = 0;
1191
1192 list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1193 dev_list) {
1194 int ret2;
1195
1196 ret2 = btrfs_open_one_device(fs_devices, device, flags, holder);
1197 if (ret2 == 0 &&
1198 (!latest_dev || device->generation > latest_dev->generation)) {
1199 latest_dev = device;
1200 } else if (ret2 == -ENODATA) {
1201 fs_devices->num_devices--;
1202 list_del(&device->dev_list);
1203 btrfs_free_device(device);
1204 }
1205 if (ret == 0 && ret2 != 0)
1206 ret = ret2;
1207 }
1208
1209 if (fs_devices->open_devices == 0) {
1210 if (ret)
1211 return ret;
1212 return -EINVAL;
1213 }
1214
1215 fs_devices->opened = 1;
1216 fs_devices->latest_dev = latest_dev;
1217 fs_devices->total_rw_bytes = 0;
1218 fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1219 fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1220
1221 return 0;
1222 }
1223
1224 static int devid_cmp(void *priv, const struct list_head *a,
1225 const struct list_head *b)
1226 {
1227 const struct btrfs_device *dev1, *dev2;
1228
1229 dev1 = list_entry(a, struct btrfs_device, dev_list);
1230 dev2 = list_entry(b, struct btrfs_device, dev_list);
1231
1232 if (dev1->devid < dev2->devid)
1233 return -1;
1234 else if (dev1->devid > dev2->devid)
1235 return 1;
1236 return 0;
1237 }
1238
1239 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1240 blk_mode_t flags, void *holder)
1241 {
1242 int ret;
1243
1244 lockdep_assert_held(&uuid_mutex);
1245 /*
1246 * The device_list_mutex cannot be taken here in case opening the
1247 * underlying device takes further locks like open_mutex.
1248 *
1249 * We also don't need the lock here as this is called during mount and
1250 * exclusion is provided by uuid_mutex
1251 */
1252
1253 if (fs_devices->opened) {
1254 fs_devices->opened++;
1255 ret = 0;
1256 } else {
1257 list_sort(NULL, &fs_devices->devices, devid_cmp);
1258 ret = open_fs_devices(fs_devices, flags, holder);
1259 }
1260
1261 return ret;
1262 }
1263
1264 void btrfs_release_disk_super(struct btrfs_super_block *super)
1265 {
1266 struct page *page = virt_to_page(super);
1267
1268 put_page(page);
1269 }
1270
1271 static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1272 u64 bytenr, u64 bytenr_orig)
1273 {
1274 struct btrfs_super_block *disk_super;
1275 struct page *page;
1276 void *p;
1277 pgoff_t index;
1278
1279 /* make sure our super fits in the device */
1280 if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1281 return ERR_PTR(-EINVAL);
1282
1283 /* make sure our super fits in the page */
1284 if (sizeof(*disk_super) > PAGE_SIZE)
1285 return ERR_PTR(-EINVAL);
1286
1287 /* make sure our super doesn't straddle pages on disk */
1288 index = bytenr >> PAGE_SHIFT;
1289 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1290 return ERR_PTR(-EINVAL);
1291
1292 /* pull in the page with our super */
1293 page = read_cache_page_gfp(bdev->bd_mapping, index, GFP_KERNEL);
1294
1295 if (IS_ERR(page))
1296 return ERR_CAST(page);
1297
1298 p = page_address(page);
1299
1300 /* align our pointer to the offset of the super block */
1301 disk_super = p + offset_in_page(bytenr);
1302
1303 if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1304 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1305 btrfs_release_disk_super(p);
1306 return ERR_PTR(-EINVAL);
1307 }
1308
1309 if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1310 disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1311
1312 return disk_super;
1313 }
1314
1315 int btrfs_forget_devices(dev_t devt)
1316 {
1317 int ret;
1318
1319 mutex_lock(&uuid_mutex);
1320 ret = btrfs_free_stale_devices(devt, NULL);
1321 mutex_unlock(&uuid_mutex);
1322
1323 return ret;
1324 }
1325
1326 static bool btrfs_skip_registration(struct btrfs_super_block *disk_super,
1327 const char *path, dev_t devt,
1328 bool mount_arg_dev)
1329 {
1330 struct btrfs_fs_devices *fs_devices;
1331
1332 /*
1333 * Do not skip device registration for mounted devices with matching
1334 * maj:min but different paths. Booting without initrd relies on
1335 * /dev/root initially, later replaced with the actual root device.
1336 * A successful scan ensures grub2-probe selects the correct device.
1337 */
1338 list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
1339 struct btrfs_device *device;
1340
1341 mutex_lock(&fs_devices->device_list_mutex);
1342
1343 if (!fs_devices->opened) {
1344 mutex_unlock(&fs_devices->device_list_mutex);
1345 continue;
1346 }
1347
1348 list_for_each_entry(device, &fs_devices->devices, dev_list) {
1349 if (device->bdev && (device->bdev->bd_dev == devt) &&
1350 strcmp(device->name->str, path) != 0) {
1351 mutex_unlock(&fs_devices->device_list_mutex);
1352
1353 /* Do not skip registration. */
1354 return false;
1355 }
1356 }
1357 mutex_unlock(&fs_devices->device_list_mutex);
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 return true;
1363
1364 return false;
1365 }
1366
1367 /*
1368 * Look for a btrfs signature on a device. This may be called out of the mount path
1369 * and we are not allowed to call set_blocksize during the scan. The superblock
1370 * is read via pagecache.
1371 *
1372 * With @mount_arg_dev it's a scan during mount time that will always register
1373 * the device or return an error. Multi-device and seeding devices are registered
1374 * in both cases.
1375 */
1376 struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags,
1377 bool mount_arg_dev)
1378 {
1379 struct btrfs_super_block *disk_super;
1380 bool new_device_added = false;
1381 struct btrfs_device *device = NULL;
1382 struct file *bdev_file;
1383 u64 bytenr;
1384 dev_t devt;
1385 int ret;
1386
1387 lockdep_assert_held(&uuid_mutex);
1388
1389 /*
1390 * Avoid an exclusive open here, as the systemd-udev may initiate the
1391 * device scan which may race with the user's mount or mkfs command,
1392 * resulting in failure.
1393 * Since the device scan is solely for reading purposes, there is no
1394 * need for an exclusive open. Additionally, the devices are read again
1395 * during the mount process. It is ok to get some inconsistent
1396 * values temporarily, as the device paths of the fsid are the only
1397 * required information for assembling the volume.
1398 */
1399 bdev_file = bdev_file_open_by_path(path, flags, NULL, NULL);
1400 if (IS_ERR(bdev_file))
1401 return ERR_CAST(bdev_file);
1402
1403 /*
1404 * We would like to check all the super blocks, but doing so would
1405 * allow a mount to succeed after a mkfs from a different filesystem.
1406 * Currently, recovery from a bad primary btrfs superblock is done
1407 * using the userspace command 'btrfs check --super'.
1408 */
1409 ret = btrfs_sb_log_location_bdev(file_bdev(bdev_file), 0, READ, &bytenr);
1410 if (ret) {
1411 device = ERR_PTR(ret);
1412 goto error_bdev_put;
1413 }
1414
1415 disk_super = btrfs_read_disk_super(file_bdev(bdev_file), bytenr,
1416 btrfs_sb_offset(0));
1417 if (IS_ERR(disk_super)) {
1418 device = ERR_CAST(disk_super);
1419 goto error_bdev_put;
1420 }
1421
1422 devt = file_bdev(bdev_file)->bd_dev;
1423 if (btrfs_skip_registration(disk_super, path, devt, mount_arg_dev)) {
1424 pr_debug("BTRFS: skip registering single non-seed device %s (%d:%d)\n",
1425 path, MAJOR(devt), MINOR(devt));
1426
1427 btrfs_free_stale_devices(devt, NULL);
1428
1429 device = NULL;
1430 goto free_disk_super;
1431 }
1432
1433 device = device_list_add(path, disk_super, &new_device_added);
1434 if (!IS_ERR(device) && new_device_added)
1435 btrfs_free_stale_devices(device->devt, device);
1436
1437 free_disk_super:
1438 btrfs_release_disk_super(disk_super);
1439
1440 error_bdev_put:
1441 fput(bdev_file);
1442
1443 return device;
1444 }
1445
1446 /*
1447 * Try to find a chunk that intersects [start, start + len] range and when one
1448 * such is found, record the end of it in *start
1449 */
1450 static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1451 u64 len)
1452 {
1453 u64 physical_start, physical_end;
1454
1455 lockdep_assert_held(&device->fs_info->chunk_mutex);
1456
1457 if (find_first_extent_bit(&device->alloc_state, *start,
1458 &physical_start, &physical_end,
1459 CHUNK_ALLOCATED, NULL)) {
1460
1461 if (in_range(physical_start, *start, len) ||
1462 in_range(*start, physical_start,
1463 physical_end + 1 - physical_start)) {
1464 *start = physical_end + 1;
1465 return true;
1466 }
1467 }
1468 return false;
1469 }
1470
1471 static u64 dev_extent_search_start(struct btrfs_device *device)
1472 {
1473 switch (device->fs_devices->chunk_alloc_policy) {
1474 case BTRFS_CHUNK_ALLOC_REGULAR:
1475 return BTRFS_DEVICE_RANGE_RESERVED;
1476 case BTRFS_CHUNK_ALLOC_ZONED:
1477 /*
1478 * We don't care about the starting region like regular
1479 * allocator, because we anyway use/reserve the first two zones
1480 * for superblock logging.
1481 */
1482 return 0;
1483 default:
1484 BUG();
1485 }
1486 }
1487
1488 static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1489 u64 *hole_start, u64 *hole_size,
1490 u64 num_bytes)
1491 {
1492 u64 zone_size = device->zone_info->zone_size;
1493 u64 pos;
1494 int ret;
1495 bool changed = false;
1496
1497 ASSERT(IS_ALIGNED(*hole_start, zone_size));
1498
1499 while (*hole_size > 0) {
1500 pos = btrfs_find_allocatable_zones(device, *hole_start,
1501 *hole_start + *hole_size,
1502 num_bytes);
1503 if (pos != *hole_start) {
1504 *hole_size = *hole_start + *hole_size - pos;
1505 *hole_start = pos;
1506 changed = true;
1507 if (*hole_size < num_bytes)
1508 break;
1509 }
1510
1511 ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1512
1513 /* Range is ensured to be empty */
1514 if (!ret)
1515 return changed;
1516
1517 /* Given hole range was invalid (outside of device) */
1518 if (ret == -ERANGE) {
1519 *hole_start += *hole_size;
1520 *hole_size = 0;
1521 return true;
1522 }
1523
1524 *hole_start += zone_size;
1525 *hole_size -= zone_size;
1526 changed = true;
1527 }
1528
1529 return changed;
1530 }
1531
1532 /*
1533 * Check if specified hole is suitable for allocation.
1534 *
1535 * @device: the device which we have the hole
1536 * @hole_start: starting position of the hole
1537 * @hole_size: the size of the hole
1538 * @num_bytes: the size of the free space that we need
1539 *
1540 * This function may modify @hole_start and @hole_size to reflect the suitable
1541 * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1542 */
1543 static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1544 u64 *hole_size, u64 num_bytes)
1545 {
1546 bool changed = false;
1547 u64 hole_end = *hole_start + *hole_size;
1548
1549 for (;;) {
1550 /*
1551 * Check before we set max_hole_start, otherwise we could end up
1552 * sending back this offset anyway.
1553 */
1554 if (contains_pending_extent(device, hole_start, *hole_size)) {
1555 if (hole_end >= *hole_start)
1556 *hole_size = hole_end - *hole_start;
1557 else
1558 *hole_size = 0;
1559 changed = true;
1560 }
1561
1562 switch (device->fs_devices->chunk_alloc_policy) {
1563 case BTRFS_CHUNK_ALLOC_REGULAR:
1564 /* No extra check */
1565 break;
1566 case BTRFS_CHUNK_ALLOC_ZONED:
1567 if (dev_extent_hole_check_zoned(device, hole_start,
1568 hole_size, num_bytes)) {
1569 changed = true;
1570 /*
1571 * The changed hole can contain pending extent.
1572 * Loop again to check that.
1573 */
1574 continue;
1575 }
1576 break;
1577 default:
1578 BUG();
1579 }
1580
1581 break;
1582 }
1583
1584 return changed;
1585 }
1586
1587 /*
1588 * Find free space in the specified device.
1589 *
1590 * @device: the device which we search the free space in
1591 * @num_bytes: the size of the free space that we need
1592 * @search_start: the position from which to begin the search
1593 * @start: store the start of the free space.
1594 * @len: the size of the free space. that we find, or the size
1595 * of the max free space if we don't find suitable free space
1596 *
1597 * This does a pretty simple search, the expectation is that it is called very
1598 * infrequently and that a given device has a small number of extents.
1599 *
1600 * @start is used to store the start of the free space if we find. But if we
1601 * don't find suitable free space, it will be used to store the start position
1602 * of the max free space.
1603 *
1604 * @len is used to store the size of the free space that we find.
1605 * But if we don't find suitable free space, it is used to store the size of
1606 * the max free space.
1607 *
1608 * NOTE: This function will search *commit* root of device tree, and does extra
1609 * check to ensure dev extents are not double allocated.
1610 * This makes the function safe to allocate dev extents but may not report
1611 * correct usable device space, as device extent freed in current transaction
1612 * is not reported as available.
1613 */
1614 static int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1615 u64 *start, u64 *len)
1616 {
1617 struct btrfs_fs_info *fs_info = device->fs_info;
1618 struct btrfs_root *root = fs_info->dev_root;
1619 struct btrfs_key key;
1620 struct btrfs_dev_extent *dev_extent;
1621 struct btrfs_path *path;
1622 u64 search_start;
1623 u64 hole_size;
1624 u64 max_hole_start;
1625 u64 max_hole_size = 0;
1626 u64 extent_end;
1627 u64 search_end = device->total_bytes;
1628 int ret;
1629 int slot;
1630 struct extent_buffer *l;
1631
1632 search_start = dev_extent_search_start(device);
1633 max_hole_start = search_start;
1634
1635 WARN_ON(device->zone_info &&
1636 !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1637
1638 path = btrfs_alloc_path();
1639 if (!path) {
1640 ret = -ENOMEM;
1641 goto out;
1642 }
1643 again:
1644 if (search_start >= search_end ||
1645 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1646 ret = -ENOSPC;
1647 goto out;
1648 }
1649
1650 path->reada = READA_FORWARD;
1651 path->search_commit_root = 1;
1652 path->skip_locking = 1;
1653
1654 key.objectid = device->devid;
1655 key.offset = search_start;
1656 key.type = BTRFS_DEV_EXTENT_KEY;
1657
1658 ret = btrfs_search_backwards(root, &key, path);
1659 if (ret < 0)
1660 goto out;
1661
1662 while (search_start < search_end) {
1663 l = path->nodes[0];
1664 slot = path->slots[0];
1665 if (slot >= btrfs_header_nritems(l)) {
1666 ret = btrfs_next_leaf(root, path);
1667 if (ret == 0)
1668 continue;
1669 if (ret < 0)
1670 goto out;
1671
1672 break;
1673 }
1674 btrfs_item_key_to_cpu(l, &key, slot);
1675
1676 if (key.objectid < device->devid)
1677 goto next;
1678
1679 if (key.objectid > device->devid)
1680 break;
1681
1682 if (key.type != BTRFS_DEV_EXTENT_KEY)
1683 goto next;
1684
1685 if (key.offset > search_end)
1686 break;
1687
1688 if (key.offset > search_start) {
1689 hole_size = key.offset - search_start;
1690 dev_extent_hole_check(device, &search_start, &hole_size,
1691 num_bytes);
1692
1693 if (hole_size > max_hole_size) {
1694 max_hole_start = search_start;
1695 max_hole_size = hole_size;
1696 }
1697
1698 /*
1699 * If this free space is greater than which we need,
1700 * it must be the max free space that we have found
1701 * until now, so max_hole_start must point to the start
1702 * of this free space and the length of this free space
1703 * is stored in max_hole_size. Thus, we return
1704 * max_hole_start and max_hole_size and go back to the
1705 * caller.
1706 */
1707 if (hole_size >= num_bytes) {
1708 ret = 0;
1709 goto out;
1710 }
1711 }
1712
1713 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1714 extent_end = key.offset + btrfs_dev_extent_length(l,
1715 dev_extent);
1716 if (extent_end > search_start)
1717 search_start = extent_end;
1718 next:
1719 path->slots[0]++;
1720 cond_resched();
1721 }
1722
1723 /*
1724 * At this point, search_start should be the end of
1725 * allocated dev extents, and when shrinking the device,
1726 * search_end may be smaller than search_start.
1727 */
1728 if (search_end > search_start) {
1729 hole_size = search_end - search_start;
1730 if (dev_extent_hole_check(device, &search_start, &hole_size,
1731 num_bytes)) {
1732 btrfs_release_path(path);
1733 goto again;
1734 }
1735
1736 if (hole_size > max_hole_size) {
1737 max_hole_start = search_start;
1738 max_hole_size = hole_size;
1739 }
1740 }
1741
1742 /* See above. */
1743 if (max_hole_size < num_bytes)
1744 ret = -ENOSPC;
1745 else
1746 ret = 0;
1747
1748 ASSERT(max_hole_start + max_hole_size <= search_end);
1749 out:
1750 btrfs_free_path(path);
1751 *start = max_hole_start;
1752 if (len)
1753 *len = max_hole_size;
1754 return ret;
1755 }
1756
1757 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1758 struct btrfs_device *device,
1759 u64 start, u64 *dev_extent_len)
1760 {
1761 struct btrfs_fs_info *fs_info = device->fs_info;
1762 struct btrfs_root *root = fs_info->dev_root;
1763 int ret;
1764 struct btrfs_path *path;
1765 struct btrfs_key key;
1766 struct btrfs_key found_key;
1767 struct extent_buffer *leaf = NULL;
1768 struct btrfs_dev_extent *extent = NULL;
1769
1770 path = btrfs_alloc_path();
1771 if (!path)
1772 return -ENOMEM;
1773
1774 key.objectid = device->devid;
1775 key.offset = start;
1776 key.type = BTRFS_DEV_EXTENT_KEY;
1777 again:
1778 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1779 if (ret > 0) {
1780 ret = btrfs_previous_item(root, path, key.objectid,
1781 BTRFS_DEV_EXTENT_KEY);
1782 if (ret)
1783 goto out;
1784 leaf = path->nodes[0];
1785 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1786 extent = btrfs_item_ptr(leaf, path->slots[0],
1787 struct btrfs_dev_extent);
1788 BUG_ON(found_key.offset > start || found_key.offset +
1789 btrfs_dev_extent_length(leaf, extent) < start);
1790 key = found_key;
1791 btrfs_release_path(path);
1792 goto again;
1793 } else if (ret == 0) {
1794 leaf = path->nodes[0];
1795 extent = btrfs_item_ptr(leaf, path->slots[0],
1796 struct btrfs_dev_extent);
1797 } else {
1798 goto out;
1799 }
1800
1801 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1802
1803 ret = btrfs_del_item(trans, root, path);
1804 if (ret == 0)
1805 set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1806 out:
1807 btrfs_free_path(path);
1808 return ret;
1809 }
1810
1811 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1812 {
1813 struct rb_node *n;
1814 u64 ret = 0;
1815
1816 read_lock(&fs_info->mapping_tree_lock);
1817 n = rb_last(&fs_info->mapping_tree.rb_root);
1818 if (n) {
1819 struct btrfs_chunk_map *map;
1820
1821 map = rb_entry(n, struct btrfs_chunk_map, rb_node);
1822 ret = map->start + map->chunk_len;
1823 }
1824 read_unlock(&fs_info->mapping_tree_lock);
1825
1826 return ret;
1827 }
1828
1829 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1830 u64 *devid_ret)
1831 {
1832 int ret;
1833 struct btrfs_key key;
1834 struct btrfs_key found_key;
1835 struct btrfs_path *path;
1836
1837 path = btrfs_alloc_path();
1838 if (!path)
1839 return -ENOMEM;
1840
1841 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1842 key.type = BTRFS_DEV_ITEM_KEY;
1843 key.offset = (u64)-1;
1844
1845 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1846 if (ret < 0)
1847 goto error;
1848
1849 if (ret == 0) {
1850 /* Corruption */
1851 btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1852 ret = -EUCLEAN;
1853 goto error;
1854 }
1855
1856 ret = btrfs_previous_item(fs_info->chunk_root, path,
1857 BTRFS_DEV_ITEMS_OBJECTID,
1858 BTRFS_DEV_ITEM_KEY);
1859 if (ret) {
1860 *devid_ret = 1;
1861 } else {
1862 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1863 path->slots[0]);
1864 *devid_ret = found_key.offset + 1;
1865 }
1866 ret = 0;
1867 error:
1868 btrfs_free_path(path);
1869 return ret;
1870 }
1871
1872 /*
1873 * the device information is stored in the chunk root
1874 * the btrfs_device struct should be fully filled in
1875 */
1876 static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1877 struct btrfs_device *device)
1878 {
1879 int ret;
1880 struct btrfs_path *path;
1881 struct btrfs_dev_item *dev_item;
1882 struct extent_buffer *leaf;
1883 struct btrfs_key key;
1884 unsigned long ptr;
1885
1886 path = btrfs_alloc_path();
1887 if (!path)
1888 return -ENOMEM;
1889
1890 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1891 key.type = BTRFS_DEV_ITEM_KEY;
1892 key.offset = device->devid;
1893
1894 btrfs_reserve_chunk_metadata(trans, true);
1895 ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1896 &key, sizeof(*dev_item));
1897 btrfs_trans_release_chunk_metadata(trans);
1898 if (ret)
1899 goto out;
1900
1901 leaf = path->nodes[0];
1902 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1903
1904 btrfs_set_device_id(leaf, dev_item, device->devid);
1905 btrfs_set_device_generation(leaf, dev_item, 0);
1906 btrfs_set_device_type(leaf, dev_item, device->type);
1907 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1908 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1909 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1910 btrfs_set_device_total_bytes(leaf, dev_item,
1911 btrfs_device_get_disk_total_bytes(device));
1912 btrfs_set_device_bytes_used(leaf, dev_item,
1913 btrfs_device_get_bytes_used(device));
1914 btrfs_set_device_group(leaf, dev_item, 0);
1915 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1916 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1917 btrfs_set_device_start_offset(leaf, dev_item, 0);
1918
1919 ptr = btrfs_device_uuid(dev_item);
1920 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1921 ptr = btrfs_device_fsid(dev_item);
1922 write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1923 ptr, BTRFS_FSID_SIZE);
1924 btrfs_mark_buffer_dirty(trans, leaf);
1925
1926 ret = 0;
1927 out:
1928 btrfs_free_path(path);
1929 return ret;
1930 }
1931
1932 /*
1933 * Function to update ctime/mtime for a given device path.
1934 * Mainly used for ctime/mtime based probe like libblkid.
1935 *
1936 * We don't care about errors here, this is just to be kind to userspace.
1937 */
1938 static void update_dev_time(const char *device_path)
1939 {
1940 struct path path;
1941 int ret;
1942
1943 ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1944 if (ret)
1945 return;
1946
1947 inode_update_time(d_inode(path.dentry), S_MTIME | S_CTIME | S_VERSION);
1948 path_put(&path);
1949 }
1950
1951 static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1952 struct btrfs_device *device)
1953 {
1954 struct btrfs_root *root = device->fs_info->chunk_root;
1955 int ret;
1956 struct btrfs_path *path;
1957 struct btrfs_key key;
1958
1959 path = btrfs_alloc_path();
1960 if (!path)
1961 return -ENOMEM;
1962
1963 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1964 key.type = BTRFS_DEV_ITEM_KEY;
1965 key.offset = device->devid;
1966
1967 btrfs_reserve_chunk_metadata(trans, false);
1968 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1969 btrfs_trans_release_chunk_metadata(trans);
1970 if (ret) {
1971 if (ret > 0)
1972 ret = -ENOENT;
1973 goto out;
1974 }
1975
1976 ret = btrfs_del_item(trans, root, path);
1977 out:
1978 btrfs_free_path(path);
1979 return ret;
1980 }
1981
1982 /*
1983 * Verify that @num_devices satisfies the RAID profile constraints in the whole
1984 * filesystem. It's up to the caller to adjust that number regarding eg. device
1985 * replace.
1986 */
1987 static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1988 u64 num_devices)
1989 {
1990 u64 all_avail;
1991 unsigned seq;
1992 int i;
1993
1994 do {
1995 seq = read_seqbegin(&fs_info->profiles_lock);
1996
1997 all_avail = fs_info->avail_data_alloc_bits |
1998 fs_info->avail_system_alloc_bits |
1999 fs_info->avail_metadata_alloc_bits;
2000 } while (read_seqretry(&fs_info->profiles_lock, seq));
2001
2002 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
2003 if (!(all_avail & btrfs_raid_array[i].bg_flag))
2004 continue;
2005
2006 if (num_devices < btrfs_raid_array[i].devs_min)
2007 return btrfs_raid_array[i].mindev_error;
2008 }
2009
2010 return 0;
2011 }
2012
2013 static struct btrfs_device * btrfs_find_next_active_device(
2014 struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
2015 {
2016 struct btrfs_device *next_device;
2017
2018 list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2019 if (next_device != device &&
2020 !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2021 && next_device->bdev)
2022 return next_device;
2023 }
2024
2025 return NULL;
2026 }
2027
2028 /*
2029 * Helper function to check if the given device is part of s_bdev / latest_dev
2030 * and replace it with the provided or the next active device, in the context
2031 * where this function called, there should be always be another device (or
2032 * this_dev) which is active.
2033 */
2034 void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2035 struct btrfs_device *next_device)
2036 {
2037 struct btrfs_fs_info *fs_info = device->fs_info;
2038
2039 if (!next_device)
2040 next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2041 device);
2042 ASSERT(next_device);
2043
2044 if (fs_info->sb->s_bdev &&
2045 (fs_info->sb->s_bdev == device->bdev))
2046 fs_info->sb->s_bdev = next_device->bdev;
2047
2048 if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2049 fs_info->fs_devices->latest_dev = next_device;
2050 }
2051
2052 /*
2053 * Return btrfs_fs_devices::num_devices excluding the device that's being
2054 * currently replaced.
2055 */
2056 static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2057 {
2058 u64 num_devices = fs_info->fs_devices->num_devices;
2059
2060 down_read(&fs_info->dev_replace.rwsem);
2061 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2062 ASSERT(num_devices > 1);
2063 num_devices--;
2064 }
2065 up_read(&fs_info->dev_replace.rwsem);
2066
2067 return num_devices;
2068 }
2069
2070 static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2071 struct block_device *bdev, int copy_num)
2072 {
2073 struct btrfs_super_block *disk_super;
2074 const size_t len = sizeof(disk_super->magic);
2075 const u64 bytenr = btrfs_sb_offset(copy_num);
2076 int ret;
2077
2078 disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2079 if (IS_ERR(disk_super))
2080 return;
2081
2082 memset(&disk_super->magic, 0, len);
2083 folio_mark_dirty(virt_to_folio(disk_super));
2084 btrfs_release_disk_super(disk_super);
2085
2086 ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2087 if (ret)
2088 btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2089 copy_num, ret);
2090 }
2091
2092 void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info, struct btrfs_device *device)
2093 {
2094 int copy_num;
2095 struct block_device *bdev = device->bdev;
2096
2097 if (!bdev)
2098 return;
2099
2100 for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2101 if (bdev_is_zoned(bdev))
2102 btrfs_reset_sb_log_zones(bdev, copy_num);
2103 else
2104 btrfs_scratch_superblock(fs_info, bdev, copy_num);
2105 }
2106
2107 /* Notify udev that device has changed */
2108 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2109
2110 /* Update ctime/mtime for device path for libblkid */
2111 update_dev_time(device->name->str);
2112 }
2113
2114 int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2115 struct btrfs_dev_lookup_args *args,
2116 struct file **bdev_file)
2117 {
2118 struct btrfs_trans_handle *trans;
2119 struct btrfs_device *device;
2120 struct btrfs_fs_devices *cur_devices;
2121 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2122 u64 num_devices;
2123 int ret = 0;
2124
2125 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2126 btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2127 return -EINVAL;
2128 }
2129
2130 /*
2131 * The device list in fs_devices is accessed without locks (neither
2132 * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2133 * filesystem and another device rm cannot run.
