Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Copyright (C) 2007 Oracle. All rights reserved.
4 : */
5 :
6 : #include <linux/sched.h>
7 : #include <linux/sched/mm.h>
8 : #include <linux/slab.h>
9 : #include <linux/ratelimit.h>
10 : #include <linux/kthread.h>
11 : #include <linux/semaphore.h>
12 : #include <linux/uuid.h>
13 : #include <linux/list_sort.h>
14 : #include <linux/namei.h>
15 : #include "misc.h"
16 : #include "ctree.h"
17 : #include "extent_map.h"
18 : #include "disk-io.h"
19 : #include "transaction.h"
20 : #include "print-tree.h"
21 : #include "volumes.h"
22 : #include "raid56.h"
23 : #include "rcu-string.h"
24 : #include "dev-replace.h"
25 : #include "sysfs.h"
26 : #include "tree-checker.h"
27 : #include "space-info.h"
28 : #include "block-group.h"
29 : #include "discard.h"
30 : #include "zoned.h"
31 : #include "fs.h"
32 : #include "accessors.h"
33 : #include "uuid-tree.h"
34 : #include "ioctl.h"
35 : #include "relocation.h"
36 : #include "scrub.h"
37 : #include "super.h"
38 :
39 : #define BTRFS_BLOCK_GROUP_STRIPE_MASK (BTRFS_BLOCK_GROUP_RAID0 | \
40 : BTRFS_BLOCK_GROUP_RAID10 | \
41 : BTRFS_BLOCK_GROUP_RAID56_MASK)
42 :
43 : const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
44 : [BTRFS_RAID_RAID10] = {
45 : .sub_stripes = 2,
46 : .dev_stripes = 1,
47 : .devs_max = 0, /* 0 == as many as possible */
48 : .devs_min = 2,
49 : .tolerated_failures = 1,
50 : .devs_increment = 2,
51 : .ncopies = 2,
52 : .nparity = 0,
53 : .raid_name = "raid10",
54 : .bg_flag = BTRFS_BLOCK_GROUP_RAID10,
55 : .mindev_error = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET,
56 : },
57 : [BTRFS_RAID_RAID1] = {
58 : .sub_stripes = 1,
59 : .dev_stripes = 1,
60 : .devs_max = 2,
61 : .devs_min = 2,
62 : .tolerated_failures = 1,
63 : .devs_increment = 2,
64 : .ncopies = 2,
65 : .nparity = 0,
66 : .raid_name = "raid1",
67 : .bg_flag = BTRFS_BLOCK_GROUP_RAID1,
68 : .mindev_error = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET,
69 : },
70 : [BTRFS_RAID_RAID1C3] = {
71 : .sub_stripes = 1,
72 : .dev_stripes = 1,
73 : .devs_max = 3,
74 : .devs_min = 3,
75 : .tolerated_failures = 2,
76 : .devs_increment = 3,
77 : .ncopies = 3,
78 : .nparity = 0,
79 : .raid_name = "raid1c3",
80 : .bg_flag = BTRFS_BLOCK_GROUP_RAID1C3,
81 : .mindev_error = BTRFS_ERROR_DEV_RAID1C3_MIN_NOT_MET,
82 : },
83 : [BTRFS_RAID_RAID1C4] = {
84 : .sub_stripes = 1,
85 : .dev_stripes = 1,
86 : .devs_max = 4,
87 : .devs_min = 4,
88 : .tolerated_failures = 3,
89 : .devs_increment = 4,
90 : .ncopies = 4,
91 : .nparity = 0,
92 : .raid_name = "raid1c4",
93 : .bg_flag = BTRFS_BLOCK_GROUP_RAID1C4,
94 : .mindev_error = BTRFS_ERROR_DEV_RAID1C4_MIN_NOT_MET,
95 : },
96 : [BTRFS_RAID_DUP] = {
97 : .sub_stripes = 1,
98 : .dev_stripes = 2,
99 : .devs_max = 1,
100 : .devs_min = 1,
101 : .tolerated_failures = 0,
102 : .devs_increment = 1,
103 : .ncopies = 2,
104 : .nparity = 0,
105 : .raid_name = "dup",
106 : .bg_flag = BTRFS_BLOCK_GROUP_DUP,
107 : .mindev_error = 0,
108 : },
109 : [BTRFS_RAID_RAID0] = {
110 : .sub_stripes = 1,
111 : .dev_stripes = 1,
112 : .devs_max = 0,
113 : .devs_min = 1,
114 : .tolerated_failures = 0,
115 : .devs_increment = 1,
116 : .ncopies = 1,
117 : .nparity = 0,
118 : .raid_name = "raid0",
119 : .bg_flag = BTRFS_BLOCK_GROUP_RAID0,
120 : .mindev_error = 0,
121 : },
122 : [BTRFS_RAID_SINGLE] = {
123 : .sub_stripes = 1,
124 : .dev_stripes = 1,
125 : .devs_max = 1,
126 : .devs_min = 1,
127 : .tolerated_failures = 0,
128 : .devs_increment = 1,
129 : .ncopies = 1,
130 : .nparity = 0,
131 : .raid_name = "single",
132 : .bg_flag = 0,
133 : .mindev_error = 0,
134 : },
135 : [BTRFS_RAID_RAID5] = {
136 : .sub_stripes = 1,
137 : .dev_stripes = 1,
138 : .devs_max = 0,
139 : .devs_min = 2,
140 : .tolerated_failures = 1,
141 : .devs_increment = 1,
142 : .ncopies = 1,
143 : .nparity = 1,
144 : .raid_name = "raid5",
145 : .bg_flag = BTRFS_BLOCK_GROUP_RAID5,
146 : .mindev_error = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET,
147 : },
148 : [BTRFS_RAID_RAID6] = {
149 : .sub_stripes = 1,
150 : .dev_stripes = 1,
151 : .devs_max = 0,
152 : .devs_min = 3,
153 : .tolerated_failures = 2,
154 : .devs_increment = 1,
155 : .ncopies = 1,
156 : .nparity = 2,
157 : .raid_name = "raid6",
158 : .bg_flag = BTRFS_BLOCK_GROUP_RAID6,
159 : .mindev_error = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET,
160 : },
161 : };
162 :
163 : /*
164 : * Convert block group flags (BTRFS_BLOCK_GROUP_*) to btrfs_raid_types, which
165 : * can be used as index to access btrfs_raid_array[].
166 : */
167 173218045 : enum btrfs_raid_types __attribute_const__ btrfs_bg_flags_to_raid_index(u64 flags)
168 : {
169 173218045 : const u64 profile = (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK);
170 :
171 173218045 : if (!profile)
172 : return BTRFS_RAID_SINGLE;
173 :
174 130419570 : return BTRFS_BG_FLAG_TO_INDEX(profile);
175 : }
176 :
177 9621 : const char *btrfs_bg_type_to_raid_name(u64 flags)
178 : {
179 9621 : const int index = btrfs_bg_flags_to_raid_index(flags);
180 :
181 9621 : if (index >= BTRFS_NR_RAID_TYPES)
182 : return NULL;
183 :
184 9621 : return btrfs_raid_array[index].raid_name;
185 : }
186 :
187 0 : int btrfs_nr_parity_stripes(u64 type)
188 : {
189 15469311 : enum btrfs_raid_types index = btrfs_bg_flags_to_raid_index(type);
190 :
191 15469311 : return btrfs_raid_array[index].nparity;
192 : }
193 :
194 : /*
195 : * Fill @buf with textual description of @bg_flags, no more than @size_buf
196 : * bytes including terminating null byte.
197 : */
198 523 : void btrfs_describe_block_groups(u64 bg_flags, char *buf, u32 size_buf)
199 : {
200 523 : int i;
201 523 : int ret;
202 523 : char *bp = buf;
203 523 : u64 flags = bg_flags;
204 523 : u32 size_bp = size_buf;
205 :
206 523 : if (!flags) {
207 0 : strcpy(bp, "NONE");
208 0 : return;
209 : }
210 :
211 : #define DESCRIBE_FLAG(flag, desc) \
212 : do { \
213 : if (flags & (flag)) { \
214 : ret = snprintf(bp, size_bp, "%s|", (desc)); \
215 : if (ret < 0 || ret >= size_bp) \
216 : goto out_overflow; \
217 : size_bp -= ret; \
218 : bp += ret; \
219 : flags &= ~(flag); \
220 : } \
221 : } while (0)
222 :
223 523 : DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_DATA, "data");
224 523 : DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_SYSTEM, "system");
225 523 : DESCRIBE_FLAG(BTRFS_BLOCK_GROUP_METADATA, "metadata");
226 :
227 523 : DESCRIBE_FLAG(BTRFS_AVAIL_ALLOC_BIT_SINGLE, "single");
228 5230 : for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
229 4707 : DESCRIBE_FLAG(btrfs_raid_array[i].bg_flag,
230 : btrfs_raid_array[i].raid_name);
231 : #undef DESCRIBE_FLAG
232 :
233 523 : if (flags) {
234 0 : ret = snprintf(bp, size_bp, "0x%llx|", flags);
235 0 : size_bp -= ret;
236 : }
237 :
238 523 : if (size_bp < size_buf)
239 523 : buf[size_buf - size_bp - 1] = '\0'; /* remove last | */
240 :
241 : /*
242 : * The text is trimmed, it's up to the caller to provide sufficiently
243 : * large buffer
244 : */
245 0 : out_overflow:;
246 : }
247 :
248 : static int init_first_rw_device(struct btrfs_trans_handle *trans);
249 : static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info);
250 : static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
251 :
252 : /*
253 : * Device locking
254 : * ==============
255 : *
256 : * There are several mutexes that protect manipulation of devices and low-level
257 : * structures like chunks but not block groups, extents or files
258 : *
259 : * uuid_mutex (global lock)
260 : * ------------------------
261 : * protects the fs_uuids list that tracks all per-fs fs_devices, resulting from
262 : * the SCAN_DEV ioctl registration or from mount either implicitly (the first
263 : * device) or requested by the device= mount option
264 : *
265 : * the mutex can be very coarse and can cover long-running operations
266 : *
267 : * protects: updates to fs_devices counters like missing devices, rw devices,
268 : * seeding, structure cloning, opening/closing devices at mount/umount time
269 : *
270 : * global::fs_devs - add, remove, updates to the global list
271 : *
272 : * does not protect: manipulation of the fs_devices::devices list in general
273 : * but in mount context it could be used to exclude list modifications by eg.
274 : * scan ioctl
275 : *
276 : * btrfs_device::name - renames (write side), read is RCU
277 : *
278 : * fs_devices::device_list_mutex (per-fs, with RCU)
279 : * ------------------------------------------------
280 : * protects updates to fs_devices::devices, ie. adding and deleting
281 : *
282 : * simple list traversal with read-only actions can be done with RCU protection
283 : *
284 : * may be used to exclude some operations from running concurrently without any
285 : * modifications to the list (see write_all_supers)
286 : *
287 : * Is not required at mount and close times, because our device list is
288 : * protected by the uuid_mutex at that point.
289 : *
290 : * balance_mutex
291 : * -------------
292 : * protects balance structures (status, state) and context accessed from
293 : * several places (internally, ioctl)
294 : *
295 : * chunk_mutex
296 : * -----------
297 : * protects chunks, adding or removing during allocation, trim or when a new
298 : * device is added/removed. Additionally it also protects post_commit_list of
299 : * individual devices, since they can be added to the transaction's
300 : * post_commit_list only with chunk_mutex held.
301 : *
302 : * cleaner_mutex
303 : * -------------
304 : * a big lock that is held by the cleaner thread and prevents running subvolume
305 : * cleaning together with relocation or delayed iputs
306 : *
307 : *
308 : * Lock nesting
309 : * ============
310 : *
311 : * uuid_mutex
312 : * device_list_mutex
313 : * chunk_mutex
314 : * balance_mutex
315 : *
316 : *
317 : * Exclusive operations
318 : * ====================
319 : *
320 : * Maintains the exclusivity of the following operations that apply to the
321 : * whole filesystem and cannot run in parallel.
322 : *
323 : * - Balance (*)
324 : * - Device add
325 : * - Device remove
326 : * - Device replace (*)
327 : * - Resize
328 : *
329 : * The device operations (as above) can be in one of the following states:
330 : *
331 : * - Running state
332 : * - Paused state
333 : * - Completed state
334 : *
335 : * Only device operations marked with (*) can go into the Paused state for the
336 : * following reasons:
337 : *
338 : * - ioctl (only Balance can be Paused through ioctl)
339 : * - filesystem remounted as read-only
340 : * - filesystem unmounted and mounted as read-only
341 : * - system power-cycle and filesystem mounted as read-only
342 : * - filesystem or device errors leading to forced read-only
343 : *
344 : * The status of exclusive operation is set and cleared atomically.
345 : * During the course of Paused state, fs_info::exclusive_operation remains set.
346 : * A device operation in Paused or Running state can be canceled or resumed
347 : * either by ioctl (Balance only) or when remounted as read-write.
348 : * The exclusive status is cleared when the device operation is canceled or
349 : * completed.
350 : */
351 :
352 : DEFINE_MUTEX(uuid_mutex);
353 : static LIST_HEAD(fs_uuids);
354 0 : struct list_head * __attribute_const__ btrfs_get_fs_uuids(void)
355 : {
356 0 : return &fs_uuids;
357 : }
358 :
359 : /*
360 : * alloc_fs_devices - allocate struct btrfs_fs_devices
361 : * @fsid: if not NULL, copy the UUID to fs_devices::fsid
362 : * @metadata_fsid: if not NULL, copy the UUID to fs_devices::metadata_fsid
363 : *
364 : * Return a pointer to a new struct btrfs_fs_devices on success, or ERR_PTR().
365 : * The returned struct is not linked onto any lists and can be destroyed with
366 : * kfree() right away.
367 : */
368 3267 : static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid,
369 : const u8 *metadata_fsid)
370 : {
371 3267 : struct btrfs_fs_devices *fs_devs;
372 :
373 3267 : ASSERT(fsid || !metadata_fsid);
374 :
375 3267 : fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL);
376 3267 : if (!fs_devs)
377 : return ERR_PTR(-ENOMEM);
378 :
379 3267 : mutex_init(&fs_devs->device_list_mutex);
380 :
381 3267 : INIT_LIST_HEAD(&fs_devs->devices);
382 3267 : INIT_LIST_HEAD(&fs_devs->alloc_list);
383 3267 : INIT_LIST_HEAD(&fs_devs->fs_list);
384 3267 : INIT_LIST_HEAD(&fs_devs->seed_list);
385 :
386 3267 : if (fsid) {
387 6534 : memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
388 6534 : memcpy(fs_devs->metadata_uuid,
389 : metadata_fsid ?: fsid, BTRFS_FSID_SIZE);
390 : }
391 :
392 : return fs_devs;
393 : }
394 :
395 3256 : static void btrfs_free_device(struct btrfs_device *device)
396 : {
397 3256 : WARN_ON(!list_empty(&device->post_commit_list));
398 3256 : rcu_string_free(device->name);
399 3256 : extent_io_tree_release(&device->alloc_state);
400 3256 : btrfs_destroy_dev_zone_info(device);
401 3256 : kfree(device);
402 3256 : }
403 :
404 3255 : static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
405 : {
406 3255 : struct btrfs_device *device;
407 :
408 3255 : WARN_ON(fs_devices->opened);
409 6497 : while (!list_empty(&fs_devices->devices)) {
410 3242 : device = list_entry(fs_devices->devices.next,
411 : struct btrfs_device, dev_list);
412 3242 : list_del(&device->dev_list);
413 3242 : btrfs_free_device(device);
414 : }
415 3255 : kfree(fs_devices);
416 3255 : }
417 :
418 0 : void __exit btrfs_cleanup_fs_uuids(void)
419 : {
420 0 : struct btrfs_fs_devices *fs_devices;
421 :
422 0 : while (!list_empty(&fs_uuids)) {
423 0 : fs_devices = list_entry(fs_uuids.next,
424 : struct btrfs_fs_devices, fs_list);
425 0 : list_del(&fs_devices->fs_list);
426 0 : free_fs_devices(fs_devices);
427 : }
428 0 : }
429 :
430 6825 : static bool match_fsid_fs_devices(const struct btrfs_fs_devices *fs_devices,
431 : const u8 *fsid, const u8 *metadata_fsid)
432 : {
433 13650 : if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) != 0)
434 : return false;
435 :
436 3453 : if (!metadata_fsid)
437 : return true;
438 :
439 0 : if (memcmp(metadata_fsid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE) != 0)
440 0 : return false;
441 :
442 : return true;
443 : }
444 :
445 6720 : static noinline struct btrfs_fs_devices *find_fsid(
446 : const u8 *fsid, const u8 *metadata_fsid)
447 : {
448 6720 : struct btrfs_fs_devices *fs_devices;
449 :
450 6720 : ASSERT(fsid);
451 :
452 : /* Handle non-split brain cases */
453 10092 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
454 6825 : if (match_fsid_fs_devices(fs_devices, fsid, metadata_fsid))
455 3453 : return fs_devices;
456 : }
457 : return NULL;
458 : }
459 :
460 : /*
461 : * First check if the metadata_uuid is different from the fsid in the given
462 : * fs_devices. Then check if the given fsid is the same as the metadata_uuid
463 : * in the fs_devices. If it is, return true; otherwise, return false.
464 : */
465 0 : static inline bool check_fsid_changed(const struct btrfs_fs_devices *fs_devices,
466 : const u8 *fsid)
467 : {
468 0 : return memcmp(fs_devices->fsid, fs_devices->metadata_uuid,
469 0 : BTRFS_FSID_SIZE) != 0 &&
470 : memcmp(fs_devices->metadata_uuid, fsid, BTRFS_FSID_SIZE) == 0;
471 : }
472 :
473 0 : static struct btrfs_fs_devices *find_fsid_with_metadata_uuid(
474 : struct btrfs_super_block *disk_super)
475 : {
476 :
477 0 : struct btrfs_fs_devices *fs_devices;
478 :
479 : /*
480 : * Handle scanned device having completed its fsid change but
481 : * belonging to a fs_devices that was created by first scanning
482 : * a device which didn't have its fsid/metadata_uuid changed
483 : * at all and the CHANGING_FSID_V2 flag set.
484 : */
485 0 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
486 0 : if (!fs_devices->fsid_change)
487 0 : continue;
488 :
489 0 : if (match_fsid_fs_devices(fs_devices, disk_super->metadata_uuid,
490 0 : fs_devices->fsid))
491 0 : return fs_devices;
492 : }
493 :
494 : /*
495 : * Handle scanned device having completed its fsid change but
496 : * belonging to a fs_devices that was created by a device that
497 : * has an outdated pair of fsid/metadata_uuid and
498 : * CHANGING_FSID_V2 flag set.
499 : */
500 0 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
501 0 : if (!fs_devices->fsid_change)
502 0 : continue;
503 :
504 0 : if (check_fsid_changed(fs_devices, disk_super->metadata_uuid))
505 0 : return fs_devices;
506 : }
507 :
508 0 : return find_fsid(disk_super->fsid, disk_super->metadata_uuid);
509 : }
510 :
511 :
512 : static int
513 3244 : btrfs_get_bdev_and_sb(const char *device_path, blk_mode_t flags, void *holder,
514 : int flush, struct block_device **bdev,
515 : struct btrfs_super_block **disk_super)
516 : {
517 3244 : int ret;
518 :
519 3244 : *bdev = blkdev_get_by_path(device_path, flags, holder, NULL);
520 :
521 3244 : if (IS_ERR(*bdev)) {
522 0 : ret = PTR_ERR(*bdev);
523 0 : goto error;
524 : }
525 :
526 3244 : if (flush)
527 3244 : sync_blockdev(*bdev);
528 3244 : ret = set_blocksize(*bdev, BTRFS_BDEV_BLOCKSIZE);
529 3244 : if (ret) {
530 0 : blkdev_put(*bdev, holder);
531 0 : goto error;
532 : }
533 3244 : invalidate_bdev(*bdev);
534 3244 : *disk_super = btrfs_read_dev_super(*bdev);
535 3244 : if (IS_ERR(*disk_super)) {
536 0 : ret = PTR_ERR(*disk_super);
537 0 : blkdev_put(*bdev, holder);
538 0 : goto error;
539 : }
540 :
541 : return 0;
542 :
543 0 : error:
544 0 : *bdev = NULL;
545 0 : return ret;
546 : }
547 :
548 : /*
549 : * Search and remove all stale devices (which are not mounted). When both
550 : * inputs are NULL, it will search and release all stale devices.
551 : *
552 : * @devt: Optional. When provided will it release all unmounted devices
553 : * matching this devt only.
554 : * @skip_device: Optional. Will skip this device when searching for the stale
555 : * devices.
556 : *
557 : * Return: 0 for success or if @devt is 0.
558 : * -EBUSY if @devt is a mounted device.
559 : * -ENOENT if @devt does not match any device in the list.
560 : */
561 3273 : static int btrfs_free_stale_devices(dev_t devt, struct btrfs_device *skip_device)
562 : {
563 3273 : struct btrfs_fs_devices *fs_devices, *tmp_fs_devices;
564 3273 : struct btrfs_device *device, *tmp_device;
565 3273 : int ret = 0;
566 :
567 3273 : lockdep_assert_held(&uuid_mutex);
568 :
569 3273 : if (devt)
570 3268 : ret = -ENOENT;
571 :
572 9918 : list_for_each_entry_safe(fs_devices, tmp_fs_devices, &fs_uuids, fs_list) {
573 :
574 6645 : mutex_lock(&fs_devices->device_list_mutex);
575 13942 : list_for_each_entry_safe(device, tmp_device,
576 : &fs_devices->devices, dev_list) {
577 7302 : if (skip_device && skip_device == device)
578 3268 : continue;
579 4034 : if (devt && devt != device->devt)
580 4015 : continue;
581 19 : if (fs_devices->opened) {
582 : /* for an already deleted device return 0 */
583 5 : if (devt && ret != 0)
584 0 : ret = -EBUSY;
585 : break;
586 : }
587 :
588 : /* delete the stale device */
589 14 : fs_devices->num_devices--;
590 14 : list_del(&device->dev_list);
591 14 : btrfs_free_device(device);
592 :
593 14 : ret = 0;
594 : }
595 6645 : mutex_unlock(&fs_devices->device_list_mutex);
596 :
597 6645 : if (fs_devices->num_devices == 0) {
598 13 : btrfs_sysfs_remove_fsid(fs_devices);
599 13 : list_del(&fs_devices->fs_list);
600 13 : free_fs_devices(fs_devices);
601 : }
602 : }
603 :
604 3273 : return ret;
605 : }
606 :
607 : /*
608 : * This is only used on mount, and we are protected from competing things
609 : * messing with our fs_devices by the uuid_mutex, thus we do not need the
610 : * fs_devices->device_list_mutex here.
611 : */
612 3244 : static int btrfs_open_one_device(struct btrfs_fs_devices *fs_devices,
613 : struct btrfs_device *device, blk_mode_t flags,
614 : void *holder)
615 : {
616 3244 : struct block_device *bdev;
617 3244 : struct btrfs_super_block *disk_super;
618 3244 : u64 devid;
619 3244 : int ret;
620 :
621 3244 : if (device->bdev)
622 : return -EINVAL;
623 3244 : if (!device->name)
624 : return -EINVAL;
625 :
626 3244 : ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
627 : &bdev, &disk_super);
628 3244 : if (ret)
629 : return ret;
630 :
631 3244 : devid = btrfs_stack_device_id(&disk_super->dev_item);
632 3244 : if (devid != device->devid)
633 0 : goto error_free_page;
634 :
635 6488 : if (memcmp(device->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE))
636 0 : goto error_free_page;
637 :
638 3244 : device->generation = btrfs_super_generation(disk_super);
639 :
640 3244 : if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
641 0 : if (btrfs_super_incompat_flags(disk_super) &
642 : BTRFS_FEATURE_INCOMPAT_METADATA_UUID) {
643 0 : pr_err(
644 : "BTRFS: Invalid seeding and uuid-changed device detected\n");
645 0 : goto error_free_page;
646 : }
647 :
648 0 : clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
649 0 : fs_devices->seeding = true;
650 : } else {
651 3244 : if (bdev_read_only(bdev))
652 0 : clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
653 : else
654 3244 : set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
655 : }
656 :
657 3244 : if (!bdev_nonrot(bdev))
658 3227 : fs_devices->rotating = true;
659 :
660 3244 : if (bdev_max_discard_sectors(bdev))
661 3240 : fs_devices->discardable = true;
662 :
663 3244 : device->bdev = bdev;
664 3244 : clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
665 3244 : device->holder = holder;
666 :
667 3244 : fs_devices->open_devices++;
668 6488 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
669 3244 : device->devid != BTRFS_DEV_REPLACE_DEVID) {
670 3244 : fs_devices->rw_devices++;
671 3244 : list_add_tail(&device->dev_alloc_list, &fs_devices->alloc_list);
672 : }
673 3244 : btrfs_release_disk_super(disk_super);
674 :
675 3244 : return 0;
676 :
677 0 : error_free_page:
678 0 : btrfs_release_disk_super(disk_super);
679 0 : blkdev_put(bdev, holder);
680 :
681 0 : return -EINVAL;
682 : }
683 :
684 : /*
685 : * Handle scanned device having its CHANGING_FSID_V2 flag set and the fs_devices
686 : * being created with a disk that has already completed its fsid change. Such
687 : * disk can belong to an fs which has its FSID changed or to one which doesn't.
688 : * Handle both cases here.
689 : */
690 0 : static struct btrfs_fs_devices *find_fsid_inprogress(
691 : struct btrfs_super_block *disk_super)
692 : {
693 0 : struct btrfs_fs_devices *fs_devices;
694 :
695 0 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
696 0 : if (fs_devices->fsid_change)
697 0 : continue;
698 :
699 0 : if (check_fsid_changed(fs_devices, disk_super->fsid))
700 0 : return fs_devices;
701 : }
702 :
703 0 : return find_fsid(disk_super->fsid, NULL);
704 : }
705 :
706 0 : static struct btrfs_fs_devices *find_fsid_changed(
707 : struct btrfs_super_block *disk_super)
708 : {
709 0 : struct btrfs_fs_devices *fs_devices;
710 :
711 : /*
712 : * Handles the case where scanned device is part of an fs that had
713 : * multiple successful changes of FSID but currently device didn't
714 : * observe it. Meaning our fsid will be different than theirs. We need
715 : * to handle two subcases :
716 : * 1 - The fs still continues to have different METADATA/FSID uuids.
717 : * 2 - The fs is switched back to its original FSID (METADATA/FSID
718 : * are equal).
719 : */
720 0 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
721 : /* Changed UUIDs */
722 0 : if (check_fsid_changed(fs_devices, disk_super->metadata_uuid) &&
723 0 : memcmp(fs_devices->fsid, disk_super->fsid,
724 : BTRFS_FSID_SIZE) != 0)
725 0 : return fs_devices;
726 :
727 : /* Unchanged UUIDs */
728 0 : if (memcmp(fs_devices->metadata_uuid, fs_devices->fsid,
729 0 : BTRFS_FSID_SIZE) == 0 &&
730 : memcmp(fs_devices->fsid, disk_super->metadata_uuid,
731 : BTRFS_FSID_SIZE) == 0)
732 0 : return fs_devices;
733 : }
734 :
735 : return NULL;
736 : }
737 :
738 6720 : static struct btrfs_fs_devices *find_fsid_reverted_metadata(
739 : struct btrfs_super_block *disk_super)
740 : {
741 6720 : struct btrfs_fs_devices *fs_devices;
742 :
743 : /*
744 : * Handle the case where the scanned device is part of an fs whose last
745 : * metadata UUID change reverted it to the original FSID. At the same
746 : * time fs_devices was first created by another constituent device
747 : * which didn't fully observe the operation. This results in an
748 : * btrfs_fs_devices created with metadata/fsid different AND
749 : * btrfs_fs_devices::fsid_change set AND the metadata_uuid of the
750 : * fs_devices equal to the FSID of the disk.
751 : */
752 18398 : list_for_each_entry(fs_devices, &fs_uuids, fs_list) {
753 11678 : if (!fs_devices->fsid_change)
754 11678 : continue;
755 :
756 0 : if (check_fsid_changed(fs_devices, disk_super->fsid))
757 0 : return fs_devices;
758 : }
759 :
760 : return NULL;
761 : }
762 : /*
763 : * Add new device to list of registered devices
764 : *
765 : * Returns:
766 : * device pointer which was just added or updated when successful
767 : * error pointer when failed
768 : */
769 6720 : static noinline struct btrfs_device *device_list_add(const char *path,
770 : struct btrfs_super_block *disk_super,
771 : bool *new_device_added)
772 : {
773 6720 : struct btrfs_device *device;
774 6720 : struct btrfs_fs_devices *fs_devices = NULL;
775 6720 : struct rcu_string *name;
776 6720 : u64 found_transid = btrfs_super_generation(disk_super);
777 6720 : u64 devid = btrfs_stack_device_id(&disk_super->dev_item);
778 6720 : dev_t path_devt;
779 6720 : int error;
780 6720 : bool has_metadata_uuid = (btrfs_super_incompat_flags(disk_super) &
781 : BTRFS_FEATURE_INCOMPAT_METADATA_UUID);
782 6720 : bool fsid_change_in_progress = (btrfs_super_flags(disk_super) &
783 : BTRFS_SUPER_FLAG_CHANGING_FSID_V2);
784 :
785 6720 : error = lookup_bdev(path, &path_devt);
786 6720 : if (error) {
787 0 : btrfs_err(NULL, "failed to lookup block device for path %s: %d",
788 : path, error);
789 0 : return ERR_PTR(error);
790 : }
791 :
792 6720 : if (fsid_change_in_progress) {
793 0 : if (!has_metadata_uuid)
794 0 : fs_devices = find_fsid_inprogress(disk_super);
795 : else
796 0 : fs_devices = find_fsid_changed(disk_super);
797 6720 : } else if (has_metadata_uuid) {
798 0 : fs_devices = find_fsid_with_metadata_uuid(disk_super);
799 : } else {
800 6720 : fs_devices = find_fsid_reverted_metadata(disk_super);
801 6720 : if (!fs_devices)
802 6720 : fs_devices = find_fsid(disk_super->fsid, NULL);
803 : }
804 :
805 :
806 6720 : if (!fs_devices) {
807 3267 : fs_devices = alloc_fs_devices(disk_super->fsid,
808 : has_metadata_uuid ? disk_super->metadata_uuid : NULL);
809 3267 : if (IS_ERR(fs_devices))
810 : return ERR_CAST(fs_devices);
811 :
812 3267 : fs_devices->fsid_change = fsid_change_in_progress;
813 :
814 3267 : mutex_lock(&fs_devices->device_list_mutex);
815 3267 : list_add(&fs_devices->fs_list, &fs_uuids);
816 :
817 3267 : device = NULL;
818 : } else {
819 3453 : struct btrfs_dev_lookup_args args = {
820 : .devid = devid,
821 3453 : .uuid = disk_super->dev_item.uuid,
822 : };
823 :
824 3453 : mutex_lock(&fs_devices->device_list_mutex);
825 3453 : device = btrfs_find_device(fs_devices, &args);
826 :
827 : /*
828 : * If this disk has been pulled into an fs devices created by
829 : * a device which had the CHANGING_FSID_V2 flag then replace the
830 : * metadata_uuid/fsid values of the fs_devices.
831 : */
832 3453 : if (fs_devices->fsid_change &&
833 0 : found_transid > fs_devices->latest_generation) {
834 0 : memcpy(fs_devices->fsid, disk_super->fsid,
835 : BTRFS_FSID_SIZE);
836 :
837 0 : if (has_metadata_uuid)
838 0 : memcpy(fs_devices->metadata_uuid,
839 : disk_super->metadata_uuid,
840 : BTRFS_FSID_SIZE);
841 : else
842 0 : memcpy(fs_devices->metadata_uuid,
843 : disk_super->fsid, BTRFS_FSID_SIZE);
844 :
845 0 : fs_devices->fsid_change = false;
846 : }
847 : }
848 :
849 6720 : if (!device) {
850 3268 : unsigned int nofs_flag;
851 :
852 3268 : if (fs_devices->opened) {
853 0 : btrfs_err(NULL,
854 : "device %s belongs to fsid %pU, and the fs is already mounted",
855 : path, fs_devices->fsid);
856 0 : mutex_unlock(&fs_devices->device_list_mutex);
857 0 : return ERR_PTR(-EBUSY);
858 : }
859 :
860 3268 : nofs_flag = memalloc_nofs_save();
861 3268 : device = btrfs_alloc_device(NULL, &devid,
862 3268 : disk_super->dev_item.uuid, path);
863 3268 : memalloc_nofs_restore(nofs_flag);
864 3268 : if (IS_ERR(device)) {
865 0 : mutex_unlock(&fs_devices->device_list_mutex);
866 : /* we can safely leave the fs_devices entry around */
867 0 : return device;
868 : }
869 :
870 3268 : device->devt = path_devt;
871 :
872 3268 : list_add_rcu(&device->dev_list, &fs_devices->devices);
873 3268 : fs_devices->num_devices++;
874 :
875 3268 : device->fs_devices = fs_devices;
876 3268 : *new_device_added = true;
877 :
878 3268 : if (disk_super->label[0])
879 2 : pr_info(
880 : "BTRFS: device label %s devid %llu transid %llu %s scanned by %s (%d)\n",
881 : disk_super->label, devid, found_transid, path,
882 : current->comm, task_pid_nr(current));
883 : else
884 3266 : pr_info(
885 : "BTRFS: device fsid %pU devid %llu transid %llu %s scanned by %s (%d)\n",
886 : disk_super->fsid, devid, found_transid, path,
887 : current->comm, task_pid_nr(current));
888 :
889 3452 : } else if (!device->name || strcmp(device->name->str, path)) {
890 : /*
891 : * When FS is already mounted.
892 : * 1. If you are here and if the device->name is NULL that
893 : * means this device was missing at time of FS mount.
894 : * 2. If you are here and if the device->name is different
895 : * from 'path' that means either
896 : * a. The same device disappeared and reappeared with
897 : * different name. or
898 : * b. The missing-disk-which-was-replaced, has
899 : * reappeared now.
900 : *
901 : * We must allow 1 and 2a above. But 2b would be a spurious
902 : * and unintentional.
903 : *
904 : * Further in case of 1 and 2a above, the disk at 'path'
905 : * would have missed some transaction when it was away and
906 : * in case of 2a the stale bdev has to be updated as well.
907 : * 2b must not be allowed at all time.
908 : */
909 :
910 : /*
911 : * For now, we do allow update to btrfs_fs_device through the
912 : * btrfs dev scan cli after FS has been mounted. We're still
913 : * tracking a problem where systems fail mount by subvolume id
914 : * when we reject replacement on a mounted FS.
915 : */
916 1224 : if (!fs_devices->opened && found_transid < device->generation) {
917 : /*
918 : * That is if the FS is _not_ mounted and if you
919 : * are here, that means there is more than one
920 : * disk with same uuid and devid.We keep the one
921 : * with larger generation number or the last-in if
922 : * generation are equal.
923 : */
924 0 : mutex_unlock(&fs_devices->device_list_mutex);
925 0 : btrfs_err(NULL,
926 : "device %s already registered with a higher generation, found %llu expect %llu",
927 : path, found_transid, device->generation);
928 0 : return ERR_PTR(-EEXIST);
929 : }
930 :
931 : /*
932 : * We are going to replace the device path for a given devid,
933 : * make sure it's the same device if the device is mounted
934 : *
935 : * NOTE: the device->fs_info may not be reliable here so pass
936 : * in a NULL to message helpers instead. This avoids a possible
937 : * use-after-free when the fs_info and fs_info->sb are already
938 : * torn down.
939 : */
940 1224 : if (device->bdev) {
941 686 : if (device->devt != path_devt) {
942 3 : mutex_unlock(&fs_devices->device_list_mutex);
943 3 : btrfs_warn_in_rcu(NULL,
944 : "duplicate device %s devid %llu generation %llu scanned by %s (%d)",
945 : path, devid, found_transid,
946 : current->comm,
947 : task_pid_nr(current));
948 3 : return ERR_PTR(-EEXIST);
949 : }
950 683 : btrfs_info_in_rcu(NULL,
951 : "devid %llu device path %s changed to %s scanned by %s (%d)",
952 : devid, btrfs_dev_name(device),
953 : path, current->comm,
954 : task_pid_nr(current));
955 : }
956 :
957 1221 : name = rcu_string_strdup(path, GFP_NOFS);
958 1221 : if (!name) {
959 0 : mutex_unlock(&fs_devices->device_list_mutex);
960 0 : return ERR_PTR(-ENOMEM);
961 : }
962 1221 : rcu_string_free(device->name);
963 1221 : rcu_assign_pointer(device->name, name);
964 2442 : if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
965 0 : fs_devices->missing_devices--;
966 0 : clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
967 : }
968 1221 : device->devt = path_devt;
969 : }
970 :
971 : /*
972 : * Unmount does not free the btrfs_device struct but would zero
973 : * generation along with most of the other members. So just update
974 : * it back. We need it to pick the disk with largest generation
975 : * (as above).
976 : */
977 6717 : if (!fs_devices->opened) {
978 5543 : device->generation = found_transid;
979 5543 : fs_devices->latest_generation = max_t(u64, found_transid,
980 : fs_devices->latest_generation);
981 : }
982 :
983 6717 : fs_devices->total_devices = btrfs_super_num_devices(disk_super);
984 :
985 6717 : mutex_unlock(&fs_devices->device_list_mutex);
986 6717 : return device;
987 : }
988 :
989 0 : static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
990 : {
991 0 : struct btrfs_fs_devices *fs_devices;
992 0 : struct btrfs_device *device;
993 0 : struct btrfs_device *orig_dev;
994 0 : int ret = 0;
995 :
996 0 : lockdep_assert_held(&uuid_mutex);
997 :
998 0 : fs_devices = alloc_fs_devices(orig->fsid, NULL);
999 0 : if (IS_ERR(fs_devices))
1000 : return fs_devices;
1001 :
1002 0 : fs_devices->total_devices = orig->total_devices;
1003 :
1004 0 : list_for_each_entry(orig_dev, &orig->devices, dev_list) {
1005 0 : const char *dev_path = NULL;
1006 :
1007 : /*
1008 : * This is ok to do without RCU read locked because we hold the
1009 : * uuid mutex so nothing we touch in here is going to disappear.
1010 : */
1011 0 : if (orig_dev->name)
1012 0 : dev_path = orig_dev->name->str;
1013 :
1014 0 : device = btrfs_alloc_device(NULL, &orig_dev->devid,
1015 0 : orig_dev->uuid, dev_path);
1016 0 : if (IS_ERR(device)) {
1017 0 : ret = PTR_ERR(device);
1018 0 : goto error;
1019 : }
1020 :
1021 0 : if (orig_dev->zone_info) {
1022 0 : struct btrfs_zoned_device_info *zone_info;
1023 :
1024 0 : zone_info = btrfs_clone_dev_zone_info(orig_dev);
1025 0 : if (!zone_info) {
1026 0 : btrfs_free_device(device);
1027 0 : ret = -ENOMEM;
1028 0 : goto error;
1029 : }
1030 : device->zone_info = zone_info;
1031 : }
1032 :
1033 0 : list_add(&device->dev_list, &fs_devices->devices);
1034 0 : device->fs_devices = fs_devices;
1035 0 : fs_devices->num_devices++;
1036 : }
1037 : return fs_devices;
1038 0 : error:
1039 0 : free_fs_devices(fs_devices);
1040 0 : return ERR_PTR(ret);
1041 : }
1042 :
1043 3218 : static void __btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices,
1044 : struct btrfs_device **latest_dev)
1045 : {
1046 3218 : struct btrfs_device *device, *next;
1047 :
1048 : /* This is the initialized path, it is safe to release the devices. */
1049 6437 : list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
1050 6438 : if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state)) {
1051 3219 : if (!test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
1052 6438 : &device->dev_state) &&
1053 0 : !test_bit(BTRFS_DEV_STATE_MISSING,
1054 3219 : &device->dev_state) &&
1055 3219 : (!*latest_dev ||
1056 1 : device->generation > (*latest_dev)->generation)) {
1057 3218 : *latest_dev = device;
1058 : }
1059 3219 : continue;
1060 : }
1061 :
1062 : /*
1063 : * We have already validated the presence of BTRFS_DEV_REPLACE_DEVID,
1064 : * in btrfs_init_dev_replace() so just continue.
1065 : */
1066 0 : if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1067 0 : continue;
1068 :
1069 0 : if (device->bdev) {
1070 0 : blkdev_put(device->bdev, device->holder);
1071 0 : device->bdev = NULL;
1072 0 : fs_devices->open_devices--;
1073 : }
1074 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1075 0 : list_del_init(&device->dev_alloc_list);
1076 0 : clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1077 0 : fs_devices->rw_devices--;
1078 : }
1079 0 : list_del_init(&device->dev_list);
1080 0 : fs_devices->num_devices--;
1081 0 : btrfs_free_device(device);
1082 : }
1083 :
1084 3218 : }
1085 :
1086 : /*
1087 : * After we have read the system tree and know devids belonging to this
1088 : * filesystem, remove the device which does not belong there.
1089 : */
1090 3218 : void btrfs_free_extra_devids(struct btrfs_fs_devices *fs_devices)
1091 : {
1092 3218 : struct btrfs_device *latest_dev = NULL;
1093 3218 : struct btrfs_fs_devices *seed_dev;
1094 :
1095 3218 : mutex_lock(&uuid_mutex);
1096 3218 : __btrfs_free_extra_devids(fs_devices, &latest_dev);
1097 :
1098 3218 : list_for_each_entry(seed_dev, &fs_devices->seed_list, seed_list)
1099 0 : __btrfs_free_extra_devids(seed_dev, &latest_dev);
1100 :
1101 3218 : fs_devices->latest_dev = latest_dev;
1102 :
1103 3218 : mutex_unlock(&uuid_mutex);
1104 3218 : }
1105 :
1106 3244 : static void btrfs_close_bdev(struct btrfs_device *device)
1107 : {
1108 3244 : if (!device->bdev)
1109 : return;
1110 :
1111 6488 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
1112 3244 : sync_blockdev(device->bdev);
1113 3244 : invalidate_bdev(device->bdev);
1114 : }
1115 :
1116 3244 : blkdev_put(device->bdev, device->holder);
1117 : }
1118 :
1119 3244 : static void btrfs_close_one_device(struct btrfs_device *device)
1120 : {
1121 3244 : struct btrfs_fs_devices *fs_devices = device->fs_devices;
1122 :
1123 6488 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
1124 3244 : device->devid != BTRFS_DEV_REPLACE_DEVID) {
1125 3244 : list_del_init(&device->dev_alloc_list);
1126 3244 : fs_devices->rw_devices--;
1127 : }
1128 :
1129 3244 : if (device->devid == BTRFS_DEV_REPLACE_DEVID)
1130 0 : clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
1131 :
1132 6488 : if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
1133 0 : clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
1134 0 : fs_devices->missing_devices--;
1135 : }
1136 :
1137 3244 : btrfs_close_bdev(device);
1138 3244 : if (device->bdev) {
1139 3244 : fs_devices->open_devices--;
1140 3244 : device->bdev = NULL;
1141 : }
1142 3244 : clear_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
1143 3244 : btrfs_destroy_dev_zone_info(device);
1144 :
1145 3244 : device->fs_info = NULL;
1146 3244 : atomic_set(&device->dev_stats_ccnt, 0);
1147 3244 : extent_io_tree_release(&device->alloc_state);
1148 :
1149 : /*
1150 : * Reset the flush error record. We might have a transient flush error
1151 : * in this mount, and if so we aborted the current transaction and set
1152 : * the fs to an error state, guaranteeing no super blocks can be further
1153 : * committed. However that error might be transient and if we unmount the
1154 : * filesystem and mount it again, we should allow the mount to succeed
1155 : * (btrfs_check_rw_degradable() should not fail) - if after mounting the
1156 : * filesystem again we still get flush errors, then we will again abort
1157 : * any transaction and set the error state, guaranteeing no commits of
1158 : * unsafe super blocks.
1159 : */
1160 3244 : device->last_flush_error = 0;
1161 :
1162 : /* Verify the device is back in a pristine state */
1163 3244 : WARN_ON(test_bit(BTRFS_DEV_STATE_FLUSH_SENT, &device->dev_state));
1164 3244 : WARN_ON(test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state));
1165 3244 : WARN_ON(!list_empty(&device->dev_alloc_list));
1166 3244 : WARN_ON(!list_empty(&device->post_commit_list));
1167 3244 : }
1168 :
1169 3471 : static void close_fs_devices(struct btrfs_fs_devices *fs_devices)
1170 : {
1171 3471 : struct btrfs_device *device, *tmp;
1172 :
1173 3471 : lockdep_assert_held(&uuid_mutex);
1174 :
1175 3471 : if (--fs_devices->opened > 0)
1176 : return;
1177 :
1178 6487 : list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list)
1179 3244 : btrfs_close_one_device(device);
1180 :
1181 3243 : WARN_ON(fs_devices->open_devices);
1182 3243 : WARN_ON(fs_devices->rw_devices);
1183 3243 : fs_devices->opened = 0;
1184 3243 : fs_devices->seeding = false;
1185 3243 : fs_devices->fs_info = NULL;
1186 : }
1187 :
1188 3471 : void btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
1189 : {
1190 3471 : LIST_HEAD(list);
1191 3471 : struct btrfs_fs_devices *tmp;
1192 :
1193 3471 : mutex_lock(&uuid_mutex);
1194 3471 : close_fs_devices(fs_devices);
1195 3471 : if (!fs_devices->opened) {
1196 3243 : list_splice_init(&fs_devices->seed_list, &list);
1197 :
1198 : /*
1199 : * If the struct btrfs_fs_devices is not assembled with any
1200 : * other device, it can be re-initialized during the next mount
1201 : * without the needing device-scan step. Therefore, it can be
1202 : * fully freed.
1203 : */
1204 3243 : if (fs_devices->num_devices == 1) {
1205 3242 : list_del(&fs_devices->fs_list);
1206 3242 : free_fs_devices(fs_devices);
1207 : }
1208 : }
1209 :
1210 :
1211 3471 : list_for_each_entry_safe(fs_devices, tmp, &list, seed_list) {
1212 0 : close_fs_devices(fs_devices);
1213 0 : list_del(&fs_devices->seed_list);
1214 0 : free_fs_devices(fs_devices);
1215 : }
1216 3471 : mutex_unlock(&uuid_mutex);
1217 3471 : }
1218 :
1219 3243 : static int open_fs_devices(struct btrfs_fs_devices *fs_devices,
1220 : blk_mode_t flags, void *holder)
1221 : {
1222 3243 : struct btrfs_device *device;
1223 3243 : struct btrfs_device *latest_dev = NULL;
1224 3243 : struct btrfs_device *tmp_device;
1225 :
1226 6487 : list_for_each_entry_safe(device, tmp_device, &fs_devices->devices,
1227 : dev_list) {
1228 3244 : int ret;
1229 :
1230 3244 : ret = btrfs_open_one_device(fs_devices, device, flags, holder);
1231 3244 : if (ret == 0 &&
1232 1 : (!latest_dev || device->generation > latest_dev->generation)) {
1233 : latest_dev = device;
1234 1 : } else if (ret == -ENODATA) {
1235 0 : fs_devices->num_devices--;
1236 0 : list_del(&device->dev_list);
1237 0 : btrfs_free_device(device);
1238 : }
1239 : }
1240 3243 : if (fs_devices->open_devices == 0)
1241 : return -EINVAL;
1242 :
1243 3243 : fs_devices->opened = 1;
1244 3243 : fs_devices->latest_dev = latest_dev;
1245 3243 : fs_devices->total_rw_bytes = 0;
1246 3243 : fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_REGULAR;
1247 3243 : fs_devices->read_policy = BTRFS_READ_POLICY_PID;
1248 :
1249 3243 : return 0;
1250 : }
1251 :
1252 1 : static int devid_cmp(void *priv, const struct list_head *a,
1253 : const struct list_head *b)
1254 : {
1255 1 : const struct btrfs_device *dev1, *dev2;
1256 :
1257 1 : dev1 = list_entry(a, struct btrfs_device, dev_list);
1258 1 : dev2 = list_entry(b, struct btrfs_device, dev_list);
1259 :
1260 1 : if (dev1->devid < dev2->devid)
1261 : return -1;
1262 1 : else if (dev1->devid > dev2->devid)
1263 1 : return 1;
1264 : return 0;
1265 : }
1266 :
1267 3471 : int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
1268 : blk_mode_t flags, void *holder)
1269 : {
1270 3471 : int ret;
1271 :
1272 3471 : lockdep_assert_held(&uuid_mutex);
1273 : /*
1274 : * The device_list_mutex cannot be taken here in case opening the
1275 : * underlying device takes further locks like open_mutex.
1276 : *
1277 : * We also don't need the lock here as this is called during mount and
1278 : * exclusion is provided by uuid_mutex
1279 : */
1280 :
1281 3471 : if (fs_devices->opened) {
1282 228 : fs_devices->opened++;
1283 228 : ret = 0;
1284 : } else {
1285 3243 : list_sort(NULL, &fs_devices->devices, devid_cmp);
1286 3243 : ret = open_fs_devices(fs_devices, flags, holder);
1287 : }
1288 :
1289 3471 : return ret;
1290 : }
1291 :
1292 13251 : void btrfs_release_disk_super(struct btrfs_super_block *super)
1293 : {
1294 13251 : struct page *page = virt_to_page(super);
1295 :
1296 13251 : put_page(page);
1297 13251 : }
1298 :
1299 6721 : static struct btrfs_super_block *btrfs_read_disk_super(struct block_device *bdev,
1300 : u64 bytenr, u64 bytenr_orig)
1301 : {
1302 6721 : struct btrfs_super_block *disk_super;
1303 6721 : struct page *page;
1304 6721 : void *p;
1305 6721 : pgoff_t index;
1306 :
1307 : /* make sure our super fits in the device */
1308 6721 : if (bytenr + PAGE_SIZE >= bdev_nr_bytes(bdev))
1309 : return ERR_PTR(-EINVAL);
1310 :
1311 : /* make sure our super fits in the page */
1312 6720 : if (sizeof(*disk_super) > PAGE_SIZE)
1313 : return ERR_PTR(-EINVAL);
1314 :
1315 : /* make sure our super doesn't straddle pages on disk */
1316 6720 : index = bytenr >> PAGE_SHIFT;
1317 6720 : if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_SHIFT != index)
1318 : return ERR_PTR(-EINVAL);
1319 :
1320 : /* pull in the page with our super */
1321 6720 : page = read_cache_page_gfp(bdev->bd_inode->i_mapping, index, GFP_KERNEL);
1322 :
1323 6720 : if (IS_ERR(page))
1324 : return ERR_CAST(page);
1325 :
1326 6720 : p = page_address(page);
1327 :
1328 : /* align our pointer to the offset of the super block */
1329 6720 : disk_super = p + offset_in_page(bytenr);
1330 :
1331 6720 : if (btrfs_super_bytenr(disk_super) != bytenr_orig ||
1332 : btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1333 0 : btrfs_release_disk_super(p);
1334 0 : return ERR_PTR(-EINVAL);
1335 : }
1336 :
1337 6720 : if (disk_super->label[0] && disk_super->label[BTRFS_LABEL_SIZE - 1])
1338 0 : disk_super->label[BTRFS_LABEL_SIZE - 1] = 0;
1339 :
1340 : return disk_super;
1341 : }
1342 :
1343 5 : int btrfs_forget_devices(dev_t devt)
1344 : {
1345 5 : int ret;
1346 :
1347 5 : mutex_lock(&uuid_mutex);
1348 5 : ret = btrfs_free_stale_devices(devt, NULL);
1349 5 : mutex_unlock(&uuid_mutex);
1350 :
1351 5 : return ret;
1352 : }
1353 :
1354 : /*
1355 : * Look for a btrfs signature on a device. This may be called out of the mount path
1356 : * and we are not allowed to call set_blocksize during the scan. The superblock
1357 : * is read via pagecache
1358 : */
1359 6721 : struct btrfs_device *btrfs_scan_one_device(const char *path, blk_mode_t flags)
1360 : {
1361 6721 : struct btrfs_super_block *disk_super;
1362 6721 : bool new_device_added = false;
1363 6721 : struct btrfs_device *device = NULL;
1364 6721 : struct block_device *bdev;
1365 6721 : u64 bytenr, bytenr_orig;
1366 6721 : int ret;
1367 :
1368 6721 : lockdep_assert_held(&uuid_mutex);
1369 :
1370 : /*
1371 : * we would like to check all the supers, but that would make
1372 : * a btrfs mount succeed after a mkfs from a different FS.
1373 : * So, we need to add a special mount option to scan for
1374 : * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1375 : */
1376 :
1377 : /*
1378 : * Avoid an exclusive open here, as the systemd-udev may initiate the
1379 : * device scan which may race with the user's mount or mkfs command,
1380 : * resulting in failure.
1381 : * Since the device scan is solely for reading purposes, there is no
1382 : * need for an exclusive open. Additionally, the devices are read again
1383 : * during the mount process. It is ok to get some inconsistent
1384 : * values temporarily, as the device paths of the fsid are the only
1385 : * required information for assembling the volume.
1386 : */
1387 6721 : bdev = blkdev_get_by_path(path, flags, NULL, NULL);
1388 6721 : if (IS_ERR(bdev))
1389 : return ERR_CAST(bdev);
1390 :
1391 6721 : bytenr_orig = btrfs_sb_offset(0);
1392 6721 : ret = btrfs_sb_log_location_bdev(bdev, 0, READ, &bytenr);
1393 6721 : if (ret) {
1394 : device = ERR_PTR(ret);
1395 : goto error_bdev_put;
1396 : }
1397 :
1398 6721 : disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr_orig);
1399 6721 : if (IS_ERR(disk_super)) {
1400 1 : device = ERR_CAST(disk_super);
1401 1 : goto error_bdev_put;
1402 : }
1403 :
1404 6720 : device = device_list_add(path, disk_super, &new_device_added);
1405 6720 : if (!IS_ERR(device) && new_device_added)
1406 3268 : btrfs_free_stale_devices(device->devt, device);
1407 :
1408 6720 : btrfs_release_disk_super(disk_super);
1409 :
1410 6721 : error_bdev_put:
1411 6721 : blkdev_put(bdev, NULL);
1412 :
1413 6721 : return device;
1414 : }
1415 :
1416 : /*
1417 : * Try to find a chunk that intersects [start, start + len] range and when one
1418 : * such is found, record the end of it in *start
1419 : */
1420 3677 : static bool contains_pending_extent(struct btrfs_device *device, u64 *start,
1421 : u64 len)
1422 : {
1423 3677 : u64 physical_start, physical_end;
1424 :
1425 3677 : lockdep_assert_held(&device->fs_info->chunk_mutex);
1426 :
1427 3677 : if (!find_first_extent_bit(&device->alloc_state, *start,
1428 : &physical_start, &physical_end,
1429 : CHUNK_ALLOCATED, NULL)) {
1430 :
1431 2519 : if (in_range(physical_start, *start, len) ||
1432 504 : in_range(*start, physical_start,
1433 : physical_end - physical_start)) {
1434 536 : *start = physical_end + 1;
1435 536 : return true;
1436 : }
1437 : }
1438 : return false;
1439 : }
1440 :
1441 1520 : static u64 dev_extent_search_start(struct btrfs_device *device, u64 start)
1442 : {
1443 1520 : switch (device->fs_devices->chunk_alloc_policy) {
1444 1520 : case BTRFS_CHUNK_ALLOC_REGULAR:
1445 1520 : return max_t(u64, start, BTRFS_DEVICE_RANGE_RESERVED);
1446 0 : case BTRFS_CHUNK_ALLOC_ZONED:
1447 : /*
1448 : * We don't care about the starting region like regular
1449 : * allocator, because we anyway use/reserve the first two zones
1450 : * for superblock logging.
1451 : */
1452 0 : return ALIGN(start, device->zone_info->zone_size);
1453 0 : default:
1454 0 : BUG();
1455 : }
1456 : }
1457 :
1458 : static bool dev_extent_hole_check_zoned(struct btrfs_device *device,
1459 : u64 *hole_start, u64 *hole_size,
1460 : u64 num_bytes)
1461 : {
1462 : u64 zone_size = device->zone_info->zone_size;
1463 : u64 pos;
1464 : int ret;
1465 : bool changed = false;
1466 :
1467 : ASSERT(IS_ALIGNED(*hole_start, zone_size));
1468 :
1469 : while (*hole_size > 0) {
1470 : pos = btrfs_find_allocatable_zones(device, *hole_start,
1471 : *hole_start + *hole_size,
1472 : num_bytes);
1473 : if (pos != *hole_start) {
1474 : *hole_size = *hole_start + *hole_size - pos;
1475 : *hole_start = pos;
1476 : changed = true;
1477 : if (*hole_size < num_bytes)
1478 : break;
1479 : }
1480 :
1481 : ret = btrfs_ensure_empty_zones(device, pos, num_bytes);
1482 :
1483 : /* Range is ensured to be empty */
1484 : if (!ret)
1485 : return changed;
1486 :
1487 : /* Given hole range was invalid (outside of device) */
1488 : if (ret == -ERANGE) {
1489 : *hole_start += *hole_size;
1490 : *hole_size = 0;
1491 : return true;
1492 : }
1493 :
1494 : *hole_start += zone_size;
1495 : *hole_size -= zone_size;
1496 : changed = true;
1497 : }
1498 :
1499 : return changed;
1500 : }
1501 :
1502 : /*
1503 : * Check if specified hole is suitable for allocation.
1504 : *
1505 : * @device: the device which we have the hole
1506 : * @hole_start: starting position of the hole
1507 : * @hole_size: the size of the hole
1508 : * @num_bytes: the size of the free space that we need
1509 : *
1510 : * This function may modify @hole_start and @hole_size to reflect the suitable
1511 : * position for allocation. Returns 1 if hole position is updated, 0 otherwise.
1512 : */
1513 3674 : static bool dev_extent_hole_check(struct btrfs_device *device, u64 *hole_start,
1514 : u64 *hole_size, u64 num_bytes)
1515 : {
1516 3674 : bool changed = false;
1517 3674 : u64 hole_end = *hole_start + *hole_size;
1518 :
1519 3674 : for (;;) {
1520 : /*
1521 : * Check before we set max_hole_start, otherwise we could end up
1522 : * sending back this offset anyway.
1523 : */
1524 3674 : if (contains_pending_extent(device, hole_start, *hole_size)) {
1525 534 : if (hole_end >= *hole_start)
1526 504 : *hole_size = hole_end - *hole_start;
1527 : else
1528 30 : *hole_size = 0;
1529 : changed = true;
1530 : }
1531 :
1532 3674 : switch (device->fs_devices->chunk_alloc_policy) {
1533 : case BTRFS_CHUNK_ALLOC_REGULAR:
1534 : /* No extra check */
1535 : break;
1536 : case BTRFS_CHUNK_ALLOC_ZONED:
1537 : if (dev_extent_hole_check_zoned(device, hole_start,
1538 : hole_size, num_bytes)) {
1539 : changed = true;
1540 : /*
1541 : * The changed hole can contain pending extent.
1542 : * Loop again to check that.
1543 : */
1544 : continue;
1545 : }
1546 : break;
1547 0 : default:
1548 0 : BUG();
1549 : }
1550 :
1551 3674 : break;
1552 : }
1553 :
1554 3674 : return changed;
1555 : }
1556 :
1557 : /*
1558 : * Find free space in the specified device.
1559 : *
1560 : * @device: the device which we search the free space in
1561 : * @num_bytes: the size of the free space that we need
1562 : * @search_start: the position from which to begin the search
1563 : * @start: store the start of the free space.
1564 : * @len: the size of the free space. that we find, or the size
1565 : * of the max free space if we don't find suitable free space
1566 : *
1567 : * This does a pretty simple search, the expectation is that it is called very
1568 : * infrequently and that a given device has a small number of extents.
1569 : *
1570 : * @start is used to store the start of the free space if we find. But if we
1571 : * don't find suitable free space, it will be used to store the start position
1572 : * of the max free space.
1573 : *
1574 : * @len is used to store the size of the free space that we find.
1575 : * But if we don't find suitable free space, it is used to store the size of
1576 : * the max free space.
1577 : *
1578 : * NOTE: This function will search *commit* root of device tree, and does extra
1579 : * check to ensure dev extents are not double allocated.
1580 : * This makes the function safe to allocate dev extents but may not report
1581 : * correct usable device space, as device extent freed in current transaction
1582 : * is not reported as available.
1583 : */
1584 1520 : static int find_free_dev_extent_start(struct btrfs_device *device,
1585 : u64 num_bytes, u64 search_start, u64 *start,
1586 : u64 *len)
1587 : {
1588 1520 : struct btrfs_fs_info *fs_info = device->fs_info;
1589 1520 : struct btrfs_root *root = fs_info->dev_root;
1590 1520 : struct btrfs_key key;
1591 1520 : struct btrfs_dev_extent *dev_extent;
1592 1520 : struct btrfs_path *path;
1593 1520 : u64 hole_size;
1594 1520 : u64 max_hole_start;
1595 1520 : u64 max_hole_size;
1596 1520 : u64 extent_end;
1597 1520 : u64 search_end = device->total_bytes;
1598 1520 : int ret;
1599 1520 : int slot;
1600 1520 : struct extent_buffer *l;
1601 :
1602 1520 : search_start = dev_extent_search_start(device, search_start);
1603 :
1604 3040 : WARN_ON(device->zone_info &&
1605 : !IS_ALIGNED(num_bytes, device->zone_info->zone_size));
1606 :
1607 1520 : path = btrfs_alloc_path();
1608 1520 : if (!path)
1609 : return -ENOMEM;
1610 :
1611 : max_hole_start = search_start;
1612 : max_hole_size = 0;
1613 :
1614 2020 : again:
1615 4005 : if (search_start >= search_end ||
1616 1985 : test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
1617 35 : ret = -ENOSPC;
1618 35 : goto out;
1619 : }
1620 :
1621 1985 : path->reada = READA_FORWARD;
1622 1985 : path->search_commit_root = 1;
1623 1985 : path->skip_locking = 1;
1624 :
1625 1985 : key.objectid = device->devid;
1626 1985 : key.offset = search_start;
1627 1985 : key.type = BTRFS_DEV_EXTENT_KEY;
1628 :
1629 1985 : ret = btrfs_search_backwards(root, &key, path);
1630 1985 : if (ret < 0)
1631 0 : goto out;
1632 :
1633 11967 : while (search_start < search_end) {
1634 11915 : l = path->nodes[0];
1635 11915 : slot = path->slots[0];
1636 11915 : if (slot >= btrfs_header_nritems(l)) {
1637 1656 : ret = btrfs_next_leaf(root, path);
1638 1656 : if (ret == 0)
1639 0 : continue;
1640 1656 : if (ret < 0)
1641 0 : goto out;
1642 :
1643 : break;
1644 : }
1645 10259 : btrfs_item_key_to_cpu(l, &key, slot);
1646 :
1647 10259 : if (key.objectid < device->devid)
1648 725 : goto next;
1649 :
1650 9534 : if (key.objectid > device->devid)
1651 : break;
1652 :
1653 9534 : if (key.type != BTRFS_DEV_EXTENT_KEY)
1654 0 : goto next;
1655 :
1656 9534 : if (key.offset > search_end)
1657 : break;
1658 :
1659 9534 : if (key.offset > search_start) {
1660 2018 : hole_size = key.offset - search_start;
1661 2018 : dev_extent_hole_check(device, &search_start, &hole_size,
1662 : num_bytes);
1663 :
1664 2018 : if (hole_size > max_hole_size) {
1665 1917 : max_hole_start = search_start;
1666 1917 : max_hole_size = hole_size;
1667 : }
1668 :
1669 : /*
1670 : * If this free space is greater than which we need,
1671 : * it must be the max free space that we have found
1672 : * until now, so max_hole_start must point to the start
1673 : * of this free space and the length of this free space
1674 : * is stored in max_hole_size. Thus, we return
1675 : * max_hole_start and max_hole_size and go back to the
1676 : * caller.
1677 : */
1678 2018 : if (hole_size >= num_bytes) {
1679 277 : ret = 0;
1680 277 : goto out;
1681 : }
1682 : }
1683 :
1684 9257 : dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1685 9257 : extent_end = key.offset + btrfs_dev_extent_length(l,
1686 : dev_extent);
1687 9257 : if (extent_end > search_start)
1688 8761 : search_start = extent_end;
1689 496 : next:
1690 9982 : path->slots[0]++;
1691 9982 : cond_resched();
1692 : }
1693 :
1694 : /*
1695 : * At this point, search_start should be the end of
1696 : * allocated dev extents, and when shrinking the device,
1697 : * search_end may be smaller than search_start.
1698 : */
1699 1708 : if (search_end > search_start) {
1700 1656 : hole_size = search_end - search_start;
1701 1656 : if (dev_extent_hole_check(device, &search_start, &hole_size,
1702 : num_bytes)) {
1703 500 : btrfs_release_path(path);
1704 500 : goto again;
1705 : }
1706 :
1707 1156 : if (hole_size > max_hole_size) {
1708 1144 : max_hole_start = search_start;
1709 1144 : max_hole_size = hole_size;
1710 : }
1711 : }
1712 :
1713 : /* See above. */
1714 1208 : if (max_hole_size < num_bytes)
1715 : ret = -ENOSPC;
1716 : else
1717 695 : ret = 0;
1718 :
1719 1208 : ASSERT(max_hole_start + max_hole_size <= search_end);
1720 1520 : out:
1721 1520 : btrfs_free_path(path);
1722 1520 : *start = max_hole_start;
1723 1520 : if (len)
1724 1520 : *len = max_hole_size;
1725 : return ret;
1726 : }
1727 :
1728 0 : int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1729 : u64 *start, u64 *len)
1730 : {
1731 : /* FIXME use last free of some kind */
1732 0 : return find_free_dev_extent_start(device, num_bytes, 0, start, len);
1733 : }
1734 :
1735 828 : static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1736 : struct btrfs_device *device,
1737 : u64 start, u64 *dev_extent_len)
1738 : {
1739 828 : struct btrfs_fs_info *fs_info = device->fs_info;
1740 828 : struct btrfs_root *root = fs_info->dev_root;
1741 828 : int ret;
1742 828 : struct btrfs_path *path;
1743 828 : struct btrfs_key key;
1744 828 : struct btrfs_key found_key;
1745 828 : struct extent_buffer *leaf = NULL;
1746 828 : struct btrfs_dev_extent *extent = NULL;
1747 :
1748 828 : path = btrfs_alloc_path();
1749 828 : if (!path)
1750 : return -ENOMEM;
1751 :
1752 828 : key.objectid = device->devid;
1753 828 : key.offset = start;
1754 828 : key.type = BTRFS_DEV_EXTENT_KEY;
1755 828 : again:
1756 828 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1757 828 : if (ret > 0) {
1758 0 : ret = btrfs_previous_item(root, path, key.objectid,
1759 : BTRFS_DEV_EXTENT_KEY);
1760 0 : if (ret)
1761 0 : goto out;
1762 0 : leaf = path->nodes[0];
1763 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1764 0 : extent = btrfs_item_ptr(leaf, path->slots[0],
1765 : struct btrfs_dev_extent);
1766 0 : BUG_ON(found_key.offset > start || found_key.offset +
1767 : btrfs_dev_extent_length(leaf, extent) < start);
1768 0 : key = found_key;
1769 0 : btrfs_release_path(path);
1770 0 : goto again;
1771 828 : } else if (ret == 0) {
1772 828 : leaf = path->nodes[0];
1773 828 : extent = btrfs_item_ptr(leaf, path->slots[0],
1774 : struct btrfs_dev_extent);
1775 : } else {
1776 0 : goto out;
1777 : }
1778 :
1779 828 : *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1780 :
1781 828 : ret = btrfs_del_item(trans, root, path);
1782 828 : if (ret == 0)
1783 828 : set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags);
1784 0 : out:
1785 828 : btrfs_free_path(path);
1786 828 : return ret;
1787 : }
1788 :
1789 1520 : static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1790 : {
1791 1520 : struct extent_map_tree *em_tree;
1792 1520 : struct extent_map *em;
1793 1520 : struct rb_node *n;
1794 1520 : u64 ret = 0;
1795 :
1796 1520 : em_tree = &fs_info->mapping_tree;
1797 1520 : read_lock(&em_tree->lock);
1798 1520 : n = rb_last(&em_tree->map.rb_root);
1799 1520 : if (n) {
1800 1520 : em = rb_entry(n, struct extent_map, rb_node);
1801 1520 : ret = em->start + em->len;
1802 : }
1803 1520 : read_unlock(&em_tree->lock);
1804 :
1805 1520 : return ret;
1806 : }
1807 :
1808 0 : static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1809 : u64 *devid_ret)
1810 : {
1811 0 : int ret;
1812 0 : struct btrfs_key key;
1813 0 : struct btrfs_key found_key;
1814 0 : struct btrfs_path *path;
1815 :
1816 0 : path = btrfs_alloc_path();
1817 0 : if (!path)
1818 : return -ENOMEM;
1819 :
1820 0 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1821 0 : key.type = BTRFS_DEV_ITEM_KEY;
1822 0 : key.offset = (u64)-1;
1823 :
1824 0 : ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1825 0 : if (ret < 0)
1826 0 : goto error;
1827 :
1828 0 : if (ret == 0) {
1829 : /* Corruption */
1830 0 : btrfs_err(fs_info, "corrupted chunk tree devid -1 matched");
1831 0 : ret = -EUCLEAN;
1832 0 : goto error;
1833 : }
1834 :
1835 0 : ret = btrfs_previous_item(fs_info->chunk_root, path,
1836 : BTRFS_DEV_ITEMS_OBJECTID,
1837 : BTRFS_DEV_ITEM_KEY);
1838 0 : if (ret) {
1839 0 : *devid_ret = 1;
1840 : } else {
1841 0 : btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1842 : path->slots[0]);
1843 0 : *devid_ret = found_key.offset + 1;
1844 : }
1845 : ret = 0;
1846 0 : error:
1847 0 : btrfs_free_path(path);
1848 0 : return ret;
1849 : }
1850 :
1851 : /*
1852 : * the device information is stored in the chunk root
1853 : * the btrfs_device struct should be fully filled in
1854 : */
1855 0 : static int btrfs_add_dev_item(struct btrfs_trans_handle *trans,
1856 : struct btrfs_device *device)
1857 : {
1858 0 : int ret;
1859 0 : struct btrfs_path *path;
1860 0 : struct btrfs_dev_item *dev_item;
1861 0 : struct extent_buffer *leaf;
1862 0 : struct btrfs_key key;
1863 0 : unsigned long ptr;
1864 :
1865 0 : path = btrfs_alloc_path();
1866 0 : if (!path)
1867 : return -ENOMEM;
1868 :
1869 0 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1870 0 : key.type = BTRFS_DEV_ITEM_KEY;
1871 0 : key.offset = device->devid;
1872 :
1873 0 : btrfs_reserve_chunk_metadata(trans, true);
1874 0 : ret = btrfs_insert_empty_item(trans, trans->fs_info->chunk_root, path,
1875 : &key, sizeof(*dev_item));
1876 0 : btrfs_trans_release_chunk_metadata(trans);
1877 0 : if (ret)
1878 0 : goto out;
1879 :
1880 0 : leaf = path->nodes[0];
1881 0 : dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1882 :
1883 0 : btrfs_set_device_id(leaf, dev_item, device->devid);
1884 0 : btrfs_set_device_generation(leaf, dev_item, 0);
1885 0 : btrfs_set_device_type(leaf, dev_item, device->type);
1886 0 : btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1887 0 : btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1888 0 : btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1889 0 : btrfs_set_device_total_bytes(leaf, dev_item,
1890 : btrfs_device_get_disk_total_bytes(device));
1891 0 : btrfs_set_device_bytes_used(leaf, dev_item,
1892 : btrfs_device_get_bytes_used(device));
1893 0 : btrfs_set_device_group(leaf, dev_item, 0);
1894 0 : btrfs_set_device_seek_speed(leaf, dev_item, 0);
1895 0 : btrfs_set_device_bandwidth(leaf, dev_item, 0);
1896 0 : btrfs_set_device_start_offset(leaf, dev_item, 0);
1897 :
1898 0 : ptr = btrfs_device_uuid(dev_item);
1899 0 : write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1900 0 : ptr = btrfs_device_fsid(dev_item);
1901 0 : write_extent_buffer(leaf, trans->fs_info->fs_devices->metadata_uuid,
1902 : ptr, BTRFS_FSID_SIZE);
1903 0 : btrfs_mark_buffer_dirty(leaf);
1904 :
1905 0 : ret = 0;
1906 0 : out:
1907 0 : btrfs_free_path(path);
1908 0 : return ret;
1909 : }
1910 :
1911 : /*
1912 : * Function to update ctime/mtime for a given device path.
1913 : * Mainly used for ctime/mtime based probe like libblkid.
1914 : *
1915 : * We don't care about errors here, this is just to be kind to userspace.
1916 : */
1917 0 : static void update_dev_time(const char *device_path)
1918 : {
1919 0 : struct path path;
1920 0 : struct timespec64 now;
1921 0 : int ret;
1922 :
1923 0 : ret = kern_path(device_path, LOOKUP_FOLLOW, &path);
1924 0 : if (ret)
1925 0 : return;
1926 :
1927 0 : now = current_time(d_inode(path.dentry));
1928 0 : inode_update_time(d_inode(path.dentry), &now, S_MTIME | S_CTIME | S_VERSION);
1929 0 : path_put(&path);
1930 : }
1931 :
1932 0 : static int btrfs_rm_dev_item(struct btrfs_trans_handle *trans,
1933 : struct btrfs_device *device)
1934 : {
1935 0 : struct btrfs_root *root = device->fs_info->chunk_root;
1936 0 : int ret;
1937 0 : struct btrfs_path *path;
1938 0 : struct btrfs_key key;
1939 :
1940 0 : path = btrfs_alloc_path();
1941 0 : if (!path)
1942 : return -ENOMEM;
1943 :
1944 0 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1945 0 : key.type = BTRFS_DEV_ITEM_KEY;
1946 0 : key.offset = device->devid;
1947 :
1948 0 : btrfs_reserve_chunk_metadata(trans, false);
1949 0 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1950 0 : btrfs_trans_release_chunk_metadata(trans);
1951 0 : if (ret) {
1952 0 : if (ret > 0)
1953 0 : ret = -ENOENT;
1954 0 : goto out;
1955 : }
1956 :
1957 0 : ret = btrfs_del_item(trans, root, path);
1958 0 : out:
1959 0 : btrfs_free_path(path);
1960 0 : return ret;
1961 : }
1962 :
1963 : /*
1964 : * Verify that @num_devices satisfies the RAID profile constraints in the whole
1965 : * filesystem. It's up to the caller to adjust that number regarding eg. device
1966 : * replace.
1967 : */
1968 1 : static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info,
1969 : u64 num_devices)
1970 : {
1971 1 : u64 all_avail;
1972 1 : unsigned seq;
1973 1 : int i;
1974 :
1975 1 : do {
1976 1 : seq = read_seqbegin(&fs_info->profiles_lock);
1977 :
1978 1 : all_avail = fs_info->avail_data_alloc_bits |
1979 1 : fs_info->avail_system_alloc_bits |
1980 1 : fs_info->avail_metadata_alloc_bits;
1981 1 : } while (read_seqretry(&fs_info->profiles_lock, seq));
1982 :
1983 4 : for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
1984 4 : if (!(all_avail & btrfs_raid_array[i].bg_flag))
1985 3 : continue;
1986 :
1987 1 : if (num_devices < btrfs_raid_array[i].devs_min)
1988 1 : return btrfs_raid_array[i].mindev_error;
1989 : }
1990 :
1991 : return 0;
1992 : }
1993 :
1994 0 : static struct btrfs_device * btrfs_find_next_active_device(
1995 : struct btrfs_fs_devices *fs_devs, struct btrfs_device *device)
1996 : {
1997 0 : struct btrfs_device *next_device;
1998 :
1999 0 : list_for_each_entry(next_device, &fs_devs->devices, dev_list) {
2000 0 : if (next_device != device &&
2001 0 : !test_bit(BTRFS_DEV_STATE_MISSING, &next_device->dev_state)
2002 0 : && next_device->bdev)
2003 0 : return next_device;
2004 : }
2005 :
2006 : return NULL;
2007 : }
2008 :
2009 : /*
2010 : * Helper function to check if the given device is part of s_bdev / latest_dev
2011 : * and replace it with the provided or the next active device, in the context
2012 : * where this function called, there should be always be another device (or
2013 : * this_dev) which is active.
2014 : */
2015 0 : void __cold btrfs_assign_next_active_device(struct btrfs_device *device,
2016 : struct btrfs_device *next_device)
2017 : {
2018 0 : struct btrfs_fs_info *fs_info = device->fs_info;
2019 :
2020 0 : if (!next_device)
2021 0 : next_device = btrfs_find_next_active_device(fs_info->fs_devices,
2022 : device);
2023 0 : ASSERT(next_device);
2024 :
2025 0 : if (fs_info->sb->s_bdev &&
2026 0 : (fs_info->sb->s_bdev == device->bdev))
2027 0 : fs_info->sb->s_bdev = next_device->bdev;
2028 :
2029 0 : if (fs_info->fs_devices->latest_dev->bdev == device->bdev)
2030 0 : fs_info->fs_devices->latest_dev = next_device;
2031 0 : }
2032 :
2033 : /*
2034 : * Return btrfs_fs_devices::num_devices excluding the device that's being
2035 : * currently replaced.
2036 : */
2037 1 : static u64 btrfs_num_devices(struct btrfs_fs_info *fs_info)
2038 : {
2039 1 : u64 num_devices = fs_info->fs_devices->num_devices;
2040 :
2041 1 : down_read(&fs_info->dev_replace.rwsem);
2042 1 : if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
2043 0 : ASSERT(num_devices > 1);
2044 0 : num_devices--;
2045 : }
2046 1 : up_read(&fs_info->dev_replace.rwsem);
2047 :
2048 1 : return num_devices;
2049 : }
2050 :
2051 0 : static void btrfs_scratch_superblock(struct btrfs_fs_info *fs_info,
2052 : struct block_device *bdev, int copy_num)
2053 : {
2054 0 : struct btrfs_super_block *disk_super;
2055 0 : const size_t len = sizeof(disk_super->magic);
2056 0 : const u64 bytenr = btrfs_sb_offset(copy_num);
2057 0 : int ret;
2058 :
2059 0 : disk_super = btrfs_read_disk_super(bdev, bytenr, bytenr);
2060 0 : if (IS_ERR(disk_super))
2061 : return;
2062 :
2063 0 : memset(&disk_super->magic, 0, len);
2064 0 : folio_mark_dirty(virt_to_folio(disk_super));
2065 0 : btrfs_release_disk_super(disk_super);
2066 :
2067 0 : ret = sync_blockdev_range(bdev, bytenr, bytenr + len - 1);
2068 0 : if (ret)
2069 0 : btrfs_warn(fs_info, "error clearing superblock number %d (%d)",
2070 : copy_num, ret);
2071 : }
2072 :
2073 0 : void btrfs_scratch_superblocks(struct btrfs_fs_info *fs_info,
2074 : struct block_device *bdev,
2075 : const char *device_path)
2076 : {
2077 0 : int copy_num;
2078 :
2079 0 : if (!bdev)
2080 : return;
2081 :
2082 0 : for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; copy_num++) {
2083 0 : if (bdev_is_zoned(bdev))
2084 : btrfs_reset_sb_log_zones(bdev, copy_num);
2085 : else
2086 0 : btrfs_scratch_superblock(fs_info, bdev, copy_num);
2087 : }
2088 :
2089 : /* Notify udev that device has changed */
2090 0 : btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
2091 :
2092 : /* Update ctime/mtime for device path for libblkid */
2093 0 : update_dev_time(device_path);
2094 : }
2095 :
2096 1 : int btrfs_rm_device(struct btrfs_fs_info *fs_info,
2097 : struct btrfs_dev_lookup_args *args,
2098 : struct block_device **bdev, void **holder)
2099 : {
2100 1 : struct btrfs_trans_handle *trans;
2101 1 : struct btrfs_device *device;
2102 1 : struct btrfs_fs_devices *cur_devices;
2103 1 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2104 1 : u64 num_devices;
2105 1 : int ret = 0;
2106 :
2107 1 : if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
2108 0 : btrfs_err(fs_info, "device remove not supported on extent tree v2 yet");
2109 0 : return -EINVAL;
2110 : }
2111 :
2112 : /*
2113 : * The device list in fs_devices is accessed without locks (neither
2114 : * uuid_mutex nor device_list_mutex) as it won't change on a mounted
2115 : * filesystem and another device rm cannot run.