2134 */
2135 num_devices = btrfs_num_devices(fs_info);
2136
2137 ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2138 if (ret)
2139 return ret;
2140
2141 device = btrfs_find_device(fs_info->fs_devices, args);
2142 if (!device) {
2143 if (args->missing)
2144 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2145 else
2146 ret = -ENOENT;
2147 return ret;
2148 }
2149
2150 if (btrfs_pinned_by_swapfile(fs_info, device)) {
2151 btrfs_warn_in_rcu(fs_info,
2152 "cannot remove device %s (devid %llu) due to active swapfile",
2153 btrfs_dev_name(device), device->devid);
2154 return -ETXTBSY;
2155 }
2156
2157 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2158 return BTRFS_ERROR_DEV_TGT_REPLACE;
2159
2160 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2161 fs_info->fs_devices->rw_devices == 1)
2162 return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2163
2164 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2165 mutex_lock(&fs_info->chunk_mutex);
2166 list_del_init(&device->dev_alloc_list);
2167 device->fs_devices->rw_devices--;
2168 mutex_unlock(&fs_info->chunk_mutex);
2169 }
2170
2171 ret = btrfs_shrink_device(device, 0);
2172 if (ret)
2173 goto error_undo;
2174
2175 trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2176 if (IS_ERR(trans)) {
2177 ret = PTR_ERR(trans);
2178 goto error_undo;
2179 }
2180
2181 ret = btrfs_rm_dev_item(trans, device);
2182 if (ret) {
2183 /* Any error in dev item removal is critical */
2184 btrfs_crit(fs_info,
2185 "failed to remove device item for devid %llu: %d",
2186 device->devid, ret);
2187 btrfs_abort_transaction(trans, ret);
2188 btrfs_end_transaction(trans);
2189 return ret;
2190 }
2191
2192 clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2193 btrfs_scrub_cancel_dev(device);
2194
2195 /*
2196 * the device list mutex makes sure that we don't change
2197 * the device list while someone else is writing out all
2198 * the device supers. Whoever is writing all supers, should
2199 * lock the device list mutex before getting the number of
2200 * devices in the super block (super_copy). Conversely,
2201 * whoever updates the number of devices in the super block
2202 * (super_copy) should hold the device list mutex.
2203 */
2204
2205 /*
2206 * In normal cases the cur_devices == fs_devices. But in case
2207 * of deleting a seed device, the cur_devices should point to
2208 * its own fs_devices listed under the fs_devices->seed_list.
2209 */
2210 cur_devices = device->fs_devices;
2211 mutex_lock(&fs_devices->device_list_mutex);
2212 list_del_rcu(&device->dev_list);
2213
2214 cur_devices->num_devices--;
2215 cur_devices->total_devices--;
2216 /* Update total_devices of the parent fs_devices if it's seed */
2217 if (cur_devices != fs_devices)
2218 fs_devices->total_devices--;
2219
2220 if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2221 cur_devices->missing_devices--;
2222
2223 btrfs_assign_next_active_device(device, NULL);
2224
2225 if (device->bdev_file) {
2226 cur_devices->open_devices--;
2227 /* remove sysfs entry */
2228 btrfs_sysfs_remove_device(device);
2229 }
2230
2231 num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2232 btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2233 mutex_unlock(&fs_devices->device_list_mutex);
2234
2235 /*
2236 * At this point, the device is zero sized and detached from the
2237 * devices list. All that's left is to zero out the old supers and
2238 * free the device.
2239 *
2240 * We cannot call btrfs_close_bdev() here because we're holding the sb
2241 * write lock, and fput() on the block device will pull in the
2242 * ->open_mutex on the block device and it's dependencies. Instead
2243 * just flush the device and let the caller do the final bdev_release.
2244 */
2245 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2246 btrfs_scratch_superblocks(fs_info, device);
2247 if (device->bdev) {
2248 sync_blockdev(device->bdev);
2249 invalidate_bdev(device->bdev);
2250 }
2251 }
2252
2253 *bdev_file = device->bdev_file;
2254 synchronize_rcu();
2255 btrfs_free_device(device);
2256
2257 /*
2258 * This can happen if cur_devices is the private seed devices list. We
2259 * cannot call close_fs_devices() here because it expects the uuid_mutex
2260 * to be held, but in fact we don't need that for the private
2261 * seed_devices, we can simply decrement cur_devices->opened and then
2262 * remove it from our list and free the fs_devices.
2263 */
2264 if (cur_devices->num_devices == 0) {
2265 list_del_init(&cur_devices->seed_list);
2266 ASSERT(cur_devices->opened == 1);
2267 cur_devices->opened--;
2268 free_fs_devices(cur_devices);
2269 }
2270
2271 ret = btrfs_commit_transaction(trans);
2272
2273 return ret;
2274
2275 error_undo:
2276 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2277 mutex_lock(&fs_info->chunk_mutex);
2278 list_add(&device->dev_alloc_list,
2279 &fs_devices->alloc_list);
2280 device->fs_devices->rw_devices++;
2281 mutex_unlock(&fs_info->chunk_mutex);
2282 }
2283 return ret;
2284 }
2285
2286 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2287 {
2288 struct btrfs_fs_devices *fs_devices;
2289
2290 lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2291
2292 /*
2293 * in case of fs with no seed, srcdev->fs_devices will point
2294 * to fs_devices of fs_info. However when the dev being replaced is
2295 * a seed dev it will point to the seed's local fs_devices. In short
2296 * srcdev will have its correct fs_devices in both the cases.
2297 */
2298 fs_devices = srcdev->fs_devices;
2299
2300 list_del_rcu(&srcdev->dev_list);
2301 list_del(&srcdev->dev_alloc_list);
2302 fs_devices->num_devices--;
2303 if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2304 fs_devices->missing_devices--;
2305
2306 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2307 fs_devices->rw_devices--;
2308
2309 if (srcdev->bdev)
2310 fs_devices->open_devices--;
2311 }
2312
2313 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2314 {
2315 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2316
2317 mutex_lock(&uuid_mutex);
2318
2319 btrfs_close_bdev(srcdev);
2320 synchronize_rcu();
2321 btrfs_free_device(srcdev);
2322
2323 /* if this is no devs we rather delete the fs_devices */
2324 if (!fs_devices->num_devices) {
2325 /*
2326 * On a mounted FS, num_devices can't be zero unless it's a
2327 * seed. In case of a seed device being replaced, the replace
2328 * target added to the sprout FS, so there will be no more
2329 * device left under the seed FS.
2330 */
2331 ASSERT(fs_devices->seeding);
2332
2333 list_del_init(&fs_devices->seed_list);
2334 close_fs_devices(fs_devices);
2335 free_fs_devices(fs_devices);
2336 }
2337 mutex_unlock(&uuid_mutex);
2338 }
2339
2340 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2341 {
2342 struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2343
2344 mutex_lock(&fs_devices->device_list_mutex);
2345
2346 btrfs_sysfs_remove_device(tgtdev);
2347
2348 if (tgtdev->bdev)
2349 fs_devices->open_devices--;
2350
2351 fs_devices->num_devices--;
2352
2353 btrfs_assign_next_active_device(tgtdev, NULL);
2354
2355 list_del_rcu(&tgtdev->dev_list);
2356
2357 mutex_unlock(&fs_devices->device_list_mutex);
2358
2359 btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev);
2360
2361 btrfs_close_bdev(tgtdev);
2362 synchronize_rcu();
2363 btrfs_free_device(tgtdev);
2364 }
2365
2366 /*
2367 * Populate args from device at path.
2368 *
2369 * @fs_info: the filesystem
2370 * @args: the args to populate
2371 * @path: the path to the device
2372 *
2373 * This will read the super block of the device at @path and populate @args with
2374 * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2375 * lookup a device to operate on, but need to do it before we take any locks.
2376 * This properly handles the special case of "missing" that a user may pass in,
2377 * and does some basic sanity checks. The caller must make sure that @path is
2378 * properly NUL terminated before calling in, and must call
2379 * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2380 * uuid buffers.
2381 *
2382 * Return: 0 for success, -errno for failure
2383 */
2384 int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2385 struct btrfs_dev_lookup_args *args,
2386 const char *path)
2387 {
2388 struct btrfs_super_block *disk_super;
2389 struct file *bdev_file;
2390 int ret;
2391
2392 if (!path || !path[0])
2393 return -EINVAL;
2394 if (!strcmp(path, "missing")) {
2395 args->missing = true;
2396 return 0;
2397 }
2398
2399 args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2400 args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2401 if (!args->uuid || !args->fsid) {
2402 btrfs_put_dev_args_from_path(args);
2403 return -ENOMEM;
2404 }
2405
2406 ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2407 &bdev_file, &disk_super);
2408 if (ret) {
2409 btrfs_put_dev_args_from_path(args);
2410 return ret;
2411 }
2412
2413 args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2414 memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2415 if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2416 memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2417 else
2418 memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2419 btrfs_release_disk_super(disk_super);
2420 fput(bdev_file);
2421 return 0;
2422 }
2423
2424 /*
2425 * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2426 * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2427 * that don't need to be freed.
2428 */
2429 void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2430 {
2431 kfree(args->uuid);
2432 kfree(args->fsid);
2433 args->uuid = NULL;
2434 args->fsid = NULL;
2435 }
2436
2437 struct btrfs_device *btrfs_find_device_by_devspec(
2438 struct btrfs_fs_info *fs_info, u64 devid,
2439 const char *device_path)
2440 {
2441 BTRFS_DEV_LOOKUP_ARGS(args);
2442 struct btrfs_device *device;
2443 int ret;
2444
2445 if (devid) {
2446 args.devid = devid;
2447 device = btrfs_find_device(fs_info->fs_devices, &args);
2448 if (!device)
2449 return ERR_PTR(-ENOENT);
2450 return device;
2451 }
2452
2453 ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2454 if (ret)
2455 return ERR_PTR(ret);
2456 device = btrfs_find_device(fs_info->fs_devices, &args);
2457 btrfs_put_dev_args_from_path(&args);
2458 if (!device)
2459 return ERR_PTR(-ENOENT);
2460 return device;
2461 }
2462
2463 static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2464 {
2465 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2466 struct btrfs_fs_devices *old_devices;
2467 struct btrfs_fs_devices *seed_devices;
2468
2469 lockdep_assert_held(&uuid_mutex);
2470 if (!fs_devices->seeding)
2471 return ERR_PTR(-EINVAL);
2472
2473 /*
2474 * Private copy of the seed devices, anchored at
2475 * fs_info->fs_devices->seed_list
2476 */
2477 seed_devices = alloc_fs_devices(NULL);
2478 if (IS_ERR(seed_devices))
2479 return seed_devices;
2480
2481 /*
2482 * It's necessary to retain a copy of the original seed fs_devices in
2483 * fs_uuids so that filesystems which have been seeded can successfully
2484 * reference the seed device from open_seed_devices. This also supports
2485 * multiple fs seed.
2486 */
2487 old_devices = clone_fs_devices(fs_devices);
2488 if (IS_ERR(old_devices)) {
2489 kfree(seed_devices);
2490 return old_devices;
2491 }
2492
2493 list_add(&old_devices->fs_list, &fs_uuids);
2494
2495 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2496 seed_devices->opened = 1;
2497 INIT_LIST_HEAD(&seed_devices->devices);
2498 INIT_LIST_HEAD(&seed_devices->alloc_list);
2499 mutex_init(&seed_devices->device_list_mutex);
2500
2501 return seed_devices;
2502 }
2503
2504 /*
2505 * Splice seed devices into the sprout fs_devices.
2506 * Generate a new fsid for the sprouted read-write filesystem.
2507 */
2508 static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2509 struct btrfs_fs_devices *seed_devices)
2510 {
2511 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2512 struct btrfs_super_block *disk_super = fs_info->super_copy;
2513 struct btrfs_device *device;
2514 u64 super_flags;
2515
2516 /*
2517 * We are updating the fsid, the thread leading to device_list_add()
2518 * could race, so uuid_mutex is needed.
2519 */
2520 lockdep_assert_held(&uuid_mutex);
2521
2522 /*
2523 * The threads listed below may traverse dev_list but can do that without
2524 * device_list_mutex:
2525 * - All device ops and balance - as we are in btrfs_exclop_start.
2526 * - Various dev_list readers - are using RCU.
2527 * - btrfs_ioctl_fitrim() - is using RCU.
2528 *
2529 * For-read threads as below are using device_list_mutex:
2530 * - Readonly scrub btrfs_scrub_dev()
2531 * - Readonly scrub btrfs_scrub_progress()
2532 * - btrfs_get_dev_stats()
2533 */
2534 lockdep_assert_held(&fs_devices->device_list_mutex);
2535
2536 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2537 synchronize_rcu);
2538 list_for_each_entry(device, &seed_devices->devices, dev_list)
2539 device->fs_devices = seed_devices;
2540
2541 fs_devices->seeding = false;
2542 fs_devices->num_devices = 0;
2543 fs_devices->open_devices = 0;
2544 fs_devices->missing_devices = 0;
2545 fs_devices->rotating = false;
2546 list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2547
2548 generate_random_uuid(fs_devices->fsid);
2549 memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2550 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2551
2552 super_flags = btrfs_super_flags(disk_super) &
2553 ~BTRFS_SUPER_FLAG_SEEDING;
2554 btrfs_set_super_flags(disk_super, super_flags);
2555 }
2556
2557 /*
2558 * Store the expected generation for seed devices in device items.
2559 */
2560 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2561 {
2562 BTRFS_DEV_LOOKUP_ARGS(args);
2563 struct btrfs_fs_info *fs_info = trans->fs_info;
2564 struct btrfs_root *root = fs_info->chunk_root;
2565 struct btrfs_path *path;
2566 struct extent_buffer *leaf;
2567 struct btrfs_dev_item *dev_item;
2568 struct btrfs_device *device;
2569 struct btrfs_key key;
2570 u8 fs_uuid[BTRFS_FSID_SIZE];
2571 u8 dev_uuid[BTRFS_UUID_SIZE];
2572 int ret;
2573
2574 path = btrfs_alloc_path();
2575 if (!path)
2576 return -ENOMEM;
2577
2578 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2579 key.offset = 0;
2580 key.type = BTRFS_DEV_ITEM_KEY;
2581
2582 while (1) {
2583 btrfs_reserve_chunk_metadata(trans, false);
2584 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2585 btrfs_trans_release_chunk_metadata(trans);
2586 if (ret < 0)
2587 goto error;
2588
2589 leaf = path->nodes[0];
2590 next_slot:
2591 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2592 ret = btrfs_next_leaf(root, path);
2593 if (ret > 0)
2594 break;
2595 if (ret < 0)
2596 goto error;
2597 leaf = path->nodes[0];
2598 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2599 btrfs_release_path(path);
2600 continue;
2601 }
2602
2603 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2604 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2605 key.type != BTRFS_DEV_ITEM_KEY)
2606 break;
2607
2608 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2609 struct btrfs_dev_item);
2610 args.devid = btrfs_device_id(leaf, dev_item);
2611 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2612 BTRFS_UUID_SIZE);
2613 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2614 BTRFS_FSID_SIZE);
2615 args.uuid = dev_uuid;
2616 args.fsid = fs_uuid;
2617 device = btrfs_find_device(fs_info->fs_devices, &args);
2618 BUG_ON(!device); /* Logic error */
2619
2620 if (device->fs_devices->seeding) {
2621 btrfs_set_device_generation(leaf, dev_item,
2622 device->generation);
2623 btrfs_mark_buffer_dirty(trans, leaf);
2624 }
2625
2626 path->slots[0]++;
2627 goto next_slot;
2628 }
2629 ret = 0;
2630 error:
2631 btrfs_free_path(path);
2632 return ret;
2633 }
2634
2635 int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2636 {
2637 struct btrfs_root *root = fs_info->dev_root;
2638 struct btrfs_trans_handle *trans;
2639 struct btrfs_device *device;
2640 struct file *bdev_file;
2641 struct super_block *sb = fs_info->sb;
2642 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2643 struct btrfs_fs_devices *seed_devices = NULL;
2644 u64 orig_super_total_bytes;
2645 u64 orig_super_num_devices;
2646 int ret = 0;
2647 bool seeding_dev = false;
2648 bool locked = false;
2649
2650 if (sb_rdonly(sb) && !fs_devices->seeding)
2651 return -EROFS;
2652
2653 bdev_file = bdev_file_open_by_path(device_path, BLK_OPEN_WRITE,
2654 fs_info->bdev_holder, NULL);
2655 if (IS_ERR(bdev_file))
2656 return PTR_ERR(bdev_file);
2657
2658 if (!btrfs_check_device_zone_type(fs_info, file_bdev(bdev_file))) {
2659 ret = -EINVAL;
2660 goto error;
2661 }
2662
2663 if (fs_devices->seeding) {
2664 seeding_dev = true;
2665 down_write(&sb->s_umount);
2666 mutex_lock(&uuid_mutex);
2667 locked = true;
2668 }
2669
2670 sync_blockdev(file_bdev(bdev_file));
2671
2672 rcu_read_lock();
2673 list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2674 if (device->bdev == file_bdev(bdev_file)) {
2675 ret = -EEXIST;
2676 rcu_read_unlock();
2677 goto error;
2678 }
2679 }
2680 rcu_read_unlock();
2681
2682 device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2683 if (IS_ERR(device)) {
2684 /* we can safely leave the fs_devices entry around */
2685 ret = PTR_ERR(device);
2686 goto error;
2687 }
2688
2689 device->fs_info = fs_info;
2690 device->bdev_file = bdev_file;
2691 device->bdev = file_bdev(bdev_file);
2692 ret = lookup_bdev(device_path, &device->devt);
2693 if (ret)
2694 goto error_free_device;
2695
2696 ret = btrfs_get_dev_zone_info(device, false);
2697 if (ret)
2698 goto error_free_device;
2699
2700 trans = btrfs_start_transaction(root, 0);
2701 if (IS_ERR(trans)) {
2702 ret = PTR_ERR(trans);
2703 goto error_free_zone;
2704 }
2705
2706 set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2707 device->generation = trans->transid;
2708 device->io_width = fs_info->sectorsize;
2709 device->io_align = fs_info->sectorsize;
2710 device->sector_size = fs_info->sectorsize;
2711 device->total_bytes =
2712 round_down(bdev_nr_bytes(device->bdev), fs_info->sectorsize);
2713 device->disk_total_bytes = device->total_bytes;
2714 device->commit_total_bytes = device->total_bytes;
2715 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2716 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2717 device->dev_stats_valid = 1;
2718 set_blocksize(device->bdev_file, BTRFS_BDEV_BLOCKSIZE);
2719
2720 if (seeding_dev) {
2721 btrfs_clear_sb_rdonly(sb);
2722
2723 /* GFP_KERNEL allocation must not be under device_list_mutex */
2724 seed_devices = btrfs_init_sprout(fs_info);
2725 if (IS_ERR(seed_devices)) {
2726 ret = PTR_ERR(seed_devices);
2727 btrfs_abort_transaction(trans, ret);
2728 goto error_trans;
2729 }
2730 }
2731
2732 mutex_lock(&fs_devices->device_list_mutex);
2733 if (seeding_dev) {
2734 btrfs_setup_sprout(fs_info, seed_devices);
2735 btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2736 device);
2737 }
2738
2739 device->fs_devices = fs_devices;
2740
2741 mutex_lock(&fs_info->chunk_mutex);
2742 list_add_rcu(&device->dev_list, &fs_devices->devices);
2743 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2744 fs_devices->num_devices++;
2745 fs_devices->open_devices++;
2746 fs_devices->rw_devices++;
2747 fs_devices->total_devices++;
2748 fs_devices->total_rw_bytes += device->total_bytes;
2749
2750 atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2751
2752 if (!bdev_nonrot(device->bdev))
2753 fs_devices->rotating = true;
2754
2755 orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2756 btrfs_set_super_total_bytes(fs_info->super_copy,
2757 round_down(orig_super_total_bytes + device->total_bytes,
2758 fs_info->sectorsize));
2759
2760 orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2761 btrfs_set_super_num_devices(fs_info->super_copy,
2762 orig_super_num_devices + 1);
2763
2764 /*
2765 * we've got more storage, clear any full flags on the space
2766 * infos
2767 */
2768 btrfs_clear_space_info_full(fs_info);
2769
2770 mutex_unlock(&fs_info->chunk_mutex);
2771
2772 /* Add sysfs device entry */
2773 btrfs_sysfs_add_device(device);
2774
2775 mutex_unlock(&fs_devices->device_list_mutex);
2776
2777 if (seeding_dev) {
2778 mutex_lock(&fs_info->chunk_mutex);
2779 ret = init_first_rw_device(trans);
2780 mutex_unlock(&fs_info->chunk_mutex);
2781 if (ret) {
2782 btrfs_abort_transaction(trans, ret);
2783 goto error_sysfs;
2784 }
2785 }
2786
2787 ret = btrfs_add_dev_item(trans, device);
2788 if (ret) {
2789 btrfs_abort_transaction(trans, ret);
2790 goto error_sysfs;
2791 }
2792
2793 if (seeding_dev) {
2794 ret = btrfs_finish_sprout(trans);
2795 if (ret) {
2796 btrfs_abort_transaction(trans, ret);
2797 goto error_sysfs;
2798 }
2799
2800 /*
2801 * fs_devices now represents the newly sprouted filesystem and
2802 * its fsid has been changed by btrfs_sprout_splice().
2803 */
2804 btrfs_sysfs_update_sprout_fsid(fs_devices);
2805 }
2806
2807 ret = btrfs_commit_transaction(trans);
2808
2809 if (seeding_dev) {
2810 mutex_unlock(&uuid_mutex);
2811 up_write(&sb->s_umount);
2812 locked = false;
2813
2814 if (ret) /* transaction commit */
2815 return ret;
2816
2817 ret = btrfs_relocate_sys_chunks(fs_info);
2818 if (ret < 0)
2819 btrfs_handle_fs_error(fs_info, ret,
2820 "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2821 trans = btrfs_attach_transaction(root);
2822 if (IS_ERR(trans)) {
2823 if (PTR_ERR(trans) == -ENOENT)
2824 return 0;
2825 ret = PTR_ERR(trans);
2826 trans = NULL;
2827 goto error_sysfs;
2828 }
2829 ret = btrfs_commit_transaction(trans);
2830 }
2831
2832 /*
2833 * Now that we have written a new super block to this device, check all
2834 * other fs_devices list if device_path alienates any other scanned
2835 * device.
2836 * We can ignore the return value as it typically returns -EINVAL and
2837 * only succeeds if the device was an alien.
2838 */
2839 btrfs_forget_devices(device->devt);
2840
2841 /* Update ctime/mtime for blkid or udev */
2842 update_dev_time(device_path);
2843
2844 return ret;
2845
2846 error_sysfs:
2847 btrfs_sysfs_remove_device(device);
2848 mutex_lock(&fs_info->fs_devices->device_list_mutex);
2849 mutex_lock(&fs_info->chunk_mutex);
2850 list_del_rcu(&device->dev_list);
2851 list_del(&device->dev_alloc_list);
2852 fs_info->fs_devices->num_devices--;
2853 fs_info->fs_devices->open_devices--;
2854 fs_info->fs_devices->rw_devices--;
2855 fs_info->fs_devices->total_devices--;
2856 fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2857 atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2858 btrfs_set_super_total_bytes(fs_info->super_copy,
2859 orig_super_total_bytes);
2860 btrfs_set_super_num_devices(fs_info->super_copy,
2861 orig_super_num_devices);
2862 mutex_unlock(&fs_info->chunk_mutex);
2863 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2864 error_trans:
2865 if (seeding_dev)
2866 btrfs_set_sb_rdonly(sb);
2867 if (trans)
2868 btrfs_end_transaction(trans);
2869 error_free_zone:
2870 btrfs_destroy_dev_zone_info(device);
2871 error_free_device:
2872 btrfs_free_device(device);
2873 error:
2874 fput(bdev_file);
2875 if (locked) {
2876 mutex_unlock(&uuid_mutex);
2877 up_write(&sb->s_umount);
2878 }
2879 return ret;
2880 }
2881
2882 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2883 struct btrfs_device *device)
2884 {
2885 int ret;
2886 struct btrfs_path *path;
2887 struct btrfs_root *root = device->fs_info->chunk_root;
2888 struct btrfs_dev_item *dev_item;
2889 struct extent_buffer *leaf;
2890 struct btrfs_key key;
2891
2892 path = btrfs_alloc_path();
2893 if (!path)
2894 return -ENOMEM;
2895
2896 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2897 key.type = BTRFS_DEV_ITEM_KEY;
2898 key.offset = device->devid;
2899
2900 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2901 if (ret < 0)
2902 goto out;
2903
2904 if (ret > 0) {
2905 ret = -ENOENT;
2906 goto out;
2907 }
2908
2909 leaf = path->nodes[0];
2910 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2911
2912 btrfs_set_device_id(leaf, dev_item, device->devid);
2913 btrfs_set_device_type(leaf, dev_item, device->type);
2914 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2915 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2916 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2917 btrfs_set_device_total_bytes(leaf, dev_item,
2918 btrfs_device_get_disk_total_bytes(device));
2919 btrfs_set_device_bytes_used(leaf, dev_item,
2920 btrfs_device_get_bytes_used(device));
2921 btrfs_mark_buffer_dirty(trans, leaf);
2922
2923 out:
2924 btrfs_free_path(path);
2925 return ret;
2926 }
2927
2928 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2929 struct btrfs_device *device, u64 new_size)
2930 {
2931 struct btrfs_fs_info *fs_info = device->fs_info;
2932 struct btrfs_super_block *super_copy = fs_info->super_copy;
2933 u64 old_total;
2934 u64 diff;
2935 int ret;
2936
2937 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2938 return -EACCES;
2939
2940 new_size = round_down(new_size, fs_info->sectorsize);
2941
2942 mutex_lock(&fs_info->chunk_mutex);
2943 old_total = btrfs_super_total_bytes(super_copy);
2944 diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2945
2946 if (new_size <= device->total_bytes ||
2947 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2948 mutex_unlock(&fs_info->chunk_mutex);
2949 return -EINVAL;
2950 }
2951
2952 btrfs_set_super_total_bytes(super_copy,
2953 round_down(old_total + diff, fs_info->sectorsize));
2954 device->fs_devices->total_rw_bytes += diff;
2955 atomic64_add(diff, &fs_info->free_chunk_space);
2956
2957 btrfs_device_set_total_bytes(device, new_size);
2958 btrfs_device_set_disk_total_bytes(device, new_size);
2959 btrfs_clear_space_info_full(device->fs_info);
2960 if (list_empty(&device->post_commit_list))
2961 list_add_tail(&device->post_commit_list,
2962 &trans->transaction->dev_update_list);
2963 mutex_unlock(&fs_info->chunk_mutex);
2964
2965 btrfs_reserve_chunk_metadata(trans, false);
2966 ret = btrfs_update_device(trans, device);
2967 btrfs_trans_release_chunk_metadata(trans);
2968
2969 return ret;
2970 }
2971
2972 static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2973 {
2974 struct btrfs_fs_info *fs_info = trans->fs_info;
2975 struct btrfs_root *root = fs_info->chunk_root;
2976 int ret;
2977 struct btrfs_path *path;
2978 struct btrfs_key key;
2979
2980 path = btrfs_alloc_path();
2981 if (!path)
2982 return -ENOMEM;
2983
2984 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2985 key.offset = chunk_offset;
2986 key.type = BTRFS_CHUNK_ITEM_KEY;
2987
2988 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2989 if (ret < 0)
2990 goto out;
2991 else if (ret > 0) { /* Logic error or corruption */
2992 btrfs_err(fs_info, "failed to lookup chunk %llu when freeing",
2993 chunk_offset);
2994 btrfs_abort_transaction(trans, -ENOENT);
2995 ret = -EUCLEAN;
2996 goto out;
2997 }
2998
2999 ret = btrfs_del_item(trans, root, path);
3000 if (ret < 0) {
3001 btrfs_err(fs_info, "failed to delete chunk %llu item", chunk_offset);
3002 btrfs_abort_transaction(trans, ret);
3003 goto out;
3004 }
3005 out:
3006 btrfs_free_path(path);
3007 return ret;
3008 }
3009
3010 static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3011 {
3012 struct btrfs_super_block *super_copy = fs_info->super_copy;
3013 struct btrfs_disk_key *disk_key;
3014 struct btrfs_chunk *chunk;
3015 u8 *ptr;
3016 int ret = 0;
3017 u32 num_stripes;
3018 u32 array_size;
3019 u32 len = 0;
3020 u32 cur;
3021 struct btrfs_key key;
3022
3023 lockdep_assert_held(&fs_info->chunk_mutex);
3024 array_size = btrfs_super_sys_array_size(super_copy);
3025
3026 ptr = super_copy->sys_chunk_array;
3027 cur = 0;
3028
3029 while (cur < array_size) {
3030 disk_key = (struct btrfs_disk_key *)ptr;
3031 btrfs_disk_key_to_cpu(&key, disk_key);
3032
3033 len = sizeof(*disk_key);
3034
3035 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3036 chunk = (struct btrfs_chunk *)(ptr + len);
3037 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3038 len += btrfs_chunk_item_size(num_stripes);
3039 } else {
3040 ret = -EIO;
3041 break;
3042 }
3043 if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3044 key.offset == chunk_offset) {
3045 memmove(ptr, ptr + len, array_size - (cur + len));
3046 array_size -= len;
3047 btrfs_set_super_sys_array_size(super_copy, array_size);
3048 } else {
3049 ptr += len;
3050 cur += len;
3051 }
3052 }
3053 return ret;
3054 }
3055
3056 struct btrfs_chunk_map *btrfs_find_chunk_map_nolock(struct btrfs_fs_info *fs_info,
3057 u64 logical, u64 length)
3058 {
3059 struct rb_node *node = fs_info->mapping_tree.rb_root.rb_node;
3060 struct rb_node *prev = NULL;
3061 struct rb_node *orig_prev;
3062 struct btrfs_chunk_map *map;
3063 struct btrfs_chunk_map *prev_map = NULL;
3064
3065 while (node) {
3066 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
3067 prev = node;
3068 prev_map = map;
3069
3070 if (logical < map->start) {
3071 node = node->rb_left;
3072 } else if (logical >= map->start + map->chunk_len) {
3073 node = node->rb_right;
3074 } else {
3075 refcount_inc(&map->refs);
3076 return map;
3077 }
3078 }
3079
3080 if (!prev)
3081 return NULL;
3082
3083 orig_prev = prev;
3084 while (prev && logical >= prev_map->start + prev_map->chunk_len) {
3085 prev = rb_next(prev);
3086 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3087 }
3088
3089 if (!prev) {
3090 prev = orig_prev;
3091 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3092 while (prev && logical < prev_map->start) {
3093 prev = rb_prev(prev);
3094 prev_map = rb_entry(prev, struct btrfs_chunk_map, rb_node);
3095 }
3096 }
3097
3098 if (prev) {
3099 u64 end = logical + length;
3100
3101 /*
3102 * Caller can pass a U64_MAX length when it wants to get any
3103 * chunk starting at an offset of 'logical' or higher, so deal
3104 * with underflow by resetting the end offset to U64_MAX.
3105 */
3106 if (end < logical)
3107 end = U64_MAX;
3108
3109 if (end > prev_map->start &&
3110 logical < prev_map->start + prev_map->chunk_len) {
3111 refcount_inc(&prev_map->refs);
3112 return prev_map;
3113 }
3114 }
3115
3116 return NULL;
3117 }
3118
3119 struct btrfs_chunk_map *btrfs_find_chunk_map(struct btrfs_fs_info *fs_info,
3120 u64 logical, u64 length)
3121 {
3122 struct btrfs_chunk_map *map;
3123
3124 read_lock(&fs_info->mapping_tree_lock);
3125 map = btrfs_find_chunk_map_nolock(fs_info, logical, length);
3126 read_unlock(&fs_info->mapping_tree_lock);
3127
3128 return map;
3129 }
3130
3131 /*
3132 * Find the mapping containing the given logical extent.
3133 *
3134 * @logical: Logical block offset in bytes.
3135 * @length: Length of extent in bytes.
3136 *
3137 * Return: Chunk mapping or ERR_PTR.
3138 */
3139 struct btrfs_chunk_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3140 u64 logical, u64 length)
3141 {
3142 struct btrfs_chunk_map *map;
3143
3144 map = btrfs_find_chunk_map(fs_info, logical, length);
3145
3146 if (unlikely(!map)) {
3147 btrfs_crit(fs_info,
3148 "unable to find chunk map for logical %llu length %llu",
3149 logical, length);
3150 return ERR_PTR(-EINVAL);
3151 }
3152
3153 if (unlikely(map->start > logical || map->start + map->chunk_len <= logical)) {
3154 btrfs_crit(fs_info,
3155 "found a bad chunk map, wanted %llu-%llu, found %llu-%llu",
3156 logical, logical + length, map->start,
3157 map->start + map->chunk_len);
3158 btrfs_free_chunk_map(map);
3159 return ERR_PTR(-EINVAL);
3160 }
3161
3162 /* Callers are responsible for dropping the reference. */
3163 return map;
3164 }
3165
3166 static int remove_chunk_item(struct btrfs_trans_handle *trans,
3167 struct btrfs_chunk_map *map, u64 chunk_offset)
3168 {
3169 int i;
3170
3171 /*
3172 * Removing chunk items and updating the device items in the chunks btree
3173 * requires holding the chunk_mutex.
3174 * See the comment at btrfs_chunk_alloc() for the details.
3175 */
3176 lockdep_assert_held(&trans->fs_info->chunk_mutex);
3177
3178 for (i = 0; i < map->num_stripes; i++) {
3179 int ret;
3180
3181 ret = btrfs_update_device(trans, map->stripes[i].dev);
3182 if (ret)
3183 return ret;
3184 }
3185
3186 return btrfs_free_chunk(trans, chunk_offset);
3187 }
3188
3189 int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3190 {
3191 struct btrfs_fs_info *fs_info = trans->fs_info;
3192 struct btrfs_chunk_map *map;
3193 u64 dev_extent_len = 0;
3194 int i, ret = 0;
3195 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3196
3197 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3198 if (IS_ERR(map)) {
3199 /*
3200 * This is a logic error, but we don't want to just rely on the
3201 * user having built with ASSERT enabled, so if ASSERT doesn't
3202 * do anything we still error out.
3203 */
3204 ASSERT(0);
3205 return PTR_ERR(map);
3206 }
3207
3208 /*
3209 * First delete the device extent items from the devices btree.
3210 * We take the device_list_mutex to avoid racing with the finishing phase
3211 * of a device replace operation. See the comment below before acquiring
3212 * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3213 * because that can result in a deadlock when deleting the device extent
3214 * items from the devices btree - COWing an extent buffer from the btree
3215 * may result in allocating a new metadata chunk, which would attempt to
3216 * lock again fs_info->chunk_mutex.
3217 */
3218 mutex_lock(&fs_devices->device_list_mutex);
3219 for (i = 0; i < map->num_stripes; i++) {
3220 struct btrfs_device *device = map->stripes[i].dev;
3221 ret = btrfs_free_dev_extent(trans, device,
3222 map->stripes[i].physical,
3223 &dev_extent_len);
3224 if (ret) {
3225 mutex_unlock(&fs_devices->device_list_mutex);
3226 btrfs_abort_transaction(trans, ret);
3227 goto out;
3228 }
3229
3230 if (device->bytes_used > 0) {
3231 mutex_lock(&fs_info->chunk_mutex);
3232 btrfs_device_set_bytes_used(device,
3233 device->bytes_used - dev_extent_len);
3234 atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3235 btrfs_clear_space_info_full(fs_info);
3236 mutex_unlock(&fs_info->chunk_mutex);
3237 }
3238 }
3239 mutex_unlock(&fs_devices->device_list_mutex);
3240
3241 /*
3242 * We acquire fs_info->chunk_mutex for 2 reasons:
3243 *
3244 * 1) Just like with the first phase of the chunk allocation, we must
3245 * reserve system space, do all chunk btree updates and deletions, and
3246 * update the system chunk array in the superblock while holding this
3247 * mutex. This is for similar reasons as explained on the comment at
3248 * the top of btrfs_chunk_alloc();
3249 *
3250 * 2) Prevent races with the final phase of a device replace operation
3251 * that replaces the device object associated with the map's stripes,
3252 * because the device object's id can change at any time during that
3253 * final phase of the device replace operation
3254 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3255 * replaced device and then see it with an ID of
3256 * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3257 * the device item, which does not exists on the chunk btree.
3258 * The finishing phase of device replace acquires both the
3259 * device_list_mutex and the chunk_mutex, in that order, so we are
3260 * safe by just acquiring the chunk_mutex.
3261 */
3262 trans->removing_chunk = true;
3263 mutex_lock(&fs_info->chunk_mutex);
3264
3265 check_system_chunk(trans, map->type);
3266
3267 ret = remove_chunk_item(trans, map, chunk_offset);
3268 /*
3269 * Normally we should not get -ENOSPC since we reserved space before
3270 * through the call to check_system_chunk().
3271 *
3272 * Despite our system space_info having enough free space, we may not
3273 * be able to allocate extents from its block groups, because all have
3274 * an incompatible profile, which will force us to allocate a new system
3275 * block group with the right profile, or right after we called
3276 * check_system_space() above, a scrub turned the only system block group
3277 * with enough free space into RO mode.
3278 * This is explained with more detail at do_chunk_alloc().
3279 *
3280 * So if we get -ENOSPC, allocate a new system chunk and retry once.
3281 */
3282 if (ret == -ENOSPC) {
3283 const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3284 struct btrfs_block_group *sys_bg;
3285
3286 sys_bg = btrfs_create_chunk(trans, sys_flags);
3287 if (IS_ERR(sys_bg)) {
3288 ret = PTR_ERR(sys_bg);
3289 btrfs_abort_transaction(trans, ret);
3290 goto out;
3291 }
3292
3293 ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3294 if (ret) {
3295 btrfs_abort_transaction(trans, ret);
3296 goto out;
3297 }
3298
3299 ret = remove_chunk_item(trans, map, chunk_offset);
3300 if (ret) {
3301 btrfs_abort_transaction(trans, ret);
3302 goto out;
3303 }
3304 } else if (ret) {
3305 btrfs_abort_transaction(trans, ret);
3306 goto out;
3307 }
3308
3309 trace_btrfs_chunk_free(fs_info, map, chunk_offset, map->chunk_len);
3310
3311 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3312 ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3313 if (ret) {
3314 btrfs_abort_transaction(trans, ret);
3315 goto out;
3316 }
3317 }
3318
3319 mutex_unlock(&fs_info->chunk_mutex);
3320 trans->removing_chunk = false;
3321
3322 /*
3323 * We are done with chunk btree updates and deletions, so release the
3324 * system space we previously reserved (with check_system_chunk()).
3325 */
3326 btrfs_trans_release_chunk_metadata(trans);
3327
3328 ret = btrfs_remove_block_group(trans, map);
3329 if (ret) {
3330 btrfs_abort_transaction(trans, ret);
3331 goto out;
3332 }
3333
3334 out:
3335 if (trans->removing_chunk) {
3336 mutex_unlock(&fs_info->chunk_mutex);
3337 trans->removing_chunk = false;
3338 }
3339 /* once for us */
3340 btrfs_free_chunk_map(map);
3341 return ret;
3342 }
3343
3344 int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3345 {
3346 struct btrfs_root *root = fs_info->chunk_root;
3347 struct btrfs_trans_handle *trans;
3348 struct btrfs_block_group *block_group;
3349 u64 length;
3350 int ret;
3351
3352 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3353 btrfs_err(fs_info,
3354 "relocate: not supported on extent tree v2 yet");
3355 return -EINVAL;
3356 }
3357
3358 /*
3359 * Prevent races with automatic removal of unused block groups.
3360 * After we relocate and before we remove the chunk with offset
3361 * chunk_offset, automatic removal of the block group can kick in,
3362 * resulting in a failure when calling btrfs_remove_chunk() below.
3363 *
3364 * Make sure to acquire this mutex before doing a tree search (dev
3365 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3366 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3367 * we release the path used to search the chunk/dev tree and before
3368 * the current task acquires this mutex and calls us.
3369 */
3370 lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3371
3372 /* step one, relocate all the extents inside this chunk */
3373 btrfs_scrub_pause(fs_info);
3374 ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3375 btrfs_scrub_continue(fs_info);
3376 if (ret) {
3377 /*
3378 * If we had a transaction abort, stop all running scrubs.
3379 * See transaction.c:cleanup_transaction() why we do it here.
3380 */
3381 if (BTRFS_FS_ERROR(fs_info))
3382 btrfs_scrub_cancel(fs_info);
3383 return ret;
3384 }
3385
3386 block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3387 if (!block_group)
3388 return -ENOENT;
3389 btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3390 length = block_group->length;
3391 btrfs_put_block_group(block_group);
3392
3393 /*
3394 * On a zoned file system, discard the whole block group, this will
3395 * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3396 * resetting the zone fails, don't treat it as a fatal problem from the
3397 * filesystem's point of view.
3398 */
3399 if (btrfs_is_zoned(fs_info)) {
3400 ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3401 if (ret)
3402 btrfs_info(fs_info,
3403 "failed to reset zone %llu after relocation",
3404 chunk_offset);
3405 }
3406
3407 trans = btrfs_start_trans_remove_block_group(root->fs_info,
3408 chunk_offset);
3409 if (IS_ERR(trans)) {
3410 ret = PTR_ERR(trans);
3411 btrfs_handle_fs_error(root->fs_info, ret, NULL);
3412 return ret;
3413 }
3414
3415 /*
3416 * step two, delete the device extents and the
3417 * chunk tree entries
3418 */
3419 ret = btrfs_remove_chunk(trans, chunk_offset);
3420 btrfs_end_transaction(trans);
3421 return ret;
3422 }
3423
3424 static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3425 {
3426 struct btrfs_root *chunk_root = fs_info->chunk_root;
3427 struct btrfs_path *path;
3428 struct extent_buffer *leaf;
3429 struct btrfs_chunk *chunk;
3430 struct btrfs_key key;
3431 struct btrfs_key found_key;
3432 u64 chunk_type;
3433 bool retried = false;
3434 int failed = 0;
3435 int ret;
3436
3437 path = btrfs_alloc_path();
3438 if (!path)
3439 return -ENOMEM;
3440
3441 again:
3442 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3443 key.offset = (u64)-1;
3444 key.type = BTRFS_CHUNK_ITEM_KEY;
3445
3446 while (1) {
3447 mutex_lock(&fs_info->reclaim_bgs_lock);
3448 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3449 if (ret < 0) {
3450 mutex_unlock(&fs_info->reclaim_bgs_lock);
3451 goto error;
3452 }
3453 if (ret == 0) {
3454 /*
3455 * On the first search we would find chunk tree with
3456 * offset -1, which is not possible. On subsequent
3457 * loops this would find an existing item on an invalid
3458 * offset (one less than the previous one, wrong
3459 * alignment and size).
3460 */
3461 ret = -EUCLEAN;
3462 mutex_unlock(&fs_info->reclaim_bgs_lock);
3463 goto error;
3464 }
3465
3466 ret = btrfs_previous_item(chunk_root, path, key.objectid,
3467 key.type);
3468 if (ret)
3469 mutex_unlock(&fs_info->reclaim_bgs_lock);
3470 if (ret < 0)
3471 goto error;
3472 if (ret > 0)
3473 break;
3474
3475 leaf = path->nodes[0];
3476 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3477
3478 chunk = btrfs_item_ptr(leaf, path->slots[0],
3479 struct btrfs_chunk);
3480 chunk_type = btrfs_chunk_type(leaf, chunk);
3481 btrfs_release_path(path);
3482
3483 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3484 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3485 if (ret == -ENOSPC)
3486 failed++;
3487 else
3488 BUG_ON(ret);
3489 }
3490 mutex_unlock(&fs_info->reclaim_bgs_lock);
3491
3492 if (found_key.offset == 0)
3493 break;
3494 key.offset = found_key.offset - 1;
3495 }
3496 ret = 0;
3497 if (failed && !retried) {
3498 failed = 0;
3499 retried = true;
3500 goto again;
3501 } else if (WARN_ON(failed && retried)) {
3502 ret = -ENOSPC;
3503 }
3504 error:
3505 btrfs_free_path(path);
3506 return ret;
3507 }
3508
3509 /*
3510 * return 1 : allocate a data chunk successfully,
3511 * return <0: errors during allocating a data chunk,
3512 * return 0 : no need to allocate a data chunk.
3513 */
3514 static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3515 u64 chunk_offset)
3516 {
3517 struct btrfs_block_group *cache;
3518 u64 bytes_used;
3519 u64 chunk_type;
3520
3521 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3522 ASSERT(cache);
3523 chunk_type = cache->flags;
3524 btrfs_put_block_group(cache);
3525
3526 if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3527 return 0;
3528
3529 spin_lock(&fs_info->data_sinfo->lock);
3530 bytes_used = fs_info->data_sinfo->bytes_used;
3531 spin_unlock(&fs_info->data_sinfo->lock);
3532
3533 if (!bytes_used) {
3534 struct btrfs_trans_handle *trans;
3535 int ret;
3536
3537 trans = btrfs_join_transaction(fs_info->tree_root);
3538 if (IS_ERR(trans))
3539 return PTR_ERR(trans);
3540
3541 ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3542 btrfs_end_transaction(trans);
3543 if (ret < 0)
3544 return ret;
3545 return 1;
3546 }
3547
3548 return 0;
3549 }
3550
3551 static void btrfs_disk_balance_args_to_cpu(struct btrfs_balance_args *cpu,
3552 const struct btrfs_disk_balance_args *disk)
3553 {
3554 memset(cpu, 0, sizeof(*cpu));
3555
3556 cpu->profiles = le64_to_cpu(disk->profiles);
3557 cpu->usage = le64_to_cpu(disk->usage);
3558 cpu->devid = le64_to_cpu(disk->devid);
3559 cpu->pstart = le64_to_cpu(disk->pstart);
3560 cpu->pend = le64_to_cpu(disk->pend);
3561 cpu->vstart = le64_to_cpu(disk->vstart);
3562 cpu->vend = le64_to_cpu(disk->vend);
3563 cpu->target = le64_to_cpu(disk->target);
3564 cpu->flags = le64_to_cpu(disk->flags);
3565 cpu->limit = le64_to_cpu(disk->limit);
3566 cpu->stripes_min = le32_to_cpu(disk->stripes_min);
3567 cpu->stripes_max = le32_to_cpu(disk->stripes_max);
3568 }
3569
3570 static void btrfs_cpu_balance_args_to_disk(struct btrfs_disk_balance_args *disk,
3571 const struct btrfs_balance_args *cpu)
3572 {
3573 memset(disk, 0, sizeof(*disk));
3574
3575 disk->profiles = cpu_to_le64(cpu->profiles);
3576 disk->usage = cpu_to_le64(cpu->usage);
3577 disk->devid = cpu_to_le64(cpu->devid);
3578 disk->pstart = cpu_to_le64(cpu->pstart);
3579 disk->pend = cpu_to_le64(cpu->pend);
3580 disk->vstart = cpu_to_le64(cpu->vstart);
3581 disk->vend = cpu_to_le64(cpu->vend);
3582 disk->target = cpu_to_le64(cpu->target);
3583 disk->flags = cpu_to_le64(cpu->flags);
3584 disk->limit = cpu_to_le64(cpu->limit);
3585 disk->stripes_min = cpu_to_le32(cpu->stripes_min);
3586 disk->stripes_max = cpu_to_le32(cpu->stripes_max);
3587 }
3588
3589 static int insert_balance_item(struct btrfs_fs_info *fs_info,
3590 struct btrfs_balance_control *bctl)
3591 {
3592 struct btrfs_root *root = fs_info->tree_root;
3593 struct btrfs_trans_handle *trans;
3594 struct btrfs_balance_item *item;
3595 struct btrfs_disk_balance_args disk_bargs;
3596 struct btrfs_path *path;
3597 struct extent_buffer *leaf;
3598 struct btrfs_key key;
3599 int ret, err;
3600
3601 path = btrfs_alloc_path();
3602 if (!path)
3603 return -ENOMEM;
3604
3605 trans = btrfs_start_transaction(root, 0);
3606 if (IS_ERR(trans)) {
3607 btrfs_free_path(path);
3608 return PTR_ERR(trans);
3609 }
3610
3611 key.objectid = BTRFS_BALANCE_OBJECTID;
3612 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3613 key.offset = 0;
3614
3615 ret = btrfs_insert_empty_item(trans, root, path, &key,
3616 sizeof(*item));
3617 if (ret)
3618 goto out;
3619
3620 leaf = path->nodes[0];
3621 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3622
3623 memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3624
3625 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3626 btrfs_set_balance_data(leaf, item, &disk_bargs);
3627 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3628 btrfs_set_balance_meta(leaf, item, &disk_bargs);
3629 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3630 btrfs_set_balance_sys(leaf, item, &disk_bargs);
3631
3632 btrfs_set_balance_flags(leaf, item, bctl->flags);
3633
3634 btrfs_mark_buffer_dirty(trans, leaf);
3635 out:
3636 btrfs_free_path(path);
3637 err = btrfs_commit_transaction(trans);
3638 if (err && !ret)
3639 ret = err;
3640 return ret;
3641 }
3642
3643 static int del_balance_item(struct btrfs_fs_info *fs_info)
3644 {
3645 struct btrfs_root *root = fs_info->tree_root;
3646 struct btrfs_trans_handle *trans;
3647 struct btrfs_path *path;
3648 struct btrfs_key key;
3649 int ret, err;
3650
3651 path = btrfs_alloc_path();
3652 if (!path)
3653 return -ENOMEM;
3654
3655 trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3656 if (IS_ERR(trans)) {
3657 btrfs_free_path(path);
3658 return PTR_ERR(trans);
3659 }
3660
3661 key.objectid = BTRFS_BALANCE_OBJECTID;
3662 key.type = BTRFS_TEMPORARY_ITEM_KEY;
3663 key.offset = 0;
3664
3665 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3666 if (ret < 0)
3667 goto out;
3668 if (ret > 0) {
3669 ret = -ENOENT;
3670 goto out;
3671 }
3672
3673 ret = btrfs_del_item(trans, root, path);
3674 out:
3675 btrfs_free_path(path);
3676 err = btrfs_commit_transaction(trans);
3677 if (err && !ret)
3678 ret = err;
3679 return ret;
3680 }
3681
3682 /*
3683 * This is a heuristic used to reduce the number of chunks balanced on
3684 * resume after balance was interrupted.
3685 */
3686 static void update_balance_args(struct btrfs_balance_control *bctl)
3687 {
3688 /*
3689 * Turn on soft mode for chunk types that were being converted.
3690 */
3691 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3692 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3693 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3694 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3695 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3696 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3697
3698 /*
3699 * Turn on usage filter if is not already used. The idea is
3700 * that chunks that we have already balanced should be
3701 * reasonably full. Don't do it for chunks that are being
3702 * converted - that will keep us from relocating unconverted
3703 * (albeit full) chunks.
3704 */
3705 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3706 !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3707 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3708 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3709 bctl->data.usage = 90;
3710 }
3711 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3712 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3713 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3714 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3715 bctl->sys.usage = 90;
3716 }
3717 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3718 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3719 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3720 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3721 bctl->meta.usage = 90;
3722 }
3723 }
3724
3725 /*
3726 * Clear the balance status in fs_info and delete the balance item from disk.
3727 */
3728 static void reset_balance_state(struct btrfs_fs_info *fs_info)
3729 {
3730 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3731 int ret;
3732
3733 ASSERT(fs_info->balance_ctl);
3734
3735 spin_lock(&fs_info->balance_lock);
3736 fs_info->balance_ctl = NULL;
3737 spin_unlock(&fs_info->balance_lock);
3738
3739 kfree(bctl);
3740 ret = del_balance_item(fs_info);
3741 if (ret)
3742 btrfs_handle_fs_error(fs_info, ret, NULL);
3743 }
3744
3745 /*
3746 * Balance filters. Return 1 if chunk should be filtered out
3747 * (should not be balanced).
3748 */
3749 static int chunk_profiles_filter(u64 chunk_type,
3750 struct btrfs_balance_args *bargs)
3751 {
3752 chunk_type = chunk_to_extended(chunk_type) &
3753 BTRFS_EXTENDED_PROFILE_MASK;
3754
3755 if (bargs->profiles & chunk_type)
3756 return 0;
3757
3758 return 1;
3759 }
3760
3761 static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3762 struct btrfs_balance_args *bargs)
3763 {
3764 struct btrfs_block_group *cache;
3765 u64 chunk_used;
3766 u64 user_thresh_min;
3767 u64 user_thresh_max;
3768 int ret = 1;
3769
3770 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3771 chunk_used = cache->used;
3772
3773 if (bargs->usage_min == 0)
3774 user_thresh_min = 0;
3775 else
3776 user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3777
3778 if (bargs->usage_max == 0)
3779 user_thresh_max = 1;
3780 else if (bargs->usage_max > 100)
3781 user_thresh_max = cache->length;
3782 else
3783 user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3784
3785 if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3786 ret = 0;
3787
3788 btrfs_put_block_group(cache);
3789 return ret;
3790 }
3791
3792 static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3793 u64 chunk_offset, struct btrfs_balance_args *bargs)
3794 {
3795 struct btrfs_block_group *cache;
3796 u64 chunk_used, user_thresh;
3797 int ret = 1;
3798
3799 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3800 chunk_used = cache->used;
3801
3802 if (bargs->usage_min == 0)
3803 user_thresh = 1;
3804 else if (bargs->usage > 100)
3805 user_thresh = cache->length;
3806 else
3807 user_thresh = mult_perc(cache->length, bargs->usage);
3808
3809 if (chunk_used < user_thresh)
3810 ret = 0;
3811
3812 btrfs_put_block_group(cache);
3813 return ret;
3814 }
3815
3816 static int chunk_devid_filter(struct extent_buffer *leaf,
3817 struct btrfs_chunk *chunk,
3818 struct btrfs_balance_args *bargs)
3819 {
3820 struct btrfs_stripe *stripe;
3821 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3822 int i;
3823
3824 for (i = 0; i < num_stripes; i++) {
3825 stripe = btrfs_stripe_nr(chunk, i);
3826 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3827 return 0;
3828 }
3829
3830 return 1;
3831 }
3832
3833 static u64 calc_data_stripes(u64 type, int num_stripes)
3834 {
3835 const int index = btrfs_bg_flags_to_raid_index(type);
3836 const int ncopies = btrfs_raid_array[index].ncopies;
3837 const int nparity = btrfs_raid_array[index].nparity;
3838
3839 return (num_stripes - nparity) / ncopies;
3840 }
3841
3842 /* [pstart, pend) */
3843 static int chunk_drange_filter(struct extent_buffer *leaf,
3844 struct btrfs_chunk *chunk,
3845 struct btrfs_balance_args *bargs)
3846 {
3847 struct btrfs_stripe *stripe;
3848 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3849 u64 stripe_offset;
3850 u64 stripe_length;
3851 u64 type;
3852 int factor;
3853 int i;
3854
3855 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3856 return 0;
3857
3858 type = btrfs_chunk_type(leaf, chunk);
3859 factor = calc_data_stripes(type, num_stripes);
3860
3861 for (i = 0; i < num_stripes; i++) {
3862 stripe = btrfs_stripe_nr(chunk, i);
3863 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3864 continue;
3865
3866 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3867 stripe_length = btrfs_chunk_length(leaf, chunk);
3868 stripe_length = div_u64(stripe_length, factor);
3869
3870 if (stripe_offset < bargs->pend &&
3871 stripe_offset + stripe_length > bargs->pstart)
3872 return 0;
3873 }
3874
3875 return 1;
3876 }
3877
3878 /* [vstart, vend) */
3879 static int chunk_vrange_filter(struct extent_buffer *leaf,
3880 struct btrfs_chunk *chunk,
3881 u64 chunk_offset,
3882 struct btrfs_balance_args *bargs)
3883 {
3884 if (chunk_offset < bargs->vend &&
3885 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3886 /* at least part of the chunk is inside this vrange */
3887 return 0;
3888
3889 return 1;
3890 }
3891
3892 static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3893 struct btrfs_chunk *chunk,
3894 struct btrfs_balance_args *bargs)
3895 {
3896 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3897
3898 if (bargs->stripes_min <= num_stripes
3899 && num_stripes <= bargs->stripes_max)
3900 return 0;
3901
3902 return 1;
3903 }
3904
3905 static int chunk_soft_convert_filter(u64 chunk_type,
3906 struct btrfs_balance_args *bargs)
3907 {
3908 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3909 return 0;
3910
3911 chunk_type = chunk_to_extended(chunk_type) &
3912 BTRFS_EXTENDED_PROFILE_MASK;
3913
3914 if (bargs->target == chunk_type)
3915 return 1;
3916
3917 return 0;
3918 }
3919
3920 static int should_balance_chunk(struct extent_buffer *leaf,
3921 struct btrfs_chunk *chunk, u64 chunk_offset)
3922 {
3923 struct btrfs_fs_info *fs_info = leaf->fs_info;
3924 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3925 struct btrfs_balance_args *bargs = NULL;
3926 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3927
3928 /* type filter */
3929 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3930 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3931 return 0;
3932 }
3933
3934 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3935 bargs = &bctl->data;
3936 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3937 bargs = &bctl->sys;
3938 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3939 bargs = &bctl->meta;
3940
3941 /* profiles filter */
3942 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3943 chunk_profiles_filter(chunk_type, bargs)) {
3944 return 0;
3945 }
3946
3947 /* usage filter */
3948 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3949 chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3950 return 0;
3951 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3952 chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3953 return 0;
3954 }
3955
3956 /* devid filter */
3957 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3958 chunk_devid_filter(leaf, chunk, bargs)) {
3959 return 0;
3960 }
3961
3962 /* drange filter, makes sense only with devid filter */
3963 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3964 chunk_drange_filter(leaf, chunk, bargs)) {
3965 return 0;
3966 }
3967
3968 /* vrange filter */
3969 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3970 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3971 return 0;
3972 }
3973
3974 /* stripes filter */
3975 if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3976 chunk_stripes_range_filter(leaf, chunk, bargs)) {
3977 return 0;
3978 }
3979
3980 /* soft profile changing mode */
3981 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3982 chunk_soft_convert_filter(chunk_type, bargs)) {
3983 return 0;
3984 }
3985
3986 /*
3987 * limited by count, must be the last filter
3988 */
3989 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3990 if (bargs->limit == 0)
3991 return 0;
3992 else
3993 bargs->limit--;
3994 } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3995 /*
3996 * Same logic as the 'limit' filter; the minimum cannot be
3997 * determined here because we do not have the global information
3998 * about the count of all chunks that satisfy the filters.