2116 : */
2117 1 : num_devices = btrfs_num_devices(fs_info);
2118 :
2119 1 : ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1);
2120 1 : if (ret)
2121 : return ret;
2122 :
2123 1 : device = btrfs_find_device(fs_info->fs_devices, args);
2124 1 : if (!device) {
2125 1 : if (args->missing)
2126 : ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
2127 : else
2128 1 : ret = -ENOENT;
2129 1 : return ret;
2130 : }
2131 :
2132 0 : if (btrfs_pinned_by_swapfile(fs_info, device)) {
2133 0 : btrfs_warn_in_rcu(fs_info,
2134 : "cannot remove device %s (devid %llu) due to active swapfile",
2135 : btrfs_dev_name(device), device->devid);
2136 0 : return -ETXTBSY;
2137 : }
2138 :
2139 0 : if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
2140 : return BTRFS_ERROR_DEV_TGT_REPLACE;
2141 :
2142 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
2143 0 : fs_info->fs_devices->rw_devices == 1)
2144 : return BTRFS_ERROR_DEV_ONLY_WRITABLE;
2145 :
2146 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2147 0 : mutex_lock(&fs_info->chunk_mutex);
2148 0 : list_del_init(&device->dev_alloc_list);
2149 0 : device->fs_devices->rw_devices--;
2150 0 : mutex_unlock(&fs_info->chunk_mutex);
2151 : }
2152 :
2153 0 : ret = btrfs_shrink_device(device, 0);
2154 0 : if (ret)
2155 0 : goto error_undo;
2156 :
2157 0 : trans = btrfs_start_transaction(fs_info->chunk_root, 0);
2158 0 : if (IS_ERR(trans)) {
2159 0 : ret = PTR_ERR(trans);
2160 0 : goto error_undo;
2161 : }
2162 :
2163 0 : ret = btrfs_rm_dev_item(trans, device);
2164 0 : if (ret) {
2165 : /* Any error in dev item removal is critical */
2166 0 : btrfs_crit(fs_info,
2167 : "failed to remove device item for devid %llu: %d",
2168 : device->devid, ret);
2169 0 : btrfs_abort_transaction(trans, ret);
2170 0 : btrfs_end_transaction(trans);
2171 0 : return ret;
2172 : }
2173 :
2174 0 : clear_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2175 0 : btrfs_scrub_cancel_dev(device);
2176 :
2177 : /*
2178 : * the device list mutex makes sure that we don't change
2179 : * the device list while someone else is writing out all
2180 : * the device supers. Whoever is writing all supers, should
2181 : * lock the device list mutex before getting the number of
2182 : * devices in the super block (super_copy). Conversely,
2183 : * whoever updates the number of devices in the super block
2184 : * (super_copy) should hold the device list mutex.
2185 : */
2186 :
2187 : /*
2188 : * In normal cases the cur_devices == fs_devices. But in case
2189 : * of deleting a seed device, the cur_devices should point to
2190 : * its own fs_devices listed under the fs_devices->seed_list.
2191 : */
2192 0 : cur_devices = device->fs_devices;
2193 0 : mutex_lock(&fs_devices->device_list_mutex);
2194 0 : list_del_rcu(&device->dev_list);
2195 :
2196 0 : cur_devices->num_devices--;
2197 0 : cur_devices->total_devices--;
2198 : /* Update total_devices of the parent fs_devices if it's seed */
2199 0 : if (cur_devices != fs_devices)
2200 0 : fs_devices->total_devices--;
2201 :
2202 0 : if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state))
2203 0 : cur_devices->missing_devices--;
2204 :
2205 0 : btrfs_assign_next_active_device(device, NULL);
2206 :
2207 0 : if (device->bdev) {
2208 0 : cur_devices->open_devices--;
2209 : /* remove sysfs entry */
2210 0 : btrfs_sysfs_remove_device(device);
2211 : }
2212 :
2213 0 : num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1;
2214 0 : btrfs_set_super_num_devices(fs_info->super_copy, num_devices);
2215 0 : mutex_unlock(&fs_devices->device_list_mutex);
2216 :
2217 : /*
2218 : * At this point, the device is zero sized and detached from the
2219 : * devices list. All that's left is to zero out the old supers and
2220 : * free the device.
2221 : *
2222 : * We cannot call btrfs_close_bdev() here because we're holding the sb
2223 : * write lock, and blkdev_put() will pull in the ->open_mutex on the
2224 : * block device and it's dependencies. Instead just flush the device
2225 : * and let the caller do the final blkdev_put.
2226 : */
2227 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2228 0 : btrfs_scratch_superblocks(fs_info, device->bdev,
2229 0 : device->name->str);
2230 0 : if (device->bdev) {
2231 0 : sync_blockdev(device->bdev);
2232 0 : invalidate_bdev(device->bdev);
2233 : }
2234 : }
2235 :
2236 0 : *bdev = device->bdev;
2237 0 : *holder = device->holder;
2238 0 : synchronize_rcu();
2239 0 : btrfs_free_device(device);
2240 :
2241 : /*
2242 : * This can happen if cur_devices is the private seed devices list. We
2243 : * cannot call close_fs_devices() here because it expects the uuid_mutex
2244 : * to be held, but in fact we don't need that for the private
2245 : * seed_devices, we can simply decrement cur_devices->opened and then
2246 : * remove it from our list and free the fs_devices.
2247 : */
2248 0 : if (cur_devices->num_devices == 0) {
2249 0 : list_del_init(&cur_devices->seed_list);
2250 0 : ASSERT(cur_devices->opened == 1);
2251 0 : cur_devices->opened--;
2252 0 : free_fs_devices(cur_devices);
2253 : }
2254 :
2255 0 : ret = btrfs_commit_transaction(trans);
2256 :
2257 0 : return ret;
2258 :
2259 0 : error_undo:
2260 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
2261 0 : mutex_lock(&fs_info->chunk_mutex);
2262 0 : list_add(&device->dev_alloc_list,
2263 : &fs_devices->alloc_list);
2264 0 : device->fs_devices->rw_devices++;
2265 0 : mutex_unlock(&fs_info->chunk_mutex);
2266 : }
2267 : return ret;
2268 : }
2269 :
2270 0 : void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_device *srcdev)
2271 : {
2272 0 : struct btrfs_fs_devices *fs_devices;
2273 :
2274 0 : lockdep_assert_held(&srcdev->fs_info->fs_devices->device_list_mutex);
2275 :
2276 : /*
2277 : * in case of fs with no seed, srcdev->fs_devices will point
2278 : * to fs_devices of fs_info. However when the dev being replaced is
2279 : * a seed dev it will point to the seed's local fs_devices. In short
2280 : * srcdev will have its correct fs_devices in both the cases.
2281 : */
2282 0 : fs_devices = srcdev->fs_devices;
2283 :
2284 0 : list_del_rcu(&srcdev->dev_list);
2285 0 : list_del(&srcdev->dev_alloc_list);
2286 0 : fs_devices->num_devices--;
2287 0 : if (test_bit(BTRFS_DEV_STATE_MISSING, &srcdev->dev_state))
2288 0 : fs_devices->missing_devices--;
2289 :
2290 0 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &srcdev->dev_state))
2291 0 : fs_devices->rw_devices--;
2292 :
2293 0 : if (srcdev->bdev)
2294 0 : fs_devices->open_devices--;
2295 0 : }
2296 :
2297 0 : void btrfs_rm_dev_replace_free_srcdev(struct btrfs_device *srcdev)
2298 : {
2299 0 : struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
2300 :
2301 0 : mutex_lock(&uuid_mutex);
2302 :
2303 0 : btrfs_close_bdev(srcdev);
2304 0 : synchronize_rcu();
2305 0 : btrfs_free_device(srcdev);
2306 :
2307 : /* if this is no devs we rather delete the fs_devices */
2308 0 : if (!fs_devices->num_devices) {
2309 : /*
2310 : * On a mounted FS, num_devices can't be zero unless it's a
2311 : * seed. In case of a seed device being replaced, the replace
2312 : * target added to the sprout FS, so there will be no more
2313 : * device left under the seed FS.
2314 : */
2315 0 : ASSERT(fs_devices->seeding);
2316 :
2317 0 : list_del_init(&fs_devices->seed_list);
2318 0 : close_fs_devices(fs_devices);
2319 0 : free_fs_devices(fs_devices);
2320 : }
2321 0 : mutex_unlock(&uuid_mutex);
2322 0 : }
2323 :
2324 0 : void btrfs_destroy_dev_replace_tgtdev(struct btrfs_device *tgtdev)
2325 : {
2326 0 : struct btrfs_fs_devices *fs_devices = tgtdev->fs_info->fs_devices;
2327 :
2328 0 : mutex_lock(&fs_devices->device_list_mutex);
2329 :
2330 0 : btrfs_sysfs_remove_device(tgtdev);
2331 :
2332 0 : if (tgtdev->bdev)
2333 0 : fs_devices->open_devices--;
2334 :
2335 0 : fs_devices->num_devices--;
2336 :
2337 0 : btrfs_assign_next_active_device(tgtdev, NULL);
2338 :
2339 0 : list_del_rcu(&tgtdev->dev_list);
2340 :
2341 0 : mutex_unlock(&fs_devices->device_list_mutex);
2342 :
2343 0 : btrfs_scratch_superblocks(tgtdev->fs_info, tgtdev->bdev,
2344 0 : tgtdev->name->str);
2345 :
2346 0 : btrfs_close_bdev(tgtdev);
2347 0 : synchronize_rcu();
2348 0 : btrfs_free_device(tgtdev);
2349 0 : }
2350 :
2351 : /*
2352 : * Populate args from device at path.
2353 : *
2354 : * @fs_info: the filesystem
2355 : * @args: the args to populate
2356 : * @path: the path to the device
2357 : *
2358 : * This will read the super block of the device at @path and populate @args with
2359 : * the devid, fsid, and uuid. This is meant to be used for ioctls that need to
2360 : * lookup a device to operate on, but need to do it before we take any locks.
2361 : * This properly handles the special case of "missing" that a user may pass in,
2362 : * and does some basic sanity checks. The caller must make sure that @path is
2363 : * properly NUL terminated before calling in, and must call
2364 : * btrfs_put_dev_args_from_path() in order to free up the temporary fsid and
2365 : * uuid buffers.
2366 : *
2367 : * Return: 0 for success, -errno for failure
2368 : */
2369 0 : int btrfs_get_dev_args_from_path(struct btrfs_fs_info *fs_info,
2370 : struct btrfs_dev_lookup_args *args,
2371 : const char *path)
2372 : {
2373 0 : struct btrfs_super_block *disk_super;
2374 0 : struct block_device *bdev;
2375 0 : int ret;
2376 :
2377 0 : if (!path || !path[0])
2378 : return -EINVAL;
2379 0 : if (!strcmp(path, "missing")) {
2380 0 : args->missing = true;
2381 0 : return 0;
2382 : }
2383 :
2384 0 : args->uuid = kzalloc(BTRFS_UUID_SIZE, GFP_KERNEL);
2385 0 : args->fsid = kzalloc(BTRFS_FSID_SIZE, GFP_KERNEL);
2386 0 : if (!args->uuid || !args->fsid) {
2387 0 : btrfs_put_dev_args_from_path(args);
2388 0 : return -ENOMEM;
2389 : }
2390 :
2391 0 : ret = btrfs_get_bdev_and_sb(path, BLK_OPEN_READ, NULL, 0,
2392 : &bdev, &disk_super);
2393 0 : if (ret) {
2394 0 : btrfs_put_dev_args_from_path(args);
2395 0 : return ret;
2396 : }
2397 :
2398 0 : args->devid = btrfs_stack_device_id(&disk_super->dev_item);
2399 0 : memcpy(args->uuid, disk_super->dev_item.uuid, BTRFS_UUID_SIZE);
2400 0 : if (btrfs_fs_incompat(fs_info, METADATA_UUID))
2401 0 : memcpy(args->fsid, disk_super->metadata_uuid, BTRFS_FSID_SIZE);
2402 : else
2403 0 : memcpy(args->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
2404 0 : btrfs_release_disk_super(disk_super);
2405 0 : blkdev_put(bdev, NULL);
2406 0 : return 0;
2407 : }
2408 :
2409 : /*
2410 : * Only use this jointly with btrfs_get_dev_args_from_path() because we will
2411 : * allocate our ->uuid and ->fsid pointers, everybody else uses local variables
2412 : * that don't need to be freed.
2413 : */
2414 1 : void btrfs_put_dev_args_from_path(struct btrfs_dev_lookup_args *args)
2415 : {
2416 1 : kfree(args->uuid);
2417 1 : kfree(args->fsid);
2418 1 : args->uuid = NULL;
2419 1 : args->fsid = NULL;
2420 1 : }
2421 :
2422 0 : struct btrfs_device *btrfs_find_device_by_devspec(
2423 : struct btrfs_fs_info *fs_info, u64 devid,
2424 : const char *device_path)
2425 : {
2426 0 : BTRFS_DEV_LOOKUP_ARGS(args);
2427 0 : struct btrfs_device *device;
2428 0 : int ret;
2429 :
2430 0 : if (devid) {
2431 0 : args.devid = devid;
2432 0 : device = btrfs_find_device(fs_info->fs_devices, &args);
2433 0 : if (!device)
2434 : return ERR_PTR(-ENOENT);
2435 0 : return device;
2436 : }
2437 :
2438 0 : ret = btrfs_get_dev_args_from_path(fs_info, &args, device_path);
2439 0 : if (ret)
2440 0 : return ERR_PTR(ret);
2441 0 : device = btrfs_find_device(fs_info->fs_devices, &args);
2442 0 : btrfs_put_dev_args_from_path(&args);
2443 0 : if (!device)
2444 0 : return ERR_PTR(-ENOENT);
2445 : return device;
2446 : }
2447 :
2448 0 : static struct btrfs_fs_devices *btrfs_init_sprout(struct btrfs_fs_info *fs_info)
2449 : {
2450 0 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2451 0 : struct btrfs_fs_devices *old_devices;
2452 0 : struct btrfs_fs_devices *seed_devices;
2453 :
2454 0 : lockdep_assert_held(&uuid_mutex);
2455 0 : if (!fs_devices->seeding)
2456 : return ERR_PTR(-EINVAL);
2457 :
2458 : /*
2459 : * Private copy of the seed devices, anchored at
2460 : * fs_info->fs_devices->seed_list
2461 : */
2462 0 : seed_devices = alloc_fs_devices(NULL, NULL);
2463 0 : if (IS_ERR(seed_devices))
2464 : return seed_devices;
2465 :
2466 : /*
2467 : * It's necessary to retain a copy of the original seed fs_devices in
2468 : * fs_uuids so that filesystems which have been seeded can successfully
2469 : * reference the seed device from open_seed_devices. This also supports
2470 : * multiple fs seed.
2471 : */
2472 0 : old_devices = clone_fs_devices(fs_devices);
2473 0 : if (IS_ERR(old_devices)) {
2474 0 : kfree(seed_devices);
2475 0 : return old_devices;
2476 : }
2477 :
2478 0 : list_add(&old_devices->fs_list, &fs_uuids);
2479 :
2480 0 : memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2481 0 : seed_devices->opened = 1;
2482 0 : INIT_LIST_HEAD(&seed_devices->devices);
2483 0 : INIT_LIST_HEAD(&seed_devices->alloc_list);
2484 0 : mutex_init(&seed_devices->device_list_mutex);
2485 :
2486 0 : return seed_devices;
2487 : }
2488 :
2489 : /*
2490 : * Splice seed devices into the sprout fs_devices.
2491 : * Generate a new fsid for the sprouted read-write filesystem.
2492 : */
2493 0 : static void btrfs_setup_sprout(struct btrfs_fs_info *fs_info,
2494 : struct btrfs_fs_devices *seed_devices)
2495 : {
2496 0 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2497 0 : struct btrfs_super_block *disk_super = fs_info->super_copy;
2498 0 : struct btrfs_device *device;
2499 0 : u64 super_flags;
2500 :
2501 : /*
2502 : * We are updating the fsid, the thread leading to device_list_add()
2503 : * could race, so uuid_mutex is needed.
2504 : */
2505 0 : lockdep_assert_held(&uuid_mutex);
2506 :
2507 : /*
2508 : * The threads listed below may traverse dev_list but can do that without
2509 : * device_list_mutex:
2510 : * - All device ops and balance - as we are in btrfs_exclop_start.
2511 : * - Various dev_list readers - are using RCU.
2512 : * - btrfs_ioctl_fitrim() - is using RCU.
2513 : *
2514 : * For-read threads as below are using device_list_mutex:
2515 : * - Readonly scrub btrfs_scrub_dev()
2516 : * - Readonly scrub btrfs_scrub_progress()
2517 : * - btrfs_get_dev_stats()
2518 : */
2519 0 : lockdep_assert_held(&fs_devices->device_list_mutex);
2520 :
2521 0 : list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2522 : synchronize_rcu);
2523 0 : list_for_each_entry(device, &seed_devices->devices, dev_list)
2524 0 : device->fs_devices = seed_devices;
2525 :
2526 0 : fs_devices->seeding = false;
2527 0 : fs_devices->num_devices = 0;
2528 0 : fs_devices->open_devices = 0;
2529 0 : fs_devices->missing_devices = 0;
2530 0 : fs_devices->rotating = false;
2531 0 : list_add(&seed_devices->seed_list, &fs_devices->seed_list);
2532 :
2533 0 : generate_random_uuid(fs_devices->fsid);
2534 0 : memcpy(fs_devices->metadata_uuid, fs_devices->fsid, BTRFS_FSID_SIZE);
2535 0 : memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2536 :
2537 0 : super_flags = btrfs_super_flags(disk_super) &
2538 : ~BTRFS_SUPER_FLAG_SEEDING;
2539 0 : btrfs_set_super_flags(disk_super, super_flags);
2540 0 : }
2541 :
2542 : /*
2543 : * Store the expected generation for seed devices in device items.
2544 : */
2545 0 : static int btrfs_finish_sprout(struct btrfs_trans_handle *trans)
2546 : {
2547 0 : BTRFS_DEV_LOOKUP_ARGS(args);
2548 0 : struct btrfs_fs_info *fs_info = trans->fs_info;
2549 0 : struct btrfs_root *root = fs_info->chunk_root;
2550 0 : struct btrfs_path *path;
2551 0 : struct extent_buffer *leaf;
2552 0 : struct btrfs_dev_item *dev_item;
2553 0 : struct btrfs_device *device;
2554 0 : struct btrfs_key key;
2555 0 : u8 fs_uuid[BTRFS_FSID_SIZE];
2556 0 : u8 dev_uuid[BTRFS_UUID_SIZE];
2557 0 : int ret;
2558 :
2559 0 : path = btrfs_alloc_path();
2560 0 : if (!path)
2561 : return -ENOMEM;
2562 :
2563 0 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2564 0 : key.offset = 0;
2565 0 : key.type = BTRFS_DEV_ITEM_KEY;
2566 :
2567 0 : while (1) {
2568 0 : btrfs_reserve_chunk_metadata(trans, false);
2569 0 : ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2570 0 : btrfs_trans_release_chunk_metadata(trans);
2571 0 : if (ret < 0)
2572 0 : goto error;
2573 :
2574 0 : leaf = path->nodes[0];
2575 0 : next_slot:
2576 0 : if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2577 0 : ret = btrfs_next_leaf(root, path);
2578 0 : if (ret > 0)
2579 : break;
2580 0 : if (ret < 0)
2581 0 : goto error;
2582 0 : leaf = path->nodes[0];
2583 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2584 0 : btrfs_release_path(path);
2585 0 : continue;
2586 : }
2587 :
2588 0 : btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2589 0 : if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2590 0 : key.type != BTRFS_DEV_ITEM_KEY)
2591 : break;
2592 :
2593 0 : dev_item = btrfs_item_ptr(leaf, path->slots[0],
2594 : struct btrfs_dev_item);
2595 0 : args.devid = btrfs_device_id(leaf, dev_item);
2596 0 : read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2597 : BTRFS_UUID_SIZE);
2598 0 : read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2599 : BTRFS_FSID_SIZE);
2600 0 : args.uuid = dev_uuid;
2601 0 : args.fsid = fs_uuid;
2602 0 : device = btrfs_find_device(fs_info->fs_devices, &args);
2603 0 : BUG_ON(!device); /* Logic error */
2604 :
2605 0 : if (device->fs_devices->seeding) {
2606 0 : btrfs_set_device_generation(leaf, dev_item,
2607 : device->generation);
2608 0 : btrfs_mark_buffer_dirty(leaf);
2609 : }
2610 :
2611 0 : path->slots[0]++;
2612 0 : goto next_slot;
2613 : }
2614 : ret = 0;
2615 0 : error:
2616 0 : btrfs_free_path(path);
2617 0 : return ret;
2618 : }
2619 :
2620 0 : int btrfs_init_new_device(struct btrfs_fs_info *fs_info, const char *device_path)
2621 : {
2622 0 : struct btrfs_root *root = fs_info->dev_root;
2623 0 : struct btrfs_trans_handle *trans;
2624 0 : struct btrfs_device *device;
2625 0 : struct block_device *bdev;
2626 0 : struct super_block *sb = fs_info->sb;
2627 0 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2628 0 : struct btrfs_fs_devices *seed_devices = NULL;
2629 0 : u64 orig_super_total_bytes;
2630 0 : u64 orig_super_num_devices;
2631 0 : int ret = 0;
2632 0 : bool seeding_dev = false;
2633 0 : bool locked = false;
2634 :
2635 0 : if (sb_rdonly(sb) && !fs_devices->seeding)
2636 : return -EROFS;
2637 :
2638 0 : bdev = blkdev_get_by_path(device_path, BLK_OPEN_WRITE,
2639 : fs_info->bdev_holder, NULL);
2640 0 : if (IS_ERR(bdev))
2641 0 : return PTR_ERR(bdev);
2642 :
2643 0 : if (!btrfs_check_device_zone_type(fs_info, bdev)) {
2644 : ret = -EINVAL;
2645 : goto error;
2646 : }
2647 :
2648 0 : if (fs_devices->seeding) {
2649 0 : seeding_dev = true;
2650 0 : down_write(&sb->s_umount);
2651 0 : mutex_lock(&uuid_mutex);
2652 0 : locked = true;
2653 : }
2654 :
2655 0 : sync_blockdev(bdev);
2656 :
2657 0 : rcu_read_lock();
2658 0 : list_for_each_entry_rcu(device, &fs_devices->devices, dev_list) {
2659 0 : if (device->bdev == bdev) {
2660 0 : ret = -EEXIST;
2661 0 : rcu_read_unlock();
2662 0 : goto error;
2663 : }
2664 : }
2665 0 : rcu_read_unlock();
2666 :
2667 0 : device = btrfs_alloc_device(fs_info, NULL, NULL, device_path);
2668 0 : if (IS_ERR(device)) {
2669 : /* we can safely leave the fs_devices entry around */
2670 0 : ret = PTR_ERR(device);
2671 0 : goto error;
2672 : }
2673 :
2674 0 : device->fs_info = fs_info;
2675 0 : device->bdev = bdev;
2676 0 : ret = lookup_bdev(device_path, &device->devt);
2677 0 : if (ret)
2678 0 : goto error_free_device;
2679 :
2680 0 : ret = btrfs_get_dev_zone_info(device, false);
2681 0 : if (ret)
2682 : goto error_free_device;
2683 :
2684 0 : trans = btrfs_start_transaction(root, 0);
2685 0 : if (IS_ERR(trans)) {
2686 0 : ret = PTR_ERR(trans);
2687 0 : goto error_free_zone;
2688 : }
2689 :
2690 0 : set_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state);
2691 0 : device->generation = trans->transid;
2692 0 : device->io_width = fs_info->sectorsize;
2693 0 : device->io_align = fs_info->sectorsize;
2694 0 : device->sector_size = fs_info->sectorsize;
2695 0 : device->total_bytes =
2696 0 : round_down(bdev_nr_bytes(bdev), fs_info->sectorsize);
2697 0 : device->disk_total_bytes = device->total_bytes;
2698 0 : device->commit_total_bytes = device->total_bytes;
2699 0 : set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
2700 0 : clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
2701 0 : device->holder = fs_info->bdev_holder;
2702 0 : device->dev_stats_valid = 1;
2703 0 : set_blocksize(device->bdev, BTRFS_BDEV_BLOCKSIZE);
2704 :
2705 0 : if (seeding_dev) {
2706 0 : btrfs_clear_sb_rdonly(sb);
2707 :
2708 : /* GFP_KERNEL allocation must not be under device_list_mutex */
2709 0 : seed_devices = btrfs_init_sprout(fs_info);
2710 0 : if (IS_ERR(seed_devices)) {
2711 0 : ret = PTR_ERR(seed_devices);
2712 0 : btrfs_abort_transaction(trans, ret);
2713 0 : goto error_trans;
2714 : }
2715 : }
2716 :
2717 0 : mutex_lock(&fs_devices->device_list_mutex);
2718 0 : if (seeding_dev) {
2719 0 : btrfs_setup_sprout(fs_info, seed_devices);
2720 0 : btrfs_assign_next_active_device(fs_info->fs_devices->latest_dev,
2721 : device);
2722 : }
2723 :
2724 0 : device->fs_devices = fs_devices;
2725 :
2726 0 : mutex_lock(&fs_info->chunk_mutex);
2727 0 : list_add_rcu(&device->dev_list, &fs_devices->devices);
2728 0 : list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
2729 0 : fs_devices->num_devices++;
2730 0 : fs_devices->open_devices++;
2731 0 : fs_devices->rw_devices++;
2732 0 : fs_devices->total_devices++;
2733 0 : fs_devices->total_rw_bytes += device->total_bytes;
2734 :
2735 0 : atomic64_add(device->total_bytes, &fs_info->free_chunk_space);
2736 :
2737 0 : if (!bdev_nonrot(bdev))
2738 0 : fs_devices->rotating = true;
2739 :
2740 0 : orig_super_total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
2741 0 : btrfs_set_super_total_bytes(fs_info->super_copy,
2742 0 : round_down(orig_super_total_bytes + device->total_bytes,
2743 : fs_info->sectorsize));
2744 :
2745 0 : orig_super_num_devices = btrfs_super_num_devices(fs_info->super_copy);
2746 0 : btrfs_set_super_num_devices(fs_info->super_copy,
2747 : orig_super_num_devices + 1);
2748 :
2749 : /*
2750 : * we've got more storage, clear any full flags on the space
2751 : * infos
2752 : */
2753 0 : btrfs_clear_space_info_full(fs_info);
2754 :
2755 0 : mutex_unlock(&fs_info->chunk_mutex);
2756 :
2757 : /* Add sysfs device entry */
2758 0 : btrfs_sysfs_add_device(device);
2759 :
2760 0 : mutex_unlock(&fs_devices->device_list_mutex);
2761 :
2762 0 : if (seeding_dev) {
2763 0 : mutex_lock(&fs_info->chunk_mutex);
2764 0 : ret = init_first_rw_device(trans);
2765 0 : mutex_unlock(&fs_info->chunk_mutex);
2766 0 : if (ret) {
2767 0 : btrfs_abort_transaction(trans, ret);
2768 0 : goto error_sysfs;
2769 : }
2770 : }
2771 :
2772 0 : ret = btrfs_add_dev_item(trans, device);
2773 0 : if (ret) {
2774 0 : btrfs_abort_transaction(trans, ret);
2775 0 : goto error_sysfs;
2776 : }
2777 :
2778 0 : if (seeding_dev) {
2779 0 : ret = btrfs_finish_sprout(trans);
2780 0 : if (ret) {
2781 0 : btrfs_abort_transaction(trans, ret);
2782 0 : goto error_sysfs;
2783 : }
2784 :
2785 : /*
2786 : * fs_devices now represents the newly sprouted filesystem and
2787 : * its fsid has been changed by btrfs_sprout_splice().
2788 : */
2789 0 : btrfs_sysfs_update_sprout_fsid(fs_devices);
2790 : }
2791 :
2792 0 : ret = btrfs_commit_transaction(trans);
2793 :
2794 0 : if (seeding_dev) {
2795 0 : mutex_unlock(&uuid_mutex);
2796 0 : up_write(&sb->s_umount);
2797 0 : locked = false;
2798 :
2799 0 : if (ret) /* transaction commit */
2800 : return ret;
2801 :
2802 0 : ret = btrfs_relocate_sys_chunks(fs_info);
2803 0 : if (ret < 0)
2804 0 : btrfs_handle_fs_error(fs_info, ret,
2805 : "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command.");
2806 0 : trans = btrfs_attach_transaction(root);
2807 0 : if (IS_ERR(trans)) {
2808 0 : if (PTR_ERR(trans) == -ENOENT)
2809 : return 0;
2810 0 : ret = PTR_ERR(trans);
2811 0 : trans = NULL;
2812 0 : goto error_sysfs;
2813 : }
2814 0 : ret = btrfs_commit_transaction(trans);
2815 : }
2816 :
2817 : /*
2818 : * Now that we have written a new super block to this device, check all
2819 : * other fs_devices list if device_path alienates any other scanned
2820 : * device.
2821 : * We can ignore the return value as it typically returns -EINVAL and
2822 : * only succeeds if the device was an alien.
2823 : */
2824 0 : btrfs_forget_devices(device->devt);
2825 :
2826 : /* Update ctime/mtime for blkid or udev */
2827 0 : update_dev_time(device_path);
2828 :
2829 0 : return ret;
2830 :
2831 0 : error_sysfs:
2832 0 : btrfs_sysfs_remove_device(device);
2833 0 : mutex_lock(&fs_info->fs_devices->device_list_mutex);
2834 0 : mutex_lock(&fs_info->chunk_mutex);
2835 0 : list_del_rcu(&device->dev_list);
2836 0 : list_del(&device->dev_alloc_list);
2837 0 : fs_info->fs_devices->num_devices--;
2838 0 : fs_info->fs_devices->open_devices--;
2839 0 : fs_info->fs_devices->rw_devices--;
2840 0 : fs_info->fs_devices->total_devices--;
2841 0 : fs_info->fs_devices->total_rw_bytes -= device->total_bytes;
2842 0 : atomic64_sub(device->total_bytes, &fs_info->free_chunk_space);
2843 0 : btrfs_set_super_total_bytes(fs_info->super_copy,
2844 : orig_super_total_bytes);
2845 0 : btrfs_set_super_num_devices(fs_info->super_copy,
2846 : orig_super_num_devices);
2847 0 : mutex_unlock(&fs_info->chunk_mutex);
2848 0 : mutex_unlock(&fs_info->fs_devices->device_list_mutex);
2849 0 : error_trans:
2850 0 : if (seeding_dev)
2851 0 : btrfs_set_sb_rdonly(sb);
2852 0 : if (trans)
2853 0 : btrfs_end_transaction(trans);
2854 0 : error_free_zone:
2855 : btrfs_destroy_dev_zone_info(device);
2856 0 : error_free_device:
2857 0 : btrfs_free_device(device);
2858 0 : error:
2859 0 : blkdev_put(bdev, fs_info->bdev_holder);
2860 0 : if (locked) {
2861 0 : mutex_unlock(&uuid_mutex);
2862 0 : up_write(&sb->s_umount);
2863 : }
2864 : return ret;
2865 : }
2866 :
2867 2631 : static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2868 : struct btrfs_device *device)
2869 : {
2870 2631 : int ret;
2871 2631 : struct btrfs_path *path;
2872 2631 : struct btrfs_root *root = device->fs_info->chunk_root;
2873 2631 : struct btrfs_dev_item *dev_item;
2874 2631 : struct extent_buffer *leaf;
2875 2631 : struct btrfs_key key;
2876 :
2877 2631 : path = btrfs_alloc_path();
2878 2631 : if (!path)
2879 : return -ENOMEM;
2880 :
2881 2631 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2882 2631 : key.type = BTRFS_DEV_ITEM_KEY;
2883 2631 : key.offset = device->devid;
2884 :
2885 2631 : ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2886 2631 : if (ret < 0)
2887 0 : goto out;
2888 :
2889 2631 : if (ret > 0) {
2890 0 : ret = -ENOENT;
2891 0 : goto out;
2892 : }
2893 :
2894 2631 : leaf = path->nodes[0];
2895 2631 : dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2896 :
2897 2631 : btrfs_set_device_id(leaf, dev_item, device->devid);
2898 2631 : btrfs_set_device_type(leaf, dev_item, device->type);
2899 2631 : btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2900 2631 : btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2901 2631 : btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2902 2631 : btrfs_set_device_total_bytes(leaf, dev_item,
2903 : btrfs_device_get_disk_total_bytes(device));
2904 2631 : btrfs_set_device_bytes_used(leaf, dev_item,
2905 : btrfs_device_get_bytes_used(device));
2906 2631 : btrfs_mark_buffer_dirty(leaf);
2907 :
2908 2631 : out:
2909 2631 : btrfs_free_path(path);
2910 2631 : return ret;
2911 : }
2912 :
2913 1 : int btrfs_grow_device(struct btrfs_trans_handle *trans,
2914 : struct btrfs_device *device, u64 new_size)
2915 : {
2916 1 : struct btrfs_fs_info *fs_info = device->fs_info;
2917 1 : struct btrfs_super_block *super_copy = fs_info->super_copy;
2918 1 : u64 old_total;
2919 1 : u64 diff;
2920 1 : int ret;
2921 :
2922 1 : if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
2923 : return -EACCES;
2924 :
2925 1 : new_size = round_down(new_size, fs_info->sectorsize);
2926 :
2927 1 : mutex_lock(&fs_info->chunk_mutex);
2928 1 : old_total = btrfs_super_total_bytes(super_copy);
2929 1 : diff = round_down(new_size - device->total_bytes, fs_info->sectorsize);
2930 :
2931 1 : if (new_size <= device->total_bytes ||
2932 0 : test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
2933 0 : mutex_unlock(&fs_info->chunk_mutex);
2934 0 : return -EINVAL;
2935 : }
2936 :
2937 1 : btrfs_set_super_total_bytes(super_copy,
2938 1 : round_down(old_total + diff, fs_info->sectorsize));
2939 1 : device->fs_devices->total_rw_bytes += diff;
2940 :
2941 1 : btrfs_device_set_total_bytes(device, new_size);
2942 1 : btrfs_device_set_disk_total_bytes(device, new_size);
2943 1 : btrfs_clear_space_info_full(device->fs_info);
2944 1 : if (list_empty(&device->post_commit_list))
2945 1 : list_add_tail(&device->post_commit_list,
2946 1 : &trans->transaction->dev_update_list);
2947 1 : mutex_unlock(&fs_info->chunk_mutex);
2948 :
2949 1 : btrfs_reserve_chunk_metadata(trans, false);
2950 1 : ret = btrfs_update_device(trans, device);
2951 1 : btrfs_trans_release_chunk_metadata(trans);
2952 :
2953 1 : return ret;
2954 : }
2955 :
2956 533 : static int btrfs_free_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
2957 : {
2958 533 : struct btrfs_fs_info *fs_info = trans->fs_info;
2959 533 : struct btrfs_root *root = fs_info->chunk_root;
2960 533 : int ret;
2961 533 : struct btrfs_path *path;
2962 533 : struct btrfs_key key;
2963 :
2964 533 : path = btrfs_alloc_path();
2965 533 : if (!path)
2966 : return -ENOMEM;
2967 :
2968 533 : key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2969 533 : key.offset = chunk_offset;
2970 533 : key.type = BTRFS_CHUNK_ITEM_KEY;
2971 :
2972 533 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2973 533 : if (ret < 0)
2974 0 : goto out;
2975 533 : else if (ret > 0) { /* Logic error or corruption */
2976 0 : btrfs_handle_fs_error(fs_info, -ENOENT,
2977 : "Failed lookup while freeing chunk.");
2978 0 : ret = -ENOENT;
2979 0 : goto out;
2980 : }
2981 :
2982 533 : ret = btrfs_del_item(trans, root, path);
2983 533 : if (ret < 0)
2984 0 : btrfs_handle_fs_error(fs_info, ret,
2985 : "Failed to delete chunk item.");
2986 533 : out:
2987 533 : btrfs_free_path(path);
2988 533 : return ret;
2989 : }
2990 :
2991 90 : static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
2992 : {
2993 90 : struct btrfs_super_block *super_copy = fs_info->super_copy;
2994 90 : struct btrfs_disk_key *disk_key;
2995 90 : struct btrfs_chunk *chunk;
2996 90 : u8 *ptr;
2997 90 : int ret = 0;
2998 90 : u32 num_stripes;
2999 90 : u32 array_size;
3000 90 : u32 len = 0;
3001 90 : u32 cur;
3002 90 : struct btrfs_key key;
3003 :
3004 90 : lockdep_assert_held(&fs_info->chunk_mutex);
3005 90 : array_size = btrfs_super_sys_array_size(super_copy);
3006 :
3007 90 : ptr = super_copy->sys_chunk_array;
3008 90 : cur = 0;
3009 :
3010 270 : while (cur < array_size) {
3011 180 : disk_key = (struct btrfs_disk_key *)ptr;
3012 180 : btrfs_disk_key_to_cpu(&key, disk_key);
3013 :
3014 180 : len = sizeof(*disk_key);
3015 :
3016 180 : if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3017 180 : chunk = (struct btrfs_chunk *)(ptr + len);
3018 180 : num_stripes = btrfs_stack_chunk_num_stripes(chunk);
3019 180 : len += btrfs_chunk_item_size(num_stripes);
3020 : } else {
3021 : ret = -EIO;
3022 : break;
3023 : }
3024 180 : if (key.objectid == BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
3025 180 : key.offset == chunk_offset) {
3026 180 : memmove(ptr, ptr + len, array_size - (cur + len));
3027 90 : array_size -= len;
3028 90 : btrfs_set_super_sys_array_size(super_copy, array_size);
3029 : } else {
3030 90 : ptr += len;
3031 90 : cur += len;
3032 : }
3033 : }
3034 90 : return ret;
3035 : }
3036 :
3037 : /*
3038 : * btrfs_get_chunk_map() - Find the mapping containing the given logical extent.
3039 : * @logical: Logical block offset in bytes.
3040 : * @length: Length of extent in bytes.
3041 : *
3042 : * Return: Chunk mapping or ERR_PTR.
3043 : */
3044 31176432 : struct extent_map *btrfs_get_chunk_map(struct btrfs_fs_info *fs_info,
3045 : u64 logical, u64 length)
3046 : {
3047 31176432 : struct extent_map_tree *em_tree;
3048 31176432 : struct extent_map *em;
3049 :
3050 31176432 : em_tree = &fs_info->mapping_tree;
3051 31176432 : read_lock(&em_tree->lock);
3052 31177746 : em = lookup_extent_mapping(em_tree, logical, length);
3053 31177556 : read_unlock(&em_tree->lock);
3054 :
3055 31177347 : if (!em) {
3056 0 : btrfs_crit(fs_info, "unable to find logical %llu length %llu",
3057 : logical, length);
3058 0 : return ERR_PTR(-EINVAL);
3059 : }
3060 :
3061 31177347 : if (em->start > logical || em->start + em->len < logical) {
3062 0 : btrfs_crit(fs_info,
3063 : "found a bad mapping, wanted %llu-%llu, found %llu-%llu",
3064 : logical, length, em->start, em->start + em->len);
3065 0 : free_extent_map(em);
3066 0 : return ERR_PTR(-EINVAL);
3067 : }
3068 :
3069 : /* callers are responsible for dropping em's ref. */
3070 : return em;
3071 : }
3072 :
3073 533 : static int remove_chunk_item(struct btrfs_trans_handle *trans,
3074 : struct map_lookup *map, u64 chunk_offset)
3075 : {
3076 533 : int i;
3077 :
3078 : /*
3079 : * Removing chunk items and updating the device items in the chunks btree
3080 : * requires holding the chunk_mutex.
3081 : * See the comment at btrfs_chunk_alloc() for the details.