3999 */
4000 if (bargs->limit_max == 0)
4001 return 0;
4002 else
4003 bargs->limit_max--;
4004 }
4005
4006 return 1;
4007 }
4008
4009 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
4010 {
4011 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4012 struct btrfs_root *chunk_root = fs_info->chunk_root;
4013 u64 chunk_type;
4014 struct btrfs_chunk *chunk;
4015 struct btrfs_path *path = NULL;
4016 struct btrfs_key key;
4017 struct btrfs_key found_key;
4018 struct extent_buffer *leaf;
4019 int slot;
4020 int ret;
4021 int enospc_errors = 0;
4022 bool counting = true;
4023 /* The single value limit and min/max limits use the same bytes in the */
4024 u64 limit_data = bctl->data.limit;
4025 u64 limit_meta = bctl->meta.limit;
4026 u64 limit_sys = bctl->sys.limit;
4027 u32 count_data = 0;
4028 u32 count_meta = 0;
4029 u32 count_sys = 0;
4030 int chunk_reserved = 0;
4031
4032 path = btrfs_alloc_path();
4033 if (!path) {
4034 ret = -ENOMEM;
4035 goto error;
4036 }
4037
4038 /* zero out stat counters */
4039 spin_lock(&fs_info->balance_lock);
4040 memset(&bctl->stat, 0, sizeof(bctl->stat));
4041 spin_unlock(&fs_info->balance_lock);
4042 again:
4043 if (!counting) {
4044 /*
4045 * The single value limit and min/max limits use the same bytes
4046 * in the
4047 */
4048 bctl->data.limit = limit_data;
4049 bctl->meta.limit = limit_meta;
4050 bctl->sys.limit = limit_sys;
4051 }
4052 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4053 key.offset = (u64)-1;
4054 key.type = BTRFS_CHUNK_ITEM_KEY;
4055
4056 while (1) {
4057 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
4058 atomic_read(&fs_info->balance_cancel_req)) {
4059 ret = -ECANCELED;
4060 goto error;
4061 }
4062
4063 mutex_lock(&fs_info->reclaim_bgs_lock);
4064 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
4065 if (ret < 0) {
4066 mutex_unlock(&fs_info->reclaim_bgs_lock);
4067 goto error;
4068 }
4069
4070 /*
4071 * this shouldn't happen, it means the last relocate
4072 * failed
4073 */
4074 if (ret == 0)
4075 BUG(); /* FIXME break ? */
4076
4077 ret = btrfs_previous_item(chunk_root, path, 0,
4078 BTRFS_CHUNK_ITEM_KEY);
4079 if (ret) {
4080 mutex_unlock(&fs_info->reclaim_bgs_lock);
4081 ret = 0;
4082 break;
4083 }
4084
4085 leaf = path->nodes[0];
4086 slot = path->slots[0];
4087 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4088
4089 if (found_key.objectid != key.objectid) {
4090 mutex_unlock(&fs_info->reclaim_bgs_lock);
4091 break;
4092 }
4093
4094 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4095 chunk_type = btrfs_chunk_type(leaf, chunk);
4096
4097 if (!counting) {
4098 spin_lock(&fs_info->balance_lock);
4099 bctl->stat.considered++;
4100 spin_unlock(&fs_info->balance_lock);
4101 }
4102
4103 ret = should_balance_chunk(leaf, chunk, found_key.offset);
4104
4105 btrfs_release_path(path);
4106 if (!ret) {
4107 mutex_unlock(&fs_info->reclaim_bgs_lock);
4108 goto loop;
4109 }
4110
4111 if (counting) {
4112 mutex_unlock(&fs_info->reclaim_bgs_lock);
4113 spin_lock(&fs_info->balance_lock);
4114 bctl->stat.expected++;
4115 spin_unlock(&fs_info->balance_lock);
4116
4117 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
4118 count_data++;
4119 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
4120 count_sys++;
4121 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
4122 count_meta++;
4123
4124 goto loop;
4125 }
4126
4127 /*
4128 * Apply limit_min filter, no need to check if the LIMITS
4129 * filter is used, limit_min is 0 by default
4130 */
4131 if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
4132 count_data < bctl->data.limit_min)
4133 || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
4134 count_meta < bctl->meta.limit_min)
4135 || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
4136 count_sys < bctl->sys.limit_min)) {
4137 mutex_unlock(&fs_info->reclaim_bgs_lock);
4138 goto loop;
4139 }
4140
4141 if (!chunk_reserved) {
4142 /*
4143 * We may be relocating the only data chunk we have,
4144 * which could potentially end up with losing data's
4145 * raid profile, so lets allocate an empty one in
4146 * advance.
4147 */
4148 ret = btrfs_may_alloc_data_chunk(fs_info,
4149 found_key.offset);
4150 if (ret < 0) {
4151 mutex_unlock(&fs_info->reclaim_bgs_lock);
4152 goto error;
4153 } else if (ret == 1) {
4154 chunk_reserved = 1;
4155 }
4156 }
4157
4158 ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4159 mutex_unlock(&fs_info->reclaim_bgs_lock);
4160 if (ret == -ENOSPC) {
4161 enospc_errors++;
4162 } else if (ret == -ETXTBSY) {
4163 btrfs_info(fs_info,
4164 "skipping relocation of block group %llu due to active swapfile",
4165 found_key.offset);
4166 ret = 0;
4167 } else if (ret) {
4168 goto error;
4169 } else {
4170 spin_lock(&fs_info->balance_lock);
4171 bctl->stat.completed++;
4172 spin_unlock(&fs_info->balance_lock);
4173 }
4174 loop:
4175 if (found_key.offset == 0)
4176 break;
4177 key.offset = found_key.offset - 1;
4178 }
4179
4180 if (counting) {
4181 btrfs_release_path(path);
4182 counting = false;
4183 goto again;
4184 }
4185 error:
4186 btrfs_free_path(path);
4187 if (enospc_errors) {
4188 btrfs_info(fs_info, "%d enospc errors during balance",
4189 enospc_errors);
4190 if (!ret)
4191 ret = -ENOSPC;
4192 }
4193
4194 return ret;
4195 }
4196
4197 /*
4198 * See if a given profile is valid and reduced.
4199 *
4200 * @flags: profile to validate
4201 * @extended: if true @flags is treated as an extended profile
4202 */
4203 static int alloc_profile_is_valid(u64 flags, int extended)
4204 {
4205 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4206 BTRFS_BLOCK_GROUP_PROFILE_MASK);
4207
4208 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4209
4210 /* 1) check that all other bits are zeroed */
4211 if (flags & ~mask)
4212 return 0;
4213
4214 /* 2) see if profile is reduced */
4215 if (flags == 0)
4216 return !extended; /* "" is valid for usual profiles */
4217
4218 return has_single_bit_set(flags);
4219 }
4220
4221 /*
4222 * Validate target profile against allowed profiles and return true if it's OK.
4223 * Otherwise print the error message and return false.
4224 */
4225 static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4226 const struct btrfs_balance_args *bargs,
4227 u64 allowed, const char *type)
4228 {
4229 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4230 return true;
4231
4232 /* Profile is valid and does not have bits outside of the allowed set */
4233 if (alloc_profile_is_valid(bargs->target, 1) &&
4234 (bargs->target & ~allowed) == 0)
4235 return true;
4236
4237 btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4238 type, btrfs_bg_type_to_raid_name(bargs->target));
4239 return false;
4240 }
4241
4242 /*
4243 * Fill @buf with textual description of balance filter flags @bargs, up to
4244 * @size_buf including the terminating null. The output may be trimmed if it
4245 * does not fit into the provided buffer.
4246 */
4247 static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4248 u32 size_buf)
4249 {
4250 int ret;
4251 u32 size_bp = size_buf;
4252 char *bp = buf;
4253 u64 flags = bargs->flags;
4254 char tmp_buf[128] = {'\0'};
4255
4256 if (!flags)
4257 return;
4258
4259 #define CHECK_APPEND_NOARG(a) \
4260 do { \
4261 ret = snprintf(bp, size_bp, (a)); \
4262 if (ret < 0 || ret >= size_bp) \
4263 goto out_overflow; \
4264 size_bp -= ret; \
4265 bp += ret; \
4266 } while (0)
4267
4268 #define CHECK_APPEND_1ARG(a, v1) \
4269 do { \
4270 ret = snprintf(bp, size_bp, (a), (v1)); \
4271 if (ret < 0 || ret >= size_bp) \
4272 goto out_overflow; \
4273 size_bp -= ret; \
4274 bp += ret; \
4275 } while (0)
4276
4277 #define CHECK_APPEND_2ARG(a, v1, v2) \
4278 do { \
4279 ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4280 if (ret < 0 || ret >= size_bp) \
4281 goto out_overflow; \
4282 size_bp -= ret; \
4283 bp += ret; \
4284 } while (0)
4285
4286 if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4287 CHECK_APPEND_1ARG("convert=%s,",
4288 btrfs_bg_type_to_raid_name(bargs->target));
4289
4290 if (flags & BTRFS_BALANCE_ARGS_SOFT)
4291 CHECK_APPEND_NOARG("soft,");
4292
4293 if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4294 btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4295 sizeof(tmp_buf));
4296 CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4297 }
4298
4299 if (flags & BTRFS_BALANCE_ARGS_USAGE)
4300 CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4301
4302 if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4303 CHECK_APPEND_2ARG("usage=%u..%u,",
4304 bargs->usage_min, bargs->usage_max);
4305
4306 if (flags & BTRFS_BALANCE_ARGS_DEVID)
4307 CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4308
4309 if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4310 CHECK_APPEND_2ARG("drange=%llu..%llu,",
4311 bargs->pstart, bargs->pend);
4312
4313 if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4314 CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4315 bargs->vstart, bargs->vend);
4316
4317 if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4318 CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4319
4320 if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4321 CHECK_APPEND_2ARG("limit=%u..%u,",
4322 bargs->limit_min, bargs->limit_max);
4323
4324 if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4325 CHECK_APPEND_2ARG("stripes=%u..%u,",
4326 bargs->stripes_min, bargs->stripes_max);
4327
4328 #undef CHECK_APPEND_2ARG
4329 #undef CHECK_APPEND_1ARG
4330 #undef CHECK_APPEND_NOARG
4331
4332 out_overflow:
4333
4334 if (size_bp < size_buf)
4335 buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4336 else
4337 buf[0] = '\0';
4338 }
4339
4340 static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4341 {
4342 u32 size_buf = 1024;
4343 char tmp_buf[192] = {'\0'};
4344 char *buf;
4345 char *bp;
4346 u32 size_bp = size_buf;
4347 int ret;
4348 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4349
4350 buf = kzalloc(size_buf, GFP_KERNEL);
4351 if (!buf)
4352 return;
4353
4354 bp = buf;
4355
4356 #define CHECK_APPEND_1ARG(a, v1) \
4357 do { \
4358 ret = snprintf(bp, size_bp, (a), (v1)); \
4359 if (ret < 0 || ret >= size_bp) \
4360 goto out_overflow; \
4361 size_bp -= ret; \
4362 bp += ret; \
4363 } while (0)
4364
4365 if (bctl->flags & BTRFS_BALANCE_FORCE)
4366 CHECK_APPEND_1ARG("%s", "-f ");
4367
4368 if (bctl->flags & BTRFS_BALANCE_DATA) {
4369 describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4370 CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4371 }
4372
4373 if (bctl->flags & BTRFS_BALANCE_METADATA) {
4374 describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4375 CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4376 }
4377
4378 if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4379 describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4380 CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4381 }
4382
4383 #undef CHECK_APPEND_1ARG
4384
4385 out_overflow:
4386
4387 if (size_bp < size_buf)
4388 buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4389 btrfs_info(fs_info, "balance: %s %s",
4390 (bctl->flags & BTRFS_BALANCE_RESUME) ?
4391 "resume" : "start", buf);
4392
4393 kfree(buf);
4394 }
4395
4396 /*
4397 * Should be called with balance mutexe held
4398 */
4399 int btrfs_balance(struct btrfs_fs_info *fs_info,
4400 struct btrfs_balance_control *bctl,
4401 struct btrfs_ioctl_balance_args *bargs)
4402 {
4403 u64 meta_target, data_target;
4404 u64 allowed;
4405 int mixed = 0;
4406 int ret;
4407 u64 num_devices;
4408 unsigned seq;
4409 bool reducing_redundancy;
4410 bool paused = false;
4411 int i;
4412
4413 if (btrfs_fs_closing(fs_info) ||
4414 atomic_read(&fs_info->balance_pause_req) ||
4415 btrfs_should_cancel_balance(fs_info)) {
4416 ret = -EINVAL;
4417 goto out;
4418 }
4419
4420 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4421 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4422 mixed = 1;
4423
4424 /*
4425 * In case of mixed groups both data and meta should be picked,
4426 * and identical options should be given for both of them.
4427 */
4428 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4429 if (mixed && (bctl->flags & allowed)) {
4430 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4431 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4432 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4433 btrfs_err(fs_info,
4434 "balance: mixed groups data and metadata options must be the same");
4435 ret = -EINVAL;
4436 goto out;
4437 }
4438 }
4439
4440 /*
4441 * rw_devices will not change at the moment, device add/delete/replace
4442 * are exclusive
4443 */
4444 num_devices = fs_info->fs_devices->rw_devices;
4445
4446 /*
4447 * SINGLE profile on-disk has no profile bit, but in-memory we have a
4448 * special bit for it, to make it easier to distinguish. Thus we need
4449 * to set it manually, or balance would refuse the profile.
4450 */
4451 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4452 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4453 if (num_devices >= btrfs_raid_array[i].devs_min)
4454 allowed |= btrfs_raid_array[i].bg_flag;
4455
4456 if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4457 !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4458 !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4459 ret = -EINVAL;
4460 goto out;
4461 }
4462
4463 /*
4464 * Allow to reduce metadata or system integrity only if force set for
4465 * profiles with redundancy (copies, parity)
4466 */
4467 allowed = 0;
4468 for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4469 if (btrfs_raid_array[i].ncopies >= 2 ||
4470 btrfs_raid_array[i].tolerated_failures >= 1)
4471 allowed |= btrfs_raid_array[i].bg_flag;
4472 }
4473 do {
4474 seq = read_seqbegin(&fs_info->profiles_lock);
4475
4476 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4477 (fs_info->avail_system_alloc_bits & allowed) &&
4478 !(bctl->sys.target & allowed)) ||
4479 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4480 (fs_info->avail_metadata_alloc_bits & allowed) &&
4481 !(bctl->meta.target & allowed)))
4482 reducing_redundancy = true;
4483 else
4484 reducing_redundancy = false;
4485
4486 /* if we're not converting, the target field is uninitialized */
4487 meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4488 bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4489 data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4490 bctl->data.target : fs_info->avail_data_alloc_bits;
4491 } while (read_seqretry(&fs_info->profiles_lock, seq));
4492
4493 if (reducing_redundancy) {
4494 if (bctl->flags & BTRFS_BALANCE_FORCE) {
4495 btrfs_info(fs_info,
4496 "balance: force reducing metadata redundancy");
4497 } else {
4498 btrfs_err(fs_info,
4499 "balance: reduces metadata redundancy, use --force if you want this");
4500 ret = -EINVAL;
4501 goto out;
4502 }
4503 }
4504
4505 if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4506 btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4507 btrfs_warn(fs_info,
4508 "balance: metadata profile %s has lower redundancy than data profile %s",
4509 btrfs_bg_type_to_raid_name(meta_target),
4510 btrfs_bg_type_to_raid_name(data_target));
4511 }
4512
4513 ret = insert_balance_item(fs_info, bctl);
4514 if (ret && ret != -EEXIST)
4515 goto out;
4516
4517 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4518 BUG_ON(ret == -EEXIST);
4519 BUG_ON(fs_info->balance_ctl);
4520 spin_lock(&fs_info->balance_lock);
4521 fs_info->balance_ctl = bctl;
4522 spin_unlock(&fs_info->balance_lock);
4523 } else {
4524 BUG_ON(ret != -EEXIST);
4525 spin_lock(&fs_info->balance_lock);
4526 update_balance_args(bctl);
4527 spin_unlock(&fs_info->balance_lock);
4528 }
4529
4530 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4531 set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4532 describe_balance_start_or_resume(fs_info);
4533 mutex_unlock(&fs_info->balance_mutex);
4534
4535 ret = __btrfs_balance(fs_info);
4536
4537 mutex_lock(&fs_info->balance_mutex);
4538 if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4539 btrfs_info(fs_info, "balance: paused");
4540 btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4541 paused = true;
4542 }
4543 /*
4544 * Balance can be canceled by:
4545 *
4546 * - Regular cancel request
4547 * Then ret == -ECANCELED and balance_cancel_req > 0
4548 *
4549 * - Fatal signal to "btrfs" process
4550 * Either the signal caught by wait_reserve_ticket() and callers
4551 * got -EINTR, or caught by btrfs_should_cancel_balance() and
4552 * got -ECANCELED.
4553 * Either way, in this case balance_cancel_req = 0, and
4554 * ret == -EINTR or ret == -ECANCELED.
4555 *
4556 * So here we only check the return value to catch canceled balance.
4557 */
4558 else if (ret == -ECANCELED || ret == -EINTR)
4559 btrfs_info(fs_info, "balance: canceled");
4560 else
4561 btrfs_info(fs_info, "balance: ended with status: %d", ret);
4562
4563 clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4564
4565 if (bargs) {
4566 memset(bargs, 0, sizeof(*bargs));
4567 btrfs_update_ioctl_balance_args(fs_info, bargs);
4568 }
4569
4570 /* We didn't pause, we can clean everything up. */
4571 if (!paused) {
4572 reset_balance_state(fs_info);
4573 btrfs_exclop_finish(fs_info);
4574 }
4575
4576 wake_up(&fs_info->balance_wait_q);
4577
4578 return ret;
4579 out:
4580 if (bctl->flags & BTRFS_BALANCE_RESUME)
4581 reset_balance_state(fs_info);
4582 else
4583 kfree(bctl);
4584 btrfs_exclop_finish(fs_info);
4585
4586 return ret;
4587 }
4588
4589 static int balance_kthread(void *data)
4590 {
4591 struct btrfs_fs_info *fs_info = data;
4592 int ret = 0;
4593
4594 sb_start_write(fs_info->sb);
4595 mutex_lock(&fs_info->balance_mutex);
4596 if (fs_info->balance_ctl)
4597 ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4598 mutex_unlock(&fs_info->balance_mutex);
4599 sb_end_write(fs_info->sb);
4600
4601 return ret;
4602 }
4603
4604 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4605 {
4606 struct task_struct *tsk;
4607
4608 mutex_lock(&fs_info->balance_mutex);
4609 if (!fs_info->balance_ctl) {
4610 mutex_unlock(&fs_info->balance_mutex);
4611 return 0;
4612 }
4613 mutex_unlock(&fs_info->balance_mutex);
4614
4615 if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4616 btrfs_info(fs_info, "balance: resume skipped");
4617 return 0;
4618 }
4619
4620 spin_lock(&fs_info->super_lock);
4621 ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4622 fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4623 spin_unlock(&fs_info->super_lock);
4624 /*
4625 * A ro->rw remount sequence should continue with the paused balance
4626 * regardless of who pauses it, system or the user as of now, so set
4627 * the resume flag.
4628 */
4629 spin_lock(&fs_info->balance_lock);
4630 fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4631 spin_unlock(&fs_info->balance_lock);
4632
4633 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4634 return PTR_ERR_OR_ZERO(tsk);
4635 }
4636
4637 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4638 {
4639 struct btrfs_balance_control *bctl;
4640 struct btrfs_balance_item *item;
4641 struct btrfs_disk_balance_args disk_bargs;
4642 struct btrfs_path *path;
4643 struct extent_buffer *leaf;
4644 struct btrfs_key key;
4645 int ret;
4646
4647 path = btrfs_alloc_path();
4648 if (!path)
4649 return -ENOMEM;
4650
4651 key.objectid = BTRFS_BALANCE_OBJECTID;
4652 key.type = BTRFS_TEMPORARY_ITEM_KEY;
4653 key.offset = 0;
4654
4655 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4656 if (ret < 0)
4657 goto out;
4658 if (ret > 0) { /* ret = -ENOENT; */
4659 ret = 0;
4660 goto out;
4661 }
4662
4663 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4664 if (!bctl) {
4665 ret = -ENOMEM;
4666 goto out;
4667 }
4668
4669 leaf = path->nodes[0];
4670 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4671
4672 bctl->flags = btrfs_balance_flags(leaf, item);
4673 bctl->flags |= BTRFS_BALANCE_RESUME;
4674
4675 btrfs_balance_data(leaf, item, &disk_bargs);
4676 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4677 btrfs_balance_meta(leaf, item, &disk_bargs);
4678 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4679 btrfs_balance_sys(leaf, item, &disk_bargs);
4680 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4681
4682 /*
4683 * This should never happen, as the paused balance state is recovered
4684 * during mount without any chance of other exclusive ops to collide.
4685 *
4686 * This gives the exclusive op status to balance and keeps in paused
4687 * state until user intervention (cancel or umount). If the ownership
4688 * cannot be assigned, show a message but do not fail. The balance
4689 * is in a paused state and must have fs_info::balance_ctl properly
4690 * set up.
4691 */
4692 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4693 btrfs_warn(fs_info,
4694 "balance: cannot set exclusive op status, resume manually");
4695
4696 btrfs_release_path(path);
4697
4698 mutex_lock(&fs_info->balance_mutex);
4699 BUG_ON(fs_info->balance_ctl);
4700 spin_lock(&fs_info->balance_lock);
4701 fs_info->balance_ctl = bctl;
4702 spin_unlock(&fs_info->balance_lock);
4703 mutex_unlock(&fs_info->balance_mutex);
4704 out:
4705 btrfs_free_path(path);
4706 return ret;
4707 }
4708
4709 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4710 {
4711 int ret = 0;
4712
4713 mutex_lock(&fs_info->balance_mutex);
4714 if (!fs_info->balance_ctl) {
4715 mutex_unlock(&fs_info->balance_mutex);
4716 return -ENOTCONN;
4717 }
4718
4719 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4720 atomic_inc(&fs_info->balance_pause_req);
4721 mutex_unlock(&fs_info->balance_mutex);
4722
4723 wait_event(fs_info->balance_wait_q,
4724 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4725
4726 mutex_lock(&fs_info->balance_mutex);
4727 /* we are good with balance_ctl ripped off from under us */
4728 BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4729 atomic_dec(&fs_info->balance_pause_req);
4730 } else {
4731 ret = -ENOTCONN;
4732 }
4733
4734 mutex_unlock(&fs_info->balance_mutex);
4735 return ret;
4736 }
4737
4738 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4739 {
4740 mutex_lock(&fs_info->balance_mutex);
4741 if (!fs_info->balance_ctl) {
4742 mutex_unlock(&fs_info->balance_mutex);
4743 return -ENOTCONN;
4744 }
4745
4746 /*
4747 * A paused balance with the item stored on disk can be resumed at
4748 * mount time if the mount is read-write. Otherwise it's still paused
4749 * and we must not allow cancelling as it deletes the item.
4750 */
4751 if (sb_rdonly(fs_info->sb)) {
4752 mutex_unlock(&fs_info->balance_mutex);
4753 return -EROFS;
4754 }
4755
4756 atomic_inc(&fs_info->balance_cancel_req);
4757 /*
4758 * if we are running just wait and return, balance item is
4759 * deleted in btrfs_balance in this case
4760 */
4761 if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4762 mutex_unlock(&fs_info->balance_mutex);
4763 wait_event(fs_info->balance_wait_q,
4764 !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4765 mutex_lock(&fs_info->balance_mutex);
4766 } else {
4767 mutex_unlock(&fs_info->balance_mutex);
4768 /*
4769 * Lock released to allow other waiters to continue, we'll
4770 * reexamine the status again.
4771 */
4772 mutex_lock(&fs_info->balance_mutex);
4773
4774 if (fs_info->balance_ctl) {
4775 reset_balance_state(fs_info);
4776 btrfs_exclop_finish(fs_info);
4777 btrfs_info(fs_info, "balance: canceled");
4778 }
4779 }
4780
4781 ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4782 atomic_dec(&fs_info->balance_cancel_req);
4783 mutex_unlock(&fs_info->balance_mutex);
4784 return 0;
4785 }
4786
4787 int btrfs_uuid_scan_kthread(void *data)
4788 {
4789 struct btrfs_fs_info *fs_info = data;
4790 struct btrfs_root *root = fs_info->tree_root;
4791 struct btrfs_key key;
4792 struct btrfs_path *path = NULL;
4793 int ret = 0;
4794 struct extent_buffer *eb;
4795 int slot;
4796 struct btrfs_root_item root_item;
4797 u32 item_size;
4798 struct btrfs_trans_handle *trans = NULL;
4799 bool closing = false;
4800
4801 path = btrfs_alloc_path();
4802 if (!path) {
4803 ret = -ENOMEM;
4804 goto out;
4805 }
4806
4807 key.objectid = 0;
4808 key.type = BTRFS_ROOT_ITEM_KEY;
4809 key.offset = 0;
4810
4811 while (1) {
4812 if (btrfs_fs_closing(fs_info)) {
4813 closing = true;
4814 break;
4815 }
4816 ret = btrfs_search_forward(root, &key, path,
4817 BTRFS_OLDEST_GENERATION);
4818 if (ret) {
4819 if (ret > 0)
4820 ret = 0;
4821 break;
4822 }
4823
4824 if (key.type != BTRFS_ROOT_ITEM_KEY ||
4825 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4826 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4827 key.objectid > BTRFS_LAST_FREE_OBJECTID)
4828 goto skip;
4829
4830 eb = path->nodes[0];
4831 slot = path->slots[0];
4832 item_size = btrfs_item_size(eb, slot);
4833 if (item_size < sizeof(root_item))
4834 goto skip;
4835
4836 read_extent_buffer(eb, &root_item,
4837 btrfs_item_ptr_offset(eb, slot),
4838 (int)sizeof(root_item));
4839 if (btrfs_root_refs(&root_item) == 0)
4840 goto skip;
4841
4842 if (!btrfs_is_empty_uuid(root_item.uuid) ||
4843 !btrfs_is_empty_uuid(root_item.received_uuid)) {
4844 if (trans)
4845 goto update_tree;
4846
4847 btrfs_release_path(path);
4848 /*
4849 * 1 - subvol uuid item
4850 * 1 - received_subvol uuid item
4851 */
4852 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4853 if (IS_ERR(trans)) {
4854 ret = PTR_ERR(trans);
4855 break;
4856 }
4857 continue;
4858 } else {
4859 goto skip;
4860 }
4861 update_tree:
4862 btrfs_release_path(path);
4863 if (!btrfs_is_empty_uuid(root_item.uuid)) {
4864 ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4865 BTRFS_UUID_KEY_SUBVOL,
4866 key.objectid);
4867 if (ret < 0) {
4868 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4869 ret);
4870 break;
4871 }
4872 }
4873
4874 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4875 ret = btrfs_uuid_tree_add(trans,
4876 root_item.received_uuid,
4877 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4878 key.objectid);
4879 if (ret < 0) {
4880 btrfs_warn(fs_info, "uuid_tree_add failed %d",
4881 ret);
4882 break;
4883 }
4884 }
4885
4886 skip:
4887 btrfs_release_path(path);
4888 if (trans) {
4889 ret = btrfs_end_transaction(trans);
4890 trans = NULL;
4891 if (ret)
4892 break;
4893 }
4894
4895 if (key.offset < (u64)-1) {
4896 key.offset++;
4897 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4898 key.offset = 0;
4899 key.type = BTRFS_ROOT_ITEM_KEY;
4900 } else if (key.objectid < (u64)-1) {
4901 key.offset = 0;
4902 key.type = BTRFS_ROOT_ITEM_KEY;
4903 key.objectid++;
4904 } else {
4905 break;
4906 }
4907 cond_resched();
4908 }
4909
4910 out:
4911 btrfs_free_path(path);
4912 if (trans && !IS_ERR(trans))
4913 btrfs_end_transaction(trans);
4914 if (ret)
4915 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4916 else if (!closing)
4917 set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4918 up(&fs_info->uuid_tree_rescan_sem);
4919 return 0;
4920 }
4921
4922 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4923 {
4924 struct btrfs_trans_handle *trans;
4925 struct btrfs_root *tree_root = fs_info->tree_root;
4926 struct btrfs_root *uuid_root;
4927 struct task_struct *task;
4928 int ret;
4929
4930 /*
4931 * 1 - root node
4932 * 1 - root item
4933 */
4934 trans = btrfs_start_transaction(tree_root, 2);
4935 if (IS_ERR(trans))
4936 return PTR_ERR(trans);
4937
4938 uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4939 if (IS_ERR(uuid_root)) {
4940 ret = PTR_ERR(uuid_root);
4941 btrfs_abort_transaction(trans, ret);
4942 btrfs_end_transaction(trans);
4943 return ret;
4944 }
4945
4946 fs_info->uuid_root = uuid_root;
4947
4948 ret = btrfs_commit_transaction(trans);
4949 if (ret)
4950 return ret;
4951
4952 down(&fs_info->uuid_tree_rescan_sem);
4953 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4954 if (IS_ERR(task)) {
4955 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4956 btrfs_warn(fs_info, "failed to start uuid_scan task");
4957 up(&fs_info->uuid_tree_rescan_sem);
4958 return PTR_ERR(task);
4959 }
4960
4961 return 0;
4962 }
4963
4964 /*
4965 * shrinking a device means finding all of the device extents past
4966 * the new size, and then following the back refs to the chunks.
4967 * The chunk relocation code actually frees the device extent
4968 */
4969 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4970 {
4971 struct btrfs_fs_info *fs_info = device->fs_info;
4972 struct btrfs_root *root = fs_info->dev_root;
4973 struct btrfs_trans_handle *trans;
4974 struct btrfs_dev_extent *dev_extent = NULL;
4975 struct btrfs_path *path;
4976 u64 length;
4977 u64 chunk_offset;
4978 int ret;
4979 int slot;
4980 int failed = 0;
4981 bool retried = false;
4982 struct extent_buffer *l;
4983 struct btrfs_key key;
4984 struct btrfs_super_block *super_copy = fs_info->super_copy;
4985 u64 old_total = btrfs_super_total_bytes(super_copy);
4986 u64 old_size = btrfs_device_get_total_bytes(device);
4987 u64 diff;
4988 u64 start;
4989 u64 free_diff = 0;
4990
4991 new_size = round_down(new_size, fs_info->sectorsize);
4992 start = new_size;
4993 diff = round_down(old_size - new_size, fs_info->sectorsize);
4994
4995 if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4996 return -EINVAL;
4997
4998 path = btrfs_alloc_path();
4999 if (!path)
5000 return -ENOMEM;
5001
5002 path->reada = READA_BACK;
5003
5004 trans = btrfs_start_transaction(root, 0);
5005 if (IS_ERR(trans)) {
5006 btrfs_free_path(path);
5007 return PTR_ERR(trans);
5008 }
5009
5010 mutex_lock(&fs_info->chunk_mutex);
5011
5012 btrfs_device_set_total_bytes(device, new_size);
5013 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5014 device->fs_devices->total_rw_bytes -= diff;
5015
5016 /*
5017 * The new free_chunk_space is new_size - used, so we have to
5018 * subtract the delta of the old free_chunk_space which included
5019 * old_size - used. If used > new_size then just subtract this
5020 * entire device's free space.
5021 */
5022 if (device->bytes_used < new_size)
5023 free_diff = (old_size - device->bytes_used) -
5024 (new_size - device->bytes_used);
5025 else
5026 free_diff = old_size - device->bytes_used;
5027 atomic64_sub(free_diff, &fs_info->free_chunk_space);
5028 }
5029
5030 /*
5031 * Once the device's size has been set to the new size, ensure all
5032 * in-memory chunks are synced to disk so that the loop below sees them
5033 * and relocates them accordingly.
5034 */
5035 if (contains_pending_extent(device, &start, diff)) {
5036 mutex_unlock(&fs_info->chunk_mutex);
5037 ret = btrfs_commit_transaction(trans);
5038 if (ret)
5039 goto done;
5040 } else {
5041 mutex_unlock(&fs_info->chunk_mutex);
5042 btrfs_end_transaction(trans);
5043 }
5044
5045 again:
5046 key.objectid = device->devid;
5047 key.offset = (u64)-1;
5048 key.type = BTRFS_DEV_EXTENT_KEY;
5049
5050 do {
5051 mutex_lock(&fs_info->reclaim_bgs_lock);
5052 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5053 if (ret < 0) {
5054 mutex_unlock(&fs_info->reclaim_bgs_lock);
5055 goto done;
5056 }
5057
5058 ret = btrfs_previous_item(root, path, 0, key.type);
5059 if (ret) {
5060 mutex_unlock(&fs_info->reclaim_bgs_lock);
5061 if (ret < 0)
5062 goto done;
5063 ret = 0;
5064 btrfs_release_path(path);
5065 break;
5066 }
5067
5068 l = path->nodes[0];
5069 slot = path->slots[0];
5070 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
5071
5072 if (key.objectid != device->devid) {
5073 mutex_unlock(&fs_info->reclaim_bgs_lock);
5074 btrfs_release_path(path);
5075 break;
5076 }
5077
5078 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
5079 length = btrfs_dev_extent_length(l, dev_extent);
5080
5081 if (key.offset + length <= new_size) {
5082 mutex_unlock(&fs_info->reclaim_bgs_lock);
5083 btrfs_release_path(path);
5084 break;
5085 }
5086
5087 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
5088 btrfs_release_path(path);
5089
5090 /*
5091 * We may be relocating the only data chunk we have,
5092 * which could potentially end up with losing data's
5093 * raid profile, so lets allocate an empty one in
5094 * advance.
5095 */
5096 ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
5097 if (ret < 0) {
5098 mutex_unlock(&fs_info->reclaim_bgs_lock);
5099 goto done;
5100 }
5101
5102 ret = btrfs_relocate_chunk(fs_info, chunk_offset);
5103 mutex_unlock(&fs_info->reclaim_bgs_lock);
5104 if (ret == -ENOSPC) {
5105 failed++;
5106 } else if (ret) {
5107 if (ret == -ETXTBSY) {
5108 btrfs_warn(fs_info,
5109 "could not shrink block group %llu due to active swapfile",
5110 chunk_offset);
5111 }
5112 goto done;
5113 }
5114 } while (key.offset-- > 0);
5115
5116 if (failed && !retried) {
5117 failed = 0;
5118 retried = true;
5119 goto again;
5120 } else if (failed && retried) {
5121 ret = -ENOSPC;
5122 goto done;
5123 }
5124
5125 /* Shrinking succeeded, else we would be at "done". */
5126 trans = btrfs_start_transaction(root, 0);
5127 if (IS_ERR(trans)) {
5128 ret = PTR_ERR(trans);
5129 goto done;
5130 }
5131
5132 mutex_lock(&fs_info->chunk_mutex);
5133 /* Clear all state bits beyond the shrunk device size */
5134 clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
5135 CHUNK_STATE_MASK);
5136
5137 btrfs_device_set_disk_total_bytes(device, new_size);
5138 if (list_empty(&device->post_commit_list))
5139 list_add_tail(&device->post_commit_list,
5140 &trans->transaction->dev_update_list);
5141
5142 WARN_ON(diff > old_total);
5143 btrfs_set_super_total_bytes(super_copy,
5144 round_down(old_total - diff, fs_info->sectorsize));
5145 mutex_unlock(&fs_info->chunk_mutex);
5146
5147 btrfs_reserve_chunk_metadata(trans, false);
5148 /* Now btrfs_update_device() will change the on-disk size. */
5149 ret = btrfs_update_device(trans, device);
5150 btrfs_trans_release_chunk_metadata(trans);
5151 if (ret < 0) {
5152 btrfs_abort_transaction(trans, ret);
5153 btrfs_end_transaction(trans);
5154 } else {
5155 ret = btrfs_commit_transaction(trans);
5156 }
5157 done:
5158 btrfs_free_path(path);
5159 if (ret) {
5160 mutex_lock(&fs_info->chunk_mutex);
5161 btrfs_device_set_total_bytes(device, old_size);
5162 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5163 device->fs_devices->total_rw_bytes += diff;
5164 atomic64_add(free_diff, &fs_info->free_chunk_space);
5165 }
5166 mutex_unlock(&fs_info->chunk_mutex);
5167 }
5168 return ret;
5169 }
5170
5171 static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5172 struct btrfs_key *key,
5173 struct btrfs_chunk *chunk, int item_size)
5174 {
5175 struct btrfs_super_block *super_copy = fs_info->super_copy;
5176 struct btrfs_disk_key disk_key;
5177 u32 array_size;
5178 u8 *ptr;
5179
5180 lockdep_assert_held(&fs_info->chunk_mutex);
5181
5182 array_size = btrfs_super_sys_array_size(super_copy);
5183 if (array_size + item_size + sizeof(disk_key)
5184 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5185 return -EFBIG;
5186
5187 ptr = super_copy->sys_chunk_array + array_size;
5188 btrfs_cpu_key_to_disk(&disk_key, key);
5189 memcpy(ptr, &disk_key, sizeof(disk_key));
5190 ptr += sizeof(disk_key);
5191 memcpy(ptr, chunk, item_size);
5192 item_size += sizeof(disk_key);
5193 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5194
5195 return 0;
5196 }
5197
5198 /*
5199 * sort the devices in descending order by max_avail, total_avail
5200 */
5201 static int btrfs_cmp_device_info(const void *a, const void *b)
5202 {
5203 const struct btrfs_device_info *di_a = a;
5204 const struct btrfs_device_info *di_b = b;
5205
5206 if (di_a->max_avail > di_b->max_avail)
5207 return -1;
5208 if (di_a->max_avail < di_b->max_avail)
5209 return 1;
5210 if (di_a->total_avail > di_b->total_avail)
5211 return -1;
5212 if (di_a->total_avail < di_b->total_avail)
5213 return 1;
5214 return 0;
5215 }
5216
5217 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5218 {
5219 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5220 return;
5221
5222 btrfs_set_fs_incompat(info, RAID56);
5223 }
5224
5225 static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5226 {
5227 if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5228 return;
5229
5230 btrfs_set_fs_incompat(info, RAID1C34);
5231 }
5232
5233 /*
5234 * Structure used internally for btrfs_create_chunk() function.
5235 * Wraps needed parameters.
5236 */
5237 struct alloc_chunk_ctl {
5238 u64 start;
5239 u64 type;
5240 /* Total number of stripes to allocate */
5241 int num_stripes;
5242 /* sub_stripes info for map */
5243 int sub_stripes;
5244 /* Stripes per device */
5245 int dev_stripes;
5246 /* Maximum number of devices to use */
5247 int devs_max;
5248 /* Minimum number of devices to use */
5249 int devs_min;
5250 /* ndevs has to be a multiple of this */
5251 int devs_increment;
5252 /* Number of copies */
5253 int ncopies;
5254 /* Number of stripes worth of bytes to store parity information */
5255 int nparity;
5256 u64 max_stripe_size;
5257 u64 max_chunk_size;
5258 u64 dev_extent_min;
5259 u64 stripe_size;
5260 u64 chunk_size;
5261 int ndevs;
5262 };
5263
5264 static void init_alloc_chunk_ctl_policy_regular(
5265 struct btrfs_fs_devices *fs_devices,
5266 struct alloc_chunk_ctl *ctl)
5267 {
5268 struct btrfs_space_info *space_info;
5269
5270 space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5271 ASSERT(space_info);
5272
5273 ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5274 ctl->max_stripe_size = min_t(u64, ctl->max_chunk_size, SZ_1G);
5275
5276 if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5277 ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5278
5279 /* We don't want a chunk larger than 10% of writable space */
5280 ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5281 ctl->max_chunk_size);
5282 ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5283 }
5284
5285 static void init_alloc_chunk_ctl_policy_zoned(
5286 struct btrfs_fs_devices *fs_devices,
5287 struct alloc_chunk_ctl *ctl)
5288 {
5289 u64 zone_size = fs_devices->fs_info->zone_size;
5290 u64 limit;
5291 int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5292 int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5293 u64 min_chunk_size = min_data_stripes * zone_size;
5294 u64 type = ctl->type;
5295
5296 ctl->max_stripe_size = zone_size;
5297 if (type & BTRFS_BLOCK_GROUP_DATA) {
5298 ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5299 zone_size);
5300 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5301 ctl->max_chunk_size = ctl->max_stripe_size;
5302 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5303 ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5304 ctl->devs_max = min_t(int, ctl->devs_max,
5305 BTRFS_MAX_DEVS_SYS_CHUNK);
5306 } else {
5307 BUG();
5308 }
5309
5310 /* We don't want a chunk larger than 10% of writable space */
5311 limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5312 zone_size),
5313 min_chunk_size);
5314 ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5315 ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5316 }
5317
5318 static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5319 struct alloc_chunk_ctl *ctl)
5320 {
5321 int index = btrfs_bg_flags_to_raid_index(ctl->type);
5322
5323 ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5324 ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5325 ctl->devs_max = btrfs_raid_array[index].devs_max;
5326 if (!ctl->devs_max)
5327 ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5328 ctl->devs_min = btrfs_raid_array[index].devs_min;
5329 ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5330 ctl->ncopies = btrfs_raid_array[index].ncopies;
5331 ctl->nparity = btrfs_raid_array[index].nparity;
5332 ctl->ndevs = 0;
5333
5334 switch (fs_devices->chunk_alloc_policy) {
5335 case BTRFS_CHUNK_ALLOC_REGULAR:
5336 init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5337 break;
5338 case BTRFS_CHUNK_ALLOC_ZONED:
5339 init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5340 break;
5341 default:
5342 BUG();
5343 }
5344 }
5345
5346 static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5347 struct alloc_chunk_ctl *ctl,
5348 struct btrfs_device_info *devices_info)
5349 {
5350 struct btrfs_fs_info *info = fs_devices->fs_info;
5351 struct btrfs_device *device;
5352 u64 total_avail;
5353 u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5354 int ret;
5355 int ndevs = 0;
5356 u64 max_avail;
5357 u64 dev_offset;
5358
5359 /*
5360 * in the first pass through the devices list, we gather information
5361 * about the available holes on each device.
5362 */
5363 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5364 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5365 WARN(1, KERN_ERR
5366 "BTRFS: read-only device in alloc_list\n");
5367 continue;
5368 }
5369
5370 if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5371 &device->dev_state) ||
5372 test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5373 continue;
5374
5375 if (device->total_bytes > device->bytes_used)
5376 total_avail = device->total_bytes - device->bytes_used;
5377 else
5378 total_avail = 0;
5379
5380 /* If there is no space on this device, skip it. */
5381 if (total_avail < ctl->dev_extent_min)
5382 continue;
5383
5384 ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5385 &max_avail);
5386 if (ret && ret != -ENOSPC)
5387 return ret;
5388
5389 if (ret == 0)
5390 max_avail = dev_extent_want;
5391
5392 if (max_avail < ctl->dev_extent_min) {
5393 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5394 btrfs_debug(info,
5395 "%s: devid %llu has no free space, have=%llu want=%llu",
5396 __func__, device->devid, max_avail,
5397 ctl->dev_extent_min);
5398 continue;
5399 }
5400
5401 if (ndevs == fs_devices->rw_devices) {
5402 WARN(1, "%s: found more than %llu devices\n",
5403 __func__, fs_devices->rw_devices);
5404 break;
5405 }
5406 devices_info[ndevs].dev_offset = dev_offset;
5407 devices_info[ndevs].max_avail = max_avail;
5408 devices_info[ndevs].total_avail = total_avail;
5409 devices_info[ndevs].dev = device;
5410 ++ndevs;
5411 }
5412 ctl->ndevs = ndevs;
5413
5414 /*
5415 * now sort the devices by hole size / available space
5416 */
5417 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5418 btrfs_cmp_device_info, NULL);
5419
5420 return 0;
5421 }
5422
5423 static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5424 struct btrfs_device_info *devices_info)
5425 {
5426 /* Number of stripes that count for block group size */
5427 int data_stripes;
5428
5429 /*
5430 * The primary goal is to maximize the number of stripes, so use as
5431 * many devices as possible, even if the stripes are not maximum sized.
5432 *
5433 * The DUP profile stores more than one stripe per device, the
5434 * max_avail is the total size so we have to adjust.
5435 */
5436 ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5437 ctl->dev_stripes);
5438 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5439
5440 /* This will have to be fixed for RAID1 and RAID10 over more drives */
5441 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5442
5443 /*
5444 * Use the number of data stripes to figure out how big this chunk is
5445 * really going to be in terms of logical address space, and compare
5446 * that answer with the max chunk size. If it's higher, we try to
5447 * reduce stripe_size.
5448 */
5449 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5450 /*
5451 * Reduce stripe_size, round it up to a 16MB boundary again and
5452 * then use it, unless it ends up being even bigger than the
5453 * previous value we had already.
5454 */
5455 ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5456 data_stripes), SZ_16M),
5457 ctl->stripe_size);
5458 }
5459
5460 /* Stripe size should not go beyond 1G. */
5461 ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5462
5463 /* Align to BTRFS_STRIPE_LEN */
5464 ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5465 ctl->chunk_size = ctl->stripe_size * data_stripes;
5466
5467 return 0;
5468 }
5469
5470 static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5471 struct btrfs_device_info *devices_info)
5472 {
5473 u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5474 /* Number of stripes that count for block group size */
5475 int data_stripes;
5476
5477 /*
5478 * It should hold because:
5479 * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5480 */
5481 ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5482
5483 ctl->stripe_size = zone_size;
5484 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5485 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5486
5487 /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5488 if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5489 ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5490 ctl->stripe_size) + ctl->nparity,
5491 ctl->dev_stripes);
5492 ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5493 data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5494 ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5495 }
5496
5497 ctl->chunk_size = ctl->stripe_size * data_stripes;
5498
5499 return 0;
5500 }
5501
5502 static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5503 struct alloc_chunk_ctl *ctl,
5504 struct btrfs_device_info *devices_info)
5505 {
5506 struct btrfs_fs_info *info = fs_devices->fs_info;
5507
5508 /*
5509 * Round down to number of usable stripes, devs_increment can be any
5510 * number so we can't use round_down() that requires power of 2, while
5511 * rounddown is safe.
5512 */
5513 ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5514
5515 if (ctl->ndevs < ctl->devs_min) {
5516 if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5517 btrfs_debug(info,
5518 "%s: not enough devices with free space: have=%d minimum required=%d",
5519 __func__, ctl->ndevs, ctl->devs_min);
5520 }
5521 return -ENOSPC;
5522 }
5523
5524 ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5525
5526 switch (fs_devices->chunk_alloc_policy) {
5527 case BTRFS_CHUNK_ALLOC_REGULAR:
5528 return decide_stripe_size_regular(ctl, devices_info);
5529 case BTRFS_CHUNK_ALLOC_ZONED:
5530 return decide_stripe_size_zoned(ctl, devices_info);
5531 default:
5532 BUG();
5533 }
5534 }
5535
5536 static void chunk_map_device_set_bits(struct btrfs_chunk_map *map, unsigned int bits)
5537 {
5538 for (int i = 0; i < map->num_stripes; i++) {
5539 struct btrfs_io_stripe *stripe = &map->stripes[i];
5540 struct btrfs_device *device = stripe->dev;
5541
5542 set_extent_bit(&device->alloc_state, stripe->physical,
5543 stripe->physical + map->stripe_size - 1,
5544 bits | EXTENT_NOWAIT, NULL);
5545 }
5546 }
5547
5548 static void chunk_map_device_clear_bits(struct btrfs_chunk_map *map, unsigned int bits)
5549 {
5550 for (int i = 0; i < map->num_stripes; i++) {
5551 struct btrfs_io_stripe *stripe = &map->stripes[i];
5552 struct btrfs_device *device = stripe->dev;
5553
5554 __clear_extent_bit(&device->alloc_state, stripe->physical,
5555 stripe->physical + map->stripe_size - 1,
5556 bits | EXTENT_NOWAIT,
5557 NULL, NULL);
5558 }
5559 }
5560
5561 void btrfs_remove_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5562 {
5563 write_lock(&fs_info->mapping_tree_lock);
5564 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5565 RB_CLEAR_NODE(&map->rb_node);
5566 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5567 write_unlock(&fs_info->mapping_tree_lock);
5568
5569 /* Once for the tree reference. */
5570 btrfs_free_chunk_map(map);
5571 }
5572
5573 EXPORT_FOR_TESTS
5574 int btrfs_add_chunk_map(struct btrfs_fs_info *fs_info, struct btrfs_chunk_map *map)
5575 {
5576 struct rb_node **p;
5577 struct rb_node *parent = NULL;
5578 bool leftmost = true;
5579
5580 write_lock(&fs_info->mapping_tree_lock);
5581 p = &fs_info->mapping_tree.rb_root.rb_node;
5582 while (*p) {
5583 struct btrfs_chunk_map *entry;
5584
5585 parent = *p;
5586 entry = rb_entry(parent, struct btrfs_chunk_map, rb_node);
5587
5588 if (map->start < entry->start) {
5589 p = &(*p)->rb_left;
5590 } else if (map->start > entry->start) {
5591 p = &(*p)->rb_right;
5592 leftmost = false;
5593 } else {
5594 write_unlock(&fs_info->mapping_tree_lock);
5595 return -EEXIST;
5596 }
5597 }
5598 rb_link_node(&map->rb_node, parent, p);
5599 rb_insert_color_cached(&map->rb_node, &fs_info->mapping_tree, leftmost);
5600 chunk_map_device_set_bits(map, CHUNK_ALLOCATED);
5601 chunk_map_device_clear_bits(map, CHUNK_TRIMMED);
5602 write_unlock(&fs_info->mapping_tree_lock);
5603
5604 return 0;
5605 }
5606
5607 EXPORT_FOR_TESTS
5608 struct btrfs_chunk_map *btrfs_alloc_chunk_map(int num_stripes, gfp_t gfp)
5609 {
5610 struct btrfs_chunk_map *map;
5611
5612 map = kmalloc(btrfs_chunk_map_size(num_stripes), gfp);
5613 if (!map)
5614 return NULL;
5615
5616 refcount_set(&map->refs, 1);
5617 RB_CLEAR_NODE(&map->rb_node);
5618
5619 return map;
5620 }
5621
5622 static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5623 struct alloc_chunk_ctl *ctl,
5624 struct btrfs_device_info *devices_info)
5625 {
5626 struct btrfs_fs_info *info = trans->fs_info;
5627 struct btrfs_chunk_map *map;
5628 struct btrfs_block_group *block_group;
5629 u64 start = ctl->start;
5630 u64 type = ctl->type;
5631 int ret;
5632
5633 map = btrfs_alloc_chunk_map(ctl->num_stripes, GFP_NOFS);
5634 if (!map)
5635 return ERR_PTR(-ENOMEM);
5636
5637 map->start = start;
5638 map->chunk_len = ctl->chunk_size;
5639 map->stripe_size = ctl->stripe_size;
5640 map->type = type;
5641 map->io_align = BTRFS_STRIPE_LEN;
5642 map->io_width = BTRFS_STRIPE_LEN;
5643 map->sub_stripes = ctl->sub_stripes;
5644 map->num_stripes = ctl->num_stripes;
5645
5646 for (int i = 0; i < ctl->ndevs; i++) {
5647 for (int j = 0; j < ctl->dev_stripes; j++) {
5648 int s = i * ctl->dev_stripes + j;
5649 map->stripes[s].dev = devices_info[i].dev;
5650 map->stripes[s].physical = devices_info[i].dev_offset +
5651 j * ctl->stripe_size;
5652 }
5653 }
5654
5655 trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5656
5657 ret = btrfs_add_chunk_map(info, map);
5658 if (ret) {
5659 btrfs_free_chunk_map(map);
5660 return ERR_PTR(ret);
5661 }
5662
5663 block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5664 if (IS_ERR(block_group)) {
5665 btrfs_remove_chunk_map(info, map);
5666 return block_group;
5667 }
5668
5669 for (int i = 0; i < map->num_stripes; i++) {
5670 struct btrfs_device *dev = map->stripes[i].dev;
5671
5672 btrfs_device_set_bytes_used(dev,
5673 dev->bytes_used + ctl->stripe_size);
5674 if (list_empty(&dev->post_commit_list))
5675 list_add_tail(&dev->post_commit_list,
5676 &trans->transaction->dev_update_list);
5677 }
5678
5679 atomic64_sub(ctl->stripe_size * map->num_stripes,
5680 &info->free_chunk_space);
5681
5682 check_raid56_incompat_flag(info, type);
5683 check_raid1c34_incompat_flag(info, type);
5684
5685 return block_group;
5686 }
5687
5688 struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5689 u64 type)
5690 {
5691 struct btrfs_fs_info *info = trans->fs_info;
5692 struct btrfs_fs_devices *fs_devices = info->fs_devices;
5693 struct btrfs_device_info *devices_info = NULL;
5694 struct alloc_chunk_ctl ctl;
5695 struct btrfs_block_group *block_group;
5696 int ret;
5697
5698 lockdep_assert_held(&info->chunk_mutex);
5699
5700 if (!alloc_profile_is_valid(type, 0)) {
5701 ASSERT(0);
5702 return ERR_PTR(-EINVAL);
5703 }
5704
5705 if (list_empty(&fs_devices->alloc_list)) {
5706 if (btrfs_test_opt(info, ENOSPC_DEBUG))
5707 btrfs_debug(info, "%s: no writable device", __func__);
5708 return ERR_PTR(-ENOSPC);
5709 }
5710
5711 if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5712 btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5713 ASSERT(0);
5714 return ERR_PTR(-EINVAL);
5715 }
5716
5717 ctl.start = find_next_chunk(info);
5718 ctl.type = type;
5719 init_alloc_chunk_ctl(fs_devices, &ctl);
5720
5721 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5722 GFP_NOFS);
5723 if (!devices_info)
5724 return ERR_PTR(-ENOMEM);
5725
5726 ret = gather_device_info(fs_devices, &ctl, devices_info);
5727 if (ret < 0) {
5728 block_group = ERR_PTR(ret);
5729 goto out;
5730 }
5731
5732 ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5733 if (ret < 0) {
5734 block_group = ERR_PTR(ret);
5735 goto out;
5736 }
5737
5738 block_group = create_chunk(trans, &ctl, devices_info);
5739
5740 out:
5741 kfree(devices_info);
5742 return block_group;
5743 }
5744
5745 /*
5746 * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5747 * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5748 * chunks.