3082 : */
3083 533 : lockdep_assert_held(&trans->fs_info->chunk_mutex);
3084 :
3085 1361 : for (i = 0; i < map->num_stripes; i++) {
3086 828 : int ret;
3087 :
3088 828 : ret = btrfs_update_device(trans, map->stripes[i].dev);
3089 828 : if (ret)
3090 0 : return ret;
3091 : }
3092 :
3093 533 : return btrfs_free_chunk(trans, chunk_offset);
3094 : }
3095 :
3096 533 : int btrfs_remove_chunk(struct btrfs_trans_handle *trans, u64 chunk_offset)
3097 : {
3098 533 : struct btrfs_fs_info *fs_info = trans->fs_info;
3099 533 : struct extent_map *em;
3100 533 : struct map_lookup *map;
3101 533 : u64 dev_extent_len = 0;
3102 533 : int i, ret = 0;
3103 533 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
3104 :
3105 533 : em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
3106 533 : if (IS_ERR(em)) {
3107 : /*
3108 : * This is a logic error, but we don't want to just rely on the
3109 : * user having built with ASSERT enabled, so if ASSERT doesn't
3110 : * do anything we still error out.
3111 : */
3112 0 : ASSERT(0);
3113 0 : return PTR_ERR(em);
3114 : }
3115 533 : map = em->map_lookup;
3116 :
3117 : /*
3118 : * First delete the device extent items from the devices btree.
3119 : * We take the device_list_mutex to avoid racing with the finishing phase
3120 : * of a device replace operation. See the comment below before acquiring
3121 : * fs_info->chunk_mutex. Note that here we do not acquire the chunk_mutex
3122 : * because that can result in a deadlock when deleting the device extent
3123 : * items from the devices btree - COWing an extent buffer from the btree
3124 : * may result in allocating a new metadata chunk, which would attempt to
3125 : * lock again fs_info->chunk_mutex.
3126 : */
3127 533 : mutex_lock(&fs_devices->device_list_mutex);
3128 1894 : for (i = 0; i < map->num_stripes; i++) {
3129 828 : struct btrfs_device *device = map->stripes[i].dev;
3130 828 : ret = btrfs_free_dev_extent(trans, device,
3131 : map->stripes[i].physical,
3132 : &dev_extent_len);
3133 828 : if (ret) {
3134 0 : mutex_unlock(&fs_devices->device_list_mutex);
3135 0 : btrfs_abort_transaction(trans, ret);
3136 0 : goto out;
3137 : }
3138 :
3139 828 : if (device->bytes_used > 0) {
3140 828 : mutex_lock(&fs_info->chunk_mutex);
3141 828 : btrfs_device_set_bytes_used(device,
3142 828 : device->bytes_used - dev_extent_len);
3143 828 : atomic64_add(dev_extent_len, &fs_info->free_chunk_space);
3144 828 : btrfs_clear_space_info_full(fs_info);
3145 828 : mutex_unlock(&fs_info->chunk_mutex);
3146 : }
3147 : }
3148 533 : mutex_unlock(&fs_devices->device_list_mutex);
3149 :
3150 : /*
3151 : * We acquire fs_info->chunk_mutex for 2 reasons:
3152 : *
3153 : * 1) Just like with the first phase of the chunk allocation, we must
3154 : * reserve system space, do all chunk btree updates and deletions, and
3155 : * update the system chunk array in the superblock while holding this
3156 : * mutex. This is for similar reasons as explained on the comment at
3157 : * the top of btrfs_chunk_alloc();
3158 : *
3159 : * 2) Prevent races with the final phase of a device replace operation
3160 : * that replaces the device object associated with the map's stripes,
3161 : * because the device object's id can change at any time during that
3162 : * final phase of the device replace operation
3163 : * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
3164 : * replaced device and then see it with an ID of
3165 : * BTRFS_DEV_REPLACE_DEVID, which would cause a failure when updating
3166 : * the device item, which does not exists on the chunk btree.
3167 : * The finishing phase of device replace acquires both the
3168 : * device_list_mutex and the chunk_mutex, in that order, so we are
3169 : * safe by just acquiring the chunk_mutex.
3170 : */
3171 533 : trans->removing_chunk = true;
3172 533 : mutex_lock(&fs_info->chunk_mutex);
3173 :
3174 533 : check_system_chunk(trans, map->type);
3175 :
3176 533 : ret = remove_chunk_item(trans, map, chunk_offset);
3177 : /*
3178 : * Normally we should not get -ENOSPC since we reserved space before
3179 : * through the call to check_system_chunk().
3180 : *
3181 : * Despite our system space_info having enough free space, we may not
3182 : * be able to allocate extents from its block groups, because all have
3183 : * an incompatible profile, which will force us to allocate a new system
3184 : * block group with the right profile, or right after we called
3185 : * check_system_space() above, a scrub turned the only system block group
3186 : * with enough free space into RO mode.
3187 : * This is explained with more detail at do_chunk_alloc().
3188 : *
3189 : * So if we get -ENOSPC, allocate a new system chunk and retry once.
3190 : */
3191 533 : if (ret == -ENOSPC) {
3192 0 : const u64 sys_flags = btrfs_system_alloc_profile(fs_info);
3193 0 : struct btrfs_block_group *sys_bg;
3194 :
3195 0 : sys_bg = btrfs_create_chunk(trans, sys_flags);
3196 0 : if (IS_ERR(sys_bg)) {
3197 0 : ret = PTR_ERR(sys_bg);
3198 0 : btrfs_abort_transaction(trans, ret);
3199 0 : goto out;
3200 : }
3201 :
3202 0 : ret = btrfs_chunk_alloc_add_chunk_item(trans, sys_bg);
3203 0 : if (ret) {
3204 0 : btrfs_abort_transaction(trans, ret);
3205 0 : goto out;
3206 : }
3207 :
3208 0 : ret = remove_chunk_item(trans, map, chunk_offset);
3209 0 : if (ret) {
3210 0 : btrfs_abort_transaction(trans, ret);
3211 0 : goto out;
3212 : }
3213 533 : } else if (ret) {
3214 0 : btrfs_abort_transaction(trans, ret);
3215 0 : goto out;
3216 : }
3217 :
3218 533 : trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len);
3219 :
3220 533 : if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3221 90 : ret = btrfs_del_sys_chunk(fs_info, chunk_offset);
3222 90 : if (ret) {
3223 0 : btrfs_abort_transaction(trans, ret);
3224 0 : goto out;
3225 : }
3226 : }
3227 :
3228 533 : mutex_unlock(&fs_info->chunk_mutex);
3229 533 : trans->removing_chunk = false;
3230 :
3231 : /*
3232 : * We are done with chunk btree updates and deletions, so release the
3233 : * system space we previously reserved (with check_system_chunk()).
3234 : */
3235 533 : btrfs_trans_release_chunk_metadata(trans);
3236 :
3237 533 : ret = btrfs_remove_block_group(trans, chunk_offset, em);
3238 533 : if (ret) {
3239 0 : btrfs_abort_transaction(trans, ret);
3240 0 : goto out;
3241 : }
3242 :
3243 533 : out:
3244 533 : if (trans->removing_chunk) {
3245 0 : mutex_unlock(&fs_info->chunk_mutex);
3246 0 : trans->removing_chunk = false;
3247 : }
3248 : /* once for us */
3249 533 : free_extent_map(em);
3250 533 : return ret;
3251 : }
3252 :
3253 525 : int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset)
3254 : {
3255 525 : struct btrfs_root *root = fs_info->chunk_root;
3256 525 : struct btrfs_trans_handle *trans;
3257 525 : struct btrfs_block_group *block_group;
3258 525 : u64 length;
3259 525 : int ret;
3260 :
3261 525 : if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
3262 0 : btrfs_err(fs_info,
3263 : "relocate: not supported on extent tree v2 yet");
3264 0 : return -EINVAL;
3265 : }
3266 :
3267 : /*
3268 : * Prevent races with automatic removal of unused block groups.
3269 : * After we relocate and before we remove the chunk with offset
3270 : * chunk_offset, automatic removal of the block group can kick in,
3271 : * resulting in a failure when calling btrfs_remove_chunk() below.
3272 : *
3273 : * Make sure to acquire this mutex before doing a tree search (dev
3274 : * or chunk trees) to find chunks. Otherwise the cleaner kthread might
3275 : * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
3276 : * we release the path used to search the chunk/dev tree and before
3277 : * the current task acquires this mutex and calls us.
3278 : */
3279 525 : lockdep_assert_held(&fs_info->reclaim_bgs_lock);
3280 :
3281 : /* step one, relocate all the extents inside this chunk */
3282 525 : btrfs_scrub_pause(fs_info);
3283 525 : ret = btrfs_relocate_block_group(fs_info, chunk_offset);
3284 525 : btrfs_scrub_continue(fs_info);
3285 525 : if (ret) {
3286 : /*
3287 : * If we had a transaction abort, stop all running scrubs.
3288 : * See transaction.c:cleanup_transaction() why we do it here.
3289 : */
3290 4 : if (BTRFS_FS_ERROR(fs_info))
3291 0 : btrfs_scrub_cancel(fs_info);
3292 4 : return ret;
3293 : }
3294 :
3295 521 : block_group = btrfs_lookup_block_group(fs_info, chunk_offset);
3296 521 : if (!block_group)
3297 : return -ENOENT;
3298 521 : btrfs_discard_cancel_work(&fs_info->discard_ctl, block_group);
3299 521 : length = block_group->length;
3300 521 : btrfs_put_block_group(block_group);
3301 :
3302 : /*
3303 : * On a zoned file system, discard the whole block group, this will
3304 : * trigger a REQ_OP_ZONE_RESET operation on the device zone. If
3305 : * resetting the zone fails, don't treat it as a fatal problem from the
3306 : * filesystem's point of view.
3307 : */
3308 521 : if (btrfs_is_zoned(fs_info)) {
3309 : ret = btrfs_discard_extent(fs_info, chunk_offset, length, NULL);
3310 : if (ret)
3311 : btrfs_info(fs_info,
3312 : "failed to reset zone %llu after relocation",
3313 : chunk_offset);
3314 : }
3315 :
3316 521 : trans = btrfs_start_trans_remove_block_group(root->fs_info,
3317 : chunk_offset);
3318 521 : if (IS_ERR(trans)) {
3319 0 : ret = PTR_ERR(trans);
3320 0 : btrfs_handle_fs_error(root->fs_info, ret, NULL);
3321 0 : return ret;
3322 : }
3323 :
3324 : /*
3325 : * step two, delete the device extents and the
3326 : * chunk tree entries
3327 : */
3328 521 : ret = btrfs_remove_chunk(trans, chunk_offset);
3329 521 : btrfs_end_transaction(trans);
3330 521 : return ret;
3331 : }
3332 :
3333 0 : static int btrfs_relocate_sys_chunks(struct btrfs_fs_info *fs_info)
3334 : {
3335 0 : struct btrfs_root *chunk_root = fs_info->chunk_root;
3336 0 : struct btrfs_path *path;
3337 0 : struct extent_buffer *leaf;
3338 0 : struct btrfs_chunk *chunk;
3339 0 : struct btrfs_key key;
3340 0 : struct btrfs_key found_key;
3341 0 : u64 chunk_type;
3342 0 : bool retried = false;
3343 0 : int failed = 0;
3344 0 : int ret;
3345 :
3346 0 : path = btrfs_alloc_path();
3347 0 : if (!path)
3348 : return -ENOMEM;
3349 :
3350 0 : again:
3351 0 : key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3352 0 : key.offset = (u64)-1;
3353 0 : key.type = BTRFS_CHUNK_ITEM_KEY;
3354 :
3355 0 : while (1) {
3356 0 : mutex_lock(&fs_info->reclaim_bgs_lock);
3357 0 : ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3358 0 : if (ret < 0) {
3359 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3360 0 : goto error;
3361 : }
3362 0 : BUG_ON(ret == 0); /* Corruption */
3363 :
3364 0 : ret = btrfs_previous_item(chunk_root, path, key.objectid,
3365 0 : key.type);
3366 0 : if (ret)
3367 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3368 0 : if (ret < 0)
3369 0 : goto error;
3370 0 : if (ret > 0)
3371 : break;
3372 :
3373 0 : leaf = path->nodes[0];
3374 0 : btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3375 :
3376 0 : chunk = btrfs_item_ptr(leaf, path->slots[0],
3377 : struct btrfs_chunk);
3378 0 : chunk_type = btrfs_chunk_type(leaf, chunk);
3379 0 : btrfs_release_path(path);
3380 :
3381 0 : if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
3382 0 : ret = btrfs_relocate_chunk(fs_info, found_key.offset);
3383 0 : if (ret == -ENOSPC)
3384 0 : failed++;
3385 : else
3386 0 : BUG_ON(ret);
3387 : }
3388 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3389 :
3390 0 : if (found_key.offset == 0)
3391 : break;
3392 0 : key.offset = found_key.offset - 1;
3393 : }
3394 0 : ret = 0;
3395 0 : if (failed && !retried) {
3396 0 : failed = 0;
3397 0 : retried = true;
3398 0 : goto again;
3399 0 : } else if (WARN_ON(failed && retried)) {
3400 0 : ret = -ENOSPC;
3401 : }
3402 0 : error:
3403 0 : btrfs_free_path(path);
3404 0 : return ret;
3405 : }
3406 :
3407 : /*
3408 : * return 1 : allocate a data chunk successfully,
3409 : * return <0: errors during allocating a data chunk,
3410 : * return 0 : no need to allocate a data chunk.
3411 : */
3412 505 : static int btrfs_may_alloc_data_chunk(struct btrfs_fs_info *fs_info,
3413 : u64 chunk_offset)
3414 : {
3415 505 : struct btrfs_block_group *cache;
3416 505 : u64 bytes_used;
3417 505 : u64 chunk_type;
3418 :
3419 505 : cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3420 505 : ASSERT(cache);
3421 505 : chunk_type = cache->flags;
3422 505 : btrfs_put_block_group(cache);
3423 :
3424 505 : if (!(chunk_type & BTRFS_BLOCK_GROUP_DATA))
3425 : return 0;
3426 :
3427 229 : spin_lock(&fs_info->data_sinfo->lock);
3428 229 : bytes_used = fs_info->data_sinfo->bytes_used;
3429 229 : spin_unlock(&fs_info->data_sinfo->lock);
3430 :
3431 229 : if (!bytes_used) {
3432 23 : struct btrfs_trans_handle *trans;
3433 23 : int ret;
3434 :
3435 23 : trans = btrfs_join_transaction(fs_info->tree_root);
3436 23 : if (IS_ERR(trans))
3437 0 : return PTR_ERR(trans);
3438 :
3439 23 : ret = btrfs_force_chunk_alloc(trans, BTRFS_BLOCK_GROUP_DATA);
3440 23 : btrfs_end_transaction(trans);
3441 23 : if (ret < 0)
3442 : return ret;
3443 23 : return 1;
3444 : }
3445 :
3446 : return 0;
3447 : }
3448 :
3449 191 : static int insert_balance_item(struct btrfs_fs_info *fs_info,
3450 : struct btrfs_balance_control *bctl)
3451 : {
3452 191 : struct btrfs_root *root = fs_info->tree_root;
3453 191 : struct btrfs_trans_handle *trans;
3454 191 : struct btrfs_balance_item *item;
3455 191 : struct btrfs_disk_balance_args disk_bargs;
3456 191 : struct btrfs_path *path;
3457 191 : struct extent_buffer *leaf;
3458 191 : struct btrfs_key key;
3459 191 : int ret, err;
3460 :
3461 191 : path = btrfs_alloc_path();
3462 191 : if (!path)
3463 : return -ENOMEM;
3464 :
3465 191 : trans = btrfs_start_transaction(root, 0);
3466 191 : if (IS_ERR(trans)) {
3467 0 : btrfs_free_path(path);
3468 0 : return PTR_ERR(trans);
3469 : }
3470 :
3471 191 : key.objectid = BTRFS_BALANCE_OBJECTID;
3472 191 : key.type = BTRFS_TEMPORARY_ITEM_KEY;
3473 191 : key.offset = 0;
3474 :
3475 191 : ret = btrfs_insert_empty_item(trans, root, path, &key,
3476 : sizeof(*item));
3477 191 : if (ret)
3478 0 : goto out;
3479 :
3480 191 : leaf = path->nodes[0];
3481 191 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3482 :
3483 191 : memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3484 :
3485 191 : btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
3486 191 : btrfs_set_balance_data(leaf, item, &disk_bargs);
3487 191 : btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
3488 191 : btrfs_set_balance_meta(leaf, item, &disk_bargs);
3489 191 : btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
3490 191 : btrfs_set_balance_sys(leaf, item, &disk_bargs);
3491 :
3492 191 : btrfs_set_balance_flags(leaf, item, bctl->flags);
3493 :
3494 191 : btrfs_mark_buffer_dirty(leaf);
3495 191 : out:
3496 191 : btrfs_free_path(path);
3497 191 : err = btrfs_commit_transaction(trans);
3498 191 : if (err && !ret)
3499 0 : ret = err;
3500 : return ret;
3501 : }
3502 :
3503 191 : static int del_balance_item(struct btrfs_fs_info *fs_info)
3504 : {
3505 191 : struct btrfs_root *root = fs_info->tree_root;
3506 191 : struct btrfs_trans_handle *trans;
3507 191 : struct btrfs_path *path;
3508 191 : struct btrfs_key key;
3509 191 : int ret, err;
3510 :
3511 191 : path = btrfs_alloc_path();
3512 191 : if (!path)
3513 : return -ENOMEM;
3514 :
3515 191 : trans = btrfs_start_transaction_fallback_global_rsv(root, 0);
3516 191 : if (IS_ERR(trans)) {
3517 0 : btrfs_free_path(path);
3518 0 : return PTR_ERR(trans);
3519 : }
3520 :
3521 191 : key.objectid = BTRFS_BALANCE_OBJECTID;
3522 191 : key.type = BTRFS_TEMPORARY_ITEM_KEY;
3523 191 : key.offset = 0;
3524 :
3525 191 : ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3526 191 : if (ret < 0)
3527 0 : goto out;
3528 191 : if (ret > 0) {
3529 0 : ret = -ENOENT;
3530 0 : goto out;
3531 : }
3532 :
3533 191 : ret = btrfs_del_item(trans, root, path);
3534 191 : out:
3535 191 : btrfs_free_path(path);
3536 191 : err = btrfs_commit_transaction(trans);
3537 191 : if (err && !ret)
3538 0 : ret = err;
3539 : return ret;
3540 : }
3541 :
3542 : /*
3543 : * This is a heuristic used to reduce the number of chunks balanced on
3544 : * resume after balance was interrupted.
3545 : */
3546 0 : static void update_balance_args(struct btrfs_balance_control *bctl)
3547 : {
3548 : /*
3549 : * Turn on soft mode for chunk types that were being converted.
3550 : */
3551 0 : if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
3552 0 : bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
3553 0 : if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
3554 0 : bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
3555 0 : if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
3556 0 : bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
3557 :
3558 : /*
3559 : * Turn on usage filter if is not already used. The idea is
3560 : * that chunks that we have already balanced should be
3561 : * reasonably full. Don't do it for chunks that are being
3562 : * converted - that will keep us from relocating unconverted
3563 : * (albeit full) chunks.
3564 : */
3565 0 : if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3566 0 : !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3567 : !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3568 0 : bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3569 0 : bctl->data.usage = 90;
3570 : }
3571 0 : if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3572 0 : !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3573 : !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3574 0 : bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3575 0 : bctl->sys.usage = 90;
3576 : }
3577 0 : if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3578 0 : !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3579 : !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3580 0 : bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3581 0 : bctl->meta.usage = 90;
3582 : }
3583 0 : }
3584 :
3585 : /*
3586 : * Clear the balance status in fs_info and delete the balance item from disk.
3587 : */
3588 191 : static void reset_balance_state(struct btrfs_fs_info *fs_info)
3589 : {
3590 191 : struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3591 191 : int ret;
3592 :
3593 191 : BUG_ON(!fs_info->balance_ctl);
3594 :
3595 191 : spin_lock(&fs_info->balance_lock);
3596 191 : fs_info->balance_ctl = NULL;
3597 191 : spin_unlock(&fs_info->balance_lock);
3598 :
3599 191 : kfree(bctl);
3600 191 : ret = del_balance_item(fs_info);
3601 191 : if (ret)
3602 0 : btrfs_handle_fs_error(fs_info, ret, NULL);
3603 191 : }
3604 :
3605 : /*
3606 : * Balance filters. Return 1 if chunk should be filtered out
3607 : * (should not be balanced).
3608 : */
3609 : static int chunk_profiles_filter(u64 chunk_type,
3610 : struct btrfs_balance_args *bargs)
3611 : {
3612 0 : chunk_type = chunk_to_extended(chunk_type) &
3613 : BTRFS_EXTENDED_PROFILE_MASK;
3614 :
3615 0 : if (bargs->profiles & chunk_type)
3616 : return 0;
3617 :
3618 : return 1;
3619 : }
3620 :
3621 0 : static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3622 : struct btrfs_balance_args *bargs)
3623 : {
3624 0 : struct btrfs_block_group *cache;
3625 0 : u64 chunk_used;
3626 0 : u64 user_thresh_min;
3627 0 : u64 user_thresh_max;
3628 0 : int ret = 1;
3629 :
3630 0 : cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3631 0 : chunk_used = cache->used;
3632 :
3633 0 : if (bargs->usage_min == 0)
3634 : user_thresh_min = 0;
3635 : else
3636 0 : user_thresh_min = mult_perc(cache->length, bargs->usage_min);
3637 :
3638 0 : if (bargs->usage_max == 0)
3639 : user_thresh_max = 1;
3640 0 : else if (bargs->usage_max > 100)
3641 0 : user_thresh_max = cache->length;
3642 : else
3643 0 : user_thresh_max = mult_perc(cache->length, bargs->usage_max);
3644 :
3645 0 : if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max)
3646 0 : ret = 0;
3647 :
3648 0 : btrfs_put_block_group(cache);
3649 0 : return ret;
3650 : }
3651 :
3652 2 : static int chunk_usage_filter(struct btrfs_fs_info *fs_info,
3653 : u64 chunk_offset, struct btrfs_balance_args *bargs)
3654 : {
3655 2 : struct btrfs_block_group *cache;
3656 2 : u64 chunk_used, user_thresh;
3657 2 : int ret = 1;
3658 :
3659 2 : cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3660 2 : chunk_used = cache->used;
3661 :
3662 2 : if (bargs->usage_min == 0)
3663 : user_thresh = 1;
3664 0 : else if (bargs->usage > 100)
3665 0 : user_thresh = cache->length;
3666 : else
3667 0 : user_thresh = mult_perc(cache->length, bargs->usage);
3668 :
3669 2 : if (chunk_used < user_thresh)
3670 2 : ret = 0;
3671 :
3672 2 : btrfs_put_block_group(cache);
3673 2 : return ret;
3674 : }
3675 :
3676 0 : static int chunk_devid_filter(struct extent_buffer *leaf,
3677 : struct btrfs_chunk *chunk,
3678 : struct btrfs_balance_args *bargs)
3679 : {
3680 0 : struct btrfs_stripe *stripe;
3681 0 : int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3682 0 : int i;
3683 :
3684 0 : for (i = 0; i < num_stripes; i++) {
3685 0 : stripe = btrfs_stripe_nr(chunk, i);
3686 0 : if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3687 : return 0;
3688 : }
3689 :
3690 : return 1;
3691 : }
3692 :
3693 65135 : static u64 calc_data_stripes(u64 type, int num_stripes)
3694 : {
3695 65135 : const int index = btrfs_bg_flags_to_raid_index(type);
3696 65135 : const int ncopies = btrfs_raid_array[index].ncopies;
3697 65135 : const int nparity = btrfs_raid_array[index].nparity;
3698 :
3699 65135 : return (num_stripes - nparity) / ncopies;
3700 : }
3701 :
3702 : /* [pstart, pend) */
3703 0 : static int chunk_drange_filter(struct extent_buffer *leaf,
3704 : struct btrfs_chunk *chunk,
3705 : struct btrfs_balance_args *bargs)
3706 : {
3707 0 : struct btrfs_stripe *stripe;
3708 0 : int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3709 0 : u64 stripe_offset;
3710 0 : u64 stripe_length;
3711 0 : u64 type;
3712 0 : int factor;
3713 0 : int i;
3714 :
3715 0 : if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3716 : return 0;
3717 :
3718 0 : type = btrfs_chunk_type(leaf, chunk);
3719 0 : factor = calc_data_stripes(type, num_stripes);
3720 :
3721 0 : for (i = 0; i < num_stripes; i++) {
3722 0 : stripe = btrfs_stripe_nr(chunk, i);
3723 0 : if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3724 0 : continue;
3725 :
3726 0 : stripe_offset = btrfs_stripe_offset(leaf, stripe);
3727 0 : stripe_length = btrfs_chunk_length(leaf, chunk);
3728 0 : stripe_length = div_u64(stripe_length, factor);
3729 :
3730 0 : if (stripe_offset < bargs->pend &&
3731 0 : stripe_offset + stripe_length > bargs->pstart)
3732 : return 0;
3733 : }
3734 :
3735 : return 1;
3736 : }
3737 :
3738 : /* [vstart, vend) */
3739 0 : static int chunk_vrange_filter(struct extent_buffer *leaf,
3740 : struct btrfs_chunk *chunk,
3741 : u64 chunk_offset,
3742 : struct btrfs_balance_args *bargs)
3743 : {
3744 0 : if (chunk_offset < bargs->vend &&
3745 0 : chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3746 : /* at least part of the chunk is inside this vrange */
3747 0 : return 0;
3748 :
3749 : return 1;
3750 : }
3751 :
3752 0 : static int chunk_stripes_range_filter(struct extent_buffer *leaf,
3753 : struct btrfs_chunk *chunk,
3754 : struct btrfs_balance_args *bargs)
3755 : {
3756 0 : int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3757 :
3758 0 : if (bargs->stripes_min <= num_stripes
3759 0 : && num_stripes <= bargs->stripes_max)
3760 0 : return 0;
3761 :
3762 : return 1;
3763 : }
3764 :
3765 0 : static int chunk_soft_convert_filter(u64 chunk_type,
3766 : struct btrfs_balance_args *bargs)
3767 : {
3768 0 : if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3769 : return 0;
3770 :
3771 0 : chunk_type = chunk_to_extended(chunk_type) &
3772 : BTRFS_EXTENDED_PROFILE_MASK;
3773 :
3774 0 : if (bargs->target == chunk_type)
3775 0 : return 1;
3776 :
3777 : return 0;
3778 : }
3779 :
3780 2158 : static int should_balance_chunk(struct extent_buffer *leaf,
3781 : struct btrfs_chunk *chunk, u64 chunk_offset)
3782 : {
3783 2158 : struct btrfs_fs_info *fs_info = leaf->fs_info;
3784 2158 : struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3785 2158 : struct btrfs_balance_args *bargs = NULL;
3786 2158 : u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3787 :
3788 : /* type filter */
3789 2158 : if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3790 2158 : (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3791 : return 0;
3792 : }
3793 :
3794 1048 : if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3795 456 : bargs = &bctl->data;
3796 592 : else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3797 180 : bargs = &bctl->sys;
3798 412 : else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3799 412 : bargs = &bctl->meta;
3800 :
3801 : /* profiles filter */
3802 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3803 : chunk_profiles_filter(chunk_type, bargs)) {
3804 : return 0;
3805 : }
3806 :
3807 : /* usage filter */
3808 1050 : if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3809 2 : chunk_usage_filter(fs_info, chunk_offset, bargs)) {
3810 : return 0;
3811 1048 : } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) &&
3812 0 : chunk_usage_range_filter(fs_info, chunk_offset, bargs)) {
3813 : return 0;
3814 : }
3815 :
3816 : /* devid filter */
3817 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3818 0 : chunk_devid_filter(leaf, chunk, bargs)) {
3819 : return 0;
3820 : }
3821 :
3822 : /* drange filter, makes sense only with devid filter */
3823 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3824 0 : chunk_drange_filter(leaf, chunk, bargs)) {
3825 : return 0;
3826 : }
3827 :
3828 : /* vrange filter */
3829 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3830 0 : chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3831 : return 0;
3832 : }
3833 :
3834 : /* stripes filter */
3835 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) &&
3836 0 : chunk_stripes_range_filter(leaf, chunk, bargs)) {
3837 : return 0;
3838 : }
3839 :
3840 : /* soft profile changing mode */
3841 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3842 0 : chunk_soft_convert_filter(chunk_type, bargs)) {
3843 : return 0;
3844 : }
3845 :
3846 : /*
3847 : * limited by count, must be the last filter
3848 : */
3849 1048 : if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3850 0 : if (bargs->limit == 0)
3851 : return 0;
3852 : else
3853 0 : bargs->limit--;
3854 1048 : } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) {
3855 : /*
3856 : * Same logic as the 'limit' filter; the minimum cannot be
3857 : * determined here because we do not have the global information
3858 : * about the count of all chunks that satisfy the filters.
3859 : */
3860 0 : if (bargs->limit_max == 0)
3861 : return 0;
3862 : else
3863 0 : bargs->limit_max--;
3864 : }
3865 :
3866 : return 1;
3867 : }
3868 :
3869 191 : static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3870 : {
3871 191 : struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3872 191 : struct btrfs_root *chunk_root = fs_info->chunk_root;
3873 191 : u64 chunk_type;
3874 191 : struct btrfs_chunk *chunk;
3875 191 : struct btrfs_path *path = NULL;
3876 191 : struct btrfs_key key;
3877 191 : struct btrfs_key found_key;
3878 191 : struct extent_buffer *leaf;
3879 191 : int slot;
3880 191 : int ret;
3881 191 : int enospc_errors = 0;
3882 191 : bool counting = true;
3883 : /* The single value limit and min/max limits use the same bytes in the */
3884 191 : u64 limit_data = bctl->data.limit;
3885 191 : u64 limit_meta = bctl->meta.limit;
3886 191 : u64 limit_sys = bctl->sys.limit;
3887 191 : u32 count_data = 0;
3888 191 : u32 count_meta = 0;
3889 191 : u32 count_sys = 0;
3890 191 : int chunk_reserved = 0;
3891 :
3892 191 : path = btrfs_alloc_path();
3893 191 : if (!path) {
3894 0 : ret = -ENOMEM;
3895 0 : goto error;
3896 : }
3897 :
3898 : /* zero out stat counters */
3899 191 : spin_lock(&fs_info->balance_lock);
3900 191 : memset(&bctl->stat, 0, sizeof(bctl->stat));
3901 191 : spin_unlock(&fs_info->balance_lock);
3902 382 : again:
3903 382 : if (!counting) {
3904 : /*
3905 : * The single value limit and min/max limits use the same bytes
3906 : * in the
3907 : */
3908 191 : bctl->data.limit = limit_data;
3909 191 : bctl->meta.limit = limit_meta;
3910 191 : bctl->sys.limit = limit_sys;
3911 : }
3912 382 : key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3913 382 : key.offset = (u64)-1;
3914 382 : key.type = BTRFS_CHUNK_ITEM_KEY;
3915 :
3916 4694 : while (1) {
3917 2538 : if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3918 : atomic_read(&fs_info->balance_cancel_req)) {
3919 0 : ret = -ECANCELED;
3920 0 : goto error;
3921 : }
3922 :
3923 2538 : mutex_lock(&fs_info->reclaim_bgs_lock);
3924 2538 : ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3925 2538 : if (ret < 0) {
3926 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3927 0 : goto error;
3928 : }
3929 :
3930 : /*
3931 : * this shouldn't happen, it means the last relocate
3932 : * failed
3933 : */
3934 2538 : if (ret == 0)
3935 0 : BUG(); /* FIXME break ? */
3936 :
3937 2538 : ret = btrfs_previous_item(chunk_root, path, 0,
3938 : BTRFS_CHUNK_ITEM_KEY);
3939 2538 : if (ret) {
3940 380 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3941 380 : ret = 0;
3942 380 : break;
3943 : }
3944 :
3945 2158 : leaf = path->nodes[0];
3946 2158 : slot = path->slots[0];
3947 2158 : btrfs_item_key_to_cpu(leaf, &found_key, slot);
3948 :
3949 2158 : if (found_key.objectid != key.objectid) {
3950 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3951 0 : break;
3952 : }
3953 :
3954 2158 : chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3955 2158 : chunk_type = btrfs_chunk_type(leaf, chunk);
3956 :
3957 2158 : if (!counting) {
3958 1076 : spin_lock(&fs_info->balance_lock);
3959 1076 : bctl->stat.considered++;
3960 1076 : spin_unlock(&fs_info->balance_lock);
3961 : }
3962 :
3963 2158 : ret = should_balance_chunk(leaf, chunk, found_key.offset);
3964 :
3965 2158 : btrfs_release_path(path);
3966 2158 : if (!ret) {
3967 1110 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3968 1110 : goto loop;
3969 : }
3970 :
3971 1048 : if (counting) {
3972 525 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3973 525 : spin_lock(&fs_info->balance_lock);
3974 525 : bctl->stat.expected++;
3975 525 : spin_unlock(&fs_info->balance_lock);
3976 :
3977 525 : if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3978 229 : count_data++;
3979 296 : else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3980 90 : count_sys++;
3981 206 : else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3982 206 : count_meta++;
3983 :
3984 525 : goto loop;
3985 : }
3986 :
3987 : /*
3988 : * Apply limit_min filter, no need to check if the LIMITS
3989 : * filter is used, limit_min is 0 by default
3990 : */
3991 523 : if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) &&
3992 227 : count_data < bctl->data.limit_min)
3993 523 : || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) &&
3994 206 : count_meta < bctl->meta.limit_min)
3995 523 : || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) &&
3996 90 : count_sys < bctl->sys.limit_min)) {
3997 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
3998 0 : goto loop;
3999 : }
4000 :
4001 523 : if (!chunk_reserved) {
4002 : /*
4003 : * We may be relocating the only data chunk we have,
4004 : * which could potentially end up with losing data's
4005 : * raid profile, so lets allocate an empty one in
4006 : * advance.
4007 : */
4008 503 : ret = btrfs_may_alloc_data_chunk(fs_info,
4009 : found_key.offset);
4010 503 : if (ret < 0) {
4011 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4012 0 : goto error;
4013 503 : } else if (ret == 1) {
4014 23 : chunk_reserved = 1;
4015 : }
4016 : }
4017 :
4018 523 : ret = btrfs_relocate_chunk(fs_info, found_key.offset);
4019 523 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4020 523 : if (ret == -ENOSPC) {
4021 0 : enospc_errors++;
4022 523 : } else if (ret == -ETXTBSY) {
4023 1 : btrfs_info(fs_info,
4024 : "skipping relocation of block group %llu due to active swapfile",
4025 : found_key.offset);
4026 1 : ret = 0;
4027 522 : } else if (ret) {
4028 2 : goto error;
4029 : } else {
4030 520 : spin_lock(&fs_info->balance_lock);
4031 520 : bctl->stat.completed++;
4032 520 : spin_unlock(&fs_info->balance_lock);
4033 : }
4034 2156 : loop:
4035 2156 : if (found_key.offset == 0)
4036 : break;
4037 2156 : key.offset = found_key.offset - 1;
4038 : }
4039 :
4040 380 : if (counting) {
4041 191 : btrfs_release_path(path);
4042 191 : counting = false;
4043 191 : goto again;
4044 : }
4045 189 : error:
4046 191 : btrfs_free_path(path);
4047 191 : if (enospc_errors) {
4048 0 : btrfs_info(fs_info, "%d enospc errors during balance",
4049 : enospc_errors);
4050 0 : if (!ret)
4051 0 : ret = -ENOSPC;
4052 : }
4053 :
4054 191 : return ret;
4055 : }
4056 :
4057 : /*
4058 : * See if a given profile is valid and reduced.
4059 : *
4060 : * @flags: profile to validate
4061 : * @extended: if true @flags is treated as an extended profile
4062 : */
4063 : static int alloc_profile_is_valid(u64 flags, int extended)
4064 : {
4065 1520 : u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
4066 : BTRFS_BLOCK_GROUP_PROFILE_MASK);
4067 :
4068 1520 : flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
4069 :
4070 : /* 1) check that all other bits are zeroed */
4071 1520 : if (flags & ~mask)
4072 : return 0;
4073 :
4074 : /* 2) see if profile is reduced */
4075 1520 : if (flags == 0)
4076 : return !extended; /* "0" is valid for usual profiles */
4077 :
4078 336 : return has_single_bit_set(flags);
4079 : }
4080 :
4081 : /*
4082 : * Validate target profile against allowed profiles and return true if it's OK.
4083 : * Otherwise print the error message and return false.
4084 : */
4085 573 : static inline int validate_convert_profile(struct btrfs_fs_info *fs_info,
4086 : const struct btrfs_balance_args *bargs,
4087 : u64 allowed, const char *type)
4088 : {
4089 573 : if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
4090 : return true;
4091 :
4092 : /* Profile is valid and does not have bits outside of the allowed set */
4093 0 : if (alloc_profile_is_valid(bargs->target, 1) &&
4094 0 : (bargs->target & ~allowed) == 0)
4095 : return true;
4096 :
4097 0 : btrfs_err(fs_info, "balance: invalid convert %s profile %s",
4098 : type, btrfs_bg_type_to_raid_name(bargs->target));
4099 0 : return false;
4100 : }
4101 :
4102 : /*
4103 : * Fill @buf with textual description of balance filter flags @bargs, up to
4104 : * @size_buf including the terminating null. The output may be trimmed if it
4105 : * does not fit into the provided buffer.
4106 : */
4107 310 : static void describe_balance_args(struct btrfs_balance_args *bargs, char *buf,
4108 : u32 size_buf)
4109 : {
4110 310 : int ret;
4111 310 : u32 size_bp = size_buf;
4112 310 : char *bp = buf;
4113 310 : u64 flags = bargs->flags;
4114 310 : char tmp_buf[128] = {'\0'};
4115 :
4116 310 : if (!flags)
4117 309 : return;
4118 :
4119 : #define CHECK_APPEND_NOARG(a) \
4120 : do { \
4121 : ret = snprintf(bp, size_bp, (a)); \
4122 : if (ret < 0 || ret >= size_bp) \
4123 : goto out_overflow; \
4124 : size_bp -= ret; \
4125 : bp += ret; \
4126 : } while (0)
4127 :
4128 : #define CHECK_APPEND_1ARG(a, v1) \
4129 : do { \
4130 : ret = snprintf(bp, size_bp, (a), (v1)); \
4131 : if (ret < 0 || ret >= size_bp) \
4132 : goto out_overflow; \
4133 : size_bp -= ret; \
4134 : bp += ret; \
4135 : } while (0)
4136 :
4137 : #define CHECK_APPEND_2ARG(a, v1, v2) \
4138 : do { \
4139 : ret = snprintf(bp, size_bp, (a), (v1), (v2)); \
4140 : if (ret < 0 || ret >= size_bp) \
4141 : goto out_overflow; \
4142 : size_bp -= ret; \
4143 : bp += ret; \
4144 : } while (0)
4145 :
4146 1 : if (flags & BTRFS_BALANCE_ARGS_CONVERT)
4147 0 : CHECK_APPEND_1ARG("convert=%s,",
4148 : btrfs_bg_type_to_raid_name(bargs->target));
4149 :
4150 1 : if (flags & BTRFS_BALANCE_ARGS_SOFT)
4151 0 : CHECK_APPEND_NOARG("soft,");
4152 :
4153 1 : if (flags & BTRFS_BALANCE_ARGS_PROFILES) {
4154 0 : btrfs_describe_block_groups(bargs->profiles, tmp_buf,
4155 : sizeof(tmp_buf));
4156 0 : CHECK_APPEND_1ARG("profiles=%s,", tmp_buf);
4157 : }
4158 :
4159 1 : if (flags & BTRFS_BALANCE_ARGS_USAGE)
4160 1 : CHECK_APPEND_1ARG("usage=%llu,", bargs->usage);
4161 :
4162 1 : if (flags & BTRFS_BALANCE_ARGS_USAGE_RANGE)
4163 0 : CHECK_APPEND_2ARG("usage=%u..%u,",
4164 : bargs->usage_min, bargs->usage_max);
4165 :
4166 1 : if (flags & BTRFS_BALANCE_ARGS_DEVID)
4167 0 : CHECK_APPEND_1ARG("devid=%llu,", bargs->devid);
4168 :
4169 1 : if (flags & BTRFS_BALANCE_ARGS_DRANGE)
4170 0 : CHECK_APPEND_2ARG("drange=%llu..%llu,",
4171 : bargs->pstart, bargs->pend);
4172 :
4173 1 : if (flags & BTRFS_BALANCE_ARGS_VRANGE)
4174 0 : CHECK_APPEND_2ARG("vrange=%llu..%llu,",
4175 : bargs->vstart, bargs->vend);
4176 :
4177 1 : if (flags & BTRFS_BALANCE_ARGS_LIMIT)
4178 0 : CHECK_APPEND_1ARG("limit=%llu,", bargs->limit);
4179 :
4180 1 : if (flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)
4181 0 : CHECK_APPEND_2ARG("limit=%u..%u,",
4182 : bargs->limit_min, bargs->limit_max);
4183 :
4184 1 : if (flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE)
4185 0 : CHECK_APPEND_2ARG("stripes=%u..%u,",
4186 : bargs->stripes_min, bargs->stripes_max);
4187 :
4188 : #undef CHECK_APPEND_2ARG
4189 : #undef CHECK_APPEND_1ARG
4190 : #undef CHECK_APPEND_NOARG
4191 :
4192 1 : out_overflow:
4193 :
4194 1 : if (size_bp < size_buf)
4195 1 : buf[size_buf - size_bp - 1] = '\0'; /* remove last , */
4196 : else
4197 0 : buf[0] = '\0';
4198 : }
4199 :
4200 191 : static void describe_balance_start_or_resume(struct btrfs_fs_info *fs_info)
4201 : {
4202 191 : u32 size_buf = 1024;
4203 191 : char tmp_buf[192] = {'\0'};
4204 191 : char *buf;
4205 191 : char *bp;
4206 191 : u32 size_bp = size_buf;
4207 191 : int ret;
4208 191 : struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4209 :
4210 191 : buf = kzalloc(size_buf, GFP_KERNEL);
4211 191 : if (!buf)
4212 0 : return;
4213 :
4214 191 : bp = buf;
4215 :
4216 : #define CHECK_APPEND_1ARG(a, v1) \
4217 : do { \
4218 : ret = snprintf(bp, size_bp, (a), (v1)); \
4219 : if (ret < 0 || ret >= size_bp) \
4220 : goto out_overflow; \
4221 : size_bp -= ret; \
4222 : bp += ret; \
4223 : } while (0)
4224 :
4225 191 : if (bctl->flags & BTRFS_BALANCE_FORCE)
4226 58 : CHECK_APPEND_1ARG("%s", "-f ");
4227 :
4228 191 : if (bctl->flags & BTRFS_BALANCE_DATA) {
4229 130 : describe_balance_args(&bctl->data, tmp_buf, sizeof(tmp_buf));
4230 130 : CHECK_APPEND_1ARG("-d%s ", tmp_buf);
4231 : }
4232 :
4233 191 : if (bctl->flags & BTRFS_BALANCE_METADATA) {
4234 90 : describe_balance_args(&bctl->meta, tmp_buf, sizeof(tmp_buf));
4235 90 : CHECK_APPEND_1ARG("-m%s ", tmp_buf);
4236 : }
4237 :
4238 191 : if (bctl->flags & BTRFS_BALANCE_SYSTEM) {
4239 90 : describe_balance_args(&bctl->sys, tmp_buf, sizeof(tmp_buf));
4240 90 : CHECK_APPEND_1ARG("-s%s ", tmp_buf);
4241 : }
4242 :
4243 : #undef CHECK_APPEND_1ARG
4244 :
4245 101 : out_overflow:
4246 :
4247 191 : if (size_bp < size_buf)
4248 191 : buf[size_buf - size_bp - 1] = '\0'; /* remove last " " */
4249 382 : btrfs_info(fs_info, "balance: %s %s",
4250 : (bctl->flags & BTRFS_BALANCE_RESUME) ?
4251 : "resume" : "start", buf);
4252 :
4253 191 : kfree(buf);
4254 : }
4255 :
4256 : /*
4257 : * Should be called with balance mutexe held
4258 : */
4259 191 : int btrfs_balance(struct btrfs_fs_info *fs_info,
4260 : struct btrfs_balance_control *bctl,
4261 : struct btrfs_ioctl_balance_args *bargs)
4262 : {
4263 191 : u64 meta_target, data_target;
4264 191 : u64 allowed;
4265 191 : int mixed = 0;
4266 191 : int ret;
4267 191 : u64 num_devices;
4268 191 : unsigned seq;
4269 191 : bool reducing_redundancy;
4270 191 : bool paused = false;
4271 191 : int i;
4272 :
4273 191 : if (btrfs_fs_closing(fs_info) ||
4274 191 : atomic_read(&fs_info->balance_pause_req) ||
4275 191 : btrfs_should_cancel_balance(fs_info)) {
4276 0 : ret = -EINVAL;
4277 0 : goto out;
4278 : }
4279 :
4280 191 : allowed = btrfs_super_incompat_flags(fs_info->super_copy);
4281 191 : if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
4282 0 : mixed = 1;
4283 :
4284 : /*
4285 : * In case of mixed groups both data and meta should be picked,
4286 : * and identical options should be given for both of them.
4287 : */
4288 0 : allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
4289 0 : if (mixed && (bctl->flags & allowed)) {
4290 0 : if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
4291 0 : !(bctl->flags & BTRFS_BALANCE_METADATA) ||
4292 0 : memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
4293 0 : btrfs_err(fs_info,
4294 : "balance: mixed groups data and metadata options must be the same");
4295 0 : ret = -EINVAL;
4296 0 : goto out;
4297 : }
4298 : }
4299 :
4300 : /*
4301 : * rw_devices will not change at the moment, device add/delete/replace
4302 : * are exclusive
4303 : */
4304 191 : num_devices = fs_info->fs_devices->rw_devices;
4305 :
4306 : /*
4307 : * SINGLE profile on-disk has no profile bit, but in-memory we have a
4308 : * special bit for it, to make it easier to distinguish. Thus we need
4309 : * to set it manually, or balance would refuse the profile.
4310 : */
4311 191 : allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
4312 1910 : for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++)
4313 1719 : if (num_devices >= btrfs_raid_array[i].devs_min)
4314 573 : allowed |= btrfs_raid_array[i].bg_flag;
4315 :
4316 382 : if (!validate_convert_profile(fs_info, &bctl->data, allowed, "data") ||
4317 382 : !validate_convert_profile(fs_info, &bctl->meta, allowed, "metadata") ||
4318 191 : !validate_convert_profile(fs_info, &bctl->sys, allowed, "system")) {
4319 0 : ret = -EINVAL;
4320 0 : goto out;
4321 : }
4322 :
4323 : /*
4324 : * Allow to reduce metadata or system integrity only if force set for
4325 : * profiles with redundancy (copies, parity)
4326 : */
4327 : allowed = 0;
4328 1910 : for (i = 0; i < ARRAY_SIZE(btrfs_raid_array); i++) {
4329 1719 : if (btrfs_raid_array[i].ncopies >= 2 ||
4330 764 : btrfs_raid_array[i].tolerated_failures >= 1)
4331 1337 : allowed |= btrfs_raid_array[i].bg_flag;
4332 : }
4333 191 : do {
4334 191 : seq = read_seqbegin(&fs_info->profiles_lock);
4335 :
4336 191 : if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4337 0 : (fs_info->avail_system_alloc_bits & allowed) &&
4338 0 : !(bctl->sys.target & allowed)) ||
4339 191 : ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
4340 0 : (fs_info->avail_metadata_alloc_bits & allowed) &&
4341 0 : !(bctl->meta.target & allowed)))
4342 : reducing_redundancy = true;
4343 : else
4344 191 : reducing_redundancy = false;
4345 :
4346 : /* if we're not converting, the target field is uninitialized */
4347 191 : meta_target = (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4348 191 : bctl->meta.target : fs_info->avail_metadata_alloc_bits;
4349 191 : data_target = (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) ?
4350 191 : bctl->data.target : fs_info->avail_data_alloc_bits;
4351 191 : } while (read_seqretry(&fs_info->profiles_lock, seq));
4352 :
4353 191 : if (reducing_redundancy) {
4354 0 : if (bctl->flags & BTRFS_BALANCE_FORCE) {
4355 0 : btrfs_info(fs_info,
4356 : "balance: force reducing metadata redundancy");
4357 : } else {
4358 0 : btrfs_err(fs_info,
4359 : "balance: reduces metadata redundancy, use --force if you want this");
4360 0 : ret = -EINVAL;
4361 0 : goto out;
4362 : }
4363 : }
4364 :
4365 382 : if (btrfs_get_num_tolerated_disk_barrier_failures(meta_target) <
4366 191 : btrfs_get_num_tolerated_disk_barrier_failures(data_target)) {
4367 0 : btrfs_warn(fs_info,
4368 : "balance: metadata profile %s has lower redundancy than data profile %s",
4369 : btrfs_bg_type_to_raid_name(meta_target),
4370 : btrfs_bg_type_to_raid_name(data_target));
4371 : }
4372 :
4373 191 : ret = insert_balance_item(fs_info, bctl);
4374 191 : if (ret && ret != -EEXIST)
4375 0 : goto out;
4376 :
4377 191 : if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
4378 191 : BUG_ON(ret == -EEXIST);
4379 191 : BUG_ON(fs_info->balance_ctl);
4380 191 : spin_lock(&fs_info->balance_lock);
4381 191 : fs_info->balance_ctl = bctl;
4382 191 : spin_unlock(&fs_info->balance_lock);
4383 : } else {
4384 0 : BUG_ON(ret != -EEXIST);
4385 0 : spin_lock(&fs_info->balance_lock);
4386 0 : update_balance_args(bctl);
4387 0 : spin_unlock(&fs_info->balance_lock);
4388 : }
4389 :
4390 191 : ASSERT(!test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4391 191 : set_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4392 191 : describe_balance_start_or_resume(fs_info);
4393 191 : mutex_unlock(&fs_info->balance_mutex);
4394 :
4395 191 : ret = __btrfs_balance(fs_info);
4396 :
4397 191 : mutex_lock(&fs_info->balance_mutex);
4398 191 : if (ret == -ECANCELED && atomic_read(&fs_info->balance_pause_req)) {
4399 0 : btrfs_info(fs_info, "balance: paused");
4400 0 : btrfs_exclop_balance(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED);
4401 0 : paused = true;
4402 : }
4403 : /*
4404 : * Balance can be canceled by:
4405 : *
4406 : * - Regular cancel request
4407 : * Then ret == -ECANCELED and balance_cancel_req > 0
4408 : *
4409 : * - Fatal signal to "btrfs" process
4410 : * Either the signal caught by wait_reserve_ticket() and callers
4411 : * got -EINTR, or caught by btrfs_should_cancel_balance() and
4412 : * got -ECANCELED.
4413 : * Either way, in this case balance_cancel_req = 0, and
4414 : * ret == -EINTR or ret == -ECANCELED.
4415 : *
4416 : * So here we only check the return value to catch canceled balance.
4417 : */
4418 191 : else if (ret == -ECANCELED || ret == -EINTR)
4419 2 : btrfs_info(fs_info, "balance: canceled");
4420 : else
4421 189 : btrfs_info(fs_info, "balance: ended with status: %d", ret);
4422 :
4423 191 : clear_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags);
4424 :
4425 191 : if (bargs) {
4426 191 : memset(bargs, 0, sizeof(*bargs));
4427 191 : btrfs_update_ioctl_balance_args(fs_info, bargs);
4428 : }
4429 :
4430 : /* We didn't pause, we can clean everything up. */
4431 191 : if (!paused) {
4432 191 : reset_balance_state(fs_info);
4433 191 : btrfs_exclop_finish(fs_info);
4434 : }
4435 :
4436 191 : wake_up(&fs_info->balance_wait_q);
4437 :
4438 191 : return ret;
4439 0 : out:
4440 0 : if (bctl->flags & BTRFS_BALANCE_RESUME)
4441 0 : reset_balance_state(fs_info);
4442 : else
4443 0 : kfree(bctl);
4444 0 : btrfs_exclop_finish(fs_info);
4445 :
4446 0 : return ret;
4447 : }
4448 :
4449 0 : static int balance_kthread(void *data)
4450 : {
4451 0 : struct btrfs_fs_info *fs_info = data;
4452 0 : int ret = 0;
4453 :
4454 0 : sb_start_write(fs_info->sb);
4455 0 : mutex_lock(&fs_info->balance_mutex);
4456 0 : if (fs_info->balance_ctl)
4457 0 : ret = btrfs_balance(fs_info, fs_info->balance_ctl, NULL);
4458 0 : mutex_unlock(&fs_info->balance_mutex);
4459 0 : sb_end_write(fs_info->sb);
4460 :
4461 0 : return ret;
4462 : }
4463 :
4464 3181 : int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
4465 : {
4466 3181 : struct task_struct *tsk;
4467 :
4468 3181 : mutex_lock(&fs_info->balance_mutex);
4469 3181 : if (!fs_info->balance_ctl) {
4470 3181 : mutex_unlock(&fs_info->balance_mutex);
4471 3181 : return 0;
4472 : }
4473 0 : mutex_unlock(&fs_info->balance_mutex);
4474 :
4475 0 : if (btrfs_test_opt(fs_info, SKIP_BALANCE)) {
4476 0 : btrfs_info(fs_info, "balance: resume skipped");
4477 0 : return 0;
4478 : }
4479 :
4480 0 : spin_lock(&fs_info->super_lock);
4481 0 : ASSERT(fs_info->exclusive_operation == BTRFS_EXCLOP_BALANCE_PAUSED);
4482 0 : fs_info->exclusive_operation = BTRFS_EXCLOP_BALANCE;
4483 0 : spin_unlock(&fs_info->super_lock);
4484 : /*
4485 : * A ro->rw remount sequence should continue with the paused balance
4486 : * regardless of who pauses it, system or the user as of now, so set
4487 : * the resume flag.
4488 : */
4489 0 : spin_lock(&fs_info->balance_lock);
4490 0 : fs_info->balance_ctl->flags |= BTRFS_BALANCE_RESUME;
4491 0 : spin_unlock(&fs_info->balance_lock);
4492 :
4493 0 : tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
4494 0 : return PTR_ERR_OR_ZERO(tsk);
4495 : }
4496 :
4497 3217 : int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
4498 : {
4499 3217 : struct btrfs_balance_control *bctl;
4500 3217 : struct btrfs_balance_item *item;
4501 3217 : struct btrfs_disk_balance_args disk_bargs;
4502 3217 : struct btrfs_path *path;
4503 3217 : struct extent_buffer *leaf;
4504 3217 : struct btrfs_key key;
4505 3217 : int ret;
4506 :
4507 3217 : path = btrfs_alloc_path();
4508 3217 : if (!path)
4509 : return -ENOMEM;
4510 :
4511 3217 : key.objectid = BTRFS_BALANCE_OBJECTID;
4512 3217 : key.type = BTRFS_TEMPORARY_ITEM_KEY;
4513 3217 : key.offset = 0;
4514 :
4515 3217 : ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4516 3217 : if (ret < 0)
4517 0 : goto out;
4518 3217 : if (ret > 0) { /* ret = -ENOENT; */
4519 3217 : ret = 0;
4520 3217 : goto out;
4521 : }
4522 :
4523 0 : bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
4524 0 : if (!bctl) {
4525 0 : ret = -ENOMEM;
4526 0 : goto out;
4527 : }
4528 :
4529 0 : leaf = path->nodes[0];
4530 0 : item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
4531 :
4532 0 : bctl->flags = btrfs_balance_flags(leaf, item);
4533 0 : bctl->flags |= BTRFS_BALANCE_RESUME;
4534 :
4535 0 : btrfs_balance_data(leaf, item, &disk_bargs);
4536 0 : btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
4537 0 : btrfs_balance_meta(leaf, item, &disk_bargs);
4538 0 : btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
4539 0 : btrfs_balance_sys(leaf, item, &disk_bargs);
4540 0 : btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
4541 :
4542 : /*
4543 : * This should never happen, as the paused balance state is recovered
4544 : * during mount without any chance of other exclusive ops to collide.
4545 : *
4546 : * This gives the exclusive op status to balance and keeps in paused
4547 : * state until user intervention (cancel or umount). If the ownership
4548 : * cannot be assigned, show a message but do not fail. The balance
4549 : * is in a paused state and must have fs_info::balance_ctl properly
4550 : * set up.
4551 : */
4552 0 : if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE_PAUSED))
4553 0 : btrfs_warn(fs_info,
4554 : "balance: cannot set exclusive op status, resume manually");
4555 :
4556 0 : btrfs_release_path(path);
4557 :
4558 0 : mutex_lock(&fs_info->balance_mutex);
4559 0 : BUG_ON(fs_info->balance_ctl);
4560 0 : spin_lock(&fs_info->balance_lock);
4561 0 : fs_info->balance_ctl = bctl;
4562 0 : spin_unlock(&fs_info->balance_lock);
4563 0 : mutex_unlock(&fs_info->balance_mutex);
4564 3217 : out:
4565 3217 : btrfs_free_path(path);
4566 3217 : return ret;
4567 : }
4568 :
4569 3221 : int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
4570 : {
4571 3221 : int ret = 0;
4572 :
4573 3221 : mutex_lock(&fs_info->balance_mutex);
4574 3221 : if (!fs_info->balance_ctl) {
4575 3221 : mutex_unlock(&fs_info->balance_mutex);
4576 3221 : return -ENOTCONN;
4577 : }
4578 :
4579 0 : if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4580 0 : atomic_inc(&fs_info->balance_pause_req);
4581 0 : mutex_unlock(&fs_info->balance_mutex);
4582 :
4583 0 : wait_event(fs_info->balance_wait_q,
4584 : !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4585 :
4586 0 : mutex_lock(&fs_info->balance_mutex);
4587 : /* we are good with balance_ctl ripped off from under us */
4588 0 : BUG_ON(test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4589 0 : atomic_dec(&fs_info->balance_pause_req);
4590 : } else {
4591 : ret = -ENOTCONN;
4592 : }
4593 :
4594 0 : mutex_unlock(&fs_info->balance_mutex);
4595 0 : return ret;
4596 : }
4597 :
4598 12 : int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
4599 : {
4600 12 : mutex_lock(&fs_info->balance_mutex);
4601 12 : if (!fs_info->balance_ctl) {
4602 10 : mutex_unlock(&fs_info->balance_mutex);
4603 10 : return -ENOTCONN;
4604 : }
4605 :
4606 : /*
4607 : * A paused balance with the item stored on disk can be resumed at
4608 : * mount time if the mount is read-write. Otherwise it's still paused
4609 : * and we must not allow cancelling as it deletes the item.
4610 : */
4611 2 : if (sb_rdonly(fs_info->sb)) {
4612 0 : mutex_unlock(&fs_info->balance_mutex);
4613 0 : return -EROFS;
4614 : }
4615 :
4616 2 : atomic_inc(&fs_info->balance_cancel_req);
4617 : /*
4618 : * if we are running just wait and return, balance item is
4619 : * deleted in btrfs_balance in this case
4620 : */
4621 4 : if (test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags)) {
4622 2 : mutex_unlock(&fs_info->balance_mutex);
4623 4 : wait_event(fs_info->balance_wait_q,
4624 : !test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4625 2 : mutex_lock(&fs_info->balance_mutex);
4626 : } else {
4627 0 : mutex_unlock(&fs_info->balance_mutex);
4628 : /*
4629 : * Lock released to allow other waiters to continue, we'll
4630 : * reexamine the status again.
4631 : */
4632 0 : mutex_lock(&fs_info->balance_mutex);
4633 :
4634 0 : if (fs_info->balance_ctl) {
4635 0 : reset_balance_state(fs_info);
4636 0 : btrfs_exclop_finish(fs_info);
4637 0 : btrfs_info(fs_info, "balance: canceled");
4638 : }
4639 : }
4640 :
4641 2 : BUG_ON(fs_info->balance_ctl ||
4642 : test_bit(BTRFS_FS_BALANCE_RUNNING, &fs_info->flags));
4643 2 : atomic_dec(&fs_info->balance_cancel_req);
4644 2 : mutex_unlock(&fs_info->balance_mutex);
4645 2 : return 0;
4646 : }
4647 :
4648 995 : int btrfs_uuid_scan_kthread(void *data)
4649 : {
4650 995 : struct btrfs_fs_info *fs_info = data;
4651 995 : struct btrfs_root *root = fs_info->tree_root;
4652 995 : struct btrfs_key key;
4653 995 : struct btrfs_path *path = NULL;
4654 995 : int ret = 0;
4655 995 : struct extent_buffer *eb;
4656 995 : int slot;
4657 995 : struct btrfs_root_item root_item;
4658 995 : u32 item_size;
4659 995 : struct btrfs_trans_handle *trans = NULL;
4660 995 : bool closing = false;
4661 :
4662 995 : path = btrfs_alloc_path();
4663 995 : if (!path) {
4664 0 : ret = -ENOMEM;
4665 0 : goto out;
4666 : }
4667 :
4668 995 : key.objectid = 0;
4669 995 : key.type = BTRFS_ROOT_ITEM_KEY;
4670 995 : key.offset = 0;
4671 :
4672 12955 : while (1) {
4673 12955 : if (btrfs_fs_closing(fs_info)) {
4674 : closing = true;
4675 : break;
4676 : }
4677 12955 : ret = btrfs_search_forward(root, &key, path,
4678 : BTRFS_OLDEST_GENERATION);
4679 12955 : if (ret) {
4680 995 : if (ret > 0)
4681 : ret = 0;
4682 : break;
4683 : }
4684 :
4685 11960 : if (key.type != BTRFS_ROOT_ITEM_KEY ||
4686 7955 : (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
4687 2989 : key.objectid != BTRFS_FS_TREE_OBJECTID) ||
4688 : key.objectid > BTRFS_LAST_FREE_OBJECTID)
4689 9966 : goto skip;
4690 :
4691 1994 : eb = path->nodes[0];
4692 1994 : slot = path->slots[0];
4693 1994 : item_size = btrfs_item_size(eb, slot);
4694 1994 : if (item_size < sizeof(root_item))
4695 0 : goto skip;
4696 :
4697 1994 : read_extent_buffer(eb, &root_item,
4698 1994 : btrfs_item_ptr_offset(eb, slot),
4699 : (int)sizeof(root_item));
4700 1994 : if (btrfs_root_refs(&root_item) == 0)
4701 0 : goto skip;
4702 :
4703 1994 : if (!btrfs_is_empty_uuid(root_item.uuid) ||
4704 12 : !btrfs_is_empty_uuid(root_item.received_uuid)) {
4705 1982 : if (trans)
4706 991 : goto update_tree;
4707 :
4708 991 : btrfs_release_path(path);
4709 : /*
4710 : * 1 - subvol uuid item
4711 : * 1 - received_subvol uuid item
4712 : */
4713 991 : trans = btrfs_start_transaction(fs_info->uuid_root, 2);
4714 991 : if (IS_ERR(trans)) {
4715 0 : ret = PTR_ERR(trans);
4716 0 : break;
4717 : }
4718 991 : continue;
4719 : } else {
4720 12 : goto skip;
4721 : }
4722 : update_tree:
4723 991 : btrfs_release_path(path);
4724 991 : if (!btrfs_is_empty_uuid(root_item.uuid)) {
4725 991 : ret = btrfs_uuid_tree_add(trans, root_item.uuid,
4726 : BTRFS_UUID_KEY_SUBVOL,
4727 : key.objectid);
4728 991 : if (ret < 0) {
4729 0 : btrfs_warn(fs_info, "uuid_tree_add failed %d",
4730 : ret);
4731 0 : break;
4732 : }
4733 : }
4734 :
4735 991 : if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
4736 0 : ret = btrfs_uuid_tree_add(trans,
4737 : root_item.received_uuid,
4738 : BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4739 : key.objectid);
4740 0 : if (ret < 0) {
4741 0 : btrfs_warn(fs_info, "uuid_tree_add failed %d",
4742 : ret);
4743 0 : break;
4744 : }
4745 : }
4746 :
4747 991 : skip:
4748 10969 : btrfs_release_path(path);
4749 10969 : if (trans) {
4750 991 : ret = btrfs_end_transaction(trans);
4751 991 : trans = NULL;
4752 991 : if (ret)
4753 : break;
4754 : }
4755 :
4756 10969 : if (key.offset < (u64)-1) {
4757 10969 : key.offset++;
4758 0 : } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
4759 0 : key.offset = 0;
4760 0 : key.type = BTRFS_ROOT_ITEM_KEY;
4761 0 : } else if (key.objectid < (u64)-1) {
4762 0 : key.offset = 0;
4763 0 : key.type = BTRFS_ROOT_ITEM_KEY;
4764 0 : key.objectid++;
4765 : } else {
4766 : break;
4767 : }
4768 10969 : cond_resched();
4769 : }
4770 :
4771 0 : out:
4772 995 : btrfs_free_path(path);
4773 995 : if (trans && !IS_ERR(trans))
4774 0 : btrfs_end_transaction(trans);
4775 995 : if (ret)
4776 0 : btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
4777 995 : else if (!closing)
4778 995 : set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags);
4779 995 : up(&fs_info->uuid_tree_rescan_sem);
4780 995 : return 0;
4781 : }
4782 :
4783 3 : int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4784 : {
4785 3 : struct btrfs_trans_handle *trans;
4786 3 : struct btrfs_root *tree_root = fs_info->tree_root;
4787 3 : struct btrfs_root *uuid_root;
4788 3 : struct task_struct *task;
4789 3 : int ret;
4790 :
4791 : /*
4792 : * 1 - root node
4793 : * 1 - root item
4794 : */
4795 3 : trans = btrfs_start_transaction(tree_root, 2);
4796 3 : if (IS_ERR(trans))
4797 0 : return PTR_ERR(trans);
4798 :
4799 3 : uuid_root = btrfs_create_tree(trans, BTRFS_UUID_TREE_OBJECTID);
4800 3 : if (IS_ERR(uuid_root)) {
4801 0 : ret = PTR_ERR(uuid_root);
4802 0 : btrfs_abort_transaction(trans, ret);
4803 0 : btrfs_end_transaction(trans);
4804 0 : return ret;
4805 : }
4806 :
4807 3 : fs_info->uuid_root = uuid_root;
4808 :
4809 3 : ret = btrfs_commit_transaction(trans);
4810 3 : if (ret)
4811 : return ret;
4812 :
4813 3 : down(&fs_info->uuid_tree_rescan_sem);
4814 3 : task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4815 3 : if (IS_ERR(task)) {
4816 : /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4817 0 : btrfs_warn(fs_info, "failed to start uuid_scan task");
4818 0 : up(&fs_info->uuid_tree_rescan_sem);
4819 0 : return PTR_ERR(task);
4820 : }
4821 :
4822 : return 0;
4823 : }
4824 :
4825 : /*
4826 : * shrinking a device means finding all of the device extents past
4827 : * the new size, and then following the back refs to the chunks.
4828 : * The chunk relocation code actually frees the device extent
4829 : */
4830 3 : int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4831 : {
4832 3 : struct btrfs_fs_info *fs_info = device->fs_info;
4833 3 : struct btrfs_root *root = fs_info->dev_root;
4834 3 : struct btrfs_trans_handle *trans;
4835 3 : struct btrfs_dev_extent *dev_extent = NULL;
4836 3 : struct btrfs_path *path;
4837 3 : u64 length;
4838 3 : u64 chunk_offset;
4839 3 : int ret;
4840 3 : int slot;
4841 3 : int failed = 0;
4842 3 : bool retried = false;
4843 3 : struct extent_buffer *l;
4844 3 : struct btrfs_key key;
4845 3 : struct btrfs_super_block *super_copy = fs_info->super_copy;
4846 3 : u64 old_total = btrfs_super_total_bytes(super_copy);
4847 3 : u64 old_size = btrfs_device_get_total_bytes(device);
4848 3 : u64 diff;
4849 3 : u64 start;
4850 :
4851 3 : new_size = round_down(new_size, fs_info->sectorsize);
4852 3 : start = new_size;
4853 3 : diff = round_down(old_size - new_size, fs_info->sectorsize);
4854 :
4855 6 : if (test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
4856 : return -EINVAL;
4857 :
4858 3 : path = btrfs_alloc_path();
4859 3 : if (!path)
4860 : return -ENOMEM;
4861 :
4862 3 : path->reada = READA_BACK;
4863 :
4864 3 : trans = btrfs_start_transaction(root, 0);
4865 3 : if (IS_ERR(trans)) {
4866 0 : btrfs_free_path(path);
4867 0 : return PTR_ERR(trans);
4868 : }
4869 :
4870 3 : mutex_lock(&fs_info->chunk_mutex);
4871 :
4872 3 : btrfs_device_set_total_bytes(device, new_size);
4873 6 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
4874 3 : device->fs_devices->total_rw_bytes -= diff;
4875 3 : atomic64_sub(diff, &fs_info->free_chunk_space);
4876 : }
4877 :
4878 : /*
4879 : * Once the device's size has been set to the new size, ensure all
4880 : * in-memory chunks are synced to disk so that the loop below sees them
4881 : * and relocates them accordingly.
4882 : */
4883 3 : if (contains_pending_extent(device, &start, diff)) {
4884 2 : mutex_unlock(&fs_info->chunk_mutex);
4885 2 : ret = btrfs_commit_transaction(trans);
4886 2 : if (ret)
4887 0 : goto done;
4888 : } else {
4889 1 : mutex_unlock(&fs_info->chunk_mutex);
4890 1 : btrfs_end_transaction(trans);
4891 : }
4892 :
4893 : again:
4894 3 : key.objectid = device->devid;
4895 3 : key.offset = (u64)-1;
4896 3 : key.type = BTRFS_DEV_EXTENT_KEY;
4897 :
4898 4 : do {
4899 4 : mutex_lock(&fs_info->reclaim_bgs_lock);
4900 4 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4901 4 : if (ret < 0) {
4902 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4903 0 : goto done;
4904 : }
4905 :
4906 4 : ret = btrfs_previous_item(root, path, 0, key.type);
4907 4 : if (ret) {
4908 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4909 0 : if (ret < 0)
4910 0 : goto done;
4911 0 : ret = 0;
4912 0 : btrfs_release_path(path);
4913 0 : break;
4914 : }
4915 :
4916 4 : l = path->nodes[0];
4917 4 : slot = path->slots[0];
4918 4 : btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4919 :
4920 4 : if (key.objectid != device->devid) {
4921 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4922 0 : btrfs_release_path(path);
4923 0 : break;
4924 : }
4925 :
4926 4 : dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4927 4 : length = btrfs_dev_extent_length(l, dev_extent);
4928 :
4929 4 : if (key.offset + length <= new_size) {
4930 2 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4931 2 : btrfs_release_path(path);
4932 2 : break;
4933 : }
4934 :
4935 2 : chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4936 2 : btrfs_release_path(path);
4937 :
4938 : /*
4939 : * We may be relocating the only data chunk we have,
4940 : * which could potentially end up with losing data's
4941 : * raid profile, so lets allocate an empty one in
4942 : * advance.
4943 : */
4944 2 : ret = btrfs_may_alloc_data_chunk(fs_info, chunk_offset);
4945 2 : if (ret < 0) {
4946 0 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4947 0 : goto done;
4948 : }
4949 :
4950 2 : ret = btrfs_relocate_chunk(fs_info, chunk_offset);
4951 2 : mutex_unlock(&fs_info->reclaim_bgs_lock);
4952 2 : if (ret == -ENOSPC) {
4953 0 : failed++;
4954 2 : } else if (ret) {
4955 1 : if (ret == -ETXTBSY) {
4956 1 : btrfs_warn(fs_info,
4957 : "could not shrink block group %llu due to active swapfile",
4958 : chunk_offset);
4959 : }
4960 1 : goto done;
4961 : }
4962 1 : } while (key.offset-- > 0);
4963 :
4964 2 : if (failed && !retried) {
4965 0 : failed = 0;
4966 0 : retried = true;
4967 0 : goto again;
4968 2 : } else if (failed && retried) {
4969 0 : ret = -ENOSPC;
4970 0 : goto done;
4971 : }
4972 :
4973 : /* Shrinking succeeded, else we would be at "done". */
4974 2 : trans = btrfs_start_transaction(root, 0);
4975 2 : if (IS_ERR(trans)) {
4976 0 : ret = PTR_ERR(trans);
4977 0 : goto done;
4978 : }
4979 :
4980 2 : mutex_lock(&fs_info->chunk_mutex);
4981 : /* Clear all state bits beyond the shrunk device size */
4982 2 : clear_extent_bits(&device->alloc_state, new_size, (u64)-1,
4983 : CHUNK_STATE_MASK);
4984 :
4985 2 : btrfs_device_set_disk_total_bytes(device, new_size);
4986 2 : if (list_empty(&device->post_commit_list))
4987 2 : list_add_tail(&device->post_commit_list,
4988 2 : &trans->transaction->dev_update_list);
4989 :
4990 2 : WARN_ON(diff > old_total);
4991 2 : btrfs_set_super_total_bytes(super_copy,
4992 2 : round_down(old_total - diff, fs_info->sectorsize));
4993 2 : mutex_unlock(&fs_info->chunk_mutex);
4994 :
4995 2 : btrfs_reserve_chunk_metadata(trans, false);
4996 : /* Now btrfs_update_device() will change the on-disk size. */
4997 2 : ret = btrfs_update_device(trans, device);
4998 2 : btrfs_trans_release_chunk_metadata(trans);
4999 2 : if (ret < 0) {
5000 0 : btrfs_abort_transaction(trans, ret);
5001 0 : btrfs_end_transaction(trans);
5002 : } else {
5003 2 : ret = btrfs_commit_transaction(trans);
5004 : }
5005 3 : done:
5006 3 : btrfs_free_path(path);
5007 3 : if (ret) {
5008 1 : mutex_lock(&fs_info->chunk_mutex);
5009 1 : btrfs_device_set_total_bytes(device, old_size);
5010 2 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
5011 1 : device->fs_devices->total_rw_bytes += diff;
5012 1 : atomic64_add(diff, &fs_info->free_chunk_space);
5013 1 : mutex_unlock(&fs_info->chunk_mutex);
5014 : }
5015 : return ret;
5016 : }
5017 :
5018 93 : static int btrfs_add_system_chunk(struct btrfs_fs_info *fs_info,
5019 : struct btrfs_key *key,
5020 : struct btrfs_chunk *chunk, int item_size)
5021 : {
5022 93 : struct btrfs_super_block *super_copy = fs_info->super_copy;
5023 93 : struct btrfs_disk_key disk_key;
5024 93 : u32 array_size;
5025 93 : u8 *ptr;
5026 :
5027 93 : lockdep_assert_held(&fs_info->chunk_mutex);
5028 :
5029 93 : array_size = btrfs_super_sys_array_size(super_copy);
5030 93 : if (array_size + item_size + sizeof(disk_key)
5031 : > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
5032 : return -EFBIG;
5033 :
5034 93 : ptr = super_copy->sys_chunk_array + array_size;
5035 93 : btrfs_cpu_key_to_disk(&disk_key, key);
5036 186 : memcpy(ptr, &disk_key, sizeof(disk_key));
5037 93 : ptr += sizeof(disk_key);
5038 186 : memcpy(ptr, chunk, item_size);
5039 93 : item_size += sizeof(disk_key);
5040 93 : btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
5041 :
5042 93 : return 0;
5043 : }
5044 :
5045 : /*
5046 : * sort the devices in descending order by max_avail, total_avail
5047 : */
5048 0 : static int btrfs_cmp_device_info(const void *a, const void *b)
5049 : {
5050 0 : const struct btrfs_device_info *di_a = a;
5051 0 : const struct btrfs_device_info *di_b = b;
5052 :
5053 0 : if (di_a->max_avail > di_b->max_avail)
5054 : return -1;
5055 0 : if (di_a->max_avail < di_b->max_avail)
5056 : return 1;
5057 0 : if (di_a->total_avail > di_b->total_avail)
5058 : return -1;
5059 0 : if (di_a->total_avail < di_b->total_avail)
5060 0 : return 1;
5061 : return 0;
5062 : }
5063 :
5064 1470 : static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
5065 : {
5066 1470 : if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5067 : return;
5068 :
5069 0 : btrfs_set_fs_incompat(info, RAID56);
5070 : }
5071 :
5072 1470 : static void check_raid1c34_incompat_flag(struct btrfs_fs_info *info, u64 type)
5073 : {
5074 1470 : if (!(type & (BTRFS_BLOCK_GROUP_RAID1C3 | BTRFS_BLOCK_GROUP_RAID1C4)))
5075 : return;
5076 :
5077 0 : btrfs_set_fs_incompat(info, RAID1C34);
5078 : }
5079 :
5080 : /*
5081 : * Structure used internally for btrfs_create_chunk() function.
5082 : * Wraps needed parameters.
5083 : */
5084 : struct alloc_chunk_ctl {
5085 : u64 start;
5086 : u64 type;
5087 : /* Total number of stripes to allocate */
5088 : int num_stripes;
5089 : /* sub_stripes info for map */
5090 : int sub_stripes;
5091 : /* Stripes per device */
5092 : int dev_stripes;
5093 : /* Maximum number of devices to use */
5094 : int devs_max;
5095 : /* Minimum number of devices to use */
5096 : int devs_min;
5097 : /* ndevs has to be a multiple of this */
5098 : int devs_increment;
5099 : /* Number of copies */
5100 : int ncopies;
5101 : /* Number of stripes worth of bytes to store parity information */
5102 : int nparity;
5103 : u64 max_stripe_size;
5104 : u64 max_chunk_size;
5105 : u64 dev_extent_min;
5106 : u64 stripe_size;
5107 : u64 chunk_size;
5108 : int ndevs;
5109 : };
5110 :
5111 1520 : static void init_alloc_chunk_ctl_policy_regular(
5112 : struct btrfs_fs_devices *fs_devices,
5113 : struct alloc_chunk_ctl *ctl)
5114 : {
5115 1520 : struct btrfs_space_info *space_info;
5116 :
5117 1520 : space_info = btrfs_find_space_info(fs_devices->fs_info, ctl->type);
5118 1520 : ASSERT(space_info);
5119 :
5120 1520 : ctl->max_chunk_size = READ_ONCE(space_info->chunk_size);
5121 1520 : ctl->max_stripe_size = ctl->max_chunk_size;
5122 :
5123 1520 : if (ctl->type & BTRFS_BLOCK_GROUP_SYSTEM)
5124 93 : ctl->devs_max = min_t(int, ctl->devs_max, BTRFS_MAX_DEVS_SYS_CHUNK);
5125 :
5126 : /* We don't want a chunk larger than 10% of writable space */
5127 1520 : ctl->max_chunk_size = min(mult_perc(fs_devices->total_rw_bytes, 10),
5128 : ctl->max_chunk_size);
5129 1520 : ctl->dev_extent_min = btrfs_stripe_nr_to_offset(ctl->dev_stripes);
5130 1520 : }
5131 :
5132 0 : static void init_alloc_chunk_ctl_policy_zoned(
5133 : struct btrfs_fs_devices *fs_devices,
5134 : struct alloc_chunk_ctl *ctl)
5135 : {
5136 0 : u64 zone_size = fs_devices->fs_info->zone_size;
5137 0 : u64 limit;
5138 0 : int min_num_stripes = ctl->devs_min * ctl->dev_stripes;
5139 0 : int min_data_stripes = (min_num_stripes - ctl->nparity) / ctl->ncopies;
5140 0 : u64 min_chunk_size = min_data_stripes * zone_size;
5141 0 : u64 type = ctl->type;
5142 :
5143 0 : ctl->max_stripe_size = zone_size;
5144 0 : if (type & BTRFS_BLOCK_GROUP_DATA) {
5145 0 : ctl->max_chunk_size = round_down(BTRFS_MAX_DATA_CHUNK_SIZE,
5146 : zone_size);
5147 0 : } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
5148 0 : ctl->max_chunk_size = ctl->max_stripe_size;
5149 0 : } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
5150 0 : ctl->max_chunk_size = 2 * ctl->max_stripe_size;
5151 0 : ctl->devs_max = min_t(int, ctl->devs_max,
5152 : BTRFS_MAX_DEVS_SYS_CHUNK);
5153 : } else {
5154 0 : BUG();
5155 : }
5156 :
5157 : /* We don't want a chunk larger than 10% of writable space */
5158 0 : limit = max(round_down(mult_perc(fs_devices->total_rw_bytes, 10),
5159 : zone_size),
5160 : min_chunk_size);
5161 0 : ctl->max_chunk_size = min(limit, ctl->max_chunk_size);
5162 0 : ctl->dev_extent_min = zone_size * ctl->dev_stripes;
5163 0 : }
5164 :
5165 1520 : static void init_alloc_chunk_ctl(struct btrfs_fs_devices *fs_devices,
5166 : struct alloc_chunk_ctl *ctl)
5167 : {
5168 1520 : int index = btrfs_bg_flags_to_raid_index(ctl->type);
5169 :
5170 1520 : ctl->sub_stripes = btrfs_raid_array[index].sub_stripes;
5171 1520 : ctl->dev_stripes = btrfs_raid_array[index].dev_stripes;
5172 1520 : ctl->devs_max = btrfs_raid_array[index].devs_max;
5173 1520 : if (!ctl->devs_max)
5174 0 : ctl->devs_max = BTRFS_MAX_DEVS(fs_devices->fs_info);
5175 1520 : ctl->devs_min = btrfs_raid_array[index].devs_min;
5176 1520 : ctl->devs_increment = btrfs_raid_array[index].devs_increment;
5177 1520 : ctl->ncopies = btrfs_raid_array[index].ncopies;
5178 1520 : ctl->nparity = btrfs_raid_array[index].nparity;
5179 1520 : ctl->ndevs = 0;
5180 :
5181 1520 : switch (fs_devices->chunk_alloc_policy) {
5182 1520 : case BTRFS_CHUNK_ALLOC_REGULAR:
5183 1520 : init_alloc_chunk_ctl_policy_regular(fs_devices, ctl);
5184 1520 : break;
5185 0 : case BTRFS_CHUNK_ALLOC_ZONED:
5186 0 : init_alloc_chunk_ctl_policy_zoned(fs_devices, ctl);
5187 0 : break;
5188 0 : default:
5189 0 : BUG();
5190 : }
5191 1520 : }
5192 :
5193 1520 : static int gather_device_info(struct btrfs_fs_devices *fs_devices,
5194 : struct alloc_chunk_ctl *ctl,
5195 : struct btrfs_device_info *devices_info)
5196 : {
5197 1520 : struct btrfs_fs_info *info = fs_devices->fs_info;
5198 1520 : struct btrfs_device *device;
5199 1520 : u64 total_avail;
5200 1520 : u64 dev_extent_want = ctl->max_stripe_size * ctl->dev_stripes;
5201 1520 : int ret;
5202 1520 : int ndevs = 0;
5203 1520 : u64 max_avail;
5204 1520 : u64 dev_offset;
5205 :
5206 : /*
5207 : * in the first pass through the devices list, we gather information
5208 : * about the available holes on each device.