5749 *
5750 * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5751 * phases.
5752 */
5753 int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5754 struct btrfs_block_group *bg)
5755 {
5756 struct btrfs_fs_info *fs_info = trans->fs_info;
5757 struct btrfs_root *chunk_root = fs_info->chunk_root;
5758 struct btrfs_key key;
5759 struct btrfs_chunk *chunk;
5760 struct btrfs_stripe *stripe;
5761 struct btrfs_chunk_map *map;
5762 size_t item_size;
5763 int i;
5764 int ret;
5765
5766 /*
5767 * We take the chunk_mutex for 2 reasons:
5768 *
5769 * 1) Updates and insertions in the chunk btree must be done while holding
5770 * the chunk_mutex, as well as updating the system chunk array in the
5771 * superblock. See the comment on top of btrfs_chunk_alloc() for the
5772 * details;
5773 *
5774 * 2) To prevent races with the final phase of a device replace operation
5775 * that replaces the device object associated with the map's stripes,
5776 * because the device object's id can change at any time during that
5777 * final phase of the device replace operation
5778 * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5779 * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5780 * which would cause a failure when updating the device item, which does
5781 * not exists, or persisting a stripe of the chunk item with such ID.
5782 * Here we can't use the device_list_mutex because our caller already
5783 * has locked the chunk_mutex, and the final phase of device replace
5784 * acquires both mutexes - first the device_list_mutex and then the
5785 * chunk_mutex. Using any of those two mutexes protects us from a
5786 * concurrent device replace.
5787 */
5788 lockdep_assert_held(&fs_info->chunk_mutex);
5789
5790 map = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5791 if (IS_ERR(map)) {
5792 ret = PTR_ERR(map);
5793 btrfs_abort_transaction(trans, ret);
5794 return ret;
5795 }
5796
5797 item_size = btrfs_chunk_item_size(map->num_stripes);
5798
5799 chunk = kzalloc(item_size, GFP_NOFS);
5800 if (!chunk) {
5801 ret = -ENOMEM;
5802 btrfs_abort_transaction(trans, ret);
5803 goto out;
5804 }
5805
5806 for (i = 0; i < map->num_stripes; i++) {
5807 struct btrfs_device *device = map->stripes[i].dev;
5808
5809 ret = btrfs_update_device(trans, device);
5810 if (ret)
5811 goto out;
5812 }
5813
5814 stripe = &chunk->stripe;
5815 for (i = 0; i < map->num_stripes; i++) {
5816 struct btrfs_device *device = map->stripes[i].dev;
5817 const u64 dev_offset = map->stripes[i].physical;
5818
5819 btrfs_set_stack_stripe_devid(stripe, device->devid);
5820 btrfs_set_stack_stripe_offset(stripe, dev_offset);
5821 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5822 stripe++;
5823 }
5824
5825 btrfs_set_stack_chunk_length(chunk, bg->length);
5826 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5827 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5828 btrfs_set_stack_chunk_type(chunk, map->type);
5829 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5830 btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5831 btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5832 btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5833 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5834
5835 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5836 key.type = BTRFS_CHUNK_ITEM_KEY;
5837 key.offset = bg->start;
5838
5839 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5840 if (ret)
5841 goto out;
5842
5843 set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5844
5845 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5846 ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5847 if (ret)
5848 goto out;
5849 }
5850
5851 out:
5852 kfree(chunk);
5853 btrfs_free_chunk_map(map);
5854 return ret;
5855 }
5856
5857 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5858 {
5859 struct btrfs_fs_info *fs_info = trans->fs_info;
5860 u64 alloc_profile;
5861 struct btrfs_block_group *meta_bg;
5862 struct btrfs_block_group *sys_bg;
5863
5864 /*
5865 * When adding a new device for sprouting, the seed device is read-only
5866 * so we must first allocate a metadata and a system chunk. But before
5867 * adding the block group items to the extent, device and chunk btrees,
5868 * we must first:
5869 *
5870 * 1) Create both chunks without doing any changes to the btrees, as
5871 * otherwise we would get -ENOSPC since the block groups from the
5872 * seed device are read-only;
5873 *
5874 * 2) Add the device item for the new sprout device - finishing the setup
5875 * of a new block group requires updating the device item in the chunk
5876 * btree, so it must exist when we attempt to do it. The previous step
5877 * ensures this does not fail with -ENOSPC.
5878 *
5879 * After that we can add the block group items to their btrees:
5880 * update existing device item in the chunk btree, add a new block group
5881 * item to the extent btree, add a new chunk item to the chunk btree and
5882 * finally add the new device extent items to the devices btree.
5883 */
5884
5885 alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5886 meta_bg = btrfs_create_chunk(trans, alloc_profile);
5887 if (IS_ERR(meta_bg))
5888 return PTR_ERR(meta_bg);
5889
5890 alloc_profile = btrfs_system_alloc_profile(fs_info);
5891 sys_bg = btrfs_create_chunk(trans, alloc_profile);
5892 if (IS_ERR(sys_bg))
5893 return PTR_ERR(sys_bg);
5894
5895 return 0;
5896 }
5897
5898 static inline int btrfs_chunk_max_errors(struct btrfs_chunk_map *map)
5899 {
5900 const int index = btrfs_bg_flags_to_raid_index(map->type);
5901
5902 return btrfs_raid_array[index].tolerated_failures;
5903 }
5904
5905 bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5906 {
5907 struct btrfs_chunk_map *map;
5908 int miss_ndevs = 0;
5909 int i;
5910 bool ret = true;
5911
5912 map = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5913 if (IS_ERR(map))
5914 return false;
5915
5916 for (i = 0; i < map->num_stripes; i++) {
5917 if (test_bit(BTRFS_DEV_STATE_MISSING,
5918 &map->stripes[i].dev->dev_state)) {
5919 miss_ndevs++;
5920 continue;
5921 }
5922 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5923 &map->stripes[i].dev->dev_state)) {
5924 ret = false;
5925 goto end;
5926 }
5927 }
5928
5929 /*
5930 * If the number of missing devices is larger than max errors, we can
5931 * not write the data into that chunk successfully.
5932 */
5933 if (miss_ndevs > btrfs_chunk_max_errors(map))
5934 ret = false;
5935 end:
5936 btrfs_free_chunk_map(map);
5937 return ret;
5938 }
5939
5940 void btrfs_mapping_tree_free(struct btrfs_fs_info *fs_info)
5941 {
5942 write_lock(&fs_info->mapping_tree_lock);
5943 while (!RB_EMPTY_ROOT(&fs_info->mapping_tree.rb_root)) {
5944 struct btrfs_chunk_map *map;
5945 struct rb_node *node;
5946
5947 node = rb_first_cached(&fs_info->mapping_tree);
5948 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
5949 rb_erase_cached(&map->rb_node, &fs_info->mapping_tree);
5950 RB_CLEAR_NODE(&map->rb_node);
5951 chunk_map_device_clear_bits(map, CHUNK_ALLOCATED);
5952 /* Once for the tree ref. */
5953 btrfs_free_chunk_map(map);
5954 cond_resched_rwlock_write(&fs_info->mapping_tree_lock);
5955 }
5956 write_unlock(&fs_info->mapping_tree_lock);
5957 }
5958
5959 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5960 {
5961 struct btrfs_chunk_map *map;
5962 enum btrfs_raid_types index;
5963 int ret = 1;
5964
5965 map = btrfs_get_chunk_map(fs_info, logical, len);
5966 if (IS_ERR(map))
5967 /*
5968 * We could return errors for these cases, but that could get
5969 * ugly and we'd probably do the same thing which is just not do
5970 * anything else and exit, so return 1 so the callers don't try
5971 * to use other copies.
5972 */
5973 return 1;
5974
5975 index = btrfs_bg_flags_to_raid_index(map->type);
5976
5977 /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5978 if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5979 ret = btrfs_raid_array[index].ncopies;
5980 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5981 ret = 2;
5982 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5983 /*
5984 * There could be two corrupted data stripes, we need
5985 * to loop retry in order to rebuild the correct data.
5986 *
5987 * Fail a stripe at a time on every retry except the
5988 * stripe under reconstruction.
5989 */
5990 ret = map->num_stripes;
5991 btrfs_free_chunk_map(map);
5992 return ret;
5993 }
5994
5995 unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5996 u64 logical)
5997 {
5998 struct btrfs_chunk_map *map;
5999 unsigned long len = fs_info->sectorsize;
6000
6001 if (!btrfs_fs_incompat(fs_info, RAID56))
6002 return len;
6003
6004 map = btrfs_get_chunk_map(fs_info, logical, len);
6005
6006 if (!WARN_ON(IS_ERR(map))) {
6007 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6008 len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6009 btrfs_free_chunk_map(map);
6010 }
6011 return len;
6012 }
6013
6014 int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
6015 {
6016 struct btrfs_chunk_map *map;
6017 int ret = 0;
6018
6019 if (!btrfs_fs_incompat(fs_info, RAID56))
6020 return 0;
6021
6022 map = btrfs_get_chunk_map(fs_info, logical, len);
6023
6024 if (!WARN_ON(IS_ERR(map))) {
6025 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6026 ret = 1;
6027 btrfs_free_chunk_map(map);
6028 }
6029 return ret;
6030 }
6031
6032 static int find_live_mirror(struct btrfs_fs_info *fs_info,
6033 struct btrfs_chunk_map *map, int first,
6034 int dev_replace_is_ongoing)
6035 {
6036 const enum btrfs_read_policy policy = READ_ONCE(fs_info->fs_devices->read_policy);
6037 int i;
6038 int num_stripes;
6039 int preferred_mirror;
6040 int tolerance;
6041 struct btrfs_device *srcdev;
6042
6043 ASSERT((map->type &
6044 (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
6045
6046 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
6047 num_stripes = map->sub_stripes;
6048 else
6049 num_stripes = map->num_stripes;
6050
6051 switch (policy) {
6052 default:
6053 /* Shouldn't happen, just warn and use pid instead of failing */
6054 btrfs_warn_rl(fs_info, "unknown read_policy type %u, reset to pid",
6055 policy);
6056 WRITE_ONCE(fs_info->fs_devices->read_policy, BTRFS_READ_POLICY_PID);
6057 fallthrough;
6058 case BTRFS_READ_POLICY_PID:
6059 preferred_mirror = first + (current->pid % num_stripes);
6060 break;
6061 }
6062
6063 if (dev_replace_is_ongoing &&
6064 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
6065 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
6066 srcdev = fs_info->dev_replace.srcdev;
6067 else
6068 srcdev = NULL;
6069
6070 /*
6071 * try to avoid the drive that is the source drive for a
6072 * dev-replace procedure, only choose it if no other non-missing
6073 * mirror is available
6074 */
6075 for (tolerance = 0; tolerance < 2; tolerance++) {
6076 if (map->stripes[preferred_mirror].dev->bdev &&
6077 (tolerance || map->stripes[preferred_mirror].dev != srcdev))
6078 return preferred_mirror;
6079 for (i = first; i < first + num_stripes; i++) {
6080 if (map->stripes[i].dev->bdev &&
6081 (tolerance || map->stripes[i].dev != srcdev))
6082 return i;
6083 }
6084 }
6085
6086 /* we couldn't find one that doesn't fail. Just return something
6087 * and the io error handling code will clean up eventually
6088 */
6089 return preferred_mirror;
6090 }
6091
6092 static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
6093 u64 logical,
6094 u16 total_stripes)
6095 {
6096 struct btrfs_io_context *bioc;
6097
6098 bioc = kzalloc(
6099 /* The size of btrfs_io_context */
6100 sizeof(struct btrfs_io_context) +
6101 /* Plus the variable array for the stripes */
6102 sizeof(struct btrfs_io_stripe) * (total_stripes),
6103 GFP_NOFS);
6104
6105 if (!bioc)
6106 return NULL;
6107
6108 refcount_set(&bioc->refs, 1);
6109
6110 bioc->fs_info = fs_info;
6111 bioc->replace_stripe_src = -1;
6112 bioc->full_stripe_logical = (u64)-1;
6113 bioc->logical = logical;
6114
6115 return bioc;
6116 }
6117
6118 void btrfs_get_bioc(struct btrfs_io_context *bioc)
6119 {
6120 WARN_ON(!refcount_read(&bioc->refs));
6121 refcount_inc(&bioc->refs);
6122 }
6123
6124 void btrfs_put_bioc(struct btrfs_io_context *bioc)
6125 {
6126 if (!bioc)
6127 return;
6128 if (refcount_dec_and_test(&bioc->refs))
6129 kfree(bioc);
6130 }
6131
6132 /*
6133 * Please note that, discard won't be sent to target device of device
6134 * replace.
6135 */
6136 struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
6137 u64 logical, u64 *length_ret,
6138 u32 *num_stripes)
6139 {
6140 struct btrfs_chunk_map *map;
6141 struct btrfs_discard_stripe *stripes;
6142 u64 length = *length_ret;
6143 u64 offset;
6144 u32 stripe_nr;
6145 u32 stripe_nr_end;
6146 u32 stripe_cnt;
6147 u64 stripe_end_offset;
6148 u64 stripe_offset;
6149 u32 stripe_index;
6150 u32 factor = 0;
6151 u32 sub_stripes = 0;
6152 u32 stripes_per_dev = 0;
6153 u32 remaining_stripes = 0;
6154 u32 last_stripe = 0;
6155 int ret;
6156 int i;
6157
6158 map = btrfs_get_chunk_map(fs_info, logical, length);
6159 if (IS_ERR(map))
6160 return ERR_CAST(map);
6161
6162 /* we don't discard raid56 yet */
6163 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6164 ret = -EOPNOTSUPP;
6165 goto out_free_map;
6166 }
6167
6168 offset = logical - map->start;
6169 length = min_t(u64, map->start + map->chunk_len - logical, length);
6170 *length_ret = length;
6171
6172 /*
6173 * stripe_nr counts the total number of stripes we have to stride
6174 * to get to this block
6175 */
6176 stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6177
6178 /* stripe_offset is the offset of this block in its stripe */
6179 stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
6180
6181 stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
6182 BTRFS_STRIPE_LEN_SHIFT;
6183 stripe_cnt = stripe_nr_end - stripe_nr;
6184 stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
6185 (offset + length);
6186 /*
6187 * after this, stripe_nr is the number of stripes on this
6188 * device we have to walk to find the data, and stripe_index is
6189 * the number of our device in the stripe array
6190 */
6191 *num_stripes = 1;
6192 stripe_index = 0;
6193 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6194 BTRFS_BLOCK_GROUP_RAID10)) {
6195 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
6196 sub_stripes = 1;
6197 else
6198 sub_stripes = map->sub_stripes;
6199
6200 factor = map->num_stripes / sub_stripes;
6201 *num_stripes = min_t(u64, map->num_stripes,
6202 sub_stripes * stripe_cnt);
6203 stripe_index = stripe_nr % factor;
6204 stripe_nr /= factor;
6205 stripe_index *= sub_stripes;
6206
6207 remaining_stripes = stripe_cnt % factor;
6208 stripes_per_dev = stripe_cnt / factor;
6209 last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6210 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6211 BTRFS_BLOCK_GROUP_DUP)) {
6212 *num_stripes = map->num_stripes;
6213 } else {
6214 stripe_index = stripe_nr % map->num_stripes;
6215 stripe_nr /= map->num_stripes;
6216 }
6217
6218 stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6219 if (!stripes) {
6220 ret = -ENOMEM;
6221 goto out_free_map;
6222 }
6223
6224 for (i = 0; i < *num_stripes; i++) {
6225 stripes[i].physical =
6226 map->stripes[stripe_index].physical +
6227 stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6228 stripes[i].dev = map->stripes[stripe_index].dev;
6229
6230 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6231 BTRFS_BLOCK_GROUP_RAID10)) {
6232 stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6233
6234 if (i / sub_stripes < remaining_stripes)
6235 stripes[i].length += BTRFS_STRIPE_LEN;
6236
6237 /*
6238 * Special for the first stripe and
6239 * the last stripe:
6240 *
6241 * |-------|...|-------|
6242 * |----------|
6243 * off end_off
6244 */
6245 if (i < sub_stripes)
6246 stripes[i].length -= stripe_offset;
6247
6248 if (stripe_index >= last_stripe &&
6249 stripe_index <= (last_stripe +
6250 sub_stripes - 1))
6251 stripes[i].length -= stripe_end_offset;
6252
6253 if (i == sub_stripes - 1)
6254 stripe_offset = 0;
6255 } else {
6256 stripes[i].length = length;
6257 }
6258
6259 stripe_index++;
6260 if (stripe_index == map->num_stripes) {
6261 stripe_index = 0;
6262 stripe_nr++;
6263 }
6264 }
6265
6266 btrfs_free_chunk_map(map);
6267 return stripes;
6268 out_free_map:
6269 btrfs_free_chunk_map(map);
6270 return ERR_PTR(ret);
6271 }
6272
6273 static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6274 {
6275 struct btrfs_block_group *cache;
6276 bool ret;
6277
6278 /* Non zoned filesystem does not use "to_copy" flag */
6279 if (!btrfs_is_zoned(fs_info))
6280 return false;
6281
6282 cache = btrfs_lookup_block_group(fs_info, logical);
6283
6284 ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6285
6286 btrfs_put_block_group(cache);
6287 return ret;
6288 }
6289
6290 static void handle_ops_on_dev_replace(struct btrfs_io_context *bioc,
6291 struct btrfs_dev_replace *dev_replace,
6292 u64 logical,
6293 struct btrfs_io_geometry *io_geom)
6294 {
6295 u64 srcdev_devid = dev_replace->srcdev->devid;
6296 /*
6297 * At this stage, num_stripes is still the real number of stripes,
6298 * excluding the duplicated stripes.
6299 */
6300 int num_stripes = io_geom->num_stripes;
6301 int max_errors = io_geom->max_errors;
6302 int nr_extra_stripes = 0;
6303 int i;
6304
6305 /*
6306 * A block group which has "to_copy" set will eventually be copied by
6307 * the dev-replace process. We can avoid cloning IO here.
6308 */
6309 if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6310 return;
6311
6312 /*
6313 * Duplicate the write operations while the dev-replace procedure is
6314 * running. Since the copying of the old disk to the new disk takes
6315 * place at run time while the filesystem is mounted writable, the
6316 * regular write operations to the old disk have to be duplicated to go
6317 * to the new disk as well.
6318 *
6319 * Note that device->missing is handled by the caller, and that the
6320 * write to the old disk is already set up in the stripes array.
6321 */
6322 for (i = 0; i < num_stripes; i++) {
6323 struct btrfs_io_stripe *old = &bioc->stripes[i];
6324 struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6325
6326 if (old->dev->devid != srcdev_devid)
6327 continue;
6328
6329 new->physical = old->physical;
6330 new->dev = dev_replace->tgtdev;
6331 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6332 bioc->replace_stripe_src = i;
6333 nr_extra_stripes++;
6334 }
6335
6336 /* We can only have at most 2 extra nr_stripes (for DUP). */
6337 ASSERT(nr_extra_stripes <= 2);
6338 /*
6339 * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6340 * replace.
6341 * If we have 2 extra stripes, only choose the one with smaller physical.
6342 */
6343 if (io_geom->op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6344 struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6345 struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6346
6347 /* Only DUP can have two extra stripes. */
6348 ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6349
6350 /*
6351 * Swap the last stripe stripes and reduce @nr_extra_stripes.
6352 * The extra stripe would still be there, but won't be accessed.
6353 */
6354 if (first->physical > second->physical) {
6355 swap(second->physical, first->physical);
6356 swap(second->dev, first->dev);
6357 nr_extra_stripes--;
6358 }
6359 }
6360
6361 io_geom->num_stripes = num_stripes + nr_extra_stripes;
6362 io_geom->max_errors = max_errors + nr_extra_stripes;
6363 bioc->replace_nr_stripes = nr_extra_stripes;
6364 }
6365
6366 static u64 btrfs_max_io_len(struct btrfs_chunk_map *map, u64 offset,
6367 struct btrfs_io_geometry *io_geom)
6368 {
6369 /*
6370 * Stripe_nr is the stripe where this block falls. stripe_offset is
6371 * the offset of this block in its stripe.
6372 */
6373 io_geom->stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6374 io_geom->stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6375 ASSERT(io_geom->stripe_offset < U32_MAX);
6376
6377 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6378 unsigned long full_stripe_len =
6379 btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6380
6381 /*
6382 * For full stripe start, we use previously calculated
6383 * @stripe_nr. Align it to nr_data_stripes, then multiply with
6384 * STRIPE_LEN.
6385 *
6386 * By this we can avoid u64 division completely. And we have
6387 * to go rounddown(), not round_down(), as nr_data_stripes is
6388 * not ensured to be power of 2.
6389 */
6390 io_geom->raid56_full_stripe_start = btrfs_stripe_nr_to_offset(
6391 rounddown(io_geom->stripe_nr, nr_data_stripes(map)));
6392
6393 ASSERT(io_geom->raid56_full_stripe_start + full_stripe_len > offset);
6394 ASSERT(io_geom->raid56_full_stripe_start <= offset);
6395 /*
6396 * For writes to RAID56, allow to write a full stripe set, but
6397 * no straddling of stripe sets.
6398 */
6399 if (io_geom->op == BTRFS_MAP_WRITE)
6400 return full_stripe_len - (offset - io_geom->raid56_full_stripe_start);
6401 }
6402
6403 /*
6404 * For other RAID types and for RAID56 reads, allow a single stripe (on
6405 * a single disk).
6406 */
6407 if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6408 return BTRFS_STRIPE_LEN - io_geom->stripe_offset;
6409 return U64_MAX;
6410 }
6411
6412 static int set_io_stripe(struct btrfs_fs_info *fs_info, u64 logical,
6413 u64 *length, struct btrfs_io_stripe *dst,
6414 struct btrfs_chunk_map *map,
6415 struct btrfs_io_geometry *io_geom)
6416 {
6417 dst->dev = map->stripes[io_geom->stripe_index].dev;
6418
6419 if (io_geom->op == BTRFS_MAP_READ &&
6420 btrfs_need_stripe_tree_update(fs_info, map->type))
6421 return btrfs_get_raid_extent_offset(fs_info, logical, length,
6422 map->type,
6423 io_geom->stripe_index, dst);
6424
6425 dst->physical = map->stripes[io_geom->stripe_index].physical +
6426 io_geom->stripe_offset +
6427 btrfs_stripe_nr_to_offset(io_geom->stripe_nr);
6428 return 0;
6429 }
6430
6431 static bool is_single_device_io(struct btrfs_fs_info *fs_info,
6432 const struct btrfs_io_stripe *smap,
6433 const struct btrfs_chunk_map *map,
6434 int num_alloc_stripes,
6435 enum btrfs_map_op op, int mirror_num)
6436 {
6437 if (!smap)
6438 return false;
6439
6440 if (num_alloc_stripes != 1)
6441 return false;
6442
6443 if (btrfs_need_stripe_tree_update(fs_info, map->type) && op != BTRFS_MAP_READ)
6444 return false;
6445
6446 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1)
6447 return false;
6448
6449 return true;
6450 }
6451
6452 static void map_blocks_raid0(const struct btrfs_chunk_map *map,
6453 struct btrfs_io_geometry *io_geom)
6454 {
6455 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6456 io_geom->stripe_nr /= map->num_stripes;
6457 if (io_geom->op == BTRFS_MAP_READ)
6458 io_geom->mirror_num = 1;
6459 }
6460
6461 static void map_blocks_raid1(struct btrfs_fs_info *fs_info,
6462 struct btrfs_chunk_map *map,
6463 struct btrfs_io_geometry *io_geom,
6464 bool dev_replace_is_ongoing)
6465 {
6466 if (io_geom->op != BTRFS_MAP_READ) {
6467 io_geom->num_stripes = map->num_stripes;
6468 return;
6469 }
6470
6471 if (io_geom->mirror_num) {
6472 io_geom->stripe_index = io_geom->mirror_num - 1;
6473 return;
6474 }
6475
6476 io_geom->stripe_index = find_live_mirror(fs_info, map, 0,
6477 dev_replace_is_ongoing);
6478 io_geom->mirror_num = io_geom->stripe_index + 1;
6479 }
6480
6481 static void map_blocks_dup(const struct btrfs_chunk_map *map,
6482 struct btrfs_io_geometry *io_geom)
6483 {
6484 if (io_geom->op != BTRFS_MAP_READ) {
6485 io_geom->num_stripes = map->num_stripes;
6486 return;
6487 }
6488
6489 if (io_geom->mirror_num) {
6490 io_geom->stripe_index = io_geom->mirror_num - 1;
6491 return;
6492 }
6493
6494 io_geom->mirror_num = 1;
6495 }
6496
6497 static void map_blocks_raid10(struct btrfs_fs_info *fs_info,
6498 struct btrfs_chunk_map *map,
6499 struct btrfs_io_geometry *io_geom,
6500 bool dev_replace_is_ongoing)
6501 {
6502 u32 factor = map->num_stripes / map->sub_stripes;
6503 int old_stripe_index;
6504
6505 io_geom->stripe_index = (io_geom->stripe_nr % factor) * map->sub_stripes;
6506 io_geom->stripe_nr /= factor;
6507
6508 if (io_geom->op != BTRFS_MAP_READ) {
6509 io_geom->num_stripes = map->sub_stripes;
6510 return;
6511 }
6512
6513 if (io_geom->mirror_num) {
6514 io_geom->stripe_index += io_geom->mirror_num - 1;
6515 return;
6516 }
6517
6518 old_stripe_index = io_geom->stripe_index;
6519 io_geom->stripe_index = find_live_mirror(fs_info, map,
6520 io_geom->stripe_index,
6521 dev_replace_is_ongoing);
6522 io_geom->mirror_num = io_geom->stripe_index - old_stripe_index + 1;
6523 }
6524
6525 static void map_blocks_raid56_write(struct btrfs_chunk_map *map,
6526 struct btrfs_io_geometry *io_geom,
6527 u64 logical, u64 *length)
6528 {
6529 int data_stripes = nr_data_stripes(map);
6530
6531 /*
6532 * Needs full stripe mapping.
6533 *
6534 * Push stripe_nr back to the start of the full stripe For those cases
6535 * needing a full stripe, @stripe_nr is the full stripe number.
6536 *
6537 * Originally we go raid56_full_stripe_start / full_stripe_len, but
6538 * that can be expensive. Here we just divide @stripe_nr with
6539 * @data_stripes.
6540 */
6541 io_geom->stripe_nr /= data_stripes;
6542
6543 /* RAID[56] write or recovery. Return all stripes */
6544 io_geom->num_stripes = map->num_stripes;
6545 io_geom->max_errors = btrfs_chunk_max_errors(map);
6546
6547 /* Return the length to the full stripe end. */
6548 *length = min(logical + *length,
6549 io_geom->raid56_full_stripe_start + map->start +
6550 btrfs_stripe_nr_to_offset(data_stripes)) -
6551 logical;
6552 io_geom->stripe_index = 0;
6553 io_geom->stripe_offset = 0;
6554 }
6555
6556 static void map_blocks_raid56_read(struct btrfs_chunk_map *map,
6557 struct btrfs_io_geometry *io_geom)
6558 {
6559 int data_stripes = nr_data_stripes(map);
6560
6561 ASSERT(io_geom->mirror_num <= 1);
6562 /* Just grab the data stripe directly. */
6563 io_geom->stripe_index = io_geom->stripe_nr % data_stripes;
6564 io_geom->stripe_nr /= data_stripes;
6565
6566 /* We distribute the parity blocks across stripes. */
6567 io_geom->stripe_index =
6568 (io_geom->stripe_nr + io_geom->stripe_index) % map->num_stripes;
6569
6570 if (io_geom->op == BTRFS_MAP_READ && io_geom->mirror_num < 1)
6571 io_geom->mirror_num = 1;
6572 }
6573
6574 static void map_blocks_single(const struct btrfs_chunk_map *map,
6575 struct btrfs_io_geometry *io_geom)
6576 {
6577 io_geom->stripe_index = io_geom->stripe_nr % map->num_stripes;
6578 io_geom->stripe_nr /= map->num_stripes;
6579 io_geom->mirror_num = io_geom->stripe_index + 1;
6580 }
6581
6582 /*
6583 * Map one logical range to one or more physical ranges.
6584 *
6585 * @length: (Mandatory) mapped length of this run.
6586 * One logical range can be split into different segments
6587 * due to factors like zones and RAID0/5/6/10 stripe
6588 * boundaries.
6589 *
6590 * @bioc_ret: (Mandatory) returned btrfs_io_context structure.
6591 * which has one or more physical ranges (btrfs_io_stripe)
6592 * recorded inside.
6593 * Caller should call btrfs_put_bioc() to free it after use.
6594 *
6595 * @smap: (Optional) single physical range optimization.
6596 * If the map request can be fulfilled by one single
6597 * physical range, and this is parameter is not NULL,
6598 * then @bioc_ret would be NULL, and @smap would be
6599 * updated.
6600 *
6601 * @mirror_num_ret: (Mandatory) returned mirror number if the original
6602 * value is 0.
6603 *
6604 * Mirror number 0 means to choose any live mirrors.
6605 *
6606 * For non-RAID56 profiles, non-zero mirror_num means
6607 * the Nth mirror. (e.g. mirror_num 1 means the first
6608 * copy).
6609 *
6610 * For RAID56 profile, mirror 1 means rebuild from P and
6611 * the remaining data stripes.
6612 *
6613 * For RAID6 profile, mirror > 2 means mark another
6614 * data/P stripe error and rebuild from the remaining
6615 * stripes..
6616 */
6617 int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6618 u64 logical, u64 *length,
6619 struct btrfs_io_context **bioc_ret,
6620 struct btrfs_io_stripe *smap, int *mirror_num_ret)
6621 {
6622 struct btrfs_chunk_map *map;
6623 struct btrfs_io_geometry io_geom = { 0 };
6624 u64 map_offset;
6625 int ret = 0;
6626 int num_copies;
6627 struct btrfs_io_context *bioc = NULL;
6628 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6629 int dev_replace_is_ongoing = 0;
6630 u16 num_alloc_stripes;
6631 u64 max_len;
6632
6633 ASSERT(bioc_ret);
6634
6635 io_geom.mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6636 io_geom.num_stripes = 1;
6637 io_geom.stripe_index = 0;
6638 io_geom.op = op;
6639
6640 num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6641 if (io_geom.mirror_num > num_copies)
6642 return -EINVAL;
6643
6644 map = btrfs_get_chunk_map(fs_info, logical, *length);
6645 if (IS_ERR(map))
6646 return PTR_ERR(map);
6647
6648 map_offset = logical - map->start;
6649 io_geom.raid56_full_stripe_start = (u64)-1;
6650 max_len = btrfs_max_io_len(map, map_offset, &io_geom);
6651 *length = min_t(u64, map->chunk_len - map_offset, max_len);
6652
6653 down_read(&dev_replace->rwsem);
6654 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6655 /*
6656 * Hold the semaphore for read during the whole operation, write is
6657 * requested at commit time but must wait.
6658 */
6659 if (!dev_replace_is_ongoing)
6660 up_read(&dev_replace->rwsem);
6661
6662 switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
6663 case BTRFS_BLOCK_GROUP_RAID0:
6664 map_blocks_raid0(map, &io_geom);
6665 break;
6666 case BTRFS_BLOCK_GROUP_RAID1:
6667 case BTRFS_BLOCK_GROUP_RAID1C3:
6668 case BTRFS_BLOCK_GROUP_RAID1C4:
6669 map_blocks_raid1(fs_info, map, &io_geom, dev_replace_is_ongoing);
6670 break;
6671 case BTRFS_BLOCK_GROUP_DUP:
6672 map_blocks_dup(map, &io_geom);
6673 break;
6674 case BTRFS_BLOCK_GROUP_RAID10:
6675 map_blocks_raid10(fs_info, map, &io_geom, dev_replace_is_ongoing);
6676 break;
6677 case BTRFS_BLOCK_GROUP_RAID5:
6678 case BTRFS_BLOCK_GROUP_RAID6:
6679 if (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)
6680 map_blocks_raid56_write(map, &io_geom, logical, length);
6681 else
6682 map_blocks_raid56_read(map, &io_geom);
6683 break;
6684 default:
6685 /*
6686 * After this, stripe_nr is the number of stripes on this
6687 * device we have to walk to find the data, and stripe_index is
6688 * the number of our device in the stripe array
6689 */
6690 map_blocks_single(map, &io_geom);
6691 break;
6692 }
6693 if (io_geom.stripe_index >= map->num_stripes) {
6694 btrfs_crit(fs_info,
6695 "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6696 io_geom.stripe_index, map->num_stripes);
6697 ret = -EINVAL;
6698 goto out;
6699 }
6700
6701 num_alloc_stripes = io_geom.num_stripes;
6702 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6703 op != BTRFS_MAP_READ)
6704 /*
6705 * For replace case, we need to add extra stripes for extra
6706 * duplicated stripes.
6707 *
6708 * For both WRITE and GET_READ_MIRRORS, we may have at most
6709 * 2 more stripes (DUP types, otherwise 1).
6710 */
6711 num_alloc_stripes += 2;
6712
6713 /*
6714 * If this I/O maps to a single device, try to return the device and
6715 * physical block information on the stack instead of allocating an
6716 * I/O context structure.
6717 */
6718 if (is_single_device_io(fs_info, smap, map, num_alloc_stripes, op,
6719 io_geom.mirror_num)) {
6720 ret = set_io_stripe(fs_info, logical, length, smap, map, &io_geom);
6721 if (mirror_num_ret)
6722 *mirror_num_ret = io_geom.mirror_num;
6723 *bioc_ret = NULL;
6724 goto out;
6725 }
6726
6727 bioc = alloc_btrfs_io_context(fs_info, logical, num_alloc_stripes);
6728 if (!bioc) {
6729 ret = -ENOMEM;
6730 goto out;
6731 }
6732 bioc->map_type = map->type;
6733
6734 /*
6735 * For RAID56 full map, we need to make sure the stripes[] follows the
6736 * rule that data stripes are all ordered, then followed with P and Q
6737 * (if we have).
6738 *
6739 * It's still mostly the same as other profiles, just with extra rotation.
6740 */
6741 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
6742 (op != BTRFS_MAP_READ || io_geom.mirror_num > 1)) {
6743 /*
6744 * For RAID56 @stripe_nr is already the number of full stripes
6745 * before us, which is also the rotation value (needs to modulo
6746 * with num_stripes).
6747 *
6748 * In this case, we just add @stripe_nr with @i, then do the
6749 * modulo, to reduce one modulo call.
6750 */
6751 bioc->full_stripe_logical = map->start +
6752 btrfs_stripe_nr_to_offset(io_geom.stripe_nr *
6753 nr_data_stripes(map));
6754 for (int i = 0; i < io_geom.num_stripes; i++) {
6755 struct btrfs_io_stripe *dst = &bioc->stripes[i];
6756 u32 stripe_index;
6757
6758 stripe_index = (i + io_geom.stripe_nr) % io_geom.num_stripes;
6759 dst->dev = map->stripes[stripe_index].dev;
6760 dst->physical =
6761 map->stripes[stripe_index].physical +
6762 io_geom.stripe_offset +
6763 btrfs_stripe_nr_to_offset(io_geom.stripe_nr);
6764 }
6765 } else {
6766 /*
6767 * For all other non-RAID56 profiles, just copy the target
6768 * stripe into the bioc.
6769 */
6770 for (int i = 0; i < io_geom.num_stripes; i++) {
6771 ret = set_io_stripe(fs_info, logical, length,
6772 &bioc->stripes[i], map, &io_geom);
6773 if (ret < 0)
6774 break;
6775 io_geom.stripe_index++;
6776 }
6777 }
6778
6779 if (ret) {
6780 *bioc_ret = NULL;
6781 btrfs_put_bioc(bioc);
6782 goto out;
6783 }
6784
6785 if (op != BTRFS_MAP_READ)
6786 io_geom.max_errors = btrfs_chunk_max_errors(map);
6787
6788 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6789 op != BTRFS_MAP_READ) {
6790 handle_ops_on_dev_replace(bioc, dev_replace, logical, &io_geom);
6791 }
6792
6793 *bioc_ret = bioc;
6794 bioc->num_stripes = io_geom.num_stripes;
6795 bioc->max_errors = io_geom.max_errors;
6796 bioc->mirror_num = io_geom.mirror_num;
6797
6798 out:
6799 if (dev_replace_is_ongoing) {
6800 lockdep_assert_held(&dev_replace->rwsem);
6801 /* Unlock and let waiting writers proceed */
6802 up_read(&dev_replace->rwsem);
6803 }
6804 btrfs_free_chunk_map(map);
6805 return ret;
6806 }
6807
6808 static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6809 const struct btrfs_fs_devices *fs_devices)
6810 {
6811 if (args->fsid == NULL)
6812 return true;
6813 if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6814 return true;
6815 return false;
6816 }
6817
6818 static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6819 const struct btrfs_device *device)
6820 {
6821 if (args->missing) {
6822 if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6823 !device->bdev)
6824 return true;
6825 return false;
6826 }
6827
6828 if (device->devid != args->devid)
6829 return false;
6830 if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6831 return false;
6832 return true;
6833 }
6834
6835 /*
6836 * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6837 * return NULL.
6838 *
6839 * If devid and uuid are both specified, the match must be exact, otherwise
6840 * only devid is used.
6841 */
6842 struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6843 const struct btrfs_dev_lookup_args *args)
6844 {
6845 struct btrfs_device *device;
6846 struct btrfs_fs_devices *seed_devs;
6847
6848 if (dev_args_match_fs_devices(args, fs_devices)) {
6849 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6850 if (dev_args_match_device(args, device))
6851 return device;
6852 }
6853 }
6854
6855 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6856 if (!dev_args_match_fs_devices(args, seed_devs))
6857 continue;
6858 list_for_each_entry(device, &seed_devs->devices, dev_list) {
6859 if (dev_args_match_device(args, device))
6860 return device;
6861 }
6862 }
6863
6864 return NULL;
6865 }
6866
6867 static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6868 u64 devid, u8 *dev_uuid)
6869 {
6870 struct btrfs_device *device;
6871 unsigned int nofs_flag;
6872
6873 /*
6874 * We call this under the chunk_mutex, so we want to use NOFS for this
6875 * allocation, however we don't want to change btrfs_alloc_device() to
6876 * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6877 * places.
6878 */
6879
6880 nofs_flag = memalloc_nofs_save();
6881 device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6882 memalloc_nofs_restore(nofs_flag);
6883 if (IS_ERR(device))
6884 return device;
6885
6886 list_add(&device->dev_list, &fs_devices->devices);
6887 device->fs_devices = fs_devices;
6888 fs_devices->num_devices++;
6889
6890 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6891 fs_devices->missing_devices++;
6892
6893 return device;
6894 }
6895
6896 /*
6897 * Allocate new device struct, set up devid and UUID.
6898 *
6899 * @fs_info: used only for generating a new devid, can be NULL if
6900 * devid is provided (i.e. @devid != NULL).
6901 * @devid: a pointer to devid for this device. If NULL a new devid
6902 * is generated.
6903 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6904 * is generated.
6905 * @path: a pointer to device path if available, NULL otherwise.
6906 *
6907 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6908 * on error. Returned struct is not linked onto any lists and must be
6909 * destroyed with btrfs_free_device.
6910 */
6911 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6912 const u64 *devid, const u8 *uuid,
6913 const char *path)
6914 {
6915 struct btrfs_device *dev;
6916 u64 tmp;
6917
6918 if (WARN_ON(!devid && !fs_info))
6919 return ERR_PTR(-EINVAL);
6920
6921 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6922 if (!dev)
6923 return ERR_PTR(-ENOMEM);
6924
6925 INIT_LIST_HEAD(&dev->dev_list);
6926 INIT_LIST_HEAD(&dev->dev_alloc_list);
6927 INIT_LIST_HEAD(&dev->post_commit_list);
6928
6929 atomic_set(&dev->dev_stats_ccnt, 0);
6930 btrfs_device_data_ordered_init(dev);
6931 extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6932
6933 if (devid)
6934 tmp = *devid;
6935 else {
6936 int ret;
6937
6938 ret = find_next_devid(fs_info, &tmp);
6939 if (ret) {
6940 btrfs_free_device(dev);
6941 return ERR_PTR(ret);
6942 }
6943 }
6944 dev->devid = tmp;
6945
6946 if (uuid)
6947 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6948 else
6949 generate_random_uuid(dev->uuid);
6950
6951 if (path) {
6952 struct rcu_string *name;
6953
6954 name = rcu_string_strdup(path, GFP_KERNEL);
6955 if (!name) {
6956 btrfs_free_device(dev);
6957 return ERR_PTR(-ENOMEM);
6958 }
6959 rcu_assign_pointer(dev->name, name);
6960 }
6961
6962 return dev;
6963 }
6964
6965 static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6966 u64 devid, u8 *uuid, bool error)
6967 {
6968 if (error)
6969 btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6970 devid, uuid);
6971 else
6972 btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6973 devid, uuid);
6974 }
6975
6976 u64 btrfs_calc_stripe_length(const struct btrfs_chunk_map *map)
6977 {
6978 const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6979
6980 return div_u64(map->chunk_len, data_stripes);
6981 }
6982
6983 #if BITS_PER_LONG == 32
6984 /*
6985 * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6986 * can't be accessed on 32bit systems.
6987 *
6988 * This function do mount time check to reject the fs if it already has
6989 * metadata chunk beyond that limit.
6990 */
6991 static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6992 u64 logical, u64 length, u64 type)
6993 {
6994 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6995 return 0;
6996
6997 if (logical + length < MAX_LFS_FILESIZE)
6998 return 0;
6999
7000 btrfs_err_32bit_limit(fs_info);
7001 return -EOVERFLOW;
7002 }
7003
7004 /*
7005 * This is to give early warning for any metadata chunk reaching
7006 * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
7007 * Although we can still access the metadata, it's not going to be possible
7008 * once the limit is reached.
7009 */
7010 static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
7011 u64 logical, u64 length, u64 type)
7012 {
7013 if (!(type & BTRFS_BLOCK_GROUP_METADATA))
7014 return;
7015
7016 if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
7017 return;
7018
7019 btrfs_warn_32bit_limit(fs_info);
7020 }
7021 #endif
7022
7023 static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
7024 u64 devid, u8 *uuid)
7025 {
7026 struct btrfs_device *dev;
7027
7028 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7029 btrfs_report_missing_device(fs_info, devid, uuid, true);
7030 return ERR_PTR(-ENOENT);
7031 }
7032
7033 dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
7034 if (IS_ERR(dev)) {
7035 btrfs_err(fs_info, "failed to init missing device %llu: %ld",
7036 devid, PTR_ERR(dev));
7037 return dev;
7038 }
7039 btrfs_report_missing_device(fs_info, devid, uuid, false);
7040
7041 return dev;
7042 }
7043
7044 static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
7045 struct btrfs_chunk *chunk)
7046 {
7047 BTRFS_DEV_LOOKUP_ARGS(args);
7048 struct btrfs_fs_info *fs_info = leaf->fs_info;
7049 struct btrfs_chunk_map *map;
7050 u64 logical;
7051 u64 length;
7052 u64 devid;
7053 u64 type;
7054 u8 uuid[BTRFS_UUID_SIZE];
7055 int index;
7056 int num_stripes;
7057 int ret;
7058 int i;
7059
7060 logical = key->offset;
7061 length = btrfs_chunk_length(leaf, chunk);
7062 type = btrfs_chunk_type(leaf, chunk);
7063 index = btrfs_bg_flags_to_raid_index(type);
7064 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
7065
7066 #if BITS_PER_LONG == 32
7067 ret = check_32bit_meta_chunk(fs_info, logical, length, type);
7068 if (ret < 0)
7069 return ret;
7070 warn_32bit_meta_chunk(fs_info, logical, length, type);
7071 #endif
7072
7073 /*
7074 * Only need to verify chunk item if we're reading from sys chunk array,
7075 * as chunk item in tree block is already verified by tree-checker.
7076 */
7077 if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
7078 ret = btrfs_check_chunk_valid(leaf, chunk, logical);
7079 if (ret)
7080 return ret;
7081 }
7082
7083 map = btrfs_find_chunk_map(fs_info, logical, 1);
7084
7085 /* already mapped? */
7086 if (map && map->start <= logical && map->start + map->chunk_len > logical) {
7087 btrfs_free_chunk_map(map);
7088 return 0;
7089 } else if (map) {
7090 btrfs_free_chunk_map(map);
7091 }
7092
7093 map = btrfs_alloc_chunk_map(num_stripes, GFP_NOFS);
7094 if (!map)
7095 return -ENOMEM;
7096
7097 map->start = logical;
7098 map->chunk_len = length;
7099 map->num_stripes = num_stripes;
7100 map->io_width = btrfs_chunk_io_width(leaf, chunk);
7101 map->io_align = btrfs_chunk_io_align(leaf, chunk);
7102 map->type = type;
7103 /*
7104 * We can't use the sub_stripes value, as for profiles other than
7105 * RAID10, they may have 0 as sub_stripes for filesystems created by
7106 * older mkfs (<v5.4).
7107 * In that case, it can cause divide-by-zero errors later.
7108 * Since currently sub_stripes is fixed for each profile, let's
7109 * use the trusted value instead.
7110 */
7111 map->sub_stripes = btrfs_raid_array[index].sub_stripes;
7112 map->verified_stripes = 0;
7113 map->stripe_size = btrfs_calc_stripe_length(map);
7114 for (i = 0; i < num_stripes; i++) {
7115 map->stripes[i].physical =
7116 btrfs_stripe_offset_nr(leaf, chunk, i);
7117 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
7118 args.devid = devid;
7119 read_extent_buffer(leaf, uuid, (unsigned long)
7120 btrfs_stripe_dev_uuid_nr(chunk, i),
7121 BTRFS_UUID_SIZE);
7122 args.uuid = uuid;
7123 map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
7124 if (!map->stripes[i].dev) {
7125 map->stripes[i].dev = handle_missing_device(fs_info,
7126 devid, uuid);
7127 if (IS_ERR(map->stripes[i].dev)) {
7128 ret = PTR_ERR(map->stripes[i].dev);
7129 btrfs_free_chunk_map(map);
7130 return ret;
7131 }
7132 }
7133
7134 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
7135 &(map->stripes[i].dev->dev_state));
7136 }
7137
7138 ret = btrfs_add_chunk_map(fs_info, map);
7139 if (ret < 0) {
7140 btrfs_err(fs_info,
7141 "failed to add chunk map, start=%llu len=%llu: %d",
7142 map->start, map->chunk_len, ret);
7143 }
7144
7145 return ret;
7146 }
7147
7148 static void fill_device_from_item(struct extent_buffer *leaf,
7149 struct btrfs_dev_item *dev_item,
7150 struct btrfs_device *device)
7151 {
7152 unsigned long ptr;
7153
7154 device->devid = btrfs_device_id(leaf, dev_item);
7155 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
7156 device->total_bytes = device->disk_total_bytes;
7157 device->commit_total_bytes = device->disk_total_bytes;
7158 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
7159 device->commit_bytes_used = device->bytes_used;
7160 device->type = btrfs_device_type(leaf, dev_item);
7161 device->io_align = btrfs_device_io_align(leaf, dev_item);
7162 device->io_width = btrfs_device_io_width(leaf, dev_item);
7163 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
7164 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
7165 clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
7166
7167 ptr = btrfs_device_uuid(dev_item);
7168 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
7169 }
7170
7171 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
7172 u8 *fsid)
7173 {
7174 struct btrfs_fs_devices *fs_devices;
7175 int ret;
7176
7177 lockdep_assert_held(&uuid_mutex);
7178 ASSERT(fsid);
7179
7180 /* This will match only for multi-device seed fs */
7181 list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
7182 if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
7183 return fs_devices;
7184
7185
7186 fs_devices = find_fsid(fsid, NULL);
7187 if (!fs_devices) {
7188 if (!btrfs_test_opt(fs_info, DEGRADED))
7189 return ERR_PTR(-ENOENT);
7190
7191 fs_devices = alloc_fs_devices(fsid);
7192 if (IS_ERR(fs_devices))
7193 return fs_devices;
7194
7195 fs_devices->seeding = true;
7196 fs_devices->opened = 1;
7197 return fs_devices;
7198 }
7199
7200 /*
7201 * Upon first call for a seed fs fsid, just create a private copy of the
7202 * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
7203 */
7204 fs_devices = clone_fs_devices(fs_devices);
7205 if (IS_ERR(fs_devices))
7206 return fs_devices;
7207
7208 ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
7209 if (ret) {
7210 free_fs_devices(fs_devices);
7211 return ERR_PTR(ret);
7212 }
7213
7214 if (!fs_devices->seeding) {
7215 close_fs_devices(fs_devices);
7216 free_fs_devices(fs_devices);
7217 return ERR_PTR(-EINVAL);
7218 }
7219
7220 list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
7221
7222 return fs_devices;
7223 }
7224
7225 static int read_one_dev(struct extent_buffer *leaf,
7226 struct btrfs_dev_item *dev_item)
7227 {
7228 BTRFS_DEV_LOOKUP_ARGS(args);
7229 struct btrfs_fs_info *fs_info = leaf->fs_info;
7230 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7231 struct btrfs_device *device;
7232 u64 devid;
7233 int ret;
7234 u8 fs_uuid[BTRFS_FSID_SIZE];
7235 u8 dev_uuid[BTRFS_UUID_SIZE];
7236
7237 devid = btrfs_device_id(leaf, dev_item);
7238 args.devid = devid;
7239 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
7240 BTRFS_UUID_SIZE);
7241 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
7242 BTRFS_FSID_SIZE);
7243 args.uuid = dev_uuid;
7244 args.fsid = fs_uuid;
7245
7246 if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
7247 fs_devices = open_seed_devices(fs_info, fs_uuid);
7248 if (IS_ERR(fs_devices))
7249 return PTR_ERR(fs_devices);
7250 }
7251
7252 device = btrfs_find_device(fs_info->fs_devices, &args);
7253 if (!device) {
7254 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7255 btrfs_report_missing_device(fs_info, devid,
7256 dev_uuid, true);
7257 return -ENOENT;
7258 }
7259
7260 device = add_missing_dev(fs_devices, devid, dev_uuid);
7261 if (IS_ERR(device)) {
7262 btrfs_err(fs_info,
7263 "failed to add missing dev %llu: %ld",
7264 devid, PTR_ERR(device));
7265 return PTR_ERR(device);
7266 }
7267 btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
7268 } else {
7269 if (!device->bdev) {
7270 if (!btrfs_test_opt(fs_info, DEGRADED)) {
7271 btrfs_report_missing_device(fs_info,
7272 devid, dev_uuid, true);
7273 return -ENOENT;
7274 }
7275 btrfs_report_missing_device(fs_info, devid,
7276 dev_uuid, false);
7277 }
7278
7279 if (!device->bdev &&
7280 !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
7281 /*
7282 * this happens when a device that was properly setup
7283 * in the device info lists suddenly goes bad.
7284 * device->bdev is NULL, and so we have to set
7285 * device->missing to one here
7286 */
7287 device->fs_devices->missing_devices++;
7288 set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
7289 }
7290
7291 /* Move the device to its own fs_devices */
7292 if (device->fs_devices != fs_devices) {
7293 ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
7294 &device->dev_state));
7295
7296 list_move(&device->dev_list, &fs_devices->devices);
7297 device->fs_devices->num_devices--;
7298 fs_devices->num_devices++;
7299
7300 device->fs_devices->missing_devices--;
7301 fs_devices->missing_devices++;
7302
7303 device->fs_devices = fs_devices;
7304 }
7305 }
7306
7307 if (device->fs_devices != fs_info->fs_devices) {
7308 BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
7309 if (device->generation !=
7310 btrfs_device_generation(leaf, dev_item))
7311 return -EINVAL;
7312 }
7313
7314 fill_device_from_item(leaf, dev_item, device);
7315 if (device->bdev) {
7316 u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7317
7318 if (device->total_bytes > max_total_bytes) {
7319 btrfs_err(fs_info,
7320 "device total_bytes should be at most %llu but found %llu",
7321 max_total_bytes, device->total_bytes);
7322 return -EINVAL;
7323 }
7324 }
7325 set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7326 if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7327 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7328 device->fs_devices->total_rw_bytes += device->total_bytes;
7329 atomic64_add(device->total_bytes - device->bytes_used,
7330 &fs_info->free_chunk_space);
7331 }
7332 ret = 0;
7333 return ret;
7334 }
7335
7336 int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7337 {
7338 struct btrfs_super_block *super_copy = fs_info->super_copy;
7339 struct extent_buffer *sb;
7340 struct btrfs_disk_key *disk_key;
7341 struct btrfs_chunk *chunk;
7342 u8 *array_ptr;
7343 unsigned long sb_array_offset;
7344 int ret = 0;
7345 u32 num_stripes;
7346 u32 array_size;
7347 u32 len = 0;
7348 u32 cur_offset;
7349 u64 type;
7350 struct btrfs_key key;
7351
7352 ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7353
7354 /*
7355 * We allocated a dummy extent, just to use extent buffer accessors.
7356 * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7357 * that's fine, we will not go beyond system chunk array anyway.
7358 */
7359 sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7360 if (!sb)
7361 return -ENOMEM;
7362 set_extent_buffer_uptodate(sb);
7363
7364 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7365 array_size = btrfs_super_sys_array_size(super_copy);
7366
7367 array_ptr = super_copy->sys_chunk_array;
7368 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7369 cur_offset = 0;
7370
7371 while (cur_offset < array_size) {
7372 disk_key = (struct btrfs_disk_key *)array_ptr;
7373 len = sizeof(*disk_key);
7374 if (cur_offset + len > array_size)
7375 goto out_short_read;
7376
7377 btrfs_disk_key_to_cpu(&key, disk_key);
7378
7379 array_ptr += len;
7380 sb_array_offset += len;
7381 cur_offset += len;
7382
7383 if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7384 btrfs_err(fs_info,
7385 "unexpected item type %u in sys_array at offset %u",
7386 (u32)key.type, cur_offset);
7387 ret = -EIO;
7388 break;
7389 }
7390
7391 chunk = (struct btrfs_chunk *)sb_array_offset;
7392 /*
7393 * At least one btrfs_chunk with one stripe must be present,
7394 * exact stripe count check comes afterwards
7395 */
7396 len = btrfs_chunk_item_size(1);
7397 if (cur_offset + len > array_size)
7398 goto out_short_read;
7399
7400 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7401 if (!num_stripes) {
7402 btrfs_err(fs_info,
7403 "invalid number of stripes %u in sys_array at offset %u",
7404 num_stripes, cur_offset);
7405 ret = -EIO;
7406 break;
7407 }
7408
7409 type = btrfs_chunk_type(sb, chunk);
7410 if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7411 btrfs_err(fs_info,
7412 "invalid chunk type %llu in sys_array at offset %u",
7413 type, cur_offset);
7414 ret = -EIO;
7415 break;
7416 }
7417
7418 len = btrfs_chunk_item_size(num_stripes);
7419 if (cur_offset + len > array_size)
7420 goto out_short_read;
7421
7422 ret = read_one_chunk(&key, sb, chunk);
7423 if (ret)
7424 break;
7425
7426 array_ptr += len;
7427 sb_array_offset += len;
7428 cur_offset += len;
7429 }
7430 clear_extent_buffer_uptodate(sb);
7431 free_extent_buffer_stale(sb);
7432 return ret;
7433
7434 out_short_read:
7435 btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7436 len, cur_offset);
7437 clear_extent_buffer_uptodate(sb);
7438 free_extent_buffer_stale(sb);
7439 return -EIO;
7440 }
7441
7442 /*
7443 * Check if all chunks in the fs are OK for read-write degraded mount
7444 *
7445 * If the @failing_dev is specified, it's accounted as missing.
7446 *
7447 * Return true if all chunks meet the minimal RW mount requirements.
7448 * Return false if any chunk doesn't meet the minimal RW mount requirements.