5209 : */
5210 3040 : list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
5211 1520 : if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) {
5212 0 : WARN(1, KERN_ERR
5213 : "BTRFS: read-only device in alloc_list\n");
5214 0 : continue;
5215 : }
5216 :
5217 3040 : if (!test_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
5218 3040 : &device->dev_state) ||
5219 0 : test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state))
5220 0 : continue;
5221 :
5222 1520 : if (device->total_bytes > device->bytes_used)
5223 1520 : total_avail = device->total_bytes - device->bytes_used;
5224 : else
5225 : total_avail = 0;
5226 :
5227 : /* If there is no space on this device, skip it. */
5228 1520 : if (total_avail < ctl->dev_extent_min)
5229 0 : continue;
5230 :
5231 1520 : ret = find_free_dev_extent(device, dev_extent_want, &dev_offset,
5232 : &max_avail);
5233 1520 : if (ret && ret != -ENOSPC)
5234 0 : return ret;
5235 :
5236 1520 : if (ret == 0)
5237 972 : max_avail = dev_extent_want;
5238 :
5239 1520 : if (max_avail < ctl->dev_extent_min) {
5240 50 : if (btrfs_test_opt(info, ENOSPC_DEBUG))
5241 : btrfs_debug(info,
5242 : "%s: devid %llu has no free space, have=%llu want=%llu",
5243 : __func__, device->devid, max_avail,
5244 : ctl->dev_extent_min);
5245 50 : continue;
5246 : }
5247 :
5248 1470 : if (ndevs == fs_devices->rw_devices) {
5249 0 : WARN(1, "%s: found more than %llu devices\n",
5250 : __func__, fs_devices->rw_devices);
5251 0 : break;
5252 : }
5253 1470 : devices_info[ndevs].dev_offset = dev_offset;
5254 1470 : devices_info[ndevs].max_avail = max_avail;
5255 1470 : devices_info[ndevs].total_avail = total_avail;
5256 1470 : devices_info[ndevs].dev = device;
5257 1470 : ++ndevs;
5258 : }
5259 1520 : ctl->ndevs = ndevs;
5260 :
5261 : /*
5262 : * now sort the devices by hole size / available space
5263 : */
5264 1520 : sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
5265 : btrfs_cmp_device_info, NULL);
5266 :
5267 1520 : return 0;
5268 : }
5269 :
5270 1470 : static int decide_stripe_size_regular(struct alloc_chunk_ctl *ctl,
5271 : struct btrfs_device_info *devices_info)
5272 : {
5273 : /* Number of stripes that count for block group size */
5274 1470 : int data_stripes;
5275 :
5276 : /*
5277 : * The primary goal is to maximize the number of stripes, so use as
5278 : * many devices as possible, even if the stripes are not maximum sized.
5279 : *
5280 : * The DUP profile stores more than one stripe per device, the
5281 : * max_avail is the total size so we have to adjust.
5282 : */
5283 1470 : ctl->stripe_size = div_u64(devices_info[ctl->ndevs - 1].max_avail,
5284 1470 : ctl->dev_stripes);
5285 1470 : ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5286 :
5287 : /* This will have to be fixed for RAID1 and RAID10 over more drives */
5288 1470 : data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5289 :
5290 : /*
5291 : * Use the number of data stripes to figure out how big this chunk is
5292 : * really going to be in terms of logical address space, and compare
5293 : * that answer with the max chunk size. If it's higher, we try to
5294 : * reduce stripe_size.
5295 : */
5296 1470 : if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5297 : /*
5298 : * Reduce stripe_size, round it up to a 16MB boundary again and
5299 : * then use it, unless it ends up being even bigger than the
5300 : * previous value we had already.
5301 : */
5302 1067 : ctl->stripe_size = min(round_up(div_u64(ctl->max_chunk_size,
5303 : data_stripes), SZ_16M),
5304 : ctl->stripe_size);
5305 : }
5306 :
5307 : /* Stripe size should not go beyond 1G. */
5308 1470 : ctl->stripe_size = min_t(u64, ctl->stripe_size, SZ_1G);
5309 :
5310 : /* Align to BTRFS_STRIPE_LEN */
5311 1470 : ctl->stripe_size = round_down(ctl->stripe_size, BTRFS_STRIPE_LEN);
5312 1470 : ctl->chunk_size = ctl->stripe_size * data_stripes;
5313 :
5314 1470 : return 0;
5315 : }
5316 :
5317 0 : static int decide_stripe_size_zoned(struct alloc_chunk_ctl *ctl,
5318 : struct btrfs_device_info *devices_info)
5319 : {
5320 0 : u64 zone_size = devices_info[0].dev->zone_info->zone_size;
5321 : /* Number of stripes that count for block group size */
5322 0 : int data_stripes;
5323 :
5324 : /*
5325 : * It should hold because:
5326 : * dev_extent_min == dev_extent_want == zone_size * dev_stripes
5327 : */
5328 0 : ASSERT(devices_info[ctl->ndevs - 1].max_avail == ctl->dev_extent_min);
5329 :
5330 0 : ctl->stripe_size = zone_size;
5331 0 : ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5332 0 : data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5333 :
5334 : /* stripe_size is fixed in zoned filesysmte. Reduce ndevs instead. */
5335 0 : if (ctl->stripe_size * data_stripes > ctl->max_chunk_size) {
5336 0 : ctl->ndevs = div_u64(div_u64(ctl->max_chunk_size * ctl->ncopies,
5337 0 : ctl->stripe_size) + ctl->nparity,
5338 : ctl->dev_stripes);
5339 0 : ctl->num_stripes = ctl->ndevs * ctl->dev_stripes;
5340 0 : data_stripes = (ctl->num_stripes - ctl->nparity) / ctl->ncopies;
5341 0 : ASSERT(ctl->stripe_size * data_stripes <= ctl->max_chunk_size);
5342 : }
5343 :
5344 0 : ctl->chunk_size = ctl->stripe_size * data_stripes;
5345 :
5346 0 : return 0;
5347 : }
5348 :
5349 1520 : static int decide_stripe_size(struct btrfs_fs_devices *fs_devices,
5350 : struct alloc_chunk_ctl *ctl,
5351 : struct btrfs_device_info *devices_info)
5352 : {
5353 1520 : struct btrfs_fs_info *info = fs_devices->fs_info;
5354 :
5355 : /*
5356 : * Round down to number of usable stripes, devs_increment can be any
5357 : * number so we can't use round_down() that requires power of 2, while
5358 : * rounddown is safe.
5359 : */
5360 1520 : ctl->ndevs = rounddown(ctl->ndevs, ctl->devs_increment);
5361 :
5362 1520 : if (ctl->ndevs < ctl->devs_min) {
5363 : if (btrfs_test_opt(info, ENOSPC_DEBUG)) {
5364 : btrfs_debug(info,
5365 : "%s: not enough devices with free space: have=%d minimum required=%d",
5366 : __func__, ctl->ndevs, ctl->devs_min);
5367 : }
5368 : return -ENOSPC;
5369 : }
5370 :
5371 1470 : ctl->ndevs = min(ctl->ndevs, ctl->devs_max);
5372 :
5373 1470 : switch (fs_devices->chunk_alloc_policy) {
5374 1470 : case BTRFS_CHUNK_ALLOC_REGULAR:
5375 1470 : return decide_stripe_size_regular(ctl, devices_info);
5376 0 : case BTRFS_CHUNK_ALLOC_ZONED:
5377 0 : return decide_stripe_size_zoned(ctl, devices_info);
5378 0 : default:
5379 0 : BUG();
5380 : }
5381 : }
5382 :
5383 1470 : static struct btrfs_block_group *create_chunk(struct btrfs_trans_handle *trans,
5384 : struct alloc_chunk_ctl *ctl,
5385 : struct btrfs_device_info *devices_info)
5386 : {
5387 1470 : struct btrfs_fs_info *info = trans->fs_info;
5388 1470 : struct map_lookup *map = NULL;
5389 1470 : struct extent_map_tree *em_tree;
5390 1470 : struct btrfs_block_group *block_group;
5391 1470 : struct extent_map *em;
5392 1470 : u64 start = ctl->start;
5393 1470 : u64 type = ctl->type;
5394 1470 : int ret;
5395 1470 : int i;
5396 1470 : int j;
5397 :
5398 1470 : map = kmalloc(map_lookup_size(ctl->num_stripes), GFP_NOFS);
5399 1470 : if (!map)
5400 : return ERR_PTR(-ENOMEM);
5401 1470 : map->num_stripes = ctl->num_stripes;
5402 :
5403 2940 : for (i = 0; i < ctl->ndevs; ++i) {
5404 3270 : for (j = 0; j < ctl->dev_stripes; ++j) {
5405 1800 : int s = i * ctl->dev_stripes + j;
5406 1800 : map->stripes[s].dev = devices_info[i].dev;
5407 1800 : map->stripes[s].physical = devices_info[i].dev_offset +
5408 1800 : j * ctl->stripe_size;
5409 : }
5410 : }
5411 1470 : map->io_align = BTRFS_STRIPE_LEN;
5412 1470 : map->io_width = BTRFS_STRIPE_LEN;
5413 1470 : map->type = type;
5414 1470 : map->sub_stripes = ctl->sub_stripes;
5415 :
5416 1470 : trace_btrfs_chunk_alloc(info, map, start, ctl->chunk_size);
5417 :
5418 1470 : em = alloc_extent_map();
5419 1470 : if (!em) {
5420 0 : kfree(map);
5421 0 : return ERR_PTR(-ENOMEM);
5422 : }
5423 1470 : set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5424 1470 : em->map_lookup = map;
5425 1470 : em->start = start;
5426 1470 : em->len = ctl->chunk_size;
5427 1470 : em->block_start = 0;
5428 1470 : em->block_len = em->len;
5429 1470 : em->orig_block_len = ctl->stripe_size;
5430 :
5431 1470 : em_tree = &info->mapping_tree;
5432 1470 : write_lock(&em_tree->lock);
5433 1470 : ret = add_extent_mapping(em_tree, em, 0);
5434 1470 : if (ret) {
5435 0 : write_unlock(&em_tree->lock);
5436 0 : free_extent_map(em);
5437 0 : return ERR_PTR(ret);
5438 : }
5439 1470 : write_unlock(&em_tree->lock);
5440 :
5441 1470 : block_group = btrfs_make_block_group(trans, type, start, ctl->chunk_size);
5442 1470 : if (IS_ERR(block_group))
5443 0 : goto error_del_extent;
5444 :
5445 3270 : for (i = 0; i < map->num_stripes; i++) {
5446 1800 : struct btrfs_device *dev = map->stripes[i].dev;
5447 :
5448 1800 : btrfs_device_set_bytes_used(dev,
5449 1800 : dev->bytes_used + ctl->stripe_size);
5450 1800 : if (list_empty(&dev->post_commit_list))
5451 971 : list_add_tail(&dev->post_commit_list,
5452 971 : &trans->transaction->dev_update_list);
5453 : }
5454 :
5455 1470 : atomic64_sub(ctl->stripe_size * map->num_stripes,
5456 : &info->free_chunk_space);
5457 :
5458 1470 : free_extent_map(em);
5459 1470 : check_raid56_incompat_flag(info, type);
5460 1470 : check_raid1c34_incompat_flag(info, type);
5461 :
5462 1470 : return block_group;
5463 :
5464 : error_del_extent:
5465 0 : write_lock(&em_tree->lock);
5466 0 : remove_extent_mapping(em_tree, em);
5467 0 : write_unlock(&em_tree->lock);
5468 :
5469 : /* One for our allocation */
5470 0 : free_extent_map(em);
5471 : /* One for the tree reference */
5472 0 : free_extent_map(em);
5473 :
5474 0 : return block_group;
5475 : }
5476 :
5477 1520 : struct btrfs_block_group *btrfs_create_chunk(struct btrfs_trans_handle *trans,
5478 : u64 type)
5479 : {
5480 1520 : struct btrfs_fs_info *info = trans->fs_info;
5481 1520 : struct btrfs_fs_devices *fs_devices = info->fs_devices;
5482 1520 : struct btrfs_device_info *devices_info = NULL;
5483 1520 : struct alloc_chunk_ctl ctl;
5484 1520 : struct btrfs_block_group *block_group;
5485 1520 : int ret;
5486 :
5487 1520 : lockdep_assert_held(&info->chunk_mutex);
5488 :
5489 1520 : if (!alloc_profile_is_valid(type, 0)) {
5490 : ASSERT(0);
5491 : return ERR_PTR(-EINVAL);
5492 : }
5493 :
5494 1520 : if (list_empty(&fs_devices->alloc_list)) {
5495 : if (btrfs_test_opt(info, ENOSPC_DEBUG))
5496 : btrfs_debug(info, "%s: no writable device", __func__);
5497 : return ERR_PTR(-ENOSPC);
5498 : }
5499 :
5500 1520 : if (!(type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
5501 0 : btrfs_err(info, "invalid chunk type 0x%llx requested", type);
5502 0 : ASSERT(0);
5503 0 : return ERR_PTR(-EINVAL);
5504 : }
5505 :
5506 1520 : ctl.start = find_next_chunk(info);
5507 1520 : ctl.type = type;
5508 1520 : init_alloc_chunk_ctl(fs_devices, &ctl);
5509 :
5510 1520 : devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
5511 : GFP_NOFS);
5512 1520 : if (!devices_info)
5513 : return ERR_PTR(-ENOMEM);
5514 :
5515 1520 : ret = gather_device_info(fs_devices, &ctl, devices_info);
5516 1520 : if (ret < 0) {
5517 0 : block_group = ERR_PTR(ret);
5518 0 : goto out;
5519 : }
5520 :
5521 1520 : ret = decide_stripe_size(fs_devices, &ctl, devices_info);
5522 1520 : if (ret < 0) {
5523 50 : block_group = ERR_PTR(ret);
5524 50 : goto out;
5525 : }
5526 :
5527 1470 : block_group = create_chunk(trans, &ctl, devices_info);
5528 :
5529 1520 : out:
5530 1520 : kfree(devices_info);
5531 1520 : return block_group;
5532 : }
5533 :
5534 : /*
5535 : * This function, btrfs_chunk_alloc_add_chunk_item(), typically belongs to the
5536 : * phase 1 of chunk allocation. It belongs to phase 2 only when allocating system
5537 : * chunks.
5538 : *
5539 : * See the comment at btrfs_chunk_alloc() for details about the chunk allocation
5540 : * phases.
5541 : */
5542 1470 : int btrfs_chunk_alloc_add_chunk_item(struct btrfs_trans_handle *trans,
5543 : struct btrfs_block_group *bg)
5544 : {
5545 1470 : struct btrfs_fs_info *fs_info = trans->fs_info;
5546 1470 : struct btrfs_root *chunk_root = fs_info->chunk_root;
5547 1470 : struct btrfs_key key;
5548 1470 : struct btrfs_chunk *chunk;
5549 1470 : struct btrfs_stripe *stripe;
5550 1470 : struct extent_map *em;
5551 1470 : struct map_lookup *map;
5552 1470 : size_t item_size;
5553 1470 : int i;
5554 1470 : int ret;
5555 :
5556 : /*
5557 : * We take the chunk_mutex for 2 reasons:
5558 : *
5559 : * 1) Updates and insertions in the chunk btree must be done while holding
5560 : * the chunk_mutex, as well as updating the system chunk array in the
5561 : * superblock. See the comment on top of btrfs_chunk_alloc() for the
5562 : * details;
5563 : *
5564 : * 2) To prevent races with the final phase of a device replace operation
5565 : * that replaces the device object associated with the map's stripes,
5566 : * because the device object's id can change at any time during that
5567 : * final phase of the device replace operation
5568 : * (dev-replace.c:btrfs_dev_replace_finishing()), so we could grab the
5569 : * replaced device and then see it with an ID of BTRFS_DEV_REPLACE_DEVID,
5570 : * which would cause a failure when updating the device item, which does
5571 : * not exists, or persisting a stripe of the chunk item with such ID.
5572 : * Here we can't use the device_list_mutex because our caller already
5573 : * has locked the chunk_mutex, and the final phase of device replace
5574 : * acquires both mutexes - first the device_list_mutex and then the
5575 : * chunk_mutex. Using any of those two mutexes protects us from a
5576 : * concurrent device replace.
5577 : */
5578 1470 : lockdep_assert_held(&fs_info->chunk_mutex);
5579 :
5580 1470 : em = btrfs_get_chunk_map(fs_info, bg->start, bg->length);
5581 1470 : if (IS_ERR(em)) {
5582 0 : ret = PTR_ERR(em);
5583 0 : btrfs_abort_transaction(trans, ret);
5584 0 : return ret;
5585 : }
5586 :
5587 1470 : map = em->map_lookup;
5588 1470 : item_size = btrfs_chunk_item_size(map->num_stripes);
5589 :
5590 1470 : chunk = kzalloc(item_size, GFP_NOFS);
5591 1470 : if (!chunk) {
5592 0 : ret = -ENOMEM;
5593 0 : btrfs_abort_transaction(trans, ret);
5594 0 : goto out;
5595 : }
5596 :
5597 3270 : for (i = 0; i < map->num_stripes; i++) {
5598 1800 : struct btrfs_device *device = map->stripes[i].dev;
5599 :
5600 1800 : ret = btrfs_update_device(trans, device);
5601 1800 : if (ret)
5602 0 : goto out;
5603 : }
5604 :
5605 1470 : stripe = &chunk->stripe;
5606 3270 : for (i = 0; i < map->num_stripes; i++) {
5607 1800 : struct btrfs_device *device = map->stripes[i].dev;
5608 1800 : const u64 dev_offset = map->stripes[i].physical;
5609 :
5610 1800 : btrfs_set_stack_stripe_devid(stripe, device->devid);
5611 1800 : btrfs_set_stack_stripe_offset(stripe, dev_offset);
5612 3600 : memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
5613 1800 : stripe++;
5614 : }
5615 :
5616 1470 : btrfs_set_stack_chunk_length(chunk, bg->length);
5617 1470 : btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
5618 1470 : btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
5619 1470 : btrfs_set_stack_chunk_type(chunk, map->type);
5620 1470 : btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
5621 1470 : btrfs_set_stack_chunk_io_align(chunk, BTRFS_STRIPE_LEN);
5622 1470 : btrfs_set_stack_chunk_io_width(chunk, BTRFS_STRIPE_LEN);
5623 1470 : btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize);
5624 1470 : btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
5625 :
5626 1470 : key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
5627 1470 : key.type = BTRFS_CHUNK_ITEM_KEY;
5628 1470 : key.offset = bg->start;
5629 :
5630 1470 : ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
5631 1470 : if (ret)
5632 0 : goto out;
5633 :
5634 1470 : set_bit(BLOCK_GROUP_FLAG_CHUNK_ITEM_INSERTED, &bg->runtime_flags);
5635 :
5636 1470 : if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
5637 93 : ret = btrfs_add_system_chunk(fs_info, &key, chunk, item_size);
5638 93 : if (ret)
5639 0 : goto out;
5640 : }
5641 :
5642 1470 : out:
5643 1470 : kfree(chunk);
5644 1470 : free_extent_map(em);
5645 1470 : return ret;
5646 : }
5647 :
5648 0 : static noinline int init_first_rw_device(struct btrfs_trans_handle *trans)
5649 : {
5650 0 : struct btrfs_fs_info *fs_info = trans->fs_info;
5651 0 : u64 alloc_profile;
5652 0 : struct btrfs_block_group *meta_bg;
5653 0 : struct btrfs_block_group *sys_bg;
5654 :
5655 : /*
5656 : * When adding a new device for sprouting, the seed device is read-only
5657 : * so we must first allocate a metadata and a system chunk. But before
5658 : * adding the block group items to the extent, device and chunk btrees,
5659 : * we must first:
5660 : *
5661 : * 1) Create both chunks without doing any changes to the btrees, as
5662 : * otherwise we would get -ENOSPC since the block groups from the
5663 : * seed device are read-only;
5664 : *
5665 : * 2) Add the device item for the new sprout device - finishing the setup
5666 : * of a new block group requires updating the device item in the chunk
5667 : * btree, so it must exist when we attempt to do it. The previous step
5668 : * ensures this does not fail with -ENOSPC.
5669 : *
5670 : * After that we can add the block group items to their btrees:
5671 : * update existing device item in the chunk btree, add a new block group
5672 : * item to the extent btree, add a new chunk item to the chunk btree and
5673 : * finally add the new device extent items to the devices btree.
5674 : */
5675 :
5676 0 : alloc_profile = btrfs_metadata_alloc_profile(fs_info);
5677 0 : meta_bg = btrfs_create_chunk(trans, alloc_profile);
5678 0 : if (IS_ERR(meta_bg))
5679 0 : return PTR_ERR(meta_bg);
5680 :
5681 0 : alloc_profile = btrfs_system_alloc_profile(fs_info);
5682 0 : sys_bg = btrfs_create_chunk(trans, alloc_profile);
5683 0 : if (IS_ERR(sys_bg))
5684 0 : return PTR_ERR(sys_bg);
5685 :
5686 : return 0;
5687 : }
5688 :
5689 : static inline int btrfs_chunk_max_errors(struct map_lookup *map)
5690 : {
5691 8975786 : const int index = btrfs_bg_flags_to_raid_index(map->type);
5692 :
5693 8975786 : return btrfs_raid_array[index].tolerated_failures;
5694 : }
5695 :
5696 29306 : bool btrfs_chunk_writeable(struct btrfs_fs_info *fs_info, u64 chunk_offset)
5697 : {
5698 29306 : struct extent_map *em;
5699 29306 : struct map_lookup *map;
5700 29306 : int miss_ndevs = 0;
5701 29306 : int i;
5702 29306 : bool ret = true;
5703 :
5704 29306 : em = btrfs_get_chunk_map(fs_info, chunk_offset, 1);
5705 29306 : if (IS_ERR(em))
5706 : return false;
5707 :
5708 29306 : map = em->map_lookup;
5709 65110 : for (i = 0; i < map->num_stripes; i++) {
5710 71608 : if (test_bit(BTRFS_DEV_STATE_MISSING,
5711 : &map->stripes[i].dev->dev_state)) {
5712 0 : miss_ndevs++;
5713 0 : continue;
5714 : }
5715 35804 : if (!test_bit(BTRFS_DEV_STATE_WRITEABLE,
5716 : &map->stripes[i].dev->dev_state)) {
5717 0 : ret = false;
5718 0 : goto end;
5719 : }
5720 : }
5721 :
5722 : /*
5723 : * If the number of missing devices is larger than max errors, we can
5724 : * not write the data into that chunk successfully.
5725 : */
5726 29306 : if (miss_ndevs > btrfs_chunk_max_errors(map))
5727 0 : ret = false;
5728 29306 : end:
5729 29306 : free_extent_map(em);
5730 29306 : return ret;
5731 : }
5732 :
5733 3244 : void btrfs_mapping_tree_free(struct extent_map_tree *tree)
5734 : {
5735 33457 : struct extent_map *em;
5736 :
5737 63670 : while (1) {
5738 33457 : write_lock(&tree->lock);
5739 33457 : em = lookup_extent_mapping(tree, 0, (u64)-1);
5740 33457 : if (em)
5741 30213 : remove_extent_mapping(tree, em);
5742 33457 : write_unlock(&tree->lock);
5743 33457 : if (!em)
5744 : break;
5745 : /* once for us */
5746 30213 : free_extent_map(em);
5747 : /* once for the tree */
5748 30213 : free_extent_map(em);
5749 : }
5750 3244 : }
5751 :
5752 15532489 : int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5753 : {
5754 15532489 : struct extent_map *em;
5755 15532489 : struct map_lookup *map;
5756 15532489 : enum btrfs_raid_types index;
5757 15532489 : int ret = 1;
5758 :
5759 15532489 : em = btrfs_get_chunk_map(fs_info, logical, len);
5760 15534878 : if (IS_ERR(em))
5761 : /*
5762 : * We could return errors for these cases, but that could get
5763 : * ugly and we'd probably do the same thing which is just not do
5764 : * anything else and exit, so return 1 so the callers don't try
5765 : * to use other copies.
5766 : */
5767 : return 1;
5768 :
5769 15534878 : map = em->map_lookup;
5770 15534878 : index = btrfs_bg_flags_to_raid_index(map->type);
5771 :
5772 : /* Non-RAID56, use their ncopies from btrfs_raid_array. */
5773 15534878 : if (!(map->type & BTRFS_BLOCK_GROUP_RAID56_MASK))
5774 15534878 : ret = btrfs_raid_array[index].ncopies;
5775 0 : else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
5776 : ret = 2;
5777 0 : else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5778 : /*
5779 : * There could be two corrupted data stripes, we need
5780 : * to loop retry in order to rebuild the correct data.
5781 : *
5782 : * Fail a stripe at a time on every retry except the
5783 : * stripe under reconstruction.
5784 : */
5785 0 : ret = map->num_stripes;
5786 15534878 : free_extent_map(em);
5787 15534878 : return ret;
5788 : }
5789 :
5790 30776 : unsigned long btrfs_full_stripe_len(struct btrfs_fs_info *fs_info,
5791 : u64 logical)
5792 : {
5793 30776 : struct extent_map *em;
5794 30776 : struct map_lookup *map;
5795 30776 : unsigned long len = fs_info->sectorsize;
5796 :
5797 30776 : if (!btrfs_fs_incompat(fs_info, RAID56))
5798 : return len;
5799 :
5800 0 : em = btrfs_get_chunk_map(fs_info, logical, len);
5801 :
5802 0 : if (!WARN_ON(IS_ERR(em))) {
5803 0 : map = em->map_lookup;
5804 0 : if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5805 0 : len = btrfs_stripe_nr_to_offset(nr_data_stripes(map));
5806 0 : free_extent_map(em);
5807 : }
5808 : return len;
5809 : }
5810 :
5811 0 : int btrfs_is_parity_mirror(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
5812 : {
5813 0 : struct extent_map *em;
5814 0 : struct map_lookup *map;
5815 0 : int ret = 0;
5816 :
5817 0 : if (!btrfs_fs_incompat(fs_info, RAID56))
5818 : return 0;
5819 :
5820 0 : em = btrfs_get_chunk_map(fs_info, logical, len);
5821 :
5822 0 : if(!WARN_ON(IS_ERR(em))) {
5823 0 : map = em->map_lookup;
5824 0 : if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
5825 0 : ret = 1;
5826 0 : free_extent_map(em);
5827 : }
5828 : return ret;
5829 : }
5830 :
5831 9 : static int find_live_mirror(struct btrfs_fs_info *fs_info,
5832 : struct map_lookup *map, int first,
5833 : int dev_replace_is_ongoing)
5834 : {
5835 9 : int i;
5836 9 : int num_stripes;
5837 9 : int preferred_mirror;
5838 9 : int tolerance;
5839 9 : struct btrfs_device *srcdev;
5840 :
5841 9 : ASSERT((map->type &
5842 : (BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10)));
5843 :
5844 9 : if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5845 0 : num_stripes = map->sub_stripes;
5846 : else
5847 9 : num_stripes = map->num_stripes;
5848 :
5849 9 : switch (fs_info->fs_devices->read_policy) {
5850 0 : default:
5851 : /* Shouldn't happen, just warn and use pid instead of failing */
5852 0 : btrfs_warn_rl(fs_info,
5853 : "unknown read_policy type %u, reset to pid",
5854 : fs_info->fs_devices->read_policy);
5855 0 : fs_info->fs_devices->read_policy = BTRFS_READ_POLICY_PID;
5856 9 : fallthrough;
5857 9 : case BTRFS_READ_POLICY_PID:
5858 9 : preferred_mirror = first + (current->pid % num_stripes);
5859 9 : break;
5860 : }
5861 :
5862 9 : if (dev_replace_is_ongoing &&
5863 0 : fs_info->dev_replace.cont_reading_from_srcdev_mode ==
5864 : BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
5865 0 : srcdev = fs_info->dev_replace.srcdev;
5866 : else
5867 : srcdev = NULL;
5868 :
5869 : /*
5870 : * try to avoid the drive that is the source drive for a
5871 : * dev-replace procedure, only choose it if no other non-missing
5872 : * mirror is available
5873 : */
5874 9 : for (tolerance = 0; tolerance < 2; tolerance++) {
5875 9 : if (map->stripes[preferred_mirror].dev->bdev &&
5876 9 : (tolerance || map->stripes[preferred_mirror].dev != srcdev))
5877 : return preferred_mirror;
5878 0 : for (i = first; i < first + num_stripes; i++) {
5879 0 : if (map->stripes[i].dev->bdev &&
5880 0 : (tolerance || map->stripes[i].dev != srcdev))
5881 0 : return i;
5882 : }
5883 : }
5884 :
5885 : /* we couldn't find one that doesn't fail. Just return something
5886 : * and the io error handling code will clean up eventually
5887 : */
5888 : return preferred_mirror;
5889 : }
5890 :
5891 8946486 : static struct btrfs_io_context *alloc_btrfs_io_context(struct btrfs_fs_info *fs_info,
5892 : u16 total_stripes)
5893 : {
5894 8946486 : struct btrfs_io_context *bioc;
5895 :
5896 8946486 : bioc = kzalloc(
5897 : /* The size of btrfs_io_context */
5898 : sizeof(struct btrfs_io_context) +
5899 : /* Plus the variable array for the stripes */
5900 8946486 : sizeof(struct btrfs_io_stripe) * (total_stripes),
5901 : GFP_NOFS);
5902 :
5903 8946465 : if (!bioc)
5904 : return NULL;
5905 :
5906 8946465 : refcount_set(&bioc->refs, 1);
5907 :
5908 8946465 : bioc->fs_info = fs_info;
5909 8946465 : bioc->replace_stripe_src = -1;
5910 8946465 : bioc->full_stripe_logical = (u64)-1;
5911 :
5912 8946465 : return bioc;
5913 : }
5914 :
5915 0 : void btrfs_get_bioc(struct btrfs_io_context *bioc)
5916 : {
5917 0 : WARN_ON(!refcount_read(&bioc->refs));
5918 0 : refcount_inc(&bioc->refs);
5919 0 : }
5920 :
5921 8946546 : void btrfs_put_bioc(struct btrfs_io_context *bioc)
5922 : {
5923 8946546 : if (!bioc)
5924 : return;
5925 8946546 : if (refcount_dec_and_test(&bioc->refs))
5926 8946546 : kfree(bioc);
5927 : }
5928 :
5929 : /*
5930 : * Please note that, discard won't be sent to target device of device
5931 : * replace.
5932 : */
5933 47920 : struct btrfs_discard_stripe *btrfs_map_discard(struct btrfs_fs_info *fs_info,
5934 : u64 logical, u64 *length_ret,
5935 : u32 *num_stripes)
5936 : {
5937 47920 : struct extent_map *em;
5938 47920 : struct map_lookup *map;
5939 47920 : struct btrfs_discard_stripe *stripes;
5940 47920 : u64 length = *length_ret;
5941 47920 : u64 offset;
5942 47920 : u32 stripe_nr;
5943 47920 : u32 stripe_nr_end;
5944 47920 : u32 stripe_cnt;
5945 47920 : u64 stripe_end_offset;
5946 47920 : u64 stripe_offset;
5947 47920 : u32 stripe_index;
5948 47920 : u32 factor = 0;
5949 47920 : u32 sub_stripes = 0;
5950 47920 : u32 stripes_per_dev = 0;
5951 47920 : u32 remaining_stripes = 0;
5952 47920 : u32 last_stripe = 0;
5953 47920 : int ret;
5954 47920 : int i;
5955 :
5956 47920 : em = btrfs_get_chunk_map(fs_info, logical, length);
5957 47920 : if (IS_ERR(em))
5958 : return ERR_CAST(em);
5959 :
5960 47920 : map = em->map_lookup;
5961 :
5962 : /* we don't discard raid56 yet */
5963 47920 : if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5964 0 : ret = -EOPNOTSUPP;
5965 0 : goto out_free_map;
5966 : }
5967 :
5968 47920 : offset = logical - em->start;
5969 47920 : length = min_t(u64, em->start + em->len - logical, length);
5970 47920 : *length_ret = length;
5971 :
5972 : /*
5973 : * stripe_nr counts the total number of stripes we have to stride
5974 : * to get to this block
5975 : */
5976 47920 : stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
5977 :
5978 : /* stripe_offset is the offset of this block in its stripe */
5979 47920 : stripe_offset = offset - btrfs_stripe_nr_to_offset(stripe_nr);
5980 :
5981 47920 : stripe_nr_end = round_up(offset + length, BTRFS_STRIPE_LEN) >>
5982 : BTRFS_STRIPE_LEN_SHIFT;
5983 47920 : stripe_cnt = stripe_nr_end - stripe_nr;
5984 47920 : stripe_end_offset = btrfs_stripe_nr_to_offset(stripe_nr_end) -
5985 : (offset + length);
5986 : /*
5987 : * after this, stripe_nr is the number of stripes on this
5988 : * device we have to walk to find the data, and stripe_index is
5989 : * the number of our device in the stripe array
5990 : */
5991 47920 : *num_stripes = 1;
5992 47920 : stripe_index = 0;
5993 47920 : if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5994 : BTRFS_BLOCK_GROUP_RAID10)) {
5995 0 : if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5996 : sub_stripes = 1;
5997 : else
5998 0 : sub_stripes = map->sub_stripes;
5999 :
6000 0 : factor = map->num_stripes / sub_stripes;
6001 0 : *num_stripes = min_t(u64, map->num_stripes,
6002 : sub_stripes * stripe_cnt);
6003 0 : stripe_index = stripe_nr % factor;
6004 0 : stripe_nr /= factor;
6005 0 : stripe_index *= sub_stripes;
6006 :
6007 0 : remaining_stripes = stripe_cnt % factor;
6008 0 : stripes_per_dev = stripe_cnt / factor;
6009 0 : last_stripe = ((stripe_nr_end - 1) % factor) * sub_stripes;
6010 47920 : } else if (map->type & (BTRFS_BLOCK_GROUP_RAID1_MASK |
6011 : BTRFS_BLOCK_GROUP_DUP)) {
6012 14713 : *num_stripes = map->num_stripes;
6013 : } else {
6014 33207 : stripe_index = stripe_nr % map->num_stripes;
6015 33207 : stripe_nr /= map->num_stripes;
6016 : }
6017 :
6018 47920 : stripes = kcalloc(*num_stripes, sizeof(*stripes), GFP_NOFS);
6019 47920 : if (!stripes) {
6020 0 : ret = -ENOMEM;
6021 0 : goto out_free_map;
6022 : }
6023 :
6024 110553 : for (i = 0; i < *num_stripes; i++) {
6025 62633 : stripes[i].physical =
6026 62633 : map->stripes[stripe_index].physical +
6027 62633 : stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6028 62633 : stripes[i].dev = map->stripes[stripe_index].dev;
6029 :
6030 62633 : if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
6031 : BTRFS_BLOCK_GROUP_RAID10)) {
6032 0 : stripes[i].length = btrfs_stripe_nr_to_offset(stripes_per_dev);
6033 :
6034 0 : if (i / sub_stripes < remaining_stripes)
6035 0 : stripes[i].length += BTRFS_STRIPE_LEN;
6036 :
6037 : /*
6038 : * Special for the first stripe and
6039 : * the last stripe:
6040 : *
6041 : * |-------|...|-------|
6042 : * |----------|
6043 : * off end_off
6044 : */
6045 0 : if (i < sub_stripes)
6046 0 : stripes[i].length -= stripe_offset;
6047 :
6048 0 : if (stripe_index >= last_stripe &&
6049 0 : stripe_index <= (last_stripe +
6050 0 : sub_stripes - 1))
6051 0 : stripes[i].length -= stripe_end_offset;
6052 :
6053 0 : if (i == sub_stripes - 1)
6054 0 : stripe_offset = 0;
6055 : } else {
6056 62633 : stripes[i].length = length;
6057 : }
6058 :
6059 62633 : stripe_index++;
6060 62633 : if (stripe_index == map->num_stripes) {
6061 47920 : stripe_index = 0;
6062 47920 : stripe_nr++;
6063 : }
6064 : }
6065 :
6066 47920 : free_extent_map(em);
6067 47920 : return stripes;
6068 0 : out_free_map:
6069 0 : free_extent_map(em);
6070 0 : return ERR_PTR(ret);
6071 : }
6072 :
6073 : static bool is_block_group_to_copy(struct btrfs_fs_info *fs_info, u64 logical)
6074 : {
6075 : struct btrfs_block_group *cache;
6076 : bool ret;
6077 :
6078 : /* Non zoned filesystem does not use "to_copy" flag */
6079 : if (!btrfs_is_zoned(fs_info))
6080 : return false;
6081 :
6082 : cache = btrfs_lookup_block_group(fs_info, logical);
6083 :
6084 : ret = test_bit(BLOCK_GROUP_FLAG_TO_COPY, &cache->runtime_flags);
6085 :
6086 : btrfs_put_block_group(cache);
6087 : return ret;
6088 : }
6089 :
6090 0 : static void handle_ops_on_dev_replace(enum btrfs_map_op op,
6091 : struct btrfs_io_context *bioc,
6092 : struct btrfs_dev_replace *dev_replace,
6093 : u64 logical,
6094 : int *num_stripes_ret, int *max_errors_ret)
6095 : {
6096 0 : u64 srcdev_devid = dev_replace->srcdev->devid;
6097 : /*
6098 : * At this stage, num_stripes is still the real number of stripes,
6099 : * excluding the duplicated stripes.
6100 : */
6101 0 : int num_stripes = *num_stripes_ret;
6102 0 : int nr_extra_stripes = 0;
6103 0 : int max_errors = *max_errors_ret;
6104 0 : int i;
6105 :
6106 : /*
6107 : * A block group which has "to_copy" set will eventually be copied by
6108 : * the dev-replace process. We can avoid cloning IO here.
6109 : */
6110 0 : if (is_block_group_to_copy(dev_replace->srcdev->fs_info, logical))
6111 : return;
6112 :
6113 : /*
6114 : * Duplicate the write operations while the dev-replace procedure is
6115 : * running. Since the copying of the old disk to the new disk takes
6116 : * place at run time while the filesystem is mounted writable, the
6117 : * regular write operations to the old disk have to be duplicated to go
6118 : * to the new disk as well.
6119 : *
6120 : * Note that device->missing is handled by the caller, and that the
6121 : * write to the old disk is already set up in the stripes array.