7449 */
7450 bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7451 struct btrfs_device *failing_dev)
7452 {
7453 struct btrfs_chunk_map *map;
7454 u64 next_start;
7455 bool ret = true;
7456
7457 map = btrfs_find_chunk_map(fs_info, 0, U64_MAX);
7458 /* No chunk at all? Return false anyway */
7459 if (!map) {
7460 ret = false;
7461 goto out;
7462 }
7463 while (map) {
7464 int missing = 0;
7465 int max_tolerated;
7466 int i;
7467
7468 max_tolerated =
7469 btrfs_get_num_tolerated_disk_barrier_failures(
7470 map->type);
7471 for (i = 0; i < map->num_stripes; i++) {
7472 struct btrfs_device *dev = map->stripes[i].dev;
7473
7474 if (!dev || !dev->bdev ||
7475 test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7476 dev->last_flush_error)
7477 missing++;
7478 else if (failing_dev && failing_dev == dev)
7479 missing++;
7480 }
7481 if (missing > max_tolerated) {
7482 if (!failing_dev)
7483 btrfs_warn(fs_info,
7484 "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7485 map->start, missing, max_tolerated);
7486 btrfs_free_chunk_map(map);
7487 ret = false;
7488 goto out;
7489 }
7490 next_start = map->start + map->chunk_len;
7491 btrfs_free_chunk_map(map);
7492
7493 map = btrfs_find_chunk_map(fs_info, next_start, U64_MAX - next_start);
7494 }
7495 out:
7496 return ret;
7497 }
7498
7499 static void readahead_tree_node_children(struct extent_buffer *node)
7500 {
7501 int i;
7502 const int nr_items = btrfs_header_nritems(node);
7503
7504 for (i = 0; i < nr_items; i++)
7505 btrfs_readahead_node_child(node, i);
7506 }
7507
7508 int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7509 {
7510 struct btrfs_root *root = fs_info->chunk_root;
7511 struct btrfs_path *path;
7512 struct extent_buffer *leaf;
7513 struct btrfs_key key;
7514 struct btrfs_key found_key;
7515 int ret;
7516 int slot;
7517 int iter_ret = 0;
7518 u64 total_dev = 0;
7519 u64 last_ra_node = 0;
7520
7521 path = btrfs_alloc_path();
7522 if (!path)
7523 return -ENOMEM;
7524
7525 /*
7526 * uuid_mutex is needed only if we are mounting a sprout FS
7527 * otherwise we don't need it.
7528 */
7529 mutex_lock(&uuid_mutex);
7530
7531 /*
7532 * It is possible for mount and umount to race in such a way that
7533 * we execute this code path, but open_fs_devices failed to clear
7534 * total_rw_bytes. We certainly want it cleared before reading the
7535 * device items, so clear it here.
7536 */
7537 fs_info->fs_devices->total_rw_bytes = 0;
7538
7539 /*
7540 * Lockdep complains about possible circular locking dependency between
7541 * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7542 * used for freeze procection of a fs (struct super_block.s_writers),
7543 * which we take when starting a transaction, and extent buffers of the
7544 * chunk tree if we call read_one_dev() while holding a lock on an
7545 * extent buffer of the chunk tree. Since we are mounting the filesystem
7546 * and at this point there can't be any concurrent task modifying the
7547 * chunk tree, to keep it simple, just skip locking on the chunk tree.
7548 */
7549 ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7550 path->skip_locking = 1;
7551
7552 /*
7553 * Read all device items, and then all the chunk items. All
7554 * device items are found before any chunk item (their object id
7555 * is smaller than the lowest possible object id for a chunk
7556 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7557 */
7558 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7559 key.offset = 0;
7560 key.type = 0;
7561 btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7562 struct extent_buffer *node = path->nodes[1];
7563
7564 leaf = path->nodes[0];
7565 slot = path->slots[0];
7566
7567 if (node) {
7568 if (last_ra_node != node->start) {
7569 readahead_tree_node_children(node);
7570 last_ra_node = node->start;
7571 }
7572 }
7573 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7574 struct btrfs_dev_item *dev_item;
7575 dev_item = btrfs_item_ptr(leaf, slot,
7576 struct btrfs_dev_item);
7577 ret = read_one_dev(leaf, dev_item);
7578 if (ret)
7579 goto error;
7580 total_dev++;
7581 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7582 struct btrfs_chunk *chunk;
7583
7584 /*
7585 * We are only called at mount time, so no need to take
7586 * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7587 * we always lock first fs_info->chunk_mutex before
7588 * acquiring any locks on the chunk tree. This is a
7589 * requirement for chunk allocation, see the comment on
7590 * top of btrfs_chunk_alloc() for details.
7591 */
7592 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7593 ret = read_one_chunk(&found_key, leaf, chunk);
7594 if (ret)
7595 goto error;
7596 }
7597 }
7598 /* Catch error found during iteration */
7599 if (iter_ret < 0) {
7600 ret = iter_ret;
7601 goto error;
7602 }
7603
7604 /*
7605 * After loading chunk tree, we've got all device information,
7606 * do another round of validation checks.
7607 */
7608 if (total_dev != fs_info->fs_devices->total_devices) {
7609 btrfs_warn(fs_info,
7610 "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7611 btrfs_super_num_devices(fs_info->super_copy),
7612 total_dev);
7613 fs_info->fs_devices->total_devices = total_dev;
7614 btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7615 }
7616 if (btrfs_super_total_bytes(fs_info->super_copy) <
7617 fs_info->fs_devices->total_rw_bytes) {
7618 btrfs_err(fs_info,
7619 "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7620 btrfs_super_total_bytes(fs_info->super_copy),
7621 fs_info->fs_devices->total_rw_bytes);
7622 ret = -EINVAL;
7623 goto error;
7624 }
7625 ret = 0;
7626 error:
7627 mutex_unlock(&uuid_mutex);
7628
7629 btrfs_free_path(path);
7630 return ret;
7631 }
7632
7633 int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7634 {
7635 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7636 struct btrfs_device *device;
7637 int ret = 0;
7638
7639 fs_devices->fs_info = fs_info;
7640
7641 mutex_lock(&fs_devices->device_list_mutex);
7642 list_for_each_entry(device, &fs_devices->devices, dev_list)
7643 device->fs_info = fs_info;
7644
7645 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7646 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7647 device->fs_info = fs_info;
7648 ret = btrfs_get_dev_zone_info(device, false);
7649 if (ret)
7650 break;
7651 }
7652
7653 seed_devs->fs_info = fs_info;
7654 }
7655 mutex_unlock(&fs_devices->device_list_mutex);
7656
7657 return ret;
7658 }
7659
7660 static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7661 const struct btrfs_dev_stats_item *ptr,
7662 int index)
7663 {
7664 u64 val;
7665
7666 read_extent_buffer(eb, &val,
7667 offsetof(struct btrfs_dev_stats_item, values) +
7668 ((unsigned long)ptr) + (index * sizeof(u64)),
7669 sizeof(val));
7670 return val;
7671 }
7672
7673 static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7674 struct btrfs_dev_stats_item *ptr,
7675 int index, u64 val)
7676 {
7677 write_extent_buffer(eb, &val,
7678 offsetof(struct btrfs_dev_stats_item, values) +
7679 ((unsigned long)ptr) + (index * sizeof(u64)),
7680 sizeof(val));
7681 }
7682
7683 static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7684 struct btrfs_path *path)
7685 {
7686 struct btrfs_dev_stats_item *ptr;
7687 struct extent_buffer *eb;
7688 struct btrfs_key key;
7689 int item_size;
7690 int i, ret, slot;
7691
7692 if (!device->fs_info->dev_root)
7693 return 0;
7694
7695 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7696 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7697 key.offset = device->devid;
7698 ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7699 if (ret) {
7700 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7701 btrfs_dev_stat_set(device, i, 0);
7702 device->dev_stats_valid = 1;
7703 btrfs_release_path(path);
7704 return ret < 0 ? ret : 0;
7705 }
7706 slot = path->slots[0];
7707 eb = path->nodes[0];
7708 item_size = btrfs_item_size(eb, slot);
7709
7710 ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7711
7712 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7713 if (item_size >= (1 + i) * sizeof(__le64))
7714 btrfs_dev_stat_set(device, i,
7715 btrfs_dev_stats_value(eb, ptr, i));
7716 else
7717 btrfs_dev_stat_set(device, i, 0);
7718 }
7719
7720 device->dev_stats_valid = 1;
7721 btrfs_dev_stat_print_on_load(device);
7722 btrfs_release_path(path);
7723
7724 return 0;
7725 }
7726
7727 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7728 {
7729 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7730 struct btrfs_device *device;
7731 struct btrfs_path *path = NULL;
7732 int ret = 0;
7733
7734 path = btrfs_alloc_path();
7735 if (!path)
7736 return -ENOMEM;
7737
7738 mutex_lock(&fs_devices->device_list_mutex);
7739 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7740 ret = btrfs_device_init_dev_stats(device, path);
7741 if (ret)
7742 goto out;
7743 }
7744 list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7745 list_for_each_entry(device, &seed_devs->devices, dev_list) {
7746 ret = btrfs_device_init_dev_stats(device, path);
7747 if (ret)
7748 goto out;
7749 }
7750 }
7751 out:
7752 mutex_unlock(&fs_devices->device_list_mutex);
7753
7754 btrfs_free_path(path);
7755 return ret;
7756 }
7757
7758 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7759 struct btrfs_device *device)
7760 {
7761 struct btrfs_fs_info *fs_info = trans->fs_info;
7762 struct btrfs_root *dev_root = fs_info->dev_root;
7763 struct btrfs_path *path;
7764 struct btrfs_key key;
7765 struct extent_buffer *eb;
7766 struct btrfs_dev_stats_item *ptr;
7767 int ret;
7768 int i;
7769
7770 key.objectid = BTRFS_DEV_STATS_OBJECTID;
7771 key.type = BTRFS_PERSISTENT_ITEM_KEY;
7772 key.offset = device->devid;
7773
7774 path = btrfs_alloc_path();
7775 if (!path)
7776 return -ENOMEM;
7777 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7778 if (ret < 0) {
7779 btrfs_warn_in_rcu(fs_info,
7780 "error %d while searching for dev_stats item for device %s",
7781 ret, btrfs_dev_name(device));
7782 goto out;
7783 }
7784
7785 if (ret == 0 &&
7786 btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7787 /* need to delete old one and insert a new one */
7788 ret = btrfs_del_item(trans, dev_root, path);
7789 if (ret != 0) {
7790 btrfs_warn_in_rcu(fs_info,
7791 "delete too small dev_stats item for device %s failed %d",
7792 btrfs_dev_name(device), ret);
7793 goto out;
7794 }
7795 ret = 1;
7796 }
7797
7798 if (ret == 1) {
7799 /* need to insert a new item */
7800 btrfs_release_path(path);
7801 ret = btrfs_insert_empty_item(trans, dev_root, path,
7802 &key, sizeof(*ptr));
7803 if (ret < 0) {
7804 btrfs_warn_in_rcu(fs_info,
7805 "insert dev_stats item for device %s failed %d",
7806 btrfs_dev_name(device), ret);
7807 goto out;
7808 }
7809 }
7810
7811 eb = path->nodes[0];
7812 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7813 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7814 btrfs_set_dev_stats_value(eb, ptr, i,
7815 btrfs_dev_stat_read(device, i));
7816 btrfs_mark_buffer_dirty(trans, eb);
7817
7818 out:
7819 btrfs_free_path(path);
7820 return ret;
7821 }
7822
7823 /*
7824 * called from commit_transaction. Writes all changed device stats to disk.
7825 */
7826 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7827 {
7828 struct btrfs_fs_info *fs_info = trans->fs_info;
7829 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7830 struct btrfs_device *device;
7831 int stats_cnt;
7832 int ret = 0;
7833
7834 mutex_lock(&fs_devices->device_list_mutex);
7835 list_for_each_entry(device, &fs_devices->devices, dev_list) {
7836 stats_cnt = atomic_read(&device->dev_stats_ccnt);
7837 if (!device->dev_stats_valid || stats_cnt == 0)
7838 continue;
7839
7840
7841 /*
7842 * There is a LOAD-LOAD control dependency between the value of
7843 * dev_stats_ccnt and updating the on-disk values which requires
7844 * reading the in-memory counters. Such control dependencies
7845 * require explicit read memory barriers.
7846 *
7847 * This memory barriers pairs with smp_mb__before_atomic in
7848 * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7849 * barrier implied by atomic_xchg in
7850 * btrfs_dev_stats_read_and_reset
7851 */
7852 smp_rmb();
7853
7854 ret = update_dev_stat_item(trans, device);
7855 if (!ret)
7856 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7857 }
7858 mutex_unlock(&fs_devices->device_list_mutex);
7859
7860 return ret;
7861 }
7862
7863 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7864 {
7865 btrfs_dev_stat_inc(dev, index);
7866
7867 if (!dev->dev_stats_valid)
7868 return;
7869 btrfs_err_rl_in_rcu(dev->fs_info,
7870 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7871 btrfs_dev_name(dev),
7872 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7873 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7874 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7875 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7876 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7877 }
7878
7879 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7880 {
7881 int i;
7882
7883 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7884 if (btrfs_dev_stat_read(dev, i) != 0)
7885 break;
7886 if (i == BTRFS_DEV_STAT_VALUES_MAX)
7887 return; /* all values == 0, suppress message */
7888
7889 btrfs_info_in_rcu(dev->fs_info,
7890 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7891 btrfs_dev_name(dev),
7892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7893 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7894 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7895 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7896 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7897 }
7898
7899 int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7900 struct btrfs_ioctl_get_dev_stats *stats)
7901 {
7902 BTRFS_DEV_LOOKUP_ARGS(args);
7903 struct btrfs_device *dev;
7904 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7905 int i;
7906
7907 mutex_lock(&fs_devices->device_list_mutex);
7908 args.devid = stats->devid;
7909 dev = btrfs_find_device(fs_info->fs_devices, &args);
7910 mutex_unlock(&fs_devices->device_list_mutex);
7911
7912 if (!dev) {
7913 btrfs_warn(fs_info, "get dev_stats failed, device not found");
7914 return -ENODEV;
7915 } else if (!dev->dev_stats_valid) {
7916 btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7917 return -ENODEV;
7918 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7919 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7920 if (stats->nr_items > i)
7921 stats->values[i] =
7922 btrfs_dev_stat_read_and_reset(dev, i);
7923 else
7924 btrfs_dev_stat_set(dev, i, 0);
7925 }
7926 btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7927 current->comm, task_pid_nr(current));
7928 } else {
7929 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7930 if (stats->nr_items > i)
7931 stats->values[i] = btrfs_dev_stat_read(dev, i);
7932 }
7933 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7934 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7935 return 0;
7936 }
7937
7938 /*
7939 * Update the size and bytes used for each device where it changed. This is
7940 * delayed since we would otherwise get errors while writing out the
7941 * superblocks.
7942 *
7943 * Must be invoked during transaction commit.
7944 */
7945 void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7946 {
7947 struct btrfs_device *curr, *next;
7948
7949 ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7950
7951 if (list_empty(&trans->dev_update_list))
7952 return;
7953
7954 /*
7955 * We don't need the device_list_mutex here. This list is owned by the
7956 * transaction and the transaction must complete before the device is
7957 * released.
7958 */
7959 mutex_lock(&trans->fs_info->chunk_mutex);
7960 list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7961 post_commit_list) {
7962 list_del_init(&curr->post_commit_list);
7963 curr->commit_total_bytes = curr->disk_total_bytes;
7964 curr->commit_bytes_used = curr->bytes_used;
7965 }
7966 mutex_unlock(&trans->fs_info->chunk_mutex);
7967 }
7968
7969 /*
7970 * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7971 */
7972 int btrfs_bg_type_to_factor(u64 flags)
7973 {
7974 const int index = btrfs_bg_flags_to_raid_index(flags);
7975
7976 return btrfs_raid_array[index].ncopies;
7977 }
7978
7979
7980
7981 static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7982 u64 chunk_offset, u64 devid,
7983 u64 physical_offset, u64 physical_len)
7984 {
7985 struct btrfs_dev_lookup_args args = { .devid = devid };
7986 struct btrfs_chunk_map *map;
7987 struct btrfs_device *dev;
7988 u64 stripe_len;
7989 bool found = false;
7990 int ret = 0;
7991 int i;
7992
7993 map = btrfs_find_chunk_map(fs_info, chunk_offset, 1);
7994 if (!map) {
7995 btrfs_err(fs_info,
7996 "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7997 physical_offset, devid);
7998 ret = -EUCLEAN;
7999 goto out;
8000 }
8001
8002 stripe_len = btrfs_calc_stripe_length(map);
8003 if (physical_len != stripe_len) {
8004 btrfs_err(fs_info,
8005 "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
8006 physical_offset, devid, map->start, physical_len,
8007 stripe_len);
8008 ret = -EUCLEAN;
8009 goto out;
8010 }
8011
8012 /*
8013 * Very old mkfs.btrfs (before v4.1) will not respect the reserved
8014 * space. Although kernel can handle it without problem, better to warn
8015 * the users.
8016 */
8017 if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
8018 btrfs_warn(fs_info,
8019 "devid %llu physical %llu len %llu inside the reserved space",
8020 devid, physical_offset, physical_len);
8021
8022 for (i = 0; i < map->num_stripes; i++) {
8023 if (map->stripes[i].dev->devid == devid &&
8024 map->stripes[i].physical == physical_offset) {
8025 found = true;
8026 if (map->verified_stripes >= map->num_stripes) {
8027 btrfs_err(fs_info,
8028 "too many dev extents for chunk %llu found",
8029 map->start);
8030 ret = -EUCLEAN;
8031 goto out;
8032 }
8033 map->verified_stripes++;
8034 break;
8035 }
8036 }
8037 if (!found) {
8038 btrfs_err(fs_info,
8039 "dev extent physical offset %llu devid %llu has no corresponding chunk",
8040 physical_offset, devid);
8041 ret = -EUCLEAN;
8042 }
8043
8044 /* Make sure no dev extent is beyond device boundary */
8045 dev = btrfs_find_device(fs_info->fs_devices, &args);
8046 if (!dev) {
8047 btrfs_err(fs_info, "failed to find devid %llu", devid);
8048 ret = -EUCLEAN;
8049 goto out;
8050 }
8051
8052 if (physical_offset + physical_len > dev->disk_total_bytes) {
8053 btrfs_err(fs_info,
8054 "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
8055 devid, physical_offset, physical_len,
8056 dev->disk_total_bytes);
8057 ret = -EUCLEAN;
8058 goto out;
8059 }
8060
8061 if (dev->zone_info) {
8062 u64 zone_size = dev->zone_info->zone_size;
8063
8064 if (!IS_ALIGNED(physical_offset, zone_size) ||
8065 !IS_ALIGNED(physical_len, zone_size)) {
8066 btrfs_err(fs_info,
8067 "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
8068 devid, physical_offset, physical_len);
8069 ret = -EUCLEAN;
8070 goto out;
8071 }
8072 }
8073
8074 out:
8075 btrfs_free_chunk_map(map);
8076 return ret;
8077 }
8078
8079 static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
8080 {
8081 struct rb_node *node;
8082 int ret = 0;
8083
8084 read_lock(&fs_info->mapping_tree_lock);
8085 for (node = rb_first_cached(&fs_info->mapping_tree); node; node = rb_next(node)) {
8086 struct btrfs_chunk_map *map;
8087
8088 map = rb_entry(node, struct btrfs_chunk_map, rb_node);
8089 if (map->num_stripes != map->verified_stripes) {
8090 btrfs_err(fs_info,
8091 "chunk %llu has missing dev extent, have %d expect %d",
8092 map->start, map->verified_stripes, map->num_stripes);
8093 ret = -EUCLEAN;
8094 goto out;
8095 }
8096 }
8097 out:
8098 read_unlock(&fs_info->mapping_tree_lock);
8099 return ret;
8100 }
8101
8102 /*
8103 * Ensure that all dev extents are mapped to correct chunk, otherwise
8104 * later chunk allocation/free would cause unexpected behavior.
8105 *
8106 * NOTE: This will iterate through the whole device tree, which should be of
8107 * the same size level as the chunk tree. This slightly increases mount time.
8108 */
8109 int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
8110 {
8111 struct btrfs_path *path;
8112 struct btrfs_root *root = fs_info->dev_root;
8113 struct btrfs_key key;
8114 u64 prev_devid = 0;
8115 u64 prev_dev_ext_end = 0;
8116 int ret = 0;
8117
8118 /*
8119 * We don't have a dev_root because we mounted with ignorebadroots and
8120 * failed to load the root, so we want to skip the verification in this
8121 * case for sure.
8122 *
8123 * However if the dev root is fine, but the tree itself is corrupted
8124 * we'd still fail to mount. This verification is only to make sure
8125 * writes can happen safely, so instead just bypass this check
8126 * completely in the case of IGNOREBADROOTS.
8127 */
8128 if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
8129 return 0;
8130
8131 key.objectid = 1;
8132 key.type = BTRFS_DEV_EXTENT_KEY;
8133 key.offset = 0;
8134
8135 path = btrfs_alloc_path();
8136 if (!path)
8137 return -ENOMEM;
8138
8139 path->reada = READA_FORWARD;
8140 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8141 if (ret < 0)
8142 goto out;
8143
8144 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
8145 ret = btrfs_next_leaf(root, path);
8146 if (ret < 0)
8147 goto out;
8148 /* No dev extents at all? Not good */
8149 if (ret > 0) {
8150 ret = -EUCLEAN;
8151 goto out;
8152 }
8153 }
8154 while (1) {
8155 struct extent_buffer *leaf = path->nodes[0];
8156 struct btrfs_dev_extent *dext;
8157 int slot = path->slots[0];
8158 u64 chunk_offset;
8159 u64 physical_offset;
8160 u64 physical_len;
8161 u64 devid;
8162
8163 btrfs_item_key_to_cpu(leaf, &key, slot);
8164 if (key.type != BTRFS_DEV_EXTENT_KEY)
8165 break;
8166 devid = key.objectid;
8167 physical_offset = key.offset;
8168
8169 dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
8170 chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
8171 physical_len = btrfs_dev_extent_length(leaf, dext);
8172
8173 /* Check if this dev extent overlaps with the previous one */
8174 if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
8175 btrfs_err(fs_info,
8176 "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
8177 devid, physical_offset, prev_dev_ext_end);
8178 ret = -EUCLEAN;
8179 goto out;
8180 }
8181
8182 ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
8183 physical_offset, physical_len);
8184 if (ret < 0)
8185 goto out;
8186 prev_devid = devid;
8187 prev_dev_ext_end = physical_offset + physical_len;
8188
8189 ret = btrfs_next_item(root, path);
8190 if (ret < 0)
8191 goto out;
8192 if (ret > 0) {
8193 ret = 0;
8194 break;
8195 }
8196 }
8197
8198 /* Ensure all chunks have corresponding dev extents */
8199 ret = verify_chunk_dev_extent_mapping(fs_info);
8200 out:
8201 btrfs_free_path(path);
8202 return ret;
8203 }
8204
8205 /*
8206 * Check whether the given block group or device is pinned by any inode being
8207 * used as a swapfile.
8208 */
8209 bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
8210 {
8211 struct btrfs_swapfile_pin *sp;
8212 struct rb_node *node;
8213
8214 spin_lock(&fs_info->swapfile_pins_lock);
8215 node = fs_info->swapfile_pins.rb_node;
8216 while (node) {
8217 sp = rb_entry(node, struct btrfs_swapfile_pin, node);
8218 if (ptr < sp->ptr)
8219 node = node->rb_left;
8220 else if (ptr > sp->ptr)
8221 node = node->rb_right;
8222 else
8223 break;
8224 }
8225 spin_unlock(&fs_info->swapfile_pins_lock);
8226 return node != NULL;
8227 }
8228
8229 static int relocating_repair_kthread(void *data)
8230 {
8231 struct btrfs_block_group *cache = data;
8232 struct btrfs_fs_info *fs_info = cache->fs_info;
8233 u64 target;
8234 int ret = 0;
8235
8236 target = cache->start;
8237 btrfs_put_block_group(cache);
8238
8239 sb_start_write(fs_info->sb);
8240 if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
8241 btrfs_info(fs_info,
8242 "zoned: skip relocating block group %llu to repair: EBUSY",
8243 target);
8244 sb_end_write(fs_info->sb);
8245 return -EBUSY;
8246 }
8247
8248 mutex_lock(&fs_info->reclaim_bgs_lock);
8249
8250 /* Ensure block group still exists */
8251 cache = btrfs_lookup_block_group(fs_info, target);
8252 if (!cache)
8253 goto out;
8254
8255 if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
8256 goto out;
8257
8258 ret = btrfs_may_alloc_data_chunk(fs_info, target);
8259 if (ret < 0)
8260 goto out;
8261
8262 btrfs_info(fs_info,
8263 "zoned: relocating block group %llu to repair IO failure",
8264 target);
8265 ret = btrfs_relocate_chunk(fs_info, target);
8266
8267 out:
8268 if (cache)
8269 btrfs_put_block_group(cache);
8270 mutex_unlock(&fs_info->reclaim_bgs_lock);
8271 btrfs_exclop_finish(fs_info);
8272 sb_end_write(fs_info->sb);
8273
8274 return ret;
8275 }
8276
8277 bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
8278 {
8279 struct btrfs_block_group *cache;
8280
8281 if (!btrfs_is_zoned(fs_info))
8282 return false;
8283
8284 /* Do not attempt to repair in degraded state */
8285 if (btrfs_test_opt(fs_info, DEGRADED))
8286 return true;
8287
8288 cache = btrfs_lookup_block_group(fs_info, logical);
8289 if (!cache)
8290 return true;
8291
8292 if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
8293 btrfs_put_block_group(cache);
8294 return true;
8295 }
8296
8297 kthread_run(relocating_repair_kthread, cache,
8298 "btrfs-relocating-repair");
8299
8300 return true;
8301 }
8302
8303 static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8304 struct btrfs_io_stripe *smap,
8305 u64 logical)
8306 {
8307 int data_stripes = nr_bioc_data_stripes(bioc);
8308 int i;
8309
8310 for (i = 0; i < data_stripes; i++) {
8311 u64 stripe_start = bioc->full_stripe_logical +
8312 btrfs_stripe_nr_to_offset(i);
8313
8314 if (logical >= stripe_start &&
8315 logical < stripe_start + BTRFS_STRIPE_LEN)
8316 break;
8317 }
8318 ASSERT(i < data_stripes);
8319 smap->dev = bioc->stripes[i].dev;
8320 smap->physical = bioc->stripes[i].physical +
8321 ((logical - bioc->full_stripe_logical) &
8322 BTRFS_STRIPE_LEN_MASK);
8323 }
8324
8325 /*
8326 * Map a repair write into a single device.
8327 *
8328 * A repair write is triggered by read time repair or scrub, which would only
8329 * update the contents of a single device.
8330 * Not update any other mirrors nor go through RMW path.
8331 *
8332 * Callers should ensure:
8333 *
8334 * - Call btrfs_bio_counter_inc_blocked() first
8335 * - The range does not cross stripe boundary
8336 * - Has a valid @mirror_num passed in.
8337 */
8338 int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8339 struct btrfs_io_stripe *smap, u64 logical,
8340 u32 length, int mirror_num)
8341 {
8342 struct btrfs_io_context *bioc = NULL;
8343 u64 map_length = length;
8344 int mirror_ret = mirror_num;
8345 int ret;
8346
8347 ASSERT(mirror_num > 0);
8348
8349 ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8350 &bioc, smap, &mirror_ret);
8351 if (ret < 0)
8352 return ret;
8353
8354 /* The map range should not cross stripe boundary. */
8355 ASSERT(map_length >= length);
8356
8357 /* Already mapped to single stripe. */
8358 if (!bioc)
8359 goto out;
8360
8361 /* Map the RAID56 multi-stripe writes to a single one. */
8362 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8363 map_raid56_repair_block(bioc, smap, logical);
8364 goto out;
8365 }
8366
8367 ASSERT(mirror_num <= bioc->num_stripes);
8368 smap->dev = bioc->stripes[mirror_num - 1].dev;
8369 smap->physical = bioc->stripes[mirror_num - 1].physical;
8370 out:
8371 btrfs_put_bioc(bioc);
8372 ASSERT(smap->dev);
8373 return 0;
8374 }
8375
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.