6122 : */
6123 0 : for (i = 0; i < num_stripes; i++) {
6124 0 : struct btrfs_io_stripe *old = &bioc->stripes[i];
6125 0 : struct btrfs_io_stripe *new = &bioc->stripes[num_stripes + nr_extra_stripes];
6126 :
6127 0 : if (old->dev->devid != srcdev_devid)
6128 0 : continue;
6129 :
6130 0 : new->physical = old->physical;
6131 0 : new->dev = dev_replace->tgtdev;
6132 0 : if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK)
6133 0 : bioc->replace_stripe_src = i;
6134 0 : nr_extra_stripes++;
6135 : }
6136 :
6137 : /* We can only have at most 2 extra nr_stripes (for DUP). */
6138 0 : ASSERT(nr_extra_stripes <= 2);
6139 : /*
6140 : * For GET_READ_MIRRORS, we can only return at most 1 extra stripe for
6141 : * replace.
6142 : * If we have 2 extra stripes, only choose the one with smaller physical.
6143 : */
6144 0 : if (op == BTRFS_MAP_GET_READ_MIRRORS && nr_extra_stripes == 2) {
6145 0 : struct btrfs_io_stripe *first = &bioc->stripes[num_stripes];
6146 0 : struct btrfs_io_stripe *second = &bioc->stripes[num_stripes + 1];
6147 :
6148 : /* Only DUP can have two extra stripes. */
6149 0 : ASSERT(bioc->map_type & BTRFS_BLOCK_GROUP_DUP);
6150 :
6151 : /*
6152 : * Swap the last stripe stripes and reduce @nr_extra_stripes.
6153 : * The extra stripe would still be there, but won't be accessed.
6154 : */
6155 0 : if (first->physical > second->physical) {
6156 0 : swap(second->physical, first->physical);
6157 0 : swap(second->dev, first->dev);
6158 0 : nr_extra_stripes--;
6159 : }
6160 : }
6161 :
6162 0 : *num_stripes_ret = num_stripes + nr_extra_stripes;
6163 0 : *max_errors_ret = max_errors + nr_extra_stripes;
6164 0 : bioc->replace_nr_stripes = nr_extra_stripes;
6165 : }
6166 :
6167 15469735 : static u64 btrfs_max_io_len(struct map_lookup *map, enum btrfs_map_op op,
6168 : u64 offset, u32 *stripe_nr, u64 *stripe_offset,
6169 : u64 *full_stripe_start)
6170 : {
6171 : /*
6172 : * Stripe_nr is the stripe where this block falls. stripe_offset is
6173 : * the offset of this block in its stripe.
6174 : */
6175 15469735 : *stripe_offset = offset & BTRFS_STRIPE_LEN_MASK;
6176 15469735 : *stripe_nr = offset >> BTRFS_STRIPE_LEN_SHIFT;
6177 15469735 : ASSERT(*stripe_offset < U32_MAX);
6178 :
6179 15469735 : if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6180 0 : unsigned long full_stripe_len =
6181 0 : btrfs_stripe_nr_to_offset(nr_data_stripes(map));
6182 :
6183 : /*
6184 : * For full stripe start, we use previously calculated
6185 : * @stripe_nr. Align it to nr_data_stripes, then multiply with
6186 : * STRIPE_LEN.
6187 : *
6188 : * By this we can avoid u64 division completely. And we have
6189 : * to go rounddown(), not round_down(), as nr_data_stripes is
6190 : * not ensured to be power of 2.
6191 : */
6192 0 : *full_stripe_start =
6193 : btrfs_stripe_nr_to_offset(
6194 0 : rounddown(*stripe_nr, nr_data_stripes(map)));
6195 :
6196 0 : ASSERT(*full_stripe_start + full_stripe_len > offset);
6197 0 : ASSERT(*full_stripe_start <= offset);
6198 : /*
6199 : * For writes to RAID56, allow to write a full stripe set, but
6200 : * no straddling of stripe sets.
6201 : */
6202 0 : if (op == BTRFS_MAP_WRITE)
6203 0 : return full_stripe_len - (offset - *full_stripe_start);
6204 : }
6205 :
6206 : /*
6207 : * For other RAID types and for RAID56 reads, allow a single stripe (on
6208 : * a single disk).
6209 : */
6210 15469735 : if (map->type & BTRFS_BLOCK_GROUP_STRIPE_MASK)
6211 0 : return BTRFS_STRIPE_LEN - *stripe_offset;
6212 : return U64_MAX;
6213 : }
6214 :
6215 : static void set_io_stripe(struct btrfs_io_stripe *dst, const struct map_lookup *map,
6216 : u32 stripe_index, u64 stripe_offset, u32 stripe_nr)
6217 : {
6218 24411718 : dst->dev = map->stripes[stripe_index].dev;
6219 24411718 : dst->physical = map->stripes[stripe_index].physical +
6220 24411718 : stripe_offset + btrfs_stripe_nr_to_offset(stripe_nr);
6221 : }
6222 :
6223 15467493 : int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op,
6224 : u64 logical, u64 *length,
6225 : struct btrfs_io_context **bioc_ret,
6226 : struct btrfs_io_stripe *smap, int *mirror_num_ret,
6227 : int need_raid_map)
6228 : {
6229 15467493 : struct extent_map *em;
6230 15467493 : struct map_lookup *map;
6231 15467493 : u64 map_offset;
6232 15467493 : u64 stripe_offset;
6233 15467493 : u32 stripe_nr;
6234 15467493 : u32 stripe_index;
6235 15467493 : int data_stripes;
6236 15467493 : int i;
6237 15467493 : int ret = 0;
6238 15467493 : int mirror_num = (mirror_num_ret ? *mirror_num_ret : 0);
6239 15467493 : int num_stripes;
6240 15467493 : int num_copies;
6241 15467493 : int max_errors = 0;
6242 15467493 : struct btrfs_io_context *bioc = NULL;
6243 15467493 : struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
6244 15467493 : int dev_replace_is_ongoing = 0;
6245 15467493 : u16 num_alloc_stripes;
6246 15467493 : u64 raid56_full_stripe_start = (u64)-1;
6247 15467493 : u64 max_len;
6248 :
6249 15467493 : ASSERT(bioc_ret);
6250 :
6251 15467493 : num_copies = btrfs_num_copies(fs_info, logical, fs_info->sectorsize);
6252 15469821 : if (mirror_num > num_copies)
6253 : return -EINVAL;
6254 :
6255 15469821 : em = btrfs_get_chunk_map(fs_info, logical, *length);
6256 15469487 : if (IS_ERR(em))
6257 0 : return PTR_ERR(em);
6258 :
6259 15469487 : map = em->map_lookup;
6260 15469487 : data_stripes = nr_data_stripes(map);
6261 :
6262 15469487 : map_offset = logical - em->start;
6263 15469487 : max_len = btrfs_max_io_len(map, op, map_offset, &stripe_nr,
6264 : &stripe_offset, &raid56_full_stripe_start);
6265 15469539 : *length = min_t(u64, em->len - map_offset, max_len);
6266 :
6267 15469539 : down_read(&dev_replace->rwsem);
6268 15464565 : dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
6269 : /*
6270 : * Hold the semaphore for read during the whole operation, write is
6271 : * requested at commit time but must wait.
6272 : */
6273 15464565 : if (!dev_replace_is_ongoing)
6274 15465971 : up_read(&dev_replace->rwsem);
6275 :
6276 15465858 : num_stripes = 1;
6277 15465858 : stripe_index = 0;
6278 15465858 : if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
6279 0 : stripe_index = stripe_nr % map->num_stripes;
6280 0 : stripe_nr /= map->num_stripes;
6281 0 : if (op == BTRFS_MAP_READ)
6282 0 : mirror_num = 1;
6283 15465858 : } else if (map->type & BTRFS_BLOCK_GROUP_RAID1_MASK) {
6284 9 : if (op != BTRFS_MAP_READ) {
6285 0 : num_stripes = map->num_stripes;
6286 9 : } else if (mirror_num) {
6287 0 : stripe_index = mirror_num - 1;
6288 : } else {
6289 9 : stripe_index = find_live_mirror(fs_info, map, 0,
6290 : dev_replace_is_ongoing);
6291 9 : mirror_num = stripe_index + 1;
6292 : }
6293 :
6294 15465849 : } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
6295 9037649 : if (op != BTRFS_MAP_READ) {
6296 8946501 : num_stripes = map->num_stripes;
6297 91148 : } else if (mirror_num) {
6298 174 : stripe_index = mirror_num - 1;
6299 : } else {
6300 : mirror_num = 1;
6301 : }
6302 :
6303 6428200 : } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
6304 0 : u32 factor = map->num_stripes / map->sub_stripes;
6305 :
6306 0 : stripe_index = (stripe_nr % factor) * map->sub_stripes;
6307 0 : stripe_nr /= factor;
6308 :
6309 0 : if (op != BTRFS_MAP_READ)
6310 0 : num_stripes = map->sub_stripes;
6311 0 : else if (mirror_num)
6312 0 : stripe_index += mirror_num - 1;
6313 : else {
6314 0 : int old_stripe_index = stripe_index;
6315 0 : stripe_index = find_live_mirror(fs_info, map,
6316 : stripe_index,
6317 : dev_replace_is_ongoing);
6318 0 : mirror_num = stripe_index - old_stripe_index + 1;
6319 : }
6320 :
6321 6428200 : } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
6322 0 : if (need_raid_map && (op != BTRFS_MAP_READ || mirror_num > 1)) {
6323 : /*
6324 : * Push stripe_nr back to the start of the full stripe
6325 : * For those cases needing a full stripe, @stripe_nr
6326 : * is the full stripe number.
6327 : *
6328 : * Originally we go raid56_full_stripe_start / full_stripe_len,
6329 : * but that can be expensive. Here we just divide
6330 : * @stripe_nr with @data_stripes.
6331 : */
6332 0 : stripe_nr /= data_stripes;
6333 :
6334 : /* RAID[56] write or recovery. Return all stripes */
6335 0 : num_stripes = map->num_stripes;
6336 0 : max_errors = btrfs_chunk_max_errors(map);
6337 :
6338 : /* Return the length to the full stripe end */
6339 0 : *length = min(logical + *length,
6340 : raid56_full_stripe_start + em->start +
6341 0 : btrfs_stripe_nr_to_offset(data_stripes)) -
6342 : logical;
6343 0 : stripe_index = 0;
6344 0 : stripe_offset = 0;
6345 : } else {
6346 : /*
6347 : * Mirror #0 or #1 means the original data block.
6348 : * Mirror #2 is RAID5 parity block.
6349 : * Mirror #3 is RAID6 Q block.
6350 : */
6351 0 : stripe_index = stripe_nr % data_stripes;
6352 0 : stripe_nr /= data_stripes;
6353 0 : if (mirror_num > 1)
6354 0 : stripe_index = data_stripes + mirror_num - 2;
6355 :
6356 : /* We distribute the parity blocks across stripes */
6357 0 : stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
6358 0 : if (op == BTRFS_MAP_READ && mirror_num <= 1)
6359 0 : mirror_num = 1;
6360 : }
6361 : } else {
6362 : /*
6363 : * After this, stripe_nr is the number of stripes on this
6364 : * device we have to walk to find the data, and stripe_index is
6365 : * the number of our device in the stripe array
6366 : */
6367 6428200 : stripe_index = stripe_nr % map->num_stripes;
6368 6428200 : stripe_nr /= map->num_stripes;
6369 6428200 : mirror_num = stripe_index + 1;
6370 : }
6371 15465858 : if (stripe_index >= map->num_stripes) {
6372 0 : btrfs_crit(fs_info,
6373 : "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u",
6374 : stripe_index, map->num_stripes);
6375 0 : ret = -EINVAL;
6376 0 : goto out;
6377 : }
6378 :
6379 15465858 : num_alloc_stripes = num_stripes;
6380 15465858 : if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6381 : op != BTRFS_MAP_READ)
6382 : /*
6383 : * For replace case, we need to add extra stripes for extra
6384 : * duplicated stripes.
6385 : *
6386 : * For both WRITE and GET_READ_MIRRORS, we may have at most
6387 : * 2 more stripes (DUP types, otherwise 1).
6388 : */
6389 0 : num_alloc_stripes += 2;
6390 :
6391 : /*
6392 : * If this I/O maps to a single device, try to return the device and
6393 : * physical block information on the stack instead of allocating an
6394 : * I/O context structure.
6395 : */
6396 15465858 : if (smap && num_alloc_stripes == 1 &&
6397 6518806 : !((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && mirror_num > 1) &&
6398 6518806 : (op == BTRFS_MAP_READ || !dev_replace_is_ongoing ||
6399 0 : !dev_replace->tgtdev)) {
6400 6518806 : set_io_stripe(smap, map, stripe_index, stripe_offset, stripe_nr);
6401 6518806 : if (mirror_num_ret)
6402 6518806 : *mirror_num_ret = mirror_num;
6403 6518806 : *bioc_ret = NULL;
6404 6518806 : ret = 0;
6405 6518806 : goto out;
6406 : }
6407 :
6408 8947052 : bioc = alloc_btrfs_io_context(fs_info, num_alloc_stripes);
6409 8946468 : if (!bioc) {
6410 0 : ret = -ENOMEM;
6411 0 : goto out;
6412 : }
6413 8946468 : bioc->map_type = map->type;
6414 :
6415 : /*
6416 : * For RAID56 full map, we need to make sure the stripes[] follows the
6417 : * rule that data stripes are all ordered, then followed with P and Q
6418 : * (if we have).
6419 : *
6420 : * It's still mostly the same as other profiles, just with extra rotation.
6421 : */
6422 8946468 : if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && need_raid_map &&
6423 0 : (op != BTRFS_MAP_READ || mirror_num > 1)) {
6424 : /*
6425 : * For RAID56 @stripe_nr is already the number of full stripes
6426 : * before us, which is also the rotation value (needs to modulo
6427 : * with num_stripes).
6428 : *
6429 : * In this case, we just add @stripe_nr with @i, then do the
6430 : * modulo, to reduce one modulo call.
6431 : */
6432 0 : bioc->full_stripe_logical = em->start +
6433 0 : btrfs_stripe_nr_to_offset(stripe_nr * data_stripes);
6434 0 : for (i = 0; i < num_stripes; i++)
6435 0 : set_io_stripe(&bioc->stripes[i], map,
6436 0 : (i + stripe_nr) % num_stripes,
6437 : stripe_offset, stripe_nr);
6438 : } else {
6439 : /*
6440 : * For all other non-RAID56 profiles, just copy the target
6441 : * stripe into the bioc.
6442 : */
6443 26839380 : for (i = 0; i < num_stripes; i++) {
6444 17892912 : set_io_stripe(&bioc->stripes[i], map, stripe_index,
6445 : stripe_offset, stripe_nr);
6446 17892912 : stripe_index++;
6447 : }
6448 : }
6449 :
6450 8946468 : if (op != BTRFS_MAP_READ)
6451 8946480 : max_errors = btrfs_chunk_max_errors(map);
6452 :
6453 8946468 : if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL &&
6454 : op != BTRFS_MAP_READ) {
6455 0 : handle_ops_on_dev_replace(op, bioc, dev_replace, logical,
6456 : &num_stripes, &max_errors);
6457 : }
6458 :
6459 8946468 : *bioc_ret = bioc;
6460 8946468 : bioc->num_stripes = num_stripes;
6461 8946468 : bioc->max_errors = max_errors;
6462 8946468 : bioc->mirror_num = mirror_num;
6463 :
6464 15465274 : out:
6465 15465274 : if (dev_replace_is_ongoing) {
6466 0 : lockdep_assert_held(&dev_replace->rwsem);
6467 : /* Unlock and let waiting writers proceed */
6468 0 : up_read(&dev_replace->rwsem);
6469 : }
6470 15465274 : free_extent_map(em);
6471 15465274 : return ret;
6472 : }
6473 :
6474 81558 : static bool dev_args_match_fs_devices(const struct btrfs_dev_lookup_args *args,
6475 : const struct btrfs_fs_devices *fs_devices)
6476 : {
6477 81558 : if (args->fsid == NULL)
6478 : return true;
6479 6438 : if (memcmp(fs_devices->metadata_uuid, args->fsid, BTRFS_FSID_SIZE) == 0)
6480 3219 : return true;
6481 : return false;
6482 : }
6483 :
6484 81572 : static bool dev_args_match_device(const struct btrfs_dev_lookup_args *args,
6485 : const struct btrfs_device *device)
6486 : {
6487 81572 : if (args->missing) {
6488 0 : if (test_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state) &&
6489 0 : !device->bdev)
6490 : return true;
6491 0 : return false;
6492 : }
6493 :
6494 81572 : if (device->devid != args->devid)
6495 : return false;
6496 120793 : if (args->uuid && memcmp(device->uuid, args->uuid, BTRFS_UUID_SIZE) != 0)
6497 0 : return false;
6498 : return true;
6499 : }
6500 :
6501 : /*
6502 : * Find a device specified by @devid or @uuid in the list of @fs_devices, or
6503 : * return NULL.
6504 : *
6505 : * If devid and uuid are both specified, the match must be exact, otherwise
6506 : * only devid is used.
6507 : */
6508 81558 : struct btrfs_device *btrfs_find_device(const struct btrfs_fs_devices *fs_devices,
6509 : const struct btrfs_dev_lookup_args *args)
6510 : {
6511 81558 : struct btrfs_device *device;
6512 81558 : struct btrfs_fs_devices *seed_devs;
6513 :
6514 81558 : if (dev_args_match_fs_devices(args, fs_devices)) {
6515 84815 : list_for_each_entry(device, &fs_devices->devices, dev_list) {
6516 81572 : if (dev_args_match_device(args, device))
6517 78315 : return device;
6518 : }
6519 : }
6520 :
6521 3243 : list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
6522 0 : if (!dev_args_match_fs_devices(args, seed_devs))
6523 0 : continue;
6524 0 : list_for_each_entry(device, &seed_devs->devices, dev_list) {
6525 0 : if (dev_args_match_device(args, device))
6526 0 : return device;
6527 : }
6528 : }
6529 :
6530 : return NULL;
6531 : }
6532 :
6533 0 : static struct btrfs_device *add_missing_dev(struct btrfs_fs_devices *fs_devices,
6534 : u64 devid, u8 *dev_uuid)
6535 : {
6536 0 : struct btrfs_device *device;
6537 0 : unsigned int nofs_flag;
6538 :
6539 : /*
6540 : * We call this under the chunk_mutex, so we want to use NOFS for this
6541 : * allocation, however we don't want to change btrfs_alloc_device() to
6542 : * always do NOFS because we use it in a lot of other GFP_KERNEL safe
6543 : * places.
6544 : */
6545 :
6546 0 : nofs_flag = memalloc_nofs_save();
6547 0 : device = btrfs_alloc_device(NULL, &devid, dev_uuid, NULL);
6548 0 : memalloc_nofs_restore(nofs_flag);
6549 0 : if (IS_ERR(device))
6550 : return device;
6551 :
6552 0 : list_add(&device->dev_list, &fs_devices->devices);
6553 0 : device->fs_devices = fs_devices;
6554 0 : fs_devices->num_devices++;
6555 :
6556 0 : set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6557 0 : fs_devices->missing_devices++;
6558 :
6559 0 : return device;
6560 : }
6561 :
6562 : /*
6563 : * Allocate new device struct, set up devid and UUID.
6564 : *
6565 : * @fs_info: used only for generating a new devid, can be NULL if
6566 : * devid is provided (i.e. @devid != NULL).
6567 : * @devid: a pointer to devid for this device. If NULL a new devid
6568 : * is generated.
6569 : * @uuid: a pointer to UUID for this device. If NULL a new UUID
6570 : * is generated.
6571 : * @path: a pointer to device path if available, NULL otherwise.
6572 : *
6573 : * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6574 : * on error. Returned struct is not linked onto any lists and must be
6575 : * destroyed with btrfs_free_device.
6576 : */
6577 3268 : struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6578 : const u64 *devid, const u8 *uuid,
6579 : const char *path)
6580 : {
6581 3268 : struct btrfs_device *dev;
6582 3268 : u64 tmp;
6583 :
6584 3268 : if (WARN_ON(!devid && !fs_info))
6585 : return ERR_PTR(-EINVAL);
6586 :
6587 3268 : dev = kzalloc(sizeof(*dev), GFP_KERNEL);
6588 3268 : if (!dev)
6589 : return ERR_PTR(-ENOMEM);
6590 :
6591 3268 : INIT_LIST_HEAD(&dev->dev_list);
6592 3268 : INIT_LIST_HEAD(&dev->dev_alloc_list);
6593 3268 : INIT_LIST_HEAD(&dev->post_commit_list);
6594 :
6595 3268 : atomic_set(&dev->dev_stats_ccnt, 0);
6596 3268 : btrfs_device_data_ordered_init(dev);
6597 3268 : extent_io_tree_init(fs_info, &dev->alloc_state, IO_TREE_DEVICE_ALLOC_STATE);
6598 :
6599 3268 : if (devid)
6600 3268 : tmp = *devid;
6601 : else {
6602 0 : int ret;
6603 :
6604 0 : ret = find_next_devid(fs_info, &tmp);
6605 0 : if (ret) {
6606 0 : btrfs_free_device(dev);
6607 0 : return ERR_PTR(ret);
6608 : }
6609 : }
6610 3268 : dev->devid = tmp;
6611 :
6612 3268 : if (uuid)
6613 6536 : memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6614 : else
6615 0 : generate_random_uuid(dev->uuid);
6616 :
6617 3268 : if (path) {
6618 3268 : struct rcu_string *name;
6619 :
6620 3268 : name = rcu_string_strdup(path, GFP_KERNEL);
6621 3268 : if (!name) {
6622 0 : btrfs_free_device(dev);
6623 0 : return ERR_PTR(-ENOMEM);
6624 : }
6625 3268 : rcu_assign_pointer(dev->name, name);
6626 : }
6627 :
6628 : return dev;
6629 : }
6630 :
6631 0 : static void btrfs_report_missing_device(struct btrfs_fs_info *fs_info,
6632 : u64 devid, u8 *uuid, bool error)
6633 : {
6634 0 : if (error)
6635 0 : btrfs_err_rl(fs_info, "devid %llu uuid %pU is missing",
6636 : devid, uuid);
6637 : else
6638 0 : btrfs_warn_rl(fs_info, "devid %llu uuid %pU is missing",
6639 : devid, uuid);
6640 0 : }
6641 :
6642 65135 : u64 btrfs_calc_stripe_length(const struct extent_map *em)
6643 : {
6644 65135 : const struct map_lookup *map = em->map_lookup;
6645 65135 : const int data_stripes = calc_data_stripes(map->type, map->num_stripes);
6646 :
6647 65135 : return div_u64(em->len, data_stripes);
6648 : }
6649 :
6650 : #if BITS_PER_LONG == 32
6651 : /*
6652 : * Due to page cache limit, metadata beyond BTRFS_32BIT_MAX_FILE_SIZE
6653 : * can't be accessed on 32bit systems.
6654 : *
6655 : * This function do mount time check to reject the fs if it already has
6656 : * metadata chunk beyond that limit.
6657 : */
6658 : static int check_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6659 : u64 logical, u64 length, u64 type)
6660 : {
6661 : if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6662 : return 0;
6663 :
6664 : if (logical + length < MAX_LFS_FILESIZE)
6665 : return 0;
6666 :
6667 : btrfs_err_32bit_limit(fs_info);
6668 : return -EOVERFLOW;
6669 : }
6670 :
6671 : /*
6672 : * This is to give early warning for any metadata chunk reaching
6673 : * BTRFS_32BIT_EARLY_WARN_THRESHOLD.
6674 : * Although we can still access the metadata, it's not going to be possible
6675 : * once the limit is reached.
6676 : */
6677 : static void warn_32bit_meta_chunk(struct btrfs_fs_info *fs_info,
6678 : u64 logical, u64 length, u64 type)
6679 : {
6680 : if (!(type & BTRFS_BLOCK_GROUP_METADATA))
6681 : return;
6682 :
6683 : if (logical + length < BTRFS_32BIT_EARLY_WARN_THRESHOLD)
6684 : return;
6685 :
6686 : btrfs_warn_32bit_limit(fs_info);
6687 : }
6688 : #endif
6689 :
6690 0 : static struct btrfs_device *handle_missing_device(struct btrfs_fs_info *fs_info,
6691 : u64 devid, u8 *uuid)
6692 : {
6693 0 : struct btrfs_device *dev;
6694 :
6695 0 : if (!btrfs_test_opt(fs_info, DEGRADED)) {
6696 0 : btrfs_report_missing_device(fs_info, devid, uuid, true);
6697 0 : return ERR_PTR(-ENOENT);
6698 : }
6699 :
6700 0 : dev = add_missing_dev(fs_info->fs_devices, devid, uuid);
6701 0 : if (IS_ERR(dev)) {
6702 0 : btrfs_err(fs_info, "failed to init missing device %llu: %ld",
6703 : devid, PTR_ERR(dev));
6704 0 : return dev;
6705 : }
6706 0 : btrfs_report_missing_device(fs_info, devid, uuid, false);
6707 :
6708 0 : return dev;
6709 : }
6710 :
6711 32531 : static int read_one_chunk(struct btrfs_key *key, struct extent_buffer *leaf,
6712 : struct btrfs_chunk *chunk)
6713 : {
6714 32531 : BTRFS_DEV_LOOKUP_ARGS(args);
6715 32531 : struct btrfs_fs_info *fs_info = leaf->fs_info;
6716 32531 : struct extent_map_tree *map_tree = &fs_info->mapping_tree;
6717 32531 : struct map_lookup *map;
6718 32531 : struct extent_map *em;
6719 32531 : u64 logical;
6720 32531 : u64 length;
6721 32531 : u64 devid;
6722 32531 : u64 type;
6723 32531 : u8 uuid[BTRFS_UUID_SIZE];
6724 32531 : int index;
6725 32531 : int num_stripes;
6726 32531 : int ret;
6727 32531 : int i;
6728 :
6729 32531 : logical = key->offset;
6730 32531 : length = btrfs_chunk_length(leaf, chunk);
6731 32531 : type = btrfs_chunk_type(leaf, chunk);
6732 32531 : index = btrfs_bg_flags_to_raid_index(type);
6733 32531 : num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6734 :
6735 : #if BITS_PER_LONG == 32
6736 : ret = check_32bit_meta_chunk(fs_info, logical, length, type);
6737 : if (ret < 0)
6738 : return ret;
6739 : warn_32bit_meta_chunk(fs_info, logical, length, type);
6740 : #endif
6741 :
6742 : /*
6743 : * Only need to verify chunk item if we're reading from sys chunk array,
6744 : * as chunk item in tree block is already verified by tree-checker.
6745 : */
6746 32531 : if (leaf->start == BTRFS_SUPER_INFO_OFFSET) {
6747 3222 : ret = btrfs_check_chunk_valid(leaf, chunk, logical);
6748 3222 : if (ret)
6749 : return ret;
6750 : }
6751 :
6752 32531 : read_lock(&map_tree->lock);
6753 32531 : em = lookup_extent_mapping(map_tree, logical, 1);
6754 32531 : read_unlock(&map_tree->lock);
6755 :
6756 : /* already mapped? */
6757 32531 : if (em && em->start <= logical && em->start + em->len > logical) {
6758 3222 : free_extent_map(em);
6759 3222 : return 0;
6760 29309 : } else if (em) {
6761 0 : free_extent_map(em);
6762 : }
6763 :
6764 29309 : em = alloc_extent_map();
6765 29309 : if (!em)
6766 : return -ENOMEM;
6767 29309 : map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6768 29309 : if (!map) {
6769 0 : free_extent_map(em);
6770 0 : return -ENOMEM;
6771 : }
6772 :
6773 29309 : set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6774 29309 : em->map_lookup = map;
6775 29309 : em->start = logical;
6776 29309 : em->len = length;
6777 29309 : em->orig_start = 0;
6778 29309 : em->block_start = 0;
6779 29309 : em->block_len = em->len;
6780 :
6781 29309 : map->num_stripes = num_stripes;
6782 29309 : map->io_width = btrfs_chunk_io_width(leaf, chunk);
6783 29309 : map->io_align = btrfs_chunk_io_align(leaf, chunk);
6784 29309 : map->type = type;
6785 : /*
6786 : * We can't use the sub_stripes value, as for profiles other than
6787 : * RAID10, they may have 0 as sub_stripes for filesystems created by
6788 : * older mkfs (<v5.4).
6789 : * In that case, it can cause divide-by-zero errors later.
6790 : * Since currently sub_stripes is fixed for each profile, let's
6791 : * use the trusted value instead.
6792 : */
6793 29309 : map->sub_stripes = btrfs_raid_array[index].sub_stripes;
6794 29309 : map->verified_stripes = 0;
6795 29309 : em->orig_block_len = btrfs_calc_stripe_length(em);
6796 65116 : for (i = 0; i < num_stripes; i++) {
6797 35807 : map->stripes[i].physical =
6798 : btrfs_stripe_offset_nr(leaf, chunk, i);
6799 35807 : devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6800 35807 : args.devid = devid;
6801 35807 : read_extent_buffer(leaf, uuid, (unsigned long)
6802 : btrfs_stripe_dev_uuid_nr(chunk, i),
6803 : BTRFS_UUID_SIZE);
6804 35807 : args.uuid = uuid;
6805 35807 : map->stripes[i].dev = btrfs_find_device(fs_info->fs_devices, &args);
6806 35807 : if (!map->stripes[i].dev) {
6807 0 : map->stripes[i].dev = handle_missing_device(fs_info,
6808 : devid, uuid);
6809 0 : if (IS_ERR(map->stripes[i].dev)) {
6810 0 : ret = PTR_ERR(map->stripes[i].dev);
6811 0 : free_extent_map(em);
6812 0 : return ret;
6813 : }
6814 : }
6815 :
6816 35807 : set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
6817 35807 : &(map->stripes[i].dev->dev_state));
6818 : }
6819 :
6820 29309 : write_lock(&map_tree->lock);
6821 29309 : ret = add_extent_mapping(map_tree, em, 0);
6822 29309 : write_unlock(&map_tree->lock);
6823 29309 : if (ret < 0) {
6824 0 : btrfs_err(fs_info,
6825 : "failed to add chunk map, start=%llu len=%llu: %d",
6826 : em->start, em->len, ret);
6827 : }
6828 29309 : free_extent_map(em);
6829 :
6830 29309 : return ret;
6831 : }
6832 :
6833 3219 : static void fill_device_from_item(struct extent_buffer *leaf,
6834 : struct btrfs_dev_item *dev_item,
6835 : struct btrfs_device *device)
6836 : {
6837 3219 : unsigned long ptr;
6838 :
6839 3219 : device->devid = btrfs_device_id(leaf, dev_item);
6840 3219 : device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6841 3219 : device->total_bytes = device->disk_total_bytes;
6842 3219 : device->commit_total_bytes = device->disk_total_bytes;
6843 3219 : device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6844 3219 : device->commit_bytes_used = device->bytes_used;
6845 3219 : device->type = btrfs_device_type(leaf, dev_item);
6846 3219 : device->io_align = btrfs_device_io_align(leaf, dev_item);
6847 3219 : device->io_width = btrfs_device_io_width(leaf, dev_item);
6848 3219 : device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6849 3219 : WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6850 3219 : clear_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state);
6851 :
6852 3219 : ptr = btrfs_device_uuid(dev_item);
6853 3219 : read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6854 3219 : }
6855 :
6856 0 : static struct btrfs_fs_devices *open_seed_devices(struct btrfs_fs_info *fs_info,
6857 : u8 *fsid)
6858 : {
6859 0 : struct btrfs_fs_devices *fs_devices;
6860 0 : int ret;
6861 :
6862 0 : lockdep_assert_held(&uuid_mutex);
6863 0 : ASSERT(fsid);
6864 :
6865 : /* This will match only for multi-device seed fs */
6866 0 : list_for_each_entry(fs_devices, &fs_info->fs_devices->seed_list, seed_list)
6867 0 : if (!memcmp(fs_devices->fsid, fsid, BTRFS_FSID_SIZE))
6868 0 : return fs_devices;
6869 :
6870 :
6871 0 : fs_devices = find_fsid(fsid, NULL);
6872 0 : if (!fs_devices) {
6873 0 : if (!btrfs_test_opt(fs_info, DEGRADED))
6874 : return ERR_PTR(-ENOENT);
6875 :
6876 0 : fs_devices = alloc_fs_devices(fsid, NULL);
6877 0 : if (IS_ERR(fs_devices))
6878 : return fs_devices;
6879 :
6880 0 : fs_devices->seeding = true;
6881 0 : fs_devices->opened = 1;
6882 0 : return fs_devices;
6883 : }
6884 :
6885 : /*
6886 : * Upon first call for a seed fs fsid, just create a private copy of the
6887 : * respective fs_devices and anchor it at fs_info->fs_devices->seed_list
6888 : */
6889 0 : fs_devices = clone_fs_devices(fs_devices);
6890 0 : if (IS_ERR(fs_devices))
6891 : return fs_devices;
6892 :
6893 0 : ret = open_fs_devices(fs_devices, BLK_OPEN_READ, fs_info->bdev_holder);
6894 0 : if (ret) {
6895 0 : free_fs_devices(fs_devices);
6896 0 : return ERR_PTR(ret);
6897 : }
6898 :
6899 0 : if (!fs_devices->seeding) {
6900 0 : close_fs_devices(fs_devices);
6901 0 : free_fs_devices(fs_devices);
6902 0 : return ERR_PTR(-EINVAL);
6903 : }
6904 :
6905 0 : list_add(&fs_devices->seed_list, &fs_info->fs_devices->seed_list);
6906 :
6907 0 : return fs_devices;
6908 : }
6909 :
6910 3219 : static int read_one_dev(struct extent_buffer *leaf,
6911 : struct btrfs_dev_item *dev_item)
6912 : {
6913 3219 : BTRFS_DEV_LOOKUP_ARGS(args);
6914 3219 : struct btrfs_fs_info *fs_info = leaf->fs_info;
6915 3219 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6916 3219 : struct btrfs_device *device;
6917 3219 : u64 devid;
6918 3219 : int ret;
6919 3219 : u8 fs_uuid[BTRFS_FSID_SIZE];
6920 3219 : u8 dev_uuid[BTRFS_UUID_SIZE];
6921 :
6922 3219 : devid = btrfs_device_id(leaf, dev_item);
6923 3219 : args.devid = devid;
6924 3219 : read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6925 : BTRFS_UUID_SIZE);
6926 3219 : read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6927 : BTRFS_FSID_SIZE);
6928 3219 : args.uuid = dev_uuid;
6929 3219 : args.fsid = fs_uuid;
6930 :
6931 6438 : if (memcmp(fs_uuid, fs_devices->metadata_uuid, BTRFS_FSID_SIZE)) {
6932 0 : fs_devices = open_seed_devices(fs_info, fs_uuid);
6933 0 : if (IS_ERR(fs_devices))
6934 0 : return PTR_ERR(fs_devices);
6935 : }
6936 :
6937 3219 : device = btrfs_find_device(fs_info->fs_devices, &args);
6938 3219 : if (!device) {
6939 0 : if (!btrfs_test_opt(fs_info, DEGRADED)) {
6940 0 : btrfs_report_missing_device(fs_info, devid,
6941 : dev_uuid, true);
6942 0 : return -ENOENT;
6943 : }
6944 :
6945 0 : device = add_missing_dev(fs_devices, devid, dev_uuid);
6946 0 : if (IS_ERR(device)) {
6947 0 : btrfs_err(fs_info,
6948 : "failed to add missing dev %llu: %ld",
6949 : devid, PTR_ERR(device));
6950 0 : return PTR_ERR(device);
6951 : }
6952 0 : btrfs_report_missing_device(fs_info, devid, dev_uuid, false);
6953 : } else {
6954 3219 : if (!device->bdev) {
6955 0 : if (!btrfs_test_opt(fs_info, DEGRADED)) {
6956 0 : btrfs_report_missing_device(fs_info,
6957 : devid, dev_uuid, true);
6958 0 : return -ENOENT;
6959 : }
6960 0 : btrfs_report_missing_device(fs_info, devid,
6961 : dev_uuid, false);
6962 : }
6963 :
6964 3219 : if (!device->bdev &&
6965 0 : !test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
6966 : /*
6967 : * this happens when a device that was properly setup
6968 : * in the device info lists suddenly goes bad.
6969 : * device->bdev is NULL, and so we have to set
6970 : * device->missing to one here
6971 : */
6972 0 : device->fs_devices->missing_devices++;
6973 0 : set_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
6974 : }
6975 :
6976 : /* Move the device to its own fs_devices */
6977 3219 : if (device->fs_devices != fs_devices) {
6978 0 : ASSERT(test_bit(BTRFS_DEV_STATE_MISSING,
6979 : &device->dev_state));
6980 :
6981 0 : list_move(&device->dev_list, &fs_devices->devices);
6982 0 : device->fs_devices->num_devices--;
6983 0 : fs_devices->num_devices++;
6984 :
6985 0 : device->fs_devices->missing_devices--;
6986 0 : fs_devices->missing_devices++;
6987 :
6988 0 : device->fs_devices = fs_devices;
6989 : }
6990 : }
6991 :
6992 3219 : if (device->fs_devices != fs_info->fs_devices) {
6993 0 : BUG_ON(test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state));
6994 0 : if (device->generation !=
6995 : btrfs_device_generation(leaf, dev_item))
6996 : return -EINVAL;
6997 : }
6998 :
6999 3219 : fill_device_from_item(leaf, dev_item, device);
7000 3219 : if (device->bdev) {
7001 3219 : u64 max_total_bytes = bdev_nr_bytes(device->bdev);
7002 :
7003 3219 : if (device->total_bytes > max_total_bytes) {
7004 0 : btrfs_err(fs_info,
7005 : "device total_bytes should be at most %llu but found %llu",
7006 : max_total_bytes, device->total_bytes);
7007 0 : return -EINVAL;
7008 : }
7009 : }
7010 3219 : set_bit(BTRFS_DEV_STATE_IN_FS_METADATA, &device->dev_state);
7011 6438 : if (test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state) &&
7012 0 : !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
7013 3219 : device->fs_devices->total_rw_bytes += device->total_bytes;
7014 3219 : atomic64_add(device->total_bytes - device->bytes_used,
7015 : &fs_info->free_chunk_space);
7016 : }
7017 : ret = 0;
7018 : return ret;
7019 : }
7020 :
7021 3218 : int btrfs_read_sys_array(struct btrfs_fs_info *fs_info)
7022 : {
7023 3218 : struct btrfs_super_block *super_copy = fs_info->super_copy;
7024 3218 : struct extent_buffer *sb;
7025 3218 : struct btrfs_disk_key *disk_key;
7026 3218 : struct btrfs_chunk *chunk;
7027 3218 : u8 *array_ptr;
7028 3218 : unsigned long sb_array_offset;
7029 3218 : int ret = 0;
7030 3218 : u32 num_stripes;
7031 3218 : u32 array_size;
7032 3218 : u32 len = 0;
7033 3218 : u32 cur_offset;
7034 3218 : u64 type;
7035 3218 : struct btrfs_key key;
7036 :
7037 3218 : ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize);
7038 :
7039 : /*
7040 : * We allocated a dummy extent, just to use extent buffer accessors.
7041 : * There will be unused space after BTRFS_SUPER_INFO_SIZE, but
7042 : * that's fine, we will not go beyond system chunk array anyway.
7043 : */
7044 3218 : sb = alloc_dummy_extent_buffer(fs_info, BTRFS_SUPER_INFO_OFFSET);
7045 3218 : if (!sb)
7046 : return -ENOMEM;
7047 3218 : set_extent_buffer_uptodate(sb);
7048 :
7049 3218 : write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
7050 3218 : array_size = btrfs_super_sys_array_size(super_copy);
7051 :
7052 3218 : array_ptr = super_copy->sys_chunk_array;
7053 3218 : sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
7054 3218 : cur_offset = 0;
7055 :
7056 6440 : while (cur_offset < array_size) {
7057 3222 : disk_key = (struct btrfs_disk_key *)array_ptr;
7058 3222 : len = sizeof(*disk_key);
7059 3222 : if (cur_offset + len > array_size)
7060 0 : goto out_short_read;
7061 :
7062 3222 : btrfs_disk_key_to_cpu(&key, disk_key);
7063 :
7064 3222 : array_ptr += len;
7065 3222 : sb_array_offset += len;
7066 3222 : cur_offset += len;
7067 :
7068 3222 : if (key.type != BTRFS_CHUNK_ITEM_KEY) {
7069 0 : btrfs_err(fs_info,
7070 : "unexpected item type %u in sys_array at offset %u",
7071 : (u32)key.type, cur_offset);
7072 0 : ret = -EIO;
7073 0 : break;
7074 : }
7075 :
7076 3222 : chunk = (struct btrfs_chunk *)sb_array_offset;
7077 : /*
7078 : * At least one btrfs_chunk with one stripe must be present,
7079 : * exact stripe count check comes afterwards
7080 : */
7081 3222 : len = btrfs_chunk_item_size(1);
7082 3222 : if (cur_offset + len > array_size)
7083 0 : goto out_short_read;
7084 :
7085 3222 : num_stripes = btrfs_chunk_num_stripes(sb, chunk);
7086 3222 : if (!num_stripes) {
7087 0 : btrfs_err(fs_info,
7088 : "invalid number of stripes %u in sys_array at offset %u",
7089 : num_stripes, cur_offset);
7090 0 : ret = -EIO;
7091 0 : break;
7092 : }
7093 :
7094 3222 : type = btrfs_chunk_type(sb, chunk);
7095 3222 : if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
7096 0 : btrfs_err(fs_info,
7097 : "invalid chunk type %llu in sys_array at offset %u",
7098 : type, cur_offset);
7099 0 : ret = -EIO;
7100 0 : break;
7101 : }
7102 :
7103 3222 : len = btrfs_chunk_item_size(num_stripes);
7104 3222 : if (cur_offset + len > array_size)
7105 0 : goto out_short_read;
7106 :
7107 3222 : ret = read_one_chunk(&key, sb, chunk);
7108 3222 : if (ret)
7109 : break;
7110 :
7111 3222 : array_ptr += len;
7112 3222 : sb_array_offset += len;
7113 3222 : cur_offset += len;
7114 : }
7115 3218 : clear_extent_buffer_uptodate(sb);
7116 3218 : free_extent_buffer_stale(sb);
7117 3218 : return ret;
7118 :
7119 0 : out_short_read:
7120 0 : btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u",
7121 : len, cur_offset);
7122 0 : clear_extent_buffer_uptodate(sb);
7123 0 : free_extent_buffer_stale(sb);
7124 0 : return -EIO;
7125 : }
7126 :
7127 : /*
7128 : * Check if all chunks in the fs are OK for read-write degraded mount
7129 : *
7130 : * If the @failing_dev is specified, it's accounted as missing.
7131 : *
7132 : * Return true if all chunks meet the minimal RW mount requirements.
7133 : * Return false if any chunk doesn't meet the minimal RW mount requirements.
7134 : */
7135 2 : bool btrfs_check_rw_degradable(struct btrfs_fs_info *fs_info,
7136 : struct btrfs_device *failing_dev)
7137 : {
7138 2 : struct extent_map_tree *map_tree = &fs_info->mapping_tree;
7139 2 : struct extent_map *em;
7140 2 : u64 next_start = 0;
7141 2 : bool ret = true;
7142 :
7143 2 : read_lock(&map_tree->lock);
7144 2 : em = lookup_extent_mapping(map_tree, 0, (u64)-1);
7145 2 : read_unlock(&map_tree->lock);
7146 : /* No chunk at all? Return false anyway */
7147 2 : if (!em) {
7148 0 : ret = false;
7149 0 : goto out;
7150 : }
7151 8 : while (em) {
7152 6 : struct map_lookup *map;
7153 6 : int missing = 0;
7154 6 : int max_tolerated;
7155 6 : int i;
7156 :
7157 6 : map = em->map_lookup;
7158 6 : max_tolerated =
7159 6 : btrfs_get_num_tolerated_disk_barrier_failures(
7160 : map->type);
7161 22 : for (i = 0; i < map->num_stripes; i++) {
7162 10 : struct btrfs_device *dev = map->stripes[i].dev;
7163 :
7164 20 : if (!dev || !dev->bdev ||
7165 10 : test_bit(BTRFS_DEV_STATE_MISSING, &dev->dev_state) ||
7166 10 : dev->last_flush_error)
7167 0 : missing++;
7168 10 : else if (failing_dev && failing_dev == dev)
7169 0 : missing++;
7170 : }
7171 6 : if (missing > max_tolerated) {
7172 0 : if (!failing_dev)
7173 0 : btrfs_warn(fs_info,
7174 : "chunk %llu missing %d devices, max tolerance is %d for writable mount",
7175 : em->start, missing, max_tolerated);
7176 0 : free_extent_map(em);
7177 0 : ret = false;
7178 0 : goto out;
7179 : }
7180 6 : next_start = extent_map_end(em);
7181 6 : free_extent_map(em);
7182 :
7183 6 : read_lock(&map_tree->lock);
7184 6 : em = lookup_extent_mapping(map_tree, next_start,
7185 : (u64)(-1) - next_start);
7186 6 : read_unlock(&map_tree->lock);
7187 : }
7188 2 : out:
7189 2 : return ret;
7190 : }
7191 :
7192 1 : static void readahead_tree_node_children(struct extent_buffer *node)
7193 : {
7194 1 : int i;
7195 1 : const int nr_items = btrfs_header_nritems(node);
7196 :
7197 3 : for (i = 0; i < nr_items; i++)
7198 2 : btrfs_readahead_node_child(node, i);
7199 1 : }
7200 :
7201 3218 : int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info)
7202 : {
7203 3218 : struct btrfs_root *root = fs_info->chunk_root;
7204 3218 : struct btrfs_path *path;
7205 3218 : struct extent_buffer *leaf;
7206 3218 : struct btrfs_key key;
7207 3218 : struct btrfs_key found_key;
7208 3218 : int ret;
7209 3218 : int slot;
7210 3218 : int iter_ret = 0;
7211 3218 : u64 total_dev = 0;
7212 3218 : u64 last_ra_node = 0;
7213 :
7214 3218 : path = btrfs_alloc_path();
7215 3218 : if (!path)
7216 : return -ENOMEM;
7217 :
7218 : /*
7219 : * uuid_mutex is needed only if we are mounting a sprout FS
7220 : * otherwise we don't need it.
7221 : */
7222 3218 : mutex_lock(&uuid_mutex);
7223 :
7224 : /*
7225 : * It is possible for mount and umount to race in such a way that
7226 : * we execute this code path, but open_fs_devices failed to clear
7227 : * total_rw_bytes. We certainly want it cleared before reading the
7228 : * device items, so clear it here.
7229 : */
7230 3218 : fs_info->fs_devices->total_rw_bytes = 0;
7231 :
7232 : /*
7233 : * Lockdep complains about possible circular locking dependency between
7234 : * a disk's open_mutex (struct gendisk.open_mutex), the rw semaphores
7235 : * used for freeze procection of a fs (struct super_block.s_writers),
7236 : * which we take when starting a transaction, and extent buffers of the
7237 : * chunk tree if we call read_one_dev() while holding a lock on an
7238 : * extent buffer of the chunk tree. Since we are mounting the filesystem
7239 : * and at this point there can't be any concurrent task modifying the
7240 : * chunk tree, to keep it simple, just skip locking on the chunk tree.
7241 : */
7242 3218 : ASSERT(!test_bit(BTRFS_FS_OPEN, &fs_info->flags));
7243 3218 : path->skip_locking = 1;
7244 :
7245 : /*
7246 : * Read all device items, and then all the chunk items. All
7247 : * device items are found before any chunk item (their object id
7248 : * is smaller than the lowest possible object id for a chunk
7249 : * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
7250 : */
7251 3218 : key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
7252 3218 : key.offset = 0;
7253 3218 : key.type = 0;
7254 35746 : btrfs_for_each_slot(root, &key, &found_key, path, iter_ret) {
7255 32528 : struct extent_buffer *node = path->nodes[1];
7256 :
7257 32528 : leaf = path->nodes[0];
7258 32528 : slot = path->slots[0];
7259 :
7260 32528 : if (node) {
7261 266 : if (last_ra_node != node->start) {
7262 1 : readahead_tree_node_children(node);
7263 1 : last_ra_node = node->start;
7264 : }
7265 : }
7266 32528 : if (found_key.type == BTRFS_DEV_ITEM_KEY) {
7267 3219 : struct btrfs_dev_item *dev_item;
7268 3219 : dev_item = btrfs_item_ptr(leaf, slot,
7269 : struct btrfs_dev_item);
7270 3219 : ret = read_one_dev(leaf, dev_item);
7271 3219 : if (ret)
7272 0 : goto error;
7273 3219 : total_dev++;
7274 29309 : } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
7275 29309 : struct btrfs_chunk *chunk;
7276 :
7277 : /*
7278 : * We are only called at mount time, so no need to take
7279 : * fs_info->chunk_mutex. Plus, to avoid lockdep warnings,
7280 : * we always lock first fs_info->chunk_mutex before
7281 : * acquiring any locks on the chunk tree. This is a
7282 : * requirement for chunk allocation, see the comment on
7283 : * top of btrfs_chunk_alloc() for details.
7284 : */
7285 29309 : chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
7286 29309 : ret = read_one_chunk(&found_key, leaf, chunk);
7287 29309 : if (ret)
7288 0 : goto error;
7289 : }
7290 : }
7291 : /* Catch error found during iteration */
7292 3218 : if (iter_ret < 0) {
7293 0 : ret = iter_ret;
7294 0 : goto error;
7295 : }
7296 :
7297 : /*
7298 : * After loading chunk tree, we've got all device information,
7299 : * do another round of validation checks.
7300 : */
7301 3218 : if (total_dev != fs_info->fs_devices->total_devices) {
7302 0 : btrfs_warn(fs_info,
7303 : "super block num_devices %llu mismatch with DEV_ITEM count %llu, will be repaired on next transaction commit",
7304 : btrfs_super_num_devices(fs_info->super_copy),
7305 : total_dev);
7306 0 : fs_info->fs_devices->total_devices = total_dev;
7307 0 : btrfs_set_super_num_devices(fs_info->super_copy, total_dev);
7308 : }
7309 3218 : if (btrfs_super_total_bytes(fs_info->super_copy) <
7310 3218 : fs_info->fs_devices->total_rw_bytes) {
7311 0 : btrfs_err(fs_info,
7312 : "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu",
7313 : btrfs_super_total_bytes(fs_info->super_copy),
7314 : fs_info->fs_devices->total_rw_bytes);
7315 0 : ret = -EINVAL;
7316 0 : goto error;
7317 : }
7318 : ret = 0;
7319 3218 : error:
7320 3218 : mutex_unlock(&uuid_mutex);
7321 :
7322 3218 : btrfs_free_path(path);
7323 3218 : return ret;
7324 : }
7325 :
7326 3217 : int btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
7327 : {
7328 3217 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7329 3217 : struct btrfs_device *device;
7330 3217 : int ret = 0;
7331 :
7332 3217 : fs_devices->fs_info = fs_info;
7333 :
7334 3217 : mutex_lock(&fs_devices->device_list_mutex);
7335 6435 : list_for_each_entry(device, &fs_devices->devices, dev_list)
7336 3218 : device->fs_info = fs_info;
7337 :
7338 3217 : list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7339 0 : list_for_each_entry(device, &seed_devs->devices, dev_list) {
7340 0 : device->fs_info = fs_info;
7341 0 : ret = btrfs_get_dev_zone_info(device, false);
7342 0 : if (ret)
7343 : break;
7344 : }
7345 :
7346 0 : seed_devs->fs_info = fs_info;
7347 : }
7348 3217 : mutex_unlock(&fs_devices->device_list_mutex);
7349 :
7350 3217 : return ret;
7351 : }
7352 :
7353 : static u64 btrfs_dev_stats_value(const struct extent_buffer *eb,
7354 : const struct btrfs_dev_stats_item *ptr,
7355 : int index)
7356 : {
7357 11165 : u64 val;
7358 :
7359 11165 : read_extent_buffer(eb, &val,
7360 : offsetof(struct btrfs_dev_stats_item, values) +
7361 11165 : ((unsigned long)ptr) + (index * sizeof(u64)),
7362 : sizeof(val));
7363 11165 : return val;
7364 : }
7365 :
7366 : static void btrfs_set_dev_stats_value(struct extent_buffer *eb,
7367 : struct btrfs_dev_stats_item *ptr,
7368 : int index, u64 val)
7369 : {
7370 15955 : write_extent_buffer(eb, &val,
7371 : offsetof(struct btrfs_dev_stats_item, values) +
7372 15955 : ((unsigned long)ptr) + (index * sizeof(u64)),
7373 : sizeof(val));
7374 : }
7375 :
7376 3218 : static int btrfs_device_init_dev_stats(struct btrfs_device *device,
7377 : struct btrfs_path *path)
7378 : {
7379 3218 : struct btrfs_dev_stats_item *ptr;
7380 3218 : struct extent_buffer *eb;
7381 3218 : struct btrfs_key key;
7382 3218 : int item_size;
7383 3218 : int i, ret, slot;
7384 :
7385 3218 : if (!device->fs_info->dev_root)
7386 : return 0;
7387 :
7388 3218 : key.objectid = BTRFS_DEV_STATS_OBJECTID;
7389 3218 : key.type = BTRFS_PERSISTENT_ITEM_KEY;
7390 3218 : key.offset = device->devid;
7391 3218 : ret = btrfs_search_slot(NULL, device->fs_info->dev_root, &key, path, 0, 0);
7392 3218 : if (ret) {
7393 5910 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7394 4925 : btrfs_dev_stat_set(device, i, 0);
7395 985 : device->dev_stats_valid = 1;
7396 985 : btrfs_release_path(path);
7397 985 : return ret < 0 ? ret : 0;
7398 : }
7399 2233 : slot = path->slots[0];
7400 2233 : eb = path->nodes[0];
7401 2233 : item_size = btrfs_item_size(eb, slot);
7402 :
7403 2233 : ptr = btrfs_item_ptr(eb, slot, struct btrfs_dev_stats_item);
7404 :
7405 13398 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7406 11165 : if (item_size >= (1 + i) * sizeof(__le64))
7407 11165 : btrfs_dev_stat_set(device, i,
7408 : btrfs_dev_stats_value(eb, ptr, i));
7409 : else
7410 0 : btrfs_dev_stat_set(device, i, 0);
7411 : }
7412 :
7413 2233 : device->dev_stats_valid = 1;
7414 2233 : btrfs_dev_stat_print_on_load(device);
7415 2233 : btrfs_release_path(path);
7416 :
7417 2233 : return 0;
7418 : }
7419 :
7420 3217 : int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
7421 : {
7422 3217 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices, *seed_devs;
7423 3217 : struct btrfs_device *device;
7424 3217 : struct btrfs_path *path = NULL;
7425 3217 : int ret = 0;
7426 :
7427 3217 : path = btrfs_alloc_path();
7428 3217 : if (!path)
7429 : return -ENOMEM;
7430 :
7431 3217 : mutex_lock(&fs_devices->device_list_mutex);
7432 6435 : list_for_each_entry(device, &fs_devices->devices, dev_list) {
7433 3218 : ret = btrfs_device_init_dev_stats(device, path);
7434 3218 : if (ret)
7435 0 : goto out;
7436 : }
7437 3217 : list_for_each_entry(seed_devs, &fs_devices->seed_list, seed_list) {
7438 0 : list_for_each_entry(device, &seed_devs->devices, dev_list) {
7439 0 : ret = btrfs_device_init_dev_stats(device, path);
7440 0 : if (ret)
7441 0 : goto out;
7442 : }
7443 : }
7444 3217 : out:
7445 3217 : mutex_unlock(&fs_devices->device_list_mutex);
7446 :
7447 3217 : btrfs_free_path(path);
7448 3217 : return ret;
7449 : }
7450 :
7451 3191 : static int update_dev_stat_item(struct btrfs_trans_handle *trans,
7452 : struct btrfs_device *device)
7453 : {
7454 3191 : struct btrfs_fs_info *fs_info = trans->fs_info;
7455 3191 : struct btrfs_root *dev_root = fs_info->dev_root;
7456 3191 : struct btrfs_path *path;
7457 3191 : struct btrfs_key key;
7458 3191 : struct extent_buffer *eb;
7459 3191 : struct btrfs_dev_stats_item *ptr;
7460 3191 : int ret;
7461 3191 : int i;
7462 :
7463 3191 : key.objectid = BTRFS_DEV_STATS_OBJECTID;
7464 3191 : key.type = BTRFS_PERSISTENT_ITEM_KEY;
7465 3191 : key.offset = device->devid;
7466 :
7467 3191 : path = btrfs_alloc_path();
7468 3191 : if (!path)
7469 : return -ENOMEM;
7470 3191 : ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
7471 3191 : if (ret < 0) {
7472 0 : btrfs_warn_in_rcu(fs_info,
7473 : "error %d while searching for dev_stats item for device %s",
7474 : ret, btrfs_dev_name(device));
7475 0 : goto out;
7476 : }
7477 :
7478 5403 : if (ret == 0 &&
7479 2212 : btrfs_item_size(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
7480 : /* need to delete old one and insert a new one */
7481 0 : ret = btrfs_del_item(trans, dev_root, path);
7482 0 : if (ret != 0) {
7483 0 : btrfs_warn_in_rcu(fs_info,
7484 : "delete too small dev_stats item for device %s failed %d",
7485 : btrfs_dev_name(device), ret);
7486 0 : goto out;
7487 : }
7488 : ret = 1;
7489 : }
7490 :
7491 3191 : if (ret == 1) {
7492 : /* need to insert a new item */
7493 979 : btrfs_release_path(path);
7494 979 : ret = btrfs_insert_empty_item(trans, dev_root, path,
7495 : &key, sizeof(*ptr));
7496 979 : if (ret < 0) {
7497 0 : btrfs_warn_in_rcu(fs_info,
7498 : "insert dev_stats item for device %s failed %d",
7499 : btrfs_dev_name(device), ret);
7500 0 : goto out;
7501 : }
7502 : }
7503 :
7504 3191 : eb = path->nodes[0];
7505 3191 : ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
7506 19146 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7507 15955 : btrfs_set_dev_stats_value(eb, ptr, i,
7508 : btrfs_dev_stat_read(device, i));
7509 3191 : btrfs_mark_buffer_dirty(eb);
7510 :
7511 3191 : out:
7512 3191 : btrfs_free_path(path);
7513 3191 : return ret;
7514 : }
7515 :
7516 : /*
7517 : * called from commit_transaction. Writes all changed device stats to disk.
7518 : */
7519 206349 : int btrfs_run_dev_stats(struct btrfs_trans_handle *trans)
7520 : {
7521 206349 : struct btrfs_fs_info *fs_info = trans->fs_info;
7522 206349 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7523 206349 : struct btrfs_device *device;
7524 206349 : int stats_cnt;
7525 206349 : int ret = 0;
7526 :
7527 206349 : mutex_lock(&fs_devices->device_list_mutex);
7528 412698 : list_for_each_entry(device, &fs_devices->devices, dev_list) {
7529 206349 : stats_cnt = atomic_read(&device->dev_stats_ccnt);
7530 206349 : if (!device->dev_stats_valid || stats_cnt == 0)
7531 203158 : continue;
7532 :
7533 :
7534 : /*
7535 : * There is a LOAD-LOAD control dependency between the value of
7536 : * dev_stats_ccnt and updating the on-disk values which requires
7537 : * reading the in-memory counters. Such control dependencies
7538 : * require explicit read memory barriers.
7539 : *
7540 : * This memory barriers pairs with smp_mb__before_atomic in
7541 : * btrfs_dev_stat_inc/btrfs_dev_stat_set and with the full
7542 : * barrier implied by atomic_xchg in
7543 : * btrfs_dev_stats_read_and_reset
7544 : */
7545 3191 : smp_rmb();
7546 :
7547 3191 : ret = update_dev_stat_item(trans, device);
7548 3191 : if (!ret)
7549 3191 : atomic_sub(stats_cnt, &device->dev_stats_ccnt);
7550 : }
7551 206349 : mutex_unlock(&fs_devices->device_list_mutex);
7552 :
7553 206349 : return ret;
7554 : }
7555 :
7556 247 : void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
7557 : {
7558 247 : btrfs_dev_stat_inc(dev, index);
7559 :
7560 247 : if (!dev->dev_stats_valid)
7561 : return;
7562 247 : btrfs_err_rl_in_rcu(dev->fs_info,
7563 : "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7564 : btrfs_dev_name(dev),
7565 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7566 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7567 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7568 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7569 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7570 : }
7571 :
7572 2233 : static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
7573 : {
7574 2233 : int i;
7575 :
7576 13341 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7577 11120 : if (btrfs_dev_stat_read(dev, i) != 0)
7578 : break;
7579 2233 : if (i == BTRFS_DEV_STAT_VALUES_MAX)
7580 : return; /* all values == 0, suppress message */
7581 :
7582 12 : btrfs_info_in_rcu(dev->fs_info,
7583 : "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u",
7584 : btrfs_dev_name(dev),
7585 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
7586 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
7587 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
7588 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
7589 : btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
7590 : }
7591 :
7592 2 : int btrfs_get_dev_stats(struct btrfs_fs_info *fs_info,
7593 : struct btrfs_ioctl_get_dev_stats *stats)
7594 : {
7595 2 : BTRFS_DEV_LOOKUP_ARGS(args);
7596 2 : struct btrfs_device *dev;
7597 2 : struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
7598 2 : int i;
7599 :
7600 2 : mutex_lock(&fs_devices->device_list_mutex);
7601 2 : args.devid = stats->devid;
7602 2 : dev = btrfs_find_device(fs_info->fs_devices, &args);
7603 2 : mutex_unlock(&fs_devices->device_list_mutex);
7604 :
7605 2 : if (!dev) {
7606 0 : btrfs_warn(fs_info, "get dev_stats failed, device not found");
7607 0 : return -ENODEV;
7608 2 : } else if (!dev->dev_stats_valid) {
7609 0 : btrfs_warn(fs_info, "get dev_stats failed, not yet valid");
7610 0 : return -ENODEV;
7611 2 : } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
7612 0 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
7613 0 : if (stats->nr_items > i)
7614 0 : stats->values[i] =
7615 0 : btrfs_dev_stat_read_and_reset(dev, i);
7616 : else
7617 0 : btrfs_dev_stat_set(dev, i, 0);
7618 : }
7619 0 : btrfs_info(fs_info, "device stats zeroed by %s (%d)",
7620 : current->comm, task_pid_nr(current));
7621 : } else {
7622 12 : for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
7623 10 : if (stats->nr_items > i)
7624 10 : stats->values[i] = btrfs_dev_stat_read(dev, i);
7625 : }
7626 2 : if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
7627 0 : stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
7628 : return 0;
7629 : }
7630 :
7631 : /*
7632 : * Update the size and bytes used for each device where it changed. This is
7633 : * delayed since we would otherwise get errors while writing out the
7634 : * superblocks.
7635 : *
7636 : * Must be invoked during transaction commit.
7637 : */
7638 206174 : void btrfs_commit_device_sizes(struct btrfs_transaction *trans)
7639 : {
7640 206174 : struct btrfs_device *curr, *next;
7641 :
7642 206174 : ASSERT(trans->state == TRANS_STATE_COMMIT_DOING);
7643 :
7644 206174 : if (list_empty(&trans->dev_update_list))
7645 : return;
7646 :
7647 : /*
7648 : * We don't need the device_list_mutex here. This list is owned by the
7649 : * transaction and the transaction must complete before the device is
7650 : * released.
7651 : */
7652 973 : mutex_lock(&trans->fs_info->chunk_mutex);
7653 1946 : list_for_each_entry_safe(curr, next, &trans->dev_update_list,
7654 : post_commit_list) {
7655 973 : list_del_init(&curr->post_commit_list);
7656 973 : curr->commit_total_bytes = curr->disk_total_bytes;
7657 973 : curr->commit_bytes_used = curr->bytes_used;
7658 : }
7659 973 : mutex_unlock(&trans->fs_info->chunk_mutex);
7660 : }
7661 :
7662 : /*
7663 : * Multiplicity factor for simple profiles: DUP, RAID1-like and RAID10.
7664 : */
7665 92263581 : int btrfs_bg_type_to_factor(u64 flags)
7666 : {
7667 92263581 : const int index = btrfs_bg_flags_to_raid_index(flags);
7668 :
7669 92263581 : return btrfs_raid_array[index].ncopies;
7670 : }
7671 :
7672 :
7673 :
7674 35804 : static int verify_one_dev_extent(struct btrfs_fs_info *fs_info,
7675 : u64 chunk_offset, u64 devid,
7676 : u64 physical_offset, u64 physical_len)
7677 : {
7678 35804 : struct btrfs_dev_lookup_args args = { .devid = devid };
7679 35804 : struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7680 35804 : struct extent_map *em;
7681 35804 : struct map_lookup *map;
7682 35804 : struct btrfs_device *dev;
7683 35804 : u64 stripe_len;
7684 35804 : bool found = false;
7685 35804 : int ret = 0;
7686 35804 : int i;
7687 :
7688 35804 : read_lock(&em_tree->lock);
7689 35804 : em = lookup_extent_mapping(em_tree, chunk_offset, 1);
7690 35804 : read_unlock(&em_tree->lock);
7691 :
7692 35804 : if (!em) {
7693 0 : btrfs_err(fs_info,
7694 : "dev extent physical offset %llu on devid %llu doesn't have corresponding chunk",
7695 : physical_offset, devid);
7696 0 : ret = -EUCLEAN;
7697 0 : goto out;
7698 : }
7699 :
7700 35804 : map = em->map_lookup;
7701 35804 : stripe_len = btrfs_calc_stripe_length(em);
7702 35804 : if (physical_len != stripe_len) {
7703 0 : btrfs_err(fs_info,
7704 : "dev extent physical offset %llu on devid %llu length doesn't match chunk %llu, have %llu expect %llu",
7705 : physical_offset, devid, em->start, physical_len,
7706 : stripe_len);
7707 0 : ret = -EUCLEAN;
7708 0 : goto out;
7709 : }
7710 :
7711 : /*
7712 : * Very old mkfs.btrfs (before v4.1) will not respect the reserved
7713 : * space. Although kernel can handle it without problem, better to warn
7714 : * the users.
7715 : */
7716 35804 : if (physical_offset < BTRFS_DEVICE_RANGE_RESERVED)
7717 0 : btrfs_warn(fs_info,
7718 : "devid %llu physical %llu len %llu inside the reserved space",
7719 : devid, physical_offset, physical_len);
7720 :
7721 42302 : for (i = 0; i < map->num_stripes; i++) {
7722 42302 : if (map->stripes[i].dev->devid == devid &&
7723 42299 : map->stripes[i].physical == physical_offset) {
7724 35804 : found = true;
7725 35804 : if (map->verified_stripes >= map->num_stripes) {
7726 0 : btrfs_err(fs_info,
7727 : "too many dev extents for chunk %llu found",
7728 : em->start);
7729 0 : ret = -EUCLEAN;
7730 0 : goto out;
7731 : }
7732 35804 : map->verified_stripes++;
7733 35804 : break;
7734 : }
7735 : }
7736 35804 : if (!found) {
7737 0 : btrfs_err(fs_info,
7738 : "dev extent physical offset %llu devid %llu has no corresponding chunk",
7739 : physical_offset, devid);
7740 0 : ret = -EUCLEAN;
7741 : }
7742 :
7743 : /* Make sure no dev extent is beyond device boundary */
7744 35804 : dev = btrfs_find_device(fs_info->fs_devices, &args);
7745 35804 : if (!dev) {
7746 0 : btrfs_err(fs_info, "failed to find devid %llu", devid);
7747 0 : ret = -EUCLEAN;
7748 0 : goto out;
7749 : }
7750 :
7751 35804 : if (physical_offset + physical_len > dev->disk_total_bytes) {
7752 0 : btrfs_err(fs_info,
7753 : "dev extent devid %llu physical offset %llu len %llu is beyond device boundary %llu",
7754 : devid, physical_offset, physical_len,
7755 : dev->disk_total_bytes);
7756 0 : ret = -EUCLEAN;
7757 0 : goto out;
7758 : }
7759 :
7760 35804 : if (dev->zone_info) {
7761 0 : u64 zone_size = dev->zone_info->zone_size;
7762 :
7763 0 : if (!IS_ALIGNED(physical_offset, zone_size) ||
7764 0 : !IS_ALIGNED(physical_len, zone_size)) {
7765 0 : btrfs_err(fs_info,
7766 : "zoned: dev extent devid %llu physical offset %llu len %llu is not aligned to device zone",
7767 : devid, physical_offset, physical_len);
7768 0 : ret = -EUCLEAN;
7769 0 : goto out;
7770 : }
7771 : }
7772 :
7773 35804 : out:
7774 35804 : free_extent_map(em);
7775 35804 : return ret;
7776 : }
7777 :
7778 3217 : static int verify_chunk_dev_extent_mapping(struct btrfs_fs_info *fs_info)
7779 : {
7780 3217 : struct extent_map_tree *em_tree = &fs_info->mapping_tree;
7781 3217 : struct extent_map *em;
7782 3217 : struct rb_node *node;
7783 3217 : int ret = 0;
7784 :
7785 3217 : read_lock(&em_tree->lock);
7786 32523 : for (node = rb_first_cached(&em_tree->map); node; node = rb_next(node)) {
7787 29306 : em = rb_entry(node, struct extent_map, rb_node);
7788 29306 : if (em->map_lookup->num_stripes !=
7789 29306 : em->map_lookup->verified_stripes) {
7790 0 : btrfs_err(fs_info,
7791 : "chunk %llu has missing dev extent, have %d expect %d",
7792 : em->start, em->map_lookup->verified_stripes,
7793 : em->map_lookup->num_stripes);
7794 0 : ret = -EUCLEAN;
7795 0 : goto out;
7796 : }
7797 : }
7798 3217 : out:
7799 3217 : read_unlock(&em_tree->lock);
7800 3217 : return ret;
7801 : }
7802 :
7803 : /*
7804 : * Ensure that all dev extents are mapped to correct chunk, otherwise
7805 : * later chunk allocation/free would cause unexpected behavior.
7806 : *
7807 : * NOTE: This will iterate through the whole device tree, which should be of
7808 : * the same size level as the chunk tree. This slightly increases mount time.
7809 : */
7810 3217 : int btrfs_verify_dev_extents(struct btrfs_fs_info *fs_info)
7811 : {
7812 3217 : struct btrfs_path *path;
7813 3217 : struct btrfs_root *root = fs_info->dev_root;
7814 3217 : struct btrfs_key key;
7815 3217 : u64 prev_devid = 0;
7816 3217 : u64 prev_dev_ext_end = 0;
7817 3217 : int ret = 0;
7818 :
7819 : /*
7820 : * We don't have a dev_root because we mounted with ignorebadroots and
7821 : * failed to load the root, so we want to skip the verification in this
7822 : * case for sure.
7823 : *
7824 : * However if the dev root is fine, but the tree itself is corrupted
7825 : * we'd still fail to mount. This verification is only to make sure
7826 : * writes can happen safely, so instead just bypass this check
7827 : * completely in the case of IGNOREBADROOTS.
7828 : */
7829 3217 : if (btrfs_test_opt(fs_info, IGNOREBADROOTS))
7830 : return 0;
7831 :
7832 3217 : key.objectid = 1;
7833 3217 : key.type = BTRFS_DEV_EXTENT_KEY;
7834 3217 : key.offset = 0;
7835 :
7836 3217 : path = btrfs_alloc_path();
7837 3217 : if (!path)
7838 : return -ENOMEM;
7839 :
7840 3217 : path->reada = READA_FORWARD;
7841 3217 : ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
7842 3217 : if (ret < 0)
7843 0 : goto out;
7844 :
7845 3217 : if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
7846 0 : ret = btrfs_next_leaf(root, path);
7847 0 : if (ret < 0)
7848 0 : goto out;
7849 : /* No dev extents at all? Not good */
7850 0 : if (ret > 0) {
7851 0 : ret = -EUCLEAN;
7852 0 : goto out;
7853 : }
7854 : }
7855 35804 : while (1) {
7856 35804 : struct extent_buffer *leaf = path->nodes[0];
7857 35804 : struct btrfs_dev_extent *dext;
7858 35804 : int slot = path->slots[0];
7859 35804 : u64 chunk_offset;
7860 35804 : u64 physical_offset;
7861 35804 : u64 physical_len;
7862 35804 : u64 devid;
7863 :
7864 35804 : btrfs_item_key_to_cpu(leaf, &key, slot);
7865 35804 : if (key.type != BTRFS_DEV_EXTENT_KEY)
7866 : break;
7867 35804 : devid = key.objectid;
7868 35804 : physical_offset = key.offset;
7869 :
7870 35804 : dext = btrfs_item_ptr(leaf, slot, struct btrfs_dev_extent);
7871 35804 : chunk_offset = btrfs_dev_extent_chunk_offset(leaf, dext);
7872 35804 : physical_len = btrfs_dev_extent_length(leaf, dext);
7873 :
7874 : /* Check if this dev extent overlaps with the previous one */
7875 35804 : if (devid == prev_devid && physical_offset < prev_dev_ext_end) {
7876 0 : btrfs_err(fs_info,
7877 : "dev extent devid %llu physical offset %llu overlap with previous dev extent end %llu",
7878 : devid, physical_offset, prev_dev_ext_end);
7879 0 : ret = -EUCLEAN;
7880 0 : goto out;
7881 : }
7882 :
7883 35804 : ret = verify_one_dev_extent(fs_info, chunk_offset, devid,
7884 : physical_offset, physical_len);
7885 35804 : if (ret < 0)
7886 0 : goto out;
7887 35804 : prev_devid = devid;
7888 35804 : prev_dev_ext_end = physical_offset + physical_len;
7889 :
7890 35804 : ret = btrfs_next_item(root, path);
7891 35804 : if (ret < 0)
7892 0 : goto out;
7893 35804 : if (ret > 0) {
7894 : ret = 0;
7895 : break;
7896 : }
7897 : }
7898 :
7899 : /* Ensure all chunks have corresponding dev extents */
7900 3217 : ret = verify_chunk_dev_extent_mapping(fs_info);
7901 3217 : out:
7902 3217 : btrfs_free_path(path);
7903 3217 : return ret;
7904 : }
7905 :
7906 : /*
7907 : * Check whether the given block group or device is pinned by any inode being
7908 : * used as a swapfile.
7909 : */
7910 525 : bool btrfs_pinned_by_swapfile(struct btrfs_fs_info *fs_info, void *ptr)
7911 : {
7912 525 : struct btrfs_swapfile_pin *sp;
7913 525 : struct rb_node *node;
7914 :
7915 525 : spin_lock(&fs_info->swapfile_pins_lock);
7916 525 : node = fs_info->swapfile_pins.rb_node;
7917 546 : while (node) {
7918 23 : sp = rb_entry(node, struct btrfs_swapfile_pin, node);
7919 23 : if (ptr < sp->ptr)
7920 13 : node = node->rb_left;
7921 10 : else if (ptr > sp->ptr)
7922 8 : node = node->rb_right;
7923 : else
7924 : break;
7925 : }
7926 525 : spin_unlock(&fs_info->swapfile_pins_lock);
7927 525 : return node != NULL;
7928 : }
7929 :
7930 : static int relocating_repair_kthread(void *data)
7931 : {
7932 : struct btrfs_block_group *cache = data;
7933 : struct btrfs_fs_info *fs_info = cache->fs_info;
7934 : u64 target;
7935 : int ret = 0;
7936 :
7937 : target = cache->start;
7938 : btrfs_put_block_group(cache);
7939 :
7940 : sb_start_write(fs_info->sb);
7941 : if (!btrfs_exclop_start(fs_info, BTRFS_EXCLOP_BALANCE)) {
7942 : btrfs_info(fs_info,
7943 : "zoned: skip relocating block group %llu to repair: EBUSY",
7944 : target);
7945 : sb_end_write(fs_info->sb);
7946 : return -EBUSY;
7947 : }
7948 :
7949 : mutex_lock(&fs_info->reclaim_bgs_lock);
7950 :
7951 : /* Ensure block group still exists */
7952 : cache = btrfs_lookup_block_group(fs_info, target);
7953 : if (!cache)
7954 : goto out;
7955 :
7956 : if (!test_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags))
7957 : goto out;
7958 :
7959 : ret = btrfs_may_alloc_data_chunk(fs_info, target);
7960 : if (ret < 0)
7961 : goto out;
7962 :
7963 : btrfs_info(fs_info,
7964 : "zoned: relocating block group %llu to repair IO failure",
7965 : target);
7966 : ret = btrfs_relocate_chunk(fs_info, target);
7967 :
7968 : out:
7969 : if (cache)
7970 : btrfs_put_block_group(cache);
7971 : mutex_unlock(&fs_info->reclaim_bgs_lock);
7972 : btrfs_exclop_finish(fs_info);
7973 : sb_end_write(fs_info->sb);
7974 :
7975 : return ret;
7976 : }
7977 :
7978 0 : bool btrfs_repair_one_zone(struct btrfs_fs_info *fs_info, u64 logical)
7979 : {
7980 0 : struct btrfs_block_group *cache;
7981 :
7982 0 : if (!btrfs_is_zoned(fs_info))
7983 0 : return false;
7984 :
7985 : /* Do not attempt to repair in degraded state */
7986 : if (btrfs_test_opt(fs_info, DEGRADED))
7987 : return true;
7988 :
7989 : cache = btrfs_lookup_block_group(fs_info, logical);
7990 : if (!cache)
7991 : return true;
7992 :
7993 : if (test_and_set_bit(BLOCK_GROUP_FLAG_RELOCATING_REPAIR, &cache->runtime_flags)) {
7994 : btrfs_put_block_group(cache);
7995 : return true;
7996 : }
7997 :
7998 : kthread_run(relocating_repair_kthread, cache,
7999 : "btrfs-relocating-repair");
8000 :
8001 : return true;
8002 : }
8003 :
8004 0 : static void map_raid56_repair_block(struct btrfs_io_context *bioc,
8005 : struct btrfs_io_stripe *smap,
8006 : u64 logical)
8007 : {
8008 0 : int data_stripes = nr_bioc_data_stripes(bioc);
8009 0 : int i;
8010 :
8011 0 : for (i = 0; i < data_stripes; i++) {
8012 0 : u64 stripe_start = bioc->full_stripe_logical +
8013 : btrfs_stripe_nr_to_offset(i);
8014 :
8015 0 : if (logical >= stripe_start &&
8016 0 : logical < stripe_start + BTRFS_STRIPE_LEN)
8017 : break;
8018 : }
8019 0 : ASSERT(i < data_stripes);
8020 0 : smap->dev = bioc->stripes[i].dev;
8021 0 : smap->physical = bioc->stripes[i].physical +
8022 0 : ((logical - bioc->full_stripe_logical) &
8023 : BTRFS_STRIPE_LEN_MASK);
8024 0 : }
8025 :
8026 : /*
8027 : * Map a repair write into a single device.
8028 : *
8029 : * A repair write is triggered by read time repair or scrub, which would only
8030 : * update the contents of a single device.
8031 : * Not update any other mirrors nor go through RMW path.
8032 : *
8033 : * Callers should ensure:
8034 : *
8035 : * - Call btrfs_bio_counter_inc_blocked() first
8036 : * - The range does not cross stripe boundary
8037 : * - Has a valid @mirror_num passed in.
8038 : */
8039 2 : int btrfs_map_repair_block(struct btrfs_fs_info *fs_info,
8040 : struct btrfs_io_stripe *smap, u64 logical,
8041 : u32 length, int mirror_num)
8042 : {
8043 2 : struct btrfs_io_context *bioc = NULL;
8044 2 : u64 map_length = length;
8045 2 : int mirror_ret = mirror_num;
8046 2 : int ret;
8047 :
8048 2 : ASSERT(mirror_num > 0);
8049 :
8050 2 : ret = btrfs_map_block(fs_info, BTRFS_MAP_WRITE, logical, &map_length,
8051 : &bioc, smap, &mirror_ret, true);
8052 2 : if (ret < 0)
8053 : return ret;
8054 :
8055 : /* The map range should not cross stripe boundary. */
8056 2 : ASSERT(map_length >= length);
8057 :
8058 : /* Already mapped to single stripe. */
8059 2 : if (!bioc)
8060 0 : goto out;
8061 :
8062 : /* Map the RAID56 multi-stripe writes to a single one. */
8063 2 : if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
8064 0 : map_raid56_repair_block(bioc, smap, logical);
8065 0 : goto out;
8066 : }
8067 :
8068 2 : ASSERT(mirror_num <= bioc->num_stripes);
8069 2 : smap->dev = bioc->stripes[mirror_num - 1].dev;
8070 2 : smap->physical = bioc->stripes[mirror_num - 1].physical;
8071 2 : out:
8072 2 : btrfs_put_bioc(bioc);
8073 2 : ASSERT(smap->dev);
8074 2 : return 0;
8075 : }
|
